diff options
| author | Ian Romanick <ian.d.romanick@intel.com> | 2009-08-18 12:20:36 -0700 |
|---|---|---|
| committer | Ian Romanick <ian.d.romanick@intel.com> | 2009-08-18 12:20:36 -0700 |
| commit | a512985fd81c1ed4ccc5e69aaa05015cf7ff844d (patch) | |
| tree | 69e6e898deaeaed2b4dfb5851707c68261c464de /src/mesa/drivers | |
| parent | 0b5af41c6fae2809f4567a7cecbd207e5e4f3ab5 (diff) | |
| parent | c80bc3abcd3939e5e2d45aea4b01ff22bfec244b (diff) | |
Merge branch 'master' into asm-shader-rework-1
Conflicts:
src/mesa/shader/arbprogparse.c
Diffstat (limited to 'src/mesa/drivers')
252 files changed, 10252 insertions, 6090 deletions
diff --git a/src/mesa/drivers/common/meta.c b/src/mesa/drivers/common/meta.c new file mode 100644 index 0000000000..e42beabc9b --- /dev/null +++ b/src/mesa/drivers/common/meta.c @@ -0,0 +1,1249 @@ +/* + * Mesa 3-D graphics library + * Version: 7.6 + * + * Copyright (C) 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/** + * Meta operations. Some GL operations can be expressed in terms of + * other GL operations. For example, glBlitFramebuffer() can be done + * with texture mapping and glClear() can be done with polygon rendering. + * + * \author Brian Paul + */ + + +#include "main/glheader.h" +#include "main/mtypes.h" +#include "main/imports.h" +#include "main/arrayobj.h" +#include "main/blend.h" +#include "main/bufferobj.h" +#include "main/depth.h" +#include "main/enable.h" +#include "main/image.h" +#include "main/macros.h" +#include "main/matrix.h" +#include "main/polygon.h" +#include "main/scissor.h" +#include "main/shaders.h" +#include "main/stencil.h" +#include "main/texobj.h" +#include "main/texenv.h" +#include "main/teximage.h" +#include "main/texparam.h" +#include "main/texstate.h" +#include "main/varray.h" +#include "main/viewport.h" +#include "shader/program.h" +#include "swrast/swrast.h" +#include "drivers/common/meta.h" + + +/** + * State which we may save/restore across meta ops. + * XXX this may be incomplete... + */ +struct save_state +{ + GLbitfield SavedState; /**< bitmask of META_* flags */ + + /** META_ALPHA_TEST */ + GLboolean AlphaEnabled; + + /** META_BLEND */ + GLboolean BlendEnabled; + GLboolean ColorLogicOpEnabled; + + /** META_COLOR_MASK */ + GLubyte ColorMask[4]; + + /** META_DEPTH_TEST */ + struct gl_depthbuffer_attrib Depth; + + /** META_FOG */ + GLboolean Fog; + + /** META_PIXELSTORE */ + /* XXX / TO-DO */ + + /** META_RASTERIZATION */ + GLenum FrontPolygonMode, BackPolygonMode; + GLboolean PolygonOffset; + GLboolean PolygonSmooth; + GLboolean PolygonStipple; + GLboolean PolygonCull; + + /** META_SCISSOR */ + struct gl_scissor_attrib Scissor; + + /** META_SHADER */ + GLboolean VertexProgramEnabled; + struct gl_vertex_program *VertexProgram; + GLboolean FragmentProgramEnabled; + struct gl_fragment_program *FragmentProgram; + GLuint Shader; + + /** META_STENCIL_TEST */ + struct gl_stencil_attrib Stencil; + + /** META_TRANSFORM */ + GLenum MatrixMode; + GLfloat ModelviewMatrix[16]; + GLfloat ProjectionMatrix[16]; + GLfloat TextureMatrix[16]; + GLbitfield ClipPlanesEnabled; + + /** META_TEXTURE */ + GLuint ActiveUnit; + GLuint ClientActiveUnit; + /** for unit[0] only */ + struct gl_texture_object *CurrentTexture[NUM_TEXTURE_TARGETS]; + /** mask of TEXTURE_2D_BIT, etc */ + GLbitfield TexEnabled[MAX_TEXTURE_UNITS]; + GLbitfield TexGenEnabled[MAX_TEXTURE_UNITS]; + GLuint EnvMode; /* unit[0] only */ + + /** META_VERTEX */ + struct gl_array_object *ArrayObj; + struct gl_buffer_object *ArrayBufferObj; + + /** META_VIEWPORT */ + GLint ViewportX, ViewportY, ViewportW, ViewportH; + GLclampd DepthNear, DepthFar; + + /** Miscellaneous (always disabled) */ + GLboolean Lighting; +}; + + +/** + * State for glBlitFramebufer() + */ +struct blit_state +{ + GLuint TexObj; + GLsizei TexWidth, TexHeight; + GLenum TexType; + GLuint ArrayObj; + GLuint VBO; + GLfloat verts[4][4]; /** four verts of X,Y,S,T */ +}; + + +/** + * State for glClear() + */ +struct clear_state +{ + GLuint ArrayObj; + GLuint VBO; + GLfloat verts[4][7]; /** four verts of X,Y,Z,R,G,B,A */ +}; + + +/** + * State for glCopyPixels() + */ +struct copypix_state +{ + GLuint TexObj; + GLsizei TexWidth, TexHeight; + GLenum TexType; + GLuint ArrayObj; + GLuint VBO; + GLfloat verts[4][5]; /** four verts of X,Y,Z,S,T */ +}; + + +/** + * State for glDrawPixels() + */ +struct drawpix_state +{ + GLuint TexObj; + GLsizei TexWidth, TexHeight; + GLenum TexIntFormat; + GLuint ArrayObj; + GLuint VBO; + GLfloat verts[4][5]; /** four verts of X,Y,Z,S,T */ +}; + + + +/** + * All per-context meta state. + */ +struct gl_meta_state +{ + struct save_state Save; /**< state saved during meta-ops */ + + struct blit_state Blit; /**< For _mesa_meta_blit_framebuffer() */ + struct clear_state Clear; /**< For _mesa_meta_clear() */ + struct copypix_state CopyPix; /**< For _mesa_meta_copy_pixels() */ + struct drawpix_state DrawPix; /**< For _mesa_meta_draw_pixels() */ + + /* other possible meta-ops: + * glDrawPixels() + * glBitmap() + * glGenerateMipmap() + */ +}; + + +/** + * Initialize meta-ops for a context. + * To be called once during context creation. + */ +void +_mesa_meta_init(GLcontext *ctx) +{ + ASSERT(!ctx->Meta); + + ctx->Meta = CALLOC_STRUCT(gl_meta_state); +} + + +/** + * Free context meta-op state. + * To be called once during context destruction. + */ +void +_mesa_meta_free(GLcontext *ctx) +{ + struct gl_meta_state *meta = ctx->Meta; + + if (meta->Blit.TexObj) { + _mesa_DeleteTextures(1, &meta->Blit.TexObj); + _mesa_DeleteBuffersARB(1, & meta->Blit.VBO); + _mesa_DeleteVertexArraysAPPLE(1, &meta->Blit.ArrayObj); + } + + if (meta->Clear.VBO) { + _mesa_DeleteBuffersARB(1, & meta->Clear.VBO); + _mesa_DeleteVertexArraysAPPLE(1, &meta->Clear.ArrayObj); + } + + if (meta->CopyPix.TexObj) { + _mesa_DeleteTextures(1, &meta->CopyPix.TexObj); + _mesa_DeleteBuffersARB(1, & meta->CopyPix.VBO); + _mesa_DeleteVertexArraysAPPLE(1, &meta->CopyPix.ArrayObj); + } + + if (meta->DrawPix.TexObj) { + _mesa_DeleteTextures(1, &meta->DrawPix.TexObj); + _mesa_DeleteBuffersARB(1, & meta->DrawPix.VBO); + _mesa_DeleteVertexArraysAPPLE(1, &meta->DrawPix.ArrayObj); + } + + _mesa_free(ctx->Meta); + ctx->Meta = NULL; +} + + +/** + * Enter meta state. This is like a light-weight version of glPushAttrib + * but it also resets most GL state back to default values. + * + * \param state bitmask of META_* flags indicating which attribute groups + * to save and reset to their defaults + */ +static void +_mesa_meta_begin(GLcontext *ctx, GLbitfield state) +{ + struct save_state *save = &ctx->Meta->Save; + + save->SavedState = state; + + if (state & META_ALPHA_TEST) { + save->AlphaEnabled = ctx->Color.AlphaEnabled; + if (ctx->Color.AlphaEnabled) + _mesa_Disable(GL_ALPHA_TEST); + } + + if (state & META_BLEND) { + save->BlendEnabled = ctx->Color.BlendEnabled; + if (ctx->Color.BlendEnabled) + _mesa_Disable(GL_BLEND); + save->ColorLogicOpEnabled = ctx->Color.ColorLogicOpEnabled; + if (ctx->Color.ColorLogicOpEnabled) + _mesa_Disable(GL_COLOR_LOGIC_OP); + } + + if (state & META_COLOR_MASK) { + COPY_4V(save->ColorMask, ctx->Color.ColorMask); + if (!ctx->Color.ColorMask[0] || + !ctx->Color.ColorMask[1] || + !ctx->Color.ColorMask[2] || + !ctx->Color.ColorMask[3]) + _mesa_ColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE); + } + + if (state & META_DEPTH_TEST) { + save->Depth = ctx->Depth; /* struct copy */ + if (ctx->Depth.Test) + _mesa_Disable(GL_DEPTH_TEST); + } + + if (state & META_FOG) { + save->Fog = ctx->Fog.Enabled; + if (ctx->Fog.Enabled) + _mesa_set_enable(ctx, GL_FOG, GL_FALSE); + } + + if (state & META_RASTERIZATION) { + save->FrontPolygonMode = ctx->Polygon.FrontMode; + save->BackPolygonMode = ctx->Polygon.BackMode; + save->PolygonOffset = ctx->Polygon.OffsetFill; + save->PolygonSmooth = ctx->Polygon.SmoothFlag; + save->PolygonStipple = ctx->Polygon.StippleFlag; + save->PolygonCull = ctx->Polygon.CullFlag; + _mesa_PolygonMode(GL_FRONT_AND_BACK, GL_FILL); + _mesa_set_enable(ctx, GL_POLYGON_OFFSET_FILL, GL_FALSE); + _mesa_set_enable(ctx, GL_POLYGON_SMOOTH, GL_FALSE); + _mesa_set_enable(ctx, GL_POLYGON_STIPPLE, GL_FALSE); + _mesa_set_enable(ctx, GL_CULL_FACE, GL_FALSE); + } + + if (state & META_SCISSOR) { + save->Scissor = ctx->Scissor; /* struct copy */ + } + + if (state & META_SHADER) { + if (ctx->Extensions.ARB_vertex_program) { + save->VertexProgramEnabled = ctx->VertexProgram.Enabled; + save->VertexProgram = ctx->VertexProgram.Current; + _mesa_set_enable(ctx, GL_VERTEX_PROGRAM_ARB, GL_FALSE); + } + + if (ctx->Extensions.ARB_fragment_program) { + save->FragmentProgramEnabled = ctx->FragmentProgram.Enabled; + save->FragmentProgram = ctx->FragmentProgram.Current; + _mesa_set_enable(ctx, GL_FRAGMENT_PROGRAM_ARB, GL_FALSE); + } + + if (ctx->Extensions.ARB_shader_objects) { + save->Shader = ctx->Shader.CurrentProgram ? + ctx->Shader.CurrentProgram->Name : 0; + _mesa_UseProgramObjectARB(0); + } + } + + if (state & META_STENCIL_TEST) { + save->Stencil = ctx->Stencil; /* struct copy */ + if (ctx->Stencil.Enabled) + _mesa_Disable(GL_STENCIL_TEST); + /* NOTE: other stencil state not reset */ + } + + if (state & META_TEXTURE) { + GLuint u, tgt; + + save->ActiveUnit = ctx->Texture.CurrentUnit; + save->ClientActiveUnit = ctx->Array.ActiveTexture; + save->EnvMode = ctx->Texture.Unit[0].EnvMode; + + /* Disable all texture units */ + for (u = 0; u < ctx->Const.MaxTextureUnits; u++) { + save->TexEnabled[u] = ctx->Texture.Unit[u].Enabled; + save->TexGenEnabled[u] = ctx->Texture.Unit[u].TexGenEnabled; + if (ctx->Texture.Unit[u].Enabled || + ctx->Texture.Unit[u].TexGenEnabled) { + _mesa_ActiveTextureARB(GL_TEXTURE0 + u); + _mesa_set_enable(ctx, GL_TEXTURE_1D, GL_FALSE); + _mesa_set_enable(ctx, GL_TEXTURE_2D, GL_FALSE); + _mesa_set_enable(ctx, GL_TEXTURE_3D, GL_FALSE); + _mesa_set_enable(ctx, GL_TEXTURE_CUBE_MAP, GL_FALSE); + _mesa_set_enable(ctx, GL_TEXTURE_RECTANGLE, GL_FALSE); + _mesa_set_enable(ctx, GL_TEXTURE_GEN_S, GL_FALSE); + _mesa_set_enable(ctx, GL_TEXTURE_GEN_T, GL_FALSE); + _mesa_set_enable(ctx, GL_TEXTURE_GEN_R, GL_FALSE); + _mesa_set_enable(ctx, GL_TEXTURE_GEN_Q, GL_FALSE); + } + } + + /* save current texture objects for unit[0] only */ + for (tgt = 0; tgt < NUM_TEXTURE_TARGETS; tgt++) { + _mesa_reference_texobj(&save->CurrentTexture[tgt], + ctx->Texture.Unit[0].CurrentTex[tgt]); + } + + /* set defaults for unit[0] */ + _mesa_ActiveTextureARB(GL_TEXTURE0); + _mesa_ClientActiveTextureARB(GL_TEXTURE0); + _mesa_TexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE); + } + + if (state & META_TRANSFORM) { + GLuint activeTexture = ctx->Texture.CurrentUnit; + _mesa_memcpy(save->ModelviewMatrix, ctx->ModelviewMatrixStack.Top->m, + 16 * sizeof(GLfloat)); + _mesa_memcpy(save->ProjectionMatrix, ctx->ProjectionMatrixStack.Top->m, + 16 * sizeof(GLfloat)); + _mesa_memcpy(save->TextureMatrix, ctx->TextureMatrixStack[0].Top->m, + 16 * sizeof(GLfloat)); + save->MatrixMode = ctx->Transform.MatrixMode; + /* set 1:1 vertex:pixel coordinate transform */ + _mesa_ActiveTextureARB(GL_TEXTURE0); + _mesa_MatrixMode(GL_TEXTURE); + _mesa_LoadIdentity(); + _mesa_ActiveTextureARB(GL_TEXTURE0 + activeTexture); + _mesa_MatrixMode(GL_MODELVIEW); + _mesa_LoadIdentity(); + _mesa_MatrixMode(GL_PROJECTION); + _mesa_LoadIdentity(); + _mesa_Ortho(0.0F, ctx->DrawBuffer->Width, + 0.0F, ctx->DrawBuffer->Height, + -1.0F, 1.0F); + save->ClipPlanesEnabled = ctx->Transform.ClipPlanesEnabled; + if (ctx->Transform.ClipPlanesEnabled) { + GLuint i; + for (i = 0; i < ctx->Const.MaxClipPlanes; i++) { + _mesa_set_enable(ctx, GL_CLIP_PLANE0 + i, GL_FALSE); + } + } + } + + if (state & META_VERTEX) { + /* save vertex array object state */ + _mesa_reference_array_object(ctx, &save->ArrayObj, + ctx->Array.ArrayObj); + _mesa_reference_buffer_object(ctx, &save->ArrayBufferObj, + ctx->Array.ArrayBufferObj); + /* set some default state? */ + } + + if (state & META_VIEWPORT) { + save->ViewportX = ctx->Viewport.X; + save->ViewportY = ctx->Viewport.Y; + save->ViewportW = ctx->Viewport.Width; + save->ViewportH = ctx->Viewport.Height; + _mesa_Viewport(0, 0, ctx->DrawBuffer->Width, ctx->DrawBuffer->Height); + save->DepthNear = ctx->Viewport.Near; + save->DepthFar = ctx->Viewport.Far; + _mesa_DepthRange(0.0, 1.0); + } + + /* misc */ + { + save->Lighting = ctx->Light.Enabled; + if (ctx->Light.Enabled) + _mesa_set_enable(ctx, GL_LIGHTING, GL_FALSE); + } +} + + +/** + * Leave meta state. This is like a light-weight version of glPopAttrib(). + */ +static void +_mesa_meta_end(GLcontext *ctx) +{ + struct save_state *save = &ctx->Meta->Save; + const GLbitfield state = save->SavedState; + + if (state & META_ALPHA_TEST) { + if (ctx->Color.AlphaEnabled != save->AlphaEnabled) + _mesa_set_enable(ctx, GL_ALPHA_TEST, save->AlphaEnabled); + } + + if (state & META_BLEND) { + if (ctx->Color.BlendEnabled != save->BlendEnabled) + _mesa_set_enable(ctx, GL_BLEND, save->BlendEnabled); + if (ctx->Color.ColorLogicOpEnabled != save->ColorLogicOpEnabled) + _mesa_set_enable(ctx, GL_COLOR_LOGIC_OP, save->ColorLogicOpEnabled); + } + + if (state & META_COLOR_MASK) { + if (!TEST_EQ_4V(ctx->Color.ColorMask, save->ColorMask)) + _mesa_ColorMask(save->ColorMask[0], save->ColorMask[1], + save->ColorMask[2], save->ColorMask[3]); + } + + if (state & META_DEPTH_TEST) { + if (ctx->Depth.Test != save->Depth.Test) + _mesa_set_enable(ctx, GL_DEPTH_TEST, save->Depth.Test); + _mesa_DepthFunc(save->Depth.Func); + _mesa_DepthMask(save->Depth.Mask); + } + + if (state & META_FOG) { + _mesa_set_enable(ctx, GL_FOG, save->Fog); + } + + if (state & META_RASTERIZATION) { + _mesa_PolygonMode(GL_FRONT, save->FrontPolygonMode); + _mesa_PolygonMode(GL_BACK, save->BackPolygonMode); + _mesa_set_enable(ctx, GL_POLYGON_STIPPLE, save->PolygonStipple); + _mesa_set_enable(ctx, GL_POLYGON_OFFSET_FILL, save->PolygonOffset); + _mesa_set_enable(ctx, GL_POLYGON_SMOOTH, save->PolygonSmooth); + _mesa_set_enable(ctx, GL_CULL_FACE, save->PolygonCull); + } + + if (state & META_SCISSOR) { + _mesa_set_enable(ctx, GL_SCISSOR_TEST, save->Scissor.Enabled); + _mesa_Scissor(save->Scissor.X, save->Scissor.Y, + save->Scissor.Width, save->Scissor.Height); + } + + if (state & META_SHADER) { + if (ctx->Extensions.ARB_vertex_program) { + _mesa_set_enable(ctx, GL_VERTEX_PROGRAM_ARB, + save->VertexProgramEnabled); + _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, + save->VertexProgram); + } + + if (ctx->Extensions.ARB_fragment_program) { + _mesa_set_enable(ctx, GL_FRAGMENT_PROGRAM_ARB, + save->FragmentProgramEnabled); + _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, + save->FragmentProgram); + } + + if (ctx->Extensions.ARB_shader_objects) { + _mesa_UseProgramObjectARB(save->Shader); + } + } + + if (state & META_STENCIL_TEST) { + const struct gl_stencil_attrib *stencil = &save->Stencil; + + _mesa_set_enable(ctx, GL_STENCIL_TEST, stencil->Enabled); + _mesa_ClearStencil(stencil->Clear); + if (ctx->Extensions.EXT_stencil_two_side) { + _mesa_set_enable(ctx, GL_STENCIL_TEST_TWO_SIDE_EXT, + stencil->TestTwoSide); + _mesa_ActiveStencilFaceEXT(stencil->ActiveFace + ? GL_BACK : GL_FRONT); + } + /* front state */ + _mesa_StencilFuncSeparate(GL_FRONT, + stencil->Function[0], + stencil->Ref[0], + stencil->ValueMask[0]); + _mesa_StencilMaskSeparate(GL_FRONT, stencil->WriteMask[0]); + _mesa_StencilOpSeparate(GL_FRONT, stencil->FailFunc[0], + stencil->ZFailFunc[0], + stencil->ZPassFunc[0]); + /* back state */ + _mesa_StencilFuncSeparate(GL_BACK, + stencil->Function[1], + stencil->Ref[1], + stencil->ValueMask[1]); + _mesa_StencilMaskSeparate(GL_BACK, stencil->WriteMask[1]); + _mesa_StencilOpSeparate(GL_BACK, stencil->FailFunc[1], + stencil->ZFailFunc[1], + stencil->ZPassFunc[1]); + } + + if (state & META_TEXTURE) { + GLuint u, tgt; + + ASSERT(ctx->Texture.CurrentUnit == 0); + + /* restore texenv for unit[0] */ + _mesa_TexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, save->EnvMode); + + /* restore texture objects for unit[0] only */ + for (tgt = 0; tgt < NUM_TEXTURE_TARGETS; tgt++) { + _mesa_reference_texobj(&ctx->Texture.Unit[0].CurrentTex[tgt], + save->CurrentTexture[tgt]); + } + + /* Re-enable textures, texgen */ + for (u = 0; u < ctx->Const.MaxTextureUnits; u++) { + if (save->TexEnabled[u]) { + _mesa_ActiveTextureARB(GL_TEXTURE0 + u); + + if (save->TexEnabled[u] & TEXTURE_1D_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_1D, GL_TRUE); + if (save->TexEnabled[u] & TEXTURE_2D_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_2D, GL_TRUE); + if (save->TexEnabled[u] & TEXTURE_3D_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_3D, GL_TRUE); + if (save->TexEnabled[u] & TEXTURE_CUBE_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_CUBE_MAP, GL_TRUE); + if (save->TexEnabled[u] & TEXTURE_RECT_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_RECTANGLE, GL_TRUE); + } + + if (save->TexGenEnabled[u]) { + _mesa_ActiveTextureARB(GL_TEXTURE0 + u); + + if (save->TexGenEnabled[u] & S_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_GEN_S, GL_TRUE); + if (save->TexGenEnabled[u] & T_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_GEN_T, GL_TRUE); + if (save->TexGenEnabled[u] & R_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_GEN_R, GL_TRUE); + if (save->TexGenEnabled[u] & Q_BIT) + _mesa_set_enable(ctx, GL_TEXTURE_GEN_Q, GL_TRUE); + } + } + + /* restore current unit state */ + _mesa_ActiveTextureARB(GL_TEXTURE0 + save->ActiveUnit); + _mesa_ClientActiveTextureARB(GL_TEXTURE0 + save->ClientActiveUnit); + } + + if (state & META_TRANSFORM) { + GLuint activeTexture = ctx->Texture.CurrentUnit; + _mesa_ActiveTextureARB(GL_TEXTURE0); + _mesa_MatrixMode(GL_TEXTURE); + _mesa_LoadMatrixf(save->TextureMatrix); + _mesa_ActiveTextureARB(GL_TEXTURE0 + activeTexture); + + _mesa_MatrixMode(GL_MODELVIEW); + _mesa_LoadMatrixf(save->ModelviewMatrix); + + _mesa_MatrixMode(GL_PROJECTION); + _mesa_LoadMatrixf(save->ProjectionMatrix); + + _mesa_MatrixMode(save->MatrixMode); + + save->ClipPlanesEnabled = ctx->Transform.ClipPlanesEnabled; + if (save->ClipPlanesEnabled) { + GLuint i; + for (i = 0; i < ctx->Const.MaxClipPlanes; i++) { + if (save->ClipPlanesEnabled & (1 << i)) { + _mesa_set_enable(ctx, GL_CLIP_PLANE0 + i, GL_TRUE); + } + } + } + } + + if (state & META_VERTEX) { + /* restore vertex buffer object */ + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, save->ArrayBufferObj->Name); + _mesa_reference_buffer_object(ctx, &save->ArrayBufferObj, NULL); + + /* restore vertex array object */ + _mesa_BindVertexArray(save->ArrayObj->Name); + _mesa_reference_array_object(ctx, &save->ArrayObj, NULL); + } + + if (state & META_VIEWPORT) { + _mesa_Viewport(save->ViewportX, save->ViewportY, + save->ViewportW, save->ViewportH); + _mesa_DepthRange(save->DepthNear, save->DepthFar); + } + + /* misc */ + if (save->Lighting) { + _mesa_set_enable(ctx, GL_LIGHTING, GL_TRUE); + } + if (save->Fog) { + _mesa_set_enable(ctx, GL_FOG, GL_TRUE); + } +} + + +/** + * Meta implementation of ctx->Driver.BlitFramebuffer() in terms + * of texture mapping and polygon rendering. + * Note: this function requires GL_ARB_texture_rectangle support. + */ +void +_mesa_meta_blit_framebuffer(GLcontext *ctx, + GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, + GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, + GLbitfield mask, GLenum filter) +{ + struct blit_state *blit = &ctx->Meta->Blit; + const GLint srcX = MIN2(srcX0, srcX1); + const GLint srcY = MIN2(srcY0, srcY1); + const GLint srcW = abs(srcX1 - srcX0); + const GLint srcH = abs(srcY1 - srcY0); + GLboolean srcFlipX = srcX1 < srcX0; + GLboolean srcFlipY = srcY1 < srcY0; + + ASSERT(ctx->Extensions.NV_texture_rectangle); + + if (srcW > ctx->Const.MaxTextureRectSize || + srcH > ctx->Const.MaxTextureRectSize) { + /* XXX avoid this fallback */ + _swrast_BlitFramebuffer(ctx, srcX0, srcY0, srcX1, srcY1, + dstX0, dstY0, dstX1, dstY1, mask, filter); + return; + } + + + if (srcFlipX) { + GLint tmp = dstX0; + dstX0 = dstX1; + dstX1 = tmp; + } + + if (srcFlipY) { + GLint tmp = dstY0; + dstY0 = dstY1; + dstY1 = tmp; + } + + /* only scissor effects blit so save/clear all other relevant state */ + _mesa_meta_begin(ctx, ~META_SCISSOR); + + if (blit->TexObj == 0) { + /* one-time setup */ + + /* create texture object */ + _mesa_GenTextures(1, &blit->TexObj); + _mesa_BindTexture(GL_TEXTURE_RECTANGLE, blit->TexObj); + _mesa_TexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE); + } + else { + _mesa_BindTexture(GL_TEXTURE_RECTANGLE, blit->TexObj); + } + + _mesa_TexParameteri(GL_TEXTURE_RECTANGLE, GL_TEXTURE_MIN_FILTER, filter); + _mesa_TexParameteri(GL_TEXTURE_RECTANGLE, GL_TEXTURE_MAG_FILTER, filter); + + if (blit->ArrayObj == 0) { + /* one-time setup */ + + /* create vertex array object */ + _mesa_GenVertexArrays(1, &blit->ArrayObj); + _mesa_BindVertexArray(blit->ArrayObj); + + /* create vertex array buffer */ + _mesa_GenBuffersARB(1, &blit->VBO); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, blit->VBO); + _mesa_BufferDataARB(GL_ARRAY_BUFFER_ARB, sizeof(blit->verts), + blit->verts, GL_STREAM_DRAW_ARB); + + /* setup vertex arrays */ + _mesa_VertexPointer(2, GL_FLOAT, 4 * sizeof(GLfloat), + (void*) (0 * sizeof(GLfloat))); + _mesa_TexCoordPointer(2, GL_FLOAT, 4 * sizeof(GLfloat), + (void *) (2 * sizeof(GLfloat))); + _mesa_EnableClientState(GL_VERTEX_ARRAY); + _mesa_EnableClientState(GL_TEXTURE_COORD_ARRAY); + } + else { + _mesa_BindVertexArray(blit->ArrayObj); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, blit->VBO); + } + + /* vertex positions */ + blit->verts[0][0] = (GLfloat) dstX0; + blit->verts[0][1] = (GLfloat) dstY0; + blit->verts[1][0] = (GLfloat) dstX1; + blit->verts[1][1] = (GLfloat) dstY0; + blit->verts[2][0] = (GLfloat) dstX1; + blit->verts[2][1] = (GLfloat) dstY1; + blit->verts[3][0] = (GLfloat) dstX0; + blit->verts[3][1] = (GLfloat) dstY1; + + /* texcoords */ + blit->verts[0][2] = 0.0F; + blit->verts[0][3] = 0.0F; + blit->verts[1][2] = (GLfloat) srcW; + blit->verts[1][3] = 0.0F; + blit->verts[2][2] = (GLfloat) srcW; + blit->verts[2][3] = (GLfloat) srcH; + blit->verts[3][2] = 0.0F; + blit->verts[3][3] = (GLfloat) srcH; + + /* upload new vertex data */ + _mesa_BufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, + sizeof(blit->verts), blit->verts); + + /* copy framebuffer image to texture */ + if (mask & GL_COLOR_BUFFER_BIT) { + if (blit->TexWidth == srcW && + blit->TexHeight == srcH && + blit->TexType == GL_RGBA) { + /* replace existing tex image */ + _mesa_CopyTexSubImage2D(GL_TEXTURE_RECTANGLE, 0, + 0, 0, srcX, srcY, srcW, srcH); + } + else { + /* create new tex image */ + _mesa_CopyTexImage2D(GL_TEXTURE_RECTANGLE, 0, GL_RGBA, + srcX, srcY, srcW, srcH, 0); + blit->TexWidth = srcW; + blit->TexHeight = srcH; + blit->TexType = GL_RGBA; + } + + mask &= ~GL_COLOR_BUFFER_BIT; + } + + _mesa_Enable(GL_TEXTURE_RECTANGLE); + + /* draw textured quad */ + _mesa_DrawArrays(GL_TRIANGLE_FAN, 0, 4); + + _mesa_Disable(GL_TEXTURE_RECTANGLE); + + _mesa_meta_end(ctx); + + /* XXX, TO-DO: try to handle these cases above! */ + if (mask & (GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT)) { + _swrast_BlitFramebuffer(ctx, srcX0, srcY0, srcX1, srcY1, + dstX0, dstY0, dstX1, dstY1, mask, filter); + } +} + + +/** + * Meta implementation of ctx->Driver.Clear() in terms of polygon rendering. + */ +void +_mesa_meta_clear(GLcontext *ctx, GLbitfield buffers) +{ + struct clear_state *clear = &ctx->Meta->Clear; + GLfloat z = 1.0 - 2.0 * ctx->Depth.Clear; + GLuint i; + + /* only scissor and color mask effects clearing */ + _mesa_meta_begin(ctx, ~(META_SCISSOR | META_COLOR_MASK)); + + if (clear->ArrayObj == 0) { + /* one-time setup */ + + /* create vertex array object */ + _mesa_GenVertexArrays(1, &clear->ArrayObj); + _mesa_BindVertexArray(clear->ArrayObj); + + /* create vertex array buffer */ + _mesa_GenBuffersARB(1, &clear->VBO); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, clear->VBO); + _mesa_BufferDataARB(GL_ARRAY_BUFFER_ARB, sizeof(clear->verts), + clear->verts, GL_STREAM_DRAW_ARB); + + /* setup vertex arrays */ + _mesa_VertexPointer(3, GL_FLOAT, 7 * sizeof(GLfloat), (void *) 0); + _mesa_ColorPointer(4, GL_FLOAT, 7 * sizeof(GLfloat), + (void *) (3 * sizeof(GLfloat))); + _mesa_EnableClientState(GL_VERTEX_ARRAY); + _mesa_EnableClientState(GL_COLOR_ARRAY); + } + else { + _mesa_BindVertexArray(clear->ArrayObj); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, clear->VBO); + } + + /* GL_COLOR_BUFFER_BIT */ + if (buffers & BUFFER_BITS_COLOR) { + /* leave colormask, glDrawBuffer state as-is */ + } + else { + _mesa_ColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE); + } + + /* GL_DEPTH_BUFFER_BIT */ + if (buffers & BUFFER_BIT_DEPTH) { + _mesa_set_enable(ctx, GL_DEPTH_TEST, GL_TRUE); + _mesa_DepthFunc(GL_ALWAYS); + _mesa_DepthMask(GL_TRUE); + } + else { + assert(!ctx->Depth.Test); + } + + /* GL_STENCIL_BUFFER_BIT */ + if (buffers & BUFFER_BIT_STENCIL) { + _mesa_set_enable(ctx, GL_STENCIL_TEST, GL_TRUE); + _mesa_StencilOpSeparate(GL_FRONT_AND_BACK, + GL_REPLACE, GL_REPLACE, GL_REPLACE); + _mesa_StencilFuncSeparate(GL_FRONT_AND_BACK, GL_ALWAYS, + ctx->Stencil.Clear & 0x7fffffff, + ctx->Stencil.WriteMask[0]); + } + else { + assert(!ctx->Stencil.Enabled); + } + + /* vertex positions */ + clear->verts[0][0] = (GLfloat) ctx->DrawBuffer->_Xmin; + clear->verts[0][1] = (GLfloat) ctx->DrawBuffer->_Ymin; + clear->verts[0][2] = z; + clear->verts[1][0] = (GLfloat) ctx->DrawBuffer->_Xmax; + clear->verts[1][1] = (GLfloat) ctx->DrawBuffer->_Ymin; + clear->verts[1][2] = z; + clear->verts[2][0] = (GLfloat) ctx->DrawBuffer->_Xmax; + clear->verts[2][1] = (GLfloat) ctx->DrawBuffer->_Ymax; + clear->verts[2][2] = z; + clear->verts[3][0] = (GLfloat) ctx->DrawBuffer->_Xmin; + clear->verts[3][1] = (GLfloat) ctx->DrawBuffer->_Ymax; + clear->verts[3][2] = z; + + /* vertex colors */ + for (i = 0; i < 4; i++) { + COPY_4FV(&clear->verts[i][3], ctx->Color.ClearColor); + } + + /* upload new vertex data */ + _mesa_BufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, + sizeof(clear->verts), clear->verts); + + /* draw quad */ + _mesa_DrawArrays(GL_TRIANGLE_FAN, 0, 4); + + _mesa_meta_end(ctx); +} + + +/** + * Meta implementation of ctx->Driver.CopyPixels() in terms + * of texture mapping and polygon rendering. + * Note: this function requires GL_ARB_texture_rectangle support. + */ +void +_mesa_meta_copy_pixels(GLcontext *ctx, GLint srcX, GLint srcY, + GLsizei width, GLsizei height, + GLint dstX, GLint dstY, GLenum type) +{ + const GLenum filter = GL_NEAREST; + struct copypix_state *copypix = &ctx->Meta->CopyPix; + const GLfloat z = ctx->Current.RasterPos[2]; + const GLfloat dstX1 = dstX + width * ctx->Pixel.ZoomX; + const GLfloat dstY1 = dstY + height * ctx->Pixel.ZoomY; + + ASSERT(ctx->Extensions.NV_texture_rectangle); + + if (type != GL_COLOR || + ctx->_ImageTransferState || + ctx->Fog.Enabled || + width > ctx->Const.MaxTextureRectSize || + height > ctx->Const.MaxTextureRectSize) { + /* XXX avoid this fallback */ + _swrast_CopyPixels(ctx, srcX, srcY, width, height, dstX, dstY, type); + return; + } + + /* Most GL state applies to glCopyPixels, but a there's a few things + * we need to override: + */ + _mesa_meta_begin(ctx, (META_RASTERIZATION | + META_SHADER | + META_TEXTURE | + META_TRANSFORM | + META_VERTEX | + META_VIEWPORT)); + + if (copypix->TexObj == 0) { + /* one-time setup */ + + /* create texture object */ + _mesa_GenTextures(1, ©pix->TexObj); + _mesa_BindTexture(GL_TEXTURE_RECTANGLE, copypix->TexObj); + _mesa_TexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE); + _mesa_TexParameteri(GL_TEXTURE_RECTANGLE, GL_TEXTURE_MIN_FILTER, filter); + _mesa_TexParameteri(GL_TEXTURE_RECTANGLE, GL_TEXTURE_MAG_FILTER, filter); + } + else { + _mesa_BindTexture(GL_TEXTURE_RECTANGLE, copypix->TexObj); + } + + if (copypix->ArrayObj == 0) { + /* one-time setup */ + + /* create vertex array object */ + _mesa_GenVertexArrays(1, ©pix->ArrayObj); + _mesa_BindVertexArray(copypix->ArrayObj); + + /* create vertex array buffer */ + _mesa_GenBuffersARB(1, ©pix->VBO); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, copypix->VBO); + _mesa_BufferDataARB(GL_ARRAY_BUFFER_ARB, sizeof(copypix->verts), + copypix->verts, GL_STREAM_DRAW_ARB); + + /* setup vertex arrays */ + _mesa_VertexPointer(3, GL_FLOAT, sizeof(copypix->verts[0]), + (void*) (0 * sizeof(GLfloat))); + _mesa_TexCoordPointer(2, GL_FLOAT, sizeof(copypix->verts[0]), + (void *) (3 * sizeof(GLfloat))); + _mesa_EnableClientState(GL_VERTEX_ARRAY); + _mesa_EnableClientState(GL_TEXTURE_COORD_ARRAY); + } + else { + _mesa_BindVertexArray(copypix->ArrayObj); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, copypix->VBO); + } + + /* vertex positions, texcoords */ + copypix->verts[0][0] = (GLfloat) dstX; + copypix->verts[0][1] = (GLfloat) dstY; + copypix->verts[0][2] = z; + copypix->verts[0][3] = 0.0F; + copypix->verts[0][4] = 0.0F; + copypix->verts[1][0] = (GLfloat) dstX1; + copypix->verts[1][1] = (GLfloat) dstY; + copypix->verts[1][2] = z; + copypix->verts[1][3] = (GLfloat) width; + copypix->verts[1][4] = 0.0F; + copypix->verts[2][0] = (GLfloat) dstX1; + copypix->verts[2][1] = (GLfloat) dstY1; + copypix->verts[2][2] = z; + copypix->verts[2][3] = (GLfloat) width; + copypix->verts[2][4] = (GLfloat) height; + copypix->verts[3][0] = (GLfloat) dstX; + copypix->verts[3][1] = (GLfloat) dstY1; + copypix->verts[3][2] = z; + copypix->verts[3][3] = 0.0F; + copypix->verts[3][4] = (GLfloat) height; + + /* upload new vertex data */ + _mesa_BufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, + sizeof(copypix->verts), copypix->verts); + + /* copy framebuffer image to texture */ + if (type == GL_COLOR) { + if (copypix->TexWidth == width && + copypix->TexHeight == height && + copypix->TexType == type) { + /* replace existing tex image */ + _mesa_CopyTexSubImage2D(GL_TEXTURE_RECTANGLE, 0, + 0, 0, srcX, srcY, width, height); + } + else { + /* create new tex image */ + _mesa_CopyTexImage2D(GL_TEXTURE_RECTANGLE, 0, GL_RGBA, + srcX, srcY, width, height, 0); + copypix->TexWidth = width; + copypix->TexHeight = height; + copypix->TexType = type; + } + } + else if (type == GL_DEPTH) { + /* TO-DO: Use a GL_DEPTH_COMPONENT texture and a fragment program/shader + * that replaces the fragment.z value. + */ + } + else { + ASSERT(type == GL_STENCIL); + /* have to use sw fallback */ + } + + _mesa_Enable(GL_TEXTURE_RECTANGLE); + + /* draw textured quad */ + _mesa_DrawArrays(GL_TRIANGLE_FAN, 0, 4); + + _mesa_Disable(GL_TEXTURE_RECTANGLE); + + _mesa_meta_end(ctx); +} + + + +/** + * When the glDrawPixels() image size is greater than the max rectangle + * texture size we use this function to break the glDrawPixels() image + * into tiles which fit into the max texture size. + */ +static void +tiled_draw_pixels(GLcontext *ctx, + GLint x, GLint y, GLsizei width, GLsizei height, + GLenum format, GLenum type, + const struct gl_pixelstore_attrib *unpack, + const GLvoid *pixels) +{ + const GLint maxSize = ctx->Const.MaxTextureRectSize; + struct gl_pixelstore_attrib tileUnpack = *unpack; + GLint i, j; + + for (i = 0; i < width; i += maxSize) { + const GLint tileWidth = MIN2(maxSize, width - i); + const GLint tileX = (GLint) (x + i * ctx->Pixel.ZoomX); + + tileUnpack.SkipPixels = unpack->SkipPixels + i; + + for (j = 0; j < height; j += maxSize) { + const GLint tileHeight = MIN2(maxSize, height - j); + const GLint tileY = (GLint) (y + j * ctx->Pixel.ZoomY); + + tileUnpack.SkipRows = unpack->SkipRows + j; + + _mesa_meta_draw_pixels(ctx, tileX, tileY, + tileWidth, tileHeight, + format, type, &tileUnpack, pixels); + } + } +} + + +/** + * Meta implementation of ctx->Driver.DrawPixels() in terms + * of texture mapping and polygon rendering. + * Note: this function requires GL_ARB_texture_rectangle support. + */ +void +_mesa_meta_draw_pixels(GLcontext *ctx, + GLint x, GLint y, GLsizei width, GLsizei height, + GLenum format, GLenum type, + const struct gl_pixelstore_attrib *unpack, + const GLvoid *pixels) +{ + const GLenum filter = GL_NEAREST; + struct drawpix_state *drawpix = &ctx->Meta->DrawPix; + const GLfloat z = ctx->Current.RasterPos[2]; + const GLfloat x1 = x + width * ctx->Pixel.ZoomX; + const GLfloat y1 = y + height * ctx->Pixel.ZoomY; + const struct gl_pixelstore_attrib unpackSave = ctx->Unpack; + GLenum texIntFormat; + GLboolean fallback; + + ASSERT(ctx->Extensions.NV_texture_rectangle); + + /* + * Determine if we can do the glDrawPixels with texture mapping. + */ + fallback = GL_FALSE; + if (ctx->_ImageTransferState || + ctx->Fog.Enabled) { + fallback = GL_TRUE; + } + + if (_mesa_is_color_format(format)) { + texIntFormat = GL_RGBA; + } + else { + fallback = GL_TRUE; + } + + if (fallback) { + _swrast_DrawPixels(ctx, x, y, width, height, + format, type, unpack, pixels); + return; + } + + /* + * Check image size against max texture size, draw as tiles if needed. + */ + if (width > ctx->Const.MaxTextureRectSize || + height > ctx->Const.MaxTextureRectSize) { + tiled_draw_pixels(ctx, x, y, width, height, + format, type, unpack, pixels); + return; + } + + /* Most GL state applies to glDrawPixels, but a there's a few things + * we need to override: + */ + _mesa_meta_begin(ctx, (META_RASTERIZATION | + META_SHADER | + META_TEXTURE | + META_TRANSFORM | + META_VERTEX | + META_VIEWPORT)); + + if (drawpix->TexObj == 0) { + /* one-time setup */ + + /* create texture object */ + _mesa_GenTextures(1, &drawpix->TexObj); + _mesa_BindTexture(GL_TEXTURE_RECTANGLE, drawpix->TexObj); + _mesa_TexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE); + _mesa_TexParameteri(GL_TEXTURE_RECTANGLE, GL_TEXTURE_MIN_FILTER, filter); + _mesa_TexParameteri(GL_TEXTURE_RECTANGLE, GL_TEXTURE_MAG_FILTER, filter); + } + else { + _mesa_BindTexture(GL_TEXTURE_RECTANGLE, drawpix->TexObj); + } + + if (drawpix->ArrayObj == 0) { + /* one-time setup */ + + /* create vertex array object */ + _mesa_GenVertexArrays(1, &drawpix->ArrayObj); + _mesa_BindVertexArray(drawpix->ArrayObj); + + /* create vertex array buffer */ + _mesa_GenBuffersARB(1, &drawpix->VBO); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, drawpix->VBO); + _mesa_BufferDataARB(GL_ARRAY_BUFFER_ARB, sizeof(drawpix->verts), + drawpix->verts, GL_STREAM_DRAW_ARB); + + /* setup vertex arrays */ + _mesa_VertexPointer(3, GL_FLOAT, sizeof(drawpix->verts[0]), + (void*) (0 * sizeof(GLfloat))); + _mesa_TexCoordPointer(2, GL_FLOAT, sizeof(drawpix->verts[0]), + (void *) (3 * sizeof(GLfloat))); + _mesa_EnableClientState(GL_VERTEX_ARRAY); + _mesa_EnableClientState(GL_TEXTURE_COORD_ARRAY); + } + else { + _mesa_BindVertexArray(drawpix->ArrayObj); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, drawpix->VBO); + } + + /* vertex positions, texcoords */ + drawpix->verts[0][0] = (GLfloat) x; + drawpix->verts[0][1] = (GLfloat) y; + drawpix->verts[0][2] = z; + drawpix->verts[0][3] = 0.0F; + drawpix->verts[0][4] = 0.0F; + drawpix->verts[1][0] = (GLfloat) x1; + drawpix->verts[1][1] = (GLfloat) y; + drawpix->verts[1][2] = z; + drawpix->verts[1][3] = (GLfloat) width; + drawpix->verts[1][4] = 0.0F; + drawpix->verts[2][0] = (GLfloat) x1; + drawpix->verts[2][1] = (GLfloat) y1; + drawpix->verts[2][2] = z; + drawpix->verts[2][3] = (GLfloat) width; + drawpix->verts[2][4] = (GLfloat) height; + drawpix->verts[3][0] = (GLfloat) x; + drawpix->verts[3][1] = (GLfloat) y1; + drawpix->verts[3][2] = z; + drawpix->verts[3][3] = 0.0F; + drawpix->verts[3][4] = (GLfloat) height; + + /* upload new vertex data */ + _mesa_BufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, + sizeof(drawpix->verts), drawpix->verts); + + /* set given unpack params */ + ctx->Unpack = *unpack; /* XXX bufobj */ + + /* copy pixel data to texture */ + if (drawpix->TexWidth == width && + drawpix->TexHeight == height && + drawpix->TexIntFormat == texIntFormat) { + /* replace existing tex image */ + _mesa_TexSubImage2D(GL_TEXTURE_RECTANGLE, 0, + 0, 0, width, height, format, type, pixels); + } + else { + /* create new tex image */ + _mesa_TexImage2D(GL_TEXTURE_RECTANGLE, 0, texIntFormat, + width, height, 0, format, type, pixels); + drawpix->TexWidth = width; + drawpix->TexHeight = height; + drawpix->TexIntFormat = texIntFormat; + } + + /* restore unpack params */ + ctx->Unpack = unpackSave; /* XXX bufobj */ + + _mesa_Enable(GL_TEXTURE_RECTANGLE); + + /* draw textured quad */ + _mesa_DrawArrays(GL_TRIANGLE_FAN, 0, 4); + + _mesa_Disable(GL_TEXTURE_RECTANGLE); + + _mesa_meta_end(ctx); +} diff --git a/src/mesa/drivers/common/meta.h b/src/mesa/drivers/common/meta.h new file mode 100644 index 0000000000..a9c5f98043 --- /dev/null +++ b/src/mesa/drivers/common/meta.h @@ -0,0 +1,80 @@ +/* + * Mesa 3-D graphics library + * Version: 7.6 + * + * Copyright (C) 2009 VMware, Inc. All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN + * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + + +#ifndef META_H +#define META_H + + +/** + * Flags passed to _mesa_meta_begin(). + * XXX these flags may evolve... + */ +/*@{*/ +#define META_ALPHA_TEST 0x1 +#define META_BLEND 0x2 /**< includes logicop */ +#define META_COLOR_MASK 0x4 +#define META_DEPTH_TEST 0x8 +#define META_FOG 0x10 +#define META_RASTERIZATION 0x20 +#define META_SCISSOR 0x40 +#define META_SHADER 0x80 +#define META_STENCIL_TEST 0x100 +#define META_TRANSFORM 0x200 /**< modelview, projection */ +#define META_TEXTURE 0x400 +#define META_VERTEX 0x800 +#define META_VIEWPORT 0x1000 +#define META_ALL ~0x0 +/*@}*/ + + +extern void +_mesa_meta_init(GLcontext *ctx); + +extern void +_mesa_meta_free(GLcontext *ctx); + +extern void +_mesa_meta_blit_framebuffer(GLcontext *ctx, + GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, + GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, + GLbitfield mask, GLenum filter); + +extern void +_mesa_meta_clear(GLcontext *ctx, GLbitfield buffers); + +extern void +_mesa_meta_copy_pixels(GLcontext *ctx, GLint srcx, GLint srcy, + GLsizei width, GLsizei height, + GLint dstx, GLint dsty, GLenum type); + +extern void +_mesa_meta_draw_pixels(GLcontext *ctx, + GLint x, GLint y, GLsizei width, GLsizei height, + GLenum format, GLenum type, + const struct gl_pixelstore_attrib *unpack, + const GLvoid *pixels); + + +#endif /* META_H */ diff --git a/src/mesa/drivers/dri/common/.gitignore b/src/mesa/drivers/dri/common/.gitignore new file mode 100644 index 0000000000..1edeb79fd1 --- /dev/null +++ b/src/mesa/drivers/dri/common/.gitignore @@ -0,0 +1 @@ +*.os diff --git a/src/mesa/drivers/dri/common/dri_metaops.c b/src/mesa/drivers/dri/common/dri_metaops.c index fe183c2e87..cdbea34495 100644 --- a/src/mesa/drivers/dri/common/dri_metaops.c +++ b/src/mesa/drivers/dri/common/dri_metaops.c @@ -357,6 +357,7 @@ meta_clear_tris(struct dri_metaops *meta, GLbitfield mask) GLuint saved_shader_program = 0; unsigned int saved_active_texture; struct gl_array_object *arraySave = NULL; + GLfloat saved_near, saved_far; if (!meta->clear.arrayObj) meta_init_clear(meta); @@ -370,8 +371,7 @@ meta_clear_tris(struct dri_metaops *meta, GLbitfield mask) GL_POLYGON_BIT | GL_STENCIL_BUFFER_BIT | GL_TRANSFORM_BIT | - GL_CURRENT_BIT | - GL_VIEWPORT_BIT); + GL_CURRENT_BIT); saved_active_texture = ctx->Texture.CurrentUnit; /* Disable existing GL state we don't want to apply to a clear. */ @@ -440,6 +440,8 @@ meta_clear_tris(struct dri_metaops *meta, GLbitfield mask) /* The ClearDepth value is unaffected by DepthRange, so do a default * mapping. */ + saved_near = ctx->Viewport.Near; + saved_far = ctx->Viewport.Far; _mesa_DepthRange(0.0, 1.0); /* Prepare the vertices, which are the same regardless of which buffer we're @@ -519,6 +521,7 @@ meta_clear_tris(struct dri_metaops *meta, GLbitfield mask) if (saved_shader_program) _mesa_UseProgramObjectARB(saved_shader_program); + _mesa_DepthRange(saved_near, saved_far); _mesa_PopAttrib(); /* restore current array object */ diff --git a/src/mesa/drivers/dri/common/dri_util.c b/src/mesa/drivers/dri/common/dri_util.c index 1d940603fa..e48e10d7c0 100644 --- a/src/mesa/drivers/dri/common/dri_util.c +++ b/src/mesa/drivers/dri/common/dri_util.c @@ -778,7 +778,7 @@ dri2CreateNewScreen(int scrn, int fd, if (driDriverAPI.InitScreen2 == NULL) return NULL; - psp = _mesa_malloc(sizeof(*psp)); + psp = _mesa_calloc(sizeof(*psp)); if (!psp) return NULL; diff --git a/src/mesa/drivers/dri/common/extension_helper.h b/src/mesa/drivers/dri/common/extension_helper.h index e308fd2831..08a97bb111 100644 --- a/src/mesa/drivers/dri/common/extension_helper.h +++ b/src/mesa/drivers/dri/common/extension_helper.h @@ -752,6 +752,13 @@ static const char VertexAttrib4ubNV_names[] = ""; #endif +#if defined(need_GL_APPLE_texture_range) +static const char TextureRangeAPPLE_names[] = + "iip\0" /* Parameter signature */ + "glTextureRangeAPPLE\0" + ""; +#endif + #if defined(need_GL_SUN_vertex) static const char TexCoord2fColor4fNormal3fVertex3fSUN_names[] = "ffffffffffff\0" /* Parameter signature */ @@ -1567,6 +1574,13 @@ static const char FragmentLightivSGIX_names[] = ""; #endif +#if defined(need_GL_APPLE_texture_range) +static const char GetTexParameterPointervAPPLE_names[] = + "iip\0" /* Parameter signature */ + "glGetTexParameterPointervAPPLE\0" + ""; +#endif + #if defined(need_GL_EXT_pixel_transform) static const char PixelTransformParameterfvEXT_names[] = "iip\0" /* Parameter signature */ @@ -2079,6 +2093,13 @@ static const char Uniform2fvARB_names[] = ""; #endif +#if defined(need_GL_APPLE_flush_buffer_range) +static const char BufferParameteriAPPLE_names[] = + "iii\0" /* Parameter signature */ + "glBufferParameteriAPPLE\0" + ""; +#endif + #if defined(need_GL_VERSION_1_3) static const char MultiTexCoord3dvARB_names[] = "ip\0" /* Parameter signature */ @@ -2803,6 +2824,13 @@ static const char IsProgramNV_names[] = ""; #endif +#if defined(need_GL_APPLE_flush_buffer_range) +static const char FlushMappedBufferRangeAPPLE_names[] = + "iii\0" /* Parameter signature */ + "glFlushMappedBufferRangeAPPLE\0" + ""; +#endif + #if defined(need_GL_SUN_triangle_list) static const char ReplacementCodePointerSUN_names[] = "iip\0" /* Parameter signature */ @@ -4973,6 +5001,22 @@ static const struct dri_extension_function GL_3DFX_tbuffer_functions[] = { }; #endif +#if defined(need_GL_APPLE_flush_buffer_range) +static const struct dri_extension_function GL_APPLE_flush_buffer_range_functions[] = { + { BufferParameteriAPPLE_names, BufferParameteriAPPLE_remap_index, -1 }, + { FlushMappedBufferRangeAPPLE_names, FlushMappedBufferRangeAPPLE_remap_index, -1 }, + { NULL, 0, 0 } +}; +#endif + +#if defined(need_GL_APPLE_texture_range) +static const struct dri_extension_function GL_APPLE_texture_range_functions[] = { + { TextureRangeAPPLE_names, TextureRangeAPPLE_remap_index, -1 }, + { GetTexParameterPointervAPPLE_names, GetTexParameterPointervAPPLE_remap_index, -1 }, + { NULL, 0, 0 } +}; +#endif + #if defined(need_GL_APPLE_vertex_array_object) static const struct dri_extension_function GL_APPLE_vertex_array_object_functions[] = { { DeleteVertexArraysAPPLE_names, DeleteVertexArraysAPPLE_remap_index, -1 }, diff --git a/src/mesa/drivers/dri/fb/fb_egl.c b/src/mesa/drivers/dri/fb/fb_egl.c index dee67feb5a..4e41860d8c 100644 --- a/src/mesa/drivers/dri/fb/fb_egl.c +++ b/src/mesa/drivers/dri/fb/fb_egl.c @@ -605,7 +605,7 @@ fbDestroySurface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface) { fbSurface *fs = Lookup_fbSurface(surface); _eglUnlinkSurface(&fs->Base); - if (!fs->Base.IsBound) + if (!_eglIsSurfaceBound(&fs->Base)) free(fs); return EGL_TRUE; } @@ -616,7 +616,7 @@ fbDestroyContext(_EGLDriver *drv, EGLDisplay dpy, EGLContext context) { fbContext *fc = Lookup_fbContext(context); _eglUnlinkContext(&fc->Base); - if (!fc->Base.IsBound) + if (!_eglIsContextBound(&fc->Base)) free(fc); return EGL_TRUE; } diff --git a/src/mesa/drivers/dri/i915/i830_context.h b/src/mesa/drivers/dri/i915/i830_context.h index 1bdb32049d..f73cbbf88b 100644 --- a/src/mesa/drivers/dri/i915/i830_context.h +++ b/src/mesa/drivers/dri/i915/i830_context.h @@ -40,6 +40,7 @@ #define I830_UPLOAD_BUFFERS 0x2 #define I830_UPLOAD_STIPPLE 0x4 #define I830_UPLOAD_INVARIENT 0x8 +#define I830_UPLOAD_RASTER_RULES 0x10 #define I830_UPLOAD_TEX(i) (0x10<<(i)) #define I830_UPLOAD_TEXBLEND(i) (0x100<<(i)) #define I830_UPLOAD_TEX_ALL (0x0f0) @@ -99,6 +100,11 @@ #define I830_TEXBLEND_SIZE 12 /* (4 args + op) * 2 + COLOR_FACTOR */ +enum { + I830_RASTER_RULES, + I830_RASTER_RULES_SIZE +}; + struct i830_texture_object { struct intel_texture_object intel; @@ -112,6 +118,7 @@ struct i830_hw_state GLuint Ctx[I830_CTX_SETUP_SIZE]; GLuint Buffer[I830_DEST_SETUP_SIZE]; GLuint Stipple[I830_STP_SETUP_SIZE]; + GLuint RasterRules[I830_RASTER_RULES_SIZE]; GLuint Tex[I830_TEX_UNITS][I830_TEX_SETUP_SIZE]; GLuint TexBlend[I830_TEX_UNITS][I830_TEXBLEND_SIZE]; GLuint TexBlendWordsUsed[I830_TEX_UNITS]; @@ -197,6 +204,7 @@ extern void i830InitStateFuncs(struct dd_function_table *functions); extern void i830EmitState(struct i830_context *i830); extern void i830InitState(struct i830_context *i830); +extern void i830_update_provoking_vertex(GLcontext *ctx); /* i830_metaops.c */ diff --git a/src/mesa/drivers/dri/i915/i830_reg.h b/src/mesa/drivers/dri/i915/i830_reg.h index db16871001..ae1317029a 100644 --- a/src/mesa/drivers/dri/i915/i830_reg.h +++ b/src/mesa/drivers/dri/i915/i830_reg.h @@ -420,8 +420,11 @@ #define ENABLE_LINE_STRIP_PROVOKE_VRTX (1<<8) #define ENABLE_TRI_FAN_PROVOKE_VRTX (1<<5) #define ENABLE_TRI_STRIP_PROVOKE_VRTX (1<<2) +#define LINE_STRIP_PROVOKE_VRTX_MASK (3<<6) #define LINE_STRIP_PROVOKE_VRTX(x) ((x)<<6) +#define TRI_FAN_PROVOKE_VRTX_MASK (3<<3) #define TRI_FAN_PROVOKE_VRTX(x) ((x)<<3) +#define TRI_STRIP_PROVOKE_VRTX_MASK (3<<0) #define TRI_STRIP_PROVOKE_VRTX(x) (x) /* _3DSTATE_SCISSOR_ENABLE, p200 */ diff --git a/src/mesa/drivers/dri/i915/i830_state.c b/src/mesa/drivers/dri/i915/i830_state.c index 8ef6c9144f..645ebe3057 100644 --- a/src/mesa/drivers/dri/i915/i830_state.c +++ b/src/mesa/drivers/dri/i915/i830_state.c @@ -1047,6 +1047,16 @@ i830_init_packets(struct i830_context *i830) TEXBIND_SET1(TEXCOORDSRC_VTXSET_1) | TEXBIND_SET0(TEXCOORDSRC_VTXSET_0)); + i830->state.RasterRules[I830_RASTER_RULES] = (_3DSTATE_RASTER_RULES_CMD | + ENABLE_POINT_RASTER_RULE | + OGL_POINT_RASTER_RULE | + ENABLE_LINE_STRIP_PROVOKE_VRTX | + ENABLE_TRI_FAN_PROVOKE_VRTX | + ENABLE_TRI_STRIP_PROVOKE_VRTX | + LINE_STRIP_PROVOKE_VRTX(1) | + TRI_FAN_PROVOKE_VRTX(2) | + TRI_STRIP_PROVOKE_VRTX(2)); + i830->state.Stipple[I830_STPREG_ST0] = _3DSTATE_STIPPLE; @@ -1058,6 +1068,27 @@ i830_init_packets(struct i830_context *i830) i830->state.Buffer[I830_DESTREG_SR2] = 0; } +void +i830_update_provoking_vertex(GLcontext * ctx) +{ + struct i830_context *i830 = i830_context(ctx); + + I830_STATECHANGE(i830, I830_UPLOAD_RASTER_RULES); + i830->state.RasterRules[I830_RASTER_RULES] &= ~(LINE_STRIP_PROVOKE_VRTX_MASK | + TRI_FAN_PROVOKE_VRTX_MASK | + TRI_STRIP_PROVOKE_VRTX_MASK); + + /* _NEW_LIGHT */ + if (ctx->Light.ProvokingVertex == GL_LAST_VERTEX_CONVENTION) { + i830->state.RasterRules[I830_RASTER_RULES] |= (LINE_STRIP_PROVOKE_VRTX(1) | + TRI_FAN_PROVOKE_VRTX(2) | + TRI_STRIP_PROVOKE_VRTX(2)); + } else { + i830->state.RasterRules[I830_RASTER_RULES] |= (LINE_STRIP_PROVOKE_VRTX(0) | + TRI_FAN_PROVOKE_VRTX(1) | + TRI_STRIP_PROVOKE_VRTX(0)); + } +} void i830InitStateFuncs(struct dd_function_table *functions) @@ -1101,6 +1132,7 @@ i830InitState(struct i830_context *i830) i830->current = &i830->state; i830->state.emitted = 0; i830->state.active = (I830_UPLOAD_INVARIENT | + I830_UPLOAD_RASTER_RULES | I830_UPLOAD_TEXBLEND(0) | I830_UPLOAD_STIPPLE | I830_UPLOAD_CTX | I830_UPLOAD_BUFFERS); diff --git a/src/mesa/drivers/dri/i915/i830_vtbl.c b/src/mesa/drivers/dri/i915/i830_vtbl.c index 9c6f891dd3..983f6724c9 100644 --- a/src/mesa/drivers/dri/i915/i830_vtbl.c +++ b/src/mesa/drivers/dri/i915/i830_vtbl.c @@ -299,7 +299,7 @@ i830_emit_invarient_state(struct intel_context *intel) { BATCH_LOCALS; - BEGIN_BATCH(30, IGNORE_CLIPRECTS); + BEGIN_BATCH(29, IGNORE_CLIPRECTS); OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD); OUT_BATCH(0); @@ -351,15 +351,6 @@ i830_emit_invarient_state(struct intel_context *intel) OUT_BATCH(_3DSTATE_MAP_COORD_TRANSFORM); OUT_BATCH(DISABLE_TEX_TRANSFORM | TEXTURE_SET(3)); - OUT_BATCH(_3DSTATE_RASTER_RULES_CMD | - ENABLE_POINT_RASTER_RULE | - OGL_POINT_RASTER_RULE | - ENABLE_LINE_STRIP_PROVOKE_VRTX | - ENABLE_TRI_FAN_PROVOKE_VRTX | - ENABLE_TRI_STRIP_PROVOKE_VRTX | - LINE_STRIP_PROVOKE_VRTX(1) | - TRI_FAN_PROVOKE_VRTX(2) | TRI_STRIP_PROVOKE_VRTX(2)); - OUT_BATCH(_3DSTATE_VERTEX_TRANSFORM); OUT_BATCH(DISABLE_VIEWPORT_TRANSFORM | DISABLE_PERSPECTIVE_DIVIDE); @@ -394,6 +385,9 @@ get_state_size(struct i830_hw_state *state) if (dirty & I830_UPLOAD_INVARIENT) sz += 40 * sizeof(int); + if (dirty & I830_UPLOAD_RASTER_RULES) + sz += sizeof(state->RasterRules); + if (dirty & I830_UPLOAD_CTX) sz += sizeof(state->Ctx); @@ -486,6 +480,11 @@ i830_emit_state(struct intel_context *intel) i830_emit_invarient_state(intel); } + if (dirty & I830_UPLOAD_RASTER_RULES) { + DBG("I830_UPLOAD_RASTER_RULES:\n"); + emit(intel, state->RasterRules, sizeof(state->RasterRules)); + } + if (dirty & I830_UPLOAD_CTX) { DBG("I830_UPLOAD_CTX:\n"); emit(intel, state->Ctx, sizeof(state->Ctx)); @@ -737,6 +736,13 @@ i830_assert_not_dirty( struct intel_context *intel ) assert(!get_dirty(state)); } +static void +i830_invalidate_state(struct intel_context *intel, GLuint new_state) +{ + if (new_state & _NEW_LIGHT) + i830_update_provoking_vertex(&intel->ctx); +} + void i830InitVtbl(struct i830_context *i830) { @@ -752,4 +758,5 @@ i830InitVtbl(struct i830_context *i830) i830->intel.vtbl.render_prevalidate = i830_render_prevalidate; i830->intel.vtbl.assert_not_dirty = i830_assert_not_dirty; i830->intel.vtbl.finish_batch = intel_finish_vb; + i830->intel.vtbl.invalidate_state = i830_invalidate_state; } diff --git a/src/mesa/drivers/dri/i915/i915_context.c b/src/mesa/drivers/dri/i915/i915_context.c index 367d2a3b64..bb08cf8d18 100644 --- a/src/mesa/drivers/dri/i915/i915_context.c +++ b/src/mesa/drivers/dri/i915/i915_context.c @@ -75,6 +75,10 @@ i915InvalidateState(GLcontext * ctx, GLuint new_state) if (new_state & (_NEW_FOG | _NEW_HINT | _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS)) i915_update_fog(ctx); + if (new_state & (_NEW_STENCIL | _NEW_BUFFERS | _NEW_POLYGON)) + i915_update_stencil(ctx); + if (new_state & (_NEW_LIGHT)) + i915_update_provoking_vertex(ctx); } diff --git a/src/mesa/drivers/dri/i915/i915_context.h b/src/mesa/drivers/dri/i915/i915_context.h index 87bbf5f927..8de4a9d0d3 100644 --- a/src/mesa/drivers/dri/i915/i915_context.h +++ b/src/mesa/drivers/dri/i915/i915_context.h @@ -48,6 +48,7 @@ #define I915_UPLOAD_FOG 0x20 #define I915_UPLOAD_INVARIENT 0x40 #define I915_UPLOAD_DEFAULTS 0x80 +#define I915_UPLOAD_RASTER_RULES 0x100 #define I915_UPLOAD_TEX(i) (0x00010000<<(i)) #define I915_UPLOAD_TEX_ALL (0x00ff0000) #define I915_UPLOAD_TEX_0_SHIFT 16 @@ -82,7 +83,9 @@ #define I915_CTXREG_IAB 6 #define I915_CTXREG_BLENDCOLOR0 7 #define I915_CTXREG_BLENDCOLOR1 8 -#define I915_CTX_SETUP_SIZE 9 +#define I915_CTXREG_BF_STENCIL_OPS 9 +#define I915_CTXREG_BF_STENCIL_MASKS 10 +#define I915_CTX_SETUP_SIZE 11 #define I915_FOGREG_COLOR 0 #define I915_FOGREG_MODE0 1 @@ -110,6 +113,10 @@ #define I915_DEFREG_Z1 5 #define I915_DEF_SETUP_SIZE 6 +enum { + I915_RASTER_RULES, + I915_RASTER_RULES_SETUP_SIZE, +}; #define I915_MAX_CONSTANT 32 #define I915_CONSTANT_SIZE (2+(4*I915_MAX_CONSTANT)) @@ -206,6 +213,7 @@ struct i915_hw_state GLuint Stipple[I915_STP_SETUP_SIZE]; GLuint Fog[I915_FOG_SETUP_SIZE]; GLuint Defaults[I915_DEF_SETUP_SIZE]; + GLuint RasterRules[I915_RASTER_RULES_SETUP_SIZE]; GLuint Tex[I915_TEX_UNITS][I915_TEX_SETUP_SIZE]; GLuint Constant[I915_CONSTANT_SIZE]; GLuint ConstantSize; @@ -321,6 +329,8 @@ extern void i915_print_ureg(const char *msg, GLuint ureg); extern void i915InitStateFunctions(struct dd_function_table *functions); extern void i915InitState(struct i915_context *i915); extern void i915_update_fog(GLcontext * ctx); +extern void i915_update_stencil(GLcontext * ctx); +extern void i915_update_provoking_vertex(GLcontext *ctx); /*====================================================================== diff --git a/src/mesa/drivers/dri/i915/i915_reg.h b/src/mesa/drivers/dri/i915/i915_reg.h index 84db58ea95..b5fa7fddb9 100644 --- a/src/mesa/drivers/dri/i915/i915_reg.h +++ b/src/mesa/drivers/dri/i915/i915_reg.h @@ -86,8 +86,10 @@ #define BFM_ENABLE_STENCIL_WRITE_MASK (1<<16) #define BFM_STENCIL_TEST_MASK_SHIFT 8 #define BFM_STENCIL_TEST_MASK_MASK (0xff<<8) +#define BFM_STENCIL_TEST_MASK(x) (((x)&0xff) << 8) #define BFM_STENCIL_WRITE_MASK_SHIFT 0 #define BFM_STENCIL_WRITE_MASK_MASK (0xff<<0) +#define BFM_STENCIL_WRITE_MASK(x) ((x)&0xff) @@ -295,7 +297,9 @@ #define TEXKILL_4D (1<<9) #define ENABLE_LINE_STRIP_PROVOKE_VRTX (1<<8) #define ENABLE_TRI_FAN_PROVOKE_VRTX (1<<5) +#define LINE_STRIP_PROVOKE_VRTX_MASK (3 << 6) #define LINE_STRIP_PROVOKE_VRTX(x) ((x)<<6) +#define TRI_FAN_PROVOKE_VRTX_MASK (3 << 3) #define TRI_FAN_PROVOKE_VRTX(x) ((x)<<3) /* _3DSTATE_SCISSOR_ENABLE, p256 */ diff --git a/src/mesa/drivers/dri/i915/i915_state.c b/src/mesa/drivers/dri/i915/i915_state.c index 814fb59fd3..b60efea75b 100644 --- a/src/mesa/drivers/dri/i915/i915_state.c +++ b/src/mesa/drivers/dri/i915/i915_state.c @@ -48,73 +48,119 @@ #define FILE_DEBUG_FLAG DEBUG_STATE -static void -i915StencilFuncSeparate(GLcontext * ctx, GLenum face, GLenum func, GLint ref, - GLuint mask) +void +i915_update_stencil(GLcontext * ctx) { struct i915_context *i915 = I915_CONTEXT(ctx); - int test = intel_translate_compare_func(func); + GLuint front_ref, front_writemask, front_mask; + GLenum front_func, front_fail, front_pass_z_fail, front_pass_z_pass; + GLuint back_ref, back_writemask, back_mask; + GLenum back_func, back_fail, back_pass_z_fail, back_pass_z_pass; - mask = mask & 0xff; - - DBG("%s : func: %s, ref : 0x%x, mask: 0x%x\n", __FUNCTION__, - _mesa_lookup_enum_by_nr(func), ref, mask); + I915_STATECHANGE(i915, I915_UPLOAD_CTX); + /* The 915 considers CW to be "front" for two-sided stencil, so choose + * appropriately. + */ + /* _NEW_POLYGON | _NEW_STENCIL */ + if (ctx->Polygon.FrontFace == GL_CW) { + front_ref = ctx->Stencil.Ref[0]; + front_mask = ctx->Stencil.ValueMask[0]; + front_writemask = ctx->Stencil.WriteMask[0]; + front_func = ctx->Stencil.Function[0]; + front_fail = ctx->Stencil.FailFunc[0]; + front_pass_z_fail = ctx->Stencil.ZFailFunc[0]; + front_pass_z_pass = ctx->Stencil.ZPassFunc[0]; + back_ref = ctx->Stencil.Ref[ctx->Stencil._BackFace]; + back_mask = ctx->Stencil.ValueMask[ctx->Stencil._BackFace]; + back_writemask = ctx->Stencil.WriteMask[ctx->Stencil._BackFace]; + back_func = ctx->Stencil.Function[ctx->Stencil._BackFace]; + back_fail = ctx->Stencil.FailFunc[ctx->Stencil._BackFace]; + back_pass_z_fail = ctx->Stencil.ZFailFunc[ctx->Stencil._BackFace]; + back_pass_z_pass = ctx->Stencil.ZPassFunc[ctx->Stencil._BackFace]; + } else { + front_ref = ctx->Stencil.Ref[ctx->Stencil._BackFace]; + front_mask = ctx->Stencil.ValueMask[ctx->Stencil._BackFace]; + front_writemask = ctx->Stencil.WriteMask[ctx->Stencil._BackFace]; + front_func = ctx->Stencil.Function[ctx->Stencil._BackFace]; + front_fail = ctx->Stencil.FailFunc[ctx->Stencil._BackFace]; + front_pass_z_fail = ctx->Stencil.ZFailFunc[ctx->Stencil._BackFace]; + front_pass_z_pass = ctx->Stencil.ZPassFunc[ctx->Stencil._BackFace]; + back_ref = ctx->Stencil.Ref[0]; + back_mask = ctx->Stencil.ValueMask[0]; + back_writemask = ctx->Stencil.WriteMask[0]; + back_func = ctx->Stencil.Function[0]; + back_fail = ctx->Stencil.FailFunc[0]; + back_pass_z_fail = ctx->Stencil.ZFailFunc[0]; + back_pass_z_pass = ctx->Stencil.ZPassFunc[0]; + } - I915_STATECHANGE(i915, I915_UPLOAD_CTX); - i915->state.Ctx[I915_CTXREG_STATE4] &= ~MODE4_ENABLE_STENCIL_TEST_MASK; + /* Set front state. */ + i915->state.Ctx[I915_CTXREG_STATE4] &= ~(MODE4_ENABLE_STENCIL_TEST_MASK | + MODE4_ENABLE_STENCIL_WRITE_MASK); i915->state.Ctx[I915_CTXREG_STATE4] |= (ENABLE_STENCIL_TEST_MASK | - STENCIL_TEST_MASK(mask)); + ENABLE_STENCIL_WRITE_MASK | + STENCIL_TEST_MASK(front_mask) | + STENCIL_WRITE_MASK(front_writemask)); i915->state.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_REF_MASK | - S5_STENCIL_TEST_FUNC_MASK); + S5_STENCIL_TEST_FUNC_MASK | + S5_STENCIL_FAIL_MASK | + S5_STENCIL_PASS_Z_FAIL_MASK | + S5_STENCIL_PASS_Z_PASS_MASK); + + i915->state.Ctx[I915_CTXREG_LIS5] |= + (front_ref << S5_STENCIL_REF_SHIFT) | + (intel_translate_compare_func(front_func) << S5_STENCIL_TEST_FUNC_SHIFT) | + (intel_translate_stencil_op(front_fail) << S5_STENCIL_FAIL_SHIFT) | + (intel_translate_stencil_op(front_pass_z_fail) << + S5_STENCIL_PASS_Z_FAIL_SHIFT) | + (intel_translate_stencil_op(front_pass_z_pass) << + S5_STENCIL_PASS_Z_PASS_SHIFT); + + /* Set back state if different from front. */ + if (ctx->Stencil._TestTwoSide) { + i915->state.Ctx[I915_CTXREG_BF_STENCIL_OPS] &= + ~(BFO_STENCIL_REF_MASK | + BFO_STENCIL_TEST_MASK | + BFO_STENCIL_FAIL_MASK | + BFO_STENCIL_PASS_Z_FAIL_MASK | + BFO_STENCIL_PASS_Z_PASS_MASK); + i915->state.Ctx[I915_CTXREG_BF_STENCIL_OPS] |= BFO_STENCIL_TWO_SIDE | + (back_ref << BFO_STENCIL_REF_SHIFT) | + (intel_translate_compare_func(back_func) << BFO_STENCIL_TEST_SHIFT) | + (intel_translate_stencil_op(back_fail) << BFO_STENCIL_FAIL_SHIFT) | + (intel_translate_stencil_op(back_pass_z_fail) << + BFO_STENCIL_PASS_Z_FAIL_SHIFT) | + (intel_translate_stencil_op(back_pass_z_pass) << + BFO_STENCIL_PASS_Z_PASS_SHIFT); + + i915->state.Ctx[I915_CTXREG_BF_STENCIL_MASKS] &= + ~(BFM_STENCIL_TEST_MASK_MASK | + BFM_STENCIL_WRITE_MASK_MASK); + i915->state.Ctx[I915_CTXREG_BF_STENCIL_MASKS] |= + BFM_STENCIL_TEST_MASK(back_mask) | + BFM_STENCIL_WRITE_MASK(back_writemask); + } else { + i915->state.Ctx[I915_CTXREG_BF_STENCIL_OPS] &= ~BFO_STENCIL_TWO_SIDE; + } +} - i915->state.Ctx[I915_CTXREG_LIS5] |= ((ref << S5_STENCIL_REF_SHIFT) | - (test << - S5_STENCIL_TEST_FUNC_SHIFT)); +static void +i915StencilFuncSeparate(GLcontext * ctx, GLenum face, GLenum func, GLint ref, + GLuint mask) +{ } static void i915StencilMaskSeparate(GLcontext * ctx, GLenum face, GLuint mask) { - struct i915_context *i915 = I915_CONTEXT(ctx); - - DBG("%s : mask 0x%x\n", __FUNCTION__, mask); - - mask = mask & 0xff; - - I915_STATECHANGE(i915, I915_UPLOAD_CTX); - i915->state.Ctx[I915_CTXREG_STATE4] &= ~MODE4_ENABLE_STENCIL_WRITE_MASK; - i915->state.Ctx[I915_CTXREG_STATE4] |= (ENABLE_STENCIL_WRITE_MASK | - STENCIL_WRITE_MASK(mask)); } - static void i915StencilOpSeparate(GLcontext * ctx, GLenum face, GLenum fail, GLenum zfail, GLenum zpass) { - struct i915_context *i915 = I915_CONTEXT(ctx); - int fop = intel_translate_stencil_op(fail); - int dfop = intel_translate_stencil_op(zfail); - int dpop = intel_translate_stencil_op(zpass); - - - DBG("%s: fail : %s, zfail: %s, zpass : %s\n", __FUNCTION__, - _mesa_lookup_enum_by_nr(fail), - _mesa_lookup_enum_by_nr(zfail), _mesa_lookup_enum_by_nr(zpass)); - - I915_STATECHANGE(i915, I915_UPLOAD_CTX); - - i915->state.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_FAIL_MASK | - S5_STENCIL_PASS_Z_FAIL_MASK | - S5_STENCIL_PASS_Z_PASS_MASK); - - i915->state.Ctx[I915_CTXREG_LIS5] |= ((fop << S5_STENCIL_FAIL_SHIFT) | - (dfop << - S5_STENCIL_PASS_Z_FAIL_SHIFT) | - (dpop << - S5_STENCIL_PASS_Z_PASS_SHIFT)); } static void @@ -945,6 +991,17 @@ i915_init_packets(struct i915_context *i915) _3DSTATE_CONST_BLEND_COLOR_CMD; i915->state.Ctx[I915_CTXREG_BLENDCOLOR1] = 0; + i915->state.Ctx[I915_CTXREG_BF_STENCIL_MASKS] = + _3DSTATE_BACKFACE_STENCIL_MASKS | + BFM_ENABLE_STENCIL_TEST_MASK | + BFM_ENABLE_STENCIL_WRITE_MASK | + (0xff << BFM_STENCIL_WRITE_MASK_SHIFT) | + (0xff << BFM_STENCIL_TEST_MASK_SHIFT); + i915->state.Ctx[I915_CTXREG_BF_STENCIL_OPS] = + _3DSTATE_BACKFACE_STENCIL_OPS | + BFO_ENABLE_STENCIL_REF | + BFO_ENABLE_STENCIL_FUNCS | + BFO_ENABLE_STENCIL_TWO_SIDE; } { @@ -976,6 +1033,13 @@ i915_init_packets(struct i915_context *i915) i915->state.Buffer[I915_DESTREG_SR2] = 0; } + i915->state.RasterRules[I915_RASTER_RULES] = _3DSTATE_RASTER_RULES_CMD | + ENABLE_POINT_RASTER_RULE | + OGL_POINT_RASTER_RULE | + ENABLE_LINE_STRIP_PROVOKE_VRTX | + ENABLE_TRI_FAN_PROVOKE_VRTX | + LINE_STRIP_PROVOKE_VRTX(1) | + TRI_FAN_PROVOKE_VRTX(2) | ENABLE_TEXKILL_3D_4D | TEXKILL_4D; #if 0 { @@ -996,7 +1060,33 @@ i915_init_packets(struct i915_context *i915) i915->state.active = (I915_UPLOAD_PROGRAM | I915_UPLOAD_STIPPLE | I915_UPLOAD_CTX | - I915_UPLOAD_BUFFERS | I915_UPLOAD_INVARIENT); + I915_UPLOAD_BUFFERS | + I915_UPLOAD_INVARIENT | + I915_UPLOAD_RASTER_RULES); +} + +void +i915_update_provoking_vertex(GLcontext * ctx) +{ + struct i915_context *i915 = I915_CONTEXT(ctx); + + I915_STATECHANGE(i915, I915_UPLOAD_CTX); + i915->state.Ctx[I915_CTXREG_LIS6] &= ~(S6_TRISTRIP_PV_MASK); + + I915_STATECHANGE(i915, I915_UPLOAD_RASTER_RULES); + i915->state.RasterRules[I915_RASTER_RULES] &= ~(LINE_STRIP_PROVOKE_VRTX_MASK | + TRI_FAN_PROVOKE_VRTX_MASK); + + /* _NEW_LIGHT */ + if (ctx->Light.ProvokingVertex == GL_LAST_VERTEX_CONVENTION) { + i915->state.RasterRules[I915_RASTER_RULES] |= (LINE_STRIP_PROVOKE_VRTX(1) | + TRI_FAN_PROVOKE_VRTX(2)); + i915->state.Ctx[I915_CTXREG_LIS6] |= (2 << S6_TRISTRIP_PV_SHIFT); + } else { + i915->state.RasterRules[I915_RASTER_RULES] |= (LINE_STRIP_PROVOKE_VRTX(0) | + TRI_FAN_PROVOKE_VRTX(1)); + i915->state.Ctx[I915_CTXREG_LIS6] |= (0 << S6_TRISTRIP_PV_SHIFT); + } } void diff --git a/src/mesa/drivers/dri/i915/i915_vtbl.c b/src/mesa/drivers/dri/i915/i915_vtbl.c index fe1be93a6d..9a723d3cd7 100644 --- a/src/mesa/drivers/dri/i915/i915_vtbl.c +++ b/src/mesa/drivers/dri/i915/i915_vtbl.c @@ -176,7 +176,7 @@ i915_emit_invarient_state(struct intel_context *intel) { BATCH_LOCALS; - BEGIN_BATCH(20, IGNORE_CLIPRECTS); + BEGIN_BATCH(17, IGNORE_CLIPRECTS); OUT_BATCH(_3DSTATE_AA_CMD | AA_LINE_ECAAR_WIDTH_ENABLE | @@ -200,14 +200,6 @@ i915_emit_invarient_state(struct intel_context *intel) CSB_TCB(3, 3) | CSB_TCB(4, 4) | CSB_TCB(5, 5) | CSB_TCB(6, 6) | CSB_TCB(7, 7)); - OUT_BATCH(_3DSTATE_RASTER_RULES_CMD | - ENABLE_POINT_RASTER_RULE | - OGL_POINT_RASTER_RULE | - ENABLE_LINE_STRIP_PROVOKE_VRTX | - ENABLE_TRI_FAN_PROVOKE_VRTX | - LINE_STRIP_PROVOKE_VRTX(1) | - TRI_FAN_PROVOKE_VRTX(2) | ENABLE_TEXKILL_3D_4D | TEXKILL_4D); - /* Need to initialize this to zero. */ OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | I1_LOAD_S(3) | (0)); @@ -225,11 +217,6 @@ i915_emit_invarient_state(struct intel_context *intel) OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0); /* disable indirect state */ OUT_BATCH(0); - - /* Don't support twosided stencil yet */ - OUT_BATCH(_3DSTATE_BACKFACE_STENCIL_OPS | BFO_ENABLE_STENCIL_TWO_SIDE | 0); - OUT_BATCH(0); - ADVANCE_BATCH(); } @@ -263,6 +250,9 @@ get_state_size(struct i915_hw_state *state) if (dirty & I915_UPLOAD_INVARIENT) sz += 30 * 4; + if (dirty & I915_UPLOAD_RASTER_RULES) + sz += sizeof(state->RasterRules); + if (dirty & I915_UPLOAD_CTX) sz += sizeof(state->Ctx); @@ -371,6 +361,12 @@ i915_emit_state(struct intel_context *intel) i915_emit_invarient_state(intel); } + if (dirty & I915_UPLOAD_RASTER_RULES) { + if (INTEL_DEBUG & DEBUG_STATE) + fprintf(stderr, "I915_UPLOAD_RASTER_RULES:\n"); + emit(intel, state->RasterRules, sizeof(state->RasterRules)); + } + if (dirty & I915_UPLOAD_CTX) { if (INTEL_DEBUG & DEBUG_STATE) fprintf(stderr, "I915_UPLOAD_CTX:\n"); diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile index 00a42111da..128afb5686 100644 --- a/src/mesa/drivers/dri/i965/Makefile +++ b/src/mesa/drivers/dri/i965/Makefile @@ -43,6 +43,7 @@ DRIVER_SOURCES = \ brw_clip_util.c \ brw_context.c \ brw_curbe.c \ + brw_disasm.c \ brw_draw.c \ brw_draw_upload.c \ brw_eu.c \ diff --git a/src/mesa/drivers/dri/i965/brw_clip.c b/src/mesa/drivers/dri/i965/brw_clip.c index 54d30a3f42..20a927cf38 100644 --- a/src/mesa/drivers/dri/i965/brw_clip.c +++ b/src/mesa/drivers/dri/i965/brw_clip.c @@ -166,7 +166,8 @@ static void upload_clip_prog(struct brw_context *brw) /* _NEW_POLYGON */ if (key.primitive == GL_TRIANGLES) { - if (ctx->Polygon.CullFaceMode == GL_FRONT_AND_BACK) + if (ctx->Polygon.CullFlag && + ctx->Polygon.CullFaceMode == GL_FRONT_AND_BACK) key.clip_mode = BRW_CLIPMODE_REJECT_ALL; else { GLuint fill_front = CLIP_CULL; diff --git a/src/mesa/drivers/dri/i965/brw_clip.h b/src/mesa/drivers/dri/i965/brw_clip.h index 12e8548df1..957df441ab 100644 --- a/src/mesa/drivers/dri/i965/brw_clip.h +++ b/src/mesa/drivers/dri/i965/brw_clip.h @@ -100,6 +100,8 @@ struct brw_clip_compile { struct brw_reg fixed_planes; struct brw_reg plane_equation; + + struct brw_reg ff_sync; } reg; /* 3 different ways of expressing vertex size: @@ -173,4 +175,5 @@ struct brw_reg get_tmp( struct brw_clip_compile *c ); void brw_clip_project_position(struct brw_clip_compile *c, struct brw_reg pos ); void brw_clip_ff_sync(struct brw_clip_compile *c); +void brw_clip_init_ff_sync(struct brw_clip_compile *c); #endif diff --git a/src/mesa/drivers/dri/i965/brw_clip_line.c b/src/mesa/drivers/dri/i965/brw_clip_line.c index 9abd0642aa..048ca620fa 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_line.c +++ b/src/mesa/drivers/dri/i965/brw_clip_line.c @@ -85,6 +85,10 @@ static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) i++; } + if (c->need_ff_sync) { + c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD); + i++; + } c->first_tmp = i; c->last_tmp = i; @@ -246,8 +250,6 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) brw_ADD(p, c->reg.t, c->reg.t0, c->reg.t1); brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.t, brw_imm_f(1.0)); - if (c->need_ff_sync) - brw_clip_ff_sync(c); not_culled = brw_IF(p, BRW_EXECUTE_1); { brw_clip_interp_vertex(c, newvtx0, vtx0, vtx1, c->reg.t0, GL_FALSE); @@ -265,6 +267,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) void brw_emit_line_clip( struct brw_clip_compile *c ) { brw_clip_line_alloc_regs(c); + brw_clip_init_ff_sync(c); if (c->key.do_flat_shading) brw_clip_copy_colors(c, 0, 1); diff --git a/src/mesa/drivers/dri/i965/brw_clip_point.c b/src/mesa/drivers/dri/i965/brw_clip_point.c index 9738299168..8458f61c5a 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_point.c +++ b/src/mesa/drivers/dri/i965/brw_clip_point.c @@ -50,7 +50,7 @@ void brw_emit_point_clip( struct brw_clip_compile *c ) /* Send an empty message to kill the thread: */ brw_clip_tri_alloc_regs(c, 0); - if (c->need_ff_sync) - brw_clip_ff_sync(c); + brw_clip_init_ff_sync(c); + brw_clip_kill_thread(c); } diff --git a/src/mesa/drivers/dri/i965/brw_clip_tri.c b/src/mesa/drivers/dri/i965/brw_clip_tri.c index 4c2d655fb1..0efd77225e 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_tri.c +++ b/src/mesa/drivers/dri/i965/brw_clip_tri.c @@ -119,6 +119,11 @@ void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, i++; } + if (c->need_ff_sync) { + c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD); + i++; + } + c->first_tmp = i; c->last_tmp = i; @@ -563,6 +568,7 @@ void brw_emit_tri_clip( struct brw_clip_compile *c ) brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6); brw_clip_tri_init_vertices(c); brw_clip_init_clipmask(c); + brw_clip_init_ff_sync(c); /* if -ve rhw workaround bit is set, do cliptest */ @@ -589,8 +595,6 @@ void brw_emit_tri_clip( struct brw_clip_compile *c ) else maybe_do_clip_tri(c); - if (c->need_ff_sync) - brw_clip_ff_sync(c); brw_clip_tri_emit_polygon(c); /* Send an empty message to kill the thread: diff --git a/src/mesa/drivers/dri/i965/brw_clip_unfilled.c b/src/mesa/drivers/dri/i965/brw_clip_unfilled.c index 26950383c1..ad1bfa435f 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_unfilled.c +++ b/src/mesa/drivers/dri/i965/brw_clip_unfilled.c @@ -453,6 +453,7 @@ void brw_emit_unfilled_clip( struct brw_clip_compile *c ) brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6); brw_clip_tri_init_vertices(c); + brw_clip_init_ff_sync(c); assert(c->offset[VERT_RESULT_EDGE]); @@ -496,8 +497,6 @@ void brw_emit_unfilled_clip( struct brw_clip_compile *c ) } brw_ENDIF(p, do_clip); - if (c->need_ff_sync) - brw_clip_ff_sync(c); emit_unfilled_primitives(c); brw_clip_kill_thread(c); } diff --git a/src/mesa/drivers/dri/i965/brw_clip_util.c b/src/mesa/drivers/dri/i965/brw_clip_util.c index e09efc07ed..5a73abdfee 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_util.c +++ b/src/mesa/drivers/dri/i965/brw_clip_util.c @@ -213,6 +213,8 @@ void brw_clip_emit_vue(struct brw_clip_compile *c, struct brw_compile *p = &c->func; GLuint start = c->last_mrf; + brw_clip_ff_sync(c); + assert(!(allocate && eot)); /* Cycle through mrf regs - probably futile as we have to wait for @@ -263,6 +265,7 @@ void brw_clip_kill_thread(struct brw_clip_compile *c) { struct brw_compile *p = &c->func; + brw_clip_ff_sync(c); /* Send an empty message to kill the thread and release any * allocated urb entry: */ @@ -356,17 +359,38 @@ void brw_clip_init_clipmask( struct brw_clip_compile *c ) void brw_clip_ff_sync(struct brw_clip_compile *c) { + if (c->need_ff_sync) { + struct brw_compile *p = &c->func; + struct brw_instruction *need_ff_sync; + + brw_set_conditionalmod(p, BRW_CONDITIONAL_Z); + brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1)); + need_ff_sync = brw_IF(p, BRW_EXECUTE_1); + { + brw_OR(p, c->reg.ff_sync, c->reg.ff_sync, brw_imm_ud(0x1)); + brw_ff_sync(p, + c->reg.R0, + 0, + c->reg.R0, + 1, + 1, /* used */ + 1, /* msg length */ + 1, /* response length */ + 0, /* eot */ + 1, /* write compelete */ + 0, /* urb offset */ + BRW_URB_SWIZZLE_NONE); + } + brw_ENDIF(p, need_ff_sync); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + } +} + +void brw_clip_init_ff_sync(struct brw_clip_compile *c) +{ + if (c->need_ff_sync) { struct brw_compile *p = &c->func; - brw_ff_sync(p, - c->reg.R0, - 0, - c->reg.R0, - 1, - 1, /* used */ - 1, /* msg length */ - 1, /* response length */ - 0, /* eot */ - 1, /* write compelete */ - 0, /* urb offset */ - BRW_URB_SWIZZLE_NONE); + + brw_MOV(p, c->reg.ff_sync, brw_imm_ud(0)); + } } diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index 57ddf75413..847c44ed83 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -143,6 +143,7 @@ struct brw_context; #define BRW_NEW_DEPTH_BUFFER 0x20000 #define BRW_NEW_NR_WM_SURFACES 0x40000 #define BRW_NEW_NR_VS_SURFACES 0x80000 +#define BRW_NEW_INDEX_BUFFER 0x100000 struct brw_state_flags { /** State update flags signalled by mesa internals */ @@ -438,9 +439,13 @@ struct brw_query_object { unsigned int count; }; + +/** + * brw_context is derived from intel_context. + */ struct brw_context { - struct intel_context intel; + struct intel_context intel; /**< base class, must be first field */ GLuint primitive; GLboolean emit_state_always; @@ -475,6 +480,9 @@ struct brw_context struct { struct brw_vertex_element inputs[VERT_ATTRIB_MAX]; + struct brw_vertex_element *enabled[VERT_ATTRIB_MAX]; + GLuint nr_enabled; + #define BRW_NR_UPLOAD_BUFS 17 #define BRW_UPLOAD_INIT_SIZE (128*1024) @@ -498,8 +506,15 @@ struct brw_context */ const struct _mesa_index_buffer *ib; + /* Updates to these fields are signaled by BRW_NEW_INDEX_BUFFER. */ dri_bo *bo; unsigned int offset; + unsigned int size; + /* Offset to index buffer index to use in CMD_3D_PRIM so that we can + * avoid re-uploading the IB packet over and over if we're actually + * referencing the same index buffer. + */ + unsigned int start_vertex_offset; } ib; /* Active vertex program: @@ -706,6 +721,8 @@ void brw_upload_urb_fence(struct brw_context *brw); */ void brw_upload_cs_urb_state(struct brw_context *brw); +/* brw_disasm.c */ +int brw_disasm (FILE *file, struct brw_instruction *inst); /*====================================================================== * Inline conversion functions. These are better-typed than the diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h index d166250b4f..78d457ad2b 100644 --- a/src/mesa/drivers/dri/i965/brw_defines.h +++ b/src/mesa/drivers/dri/i965/brw_defines.h @@ -471,8 +471,9 @@ #define BRW_CONDITIONAL_GE 4 #define BRW_CONDITIONAL_L 5 #define BRW_CONDITIONAL_LE 6 -#define BRW_CONDITIONAL_C 7 +#define BRW_CONDITIONAL_R 7 #define BRW_CONDITIONAL_O 8 +#define BRW_CONDITIONAL_U 9 #define BRW_DEBUG_NONE 0 #define BRW_DEBUG_BREAKPOINT 1 @@ -512,6 +513,7 @@ #define BRW_OPCODE_RSL 11 #define BRW_OPCODE_ASR 12 #define BRW_OPCODE_CMP 16 +#define BRW_OPCODE_CMPN 17 #define BRW_OPCODE_JMPI 32 #define BRW_OPCODE_IF 34 #define BRW_OPCODE_IFF 35 diff --git a/src/mesa/drivers/dri/i965/brw_disasm.c b/src/mesa/drivers/dri/i965/brw_disasm.c new file mode 100644 index 0000000000..9fef230507 --- /dev/null +++ b/src/mesa/drivers/dri/i965/brw_disasm.c @@ -0,0 +1,903 @@ +/* + * Copyright © 2008 Keith Packard + * + * Permission to use, copy, modify, distribute, and sell this software and its + * documentation for any purpose is hereby granted without fee, provided that + * the above copyright notice appear in all copies and that both that copyright + * notice and this permission notice appear in supporting documentation, and + * that the name of the copyright holders not be used in advertising or + * publicity pertaining to distribution of the software without specific, + * written prior permission. The copyright holders make no representations + * about the suitability of this software for any purpose. It is provided "as + * is" without express or implied warranty. + * + * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, + * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO + * EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY SPECIAL, INDIRECT OR + * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, + * DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER + * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE + * OF THIS SOFTWARE. + */ + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <getopt.h> +#include <unistd.h> +#include <stdarg.h> + +#include "main/mtypes.h" + +#include "brw_context.h" +#include "brw_defines.h" + +struct { + char *name; + int nsrc; + int ndst; +} opcode[128] = { + [BRW_OPCODE_MOV] = { .name = "mov", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_FRC] = { .name = "frc", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_RNDU] = { .name = "rndu", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_RNDD] = { .name = "rndd", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_RNDE] = { .name = "rnde", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_RNDZ] = { .name = "rndz", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_NOT] = { .name = "not", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_LZD] = { .name = "lzd", .nsrc = 1, .ndst = 1 }, + + [BRW_OPCODE_MUL] = { .name = "mul", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_MAC] = { .name = "mac", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_MACH] = { .name = "mach", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_LINE] = { .name = "line", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SAD2] = { .name = "sad2", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SADA2] = { .name = "sada2", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_DP4] = { .name = "dp4", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_DPH] = { .name = "dph", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_DP3] = { .name = "dp3", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_DP2] = { .name = "dp2", .nsrc = 2, .ndst = 1 }, + + [BRW_OPCODE_AVG] = { .name = "avg", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_ADD] = { .name = "add", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SEL] = { .name = "sel", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_AND] = { .name = "and", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_OR] = { .name = "or", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_XOR] = { .name = "xor", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SHR] = { .name = "shr", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_SHL] = { .name = "shl", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_ASR] = { .name = "asr", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_CMP] = { .name = "cmp", .nsrc = 2, .ndst = 1 }, + [BRW_OPCODE_CMPN] = { .name = "cmpn", .nsrc = 2, .ndst = 1 }, + + [BRW_OPCODE_SEND] = { .name = "send", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_NOP] = { .name = "nop", .nsrc = 0, .ndst = 0 }, + [BRW_OPCODE_JMPI] = { .name = "jmpi", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_IF] = { .name = "if", .nsrc = 2, .ndst = 0 }, + [BRW_OPCODE_IFF] = { .name = "iff", .nsrc = 1, .ndst = 01 }, + [BRW_OPCODE_WHILE] = { .name = "while", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_ELSE] = { .name = "else", .nsrc = 2, .ndst = 0 }, + [BRW_OPCODE_BREAK] = { .name = "break", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_CONTINUE] = { .name = "cont", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_HALT] = { .name = "halt", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_MSAVE] = { .name = "msave", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_PUSH] = { .name = "push", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_MRESTORE] = { .name = "mrest", .nsrc = 1, .ndst = 1 }, + [BRW_OPCODE_POP] = { .name = "pop", .nsrc = 2, .ndst = 0 }, + [BRW_OPCODE_WAIT] = { .name = "wait", .nsrc = 1, .ndst = 0 }, + [BRW_OPCODE_DO] = { .name = "do", .nsrc = 0, .ndst = 0 }, + [BRW_OPCODE_ENDIF] = { .name = "endif", .nsrc = 2, .ndst = 0 }, +}; + +char *conditional_modifier[16] = { + [BRW_CONDITIONAL_NONE] = "", + [BRW_CONDITIONAL_Z] = ".e", + [BRW_CONDITIONAL_NZ] = ".ne", + [BRW_CONDITIONAL_G] = ".g", + [BRW_CONDITIONAL_GE] = ".ge", + [BRW_CONDITIONAL_L] = ".l", + [BRW_CONDITIONAL_LE] = ".le", + [BRW_CONDITIONAL_R] = ".r", + [BRW_CONDITIONAL_O] = ".o", + [BRW_CONDITIONAL_U] = ".u", +}; + +char *negate[2] = { + [0] = "", + [1] = "-", +}; + +char *_abs[2] = { + [0] = "", + [1] = "(abs)", +}; + +char *vert_stride[16] = { + [0] = "0", + [1] = "1", + [2] = "2", + [3] = "4", + [4] = "8", + [5] = "16", + [6] = "32", + [15] = "VxH", +}; + +char *width[8] = { + [0] = "1", + [1] = "2", + [2] = "4", + [3] = "8", + [4] = "16", +}; + +char *horiz_stride[4] = { + [0] = "0", + [1] = "1", + [2] = "2", + [3] = "4" +}; + +char *chan_sel[4] = { + [0] = "x", + [1] = "y", + [2] = "z", + [3] = "w", +}; + +char *dest_condmod[16] = { +}; + +char *debug_ctrl[2] = { + [0] = "", + [1] = ".breakpoint" +}; + +char *saturate[2] = { + [0] = "", + [1] = ".sat" +}; + +char *exec_size[8] = { + [0] = "1", + [1] = "2", + [2] = "4", + [3] = "8", + [4] = "16", + [5] = "32" +}; + +char *pred_inv[2] = { + [0] = "+", + [1] = "-" +}; + +char *pred_ctrl_align16[16] = { + [1] = "", + [2] = ".x", + [3] = ".y", + [4] = ".z", + [5] = ".w", + [6] = ".any4h", + [7] = ".all4h", +}; + +char *pred_ctrl_align1[16] = { + [1] = "", + [2] = ".anyv", + [3] = ".allv", + [4] = ".any2h", + [5] = ".all2h", + [6] = ".any4h", + [7] = ".all4h", + [8] = ".any8h", + [9] = ".all8h", + [10] = ".any16h", + [11] = ".all16h", +}; + +char *thread_ctrl[4] = { + [0] = "", + [2] = "switch" +}; + +char *compr_ctrl[4] = { + [0] = "", + [1] = "sechalf", + [2] = "compr", +}; + +char *dep_ctrl[4] = { + [0] = "", + [1] = "NoDDClr", + [2] = "NoDDChk", + [3] = "NoDDClr,NoDDChk", +}; + +char *mask_ctrl[4] = { + [0] = "", + [1] = "nomask", +}; + +char *access_mode[2] = { + [0] = "align1", + [1] = "align16", +}; + +char *reg_encoding[8] = { + [0] = "UD", + [1] = "D", + [2] = "UW", + [3] = "W", + [4] = "UB", + [5] = "B", + [7] = "F" +}; + +char *imm_encoding[8] = { + [0] = "UD", + [1] = "D", + [2] = "UW", + [3] = "W", + [5] = "VF", + [5] = "V", + [7] = "F" +}; + +char *reg_file[4] = { + [0] = "A", + [1] = "g", + [2] = "m", + [3] = "imm", +}; + +char *writemask[16] = { + [0x0] = ".", + [0x1] = ".x", + [0x2] = ".y", + [0x3] = ".xy", + [0x4] = ".z", + [0x5] = ".xz", + [0x6] = ".yz", + [0x7] = ".xyz", + [0x8] = ".w", + [0x9] = ".xw", + [0xa] = ".yw", + [0xb] = ".xyw", + [0xc] = ".zw", + [0xd] = ".xzw", + [0xe] = ".yzw", + [0xf] = "", +}; + +char *end_of_thread[2] = { + [0] = "", + [1] = "EOT" +}; + +char *target_function[16] = { + [BRW_MESSAGE_TARGET_NULL] = "null", + [BRW_MESSAGE_TARGET_MATH] = "math", + [BRW_MESSAGE_TARGET_SAMPLER] = "sampler", + [BRW_MESSAGE_TARGET_GATEWAY] = "gateway", + [BRW_MESSAGE_TARGET_DATAPORT_READ] = "read", + [BRW_MESSAGE_TARGET_DATAPORT_WRITE] = "write", + [BRW_MESSAGE_TARGET_URB] = "urb", + [BRW_MESSAGE_TARGET_THREAD_SPAWNER] = "thread_spawner" +}; + +char *math_function[16] = { + [BRW_MATH_FUNCTION_INV] = "inv", + [BRW_MATH_FUNCTION_LOG] = "log", + [BRW_MATH_FUNCTION_EXP] = "exp", + [BRW_MATH_FUNCTION_SQRT] = "sqrt", + [BRW_MATH_FUNCTION_RSQ] = "rsq", + [BRW_MATH_FUNCTION_SIN] = "sin", + [BRW_MATH_FUNCTION_COS] = "cos", + [BRW_MATH_FUNCTION_SINCOS] = "sincos", + [BRW_MATH_FUNCTION_TAN] = "tan", + [BRW_MATH_FUNCTION_POW] = "pow", + [BRW_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER] = "intdivmod", + [BRW_MATH_FUNCTION_INT_DIV_QUOTIENT] = "intmod", + [BRW_MATH_FUNCTION_INT_DIV_REMAINDER] = "intdiv", +}; + +char *math_saturate[2] = { + [0] = "", + [1] = "sat" +}; + +char *math_signed[2] = { + [0] = "", + [1] = "signed" +}; + +char *math_scalar[2] = { + [0] = "", + [1] = "scalar" +}; + +char *math_precision[2] = { + [0] = "", + [1] = "partial_precision" +}; + +char *urb_swizzle[4] = { + [BRW_URB_SWIZZLE_NONE] = "", + [BRW_URB_SWIZZLE_INTERLEAVE] = "interleave", + [BRW_URB_SWIZZLE_TRANSPOSE] = "transpose", +}; + +char *urb_allocate[2] = { + [0] = "", + [1] = "allocate" +}; + +char *urb_used[2] = { + [0] = "", + [1] = "used" +}; + +char *urb_complete[2] = { + [0] = "", + [1] = "complete" +}; + +char *sampler_target_format[4] = { + [0] = "F", + [2] = "UD", + [3] = "D" +}; + + +static int column; + +static int string (FILE *file, char *string) +{ + fputs (string, file); + column += strlen (string); + return 0; +} + +static int format (FILE *f, char *format, ...) +{ + char buf[1024]; + va_list args; + va_start (args, format); + + vsnprintf (buf, sizeof (buf) - 1, format, args); + string (f, buf); + return 0; +} + +static int newline (FILE *f) +{ + putc ('\n', f); + column = 0; + return 0; +} + +static int pad (FILE *f, int c) +{ + do + string (f, " "); + while (column < c); + return 0; +} + +static int control (FILE *file, char *name, char *ctrl[], GLuint id, int *space) +{ + if (!ctrl[id]) { + fprintf (file, "*** invalid %s value %d ", + name, id); + return 1; + } + if (ctrl[id][0]) + { + if (space && *space) + string (file, " "); + string (file, ctrl[id]); + if (space) + *space = 1; + } + return 0; +} + +static int print_opcode (FILE *file, int id) +{ + if (!opcode[id].name) { + format (file, "*** invalid opcode value %d ", id); + return 1; + } + string (file, opcode[id].name); + return 0; +} + +static int reg (FILE *file, GLuint _reg_file, GLuint _reg_nr) +{ + int err = 0; + if (_reg_file == BRW_ARCHITECTURE_REGISTER_FILE) { + switch (_reg_nr & 0xf0) { + case BRW_ARF_NULL: + string (file, "null"); + return -1; + case BRW_ARF_ADDRESS: + format (file, "a%d", _reg_nr & 0x0f); + break; + case BRW_ARF_ACCUMULATOR: + format (file, "acc%d", _reg_nr & 0x0f); + break; + case BRW_ARF_MASK: + format (file, "mask%d", _reg_nr & 0x0f); + break; + case BRW_ARF_MASK_STACK: + format (file, "msd%d", _reg_nr & 0x0f); + break; + case BRW_ARF_STATE: + format (file, "sr%d", _reg_nr & 0x0f); + break; + case BRW_ARF_CONTROL: + format (file, "cr%d", _reg_nr & 0x0f); + break; + case BRW_ARF_NOTIFICATION_COUNT: + format (file, "n%d", _reg_nr & 0x0f); + break; + case BRW_ARF_IP: + string (file, "ip"); + return -1; + break; + default: + format (file, "ARF%d", _reg_nr); + break; + } + } else { + err |= control (file, "src reg file", reg_file, _reg_file, NULL); + format (file, "%d", _reg_nr); + } + return err; +} + +static int dest (FILE *file, struct brw_instruction *inst) +{ + int err = 0; + + if (inst->header.access_mode == BRW_ALIGN_1) + { + if (inst->bits1.da1.dest_address_mode == BRW_ADDRESS_DIRECT) + { + err |= reg (file, inst->bits1.da1.dest_reg_file, inst->bits1.da1.dest_reg_nr); + if (err == -1) + return 0; + if (inst->bits1.da1.dest_subreg_nr) + format (file, ".%d", inst->bits1.da1.dest_subreg_nr); + format (file, "<%d>", inst->bits1.da1.dest_horiz_stride); + err |= control (file, "dest reg encoding", reg_encoding, inst->bits1.da1.dest_reg_type, NULL); + } + else + { + string (file, "g[a0"); + if (inst->bits1.ia1.dest_subreg_nr) + format (file, ".%d", inst->bits1.ia1.dest_subreg_nr); + if (inst->bits1.ia1.dest_indirect_offset) + format (file, " %d", inst->bits1.ia1.dest_indirect_offset); + string (file, "]"); + format (file, "<%d>", inst->bits1.ia1.dest_horiz_stride); + err |= control (file, "dest reg encoding", reg_encoding, inst->bits1.ia1.dest_reg_type, NULL); + } + } + else + { + if (inst->bits1.da16.dest_address_mode == BRW_ADDRESS_DIRECT) + { + err |= reg (file, inst->bits1.da16.dest_reg_file, inst->bits1.da16.dest_reg_nr); + if (err == -1) + return 0; + if (inst->bits1.da16.dest_subreg_nr) + format (file, ".%d", inst->bits1.da16.dest_subreg_nr); + string (file, "<1>"); + err |= control (file, "writemask", writemask, inst->bits1.da16.dest_writemask, NULL); + err |= control (file, "dest reg encoding", reg_encoding, inst->bits1.da16.dest_reg_type, NULL); + } + else + { + err = 1; + string (file, "Indirect align16 address mode not supported"); + } + } + + return 0; +} + +static int src_align1_region (FILE *file, + GLuint _vert_stride, GLuint _width, GLuint _horiz_stride) +{ + int err = 0; + string (file, "<"); + err |= control (file, "vert stride", vert_stride, _vert_stride, NULL); + string (file, ","); + err |= control (file, "width", width, _width, NULL); + string (file, ","); + err |= control (file, "horiz_stride", horiz_stride, _horiz_stride, NULL); + string (file, ">"); + return err; +} + +static int src_da1 (FILE *file, GLuint type, GLuint _reg_file, + GLuint _vert_stride, GLuint _width, GLuint _horiz_stride, + GLuint reg_num, GLuint sub_reg_num, GLuint __abs, GLuint _negate) +{ + int err = 0; + err |= control (file, "negate", negate, _negate, NULL); + err |= control (file, "abs", _abs, __abs, NULL); + + err |= reg (file, _reg_file, reg_num); + if (err == -1) + return 0; + if (sub_reg_num) + format (file, ".%d", sub_reg_num); + src_align1_region (file, _vert_stride, _width, _horiz_stride); + err |= control (file, "src reg encoding", reg_encoding, type, NULL); + return err; +} + +static int src_ia1 (FILE *file, + GLuint type, + GLuint _reg_file, + GLint _addr_imm, + GLuint _addr_subreg_nr, + GLuint _negate, + GLuint __abs, + GLuint _addr_mode, + GLuint _horiz_stride, + GLuint _width, + GLuint _vert_stride) +{ + int err = 0; + err |= control (file, "negate", negate, _negate, NULL); + err |= control (file, "abs", _abs, __abs, NULL); + + string (file, "g[a0"); + if (_addr_subreg_nr) + format (file, ".%d", _addr_subreg_nr); + if (_addr_imm) + format (file, " %d", _addr_imm); + string (file, "]"); + src_align1_region (file, _vert_stride, _width, _horiz_stride); + err |= control (file, "src reg encoding", reg_encoding, type, NULL); + return err; +} + +static int src_da16 (FILE *file, + GLuint _reg_type, + GLuint _reg_file, + GLuint _vert_stride, + GLuint _reg_nr, + GLuint _subreg_nr, + GLuint __abs, + GLuint _negate, + GLuint swz_x, + GLuint swz_y, + GLuint swz_z, + GLuint swz_w) +{ + int err = 0; + err |= control (file, "negate", negate, _negate, NULL); + err |= control (file, "abs", _abs, __abs, NULL); + + err |= reg (file, _reg_file, _reg_nr); + if (err == -1) + return 0; + if (_subreg_nr) + format (file, ".%d", _subreg_nr); + string (file, "<"); + err |= control (file, "vert stride", vert_stride, _vert_stride, NULL); + string (file, ",1,1>"); + err |= control (file, "src da16 reg type", reg_encoding, _reg_type, NULL); + /* + * Three kinds of swizzle display: + * identity - nothing printed + * 1->all - print the single channel + * 1->1 - print the mapping + */ + if (swz_x == BRW_CHANNEL_X && + swz_y == BRW_CHANNEL_Y && + swz_z == BRW_CHANNEL_Z && + swz_w == BRW_CHANNEL_W) + { + ; + } + else if (swz_x == swz_y && swz_x == swz_z && swz_x == swz_w) + { + string (file, "."); + err |= control (file, "channel select", chan_sel, swz_x, NULL); + } + else + { + string (file, "."); + err |= control (file, "channel select", chan_sel, swz_x, NULL); + err |= control (file, "channel select", chan_sel, swz_y, NULL); + err |= control (file, "channel select", chan_sel, swz_z, NULL); + err |= control (file, "channel select", chan_sel, swz_w, NULL); + } + return err; +} + + +static int imm (FILE *file, GLuint type, struct brw_instruction *inst) { + switch (type) { + case BRW_REGISTER_TYPE_UD: + format (file, "0x%08xUD", inst->bits3.ud); + break; + case BRW_REGISTER_TYPE_D: + format (file, "%dD", inst->bits3.d); + break; + case BRW_REGISTER_TYPE_UW: + format (file, "0x%04xUW", (uint16_t) inst->bits3.ud); + break; + case BRW_REGISTER_TYPE_W: + format (file, "%dW", (int16_t) inst->bits3.d); + break; + case BRW_REGISTER_TYPE_UB: + format (file, "0x%02xUB", (int8_t) inst->bits3.ud); + break; + case BRW_REGISTER_TYPE_VF: + format (file, "Vector Float"); + break; + case BRW_REGISTER_TYPE_V: + format (file, "0x%08xV", inst->bits3.ud); + break; + case BRW_REGISTER_TYPE_F: + format (file, "%-gF", inst->bits3.f); + } + return 0; +} + +static int src0 (FILE *file, struct brw_instruction *inst) +{ + if (inst->bits1.da1.src0_reg_file == BRW_IMMEDIATE_VALUE) + return imm (file, inst->bits1.da1.src0_reg_type, + inst); + else if (inst->header.access_mode == BRW_ALIGN_1) + { + if (inst->bits2.da1.src0_address_mode == BRW_ADDRESS_DIRECT) + { + return src_da1 (file, + inst->bits1.da1.src0_reg_type, + inst->bits1.da1.src0_reg_file, + inst->bits2.da1.src0_vert_stride, + inst->bits2.da1.src0_width, + inst->bits2.da1.src0_horiz_stride, + inst->bits2.da1.src0_reg_nr, + inst->bits2.da1.src0_subreg_nr, + inst->bits2.da1.src0_abs, + inst->bits2.da1.src0_negate); + } + else + { + return src_ia1 (file, + inst->bits1.ia1.src0_reg_type, + inst->bits1.ia1.src0_reg_file, + inst->bits2.ia1.src0_indirect_offset, + inst->bits2.ia1.src0_subreg_nr, + inst->bits2.ia1.src0_negate, + inst->bits2.ia1.src0_abs, + inst->bits2.ia1.src0_address_mode, + inst->bits2.ia1.src0_horiz_stride, + inst->bits2.ia1.src0_width, + inst->bits2.ia1.src0_vert_stride); + } + } + else + { + if (inst->bits2.da16.src0_address_mode == BRW_ADDRESS_DIRECT) + { + return src_da16 (file, + inst->bits1.da16.src0_reg_type, + inst->bits1.da16.src0_reg_file, + inst->bits2.da16.src0_vert_stride, + inst->bits2.da16.src0_reg_nr, + inst->bits2.da16.src0_subreg_nr, + inst->bits2.da16.src0_abs, + inst->bits2.da16.src0_negate, + inst->bits2.da16.src0_swz_x, + inst->bits2.da16.src0_swz_y, + inst->bits2.da16.src0_swz_z, + inst->bits2.da16.src0_swz_w); + } + else + { + string (file, "Indirect align16 address mode not supported"); + return 1; + } + } +} + +static int src1 (FILE *file, struct brw_instruction *inst) +{ + if (inst->bits1.da1.src1_reg_file == BRW_IMMEDIATE_VALUE) + return imm (file, inst->bits1.da1.src1_reg_type, + inst); + else if (inst->header.access_mode == BRW_ALIGN_1) + { + if (inst->bits3.da1.src1_address_mode == BRW_ADDRESS_DIRECT) + { + return src_da1 (file, + inst->bits1.da1.src1_reg_type, + inst->bits1.da1.src1_reg_file, + inst->bits3.da1.src1_vert_stride, + inst->bits3.da1.src1_width, + inst->bits3.da1.src1_horiz_stride, + inst->bits3.da1.src1_reg_nr, + inst->bits3.da1.src1_subreg_nr, + inst->bits3.da1.src1_abs, + inst->bits3.da1.src1_negate); + } + else + { + return src_ia1 (file, + inst->bits1.ia1.src1_reg_type, + inst->bits1.ia1.src1_reg_file, + inst->bits3.ia1.src1_indirect_offset, + inst->bits3.ia1.src1_subreg_nr, + inst->bits3.ia1.src1_negate, + inst->bits3.ia1.src1_abs, + inst->bits3.ia1.src1_address_mode, + inst->bits3.ia1.src1_horiz_stride, + inst->bits3.ia1.src1_width, + inst->bits3.ia1.src1_vert_stride); + } + } + else + { + if (inst->bits3.da16.src1_address_mode == BRW_ADDRESS_DIRECT) + { + return src_da16 (file, + inst->bits1.da16.src1_reg_type, + inst->bits1.da16.src1_reg_file, + inst->bits3.da16.src1_vert_stride, + inst->bits3.da16.src1_reg_nr, + inst->bits3.da16.src1_subreg_nr, + inst->bits3.da16.src1_abs, + inst->bits3.da16.src1_negate, + inst->bits3.da16.src1_swz_x, + inst->bits3.da16.src1_swz_y, + inst->bits3.da16.src1_swz_z, + inst->bits3.da16.src1_swz_w); + } + else + { + string (file, "Indirect align16 address mode not supported"); + return 1; + } + } +} + +int brw_disasm (FILE *file, struct brw_instruction *inst) +{ + int err = 0; + int space = 0; + + if (inst->header.predicate_control) { + string (file, "("); + err |= control (file, "predicate inverse", pred_inv, inst->header.predicate_inverse, NULL); + string (file, "f0"); + if (inst->bits2.da1.flag_reg_nr) + format (file, ".%d", inst->bits2.da1.flag_reg_nr); + if (inst->header.access_mode == BRW_ALIGN_1) + err |= control (file, "predicate control align1", pred_ctrl_align1, + inst->header.predicate_control, NULL); + else + err |= control (file, "predicate control align16", pred_ctrl_align16, + inst->header.predicate_control, NULL); + string (file, ") "); + } + + err |= print_opcode (file, inst->header.opcode); + err |= control (file, "saturate", saturate, inst->header.saturate, NULL); + err |= control (file, "debug control", debug_ctrl, inst->header.debug_control, NULL); + + if (inst->header.opcode != BRW_OPCODE_SEND) + err |= control (file, "conditional modifier", conditional_modifier, + inst->header.destreg__conditionalmod, NULL); + + if (inst->header.opcode != BRW_OPCODE_NOP) { + string (file, "("); + err |= control (file, "execution size", exec_size, inst->header.execution_size, NULL); + string (file, ")"); + } + + if (inst->header.opcode == BRW_OPCODE_SEND) + format (file, " %d", inst->header.destreg__conditionalmod); + + if (opcode[inst->header.opcode].ndst > 0) { + pad (file, 16); + err |= dest (file, inst); + } + if (opcode[inst->header.opcode].nsrc > 0) { + pad (file, 32); + err |= src0 (file, inst); + } + if (opcode[inst->header.opcode].nsrc > 1) { + pad (file, 48); + err |= src1 (file, inst); + } + + if (inst->header.opcode == BRW_OPCODE_SEND) { + newline (file); + pad (file, 16); + space = 0; + err |= control (file, "target function", target_function, + inst->bits3.generic.msg_target, &space); + switch (inst->bits3.generic.msg_target) { + case BRW_MESSAGE_TARGET_MATH: + err |= control (file, "math function", math_function, + inst->bits3.math.function, &space); + err |= control (file, "math saturate", math_saturate, + inst->bits3.math.saturate, &space); + err |= control (file, "math signed", math_signed, + inst->bits3.math.int_type, &space); + err |= control (file, "math scalar", math_scalar, + inst->bits3.math.data_type, &space); + err |= control (file, "math precision", math_precision, + inst->bits3.math.precision, &space); + break; + case BRW_MESSAGE_TARGET_SAMPLER: + format (file, " (%d, %d, ", + inst->bits3.sampler.binding_table_index, + inst->bits3.sampler.sampler); + err |= control (file, "sampler target format", sampler_target_format, + inst->bits3.sampler.return_format, NULL); + string (file, ")"); + break; + case BRW_MESSAGE_TARGET_DATAPORT_WRITE: + format (file, " (%d, %d, %d, %d)", + inst->bits3.dp_write.binding_table_index, + (inst->bits3.dp_write.pixel_scoreboard_clear << 3) | + inst->bits3.dp_write.msg_control, + inst->bits3.dp_write.msg_type, + inst->bits3.dp_write.send_commit_msg); + break; + case BRW_MESSAGE_TARGET_URB: + format (file, " %d", inst->bits3.urb.offset); + space = 1; + err |= control (file, "urb swizzle", urb_swizzle, + inst->bits3.urb.swizzle_control, &space); + err |= control (file, "urb allocate", urb_allocate, + inst->bits3.urb.allocate, &space); + err |= control (file, "urb used", urb_used, + inst->bits3.urb.used, &space); + err |= control (file, "urb complete", urb_complete, + inst->bits3.urb.complete, &space); + break; + case BRW_MESSAGE_TARGET_THREAD_SPAWNER: + break; + default: + format (file, "unsupported target %d", inst->bits3.generic.msg_target); + break; + } + if (space) + string (file, " "); + format (file, "mlen %d", + inst->bits3.generic.msg_length); + format (file, " rlen %d", + inst->bits3.generic.response_length); + } + pad (file, 64); + if (inst->header.opcode != BRW_OPCODE_NOP) { + string (file, "{"); + space = 1; + err |= control(file, "access mode", access_mode, inst->header.access_mode, &space); + err |= control (file, "mask control", mask_ctrl, inst->header.mask_control, &space); + err |= control (file, "dependency control", dep_ctrl, inst->header.dependency_control, &space); + err |= control (file, "compression control", compr_ctrl, inst->header.compression_control, &space); + err |= control (file, "thread control", thread_ctrl, inst->header.thread_control, &space); + if (inst->header.opcode == BRW_OPCODE_SEND) + err |= control (file, "end of thread", end_of_thread, + inst->bits3.generic.end_of_thread, &space); + if (space) + string (file, " "); + string (file, "}"); + } + string (file, ";"); + newline (file); + return err; +} diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c index 5152c3f3a5..682094ff13 100644 --- a/src/mesa/drivers/dri/i965/brw_draw.c +++ b/src/mesa/drivers/dri/i965/brw_draw.c @@ -141,6 +141,8 @@ static void brw_emit_prim(struct brw_context *brw, prim_packet.verts_per_instance = trim(prim->mode, prim->count); prim_packet.start_vert_location = prim->start; + if (prim->indexed) + prim_packet.start_vert_location += brw->ib.start_vertex_offset; prim_packet.instance_count = 1; prim_packet.start_instance_location = 0; prim_packet.base_vert_location = 0; @@ -422,54 +424,31 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx, return retval; } -static GLboolean brw_need_rebase( GLcontext *ctx, - const struct gl_client_array *arrays[], - const struct _mesa_index_buffer *ib, - GLuint min_index ) -{ - if (min_index == 0) - return GL_FALSE; - - if (ib) { - if (!vbo_all_varyings_in_vbos(arrays)) - return GL_TRUE; - else - return GL_FALSE; - } - else { - /* Hmm. This isn't quite what I wanted. BRW can actually - * handle the mixed case well enough that we shouldn't need to - * rebase. However, it's probably not very common, nor hugely - * expensive to do it this way: - */ - if (!vbo_all_varyings_in_vbos(arrays)) - return GL_TRUE; - else - return GL_FALSE; - } -} - - void brw_draw_prims( GLcontext *ctx, const struct gl_client_array *arrays[], const struct _mesa_prim *prim, GLuint nr_prims, const struct _mesa_index_buffer *ib, + GLboolean index_bounds_valid, GLuint min_index, GLuint max_index ) { GLboolean retval; - /* Decide if we want to rebase. If so we end up recursing once - * only into this function. - */ - if (brw_need_rebase( ctx, arrays, ib, min_index )) { - vbo_rebase_prims( ctx, arrays, - prim, nr_prims, - ib, min_index, max_index, - brw_draw_prims ); - - return; + if (!vbo_all_varyings_in_vbos(arrays)) { + if (!index_bounds_valid) + vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index); + + /* Decide if we want to rebase. If so we end up recursing once + * only into this function. + */ + if (min_index != 0) { + vbo_rebase_prims(ctx, arrays, + prim, nr_prims, + ib, min_index, max_index, + brw_draw_prims ); + return; + } } /* Make a first attempt at drawing: diff --git a/src/mesa/drivers/dri/i965/brw_draw.h b/src/mesa/drivers/dri/i965/brw_draw.h index 9aebbdb1b8..2a14db217f 100644 --- a/src/mesa/drivers/dri/i965/brw_draw.h +++ b/src/mesa/drivers/dri/i965/brw_draw.h @@ -39,6 +39,7 @@ void brw_draw_prims( GLcontext *ctx, const struct _mesa_prim *prims, GLuint nr_prims, const struct _mesa_index_buffer *ib, + GLboolean index_bounds_valid, GLuint min_index, GLuint max_index ); diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c index c29f1dd5c0..d49fb0fd95 100644 --- a/src/mesa/drivers/dri/i965/brw_draw_upload.c +++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c @@ -350,9 +350,6 @@ static void brw_prepare_vertices(struct brw_context *brw) unsigned int min_index = brw->vb.min_index; unsigned int max_index = brw->vb.max_index; - struct brw_vertex_element *enabled[VERT_ATTRIB_MAX]; - GLuint nr_enabled = 0; - struct brw_vertex_element *upload[VERT_ATTRIB_MAX]; GLuint nr_uploads = 0; @@ -362,12 +359,13 @@ static void brw_prepare_vertices(struct brw_context *brw) _mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index); /* Accumulate the list of enabled arrays. */ + brw->vb.nr_enabled = 0; while (vs_inputs) { GLuint i = _mesa_ffsll(vs_inputs) - 1; struct brw_vertex_element *input = &brw->vb.inputs[i]; vs_inputs &= ~(1 << i); - enabled[nr_enabled++] = input; + brw->vb.enabled[brw->vb.nr_enabled++] = input; } /* XXX: In the rare cases where this happens we fallback all @@ -376,16 +374,15 @@ static void brw_prepare_vertices(struct brw_context *brw) * cases with > 17 vertex attributes enabled, so it probably * isn't an issue at this point. */ - if (nr_enabled >= BRW_VEP_MAX) { + if (brw->vb.nr_enabled >= BRW_VEP_MAX) { intel->Fallback = 1; return; } - for (i = 0; i < nr_enabled; i++) { - struct brw_vertex_element *input = enabled[i]; + for (i = 0; i < brw->vb.nr_enabled; i++) { + struct brw_vertex_element *input = brw->vb.enabled[i]; input->element_size = get_size(input->glarray->Type) * input->glarray->Size; - input->count = input->glarray->StrideB ? max_index + 1 - min_index : 1; if (input->glarray->BufferObj->Name != 0) { struct intel_buffer_object *intel_buffer = @@ -398,7 +395,23 @@ static void brw_prepare_vertices(struct brw_context *brw) dri_bo_reference(input->bo); input->offset = (unsigned long)input->glarray->Ptr; input->stride = input->glarray->StrideB; + input->count = input->glarray->_MaxElement; + + /* This is a common place to reach if the user mistakenly supplies + * a pointer in place of a VBO offset. If we just let it go through, + * we may end up dereferencing a pointer beyond the bounds of the + * GTT. We would hope that the VBO's max_index would save us, but + * Mesa appears to hand us min/max values not clipped to the + * array object's _MaxElement, and _MaxElement frequently appears + * to be wrong anyway. + * + * The VBO spec allows application termination in this case, and it's + * probably a service to the poor programmer to do so rather than + * trying to just not render. + */ + assert(input->offset < input->bo->size); } else { + input->count = input->glarray->StrideB ? max_index + 1 - min_index : 1; if (input->bo != NULL) { /* Already-uploaded vertex data is present from a previous * prepare_vertices, but we had to re-validate state due to @@ -466,8 +479,8 @@ static void brw_prepare_vertices(struct brw_context *brw) brw_prepare_query_begin(brw); - for (i = 0; i < nr_enabled; i++) { - struct brw_vertex_element *input = enabled[i]; + for (i = 0; i < brw->vb.nr_enabled; i++) { + struct brw_vertex_element *input = brw->vb.enabled[i]; brw_add_validated_bo(brw, input->bo); } @@ -477,34 +490,44 @@ static void brw_emit_vertices(struct brw_context *brw) { GLcontext *ctx = &brw->intel.ctx; struct intel_context *intel = intel_context(ctx); - GLbitfield vs_inputs = brw->vs.prog_data->inputs_read; - struct brw_vertex_element *enabled[VERT_ATTRIB_MAX]; GLuint i; - GLuint nr_enabled = 0; - /* Accumulate the list of enabled arrays. */ - while (vs_inputs) { - i = _mesa_ffsll(vs_inputs) - 1; - struct brw_vertex_element *input = &brw->vb.inputs[i]; + brw_emit_query_begin(brw); - vs_inputs &= ~(1 << i); - enabled[nr_enabled++] = input; + /* If the VS doesn't read any inputs (calculating vertex position from + * a state variable for some reason, for example), emit a single pad + * VERTEX_ELEMENT struct and bail. + * + * The stale VB state stays in place, but they don't do anything unless + * a VE loads from them. + */ + if (brw->vb.nr_enabled == 0) { + BEGIN_BATCH(3, IGNORE_CLIPRECTS); + OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1); + OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) | + BRW_VE0_VALID | + (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) | + (0 << BRW_VE0_SRC_OFFSET_SHIFT)); + OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) | + (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) | + (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) | + (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT)); + ADVANCE_BATCH(); + return; } - brw_emit_query_begin(brw); - /* Now emit VB and VEP state packets. * * This still defines a hardware VB for each input, even if they * are interleaved or from the same VBO. TBD if this makes a * performance difference. */ - BEGIN_BATCH(1 + nr_enabled * 4, IGNORE_CLIPRECTS); + BEGIN_BATCH(1 + brw->vb.nr_enabled * 4, IGNORE_CLIPRECTS); OUT_BATCH((CMD_VERTEX_BUFFER << 16) | - ((1 + nr_enabled * 4) - 2)); + ((1 + brw->vb.nr_enabled * 4) - 2)); - for (i = 0; i < nr_enabled; i++) { - struct brw_vertex_element *input = enabled[i]; + for (i = 0; i < brw->vb.nr_enabled; i++) { + struct brw_vertex_element *input = brw->vb.enabled[i]; OUT_BATCH((i << BRW_VB0_INDEX_SHIFT) | BRW_VB0_ACCESS_VERTEXDATA | @@ -524,15 +547,15 @@ static void brw_emit_vertices(struct brw_context *brw) input->offset + input->element_size); } } else - OUT_BATCH(brw->vb.max_index); + OUT_BATCH(input->stride ? input->count : 0); OUT_BATCH(0); /* Instance data step rate */ } ADVANCE_BATCH(); - BEGIN_BATCH(1 + nr_enabled * 2, IGNORE_CLIPRECTS); - OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + nr_enabled * 2) - 2)); - for (i = 0; i < nr_enabled; i++) { - struct brw_vertex_element *input = enabled[i]; + BEGIN_BATCH(1 + brw->vb.nr_enabled * 2, IGNORE_CLIPRECTS); + OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + brw->vb.nr_enabled * 2) - 2)); + for (i = 0; i < brw->vb.nr_enabled; i++) { + struct brw_vertex_element *input = brw->vb.enabled[i]; uint32_t format = get_surface_type(input->glarray->Type, input->glarray->Size, input->glarray->Format, @@ -589,17 +612,20 @@ static void brw_prepare_indices(struct brw_context *brw) dri_bo *bo = NULL; struct gl_buffer_object *bufferobj; GLuint offset; + GLuint ib_type_size; if (index_buffer == NULL) return; - ib_size = get_size(index_buffer->type) * index_buffer->count; + ib_type_size = get_size(index_buffer->type); + ib_size = ib_type_size * index_buffer->count; bufferobj = index_buffer->obj;; /* Turn into a proper VBO: */ if (!bufferobj->Name) { - + brw->ib.start_vertex_offset = 0; + /* Get new bufferobj, offset: */ get_space(brw, ib_size, &bo, &offset); @@ -615,6 +641,7 @@ static void brw_prepare_indices(struct brw_context *brw) } } else { offset = (GLuint) (unsigned long) index_buffer->ptr; + brw->ib.start_vertex_offset = 0; /* If the index buffer isn't aligned to its element size, we have to * rebase it into a temporary. @@ -635,39 +662,62 @@ static void brw_prepare_indices(struct brw_context *brw) bo = intel_bufferobj_buffer(intel, intel_buffer_object(bufferobj), INTEL_READ); dri_bo_reference(bo); + + /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading + * the index buffer state when we're just moving the start index + * of our drawing. + */ + brw->ib.start_vertex_offset = offset / ib_type_size; + offset = 0; + ib_size = bo->size; } } - dri_bo_unreference(brw->ib.bo); - brw->ib.bo = bo; - brw->ib.offset = offset; + if (brw->ib.bo != bo || + brw->ib.offset != offset || + brw->ib.size != ib_size) + { + drm_intel_bo_unreference(brw->ib.bo); + brw->ib.bo = bo; + brw->ib.offset = offset; + brw->ib.size = ib_size; + + brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER; + } else { + drm_intel_bo_unreference(bo); + } brw_add_validated_bo(brw, brw->ib.bo); } -static void brw_emit_indices(struct brw_context *brw) +const struct brw_tracked_state brw_indices = { + .dirty = { + .mesa = 0, + .brw = BRW_NEW_INDICES, + .cache = 0, + }, + .prepare = brw_prepare_indices, +}; + +static void brw_emit_index_buffer(struct brw_context *brw) { struct intel_context *intel = &brw->intel; const struct _mesa_index_buffer *index_buffer = brw->ib.ib; - GLuint ib_size; if (index_buffer == NULL) return; - ib_size = get_size(index_buffer->type) * index_buffer->count - 1; - /* Emit the indexbuffer packet: */ { struct brw_indexbuffer ib; memset(&ib, 0, sizeof(ib)); - + ib.header.bits.opcode = CMD_INDEX_BUFFER; ib.header.bits.length = sizeof(ib)/4 - 2; ib.header.bits.index_format = get_index_type(index_buffer->type); ib.header.bits.cut_index_enable = 0; - BEGIN_BATCH(4, IGNORE_CLIPRECTS); OUT_BATCH( ib.header.dword ); @@ -676,18 +726,17 @@ static void brw_emit_indices(struct brw_context *brw) brw->ib.offset); OUT_RELOC(brw->ib.bo, I915_GEM_DOMAIN_VERTEX, 0, - brw->ib.offset + ib_size); + brw->ib.offset + brw->ib.size); OUT_BATCH( 0 ); ADVANCE_BATCH(); } } -const struct brw_tracked_state brw_indices = { +const struct brw_tracked_state brw_index_buffer = { .dirty = { .mesa = 0, - .brw = BRW_NEW_BATCH | BRW_NEW_INDICES, + .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER, .cache = 0, }, - .prepare = brw_prepare_indices, - .emit = brw_emit_indices, + .emit = brw_emit_index_buffer, }; diff --git a/src/mesa/drivers/dri/i965/brw_eu.c b/src/mesa/drivers/dri/i965/brw_eu.c index c53efba599..1df561386e 100644 --- a/src/mesa/drivers/dri/i965/brw_eu.c +++ b/src/mesa/drivers/dri/i965/brw_eu.c @@ -62,7 +62,7 @@ void brw_set_predicate_control( struct brw_compile *p, GLuint pc ) void brw_set_conditionalmod( struct brw_compile *p, GLuint conditional ) { - p->current->header.destreg__conditonalmod = conditional; + p->current->header.destreg__conditionalmod = conditional; } void brw_set_access_mode( struct brw_compile *p, GLuint access_mode ) diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c index 2412014248..241cdc33f8 100644 --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c @@ -481,8 +481,8 @@ static struct brw_instruction *next_insn( struct brw_compile *p, /* Reset this one-shot flag: */ - if (p->current->header.destreg__conditonalmod) { - p->current->header.destreg__conditonalmod = 0; + if (p->current->header.destreg__conditionalmod) { + p->current->header.destreg__conditionalmod = 0; p->current->header.predicate_control = BRW_PREDICATE_NORMAL; } @@ -679,7 +679,7 @@ struct brw_instruction *brw_ELSE(struct brw_compile *p, assert(if_insn->header.opcode == BRW_OPCODE_IF); if_insn->bits3.if_else.jump_count = br * (insn - if_insn); - if_insn->bits3.if_else.pop_count = 1; + if_insn->bits3.if_else.pop_count = 0; if_insn->bits3.if_else.pad0 = 0; } @@ -871,7 +871,7 @@ void brw_CMP(struct brw_compile *p, { struct brw_instruction *insn = next_insn(p, BRW_OPCODE_CMP); - insn->header.destreg__conditonalmod = conditional; + insn->header.destreg__conditionalmod = conditional; brw_set_dest(insn, dest); brw_set_src0(insn, src0); brw_set_src1(insn, src1); @@ -915,7 +915,7 @@ void brw_math( struct brw_compile *p, * instructions. */ insn->header.predicate_control = 0; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src); @@ -952,7 +952,7 @@ void brw_math_16( struct brw_compile *p, brw_set_compression_control(p, BRW_COMPRESSION_NONE); insn = next_insn(p, BRW_OPCODE_SEND); - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src); @@ -969,7 +969,7 @@ void brw_math_16( struct brw_compile *p, */ insn = next_insn(p, BRW_OPCODE_SEND); insn->header.compression_control = BRW_COMPRESSION_2NDHALF; - insn->header.destreg__conditonalmod = msg_reg_nr+1; + insn->header.destreg__conditionalmod = msg_reg_nr+1; brw_set_dest(insn, offset(dest,1)); brw_set_src0(insn, src); @@ -1016,7 +1016,7 @@ void brw_dp_WRITE_16( struct brw_compile *p, insn->header.predicate_control = 0; /* XXX */ insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src); @@ -1062,7 +1062,7 @@ void brw_dp_READ_16( struct brw_compile *p, insn->header.predicate_control = 0; /* XXX */ insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); /* UW? */ brw_set_src0(insn, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW)); @@ -1116,7 +1116,7 @@ void brw_dp_READ_4( struct brw_compile *p, insn->header.predicate_control = BRW_PREDICATE_NONE; insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; insn->header.mask_control = BRW_MASK_DISABLE; /* cast dest to a uword[8] vector */ @@ -1190,7 +1190,7 @@ void brw_dp_READ_4_vs(struct brw_compile *p, insn->header.predicate_control = BRW_PREDICATE_NONE; insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; insn->header.mask_control = BRW_MASK_DISABLE; /*insn->header.access_mode = BRW_ALIGN_16;*/ @@ -1224,7 +1224,7 @@ void brw_fb_WRITE(struct brw_compile *p, insn->header.predicate_control = 0; /* XXX */ insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src0); @@ -1322,7 +1322,7 @@ void brw_SAMPLE(struct brw_compile *p, insn->header.predicate_control = 0; /* XXX */ insn->header.compression_control = BRW_COMPRESSION_NONE; - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_dest(insn, dest); brw_set_src0(insn, src0); @@ -1375,7 +1375,7 @@ void brw_urb_WRITE(struct brw_compile *p, brw_set_src0(insn, src0); brw_set_src1(insn, brw_imm_d(0)); - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_urb_message(p->brw, insn, @@ -1410,7 +1410,7 @@ void brw_ff_sync(struct brw_compile *p, brw_set_src0(insn, src0); brw_set_src1(insn, brw_imm_d(0)); - insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.destreg__conditionalmod = msg_reg_nr; brw_set_ff_sync_message(p->brw, insn, diff --git a/src/mesa/drivers/dri/i965/brw_gs_emit.c b/src/mesa/drivers/dri/i965/brw_gs_emit.c index 980eac7646..a9b2aa2eac 100644 --- a/src/mesa/drivers/dri/i965/brw_gs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_gs_emit.c @@ -101,7 +101,7 @@ static void brw_gs_emit_vue(struct brw_gs_compile *c, BRW_URB_SWIZZLE_NONE); } -void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim) +static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim) { struct brw_compile *p = &c->func; brw_MOV(p, get_element_ud(c->reg.R0, 1), brw_imm_ud(num_prim)); diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c index b5f6371c82..bc0f076073 100644 --- a/src/mesa/drivers/dri/i965/brw_sf_state.c +++ b/src/mesa/drivers/dri/i965/brw_sf_state.c @@ -113,7 +113,7 @@ struct brw_sf_unit_key { unsigned int nr_urb_entries, urb_size, sfsize; - GLenum front_face, cull_face; + GLenum front_face, cull_face, provoking_vertex; unsigned scissor:1; unsigned line_smooth:1; unsigned point_sprite:1; @@ -153,6 +153,9 @@ sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key) key->point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize); key->point_attenuated = ctx->Point._Attenuated; + /* _NEW_LIGHT */ + key->provoking_vertex = ctx->Light.ProvokingVertex; + key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0; } @@ -284,9 +287,15 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons: */ - sf.sf7.trifan_pv = 2; - sf.sf7.linestrip_pv = 1; - sf.sf7.tristrip_pv = 2; + if (key->provoking_vertex == GL_LAST_VERTEX_CONVENTION) { + sf.sf7.trifan_pv = 2; + sf.sf7.linestrip_pv = 1; + sf.sf7.tristrip_pv = 2; + } else { + sf.sf7.trifan_pv = 1; + sf.sf7.linestrip_pv = 0; + sf.sf7.tristrip_pv = 0; + } sf.sf7.line_last_pixel_enable = 0; /* Set bias for OpenGL rasterization rules: @@ -300,6 +309,9 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, &sf, sizeof(sf), NULL, NULL); + /* STATE_PREFETCH command description describes this state as being + * something loaded through the GPE (L2 ISC), so it's INSTRUCTION domain. + */ /* Emit SF program relocation */ dri_bo_emit_reloc(bo, I915_GEM_DOMAIN_INSTRUCTION, 0, @@ -340,6 +352,7 @@ static void upload_sf_unit( struct brw_context *brw ) const struct brw_tracked_state brw_sf_unit = { .dirty = { .mesa = (_NEW_POLYGON | + _NEW_LIGHT | _NEW_LINE | _NEW_POINT | _NEW_SCISSOR | diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h index bf9f6cae55..78572356a3 100644 --- a/src/mesa/drivers/dri/i965/brw_state.h +++ b/src/mesa/drivers/dri/i965/brw_state.h @@ -92,6 +92,7 @@ const struct brw_tracked_state brw_clear_batch_cache; const struct brw_tracked_state brw_drawing_rect; const struct brw_tracked_state brw_indices; const struct brw_tracked_state brw_vertices; +const struct brw_tracked_state brw_index_buffer; /** * Use same key for WM and VS surfaces. diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c index 38d9dd8991..95d42d2dcc 100644 --- a/src/mesa/drivers/dri/i965/brw_state_upload.c +++ b/src/mesa/drivers/dri/i965/brw_state_upload.c @@ -94,6 +94,7 @@ const struct brw_tracked_state *atoms[] = &brw_drawing_rect, &brw_indices, + &brw_index_buffer, &brw_vertices, &brw_constant_buffer @@ -208,6 +209,7 @@ static struct dirty_bit_map brw_bits[] = { DEFINE_BIT(BRW_NEW_PSP), DEFINE_BIT(BRW_NEW_FENCE), DEFINE_BIT(BRW_NEW_INDICES), + DEFINE_BIT(BRW_NEW_INDEX_BUFFER), DEFINE_BIT(BRW_NEW_VERTICES), DEFINE_BIT(BRW_NEW_BATCH), DEFINE_BIT(BRW_NEW_DEPTH_BUFFER), diff --git a/src/mesa/drivers/dri/i965/brw_structs.h b/src/mesa/drivers/dri/i965/brw_structs.h index 8ba7eb27b3..a6de09207b 100644 --- a/src/mesa/drivers/dri/i965/brw_structs.h +++ b/src/mesa/drivers/dri/i965/brw_structs.h @@ -1200,7 +1200,7 @@ struct brw_instruction GLuint predicate_control:4; GLuint predicate_inverse:1; GLuint execution_size:3; - GLuint destreg__conditonalmod:4; /* destreg - send, conditionalmod - others */ + GLuint destreg__conditionalmod:4; /* destreg - send, conditionalmod - others */ GLuint pad0:2; GLuint debug_control:1; GLuint saturate:1; @@ -1228,7 +1228,9 @@ struct brw_instruction GLuint dest_reg_type:3; GLuint src0_reg_file:2; GLuint src0_reg_type:3; - GLuint pad:6; + GLuint src1_reg_file:2; /* 0x00000c00 */ + GLuint src1_reg_type:3; /* 0x00007000 */ + GLuint pad:1; GLint dest_indirect_offset:10; /* offset against the deref'd address reg */ GLuint dest_subreg_nr:3; /* subnr for the address reg a0.x */ GLuint dest_horiz_stride:2; @@ -1243,7 +1245,7 @@ struct brw_instruction GLuint src0_reg_type:3; GLuint src1_reg_file:2; GLuint src1_reg_type:3; - GLuint pad0:1; + GLuint pad:1; GLuint dest_writemask:4; GLuint dest_subreg_nr:1; GLuint dest_reg_nr:8; @@ -1348,7 +1350,7 @@ struct brw_instruction GLuint src1_reg_nr:8; GLuint src1_abs:1; GLuint src1_negate:1; - GLuint pad:1; + GLuint src1_address_mode:1; GLuint src1_horiz_stride:2; GLuint src1_width:3; GLuint src1_vert_stride:4; @@ -1363,7 +1365,7 @@ struct brw_instruction GLuint src1_reg_nr:8; GLuint src1_abs:1; GLuint src1_negate:1; - GLuint pad0:1; + GLuint src1_address_mode:1; GLuint src1_swz_z:2; GLuint src1_swz_w:2; GLuint pad1:1; @@ -1377,7 +1379,7 @@ struct brw_instruction GLuint src1_subreg_nr:3; GLuint src1_abs:1; GLuint src1_negate:1; - GLuint pad0:1; + GLuint src1_address_mode:1; GLuint src1_horiz_stride:2; GLuint src1_width:3; GLuint src1_vert_stride:4; @@ -1565,6 +1567,7 @@ struct brw_instruction GLint d; GLuint ud; + float f; } bits3; }; diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c index 7f9b253534..1d2e953eb1 100644 --- a/src/mesa/drivers/dri/i965/brw_tex_layout.c +++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c @@ -65,11 +65,6 @@ GLboolean brw_miptree_layout(struct intel_context *intel, if (mt->compressed) { mt->pitch = ALIGN(mt->width0, align_w); - qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * mt->pitch * mt->cpp; - mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * 6; - } else { - qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * mt->pitch * mt->cpp; - mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * 6; } if (mt->first_level != mt->last_level) { @@ -90,6 +85,14 @@ GLboolean brw_miptree_layout(struct intel_context *intel, mt->pitch = intel_miptree_pitch_align(intel, mt, tiling, mt->pitch); + if (mt->compressed) { + qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * mt->pitch * mt->cpp; + mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) / 4 * 6; + } else { + qpitch = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * mt->pitch * mt->cpp; + mt->total_height = (y_pitch + ALIGN(minify(y_pitch), align_h) + 11 * align_h) * 6; + } + for (level = mt->first_level; level <= mt->last_level; level++) { GLuint img_height; GLuint nr_images = 6; diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c index 514f15d5e3..83167b9258 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c @@ -68,6 +68,7 @@ static void release_tmps( struct brw_vs_compile *c ) static void brw_vs_alloc_regs( struct brw_vs_compile *c ) { GLuint i, reg = 0, mrf; + int attributes_in_vue; /* Determine whether to use a real constant buffer or use a block * of GRF registers for constants. The later is faster but only @@ -128,6 +129,11 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) reg++; } } + /* If there are no inputs, we'll still be reading one attribute's worth + * because it's required -- see urb_read_length setting. + */ + if (c->nr_inputs == 0) + reg++; /* Allocate outputs. The non-position outputs go straight into message regs. */ @@ -220,11 +226,22 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) * vertex urb, so is half the amount: */ c->prog_data.urb_read_length = (c->nr_inputs + 1) / 2; + /* Setting this field to 0 leads to undefined behavior according to the + * the VS_STATE docs. Our VUEs will always have at least one attribute + * sitting in them, even if it's padding. + */ + if (c->prog_data.urb_read_length == 0) + c->prog_data.urb_read_length = 1; + + /* The VS VUEs are shared by VF (outputting our inputs) and VS, so size + * them to fit the biggest thing they need to. + */ + attributes_in_vue = MAX2(c->nr_outputs, c->nr_inputs); if (BRW_IS_IGDNG(c->func.brw)) - c->prog_data.urb_entry_size = (c->nr_outputs + 6 + 3) / 4; + c->prog_data.urb_entry_size = (attributes_in_vue + 6 + 3) / 4; else - c->prog_data.urb_entry_size = (c->nr_outputs + 2 + 3) / 4; + c->prog_data.urb_entry_size = (attributes_in_vue + 2 + 3) / 4; c->prog_data.total_grf = reg; @@ -1245,9 +1262,30 @@ post_vs_emit( struct brw_vs_compile *c, /* patch up the END code to jump past subroutines, etc */ offset = last_inst - end_inst; - brw_set_src1(end_inst, brw_imm_d(offset * 16)); + if (offset > 1) { + brw_set_src1(end_inst, brw_imm_d(offset * 16)); + } else { + end_inst->header.opcode = BRW_OPCODE_NOP; + } } +static uint32_t +get_predicate(uint32_t swizzle) +{ + switch (swizzle) { + case SWIZZLE_XXXX: + return BRW_PREDICATE_ALIGN16_REPLICATE_X; + case SWIZZLE_YYYY: + return BRW_PREDICATE_ALIGN16_REPLICATE_Y; + case SWIZZLE_ZZZZ: + return BRW_PREDICATE_ALIGN16_REPLICATE_Z; + case SWIZZLE_WWWW: + return BRW_PREDICATE_ALIGN16_REPLICATE_W; + default: + _mesa_problem(NULL, "Unexpected predicate: 0x%08x\n", swizzle); + return BRW_PREDICATE_NORMAL; + } +} /* Emit the vertex program instructions here. */ @@ -1266,7 +1304,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) GLuint file; if (INTEL_DEBUG & DEBUG_VS) { - _mesa_printf("vs-emit:\n"); + _mesa_printf("vs-mesa:\n"); _mesa_print_program(&c->vp->program.Base); _mesa_printf("\n"); } @@ -1453,7 +1491,10 @@ void brw_vs_emit(struct brw_vs_compile *c ) break; case OPCODE_IF: assert(if_depth < MAX_IF_DEPTH); - if_inst[if_depth++] = brw_IF(p, BRW_EXECUTE_8); + if_inst[if_depth] = brw_IF(p, BRW_EXECUTE_8); + if_inst[if_depth]->header.predicate_control = + get_predicate(inst->DstReg.CondSwizzle); + if_depth++; break; case OPCODE_ELSE: if_inst[if_depth-1] = brw_ELSE(p, if_inst[if_depth-1]); @@ -1541,6 +1582,19 @@ void brw_vs_emit(struct brw_vs_compile *c ) "unknown"); } + /* Set the predication update on the last instruction of the native + * instruction sequence. + * + * This would be problematic if it was set on a math instruction, + * but that shouldn't be the case with the current GLSL compiler. + */ + if (inst->CondUpdate) { + struct brw_instruction *hw_insn = &p->store[p->nr_insn - 1]; + + assert(hw_insn->header.destreg__conditionalmod == 0); + hw_insn->header.destreg__conditionalmod = BRW_CONDITIONAL_NZ; + } + if ((inst->DstReg.File == PROGRAM_OUTPUT) && (inst->DstReg.Index != VERT_RESULT_HPOS) && c->output_regs[inst->DstReg.Index].used_in_src) { @@ -1578,4 +1632,13 @@ void brw_vs_emit(struct brw_vs_compile *c ) emit_vertex_write(c); post_vs_emit(c, end_inst, last_inst); + + if (INTEL_DEBUG & DEBUG_VS) { + int i; + + _mesa_printf("vs-native:\n"); + for (i = 0; i < p->nr_insn; i++) + brw_disasm(stderr, &p->store[i]); + _mesa_printf("\n"); + } } diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c index 14e05be4f6..2292de94c4 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.c +++ b/src/mesa/drivers/dri/i965/brw_wm.c @@ -41,13 +41,13 @@ GLuint brw_wm_nr_args( GLuint opcode ) { switch (opcode) { case WM_FRONTFACING: - return 0; case WM_PIXELXY: + return 0; case WM_CINTERP: case WM_WPOSXY: + case WM_DELTAXY: return 1; case WM_LINTERP: - case WM_DELTAXY: case WM_PIXELW: return 2; case WM_FB_WRITE: @@ -171,9 +171,11 @@ static void do_wm_prog( struct brw_context *brw, * differently from "simple" shaders. */ if (fp->isGLSL) { + c->dispatch_width = 8; brw_wm_glsl_emit(brw, c); } else { + c->dispatch_width = 16; brw_wm_non_glsl_emit(brw, c); } @@ -202,6 +204,7 @@ static void brw_wm_populate_key( struct brw_context *brw, /* BRW_NEW_FRAGMENT_PROGRAM */ const struct brw_fragment_program *fp = (struct brw_fragment_program *)brw->fragment_program; + GLboolean uses_depth = (fp->program.Base.InputsRead & (1 << FRAG_ATTRIB_WPOS)) != 0; GLuint lookup = 0; GLuint line_aa; GLuint i; @@ -263,6 +266,7 @@ static void brw_wm_populate_key( struct brw_context *brw, brw_wm_lookup_iz(line_aa, lookup, + uses_depth, key); diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h index ba497432c6..ae98b5492d 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.h +++ b/src/mesa/drivers/dri/i965/brw_wm.h @@ -260,6 +260,7 @@ struct brw_wm_compile { GLuint tmp_index; GLuint tmp_max; GLuint subroutines[BRW_WM_MAX_SUBROUTINE]; + GLuint dispatch_width; /** we may need up to 3 constants per instruction (if use_const_buffer) */ struct { @@ -292,6 +293,7 @@ void brw_wm_print_program( struct brw_wm_compile *c, void brw_wm_lookup_iz( GLuint line_aa, GLuint lookup, + GLboolean ps_uses_depth, struct brw_wm_prog_key *key ); GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp); diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c index 9f82916c02..b3cf524c63 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_emit.c +++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c @@ -65,8 +65,7 @@ static INLINE struct brw_reg sechalf( struct brw_reg reg ) static void emit_pixel_xy(struct brw_compile *p, const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0) + GLuint mask) { struct brw_reg r1 = brw_vec1_grf(1, 0); struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW); @@ -98,8 +97,7 @@ static void emit_pixel_xy(struct brw_compile *p, static void emit_delta_xy(struct brw_compile *p, const struct brw_reg *dst, GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *arg1) + const struct brw_reg *arg0) { struct brw_reg r1 = brw_vec1_grf(1, 0); @@ -545,16 +543,18 @@ static void emit_dp3( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1 ) { + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert((mask & WRITEMASK_XYZW) == WRITEMASK_X); + assert(is_power_of_two(mask & WRITEMASK_XYZW)); brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]); brw_MAC(p, brw_null_reg(), arg0[1], arg1[1]); brw_set_saturate(p, (mask & SATURATE) ? 1 : 0); - brw_MAC(p, dst[0], arg0[2], arg1[2]); + brw_MAC(p, dst[dst_chan], arg0[2], arg1[2]); brw_set_saturate(p, 0); } @@ -565,17 +565,19 @@ static void emit_dp4( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1 ) { + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert((mask & WRITEMASK_XYZW) == WRITEMASK_X); + assert(is_power_of_two(mask & WRITEMASK_XYZW)); brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]); brw_MAC(p, brw_null_reg(), arg0[1], arg1[1]); brw_MAC(p, brw_null_reg(), arg0[2], arg1[2]); brw_set_saturate(p, (mask & SATURATE) ? 1 : 0); - brw_MAC(p, dst[0], arg0[3], arg1[3]); + brw_MAC(p, dst[dst_chan], arg0[3], arg1[3]); brw_set_saturate(p, 0); } @@ -632,18 +634,19 @@ static void emit_math1( struct brw_compile *p, GLuint mask, const struct brw_reg *arg0 ) { + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ - //assert((mask & WRITEMASK_XYZW) == WRITEMASK_X || - // function == BRW_MATH_FUNCTION_SINCOS); - + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + brw_MOV(p, brw_message_reg(2), arg0[0]); /* Send two messages to perform all 16 operations: */ brw_math_16(p, - dst[0], + dst[dst_chan], function, (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, 2, @@ -659,10 +662,12 @@ static void emit_math2( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1) { + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + if (!(mask & WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert((mask & WRITEMASK_XYZW) == WRITEMASK_X); + assert(is_power_of_two(mask & WRITEMASK_XYZW)); brw_push_insn_state(p); @@ -681,7 +686,7 @@ static void emit_math2( struct brw_compile *p, */ brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_math(p, - dst[0], + dst[dst_chan], function, (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, 2, @@ -691,7 +696,7 @@ static void emit_math2( struct brw_compile *p, brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF); brw_math(p, - offset(dst[0],1), + offset(dst[dst_chan],1), function, (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, 4, @@ -1194,11 +1199,11 @@ void brw_wm_emit( struct brw_wm_compile *c ) /* Generated instructions for calculating triangle interpolants: */ case WM_PIXELXY: - emit_pixel_xy(p, dst, dst_flags, args[0]); + emit_pixel_xy(p, dst, dst_flags); break; case WM_DELTAXY: - emit_delta_xy(p, dst, dst_flags, args[0], args[1]); + emit_delta_xy(p, dst, dst_flags, args[0]); break; case WM_WPOSXY: @@ -1385,4 +1390,13 @@ void brw_wm_emit( struct brw_wm_compile *c ) inst->dst[i]->hw_reg, inst->dst[i]->spill_slot); } + + if (INTEL_DEBUG & DEBUG_WM) { + int i; + + _mesa_printf("wm-native:\n"); + for (i = 0; i < p->nr_insn; i++) + brw_disasm(stderr, &p->store[i]); + _mesa_printf("\n"); + } } diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c index b9e8dd2e96..5dc076a825 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_fp.c +++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c @@ -226,9 +226,42 @@ static struct prog_instruction * emit_op(struct brw_wm_compile *c, 0, 0, 0, /* tex unit, target, shadow */ src0, src1, src2); } - +/* Many Mesa opcodes produce the same value across all the result channels. + * We'd rather not have to support that splatting in the opcode implementations, + * and brw_wm_pass*.c wants to optimize them out by shuffling references around + * anyway. We can easily get both by emitting the opcode to one channel, and + * then MOVing it to the others, which brw_wm_pass*.c already understands. + */ +static struct prog_instruction *emit_scalar_insn(struct brw_wm_compile *c, + const struct prog_instruction *inst0) +{ + struct prog_instruction *inst; + unsigned int dst_chan; + unsigned int other_channel_mask; + + if (inst0->DstReg.WriteMask == 0) + return NULL; + + dst_chan = _mesa_ffs(inst0->DstReg.WriteMask) - 1; + inst = get_fp_inst(c); + *inst = *inst0; + inst->DstReg.WriteMask = 1 << dst_chan; + + other_channel_mask = inst0->DstReg.WriteMask & ~(1 << dst_chan); + if (other_channel_mask != 0) { + inst = emit_op(c, + OPCODE_MOV, + dst_mask(inst0->DstReg, other_channel_mask), + 0, + src_swizzle1(src_reg_from_dst(inst0->DstReg), dst_chan), + src_undef(), + src_undef()); + } + return inst; +} + /*********************************************************************** * Special instructions for interpolation and other tasks @@ -376,14 +409,6 @@ static void emit_interp( struct brw_wm_compile *c, } break; case FRAG_ATTRIB_FOGC: - /* The FOGC input is really special. When a program uses glFogFragCoord, - * the results returned are supposed to be (f,0,0,1). But for Mesa GLSL, - * the glFrontFacing and glPointCoord values are also stashed in FOGC. - * So, write the interpolated fog value to X, then either 0, 1, or the - * stashed values to Y, Z, W. Note that this means that - * glFogFragCoord.yzw can be wrong in those cases! - */ - /* Interpolate the fog coordinate */ emit_op(c, WM_PINTERP, @@ -393,26 +418,40 @@ static void emit_interp( struct brw_wm_compile *c, deltas, get_pixel_w(c)); - /* Move the front facing value into FOGC.y if it's needed. */ - if (c->fp->program.UsesFrontFacing) { - emit_op(c, - WM_FRONTFACING, - dst_mask(dst, WRITEMASK_Y), - 0, - src_undef(), - src_undef(), - src_undef()); - } else { - emit_op(c, - OPCODE_MOV, - dst_mask(dst, WRITEMASK_Y), - 0, - src_swizzle1(interp, SWIZZLE_ZERO), - src_undef(), - src_undef()); - } + emit_op(c, + OPCODE_MOV, + dst_mask(dst, WRITEMASK_YZW), + 0, + src_swizzle(interp, + SWIZZLE_ZERO, + SWIZZLE_ZERO, + SWIZZLE_ZERO, + SWIZZLE_ONE), + src_undef(), + src_undef()); + break; + + case FRAG_ATTRIB_FACE: + /* XXX review/test this case */ + emit_op(c, + WM_FRONTFACING, + dst_mask(dst, WRITEMASK_X), + 0, + src_undef(), + src_undef(), + src_undef()); + break; + + case FRAG_ATTRIB_PNTC: + /* XXX review/test this case */ + emit_op(c, + WM_PINTERP, + dst_mask(dst, WRITEMASK_XY), + 0, + interp, + deltas, + get_pixel_w(c)); - /* Should do the PointCoord thing here. */ emit_op(c, OPCODE_MOV, dst_mask(dst, WRITEMASK_ZW), @@ -425,6 +464,7 @@ static void emit_interp( struct brw_wm_compile *c, src_undef(), src_undef()); break; + default: emit_op(c, WM_PINTERP, @@ -683,7 +723,7 @@ static void precalc_tex( struct brw_wm_compile *c, /* tmp0 = 1 / tmp1 */ emit_op(c, OPCODE_RCP, - tmp0, + dst_mask(tmp0, WRITEMASK_X), 0, tmp1src, src_undef(), @@ -694,7 +734,7 @@ static void precalc_tex( struct brw_wm_compile *c, tmpcoord, 0, src0, - tmp0src, + src_swizzle1(tmp0src, SWIZZLE_X), src_undef()); release_temp(c, tmp0); @@ -717,7 +757,11 @@ static void precalc_tex( struct brw_wm_compile *c, tmpcoord, 0, inst->SrcReg[0], - scale, + src_swizzle(scale, + SWIZZLE_X, + SWIZZLE_Y, + SWIZZLE_ONE, + SWIZZLE_ONE), src_undef()); coord = src_reg_from_dst(tmpcoord); @@ -1134,9 +1178,11 @@ void brw_wm_pass_fp( struct brw_wm_compile *c ) break; case OPCODE_PRINT: break; - default: - emit_insn(c, inst); + if (brw_wm_is_scalar_result(inst->Opcode)) + emit_scalar_insn(c, inst); + else + emit_insn(c, inst); break; } } diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index 19f777fe32..a5b18ec7d7 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -10,6 +10,9 @@ enum _subroutine { SUB_NOISE1, SUB_NOISE2, SUB_NOISE3, SUB_NOISE4 }; +static struct brw_reg get_dst_reg(struct brw_wm_compile *c, + const struct prog_instruction *inst, + GLuint component); /** * Determine if the given fragment program uses GLSL features such @@ -22,6 +25,7 @@ GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp) for (i = 0; i < fp->Base.NumInstructions; i++) { const struct prog_instruction *inst = &fp->Base.Instructions[i]; switch (inst->Opcode) { + case OPCODE_ARL: case OPCODE_IF: case OPCODE_ENDIF: case OPCODE_CAL: @@ -130,19 +134,6 @@ static void set_reg(struct brw_wm_compile *c, int file, int index, c->wm_regs[file][index][component].inited = GL_TRUE; } -/** - * Examine instruction's write mask to find index of first component - * enabled for writing. - */ -static int get_scalar_dst_index(const struct prog_instruction *inst) -{ - int i; - for (i = 0; i < 4; i++) - if (inst->DstReg.WriteMask & (1<<i)) - break; - return i; -} - static struct brw_reg alloc_tmp(struct brw_wm_compile *c) { struct brw_reg reg; @@ -402,6 +393,27 @@ static void prealloc_reg(struct brw_wm_compile *c) prealloc_grf(c, 126); prealloc_grf(c, 127); + for (i = 0; i < c->nr_fp_insns; i++) { + const struct prog_instruction *inst = &c->prog_instructions[i]; + struct brw_reg dst[4]; + + switch (inst->Opcode) { + case OPCODE_TEX: + case OPCODE_TXB: + /* Allocate the channels of texture results contiguously, + * since they are written out that way by the sampler unit. + */ + for (j = 0; j < 4; j++) { + dst[j] = get_dst_reg(c, inst, j); + if (j != 0) + assert(dst[j].nr == dst[j - 1].nr + 1); + } + break; + default: + break; + } + } + /* An instruction may reference up to three constants. * They'll be found in these registers. * XXX alloc these on demand! @@ -639,23 +651,6 @@ static void invoke_subroutine( struct brw_wm_compile *c, } } -static void emit_abs( struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - int i; - struct brw_compile *p = &c->func; - brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF); - for (i = 0; i < 4; i++) { - if (inst->DstReg.WriteMask & (1<<i)) { - struct brw_reg src, dst; - dst = get_dst_reg(c, inst, i); - src = get_src_reg(c, inst, 0, i); - brw_MOV(p, dst, brw_abs(src)); - } - } - brw_set_saturate(p, 0); -} - static void emit_trunc( struct brw_wm_compile *c, const struct prog_instruction *inst) { @@ -1031,12 +1026,20 @@ static void emit_dp3(struct brw_wm_compile *c, struct brw_reg src0[3], src1[3], dst; int i; struct brw_compile *p = &c->func; + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; + + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + for (i = 0; i < 3; i++) { src0[i] = get_src_reg(c, inst, 0, i); src1[i] = get_src_reg_imm(c, inst, 1, i); } - dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + dst = get_dst_reg(c, inst, dst_chan); brw_MUL(p, brw_null_reg(), src0[0], src1[0]); brw_MAC(p, brw_null_reg(), src0[1], src1[1]); brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); @@ -1050,11 +1053,19 @@ static void emit_dp4(struct brw_wm_compile *c, struct brw_reg src0[4], src1[4], dst; int i; struct brw_compile *p = &c->func; + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; + + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + for (i = 0; i < 4; i++) { src0[i] = get_src_reg(c, inst, 0, i); src1[i] = get_src_reg_imm(c, inst, 1, i); } - dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + dst = get_dst_reg(c, inst, dst_chan); brw_MUL(p, brw_null_reg(), src0[0], src1[0]); brw_MAC(p, brw_null_reg(), src0[1], src1[1]); brw_MAC(p, brw_null_reg(), src0[2], src1[2]); @@ -1069,11 +1080,19 @@ static void emit_dph(struct brw_wm_compile *c, struct brw_reg src0[4], src1[4], dst; int i; struct brw_compile *p = &c->func; + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; + + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + for (i = 0; i < 4; i++) { src0[i] = get_src_reg(c, inst, 0, i); src1[i] = get_src_reg_imm(c, inst, 1, i); } - dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + dst = get_dst_reg(c, inst, dst_chan); brw_MUL(p, brw_null_reg(), src0[0], src1[0]); brw_MAC(p, brw_null_reg(), src0[1], src1[1]); brw_MAC(p, dst, src0[2], src1[2]); @@ -1091,37 +1110,28 @@ static void emit_math1(struct brw_wm_compile *c, const struct prog_instruction *inst, GLuint func) { struct brw_compile *p = &c->func; - struct brw_reg src0, dst, tmp; - const int mark = mark_tmps( c ); - int i; + struct brw_reg src0, dst; + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; - tmp = alloc_tmp(c); + assert(is_power_of_two(mask & WRITEMASK_XYZW)); /* Get first component of source register */ + dst = get_dst_reg(c, inst, dst_chan); src0 = get_src_reg(c, inst, 0, 0); - /* tmp = func(src0) */ brw_MOV(p, brw_message_reg(2), src0); brw_math(p, - tmp, + dst, func, (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE, 2, brw_null_reg(), BRW_MATH_DATA_VECTOR, BRW_MATH_PRECISION_FULL); - - /*tmp.dw1.bits.swizzle = SWIZZLE_XXXX;*/ - - /* replicate tmp value across enabled dest channels */ - for (i = 0; i < 4; i++) { - if (inst->DstReg.WriteMask & (1 << i)) { - dst = get_dst_reg(c, inst, i); - brw_MOV(p, dst, tmp); - } - } - - release_tmps(c, mark); } static void emit_rcp(struct brw_wm_compile *c, @@ -1192,24 +1202,6 @@ static void emit_arl(struct brw_wm_compile *c, brw_set_saturate(p, 0); } -static void emit_sub(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - struct brw_reg src0, src1, dst; - GLuint mask = inst->DstReg.WriteMask; - int i; - brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0); - for (i = 0 ; i < 4; i++) { - if (mask & (1<<i)) { - dst = get_dst_reg(c, inst, i); - src0 = get_src_reg(c, inst, 0, i); - src1 = get_src_reg_imm(c, inst, 1, i); - brw_ADD(p, dst, src0, negate(src1)); - } - } - brw_set_saturate(p, 0); -} static void emit_mul(struct brw_wm_compile *c, const struct prog_instruction *inst) @@ -1321,7 +1313,15 @@ static void emit_pow(struct brw_wm_compile *c, { struct brw_compile *p = &c->func; struct brw_reg dst, src0, src1; - dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + GLuint mask = inst->DstReg.WriteMask; + int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1; + + if (!(mask & WRITEMASK_XYZW)) + return; + + assert(is_power_of_two(mask & WRITEMASK_XYZW)); + + dst = get_dst_reg(c, inst, dst_chan); src0 = get_src_reg_imm(c, inst, 0, 0); src1 = get_src_reg_imm(c, inst, 1, 0); @@ -2828,18 +2828,12 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) case WM_FRONTFACING: emit_frontfacing(c, inst); break; - case OPCODE_ABS: - emit_abs(c, inst); - break; case OPCODE_ADD: emit_add(c, inst); break; case OPCODE_ARL: emit_arl(c, inst); break; - case OPCODE_SUB: - emit_sub(c, inst); - break; case OPCODE_FRC: emit_frc(c, inst); break; @@ -3007,7 +3001,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) loop_depth--; inst0 = inst1 = brw_WHILE(p, loop_inst[loop_depth]); - /* patch all the BREAK/CONT instructions from last BEGINLOOP */ + /* patch all the BREAK/CONT instructions from last BGNLOOP */ while (inst0 > loop_inst[loop_depth]) { inst0--; if (inst0->header.opcode == BRW_OPCODE_BREAK) { @@ -3032,8 +3026,14 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) brw_set_predicate_control(p, BRW_PREDICATE_NONE); } post_wm_emit(c); -} + if (INTEL_DEBUG & DEBUG_WM) { + _mesa_printf("wm-native:\n"); + for (i = 0; i < p->nr_insn; i++) + brw_disasm(stderr, &p->store[i]); + _mesa_printf("\n"); + } +} /** * Do GPU code generation for shaders that use GLSL features such as diff --git a/src/mesa/drivers/dri/i965/brw_wm_iz.c b/src/mesa/drivers/dri/i965/brw_wm_iz.c index 8fd067abe7..5e399ac62a 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_iz.c +++ b/src/mesa/drivers/dri/i965/brw_wm_iz.c @@ -122,6 +122,7 @@ const struct { */ void brw_wm_lookup_iz( GLuint line_aa, GLuint lookup, + GLboolean ps_uses_depth, struct brw_wm_prog_key *key ) { GLuint reg = 2; @@ -131,7 +132,7 @@ void brw_wm_lookup_iz( GLuint line_aa, if (lookup & IZ_PS_COMPUTES_DEPTH_BIT) key->computes_depth = 1; - if (wm_iz_table[lookup].sd_present) { + if (wm_iz_table[lookup].sd_present || ps_uses_depth) { key->source_depth_reg = reg; reg += 2; } diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass0.c b/src/mesa/drivers/dri/i965/brw_wm_pass0.c index 92142764f5..6279258339 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass0.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass0.c @@ -257,34 +257,6 @@ static void pass0_set_dst( struct brw_wm_compile *c, } -static void pass0_set_dst_scalar( struct brw_wm_compile *c, - struct brw_wm_instruction *out, - const struct prog_instruction *inst, - GLuint writemask ) -{ - if (writemask) { - const struct prog_dst_register *dst = &inst->DstReg; - GLuint i; - - /* Compute only the first (X) value: - */ - out->writemask = WRITEMASK_X; - out->dst[0] = get_value(c); - - /* Update our tracking register file for all the components in - * writemask: - */ - for (i = 0; i < 4; i++) { - if (writemask & (1<<i)) { - pass0_set_fpreg_value(c, dst->File, dst->Index, i, out->dst[0]); - } - } - } - else - out->writemask = 0; -} - - static const struct brw_wm_ref *get_fp_src_reg_ref( struct brw_wm_compile *c, struct prog_src_register src, GLuint i ) @@ -363,10 +335,7 @@ translate_insn(struct brw_wm_compile *c, /* Dst: */ - if (brw_wm_is_scalar_result(out->opcode)) - pass0_set_dst_scalar(c, out, inst, writemask); - else - pass0_set_dst(c, out, inst, writemask); + pass0_set_dst(c, out, inst, writemask); } diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c index 3fc18ff1f3..dff466587a 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c @@ -103,6 +103,10 @@ struct wm_sampler_key { GLenum minfilter, magfilter; GLenum comparemode, comparefunc; dri_bo *sdc_bo; + + /** If target is cubemap, take context setting. + */ + GLboolean seamless_cube_map; } sampler[BRW_MAX_TEX_UNIT]; }; @@ -169,30 +173,33 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key, } } - if (key->tex_target == GL_TEXTURE_CUBE_MAP && - (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) { - /* If we're using anything but nearest sampling for a cube map, we - * need to set this wrap mode to avoid GPU lock-ups. - */ - sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE; - sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE; - sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE; - } - else if (key->tex_target == GL_TEXTURE_1D) { + sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); + sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); + sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t); + + /* Cube-maps on 965 and later must use the same wrap mode for all 3 + * coordinate dimensions. Futher, only CUBE and CLAMP are valid. + */ + if (key->tex_target == GL_TEXTURE_CUBE_MAP) { + if (key->seamless_cube_map && + (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) { + sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE; + sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE; + sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE; + } else { + sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP; + sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP; + sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP; + } + } else if (key->tex_target == GL_TEXTURE_1D) { /* There's a bug in 1D texture sampling - it actually pays * attention to the wrap_t value, though it should not. * Override the wrap_t value here to GL_REPEAT to keep * any nonexistent border pixels from floating in. */ - sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); - sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP; } - else { - sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); - sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); - sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t); - } + /* Set shadow function: */ @@ -249,6 +256,9 @@ brw_wm_sampler_populate_key(struct brw_context *brw, entry->tex_target = texObj->Target; + entry->seamless_cube_map = (texObj->Target == GL_TEXTURE_CUBE_MAP) + ? ctx->Texture.CubeMapSeamless : GL_FALSE; + entry->wrap_r = texObj->WrapR; entry->wrap_s = texObj->WrapS; entry->wrap_t = texObj->WrapT; diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.c b/src/mesa/drivers/dri/intel/intel_batchbuffer.c index 0f87fc46a4..6aa36d10b1 100644 --- a/src/mesa/drivers/dri/intel/intel_batchbuffer.c +++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.c @@ -196,6 +196,16 @@ _intel_batchbuffer_flush(struct intel_batchbuffer *batch, const char *file, struct intel_context *intel = batch->intel; GLuint used = batch->ptr - batch->map; + if (intel->first_post_swapbuffers_batch == NULL) { + intel->first_post_swapbuffers_batch = intel->batch->buf; + drm_intel_bo_reference(intel->first_post_swapbuffers_batch); + } + + if (intel->first_post_swapbuffers_batch == NULL) { + intel->first_post_swapbuffers_batch = intel->batch->buf; + drm_intel_bo_reference(intel->first_post_swapbuffers_batch); + } + if (used == 0) { batch->cliprect_mode = IGNORE_CLIPRECTS; return; diff --git a/src/mesa/drivers/dri/intel/intel_blit.c b/src/mesa/drivers/dri/intel/intel_blit.c index 2e95bd1013..979f202584 100644 --- a/src/mesa/drivers/dri/intel/intel_blit.c +++ b/src/mesa/drivers/dri/intel/intel_blit.c @@ -477,6 +477,8 @@ intelClearWithBlit(GLcontext *ctx, GLbitfield mask) BR13 |= BR13_565; } + assert(irb->region->tiling != I915_TILING_Y); + #ifndef I915 if (irb->region->tiling != I915_TILING_NONE) { CMD |= XY_DST_TILED; @@ -571,6 +573,7 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel, assert( logic_op - GL_CLEAR >= 0 ); assert( logic_op - GL_CLEAR < 0x10 ); + assert(dst_pitch > 0); if (w < 0 || h < 0) return GL_TRUE; diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c index 4abb525f78..46f1a7f720 100644 --- a/src/mesa/drivers/dri/intel/intel_context.c +++ b/src/mesa/drivers/dri/intel/intel_context.c @@ -38,6 +38,7 @@ #include "swrast_setup/swrast_setup.h" #include "tnl/tnl.h" #include "drivers/common/driverfuncs.h" +#include "drivers/common/meta.h" #include "i830_dri.h" @@ -513,7 +514,7 @@ intel_flush(GLcontext *ctx, GLboolean needs_mi_flush) * each of N places that do rendering. This has worse performances, * but it is much easier to get correct. */ - if (intel->is_front_buffer_rendering) { + if (!intel->is_front_buffer_rendering) { intel->front_buffer_dirty = GL_FALSE; } } @@ -529,7 +530,27 @@ intelFlush(GLcontext * ctx) static void intel_glFlush(GLcontext *ctx) { + struct intel_context *intel = intel_context(ctx); + intel_flush(ctx, GL_TRUE); + + /* We're using glFlush as an indicator that a frame is done, which is + * what DRI2 does before calling SwapBuffers (and means we should catch + * people doing front-buffer rendering, as well).. + * + * Wait for the swapbuffers before the one we just emitted, so we don't + * get too many swaps outstanding for apps that are GPU-heavy but not + * CPU-heavy. + * + * Unfortunately, we don't have a handle to the batch containing the swap, + * and getting our hands on that doesn't seem worth it, so we just us the + * first batch we emitted after the last swap. + */ + if (intel->first_post_swapbuffers_batch != NULL) { + drm_intel_bo_wait_rendering(intel->first_post_swapbuffers_batch); + drm_intel_bo_unreference(intel->first_post_swapbuffers_batch); + intel->first_post_swapbuffers_batch = NULL; + } } void @@ -692,6 +713,8 @@ intelInitContext(struct intel_context *intel, _swrast_allow_pixel_fog(ctx, GL_FALSE); _swrast_allow_vertex_fog(ctx, GL_TRUE); + _mesa_meta_init(ctx); + intel->hw_stencil = mesaVis->stencilBits && mesaVis->depthBits == 24; intel->hw_stipple = 1; @@ -752,7 +775,7 @@ intelInitContext(struct intel_context *intel, if (intel->use_texture_tiling && !intel->intelScreen->kernel_exec_fencing) { fprintf(stderr, "No kernel support for execution fencing, " - "disabling texture tiling"); + "disabling texture tiling\n"); intel->use_texture_tiling = GL_FALSE; } intel->use_early_z = driQueryOptionb(&intel->optionCache, "early_z"); @@ -795,6 +818,8 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv) INTEL_FIREVERTICES(intel); + _mesa_meta_free(&intel->ctx); + meta_destroy_metaops(&intel->meta); intel->vtbl.destroy(intel); @@ -814,6 +839,8 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv) intel->prim.vb = NULL; dri_bo_unreference(intel->prim.vb_bo); intel->prim.vb_bo = NULL; + dri_bo_unreference(intel->first_post_swapbuffers_batch); + intel->first_post_swapbuffers_batch = NULL; if (release_texture_heaps) { /* Nothing is currently done here to free texture heaps; diff --git a/src/mesa/drivers/dri/intel/intel_context.h b/src/mesa/drivers/dri/intel/intel_context.h index 08bea88c95..0d9db5eb1d 100644 --- a/src/mesa/drivers/dri/intel/intel_context.h +++ b/src/mesa/drivers/dri/intel/intel_context.h @@ -80,9 +80,13 @@ extern void intelFallback(struct intel_context *intel, GLuint bit, #define INTEL_MAX_FIXUP 64 + +/** + * intel_context is derived from Mesa's context class: GLcontext. + */ struct intel_context { - GLcontext ctx; /* the parent class */ + GLcontext ctx; /**< base class, must be first field */ struct { @@ -178,6 +182,7 @@ struct intel_context GLboolean ttm; struct intel_batchbuffer *batch; + drm_intel_bo *first_post_swapbuffers_batch; GLboolean no_batch_wrap; unsigned batch_id; @@ -307,7 +312,7 @@ struct intel_context __DRIdrawablePrivate *driReadDrawable; __DRIscreenPrivate *driScreen; intelScreenPrivate *intelScreen; - volatile struct drm_i915_sarea *sarea; + volatile drm_i915_sarea_t *sarea; GLuint lastStamp; diff --git a/src/mesa/drivers/dri/intel/intel_extensions.c b/src/mesa/drivers/dri/intel/intel_extensions.c index 7742609d24..9f90ef0a69 100644 --- a/src/mesa/drivers/dri/intel/intel_extensions.c +++ b/src/mesa/drivers/dri/intel/intel_extensions.c @@ -48,6 +48,7 @@ #define need_GL_EXT_framebuffer_blit #define need_GL_EXT_gpu_program_parameters #define need_GL_EXT_point_parameters +#define need_GL_EXT_provoking_vertex #define need_GL_EXT_secondary_color #define need_GL_EXT_stencil_two_side #define need_GL_APPLE_vertex_array_object @@ -71,6 +72,7 @@ static const struct dri_extension card_extensions[] = { { "GL_ARB_half_float_pixel", NULL }, { "GL_ARB_multitexture", NULL }, { "GL_ARB_point_parameters", GL_ARB_point_parameters_functions }, + { "GL_ARB_point_sprite", NULL }, { "GL_ARB_texture_border_clamp", NULL }, { "GL_ARB_texture_cube_map", NULL }, { "GL_ARB_texture_env_add", NULL }, @@ -92,6 +94,7 @@ static const struct dri_extension card_extensions[] = { { "GL_EXT_fog_coord", GL_EXT_fog_coord_functions }, { "GL_EXT_gpu_program_parameters", GL_EXT_gpu_program_parameters_functions }, { "GL_EXT_packed_depth_stencil", NULL }, + { "GL_EXT_provoking_vertex", GL_EXT_provoking_vertex_functions }, { "GL_EXT_secondary_color", GL_EXT_secondary_color_functions }, { "GL_EXT_stencil_wrap", NULL }, { "GL_EXT_texture_edge_clamp", NULL }, @@ -119,8 +122,10 @@ static const struct dri_extension i915_extensions[] = { { "GL_ARB_fragment_program", NULL }, { "GL_ARB_shadow", NULL }, { "GL_ARB_texture_non_power_of_two", NULL }, + { "GL_ATI_separate_stencil", GL_ATI_separate_stencil_functions }, { "GL_ATI_texture_env_combine3", NULL }, { "GL_EXT_shadow_funcs", NULL }, + { "GL_EXT_stencil_two_side", GL_EXT_stencil_two_side_functions }, { "GL_NV_texture_env_combine4", NULL }, { NULL, NULL } }; @@ -135,6 +140,7 @@ static const struct dri_extension brw_extensions[] = { { "GL_ARB_framebuffer_object", GL_ARB_framebuffer_object_functions}, { "GL_ARB_occlusion_query", GL_ARB_occlusion_query_functions }, { "GL_ARB_point_sprite", NULL }, + { "GL_ARB_seamless_cube_map", NULL }, { "GL_ARB_shader_objects", GL_ARB_shader_objects_functions }, { "GL_ARB_shading_language_100", GL_VERSION_2_0_functions }, { "GL_ARB_shading_language_120", GL_VERSION_2_1_functions }, diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c index 666893596e..804c034840 100644 --- a/src/mesa/drivers/dri/intel/intel_fbo.c +++ b/src/mesa/drivers/dri/intel/intel_fbo.c @@ -35,6 +35,7 @@ #include "main/context.h" #include "main/texformat.h" #include "main/texrender.h" +#include "drivers/common/meta.h" #include "intel_context.h" #include "intel_buffers.h" @@ -700,74 +701,6 @@ intel_validate_framebuffer(GLcontext *ctx, struct gl_framebuffer *fb) /** - * Called from glBlitFramebuffer(). - * For now, we're doing an approximation with glCopyPixels(). - * XXX we need to bypass all the per-fragment operations, except scissor. - */ -static void -intel_blit_framebuffer(GLcontext *ctx, - GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, - GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, - GLbitfield mask, GLenum filter) -{ - const GLfloat xZoomSave = ctx->Pixel.ZoomX; - const GLfloat yZoomSave = ctx->Pixel.ZoomY; - GLsizei width, height; - GLfloat xFlip = 1.0F, yFlip = 1.0F; - - if (srcX1 < srcX0) { - GLint tmp = srcX1; - srcX1 = srcX0; - srcX0 = tmp; - xFlip = -1.0F; - } - - if (srcY1 < srcY0) { - GLint tmp = srcY1; - srcY1 = srcY0; - srcY0 = tmp; - yFlip = -1.0F; - } - - width = srcX1 - srcX0; - height = srcY1 - srcY0; - - ctx->Pixel.ZoomX = xFlip * (dstX1 - dstX0) / (srcX1 - srcY0); - ctx->Pixel.ZoomY = yFlip * (dstY1 - dstY0) / (srcY1 - srcY0); - - if (ctx->Pixel.ZoomX < 0.0F) { - dstX0 = MAX2(dstX0, dstX1); - } - else { - dstX0 = MIN2(dstX0, dstX1); - } - - if (ctx->Pixel.ZoomY < 0.0F) { - dstY0 = MAX2(dstY0, dstY1); - } - else { - dstY0 = MIN2(dstY0, dstY1); - } - - if (mask & GL_COLOR_BUFFER_BIT) { - ctx->Driver.CopyPixels(ctx, srcX0, srcY0, width, height, - dstX0, dstY0, GL_COLOR); - } - if (mask & GL_DEPTH_BUFFER_BIT) { - ctx->Driver.CopyPixels(ctx, srcX0, srcY0, width, height, - dstX0, dstY0, GL_DEPTH); - } - if (mask & GL_STENCIL_BUFFER_BIT) { - ctx->Driver.CopyPixels(ctx, srcX0, srcY0, width, height, - dstX0, dstY0, GL_STENCIL); - } - - ctx->Pixel.ZoomX = xZoomSave; - ctx->Pixel.ZoomY = yZoomSave; -} - - -/** * Do one-time context initializations related to GL_EXT_framebuffer_object. * Hook in device driver functions. */ @@ -782,5 +715,5 @@ intel_fbo_init(struct intel_context *intel) intel->ctx.Driver.FinishRenderTexture = intel_finish_render_texture; intel->ctx.Driver.ResizeBuffers = intel_resize_buffers; intel->ctx.Driver.ValidateFramebuffer = intel_validate_framebuffer; - intel->ctx.Driver.BlitFramebuffer = intel_blit_framebuffer; + intel->ctx.Driver.BlitFramebuffer = _mesa_meta_blit_framebuffer; } diff --git a/src/mesa/drivers/dri/intel/intel_pixel_copy.c b/src/mesa/drivers/dri/intel/intel_pixel_copy.c index 5d52335dee..ca796b3655 100644 --- a/src/mesa/drivers/dri/intel/intel_pixel_copy.c +++ b/src/mesa/drivers/dri/intel/intel_pixel_copy.c @@ -31,7 +31,7 @@ #include "main/state.h" #include "main/mtypes.h" #include "main/macros.h" -#include "swrast/swrast.h" +#include "drivers/common/meta.h" #include "intel_screen.h" #include "intel_context.h" @@ -97,162 +97,6 @@ intel_check_copypixel_blit_fragment_ops(GLcontext * ctx) ctx->Color.BlendEnabled); } -#ifdef I915 -/* Doesn't work for overlapping regions. Could do a double copy or - * just fallback. - */ -static GLboolean -do_texture_copypixels(GLcontext * ctx, - GLint srcx, GLint srcy, - GLsizei width, GLsizei height, - GLint dstx, GLint dsty, GLenum type) -{ - struct intel_context *intel = intel_context(ctx); - struct intel_region *dst = intel_drawbuf_region(intel); - struct intel_region *src = copypix_src_region(intel, type); - GLenum src_format; - GLenum src_type; - - DBG("%s %d,%d %dx%d --> %d,%d\n", __FUNCTION__, - srcx, srcy, width, height, dstx, dsty); - - if (!src || !dst || type != GL_COLOR) - return GL_FALSE; - - if (ctx->_ImageTransferState) { - if (INTEL_DEBUG & DEBUG_PIXEL) - fprintf(stderr, "%s: check_color failed\n", __FUNCTION__); - return GL_FALSE; - } - - /* Can't handle overlapping regions. Don't have sufficient control - * over rasterization to pull it off in-place. Punt on these for - * now. - * - * XXX: do a copy to a temporary. - */ - if (src->buffer == dst->buffer) { - drm_clip_rect_t srcbox; - drm_clip_rect_t dstbox; - drm_clip_rect_t tmp; - - srcbox.x1 = srcx; - srcbox.y1 = srcy; - srcbox.x2 = srcx + width; - srcbox.y2 = srcy + height; - - if (ctx->Pixel.ZoomX > 0) { - dstbox.x1 = dstx; - dstbox.x2 = dstx + width * ctx->Pixel.ZoomX; - } else { - dstbox.x1 = dstx + width * ctx->Pixel.ZoomX; - dstbox.x2 = dstx; - } - if (ctx->Pixel.ZoomY > 0) { - dstbox.y1 = dsty; - dstbox.y2 = dsty + height * ctx->Pixel.ZoomY; - } else { - dstbox.y1 = dsty + height * ctx->Pixel.ZoomY; - dstbox.y2 = dsty; - } - - DBG("src %d,%d %d,%d\n", srcbox.x1, srcbox.y1, srcbox.x2, srcbox.y2); - DBG("dst %d,%d %d,%d (%dx%d) (%f,%f)\n", dstbox.x1, dstbox.y1, dstbox.x2, dstbox.y2, - width, height, ctx->Pixel.ZoomX, ctx->Pixel.ZoomY); - - if (intel_intersect_cliprects(&tmp, &srcbox, &dstbox)) { - DBG("%s: regions overlap\n", __FUNCTION__); - return GL_FALSE; - } - } - - intelFlush(&intel->ctx); - - intel->vtbl.install_meta_state(intel); - - /* Is this true? Also will need to turn depth testing on according - * to state: - */ - intel->vtbl.meta_no_stencil_write(intel); - intel->vtbl.meta_no_depth_write(intel); - - /* Set the 3d engine to draw into the destination region: - */ - intel->vtbl.meta_draw_region(intel, dst, intel->depth_region); - - intel->vtbl.meta_import_pixel_state(intel); - - if (src->cpp == 2) { - src_format = GL_RGB; - src_type = GL_UNSIGNED_SHORT_5_6_5; - } - else { - src_format = GL_BGRA; - src_type = GL_UNSIGNED_BYTE; - } - - /* Set the frontbuffer up as a large rectangular texture. - */ - if (!intel->vtbl.meta_tex_rect_source(intel, src->buffer, 0, - src->pitch, - src->height, src_format, src_type)) { - intel->vtbl.leave_meta_state(intel); - return GL_FALSE; - } - - - intel->vtbl.meta_texture_blend_replace(intel); - - LOCK_HARDWARE(intel); - - if (intel->driDrawable->numClipRects) { - __DRIdrawablePrivate *dPriv = intel->driDrawable; - - - srcy = dPriv->h - srcy - height; /* convert from gl to hardware coords */ - - srcx += dPriv->x; - srcy += dPriv->y; - - /* Clip against the source region. This is the only source - * clipping we do. XXX: Just set the texcord wrap mode to clamp - * or similar. - * - */ - if (0) { - GLint orig_x = srcx; - GLint orig_y = srcy; - - if (!_mesa_clip_to_region(0, 0, src->pitch, src->height, - &srcx, &srcy, &width, &height)) - goto out; - - dstx += srcx - orig_x; - dsty += (srcy - orig_y) * ctx->Pixel.ZoomY; - } - - /* Just use the regular cliprect mechanism... Does this need to - * even hold the lock??? - */ - intel->vtbl.meta_draw_quad(intel, - dstx, - dstx + width * ctx->Pixel.ZoomX, - dPriv->h - (dsty + height * ctx->Pixel.ZoomY), - dPriv->h - (dsty), 0, /* XXX: what z value? */ - 0x00ff00ff, - srcx, srcx + width, srcy, srcy + height); - - out: - intel->vtbl.leave_meta_state(intel); - intel_batchbuffer_emit_mi_flush(intel->batch); - } - UNLOCK_HARDWARE(intel); - - DBG("%s: success\n", __FUNCTION__); - return GL_TRUE; -} -#endif /* I915 */ - /** * CopyPixels with the blitter. Don't support zooming, pixel transfer, etc. @@ -400,12 +244,5 @@ intelCopyPixels(GLcontext * ctx, if (do_blit_copypixels(ctx, srcx, srcy, width, height, destx, desty, type)) return; -#ifdef I915 - if (do_texture_copypixels(ctx, srcx, srcy, width, height, destx, desty, type)) - return; -#endif - - DBG("fallback to _swrast_CopyPixels\n"); - - _swrast_CopyPixels(ctx, srcx, srcy, width, height, destx, desty, type); + _mesa_meta_copy_pixels(ctx, srcx, srcy, width, height, destx, desty, type); } diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c index 7525cd9c4d..497f796764 100644 --- a/src/mesa/drivers/dri/intel/intel_regions.c +++ b/src/mesa/drivers/dri/intel/intel_regions.c @@ -181,6 +181,11 @@ intel_region_alloc(struct intel_context *intel, dri_bo *buffer; struct intel_region *region; + if (tiling == I915_TILING_X) + height = ALIGN(height, 8); + else if (tiling == I915_TILING_Y) + height = ALIGN(height, 32); + if (expect_accelerated_upload) { buffer = drm_intel_bo_alloc_for_render(intel->bufmgr, "region", pitch * cpp * height, 64); @@ -452,6 +457,7 @@ void intel_region_cow(struct intel_context *intel, struct intel_region *region) { struct intel_buffer_object *pbo = region->pbo; + GLboolean ok; intel_region_release_pbo(intel, region); @@ -463,13 +469,14 @@ intel_region_cow(struct intel_context *intel, struct intel_region *region) */ LOCK_HARDWARE(intel); - assert(intelEmitCopyBlit(intel, - region->cpp, - region->pitch, pbo->buffer, 0, region->tiling, - region->pitch, region->buffer, 0, region->tiling, - 0, 0, 0, 0, - region->pitch, region->height, - GL_COPY)); + ok = intelEmitCopyBlit(intel, + region->cpp, + region->pitch, pbo->buffer, 0, region->tiling, + region->pitch, region->buffer, 0, region->tiling, + 0, 0, 0, 0, + region->pitch, region->height, + GL_COPY); + assert(ok); UNLOCK_HARDWARE(intel); } diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c index 6bbc995c1e..1b8c56e68d 100644 --- a/src/mesa/drivers/dri/intel/intel_screen.c +++ b/src/mesa/drivers/dri/intel/intel_screen.c @@ -69,7 +69,11 @@ PUBLIC const char __driConfigOptions[] = DRI_CONF_DESC_END DRI_CONF_OPT_END - DRI_CONF_TEXTURE_TILING(false) +#ifdef I915 + DRI_CONF_TEXTURE_TILING(false) +#else + DRI_CONF_TEXTURE_TILING(true) +#endif DRI_CONF_OPT_BEGIN(early_z, bool, false) DRI_CONF_DESC(en, "Enable early Z in classic mode (unstable, 945-only).") @@ -628,10 +632,10 @@ intel_init_bufmgr(intelScreenPrivate *intelScreen) /* Otherwise, use the classic buffer manager. */ if (intelScreen->bufmgr == NULL) { if (gem_disable) { - fprintf(stderr, "GEM disabled. Using classic.\n"); + _mesa_warning(NULL, "GEM disabled. Using classic."); } else { - fprintf(stderr, "Failed to initialize GEM. " - "Falling back to classic.\n"); + _mesa_warning(NULL, + "Failed to initialize GEM. Falling back to classic."); } if (intelScreen->tex.size == 0) { diff --git a/src/mesa/drivers/dri/intel/intel_span.c b/src/mesa/drivers/dri/intel/intel_span.c index 34b78ebc1a..8df4990880 100644 --- a/src/mesa/drivers/dri/intel/intel_span.c +++ b/src/mesa/drivers/dri/intel/intel_span.c @@ -501,7 +501,7 @@ intel_map_unmap_buffers(struct intel_context *intel, GLboolean map) /** - * Prepare for softare rendering. Map current read/draw framebuffers' + * Prepare for software rendering. Map current read/draw framebuffers' * renderbuffes and all currently bound texture objects. * * Old note: Moved locking out to get reasonable span performance. @@ -526,7 +526,7 @@ intelSpanRenderStart(GLcontext * ctx) } /** - * Called when done softare rendering. Unmap the buffers we mapped in + * Called when done software rendering. Unmap the buffers we mapped in * the above function. */ void diff --git a/src/mesa/drivers/dri/intel/intel_tex_subimage.c b/src/mesa/drivers/dri/intel/intel_tex_subimage.c index 1f27131dac..89037073f8 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_subimage.c +++ b/src/mesa/drivers/dri/intel/intel_tex_subimage.c @@ -44,10 +44,12 @@ intelTexSubimage(GLcontext * ctx, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLint width, GLint height, GLint depth, + GLsizei imageSize, GLenum format, GLenum type, const void *pixels, const struct gl_pixelstore_attrib *packing, struct gl_texture_object *texObj, - struct gl_texture_image *texImage) + struct gl_texture_image *texImage, + GLboolean compressed) { struct intel_context *intel = intel_context(ctx); struct intel_texture_image *intelImage = intel_texture_image(texImage); @@ -59,9 +61,14 @@ intelTexSubimage(GLcontext * ctx, intelFlush(ctx); - pixels = - _mesa_validate_pbo_teximage(ctx, dims, width, height, depth, format, - type, pixels, packing, "glTexSubImage2D"); + if (compressed) + pixels = _mesa_validate_pbo_compressed_teximage(ctx, imageSize, + pixels, packing, + "glCompressedTexImage"); + else + pixels = _mesa_validate_pbo_teximage(ctx, dims, width, height, depth, + format, type, pixels, packing, + "glTexSubImage"); if (!pixels) return; @@ -90,15 +97,28 @@ intelTexSubimage(GLcontext * ctx, assert(dstRowStride); - if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat, - texImage->TexFormat, - texImage->Data, - xoffset, yoffset, zoffset, - dstRowStride, - texImage->ImageOffsets, - width, height, depth, - format, type, pixels, packing)) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "intelTexSubImage"); + if (compressed) { + if (intelImage->mt) { + struct intel_region *dst = intelImage->mt->region; + + _mesa_copy_rect(texImage->Data, dst->cpp, dst->pitch, + xoffset, yoffset / 4, + (width + 3) & ~3, (height + 3) / 4, + pixels, (width + 3) & ~3, 0, 0); + } else + memcpy(texImage->Data, pixels, imageSize); + } + else { + if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat, + texImage->TexFormat, + texImage->Data, + xoffset, yoffset, zoffset, + dstRowStride, + texImage->ImageOffsets, + width, height, depth, + format, type, pixels, packing)) { + _mesa_error(ctx, GL_OUT_OF_MEMORY, "intelTexSubImage"); + } } _mesa_unmap_teximage_pbo(ctx, packing); @@ -132,8 +152,8 @@ intelTexSubImage3D(GLcontext * ctx, intelTexSubimage(ctx, 3, target, level, xoffset, yoffset, zoffset, - width, height, depth, - format, type, pixels, packing, texObj, texImage); + width, height, depth, 0, + format, type, pixels, packing, texObj, texImage, GL_FALSE); } @@ -152,8 +172,8 @@ intelTexSubImage2D(GLcontext * ctx, intelTexSubimage(ctx, 2, target, level, xoffset, yoffset, 0, - width, height, 1, - format, type, pixels, packing, texObj, texImage); + width, height, 1, 0, + format, type, pixels, packing, texObj, texImage, GL_FALSE); } @@ -172,8 +192,8 @@ intelTexSubImage1D(GLcontext * ctx, intelTexSubimage(ctx, 1, target, level, xoffset, 0, 0, - width, 1, 1, - format, type, pixels, packing, texObj, texImage); + width, 1, 1, 0, + format, type, pixels, packing, texObj, texImage, GL_FALSE); } static void @@ -187,8 +207,11 @@ intelCompressedTexSubImage2D(GLcontext * ctx, struct gl_texture_object *texObj, struct gl_texture_image *texImage) { - fprintf(stderr, "stubbed CompressedTexSubImage2D: %dx%d@%dx%d\n", - width, height, xoffset, yoffset); + intelTexSubimage(ctx, 2, + target, level, + xoffset, yoffset, 0, + width, height, 1, imageSize, + format, 0, pixels, &ctx->Unpack, texObj, texImage, GL_TRUE); } diff --git a/src/mesa/drivers/dri/r200/.gitignore b/src/mesa/drivers/dri/r200/.gitignore deleted file mode 100644 index 2f9cd1a987..0000000000 --- a/src/mesa/drivers/dri/r200/.gitignore +++ /dev/null @@ -1,15 +0,0 @@ -radeon_bocs_wrapper.h -radeon_bo_legacy.[ch] -radeon_chipset.h -radeon_cmdbuf.h -radeon_common.[ch] -radeon_common_context.[ch] -radeon_cs_legacy.[ch] -radeon_dma.[ch] -radeon_fbo.c -radeon_lock.[ch] -radeon_mipmap_tree.[ch] -radeon_screen.[ch] -radeon_span.[ch] -radeon_texture.[ch] -server diff --git a/src/mesa/drivers/dri/r200/Makefile b/src/mesa/drivers/dri/r200/Makefile index 4686241957..e81a1b38ac 100644 --- a/src/mesa/drivers/dri/r200/Makefile +++ b/src/mesa/drivers/dri/r200/Makefile @@ -55,41 +55,6 @@ X86_SOURCES = DRIVER_DEFINES = -DRADEON_COMMON=1 -DRADEON_COMMON_FOR_R200 -SYMLINKS = \ - server/radeon_egl.c \ - server/radeon_dri.c \ - server/radeon_dri.h \ - server/radeon.h \ - server/radeon_macros.h \ - server/radeon_reg.h - -COMMON_SYMLINKS = \ - radeon_chipset.h \ - radeon_screen.c \ - radeon_screen.h \ - radeon_bo_legacy.c \ - radeon_cs_legacy.c \ - radeon_bo_legacy.h \ - radeon_cs_legacy.h \ - radeon_bocs_wrapper.h \ - radeon_span.h \ - radeon_span.c \ - radeon_lock.c \ - radeon_lock.h \ - radeon_common.c \ - radeon_common_context.c \ - radeon_common_context.h \ - radeon_common.h \ - radeon_cmdbuf.h \ - radeon_mipmap_tree.c \ - radeon_mipmap_tree.h \ - radeon_texture.c \ - radeon_texture.h \ - radeon_dma.c \ - radeon_dma.h \ - radeon_fbo.c \ - $(CS_SOURCES) - DRI_LIB_DEPS += $(RADEON_LDFLAGS) ##### TARGETS ##### @@ -99,15 +64,4 @@ include ../Makefile.template #INCLUDES += -I../radeon/server -server: - mkdir -p server - -$(SYMLINKS): server - @[ -e $@ ] || ln -sf ../../radeon/$@ server/ - - -$(COMMON_SYMLINKS): - @[ -e $@ ] || ln -sf ../radeon/$@ ./ - -symlinks: $(SYMLINKS) $(COMMON_SYMLINKS) - +symlinks: diff --git a/src/mesa/drivers/dri/r200/r200_cmdbuf.c b/src/mesa/drivers/dri/r200/r200_cmdbuf.c index d49f4fabe7..14d6bc19c9 100644 --- a/src/mesa/drivers/dri/r200/r200_cmdbuf.c +++ b/src/mesa/drivers/dri/r200/r200_cmdbuf.c @@ -107,31 +107,37 @@ void r200SetUpAtomList( r200ContextPtr rmesa ) void r200EmitScissor(r200ContextPtr rmesa) { + unsigned x1, y1, x2, y2; + struct radeon_renderbuffer *rrb; BATCH_LOCALS(&rmesa->radeon); if (!rmesa->radeon.radeonScreen->kernel_mm) { return; } + rrb = radeon_get_colorbuffer(&rmesa->radeon); + if (!rrb || !rrb->bo) + return; + if (rmesa->radeon.state.scissor.enabled) { - BEGIN_BATCH(8); - OUT_BATCH(CP_PACKET0(R200_RE_CNTL, 0)); - OUT_BATCH(R200_SCISSOR_ENABLE | rmesa->hw.set.cmd[SET_RE_CNTL]); - OUT_BATCH(CP_PACKET0(R200_RE_AUX_SCISSOR_CNTL, 0)); - OUT_BATCH(R200_SCISSOR_ENABLE_0); - OUT_BATCH(CP_PACKET0(R200_RE_SCISSOR_TL_0, 0)); - OUT_BATCH((rmesa->radeon.state.scissor.rect.y1 << 16) | - rmesa->radeon.state.scissor.rect.x1); - OUT_BATCH(CP_PACKET0(R200_RE_SCISSOR_BR_0, 0)); - OUT_BATCH(((rmesa->radeon.state.scissor.rect.y2 - 1) << 16) | - (rmesa->radeon.state.scissor.rect.x2 - 1)); - END_BATCH(); + x1 = rmesa->radeon.state.scissor.rect.x1; + y1 = rmesa->radeon.state.scissor.rect.y1; + x2 = rmesa->radeon.state.scissor.rect.x2 - 1; + y2 = rmesa->radeon.state.scissor.rect.y2 - 1; } else { - BEGIN_BATCH(4); - OUT_BATCH(CP_PACKET0(R200_RE_CNTL, 0)); - OUT_BATCH(rmesa->hw.set.cmd[SET_RE_CNTL] & ~R200_SCISSOR_ENABLE); - OUT_BATCH(CP_PACKET0(R200_RE_AUX_SCISSOR_CNTL, 0)); - OUT_BATCH(0); - END_BATCH(); + x1 = 0; + y1 = 0; + x2 = rrb->base.Width - 1; + y2 = rrb->base.Height - 1; } + BEGIN_BATCH(8); + OUT_BATCH(CP_PACKET0(R200_RE_CNTL, 0)); + OUT_BATCH(R200_SCISSOR_ENABLE | rmesa->hw.set.cmd[SET_RE_CNTL]); + OUT_BATCH(CP_PACKET0(R200_RE_AUX_SCISSOR_CNTL, 0)); + OUT_BATCH(0); + OUT_BATCH(CP_PACKET0(R200_RE_TOP_LEFT, 0)); + OUT_BATCH((y1 << 16) | x1); + OUT_BATCH(CP_PACKET0(R200_RE_WIDTH_HEIGHT, 0)); + OUT_BATCH((y2 << 16) | x2); + END_BATCH(); } /* Fire a section of the retained (indexed_verts) buffer as a regular @@ -234,15 +240,9 @@ GLushort *r200AllocEltsOpenEnded( r200ContextPtr rmesa, radeonEmitState(&rmesa->radeon); -#ifdef RADEON_DEBUG_BO - rmesa->radeon.tcl.elt_dma_bo = radeon_bo_open(rmesa->radeon.radeonScreen->bom, - 0, R200_ELT_BUF_SZ, 4, - RADEON_GEM_DOMAIN_GTT, 0, "ELT"); -#else rmesa->radeon.tcl.elt_dma_bo = radeon_bo_open(rmesa->radeon.radeonScreen->bom, 0, R200_ELT_BUF_SZ, 4, RADEON_GEM_DOMAIN_GTT, 0); -#endif rmesa->radeon.tcl.elt_dma_offset = 0; rmesa->tcl.elt_used = min_nr * 2; diff --git a/src/mesa/drivers/dri/r200/r200_context.c b/src/mesa/drivers/dri/r200/r200_context.c index 9a92a32079..8cb287de26 100644 --- a/src/mesa/drivers/dri/r200/r200_context.c +++ b/src/mesa/drivers/dri/r200/r200_context.c @@ -500,3 +500,15 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual, } +void r200DestroyContext( __DRIcontextPrivate *driContextPriv ) +{ + int i; + r200ContextPtr rmesa = (r200ContextPtr)driContextPriv->driverPrivate; + if (rmesa) + { + for ( i = 0 ; i < R200_MAX_TEXTURE_UNITS ; i++ ) { + _math_matrix_dtr( &rmesa->TexGenMatrix[i] ); + } + } + radeonDestroyContext(driContextPriv); +} diff --git a/src/mesa/drivers/dri/r200/r200_state_init.c b/src/mesa/drivers/dri/r200/r200_state_init.c index bc871d9904..78ad5baebb 100644 --- a/src/mesa/drivers/dri/r200/r200_state_init.c +++ b/src/mesa/drivers/dri/r200/r200_state_init.c @@ -515,7 +515,7 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) if (drb) dwords += 6; if (rrb) - dwords += 6; + dwords += 8; if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) dwords += 4; @@ -546,7 +546,7 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLORPITCH, 0)); - OUT_BATCH(cbpitch); + OUT_BATCH_RELOC(cbpitch, rrb->bo, cbpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0); } if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) { @@ -563,7 +563,6 @@ static void tex_emit(GLcontext *ctx, struct radeon_state_atom *atom) uint32_t dwords = atom->cmd_size; int i = atom->idx; radeonTexObj *t = r200->state.texture.unit[i].texobj; - radeon_mipmap_level *lvl; if (t && t->mt && !t->image_override) dwords += 2; @@ -591,7 +590,6 @@ static void tex_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) uint32_t dwords = atom->cmd_size; int i = atom->idx; radeonTexObj *t = r200->state.texture.unit[i].texobj; - radeon_mipmap_level *lvl; int hastexture = 1; if (!r200->state.texture.unit[i].unitneeded) @@ -774,7 +772,7 @@ void r200InitState( r200ContextPtr rmesa ) ALLOC_STATE( afs[1], never, AFS_STATE_SIZE, "AFS/afsinst-1", 1 ); } - for (i = 0; i < 5; i++) + for (i = 0; i < 6; i++) if (rmesa->radeon.radeonScreen->kernel_mm) rmesa->hw.tex[i].emit = tex_emit_cs; else @@ -786,7 +784,7 @@ void r200InitState( r200ContextPtr rmesa ) ALLOC_STATE( cube[3], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-3", 3 ); ALLOC_STATE( cube[4], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-4", 4 ); ALLOC_STATE( cube[5], tex_cube, CUBE_STATE_SIZE, "CUBE/tex-5", 5 ); - for (i = 0; i < 5; i++) + for (i = 0; i < 6; i++) if (rmesa->radeon.radeonScreen->kernel_mm) rmesa->hw.cube[i].emit = cube_emit_cs; else diff --git a/src/mesa/drivers/dri/r200/radeon_bo_legacy.c b/src/mesa/drivers/dri/r200/radeon_bo_legacy.c new file mode 120000 index 0000000000..79ad050e6b --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_bo_legacy.c @@ -0,0 +1 @@ +../radeon/radeon_bo_legacy.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_bo_legacy.h b/src/mesa/drivers/dri/r200/radeon_bo_legacy.h new file mode 120000 index 0000000000..83b0f7ffab --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_bo_legacy.h @@ -0,0 +1 @@ +../radeon/radeon_bo_legacy.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_bocs_wrapper.h b/src/mesa/drivers/dri/r200/radeon_bocs_wrapper.h new file mode 120000 index 0000000000..ca894b2443 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_bocs_wrapper.h @@ -0,0 +1 @@ +../radeon/radeon_bocs_wrapper.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_chipset.h b/src/mesa/drivers/dri/r200/radeon_chipset.h new file mode 120000 index 0000000000..eba99001ff --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_chipset.h @@ -0,0 +1 @@ +../radeon/radeon_chipset.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_cmdbuf.h b/src/mesa/drivers/dri/r200/radeon_cmdbuf.h new file mode 120000 index 0000000000..a799e1dc6d --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_cmdbuf.h @@ -0,0 +1 @@ +../radeon/radeon_cmdbuf.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_common.c b/src/mesa/drivers/dri/r200/radeon_common.c new file mode 120000 index 0000000000..67b19ba940 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_common.c @@ -0,0 +1 @@ +../radeon/radeon_common.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_common.h b/src/mesa/drivers/dri/r200/radeon_common.h new file mode 120000 index 0000000000..5bcb696a9f --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_common.h @@ -0,0 +1 @@ +../radeon/radeon_common.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_common_context.c b/src/mesa/drivers/dri/r200/radeon_common_context.c new file mode 120000 index 0000000000..86800f3819 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_common_context.c @@ -0,0 +1 @@ +../radeon/radeon_common_context.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_common_context.h b/src/mesa/drivers/dri/r200/radeon_common_context.h new file mode 120000 index 0000000000..4d66312550 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_common_context.h @@ -0,0 +1 @@ +../radeon/radeon_common_context.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_cs_legacy.c b/src/mesa/drivers/dri/r200/radeon_cs_legacy.c new file mode 120000 index 0000000000..006720f8a4 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_cs_legacy.c @@ -0,0 +1 @@ +../radeon/radeon_cs_legacy.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_cs_legacy.h b/src/mesa/drivers/dri/r200/radeon_cs_legacy.h new file mode 120000 index 0000000000..a5f95e0a3d --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_cs_legacy.h @@ -0,0 +1 @@ +../radeon/radeon_cs_legacy.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_cs_space_drm.c b/src/mesa/drivers/dri/r200/radeon_cs_space_drm.c new file mode 120000 index 0000000000..c248ea7d1a --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_cs_space_drm.c @@ -0,0 +1 @@ +../radeon/radeon_cs_space_drm.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_dma.c b/src/mesa/drivers/dri/r200/radeon_dma.c new file mode 120000 index 0000000000..43be000625 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_dma.c @@ -0,0 +1 @@ +../radeon/radeon_dma.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_dma.h b/src/mesa/drivers/dri/r200/radeon_dma.h new file mode 120000 index 0000000000..82e50634e3 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_dma.h @@ -0,0 +1 @@ +../radeon/radeon_dma.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_fbo.c b/src/mesa/drivers/dri/r200/radeon_fbo.c new file mode 120000 index 0000000000..0d738d8d78 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_fbo.c @@ -0,0 +1 @@ +../radeon/radeon_fbo.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_lock.c b/src/mesa/drivers/dri/r200/radeon_lock.c new file mode 120000 index 0000000000..af4108a8e3 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_lock.c @@ -0,0 +1 @@ +../radeon/radeon_lock.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_lock.h b/src/mesa/drivers/dri/r200/radeon_lock.h new file mode 120000 index 0000000000..64bdf94ee7 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_lock.h @@ -0,0 +1 @@ +../radeon/radeon_lock.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_mipmap_tree.c b/src/mesa/drivers/dri/r200/radeon_mipmap_tree.c new file mode 120000 index 0000000000..31c0cfbe94 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_mipmap_tree.c @@ -0,0 +1 @@ +../radeon/radeon_mipmap_tree.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_mipmap_tree.h b/src/mesa/drivers/dri/r200/radeon_mipmap_tree.h new file mode 120000 index 0000000000..254d50cf8c --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_mipmap_tree.h @@ -0,0 +1 @@ +../radeon/radeon_mipmap_tree.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_screen.c b/src/mesa/drivers/dri/r200/radeon_screen.c new file mode 120000 index 0000000000..86161118dd --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_screen.c @@ -0,0 +1 @@ +../radeon/radeon_screen.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_screen.h b/src/mesa/drivers/dri/r200/radeon_screen.h new file mode 120000 index 0000000000..23bb6bd459 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_screen.h @@ -0,0 +1 @@ +../radeon/radeon_screen.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_span.c b/src/mesa/drivers/dri/r200/radeon_span.c new file mode 120000 index 0000000000..232868c4c9 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_span.c @@ -0,0 +1 @@ +../radeon/radeon_span.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_span.h b/src/mesa/drivers/dri/r200/radeon_span.h new file mode 120000 index 0000000000..f9d634508c --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_span.h @@ -0,0 +1 @@ +../radeon/radeon_span.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_texture.c b/src/mesa/drivers/dri/r200/radeon_texture.c new file mode 120000 index 0000000000..a822710915 --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_texture.c @@ -0,0 +1 @@ +../radeon/radeon_texture.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/radeon_texture.h b/src/mesa/drivers/dri/r200/radeon_texture.h new file mode 120000 index 0000000000..17fac3d5ea --- /dev/null +++ b/src/mesa/drivers/dri/r200/radeon_texture.h @@ -0,0 +1 @@ +../radeon/radeon_texture.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/server/radeon.h b/src/mesa/drivers/dri/r200/server/radeon.h new file mode 120000 index 0000000000..81274a54f1 --- /dev/null +++ b/src/mesa/drivers/dri/r200/server/radeon.h @@ -0,0 +1 @@ +../../radeon/server/radeon.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/server/radeon_dri.c b/src/mesa/drivers/dri/r200/server/radeon_dri.c new file mode 120000 index 0000000000..d05847d650 --- /dev/null +++ b/src/mesa/drivers/dri/r200/server/radeon_dri.c @@ -0,0 +1 @@ +../../radeon/server/radeon_dri.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/server/radeon_dri.h b/src/mesa/drivers/dri/r200/server/radeon_dri.h new file mode 120000 index 0000000000..27c591d3c9 --- /dev/null +++ b/src/mesa/drivers/dri/r200/server/radeon_dri.h @@ -0,0 +1 @@ +../../radeon/server/radeon_dri.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/server/radeon_egl.c b/src/mesa/drivers/dri/r200/server/radeon_egl.c new file mode 120000 index 0000000000..d7735a7643 --- /dev/null +++ b/src/mesa/drivers/dri/r200/server/radeon_egl.c @@ -0,0 +1 @@ +../../radeon/server/radeon_egl.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/server/radeon_macros.h b/src/mesa/drivers/dri/r200/server/radeon_macros.h new file mode 120000 index 0000000000..c56cd735b8 --- /dev/null +++ b/src/mesa/drivers/dri/r200/server/radeon_macros.h @@ -0,0 +1 @@ +../../radeon/server/radeon_macros.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r200/server/radeon_reg.h b/src/mesa/drivers/dri/r200/server/radeon_reg.h new file mode 120000 index 0000000000..e2349dcb68 --- /dev/null +++ b/src/mesa/drivers/dri/r200/server/radeon_reg.h @@ -0,0 +1 @@ +../../radeon/server/radeon_reg.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/.gitignore b/src/mesa/drivers/dri/r300/.gitignore deleted file mode 100644 index 2f9cd1a987..0000000000 --- a/src/mesa/drivers/dri/r300/.gitignore +++ /dev/null @@ -1,15 +0,0 @@ -radeon_bocs_wrapper.h -radeon_bo_legacy.[ch] -radeon_chipset.h -radeon_cmdbuf.h -radeon_common.[ch] -radeon_common_context.[ch] -radeon_cs_legacy.[ch] -radeon_dma.[ch] -radeon_fbo.c -radeon_lock.[ch] -radeon_mipmap_tree.[ch] -radeon_screen.[ch] -radeon_span.[ch] -radeon_texture.[ch] -server diff --git a/src/mesa/drivers/dri/r300/Makefile b/src/mesa/drivers/dri/r300/Makefile index a77209074a..77b3d168f3 100644 --- a/src/mesa/drivers/dri/r300/Makefile +++ b/src/mesa/drivers/dri/r300/Makefile @@ -36,7 +36,8 @@ RADEON_COMMON_SOURCES = \ radeon_cs_legacy.c \ radeon_mipmap_tree.c \ radeon_span.c \ - radeon_fbo.c + radeon_fbo.c \ + radeon_buffer_objects.c DRIVER_SOURCES = \ radeon_screen.c \ @@ -48,20 +49,12 @@ DRIVER_SOURCES = \ r300_render.c \ r300_tex.c \ r300_texstate.c \ - radeon_program.c \ - radeon_program_alu.c \ - radeon_program_pair.c \ - radeon_nqssadce.c \ r300_vertprog.c \ r300_fragprog_common.c \ - r300_fragprog.c \ - r300_fragprog_swizzle.c \ - r300_fragprog_emit.c \ - r500_fragprog.c \ - r500_fragprog_emit.c \ r300_shader.c \ r300_emit.c \ r300_swtcl.c \ + r300_queryobj.c \ $(RADEON_COMMON_SOURCES) \ $(EGL_SOURCES) \ $(CS_SOURCES) @@ -73,54 +66,18 @@ DRIVER_DEFINES = -DCOMPILE_R300 -DR200_MERGED=0 \ # -DRADEON_BO_TRACK \ -Wall -SYMLINKS = \ - server/radeon_dri.c \ - server/radeon_dri.h \ - server/radeon.h \ - server/radeon_macros.h \ - server/radeon_reg.h \ - server/radeon_egl.c - -COMMON_SYMLINKS = \ - radeon_chipset.h \ - radeon_screen.c \ - radeon_screen.h \ - radeon_span.h \ - radeon_span.c \ - radeon_bo_legacy.c \ - radeon_cs_legacy.c \ - radeon_bo_legacy.h \ - radeon_cs_legacy.h \ - radeon_bocs_wrapper.h \ - radeon_lock.c \ - radeon_lock.h \ - radeon_common.c \ - radeon_common.h \ - radeon_common_context.c \ - radeon_common_context.h \ - radeon_cmdbuf.h \ - radeon_dma.c \ - radeon_dma.h \ - radeon_mipmap_tree.c \ - radeon_mipmap_tree.h \ - radeon_texture.c \ - radeon_texture.h \ - radeon_fbo.c \ - $(CS_SOURCES) - DRI_LIB_DEPS += $(RADEON_LDFLAGS) +PIPE_DRIVERS = compiler/libr300compiler.a + ##### TARGETS ##### include ../Makefile.template -server: - mkdir -p server - -$(SYMLINKS): server - @[ -e $@ ] || ln -sf ../../radeon/$@ server/ +symlinks: -$(COMMON_SYMLINKS): - @[ -e $@ ] || ln -sf ../radeon/$@ ./ +# Mark the archive phony so that we always check for recompilation +.PHONY : compiler/libr300compiler.a -symlinks: $(SYMLINKS) $(COMMON_SYMLINKS) +compiler/libr300compiler.a: + cd compiler && $(MAKE) diff --git a/src/mesa/drivers/dri/r300/compiler/Makefile b/src/mesa/drivers/dri/r300/compiler/Makefile new file mode 100644 index 0000000000..d973844192 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/Makefile @@ -0,0 +1,75 @@ +# src/mesa/drivers/dri/r300/compiler/Makefile + +TOP = ../../../../../.. +include $(TOP)/configs/current + +LIBNAME = r300compiler + +C_SOURCES = \ + radeon_code.c \ + radeon_compiler.c \ + radeon_nqssadce.c \ + radeon_program.c \ + radeon_program_alu.c \ + radeon_program_pair.c \ + r3xx_fragprog.c \ + r300_fragprog.c \ + r300_fragprog_swizzle.c \ + r300_fragprog_emit.c \ + r500_fragprog.c \ + r500_fragprog_emit.c \ + r3xx_vertprog.c \ + r3xx_vertprog_dump.c \ + \ + memory_pool.c + + +### Basic defines ### + +OBJECTS = $(C_SOURCES:.c=.o) \ + $(CPP_SOURCES:.cpp=.o) \ + $(ASM_SOURCES:.S=.o) + +INCLUDES = \ + -I. \ + -I$(TOP)/include \ + -I$(TOP)/src/mesa \ + + +##### TARGETS ##### + +default: depend lib$(LIBNAME).a + +lib$(LIBNAME).a: $(OBJECTS) Makefile $(TOP)/configs/current + $(MKLIB) -o $(LIBNAME) -static $(OBJECTS) + +depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(SYMLINKS) + rm -f depend + touch depend + $(MKDEP) $(MKDEP_OPTIONS) $(INCLUDES) $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) 2> /dev/null + +# Emacs tags +tags: + etags `find . -name \*.[ch]` `find ../include` + +# Remove .o and backup files +clean: + rm -f $(OBJECTS) lib$(LIBNAME).a depend depend.bak + +# Dummy target +install: + @echo -n "" + +##### RULES ##### + +.c.o: + $(CC) -c $(INCLUDES) $(CFLAGS) $(LIBRARY_DEFINES) $< -o $@ + +.cpp.o: + $(CXX) -c $(INCLUDES) $(CXXFLAGS) $(LIBRARY_DEFINES) $< -o $@ + +.S.o: + $(CC) -c $(INCLUDES) $(CFLAGS) $(LIBRARY_DEFINES) $< -o $@ + + +sinclude depend diff --git a/src/mesa/drivers/dri/r300/compiler/memory_pool.c b/src/mesa/drivers/dri/r300/compiler/memory_pool.c new file mode 100644 index 0000000000..37aa2b6579 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/memory_pool.c @@ -0,0 +1,95 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "memory_pool.h" + +#include <assert.h> +#include <stdlib.h> +#include <string.h> + + +#define POOL_LARGE_ALLOC 4096 +#define POOL_ALIGN 4 + + +struct memory_block { + struct memory_block * next; +}; + +void memory_pool_init(struct memory_pool * pool) +{ + memset(pool, 0, sizeof(struct memory_pool)); +} + + +void memory_pool_destroy(struct memory_pool * pool) +{ + while(pool->blocks) { + struct memory_block * block = pool->blocks; + pool->blocks = block->next; + free(block); + } +} + +static void refill_pool(struct memory_pool * pool) +{ + unsigned int blocksize = pool->total_allocated; + struct memory_block * newblock; + + if (!blocksize) + blocksize = 2*POOL_LARGE_ALLOC; + + newblock = (struct memory_block*)malloc(blocksize); + newblock->next = pool->blocks; + pool->blocks = newblock; + + pool->head = (unsigned char*)(newblock + 1); + pool->end = ((unsigned char*)newblock) + blocksize; + pool->total_allocated += blocksize; +} + + +void * memory_pool_malloc(struct memory_pool * pool, unsigned int bytes) +{ + if (bytes < POOL_LARGE_ALLOC) { + if (pool->head + bytes > pool->end) + refill_pool(pool); + + assert(pool->head + bytes <= pool->end); + + void * ptr = pool->head; + + pool->head += bytes; + pool->head = (unsigned char*)(((unsigned long)pool->head + POOL_ALIGN - 1) & ~(POOL_ALIGN - 1)); + + return ptr; + } else { + struct memory_block * block = (struct memory_block*)malloc(bytes + sizeof(struct memory_block)); + + block->next = pool->blocks; + pool->blocks = block; + + return (block + 1); + } +} + + diff --git a/src/mesa/drivers/dri/r300/compiler/memory_pool.h b/src/mesa/drivers/dri/r300/compiler/memory_pool.h new file mode 100644 index 0000000000..ce23c319ad --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/memory_pool.h @@ -0,0 +1,49 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef MEMORY_POOL_H +#define MEMORY_POOL_H + +struct memory_block; + +/** + * Provides a pool of memory that can quickly be allocated from, at the + * cost of being unable to explicitly free one of the allocated blocks. + * Instead, the entire pool can be freed at once. + * + * The idea is to allow one to quickly allocate a flexible amount of + * memory during operations like shader compilation while avoiding + * reference counting headaches. + */ +struct memory_pool { + unsigned char * head; + unsigned char * end; + unsigned int total_allocated; + struct memory_block * blocks; +}; + + +void memory_pool_init(struct memory_pool * pool); +void memory_pool_destroy(struct memory_pool * pool); +void * memory_pool_malloc(struct memory_pool * pool, unsigned int bytes); + +#endif /* MEMORY_POOL_H */ diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c new file mode 100644 index 0000000000..6c9fba4914 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c @@ -0,0 +1,416 @@ +/* + * Copyright (C) 2005 Ben Skeggs. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "r300_fragprog.h" + +#include "shader/prog_parameter.h" + +#include "../r300_reg.h" + +static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int tmu) +{ + struct prog_src_register reg = { 0, }; + + reg.File = PROGRAM_STATE_VAR; + reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu); + reg.Swizzle = SWIZZLE_WWWW; + return reg; +} + +/** + * Transform TEX, TXP, TXB, and KIL instructions in the following way: + * - premultiply texture coordinates for RECT + * - extract operand swizzles + * - introduce a temporary register when write masks are needed + */ +GLboolean r300_transform_TEX( + struct radeon_compiler * c, + struct rc_instruction* inst, + void* data) +{ + struct r300_fragment_program_compiler *compiler = + (struct r300_fragment_program_compiler*)data; + + if (inst->I.Opcode != OPCODE_TEX && + inst->I.Opcode != OPCODE_TXB && + inst->I.Opcode != OPCODE_TXP && + inst->I.Opcode != OPCODE_KIL) + return GL_FALSE; + + /* ARB_shadow & EXT_shadow_funcs */ + if (inst->I.Opcode != OPCODE_KIL && + c->Program.ShadowSamplers & (1 << inst->I.TexSrcUnit)) { + GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; + + if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) { + inst->I.Opcode = OPCODE_MOV; + + if (comparefunc == GL_ALWAYS) { + inst->I.SrcReg[0].File = PROGRAM_BUILTIN; + inst->I.SrcReg[0].Swizzle = SWIZZLE_1111; + } else { + inst->I.SrcReg[0] = shadow_ambient(c, inst->I.TexSrcUnit); + } + + return GL_TRUE; + } else { + GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; + GLuint depthmode = compiler->state.unit[inst->I.TexSrcUnit].depth_texture_mode; + struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, inst); + struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_rcp); + struct rc_instruction * inst_cmp = rc_insert_new_instruction(c, inst_mad); + int pass, fail; + + inst_rcp->I.Opcode = OPCODE_RCP; + inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY; + inst_rcp->I.DstReg.Index = rc_find_free_temporary(c); + inst_rcp->I.DstReg.WriteMask = WRITEMASK_W; + inst_rcp->I.SrcReg[0] = inst->I.SrcReg[0]; + inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW; + + inst_cmp->I.DstReg = inst->I.DstReg; + inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.Index = rc_find_free_temporary(c); + inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + + inst_mad->I.Opcode = OPCODE_MAD; + inst_mad->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mad->I.DstReg.Index = rc_find_free_temporary(c); + inst_mad->I.SrcReg[0] = inst->I.SrcReg[0]; + inst_mad->I.SrcReg[0].Swizzle = SWIZZLE_ZZZZ; + inst_mad->I.SrcReg[1].File = PROGRAM_TEMPORARY; + inst_mad->I.SrcReg[1].Index = inst_rcp->I.DstReg.Index; + inst_mad->I.SrcReg[1].Swizzle = SWIZZLE_WWWW; + inst_mad->I.SrcReg[2].File = PROGRAM_TEMPORARY; + inst_mad->I.SrcReg[2].Index = inst->I.DstReg.Index; + if (depthmode == 0) /* GL_LUMINANCE */ + inst_mad->I.SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z); + else if (depthmode == 2) /* GL_ALPHA */ + inst_mad->I.SrcReg[2].Swizzle = SWIZZLE_WWWW; + + /* Recall that SrcReg[0] is tex, SrcReg[2] is r and: + * r < tex <=> -tex+r < 0 + * r >= tex <=> not (-tex+r < 0 */ + if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL) + inst_mad->I.SrcReg[2].Negate = inst_mad->I.SrcReg[2].Negate ^ NEGATE_XYZW; + else + inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ NEGATE_XYZW; + + inst_cmp->I.Opcode = OPCODE_CMP; + /* DstReg has been filled out above */ + inst_cmp->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_cmp->I.SrcReg[0].Index = inst_mad->I.DstReg.Index; + + if (comparefunc == GL_LESS || comparefunc == GL_GREATER) { + pass = 1; + fail = 2; + } else { + pass = 2; + fail = 1; + } + + inst_cmp->I.SrcReg[pass].File = PROGRAM_BUILTIN; + inst_cmp->I.SrcReg[pass].Swizzle = SWIZZLE_1111; + inst_cmp->I.SrcReg[fail] = shadow_ambient(c, inst->I.TexSrcUnit); + } + } + + /* Hardware uses [0..1]x[0..1] range for rectangle textures + * instead of [0..Width]x[0..Height]. + * Add a scaling instruction. + */ + if (inst->I.Opcode != OPCODE_KIL && inst->I.TexSrcTarget == TEXTURE_RECT_INDEX) { + struct rc_instruction * inst_mul = rc_insert_new_instruction(c, inst->Prev); + + inst_mul->I.Opcode = OPCODE_MUL; + inst_mul->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mul->I.DstReg.Index = rc_find_free_temporary(c); + inst_mul->I.SrcReg[0] = inst->I.SrcReg[0]; + inst_mul->I.SrcReg[1].File = PROGRAM_STATE_VAR; + inst_mul->I.SrcReg[1].Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_R300_TEXRECT_FACTOR, inst->I.TexSrcUnit); + + reset_srcreg(&inst->I.SrcReg[0]); + inst->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[0].Index = inst_mul->I.DstReg.Index; + } + + /* Cannot write texture to output registers or with masks */ + if (inst->I.Opcode != OPCODE_KIL && + (inst->I.DstReg.File != PROGRAM_TEMPORARY || inst->I.DstReg.WriteMask != WRITEMASK_XYZW)) { + struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst); + + inst_mov->I.Opcode = OPCODE_MOV; + inst_mov->I.DstReg = inst->I.DstReg; + inst_mov->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_mov->I.SrcReg[0].Index = rc_find_free_temporary(c); + + inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index; + inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + } + + + /* Cannot read texture coordinate from constants file */ + if (inst->I.SrcReg[0].File != PROGRAM_TEMPORARY && inst->I.SrcReg[0].File != PROGRAM_INPUT) { + struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev); + + inst_mov->I.Opcode = OPCODE_MOV; + inst_mov->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mov->I.DstReg.Index = rc_find_free_temporary(c); + inst_mov->I.SrcReg[0] = inst->I.SrcReg[0]; + + reset_srcreg(&inst->I.SrcReg[0]); + inst->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[0].Index = inst_mov->I.DstReg.Index; + } + + return GL_TRUE; +} + +/* just some random things... */ +void r300FragmentProgramDump(struct rX00_fragment_program_code *c) +{ + struct r300_fragment_program_code *code = &c->code.r300; + int n, i, j; + static int pc = 0; + + fprintf(stderr, "pc=%d*************************************\n", pc++); + + fprintf(stderr, "Hardware program\n"); + fprintf(stderr, "----------------\n"); + + for (n = 0; n <= (code->config & 3); n++) { + uint32_t code_addr = code->code_addr[3 - (code->config & 3) + n]; + int alu_offset = (code_addr & R300_ALU_START_MASK) >> R300_ALU_START_SHIFT; + int alu_end = (code_addr & R300_ALU_SIZE_MASK) >> R300_ALU_SIZE_SHIFT; + int tex_offset = (code_addr & R300_TEX_START_MASK) >> R300_TEX_START_SHIFT; + int tex_end = (code_addr & R300_TEX_SIZE_MASK) >> R300_TEX_SIZE_SHIFT; + + fprintf(stderr, "NODE %d: alu_offset: %d, tex_offset: %d, " + "alu_end: %d, tex_end: %d (code_addr: %08x)\n", n, + alu_offset, tex_offset, alu_end, tex_end, code_addr); + + if (n > 0 || (code->config & R300_PFS_CNTL_FIRST_NODE_HAS_TEX)) { + fprintf(stderr, " TEX:\n"); + for (i = tex_offset; + i <= tex_offset + tex_end; + ++i) { + const char *instr; + + switch ((code->tex. + inst[i] >> R300_TEX_INST_SHIFT) & + 15) { + case R300_TEX_OP_LD: + instr = "TEX"; + break; + case R300_TEX_OP_KIL: + instr = "KIL"; + break; + case R300_TEX_OP_TXP: + instr = "TXP"; + break; + case R300_TEX_OP_TXB: + instr = "TXB"; + break; + default: + instr = "UNKNOWN"; + } + + fprintf(stderr, + " %s t%i, %c%i, texture[%i] (%08x)\n", + instr, + (code->tex. + inst[i] >> R300_DST_ADDR_SHIFT) & 31, + 't', + (code->tex. + inst[i] >> R300_SRC_ADDR_SHIFT) & 31, + (code->tex. + inst[i] & R300_TEX_ID_MASK) >> + R300_TEX_ID_SHIFT, + code->tex.inst[i]); + } + } + + for (i = alu_offset; + i <= alu_offset + alu_end; ++i) { + char srcc[3][10], dstc[20]; + char srca[3][10], dsta[20]; + char argc[3][20]; + char arga[3][20]; + char flags[5], tmp[10]; + + for (j = 0; j < 3; ++j) { + int regc = code->alu.inst[i].rgb_addr >> (j * 6); + int rega = code->alu.inst[i].alpha_addr >> (j * 6); + + sprintf(srcc[j], "%c%i", + (regc & 32) ? 'c' : 't', regc & 31); + sprintf(srca[j], "%c%i", + (rega & 32) ? 'c' : 't', rega & 31); + } + + dstc[0] = 0; + sprintf(flags, "%s%s%s", + (code->alu.inst[i]. + rgb_addr & R300_ALU_DSTC_REG_X) ? "x" : "", + (code->alu.inst[i]. + rgb_addr & R300_ALU_DSTC_REG_Y) ? "y" : "", + (code->alu.inst[i]. + rgb_addr & R300_ALU_DSTC_REG_Z) ? "z" : ""); + if (flags[0] != 0) { + sprintf(dstc, "t%i.%s ", + (code->alu.inst[i]. + rgb_addr >> R300_ALU_DSTC_SHIFT) & 31, + flags); + } + sprintf(flags, "%s%s%s", + (code->alu.inst[i]. + rgb_addr & R300_ALU_DSTC_OUTPUT_X) ? "x" : "", + (code->alu.inst[i]. + rgb_addr & R300_ALU_DSTC_OUTPUT_Y) ? "y" : "", + (code->alu.inst[i]. + rgb_addr & R300_ALU_DSTC_OUTPUT_Z) ? "z" : ""); + if (flags[0] != 0) { + sprintf(tmp, "o%i.%s", + (code->alu.inst[i]. + rgb_addr >> R300_ALU_DSTC_SHIFT) & 31, + flags); + strcat(dstc, tmp); + } + + dsta[0] = 0; + if (code->alu.inst[i].alpha_addr & R300_ALU_DSTA_REG) { + sprintf(dsta, "t%i.w ", + (code->alu.inst[i]. + alpha_addr >> R300_ALU_DSTA_SHIFT) & 31); + } + if (code->alu.inst[i].alpha_addr & R300_ALU_DSTA_OUTPUT) { + sprintf(tmp, "o%i.w ", + (code->alu.inst[i]. + alpha_addr >> R300_ALU_DSTA_SHIFT) & 31); + strcat(dsta, tmp); + } + if (code->alu.inst[i].alpha_addr & R300_ALU_DSTA_DEPTH) { + strcat(dsta, "Z"); + } + + fprintf(stderr, + "%3i: xyz: %3s %3s %3s -> %-20s (%08x)\n" + " w: %3s %3s %3s -> %-20s (%08x)\n", i, + srcc[0], srcc[1], srcc[2], dstc, + code->alu.inst[i].rgb_addr, srca[0], srca[1], + srca[2], dsta, code->alu.inst[i].alpha_addr); + + for (j = 0; j < 3; ++j) { + int regc = code->alu.inst[i].rgb_inst >> (j * 7); + int rega = code->alu.inst[i].alpha_inst >> (j * 7); + int d; + char buf[20]; + + d = regc & 31; + if (d < 12) { + switch (d % 4) { + case R300_ALU_ARGC_SRC0C_XYZ: + sprintf(buf, "%s.xyz", + srcc[d / 4]); + break; + case R300_ALU_ARGC_SRC0C_XXX: + sprintf(buf, "%s.xxx", + srcc[d / 4]); + break; + case R300_ALU_ARGC_SRC0C_YYY: + sprintf(buf, "%s.yyy", + srcc[d / 4]); + break; + case R300_ALU_ARGC_SRC0C_ZZZ: + sprintf(buf, "%s.zzz", + srcc[d / 4]); + break; + } + } else if (d < 15) { + sprintf(buf, "%s.www", srca[d - 12]); + } else if (d == 20) { + sprintf(buf, "0.0"); + } else if (d == 21) { + sprintf(buf, "1.0"); + } else if (d == 22) { + sprintf(buf, "0.5"); + } else if (d >= 23 && d < 32) { + d -= 23; + switch (d / 3) { + case 0: + sprintf(buf, "%s.yzx", + srcc[d % 3]); + break; + case 1: + sprintf(buf, "%s.zxy", + srcc[d % 3]); + break; + case 2: + sprintf(buf, "%s.Wzy", + srcc[d % 3]); + break; + } + } else { + sprintf(buf, "%i", d); + } + + sprintf(argc[j], "%s%s%s%s", + (regc & 32) ? "-" : "", + (regc & 64) ? "|" : "", + buf, (regc & 64) ? "|" : ""); + + d = rega & 31; + if (d < 9) { + sprintf(buf, "%s.%c", srcc[d / 3], + 'x' + (char)(d % 3)); + } else if (d < 12) { + sprintf(buf, "%s.w", srca[d - 9]); + } else if (d == 16) { + sprintf(buf, "0.0"); + } else if (d == 17) { + sprintf(buf, "1.0"); + } else if (d == 18) { + sprintf(buf, "0.5"); + } else { + sprintf(buf, "%i", d); + } + + sprintf(arga[j], "%s%s%s%s", + (rega & 32) ? "-" : "", + (rega & 64) ? "|" : "", + buf, (rega & 64) ? "|" : ""); + } + + fprintf(stderr, " xyz: %8s %8s %8s op: %08x\n" + " w: %8s %8s %8s op: %08x\n", + argc[0], argc[1], argc[2], + code->alu.inst[i].rgb_inst, arga[0], arga[1], + arga[2], code->alu.inst[i].alpha_inst); + } + } +} diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h new file mode 100644 index 0000000000..0ac46dbd9c --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h @@ -0,0 +1,49 @@ +/* + * Copyright (C) 2005 Ben Skeggs. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +/* + * Authors: + * Ben Skeggs <darktama@iinet.net.au> + * Jerome Glisse <j.glisse@gmail.com> + */ +#ifndef __R300_FRAGPROG_H_ +#define __R300_FRAGPROG_H_ + +#include "shader/program.h" +#include "shader/prog_instruction.h" + +#include "radeon_compiler.h" +#include "radeon_program.h" + + +extern void r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler); + +extern void r300FragmentProgramDump(struct rX00_fragment_program_code *c); + +extern GLboolean r300_transform_TEX(struct radeon_compiler * c, struct rc_instruction* inst, void* data); + +#endif diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c index b75656e7ee..305dc074ee 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c @@ -40,57 +40,43 @@ #include "r300_fragprog.h" +#include "../r300_reg.h" + #include "radeon_program_pair.h" #include "r300_fragprog_swizzle.h" -#include "r300_reg.h" +struct r300_emit_state { + struct r300_fragment_program_compiler * compiler; + + unsigned current_node : 2; + unsigned node_first_tex : 8; + unsigned node_first_alu : 8; + uint32_t node_flags; +}; + #define PROG_CODE \ - struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \ - struct r300_fragment_program_code *code = &c->code->r300 + struct r300_emit_state * emit = (struct r300_emit_state*)data; \ + struct r300_fragment_program_compiler *c = emit->compiler; \ + struct r300_fragment_program_code *code = &c->code->code.r300 #define error(fmt, args...) do { \ - fprintf(stderr, "%s::%s(): " fmt "\n", \ + rc_error(&c->Base, "%s::%s(): " fmt "\n", \ __FILE__, __FUNCTION__, ##args); \ } while(0) -static GLboolean emit_const(void* data, GLuint file, GLuint index, GLuint *hwindex) -{ - PROG_CODE; - - for (*hwindex = 0; *hwindex < code->const_nr; ++*hwindex) { - if (code->constant[*hwindex].File == file && - code->constant[*hwindex].Index == index) - break; - } - - if (*hwindex >= code->const_nr) { - if (*hwindex >= R300_PFS_NUM_CONST_REGS) { - error("Out of hw constants!\n"); - return GL_FALSE; - } - - code->const_nr++; - code->constant[*hwindex].File = file; - code->constant[*hwindex].Index = index; - } - - return GL_TRUE; -} - - /** * Mark a temporary register as used. */ static void use_temporary(struct r300_fragment_program_code *code, GLuint index) { - if (index > code->max_temp_idx) - code->max_temp_idx = index; + if (index > code->pixsize) + code->pixsize = index; } -static GLuint translate_rgb_opcode(GLuint opcode) +static GLuint translate_rgb_opcode(struct r300_fragment_program_compiler * c, GLuint opcode) { switch(opcode) { case OPCODE_CMP: return R300_ALU_OUTC_CMP; @@ -109,7 +95,7 @@ static GLuint translate_rgb_opcode(GLuint opcode) } } -static GLuint translate_alpha_opcode(GLuint opcode) +static GLuint translate_alpha_opcode(struct r300_fragment_program_compiler * c, GLuint opcode) { switch(opcode) { case OPCODE_CMP: return R300_ALU_OUTA_CMP; @@ -145,63 +131,62 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) int ip = code->alu.length++; int j; - code->node[code->cur_node].alu_end++; - code->alu.inst[ip].inst0 = translate_rgb_opcode(inst->RGB.Opcode); - code->alu.inst[ip].inst2 = translate_alpha_opcode(inst->Alpha.Opcode); + code->alu.inst[ip].rgb_inst = translate_rgb_opcode(c, inst->RGB.Opcode); + code->alu.inst[ip].alpha_inst = translate_alpha_opcode(c, inst->Alpha.Opcode); for(j = 0; j < 3; ++j) { GLuint src = inst->RGB.Src[j].Index | (inst->RGB.Src[j].Constant << 5); if (!inst->RGB.Src[j].Constant) use_temporary(code, inst->RGB.Src[j].Index); - code->alu.inst[ip].inst1 |= src << (6*j); + code->alu.inst[ip].rgb_addr |= src << (6*j); src = inst->Alpha.Src[j].Index | (inst->Alpha.Src[j].Constant << 5); if (!inst->Alpha.Src[j].Constant) use_temporary(code, inst->Alpha.Src[j].Index); - code->alu.inst[ip].inst3 |= src << (6*j); + code->alu.inst[ip].alpha_addr |= src << (6*j); GLuint arg = r300FPTranslateRGBSwizzle(inst->RGB.Arg[j].Source, inst->RGB.Arg[j].Swizzle); arg |= inst->RGB.Arg[j].Abs << 6; arg |= inst->RGB.Arg[j].Negate << 5; - code->alu.inst[ip].inst0 |= arg << (7*j); + code->alu.inst[ip].rgb_inst |= arg << (7*j); arg = r300FPTranslateAlphaSwizzle(inst->Alpha.Arg[j].Source, inst->Alpha.Arg[j].Swizzle); arg |= inst->Alpha.Arg[j].Abs << 6; arg |= inst->Alpha.Arg[j].Negate << 5; - code->alu.inst[ip].inst2 |= arg << (7*j); + code->alu.inst[ip].alpha_inst |= arg << (7*j); } if (inst->RGB.Saturate) - code->alu.inst[ip].inst0 |= R300_ALU_OUTC_CLAMP; + code->alu.inst[ip].rgb_inst |= R300_ALU_OUTC_CLAMP; if (inst->Alpha.Saturate) - code->alu.inst[ip].inst2 |= R300_ALU_OUTA_CLAMP; + code->alu.inst[ip].alpha_inst |= R300_ALU_OUTA_CLAMP; if (inst->RGB.WriteMask) { use_temporary(code, inst->RGB.DestIndex); - code->alu.inst[ip].inst1 |= + code->alu.inst[ip].rgb_addr |= (inst->RGB.DestIndex << R300_ALU_DSTC_SHIFT) | (inst->RGB.WriteMask << R300_ALU_DSTC_REG_MASK_SHIFT); } if (inst->RGB.OutputWriteMask) { - code->alu.inst[ip].inst1 |= (inst->RGB.OutputWriteMask << R300_ALU_DSTC_OUTPUT_MASK_SHIFT); - code->node[code->cur_node].flags |= R300_RGBA_OUT; + code->alu.inst[ip].rgb_addr |= (inst->RGB.OutputWriteMask << R300_ALU_DSTC_OUTPUT_MASK_SHIFT); + emit->node_flags |= R300_RGBA_OUT; } if (inst->Alpha.WriteMask) { use_temporary(code, inst->Alpha.DestIndex); - code->alu.inst[ip].inst3 |= + code->alu.inst[ip].alpha_addr |= (inst->Alpha.DestIndex << R300_ALU_DSTA_SHIFT) | R300_ALU_DSTA_REG; } if (inst->Alpha.OutputWriteMask) { - code->alu.inst[ip].inst3 |= R300_ALU_DSTA_OUTPUT; - code->node[code->cur_node].flags |= R300_RGBA_OUT; + code->alu.inst[ip].alpha_addr |= R300_ALU_DSTA_OUTPUT; + emit->node_flags |= R300_RGBA_OUT; } if (inst->Alpha.DepthWriteMask) { - code->alu.inst[ip].inst3 |= R300_ALU_DSTA_DEPTH; - code->node[code->cur_node].flags |= R300_W_OUT; - c->fp->writes_depth = GL_TRUE; + code->alu.inst[ip].alpha_addr |= R300_ALU_DSTA_DEPTH; + emit->node_flags |= R300_W_OUT; + c->code->writes_depth = GL_TRUE; } return GL_TRUE; @@ -211,31 +196,50 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) /** * Finish the current node without advancing to the next one. */ -static GLboolean finish_node(struct r300_fragment_program_compiler *c) +static GLboolean finish_node(struct r300_emit_state * emit) { - struct r300_fragment_program_code *code = &c->code->r300; - struct r300_fragment_program_node *node = &code->node[code->cur_node]; + struct r300_fragment_program_compiler * c = emit->compiler; + struct r300_fragment_program_code *code = &emit->compiler->code->code.r300; - if (node->alu_end < 0) { + if (code->alu.length == emit->node_first_alu) { /* Generate a single NOP for this node */ struct radeon_pair_instruction inst; _mesa_bzero(&inst, sizeof(inst)); - if (!emit_alu(c, &inst)) + if (!emit_alu(emit, &inst)) return GL_FALSE; } - if (node->tex_end < 0) { - if (code->cur_node == 0) { - node->tex_end = 0; - } else { - error("Node %i has no TEX instructions", code->cur_node); + unsigned alu_offset = emit->node_first_alu; + unsigned alu_end = code->alu.length - alu_offset - 1; + unsigned tex_offset = emit->node_first_tex; + unsigned tex_end = code->tex.length - tex_offset - 1; + + if (code->tex.length == emit->node_first_tex) { + if (emit->current_node > 0) { + error("Node %i has no TEX instructions", emit->current_node); return GL_FALSE; } + + tex_end = 0; } else { - if (code->cur_node == 0) - code->first_node_has_tex = 1; + if (emit->current_node == 0) + code->config |= R300_PFS_CNTL_FIRST_NODE_HAS_TEX; } + /* Write the config register. + * Note: The order in which the words for each node are written + * is not correct here and needs to be fixed up once we're entirely + * done + * + * Also note that the register specification from AMD is slightly + * incorrect in its description of this register. */ + code->code_addr[emit->current_node] = + (alu_offset << R300_ALU_START_SHIFT) | + (alu_end << R300_ALU_SIZE_SHIFT) | + (tex_offset << R300_TEX_START_SHIFT) | + (tex_end << R300_TEX_SIZE_SHIFT) | + emit->node_flags; + return GL_TRUE; } @@ -248,30 +252,28 @@ static GLboolean begin_tex(void* data) { PROG_CODE; - if (code->cur_node == 0) { - if (code->node[0].alu_end < 0 && - code->node[0].tex_end < 0) - return GL_TRUE; + if (code->alu.length == emit->node_first_alu && + code->tex.length == emit->node_first_tex) { + return GL_TRUE; } - if (code->cur_node == 3) { + if (emit->current_node == 3) { error("Too many texture indirections"); return GL_FALSE; } - if (!finish_node(c)) + if (!finish_node(emit)) return GL_FALSE; - struct r300_fragment_program_node *node = &code->node[++code->cur_node]; - node->alu_offset = code->alu.length; - node->alu_end = -1; - node->tex_offset = code->tex.length; - node->tex_end = -1; + emit->current_node++; + emit->node_first_tex = code->tex.length; + emit->node_first_alu = code->alu.length; + emit->node_flags = 0; return GL_TRUE; } -static GLboolean emit_tex(void* data, struct prog_instruction* inst) +static GLboolean emit_tex(void* data, struct radeon_pair_texture_instruction* inst) { PROG_CODE; @@ -281,31 +283,30 @@ static GLboolean emit_tex(void* data, struct prog_instruction* inst) } GLuint unit = inst->TexSrcUnit; - GLuint dest = inst->DstReg.Index; + GLuint dest = inst->DestIndex; GLuint opcode; switch(inst->Opcode) { - case OPCODE_KIL: opcode = R300_TEX_OP_KIL; break; - case OPCODE_TEX: opcode = R300_TEX_OP_LD; break; - case OPCODE_TXB: opcode = R300_TEX_OP_TXB; break; - case OPCODE_TXP: opcode = R300_TEX_OP_TXP; break; + case RADEON_OPCODE_KIL: opcode = R300_TEX_OP_KIL; break; + case RADEON_OPCODE_TEX: opcode = R300_TEX_OP_LD; break; + case RADEON_OPCODE_TXB: opcode = R300_TEX_OP_TXB; break; + case RADEON_OPCODE_TXP: opcode = R300_TEX_OP_TXP; break; default: error("Unknown texture opcode %i", inst->Opcode); return GL_FALSE; } - if (inst->Opcode == OPCODE_KIL) { + if (inst->Opcode == RADEON_OPCODE_KIL) { unit = 0; dest = 0; } else { use_temporary(code, dest); } - use_temporary(code, inst->SrcReg[0].Index); + use_temporary(code, inst->SrcIndex); - code->node[code->cur_node].tex_end++; code->tex.inst[code->tex.length++] = - (inst->SrcReg[0].Index << R300_SRC_ADDR_SHIFT) | + (inst->SrcIndex << R300_SRC_ADDR_SHIFT) | (dest << R300_DST_ADDR_SHIFT) | (unit << R300_TEX_ID_SHIFT) | (opcode << R300_TEX_INST_SHIFT); @@ -314,7 +315,6 @@ static GLboolean emit_tex(void* data, struct prog_instruction* inst) static const struct radeon_pair_handler pair_handler = { - .EmitConst = &emit_const, .EmitPaired = &emit_alu, .EmitTex = &emit_tex, .BeginTexBlock = &begin_tex, @@ -325,20 +325,36 @@ static const struct radeon_pair_handler pair_handler = { * Final compilation step: Turn the intermediate radeon_program into * machine-readable instructions. */ -GLboolean r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler) +void r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler) { - struct r300_fragment_program_code *code = &compiler->code->r300; - - _mesa_bzero(code, sizeof(struct r300_fragment_program_code)); - code->node[0].alu_end = -1; - code->node[0].tex_end = -1; + struct r300_emit_state emit; + struct r300_fragment_program_code *code = &compiler->code->code.r300; - if (!radeonPairProgram(compiler->r300->radeon.glCtx, compiler->program, &pair_handler, compiler)) - return GL_FALSE; + memset(&emit, 0, sizeof(emit)); + emit.compiler = compiler; - if (!finish_node(compiler)) - return GL_FALSE; + _mesa_bzero(code, sizeof(struct r300_fragment_program_code)); - return GL_TRUE; + radeonPairProgram(compiler, &pair_handler, &emit); + if (compiler->Base.Error) + return; + + /* Finish the program */ + finish_node(&emit); + + code->config |= emit.current_node; /* FIRST_NODE_HAS_TEX set by finish_node */ + code->code_offset = + (0 << R300_PFS_CNTL_ALU_OFFSET_SHIFT) | + ((code->alu.length-1) << R300_PFS_CNTL_ALU_END_SHIFT) | + (0 << R300_PFS_CNTL_TEX_OFFSET_SHIFT) | + ((code->tex.length ? code->tex.length-1 : 0) << R300_PFS_CNTL_TEX_END_SHIFT); + + if (emit.current_node < 3) { + int shift = 3 - emit.current_node; + int i; + for(i = 0; i <= emit.current_node; ++i) + code->code_addr[shift + i] = code->code_addr[i]; + for(i = 0; i < shift; ++i) + code->code_addr[i] = 0; + } } - diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c index fc9d855bce..1b14cc3888 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.c +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c @@ -33,8 +33,9 @@ #include "r300_fragprog_swizzle.h" -#include "r300_reg.h" +#include "../r300_reg.h" #include "radeon_nqssadce.h" +#include "radeon_compiler.h" #define MAKE_SWZ3(x, y, z) (MAKE_SWIZZLE4(SWIZZLE_##x, SWIZZLE_##y, SWIZZLE_##z, SWIZZLE_ZERO)) @@ -174,18 +175,15 @@ void r300FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, } } - struct prog_instruction *inst; - - _mesa_insert_instructions(s->Program, s->IP, 1); - inst = s->Program->Instructions + s->IP++; - inst->Opcode = OPCODE_MOV; - inst->DstReg = dst; - inst->DstReg.WriteMask &= (best_matchmask | WRITEMASK_W); - inst->SrcReg[0] = src; - inst->SrcReg[0].Negate = (best_matchmask & src.Negate) ? NEGATE_XYZW : NEGATE_NONE; + struct rc_instruction *inst = rc_insert_new_instruction(s->Compiler, s->IP->Prev); + inst->I.Opcode = OPCODE_MOV; + inst->I.DstReg = dst; + inst->I.DstReg.WriteMask &= (best_matchmask | WRITEMASK_W); + inst->I.SrcReg[0] = src; + inst->I.SrcReg[0].Negate = (best_matchmask & src.Negate) ? NEGATE_XYZW : NEGATE_NONE; /* Note: We rely on NqSSA/DCE to set unused swizzle components to NIL */ - dst.WriteMask &= ~inst->DstReg.WriteMask; + dst.WriteMask &= ~inst->I.DstReg.WriteMask; } } diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.h b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.h index 231bf4eef5..231bf4eef5 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.h +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.h diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c new file mode 100644 index 0000000000..d39b82be71 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c @@ -0,0 +1,149 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "radeon_compiler.h" + +#include "shader/prog_parameter.h" +#include "shader/prog_print.h" +#include "shader/prog_statevars.h" + +#include "radeon_nqssadce.h" +#include "radeon_program_alu.h" +#include "r300_fragprog.h" +#include "r300_fragprog_swizzle.h" +#include "r500_fragprog.h" + + +static void nqssadce_init(struct nqssadce_state* s) +{ + struct r300_fragment_program_compiler * c = s->UserData; + s->Outputs[c->OutputColor].Sourced = WRITEMASK_XYZW; + s->Outputs[c->OutputDepth].Sourced = WRITEMASK_W; +} + +static void rewrite_depth_out(struct r300_fragment_program_compiler * c) +{ + struct rc_instruction *rci; + + for (rci = c->Base.Program.Instructions.Next; rci != &c->Base.Program.Instructions; rci = rci->Next) { + struct prog_instruction * inst = &rci->I; + + if (inst->DstReg.File != PROGRAM_OUTPUT || inst->DstReg.Index != c->OutputDepth) + continue; + + if (inst->DstReg.WriteMask & WRITEMASK_Z) { + inst->DstReg.WriteMask = WRITEMASK_W; + } else { + inst->DstReg.WriteMask = 0; + continue; + } + + switch (inst->Opcode) { + case OPCODE_FRC: + case OPCODE_MOV: + inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); + break; + case OPCODE_ADD: + case OPCODE_MAX: + case OPCODE_MIN: + case OPCODE_MUL: + inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); + inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]); + break; + case OPCODE_CMP: + case OPCODE_MAD: + inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); + inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]); + inst->SrcReg[2] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[2]); + break; + default: + // Scalar instructions needn't be reswizzled + break; + } + } +} + +void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c) +{ + rewrite_depth_out(c); + + if (c->is_r500) { + struct radeon_program_transformation transformations[] = { + { &r500_transform_TEX, c }, + { &radeonTransformALU, 0 }, + { &radeonTransformDeriv, 0 }, + { &radeonTransformTrigScale, 0 } + }; + radeonLocalTransform(&c->Base, 4, transformations); + } else { + struct radeon_program_transformation transformations[] = { + { &r300_transform_TEX, c }, + { &radeonTransformALU, 0 }, + { &radeonTransformTrigSimple, 0 } + }; + radeonLocalTransform(&c->Base, 3, transformations); + } + + if (c->Base.Debug) { + _mesa_printf("Fragment Program: After native rewrite:\n"); + rc_print_program(&c->Base.Program); + fflush(stdout); + } + + if (c->is_r500) { + struct radeon_nqssadce_descr nqssadce = { + .Init = &nqssadce_init, + .IsNativeSwizzle = &r500FPIsNativeSwizzle, + .BuildSwizzle = &r500FPBuildSwizzle + }; + radeonNqssaDce(&c->Base, &nqssadce, c); + } else { + struct radeon_nqssadce_descr nqssadce = { + .Init = &nqssadce_init, + .IsNativeSwizzle = &r300FPIsNativeSwizzle, + .BuildSwizzle = &r300FPBuildSwizzle + }; + radeonNqssaDce(&c->Base, &nqssadce, c); + } + + if (c->Base.Debug) { + _mesa_printf("Compiler: after NqSSA-DCE:\n"); + rc_print_program(&c->Base.Program); + fflush(stdout); + } + + if (c->is_r500) { + r500BuildFragmentProgramHwCode(c); + } else { + r300BuildFragmentProgramHwCode(c); + } + + rc_constants_copy(&c->code->constants, &c->Base.Program.Constants); + + if (c->Base.Debug) { + if (c->is_r500) { + r500FragmentProgramDump(c->code); + } else { + r300FragmentProgramDump(c->code); + } + } +} diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c new file mode 100644 index 0000000000..fc9c8f805a --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c @@ -0,0 +1,615 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "radeon_compiler.h" + +#include "../r300_reg.h" + +#include "radeon_nqssadce.h" +#include "radeon_program.h" +#include "radeon_program_alu.h" + +#include "shader/prog_print.h" + + +/* + * Take an already-setup and valid source then swizzle it appropriately to + * obtain a constant ZERO or ONE source. + */ +#define __CONST(x, y) \ + (PVS_SRC_OPERAND(t_src_index(vp, &vpi->SrcReg[x]), \ + t_swizzle(y), \ + t_swizzle(y), \ + t_swizzle(y), \ + t_swizzle(y), \ + t_src_class(vpi->SrcReg[x].File), \ + NEGATE_NONE) | (vpi->SrcReg[x].RelAddr << 4)) + + +static unsigned long t_dst_mask(GLuint mask) +{ + /* WRITEMASK_* is equivalent to VSF_FLAG_* */ + return mask & WRITEMASK_XYZW; +} + +static unsigned long t_dst_class(gl_register_file file) +{ + + switch (file) { + case PROGRAM_TEMPORARY: + return PVS_DST_REG_TEMPORARY; + case PROGRAM_OUTPUT: + return PVS_DST_REG_OUT; + case PROGRAM_ADDRESS: + return PVS_DST_REG_A0; + /* + case PROGRAM_INPUT: + case PROGRAM_LOCAL_PARAM: + case PROGRAM_ENV_PARAM: + case PROGRAM_NAMED_PARAM: + case PROGRAM_STATE_VAR: + case PROGRAM_WRITE_ONLY: + case PROGRAM_ADDRESS: + */ + default: + fprintf(stderr, "problem in %s", __FUNCTION__); + _mesa_exit(-1); + return -1; + } +} + +static unsigned long t_dst_index(struct r300_vertex_program_code *vp, + struct prog_dst_register *dst) +{ + if (dst->File == PROGRAM_OUTPUT) + return vp->outputs[dst->Index]; + + return dst->Index; +} + +static unsigned long t_src_class(gl_register_file file) +{ + switch (file) { + case PROGRAM_TEMPORARY: + return PVS_SRC_REG_TEMPORARY; + case PROGRAM_INPUT: + return PVS_SRC_REG_INPUT; + case PROGRAM_LOCAL_PARAM: + case PROGRAM_ENV_PARAM: + case PROGRAM_NAMED_PARAM: + case PROGRAM_CONSTANT: + case PROGRAM_STATE_VAR: + return PVS_SRC_REG_CONSTANT; + /* + case PROGRAM_OUTPUT: + case PROGRAM_WRITE_ONLY: + case PROGRAM_ADDRESS: + */ + default: + fprintf(stderr, "problem in %s", __FUNCTION__); + _mesa_exit(-1); + return -1; + } +} + +static GLboolean t_src_conflict(struct prog_src_register a, struct prog_src_register b) +{ + unsigned long aclass = t_src_class(a.File); + unsigned long bclass = t_src_class(b.File); + + if (aclass != bclass) + return GL_FALSE; + if (aclass == PVS_SRC_REG_TEMPORARY) + return GL_FALSE; + + if (a.RelAddr || b.RelAddr) + return GL_TRUE; + if (a.Index != b.Index) + return GL_TRUE; + + return GL_FALSE; +} + +static INLINE unsigned long t_swizzle(GLubyte swizzle) +{ + /* this is in fact a NOP as the Mesa SWIZZLE_* are all identical to VSF_IN_COMPONENT_* */ + return swizzle; +} + +static unsigned long t_src_index(struct r300_vertex_program_code *vp, + struct prog_src_register *src) +{ + if (src->File == PROGRAM_INPUT) { + assert(vp->inputs[src->Index] != -1); + return vp->inputs[src->Index]; + } else { + if (src->Index < 0) { + fprintf(stderr, + "negative offsets for indirect addressing do not work.\n"); + return 0; + } + return src->Index; + } +} + +/* these two functions should probably be merged... */ + +static unsigned long t_src(struct r300_vertex_program_code *vp, + struct prog_src_register *src) +{ + /* src->Negate uses the NEGATE_ flags from program_instruction.h, + * which equal our VSF_FLAGS_ values, so it's safe to just pass it here. + */ + return PVS_SRC_OPERAND(t_src_index(vp, src), + t_swizzle(GET_SWZ(src->Swizzle, 0)), + t_swizzle(GET_SWZ(src->Swizzle, 1)), + t_swizzle(GET_SWZ(src->Swizzle, 2)), + t_swizzle(GET_SWZ(src->Swizzle, 3)), + t_src_class(src->File), + src->Negate) | (src->RelAddr << 4); +} + +static unsigned long t_src_scalar(struct r300_vertex_program_code *vp, + struct prog_src_register *src) +{ + /* src->Negate uses the NEGATE_ flags from program_instruction.h, + * which equal our VSF_FLAGS_ values, so it's safe to just pass it here. + */ + return PVS_SRC_OPERAND(t_src_index(vp, src), + t_swizzle(GET_SWZ(src->Swizzle, 0)), + t_swizzle(GET_SWZ(src->Swizzle, 0)), + t_swizzle(GET_SWZ(src->Swizzle, 0)), + t_swizzle(GET_SWZ(src->Swizzle, 0)), + t_src_class(src->File), + src->Negate ? NEGATE_XYZW : NEGATE_NONE) | + (src->RelAddr << 4); +} + +static GLboolean valid_dst(struct r300_vertex_program_code *vp, + struct prog_dst_register *dst) +{ + if (dst->File == PROGRAM_OUTPUT && vp->outputs[dst->Index] == -1) { + return GL_FALSE; + } else if (dst->File == PROGRAM_ADDRESS) { + assert(dst->Index == 0); + } + + return GL_TRUE; +} + +static void ei_vector1(struct r300_vertex_program_code *vp, + GLuint hw_opcode, + struct prog_instruction *vpi, + GLuint * inst) +{ + inst[0] = PVS_OP_DST_OPERAND(hw_opcode, + GL_FALSE, + GL_FALSE, + t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + inst[1] = t_src(vp, &vpi->SrcReg[0]); + inst[2] = __CONST(0, SWIZZLE_ZERO); + inst[3] = __CONST(0, SWIZZLE_ZERO); +} + +static void ei_vector2(struct r300_vertex_program_code *vp, + GLuint hw_opcode, + struct prog_instruction *vpi, + GLuint * inst) +{ + inst[0] = PVS_OP_DST_OPERAND(hw_opcode, + GL_FALSE, + GL_FALSE, + t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + inst[1] = t_src(vp, &vpi->SrcReg[0]); + inst[2] = t_src(vp, &vpi->SrcReg[1]); + inst[3] = __CONST(1, SWIZZLE_ZERO); +} + +static void ei_math1(struct r300_vertex_program_code *vp, + GLuint hw_opcode, + struct prog_instruction *vpi, + GLuint * inst) +{ + inst[0] = PVS_OP_DST_OPERAND(hw_opcode, + GL_TRUE, + GL_FALSE, + t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + inst[1] = t_src_scalar(vp, &vpi->SrcReg[0]); + inst[2] = __CONST(0, SWIZZLE_ZERO); + inst[3] = __CONST(0, SWIZZLE_ZERO); +} + +static void ei_lit(struct r300_vertex_program_code *vp, + struct prog_instruction *vpi, + GLuint * inst) +{ + //LIT TMP 1.Y Z TMP 1{} {X W Z Y} TMP 1{} {Y W Z X} TMP 1{} {Y X Z W} + + inst[0] = PVS_OP_DST_OPERAND(ME_LIGHT_COEFF_DX, + GL_TRUE, + GL_FALSE, + t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + /* NOTE: Users swizzling might not work. */ + inst[1] = PVS_SRC_OPERAND(t_src_index(vp, &vpi->SrcReg[0]), t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 0)), // X + t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 3)), // W + PVS_SRC_SELECT_FORCE_0, // Z + t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 1)), // Y + t_src_class(vpi->SrcReg[0].File), + vpi->SrcReg[0].Negate ? NEGATE_XYZW : NEGATE_NONE) | + (vpi->SrcReg[0].RelAddr << 4); + inst[2] = PVS_SRC_OPERAND(t_src_index(vp, &vpi->SrcReg[0]), t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 1)), // Y + t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 3)), // W + PVS_SRC_SELECT_FORCE_0, // Z + t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 0)), // X + t_src_class(vpi->SrcReg[0].File), + vpi->SrcReg[0].Negate ? NEGATE_XYZW : NEGATE_NONE) | + (vpi->SrcReg[0].RelAddr << 4); + inst[3] = PVS_SRC_OPERAND(t_src_index(vp, &vpi->SrcReg[0]), t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 1)), // Y + t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 0)), // X + PVS_SRC_SELECT_FORCE_0, // Z + t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 3)), // W + t_src_class(vpi->SrcReg[0].File), + vpi->SrcReg[0].Negate ? NEGATE_XYZW : NEGATE_NONE) | + (vpi->SrcReg[0].RelAddr << 4); +} + +static void ei_mad(struct r300_vertex_program_code *vp, + struct prog_instruction *vpi, + GLuint * inst) +{ + inst[0] = PVS_OP_DST_OPERAND(PVS_MACRO_OP_2CLK_MADD, + GL_FALSE, + GL_TRUE, + t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + inst[1] = t_src(vp, &vpi->SrcReg[0]); + inst[2] = t_src(vp, &vpi->SrcReg[1]); + inst[3] = t_src(vp, &vpi->SrcReg[2]); +} + +static void ei_pow(struct r300_vertex_program_code *vp, + struct prog_instruction *vpi, + GLuint * inst) +{ + inst[0] = PVS_OP_DST_OPERAND(ME_POWER_FUNC_FF, + GL_TRUE, + GL_FALSE, + t_dst_index(vp, &vpi->DstReg), + t_dst_mask(vpi->DstReg.WriteMask), + t_dst_class(vpi->DstReg.File)); + inst[1] = t_src_scalar(vp, &vpi->SrcReg[0]); + inst[2] = __CONST(0, SWIZZLE_ZERO); + inst[3] = t_src_scalar(vp, &vpi->SrcReg[1]); +} + + +static void translate_vertex_program(struct r300_vertex_program_compiler * compiler) +{ + struct rc_instruction *rci; + + compiler->code->pos_end = 0; /* Not supported yet */ + compiler->code->length = 0; + + compiler->SetHwInputOutput(compiler); + + for(rci = compiler->Base.Program.Instructions.Next; rci != &compiler->Base.Program.Instructions; rci = rci->Next) { + struct prog_instruction *vpi = &rci->I; + GLuint *inst = compiler->code->body.d + compiler->code->length; + + /* Skip instructions writing to non-existing destination */ + if (!valid_dst(compiler->code, &vpi->DstReg)) + continue; + + if (compiler->code->length >= VSF_MAX_FRAGMENT_LENGTH) { + rc_error(&compiler->Base, "Vertex program has too many instructions\n"); + return; + } + + switch (vpi->Opcode) { + case OPCODE_ADD: ei_vector2(compiler->code, VE_ADD, vpi, inst); break; + case OPCODE_ARL: ei_vector1(compiler->code, VE_FLT2FIX_DX, vpi, inst); break; + case OPCODE_DP4: ei_vector2(compiler->code, VE_DOT_PRODUCT, vpi, inst); break; + case OPCODE_DST: ei_vector2(compiler->code, VE_DISTANCE_VECTOR, vpi, inst); break; + case OPCODE_EX2: ei_math1(compiler->code, ME_EXP_BASE2_FULL_DX, vpi, inst); break; + case OPCODE_EXP: ei_math1(compiler->code, ME_EXP_BASE2_DX, vpi, inst); break; + case OPCODE_FRC: ei_vector1(compiler->code, VE_FRACTION, vpi, inst); break; + case OPCODE_LG2: ei_math1(compiler->code, ME_LOG_BASE2_FULL_DX, vpi, inst); break; + case OPCODE_LIT: ei_lit(compiler->code, vpi, inst); break; + case OPCODE_LOG: ei_math1(compiler->code, ME_LOG_BASE2_DX, vpi, inst); break; + case OPCODE_MAD: ei_mad(compiler->code, vpi, inst); break; + case OPCODE_MAX: ei_vector2(compiler->code, VE_MAXIMUM, vpi, inst); break; + case OPCODE_MIN: ei_vector2(compiler->code, VE_MINIMUM, vpi, inst); break; + case OPCODE_MOV: ei_vector1(compiler->code, VE_ADD, vpi, inst); break; + case OPCODE_MUL: ei_vector2(compiler->code, VE_MULTIPLY, vpi, inst); break; + case OPCODE_POW: ei_pow(compiler->code, vpi, inst); break; + case OPCODE_RCP: ei_math1(compiler->code, ME_RECIP_DX, vpi, inst); break; + case OPCODE_RSQ: ei_math1(compiler->code, ME_RECIP_SQRT_DX, vpi, inst); break; + case OPCODE_SGE: ei_vector2(compiler->code, VE_SET_GREATER_THAN_EQUAL, vpi, inst); break; + case OPCODE_SLT: ei_vector2(compiler->code, VE_SET_LESS_THAN, vpi, inst); break; + default: + rc_error(&compiler->Base, "Unknown opcode %i\n", vpi->Opcode); + return; + } + + compiler->code->length += 4; + + if (compiler->Base.Error) + return; + } +} + +struct temporary_allocation { + GLuint Allocated:1; + GLuint HwTemp:15; + struct rc_instruction * LastRead; +}; + +static void allocate_temporary_registers(struct r300_vertex_program_compiler * compiler) +{ + struct rc_instruction *inst; + GLuint num_orig_temps = 0; + GLboolean hwtemps[VSF_MAX_FRAGMENT_TEMPS]; + struct temporary_allocation * ta; + GLuint i, j; + + compiler->code->num_temporaries = 0; + memset(hwtemps, 0, sizeof(hwtemps)); + + /* Pass 1: Count original temporaries and allocate structures */ + for(inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) { + GLuint numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode); + GLuint numdsts = _mesa_num_inst_dst_regs(inst->I.Opcode); + + for (i = 0; i < numsrcs; ++i) { + if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY) { + if (inst->I.SrcReg[i].Index >= num_orig_temps) + num_orig_temps = inst->I.SrcReg[i].Index + 1; + } + } + + if (numdsts) { + if (inst->I.DstReg.File == PROGRAM_TEMPORARY) { + if (inst->I.DstReg.Index >= num_orig_temps) + num_orig_temps = inst->I.DstReg.Index + 1; + } + } + } + + ta = (struct temporary_allocation*)memory_pool_malloc(&compiler->Base.Pool, + sizeof(struct temporary_allocation) * num_orig_temps); + memset(ta, 0, sizeof(struct temporary_allocation) * num_orig_temps); + + /* Pass 2: Determine original temporary lifetimes */ + for(inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) { + GLuint numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode); + + for (i = 0; i < numsrcs; ++i) { + if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY) + ta[inst->I.SrcReg[i].Index].LastRead = inst; + } + } + + /* Pass 3: Register allocation */ + for(inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) { + GLuint numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode); + GLuint numdsts = _mesa_num_inst_dst_regs(inst->I.Opcode); + + for (i = 0; i < numsrcs; ++i) { + if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY) { + GLuint orig = inst->I.SrcReg[i].Index; + inst->I.SrcReg[i].Index = ta[orig].HwTemp; + + if (ta[orig].Allocated && inst == ta[orig].LastRead) + hwtemps[ta[orig].HwTemp] = GL_FALSE; + } + } + + if (numdsts) { + if (inst->I.DstReg.File == PROGRAM_TEMPORARY) { + GLuint orig = inst->I.DstReg.Index; + + if (!ta[orig].Allocated) { + for(j = 0; j < VSF_MAX_FRAGMENT_TEMPS; ++j) { + if (!hwtemps[j]) + break; + } + if (j >= VSF_MAX_FRAGMENT_TEMPS) { + fprintf(stderr, "Out of hw temporaries\n"); + } else { + ta[orig].Allocated = GL_TRUE; + ta[orig].HwTemp = j; + hwtemps[j] = GL_TRUE; + + if (j >= compiler->code->num_temporaries) + compiler->code->num_temporaries = j + 1; + } + } + + inst->I.DstReg.Index = ta[orig].HwTemp; + } + } + } +} + + +/** + * Vertex engine cannot read two inputs or two constants at the same time. + * Introduce intermediate MOVs to temporary registers to account for this. + */ +static GLboolean transform_source_conflicts( + struct radeon_compiler *c, + struct rc_instruction* inst, + void* unused) +{ + GLuint num_operands = _mesa_num_inst_src_regs(inst->I.Opcode); + + if (num_operands == 3) { + if (t_src_conflict(inst->I.SrcReg[1], inst->I.SrcReg[2]) + || t_src_conflict(inst->I.SrcReg[0], inst->I.SrcReg[2])) { + int tmpreg = rc_find_free_temporary(c); + struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev); + inst_mov->I.Opcode = OPCODE_MOV; + inst_mov->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mov->I.DstReg.Index = tmpreg; + inst_mov->I.SrcReg[0] = inst->I.SrcReg[2]; + + reset_srcreg(&inst->I.SrcReg[2]); + inst->I.SrcReg[2].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[2].Index = tmpreg; + } + } + + if (num_operands >= 2) { + if (t_src_conflict(inst->I.SrcReg[1], inst->I.SrcReg[0])) { + int tmpreg = rc_find_free_temporary(c); + struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev); + inst_mov->I.Opcode = OPCODE_MOV; + inst_mov->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mov->I.DstReg.Index = tmpreg; + inst_mov->I.SrcReg[0] = inst->I.SrcReg[1]; + + reset_srcreg(&inst->I.SrcReg[1]); + inst->I.SrcReg[1].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[1].Index = tmpreg; + } + } + + return GL_TRUE; +} + +static void addArtificialOutputs(struct r300_vertex_program_compiler * compiler) +{ + int i; + + for(i = 0; i < 32; ++i) { + if ((compiler->RequiredOutputs & (1 << i)) && + !(compiler->Base.Program.OutputsWritten & (1 << i))) { + struct rc_instruction * inst = rc_insert_new_instruction(&compiler->Base, compiler->Base.Program.Instructions.Prev); + inst->I.Opcode = OPCODE_MOV; + + inst->I.DstReg.File = PROGRAM_OUTPUT; + inst->I.DstReg.Index = i; + inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + + inst->I.SrcReg[0].File = PROGRAM_CONSTANT; + inst->I.SrcReg[0].Index = 0; + inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW; + + compiler->Base.Program.OutputsWritten |= 1 << i; + } + } +} + +static void nqssadceInit(struct nqssadce_state* s) +{ + struct r300_vertex_program_compiler * compiler = s->UserData; + int i; + + for(i = 0; i < VERT_RESULT_MAX; ++i) { + if (compiler->RequiredOutputs & (1 << i)) + s->Outputs[i].Sourced = WRITEMASK_XYZW; + } +} + +static GLboolean swizzleIsNative(GLuint opcode, struct prog_src_register reg) +{ + (void) opcode; + (void) reg; + + return GL_TRUE; +} + + + +void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler) +{ + addArtificialOutputs(compiler); + + { + struct radeon_program_transformation transformations[] = { + { &r300_transform_vertex_alu, 0 }, + }; + radeonLocalTransform(&compiler->Base, 1, transformations); + } + + if (compiler->Base.Debug) { + fprintf(stderr, "Vertex program after native rewrite:\n"); + rc_print_program(&compiler->Base.Program); + fflush(stdout); + } + + { + /* Note: This pass has to be done seperately from ALU rewrite, + * otherwise non-native ALU instructions with source conflits + * will not be treated properly. + */ + struct radeon_program_transformation transformations[] = { + { &transform_source_conflicts, 0 }, + }; + radeonLocalTransform(&compiler->Base, 1, transformations); + } + + if (compiler->Base.Debug) { + fprintf(stderr, "Vertex program after source conflict resolve:\n"); + rc_print_program(&compiler->Base.Program); + fflush(stdout); + } + + { + struct radeon_nqssadce_descr nqssadce = { + .Init = &nqssadceInit, + .IsNativeSwizzle = &swizzleIsNative, + .BuildSwizzle = NULL + }; + radeonNqssaDce(&compiler->Base, &nqssadce, compiler); + + /* We need this step for reusing temporary registers */ + allocate_temporary_registers(compiler); + + if (compiler->Base.Debug) { + fprintf(stderr, "Vertex program after NQSSADCE:\n"); + rc_print_program(&compiler->Base.Program); + fflush(stdout); + } + } + + translate_vertex_program(compiler); + + rc_constants_copy(&compiler->code->constants, &compiler->Base.Program.Constants); + + compiler->code->InputsRead = compiler->Base.Program.InputsRead; + compiler->code->OutputsWritten = compiler->Base.Program.OutputsWritten; + + if (compiler->Base.Debug) { + printf("Final vertex program code:\n"); + r300_vertex_program_dump(compiler->code); + } +} diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog_dump.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog_dump.c new file mode 100644 index 0000000000..39cc6953ba --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog_dump.c @@ -0,0 +1,177 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "radeon_code.h" + +#include <stdio.h> + +static char* r300_vs_ve_ops[] = { + /* R300 vector ops */ + " VE_NO_OP", + " VE_DOT_PRODUCT", + " VE_MULTIPLY", + " VE_ADD", + " VE_MULTIPLY_ADD", + " VE_DISTANCE_FACTOR", + " VE_FRACTION", + " VE_MAXIMUM", + " VE_MINIMUM", + "VE_SET_GREATER_THAN_EQUAL", + " VE_SET_LESS_THAN", + " VE_MULTIPLYX2_ADD", + " VE_MULTIPLY_CLAMP", + " VE_FLT2FIX_DX", + " VE_FLT2FIX_DX_RND", + /* R500 vector ops */ + " VE_PRED_SET_EQ_PUSH", + " VE_PRED_SET_GT_PUSH", + " VE_PRED_SET_GTE_PUSH", + " VE_PRED_SET_NEQ_PUSH", + " VE_COND_WRITE_EQ", + " VE_COND_WRITE_GT", + " VE_COND_WRITE_GTE", + " VE_COND_WRITE_NEQ", + " VE_SET_GREATER_THAN", + " VE_SET_EQUAL", + " VE_SET_NOT_EQUAL", + " (reserved)", + " (reserved)", + " (reserved)", +}; + +static char* r300_vs_me_ops[] = { + /* R300 math ops */ + " ME_NO_OP", + " ME_EXP_BASE2_DX", + " ME_LOG_BASE2_DX", + " ME_EXP_BASEE_FF", + " ME_LIGHT_COEFF_DX", + " ME_POWER_FUNC_FF", + " ME_RECIP_DX", + " ME_RECIP_FF", + " ME_RECIP_SQRT_DX", + " ME_RECIP_SQRT_FF", + " ME_MULTIPLY", + " ME_EXP_BASE2_FULL_DX", + " ME_LOG_BASE2_FULL_DX", + " ME_POWER_FUNC_FF_CLAMP_B", + "ME_POWER_FUNC_FF_CLAMP_B1", + "ME_POWER_FUNC_FF_CLAMP_01", + " ME_SIN", + " ME_COS", + /* R500 math ops */ + " ME_LOG_BASE2_IEEE", + " ME_RECIP_IEEE", + " ME_RECIP_SQRT_IEEE", + " ME_PRED_SET_EQ", + " ME_PRED_SET_GT", + " ME_PRED_SET_GTE", + " ME_PRED_SET_NEQ", + " ME_PRED_SET_CLR", + " ME_PRED_SET_INV", + " ME_PRED_SET_POP", + " ME_PRED_SET_RESTORE", + " (reserved)", + " (reserved)", + " (reserved)", +}; + +/* XXX refactor to avoid clashing symbols */ +static char* r300_vs_src_debug[] = { + "t", + "i", + "c", + "a", +}; + +static char* r300_vs_dst_debug[] = { + "t", + "a0", + "o", + "ox", + "a", + "i", + "u", + "u", +}; + +static char* r300_vs_swiz_debug[] = { + "X", + "Y", + "Z", + "W", + "0", + "1", + "U", + "U", +}; + + +static void r300_vs_op_dump(uint32_t op) +{ + printf(" dst: %d%s op: ", + (op >> 13) & 0x7f, r300_vs_dst_debug[(op >> 8) & 0x7]); + if (op & 0x80) { + if (op & 0x1) { + printf("PVS_MACRO_OP_2CLK_M2X_ADD\n"); + } else { + printf(" PVS_MACRO_OP_2CLK_MADD\n"); + } + } else if (op & 0x40) { + printf("%s\n", r300_vs_me_ops[op & 0x1f]); + } else { + printf("%s\n", r300_vs_ve_ops[op & 0x1f]); + } +} + +static void r300_vs_src_dump(uint32_t src) +{ + printf(" reg: %d%s swiz: %s%s/%s%s/%s%s/%s%s\n", + (src >> 5) & 0x7f, r300_vs_src_debug[src & 0x3], + src & (1 << 25) ? "-" : " ", + r300_vs_swiz_debug[(src >> 13) & 0x7], + src & (1 << 26) ? "-" : " ", + r300_vs_swiz_debug[(src >> 16) & 0x7], + src & (1 << 27) ? "-" : " ", + r300_vs_swiz_debug[(src >> 19) & 0x7], + src & (1 << 28) ? "-" : " ", + r300_vs_swiz_debug[(src >> 22) & 0x7]); +} + +void r300_vertex_program_dump(struct r300_vertex_program_code * vs) +{ + unsigned instrcount = vs->length / 4; + unsigned i; + + for(i = 0; i < instrcount; i++) { + unsigned offset = i*4; + unsigned src; + + printf("%d: op: 0x%08x", i, vs->body.d[offset]); + r300_vs_op_dump(vs->body.d[offset]); + + for(src = 0; src < 3; ++src) { + printf(" src%i: 0x%08x", src, vs->body.d[offset+1+src]); + r300_vs_src_dump(vs->body.d[offset+1+src]); + } + } +} diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c index 4d58cf2162..7e2faed690 100644 --- a/src/mesa/drivers/dri/r300/r500_fragprog.c +++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c @@ -27,158 +27,141 @@ #include "r500_fragprog.h" -static void reset_srcreg(struct prog_src_register* reg) -{ - _mesa_bzero(reg, sizeof(*reg)); - reg->Swizzle = SWIZZLE_NOOP; -} +#include "../r300_reg.h" -static struct prog_src_register shadow_ambient(struct gl_program *program, int tmu) +static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int tmu) { - gl_state_index fail_value_tokens[STATE_LENGTH] = { - STATE_INTERNAL, STATE_SHADOW_AMBIENT, 0, 0, 0 - }; struct prog_src_register reg = { 0, }; - fail_value_tokens[2] = tmu; reg.File = PROGRAM_STATE_VAR; - reg.Index = _mesa_add_state_reference(program->Parameters, fail_value_tokens); + reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu); reg.Swizzle = SWIZZLE_WWWW; return reg; } /** * Transform TEX, TXP, TXB, and KIL instructions in the following way: - * - premultiply texture coordinates for RECT - * - extract operand swizzles - * - introduce a temporary register when write masks are needed - * + * - implement texture compare (shadow extensions) + * - extract non-native source / destination operands */ GLboolean r500_transform_TEX( - struct radeon_transform_context *t, - struct prog_instruction* orig_inst, void* data) + struct radeon_compiler * c, + struct rc_instruction * inst, + void* data) { struct r300_fragment_program_compiler *compiler = (struct r300_fragment_program_compiler*)data; - struct prog_instruction inst = *orig_inst; - struct prog_instruction* tgt; - GLboolean destredirect = GL_FALSE; - - if (inst.Opcode != OPCODE_TEX && - inst.Opcode != OPCODE_TXB && - inst.Opcode != OPCODE_TXP && - inst.Opcode != OPCODE_KIL) + + if (inst->I.Opcode != OPCODE_TEX && + inst->I.Opcode != OPCODE_TXB && + inst->I.Opcode != OPCODE_TXP && + inst->I.Opcode != OPCODE_KIL) return GL_FALSE; /* ARB_shadow & EXT_shadow_funcs */ - if (inst.Opcode != OPCODE_KIL && - t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) { - GLuint comparefunc = GL_NEVER + compiler->fp->state.unit[inst.TexSrcUnit].texture_compare_func; + if (inst->I.Opcode != OPCODE_KIL && + c->Program.ShadowSamplers & (1 << inst->I.TexSrcUnit)) { + GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) { - tgt = radeonAppendInstructions(t->Program, 1); + inst->I.Opcode = OPCODE_MOV; - tgt->Opcode = OPCODE_MOV; - tgt->DstReg = inst.DstReg; if (comparefunc == GL_ALWAYS) { - tgt->SrcReg[0].File = PROGRAM_BUILTIN; - tgt->SrcReg[0].Swizzle = SWIZZLE_1111; + inst->I.SrcReg[0].File = PROGRAM_BUILTIN; + inst->I.SrcReg[0].Swizzle = SWIZZLE_1111; } else { - tgt->SrcReg[0] = shadow_ambient(t->Program, inst.TexSrcUnit); + inst->I.SrcReg[0] = shadow_ambient(c, inst->I.TexSrcUnit); } + return GL_TRUE; + } else { + GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; + GLuint depthmode = compiler->state.unit[inst->I.TexSrcUnit].depth_texture_mode; + struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, inst); + struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_rcp); + struct rc_instruction * inst_cmp = rc_insert_new_instruction(c, inst_mad); + int pass, fail; + + inst_rcp->I.Opcode = OPCODE_RCP; + inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY; + inst_rcp->I.DstReg.Index = rc_find_free_temporary(c); + inst_rcp->I.DstReg.WriteMask = WRITEMASK_W; + inst_rcp->I.SrcReg[0] = inst->I.SrcReg[0]; + inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW; + + inst_cmp->I.DstReg = inst->I.DstReg; + inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.Index = rc_find_free_temporary(c); + inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + + inst_mad->I.Opcode = OPCODE_MAD; + inst_mad->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mad->I.DstReg.Index = rc_find_free_temporary(c); + inst_mad->I.SrcReg[0] = inst->I.SrcReg[0]; + inst_mad->I.SrcReg[0].Swizzle = SWIZZLE_ZZZZ; + inst_mad->I.SrcReg[1].File = PROGRAM_TEMPORARY; + inst_mad->I.SrcReg[1].Index = inst_rcp->I.DstReg.Index; + inst_mad->I.SrcReg[1].Swizzle = SWIZZLE_WWWW; + inst_mad->I.SrcReg[2].File = PROGRAM_TEMPORARY; + inst_mad->I.SrcReg[2].Index = inst->I.DstReg.Index; + if (depthmode == 0) /* GL_LUMINANCE */ + inst_mad->I.SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z); + else if (depthmode == 2) /* GL_ALPHA */ + inst_mad->I.SrcReg[2].Swizzle = SWIZZLE_WWWW; + + /* Recall that SrcReg[0] is tex, SrcReg[2] is r and: + * r < tex <=> -tex+r < 0 + * r >= tex <=> not (-tex+r < 0 */ + if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL) + inst_mad->I.SrcReg[2].Negate = inst_mad->I.SrcReg[2].Negate ^ NEGATE_XYZW; + else + inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ NEGATE_XYZW; + + inst_cmp->I.Opcode = OPCODE_CMP; + /* DstReg has been filled out above */ + inst_cmp->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_cmp->I.SrcReg[0].Index = inst_mad->I.DstReg.Index; + + if (comparefunc == GL_LESS || comparefunc == GL_GREATER) { + pass = 1; + fail = 2; + } else { + pass = 2; + fail = 1; + } + + inst_cmp->I.SrcReg[pass].File = PROGRAM_BUILTIN; + inst_cmp->I.SrcReg[pass].Swizzle = SWIZZLE_1111; + inst_cmp->I.SrcReg[fail] = shadow_ambient(c, inst->I.TexSrcUnit); } + } - inst.DstReg.File = PROGRAM_TEMPORARY; - inst.DstReg.Index = radeonFindFreeTemporary(t); - inst.DstReg.WriteMask = WRITEMASK_XYZW; - } else if (inst.Opcode != OPCODE_KIL && inst.DstReg.File != PROGRAM_TEMPORARY) { - int tempreg = radeonFindFreeTemporary(t); + /* Cannot write texture to output registers */ + if (inst->I.Opcode != OPCODE_KIL && inst->I.DstReg.File != PROGRAM_TEMPORARY) { + struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst); - inst.DstReg.File = PROGRAM_TEMPORARY; - inst.DstReg.Index = tempreg; - inst.DstReg.WriteMask = WRITEMASK_XYZW; - destredirect = GL_TRUE; - } + inst_mov->I.Opcode = OPCODE_MOV; + inst_mov->I.DstReg = inst->I.DstReg; + inst_mov->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_mov->I.SrcReg[0].Index = rc_find_free_temporary(c); - if (inst.SrcReg[0].File != PROGRAM_TEMPORARY && inst.SrcReg[0].File != PROGRAM_INPUT) { - int tmpreg = radeonFindFreeTemporary(t); - tgt = radeonAppendInstructions(t->Program, 1); - tgt->Opcode = OPCODE_MOV; - tgt->DstReg.File = PROGRAM_TEMPORARY; - tgt->DstReg.Index = tmpreg; - tgt->SrcReg[0] = inst.SrcReg[0]; - - reset_srcreg(&inst.SrcReg[0]); - inst.SrcReg[0].File = PROGRAM_TEMPORARY; - inst.SrcReg[0].Index = tmpreg; + inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index; + inst->I.DstReg.WriteMask = WRITEMASK_XYZW; } - tgt = radeonAppendInstructions(t->Program, 1); - _mesa_copy_instructions(tgt, &inst, 1); - - if (inst.Opcode != OPCODE_KIL && - t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) { - GLuint comparefunc = GL_NEVER + compiler->fp->state.unit[inst.TexSrcUnit].texture_compare_func; - GLuint depthmode = compiler->fp->state.unit[inst.TexSrcUnit].depth_texture_mode; - int rcptemp = radeonFindFreeTemporary(t); - int pass, fail; - - tgt = radeonAppendInstructions(t->Program, 3); - - tgt[0].Opcode = OPCODE_RCP; - tgt[0].DstReg.File = PROGRAM_TEMPORARY; - tgt[0].DstReg.Index = rcptemp; - tgt[0].DstReg.WriteMask = WRITEMASK_W; - tgt[0].SrcReg[0] = inst.SrcReg[0]; - tgt[0].SrcReg[0].Swizzle = SWIZZLE_WWWW; - - tgt[1].Opcode = OPCODE_MAD; - tgt[1].DstReg = inst.DstReg; - tgt[1].DstReg.WriteMask = orig_inst->DstReg.WriteMask; - tgt[1].SrcReg[0] = inst.SrcReg[0]; - tgt[1].SrcReg[0].Swizzle = SWIZZLE_ZZZZ; - tgt[1].SrcReg[1].File = PROGRAM_TEMPORARY; - tgt[1].SrcReg[1].Index = rcptemp; - tgt[1].SrcReg[1].Swizzle = SWIZZLE_WWWW; - tgt[1].SrcReg[2].File = PROGRAM_TEMPORARY; - tgt[1].SrcReg[2].Index = inst.DstReg.Index; - if (depthmode == 0) /* GL_LUMINANCE */ - tgt[1].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z); - else if (depthmode == 2) /* GL_ALPHA */ - tgt[1].SrcReg[2].Swizzle = SWIZZLE_WWWW; - - /* Recall that SrcReg[0] is tex, SrcReg[2] is r and: - * r < tex <=> -tex+r < 0 - * r >= tex <=> not (-tex+r < 0 */ - if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL) - tgt[1].SrcReg[2].Negate = tgt[0].SrcReg[2].Negate ^ NEGATE_XYZW; - else - tgt[1].SrcReg[0].Negate = tgt[0].SrcReg[0].Negate ^ NEGATE_XYZW; - - tgt[2].Opcode = OPCODE_CMP; - tgt[2].DstReg = orig_inst->DstReg; - tgt[2].SrcReg[0].File = PROGRAM_TEMPORARY; - tgt[2].SrcReg[0].Index = tgt[1].DstReg.Index; - - if (comparefunc == GL_LESS || comparefunc == GL_GREATER) { - pass = 1; - fail = 2; - } else { - pass = 2; - fail = 1; - } + /* Cannot read texture coordinate from constants file */ + if (inst->I.SrcReg[0].File != PROGRAM_TEMPORARY && inst->I.SrcReg[0].File != PROGRAM_INPUT) { + struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev); - tgt[2].SrcReg[pass].File = PROGRAM_BUILTIN; - tgt[2].SrcReg[pass].Swizzle = SWIZZLE_1111; - tgt[2].SrcReg[fail] = shadow_ambient(t->Program, inst.TexSrcUnit); - } else if (destredirect) { - tgt = radeonAppendInstructions(t->Program, 1); + inst_mov->I.Opcode = OPCODE_MOV; + inst_mov->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mov->I.DstReg.Index = rc_find_free_temporary(c); + inst_mov->I.SrcReg[0] = inst->I.SrcReg[0]; - tgt->Opcode = OPCODE_MOV; - tgt->DstReg = orig_inst->DstReg; - tgt->SrcReg[0].File = PROGRAM_TEMPORARY; - tgt->SrcReg[0].Index = inst.DstReg.Index; + reset_srcreg(&inst->I.SrcReg[0]); + inst->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[0].Index = inst_mov->I.DstReg.Index; } return GL_TRUE; @@ -249,7 +232,6 @@ GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg) */ void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src) { - struct prog_instruction *inst; GLuint negatebase[2] = { 0, 0 }; int i; @@ -260,20 +242,16 @@ void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, negatebase[GET_BIT(src.Negate, i)] |= 1 << i; } - _mesa_insert_instructions(s->Program, s->IP, (negatebase[0] ? 1 : 0) + (negatebase[1] ? 1 : 0)); - inst = s->Program->Instructions + s->IP; - for(i = 0; i <= 1; ++i) { if (!negatebase[i]) continue; - inst->Opcode = OPCODE_MOV; - inst->DstReg = dst; - inst->DstReg.WriteMask = negatebase[i]; - inst->SrcReg[0] = src; - inst->SrcReg[0].Negate = (i == 0) ? NEGATE_NONE : NEGATE_XYZW; - inst++; - s->IP++; + struct rc_instruction *inst = rc_insert_new_instruction(s->Compiler, s->IP->Prev); + inst->I.Opcode = OPCODE_MOV; + inst->I.DstReg = dst; + inst->I.DstReg.WriteMask = negatebase[i]; + inst->I.SrcReg[0] = src; + inst->I.SrcReg[0].Negate = (i == 0) ? NEGATE_NONE : NEGATE_XYZW; } } @@ -375,9 +353,9 @@ static char *to_texop(int val) return NULL; } -void r500FragmentProgramDump(union rX00_fragment_program_code *c) +void r500FragmentProgramDump(struct rX00_fragment_program_code *c) { - struct r500_fragment_program_code *code = &c->r500; + struct r500_fragment_program_code *code = &c->code.r500; fprintf(stderr, "R500 Fragment Program:\n--------\n"); int n; @@ -385,15 +363,6 @@ void r500FragmentProgramDump(union rX00_fragment_program_code *c) uint32_t inst0; char *str = NULL; - if (code->const_nr) { - fprintf(stderr, "--------\nConstants:\n"); - for (n = 0; n < code->const_nr; n++) { - fprintf(stderr, "Constant %d: %i[%i]\n", n, - code->constant[n].File, code->constant[n].Index); - } - fprintf(stderr, "--------\n"); - } - for (n = 0; n < code->inst_end+1; n++) { inst0 = inst = code->inst[n].inst0; fprintf(stderr,"%d\t0:CMN_INST 0x%08x:", n, inst); diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h index 1179bf6607..9091f65cd2 100644 --- a/src/mesa/drivers/dri/r300/r500_fragprog.h +++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h @@ -36,17 +36,20 @@ #include "shader/prog_parameter.h" #include "shader/prog_instruction.h" -#include "r300_context.h" +#include "radeon_compiler.h" #include "radeon_nqssadce.h" -extern GLboolean r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler); +extern void r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler); -extern void r500FragmentProgramDump(union rX00_fragment_program_code *c); +extern void r500FragmentProgramDump(struct rX00_fragment_program_code *c); extern GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg); extern void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src); -extern GLboolean r500_transform_TEX(struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data); +extern GLboolean r500_transform_TEX( + struct radeon_compiler * c, + struct rc_instruction * inst, + void* data); #endif diff --git a/src/mesa/drivers/dri/r300/r500_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c index 30f4514897..d694725c9b 100644 --- a/src/mesa/drivers/dri/r300/r500_fragprog_emit.c +++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c @@ -45,47 +45,22 @@ #include "r500_fragprog.h" +#include "../r300_reg.h" + #include "radeon_program_pair.h" #define PROG_CODE \ struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \ - struct r500_fragment_program_code *code = &c->code->r500 + struct r500_fragment_program_code *code = &c->code->code.r500 #define error(fmt, args...) do { \ - fprintf(stderr, "%s::%s(): " fmt "\n", \ + rc_error(&c->Base, "%s::%s(): " fmt "\n", \ __FILE__, __FUNCTION__, ##args); \ } while(0) -/** - * Callback to register hardware constants. - */ -static GLboolean emit_const(void *data, GLuint file, GLuint idx, GLuint *hwindex) -{ - PROG_CODE; - - for (*hwindex = 0; *hwindex < code->const_nr; ++*hwindex) { - if (code->constant[*hwindex].File == file && - code->constant[*hwindex].Index == idx) - break; - } - - if (*hwindex >= code->const_nr) { - if (*hwindex >= R500_PFS_NUM_CONST_REGS) { - error("Out of hw constants!\n"); - return GL_FALSE; - } - - code->const_nr++; - code->constant[*hwindex].File = file; - code->constant[*hwindex].Index = idx; - } - - return GL_TRUE; -} - -static GLuint translate_rgb_op(GLuint opcode) +static GLuint translate_rgb_op(struct r300_fragment_program_compiler *c, GLuint opcode) { switch(opcode) { case OPCODE_CMP: return R500_ALU_RGBA_OP_CMP; @@ -106,7 +81,7 @@ static GLuint translate_rgb_op(GLuint opcode) } } -static GLuint translate_alpha_op(GLuint opcode) +static GLuint translate_alpha_op(struct r300_fragment_program_compiler *c, GLuint opcode) { switch(opcode) { case OPCODE_CMP: return R500_ALPHA_OP_CMP; @@ -189,8 +164,8 @@ static GLboolean emit_paired(void *data, struct radeon_pair_instruction *inst) int ip = ++code->inst_end; - code->inst[ip].inst5 = translate_rgb_op(inst->RGB.Opcode); - code->inst[ip].inst4 = translate_alpha_op(inst->Alpha.Opcode); + code->inst[ip].inst5 = translate_rgb_op(c, inst->RGB.Opcode); + code->inst[ip].inst4 = translate_alpha_op(c, inst->Alpha.Opcode); if (inst->RGB.OutputWriteMask || inst->Alpha.OutputWriteMask || inst->Alpha.DepthWriteMask) code->inst[ip].inst0 = R500_INST_TYPE_OUT; @@ -202,7 +177,7 @@ static GLboolean emit_paired(void *data, struct radeon_pair_instruction *inst) code->inst[ip].inst0 |= (inst->RGB.OutputWriteMask << 15) | (inst->Alpha.OutputWriteMask << 18); if (inst->Alpha.DepthWriteMask) { code->inst[ip].inst4 |= R500_ALPHA_W_OMASK; - c->fp->writes_depth = GL_TRUE; + c->code->writes_depth = GL_TRUE; } code->inst[ip].inst4 |= R500_ALPHA_ADDRD(inst->Alpha.DestIndex); @@ -234,19 +209,19 @@ static GLboolean emit_paired(void *data, struct radeon_pair_instruction *inst) return GL_TRUE; } -static GLuint translate_strq_swizzle(struct prog_src_register src) +static GLuint translate_strq_swizzle(GLuint swizzle) { GLuint swiz = 0; int i; for (i = 0; i < 4; i++) - swiz |= (GET_SWZ(src.Swizzle, i) & 0x3) << i*2; + swiz |= (GET_SWZ(swizzle, i) & 0x3) << i*2; return swiz; } /** * Emit a single TEX instruction */ -static GLboolean emit_tex(void *data, struct prog_instruction *inst) +static GLboolean emit_tex(void *data, struct radeon_pair_texture_instruction *inst) { PROG_CODE; @@ -258,7 +233,7 @@ static GLboolean emit_tex(void *data, struct prog_instruction *inst) int ip = ++code->inst_end; code->inst[ip].inst0 = R500_INST_TYPE_TEX - | (inst->DstReg.WriteMask << 11) + | (inst->WriteMask << 11) | R500_INST_TEX_SEM_WAIT; code->inst[ip].inst1 = R500_TEX_ID(inst->TexSrcUnit) | R500_TEX_SEM_ACQUIRE | R500_TEX_IGNORE_UNCOVERED; @@ -267,25 +242,25 @@ static GLboolean emit_tex(void *data, struct prog_instruction *inst) code->inst[ip].inst1 |= R500_TEX_UNSCALED; switch (inst->Opcode) { - case OPCODE_KIL: + case RADEON_OPCODE_KIL: code->inst[ip].inst1 |= R500_TEX_INST_TEXKILL; break; - case OPCODE_TEX: + case RADEON_OPCODE_TEX: code->inst[ip].inst1 |= R500_TEX_INST_LD; break; - case OPCODE_TXB: + case RADEON_OPCODE_TXB: code->inst[ip].inst1 |= R500_TEX_INST_LODBIAS; break; - case OPCODE_TXP: + case RADEON_OPCODE_TXP: code->inst[ip].inst1 |= R500_TEX_INST_PROJ; break; default: error("emit_tex can't handle opcode %x\n", inst->Opcode); } - code->inst[ip].inst2 = R500_TEX_SRC_ADDR(inst->SrcReg[0].Index) - | (translate_strq_swizzle(inst->SrcReg[0]) << 8) - | R500_TEX_DST_ADDR(inst->DstReg.Index) + code->inst[ip].inst2 = R500_TEX_SRC_ADDR(inst->SrcIndex) + | (translate_strq_swizzle(inst->SrcSwizzle) << 8) + | R500_TEX_DST_ADDR(inst->DestIndex) | R500_TEX_DST_R_SWIZ_R | R500_TEX_DST_G_SWIZ_G | R500_TEX_DST_B_SWIZ_B | R500_TEX_DST_A_SWIZ_A; @@ -293,35 +268,32 @@ static GLboolean emit_tex(void *data, struct prog_instruction *inst) } static const struct radeon_pair_handler pair_handler = { - .EmitConst = emit_const, .EmitPaired = emit_paired, .EmitTex = emit_tex, .MaxHwTemps = 128 }; -GLboolean r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler) +void r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler) { - struct r500_fragment_program_code *code = &compiler->code->r500; + struct r500_fragment_program_code *code = &compiler->code->code.r500; _mesa_bzero(code, sizeof(*code)); code->max_temp_idx = 1; - code->inst_offset = 0; code->inst_end = -1; - if (!radeonPairProgram(compiler->r300->radeon.glCtx, compiler->program, &pair_handler, compiler)) - return GL_FALSE; + radeonPairProgram(compiler, &pair_handler, compiler); + if (compiler->Base.Error) + return; if ((code->inst[code->inst_end].inst0 & R500_INST_TYPE_MASK) != R500_INST_TYPE_OUT) { /* This may happen when dead-code elimination is disabled or * when most of the fragment program logic is leading to a KIL */ if (code->inst_end >= 511) { - error("Introducing fake OUT: Too many instructions"); - return GL_FALSE; + rc_error(&compiler->Base, "Introducing fake OUT: Too many instructions"); + return; } int ip = ++code->inst_end; code->inst[ip].inst0 = R500_INST_TYPE_OUT | R500_INST_TEX_SEM_WAIT; } - - return GL_TRUE; } diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.c b/src/mesa/drivers/dri/r300/compiler/radeon_code.c new file mode 100644 index 0000000000..c7923004df --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.c @@ -0,0 +1,170 @@ +/* + * Copyright (C) 2009 Nicolai Haehnle. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "main/mtypes.h" +#include "shader/prog_instruction.h" + +#include "radeon_code.h" + +void rc_constants_init(struct rc_constant_list * c) +{ + memset(c, 0, sizeof(*c)); +} + +/** + * Copy a constants structure, assuming that the destination structure + * is not initialized. + */ +void rc_constants_copy(struct rc_constant_list * dst, struct rc_constant_list * src) +{ + dst->Constants = malloc(sizeof(struct rc_constant) * src->Count); + memcpy(dst->Constants, src->Constants, sizeof(struct rc_constant) * src->Count); + dst->Count = src->Count; + dst->_Reserved = src->Count; +} + +void rc_constants_destroy(struct rc_constant_list * c) +{ + free(c->Constants); + memset(c, 0, sizeof(*c)); +} + +unsigned rc_constants_add(struct rc_constant_list * c, struct rc_constant * constant) +{ + unsigned index = c->Count; + + if (c->Count >= c->_Reserved) { + struct rc_constant * newlist; + + c->_Reserved = c->_Reserved * 2; + if (!c->_Reserved) + c->_Reserved = 16; + + newlist = malloc(sizeof(struct rc_constant) * c->_Reserved); + memcpy(newlist, c->Constants, sizeof(struct rc_constant) * c->Count); + + free(c->Constants); + c->Constants = newlist; + } + + c->Constants[index] = *constant; + c->Count++; + + return index; +} + + +/** + * Add a state vector to the constant list, while trying to avoid duplicates. + */ +unsigned rc_constants_add_state(struct rc_constant_list * c, unsigned state0, unsigned state1) +{ + unsigned index; + struct rc_constant constant; + + for(index = 0; index < c->Count; ++index) { + if (c->Constants[index].Type == RC_CONSTANT_STATE) { + if (c->Constants[index].u.State[0] == state0 && + c->Constants[index].u.State[1] == state1) + return index; + } + } + + memset(&constant, 0, sizeof(constant)); + constant.Type = RC_CONSTANT_STATE; + constant.Size = 4; + constant.u.State[0] = state0; + constant.u.State[1] = state1; + + return rc_constants_add(c, &constant); +} + + +/** + * Add an immediate vector to the constant list, while trying to avoid + * duplicates. + */ +unsigned rc_constants_add_immediate_vec4(struct rc_constant_list * c, const float * data) +{ + unsigned index; + struct rc_constant constant; + + for(index = 0; index < c->Count; ++index) { + if (c->Constants[index].Type == RC_CONSTANT_IMMEDIATE) { + if (!memcmp(c->Constants[index].u.Immediate, data, sizeof(float)*4)) + return index; + } + } + + memset(&constant, 0, sizeof(constant)); + constant.Type = RC_CONSTANT_IMMEDIATE; + constant.Size = 4; + memcpy(constant.u.Immediate, data, sizeof(float) * 4); + + return rc_constants_add(c, &constant); +} + + +/** + * Add an immediate scalar to the constant list, while trying to avoid + * duplicates. + */ +unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float data, unsigned * swizzle) +{ + unsigned index; + int free_index = -1; + struct rc_constant constant; + + for(index = 0; index < c->Count; ++index) { + if (c->Constants[index].Type == RC_CONSTANT_IMMEDIATE) { + for(unsigned comp = 0; comp < c->Constants[index].Size; ++comp) { + if (c->Constants[index].u.Immediate[comp] == data) { + *swizzle = MAKE_SWIZZLE4(comp, comp, comp, comp); + return index; + } + } + + if (c->Constants[index].Size < 4) + free_index = index; + } + } + + if (free_index >= 0) { + unsigned comp = c->Constants[free_index].Size++; + c->Constants[free_index].u.Immediate[comp] = data; + *swizzle = MAKE_SWIZZLE4(comp, comp, comp, comp); + return free_index; + } + + memset(&constant, 0, sizeof(constant)); + constant.Type = RC_CONSTANT_IMMEDIATE; + constant.Size = 1; + constant.u.Immediate[0] = data; + *swizzle = SWIZZLE_XXXX; + + return rc_constants_add(c, &constant); +} diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.h b/src/mesa/drivers/dri/r300/compiler/radeon_code.h new file mode 100644 index 0000000000..0806fb1b5c --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.h @@ -0,0 +1,206 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef RADEON_CODE_H +#define RADEON_CODE_H + +#include <stdint.h> + +#define R300_PFS_MAX_ALU_INST 64 +#define R300_PFS_MAX_TEX_INST 32 +#define R300_PFS_MAX_TEX_INDIRECT 4 +#define R300_PFS_NUM_TEMP_REGS 32 +#define R300_PFS_NUM_CONST_REGS 32 + +#define R500_PFS_MAX_INST 512 +#define R500_PFS_NUM_TEMP_REGS 128 +#define R500_PFS_NUM_CONST_REGS 256 + + +#define STATE_R300_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0) + +enum { + /** + * External constants are constants whose meaning is unknown to this + * compiler. For example, a Mesa gl_program's constants are turned + * into external constants. + */ + RC_CONSTANT_EXTERNAL = 0, + + RC_CONSTANT_IMMEDIATE, + + /** + * Constant referring to state that is known by this compiler, + * see RC_STATE_xxx, i.e. *not* arbitrary Mesa (or other) state. + */ + RC_CONSTANT_STATE +}; + +enum { + RC_STATE_SHADOW_AMBIENT = 0, + + RC_STATE_R300_WINDOW_DIMENSION, + RC_STATE_R300_TEXRECT_FACTOR +}; + +struct rc_constant { + unsigned Type:2; /**< RC_CONSTANT_xxx */ + unsigned Size:3; + + union { + unsigned External; + float Immediate[4]; + unsigned State[2]; + } u; +}; + +struct rc_constant_list { + struct rc_constant * Constants; + unsigned Count; + + unsigned _Reserved; +}; + +void rc_constants_init(struct rc_constant_list * c); +void rc_constants_copy(struct rc_constant_list * dst, struct rc_constant_list * src); +void rc_constants_destroy(struct rc_constant_list * c); +unsigned rc_constants_add(struct rc_constant_list * c, struct rc_constant * constant); +unsigned rc_constants_add_state(struct rc_constant_list * c, unsigned state1, unsigned state2); +unsigned rc_constants_add_immediate_vec4(struct rc_constant_list * c, const float * data); +unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float data, unsigned * swizzle); + +/** + * Stores state that influences the compilation of a fragment program. + */ +struct r300_fragment_program_external_state { + struct { + /** + * If the sampler is used as a shadow sampler, + * this field is: + * 0 - GL_LUMINANCE + * 1 - GL_INTENSITY + * 2 - GL_ALPHA + * depending on the depth texture mode. + */ + unsigned depth_texture_mode : 2; + + /** + * If the sampler is used as a shadow sampler, + * this field is (texture_compare_func - GL_NEVER). + * [e.g. if compare function is GL_LEQUAL, this field is 3] + * + * Otherwise, this field is 0. + */ + unsigned texture_compare_func : 3; + } unit[16]; +}; + + + +struct r300_fragment_program_node { + int tex_offset; /**< first tex instruction */ + int tex_end; /**< last tex instruction, relative to tex_offset */ + int alu_offset; /**< first ALU instruction */ + int alu_end; /**< last ALU instruction, relative to alu_offset */ + int flags; +}; + +/** + * Stores an R300 fragment program in its compiled-to-hardware form. + */ +struct r300_fragment_program_code { + struct { + int length; /**< total # of texture instructions used */ + uint32_t inst[R300_PFS_MAX_TEX_INST]; + } tex; + + struct { + int length; /**< total # of ALU instructions used */ + struct { + uint32_t rgb_inst; + uint32_t rgb_addr; + uint32_t alpha_inst; + uint32_t alpha_addr; + } inst[R300_PFS_MAX_ALU_INST]; + } alu; + + uint32_t config; /* US_CONFIG */ + uint32_t pixsize; /* US_PIXSIZE */ + uint32_t code_offset; /* US_CODE_OFFSET */ + uint32_t code_addr[4]; /* US_CODE_ADDR */ +}; + + +struct r500_fragment_program_code { + struct { + uint32_t inst0; + uint32_t inst1; + uint32_t inst2; + uint32_t inst3; + uint32_t inst4; + uint32_t inst5; + } inst[R500_PFS_MAX_INST]; + + int inst_end; /* Number of instructions - 1; also, last instruction to be executed */ + + int max_temp_idx; +}; + +struct rX00_fragment_program_code { + union { + struct r300_fragment_program_code r300; + struct r500_fragment_program_code r500; + } code; + + unsigned writes_depth:1; + + struct rc_constant_list constants; +}; + + +#define VSF_MAX_FRAGMENT_LENGTH (255*4) +#define VSF_MAX_FRAGMENT_TEMPS (14) + +#define VSF_MAX_INPUTS 32 +#define VSF_MAX_OUTPUTS 32 + +struct r300_vertex_program_code { + int length; + union { + uint32_t d[VSF_MAX_FRAGMENT_LENGTH]; + float f[VSF_MAX_FRAGMENT_LENGTH]; + } body; + + int pos_end; + int num_temporaries; /* Number of temp vars used by program */ + int inputs[VSF_MAX_INPUTS]; + int outputs[VSF_MAX_OUTPUTS]; + + struct rc_constant_list constants; + + uint32_t InputsRead; + uint32_t OutputsWritten; +}; + +void r300_vertex_program_dump(struct r300_vertex_program_code * vs); + +#endif /* RADEON_CODE_H */
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c new file mode 100644 index 0000000000..da950d5289 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c @@ -0,0 +1,262 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "radeon_compiler.h" + +#include <stdarg.h> + +#include "radeon_program.h" + + +void rc_init(struct radeon_compiler * c) +{ + memset(c, 0, sizeof(*c)); + + memory_pool_init(&c->Pool); + c->Program.Instructions.Prev = &c->Program.Instructions; + c->Program.Instructions.Next = &c->Program.Instructions; + c->Program.Instructions.I.Opcode = OPCODE_END; +} + +void rc_destroy(struct radeon_compiler * c) +{ + rc_constants_destroy(&c->Program.Constants); + memory_pool_destroy(&c->Pool); + free(c->ErrorMsg); +} + +void rc_debug(struct radeon_compiler * c, const char * fmt, ...) +{ + va_list ap; + + if (!c->Debug) + return; + + va_start(ap, fmt); + vfprintf(stderr, fmt, ap); + va_end(ap); +} + +void rc_error(struct radeon_compiler * c, const char * fmt, ...) +{ + va_list ap; + + c->Error = GL_TRUE; + + if (!c->ErrorMsg) { + /* Only remember the first error */ + char buf[1024]; + int written; + + va_start(ap, fmt); + written = vsnprintf(buf, sizeof(buf), fmt, ap); + va_end(ap); + + if (written < sizeof(buf)) { + c->ErrorMsg = strdup(buf); + } else { + c->ErrorMsg = malloc(written + 1); + + va_start(ap, fmt); + vsnprintf(c->ErrorMsg, written + 1, fmt, ap); + va_end(ap); + } + } + + if (c->Debug) { + fprintf(stderr, "r300compiler error: "); + + va_start(ap, fmt); + vfprintf(stderr, fmt, ap); + va_end(ap); + } +} + +/** + * Rewrite the program such that everything that source the given input + * register will source new_input instead. + */ +void rc_move_input(struct radeon_compiler * c, unsigned input, struct prog_src_register new_input) +{ + struct rc_instruction * inst; + + c->Program.InputsRead &= ~(1 << input); + + for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) { + const unsigned numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode); + unsigned i; + + for(i = 0; i < numsrcs; ++i) { + if (inst->I.SrcReg[i].File == PROGRAM_INPUT && inst->I.SrcReg[i].Index == input) { + inst->I.SrcReg[i].File = new_input.File; + inst->I.SrcReg[i].Index = new_input.Index; + inst->I.SrcReg[i].Swizzle = combine_swizzles(new_input.Swizzle, inst->I.SrcReg[i].Swizzle); + if (!inst->I.SrcReg[i].Abs) { + inst->I.SrcReg[i].Negate ^= new_input.Negate; + inst->I.SrcReg[i].Abs = new_input.Abs; + } + + c->Program.InputsRead |= 1 << new_input.Index; + } + } + } +} + + +/** + * Rewrite the program such that everything that writes into the given + * output register will instead write to new_output. The new_output + * writemask is honoured. + */ +void rc_move_output(struct radeon_compiler * c, unsigned output, unsigned new_output, unsigned writemask) +{ + struct rc_instruction * inst; + + c->Program.OutputsWritten &= ~(1 << output); + + for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) { + const unsigned numdsts = _mesa_num_inst_dst_regs(inst->I.Opcode); + + if (numdsts) { + if (inst->I.DstReg.File == PROGRAM_OUTPUT && inst->I.DstReg.Index == output) { + inst->I.DstReg.Index = new_output; + inst->I.DstReg.WriteMask &= writemask; + + c->Program.OutputsWritten |= 1 << new_output; + } + } + } +} + + +/** + * Rewrite the program such that a given output is duplicated. + */ +void rc_copy_output(struct radeon_compiler * c, unsigned output, unsigned dup_output) +{ + unsigned tempreg = rc_find_free_temporary(c); + struct rc_instruction * inst; + + for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) { + const unsigned numdsts = _mesa_num_inst_dst_regs(inst->I.Opcode); + + if (numdsts) { + if (inst->I.DstReg.File == PROGRAM_OUTPUT && inst->I.DstReg.Index == output) { + inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.Index = tempreg; + } + } + } + + inst = rc_insert_new_instruction(c, c->Program.Instructions.Prev); + inst->I.Opcode = OPCODE_MOV; + inst->I.DstReg.File = PROGRAM_OUTPUT; + inst->I.DstReg.Index = output; + + inst->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[0].Index = tempreg; + inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW; + + inst = rc_insert_new_instruction(c, c->Program.Instructions.Prev); + inst->I.Opcode = OPCODE_MOV; + inst->I.DstReg.File = PROGRAM_OUTPUT; + inst->I.DstReg.Index = dup_output; + + inst->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[0].Index = tempreg; + inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW; + + c->Program.OutputsWritten |= 1 << dup_output; +} + + +/** + * Introduce standard code fragment to deal with fragment.position. + */ +void rc_transform_fragment_wpos(struct radeon_compiler * c, unsigned wpos, unsigned new_input) +{ + unsigned tempregi = rc_find_free_temporary(c); + + c->Program.InputsRead &= ~(1 << wpos); + c->Program.InputsRead |= 1 << new_input; + + /* perspective divide */ + struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, &c->Program.Instructions); + inst_rcp->I.Opcode = OPCODE_RCP; + + inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY; + inst_rcp->I.DstReg.Index = tempregi; + inst_rcp->I.DstReg.WriteMask = WRITEMASK_W; + + inst_rcp->I.SrcReg[0].File = PROGRAM_INPUT; + inst_rcp->I.SrcReg[0].Index = new_input; + inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW; + + struct rc_instruction * inst_mul = rc_insert_new_instruction(c, inst_rcp); + inst_mul->I.Opcode = OPCODE_MUL; + + inst_mul->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mul->I.DstReg.Index = tempregi; + inst_mul->I.DstReg.WriteMask = WRITEMASK_XYZ; + + inst_mul->I.SrcReg[0].File = PROGRAM_INPUT; + inst_mul->I.SrcReg[0].Index = new_input; + + inst_mul->I.SrcReg[1].File = PROGRAM_TEMPORARY; + inst_mul->I.SrcReg[1].Index = tempregi; + inst_mul->I.SrcReg[1].Swizzle = SWIZZLE_WWWW; + + /* viewport transformation */ + struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_mul); + inst_mad->I.Opcode = OPCODE_MAD; + + inst_mad->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mad->I.DstReg.Index = tempregi; + inst_mad->I.DstReg.WriteMask = WRITEMASK_XYZ; + + inst_mad->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_mad->I.SrcReg[0].Index = tempregi; + inst_mad->I.SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); + + inst_mad->I.SrcReg[1].File = PROGRAM_STATE_VAR; + inst_mad->I.SrcReg[1].Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_R300_WINDOW_DIMENSION, 0); + inst_mad->I.SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); + + inst_mad->I.SrcReg[2].File = PROGRAM_STATE_VAR; + inst_mad->I.SrcReg[2].Index = inst_mad->I.SrcReg[1].Index; + inst_mad->I.SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); + + struct rc_instruction * inst; + for (inst = inst_mad->Next; inst != &c->Program.Instructions; inst = inst->Next) { + const unsigned numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode); + unsigned i; + + for(i = 0; i < numsrcs; i++) { + if (inst->I.SrcReg[i].File == PROGRAM_INPUT && + inst->I.SrcReg[i].Index == wpos) { + inst->I.SrcReg[i].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[i].Index = tempregi; + } + } + } +} + diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h new file mode 100644 index 0000000000..e63ab8840a --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h @@ -0,0 +1,108 @@ +/* + * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef RADEON_COMPILER_H +#define RADEON_COMPILER_H + +#include "main/mtypes.h" +#include "shader/prog_instruction.h" + +#include "memory_pool.h" +#include "radeon_code.h" + + +struct rc_instruction { + struct rc_instruction * Prev; + struct rc_instruction * Next; + struct prog_instruction I; +}; + +struct rc_program { + /** + * Instructions.Next points to the first instruction, + * Instructions.Prev points to the last instruction. + */ + struct rc_instruction Instructions; + + /* Long term, we should probably remove InputsRead & OutputsWritten, + * since updating dependent state can be fragile, and they aren't + * actually used very often. */ + uint32_t InputsRead; + uint32_t OutputsWritten; + uint32_t ShadowSamplers; /**< Texture units used for shadow sampling. */ + + struct rc_constant_list Constants; +}; + +struct radeon_compiler { + struct memory_pool Pool; + struct rc_program Program; + unsigned Debug:1; + unsigned Error:1; + char * ErrorMsg; +}; + +void rc_init(struct radeon_compiler * c); +void rc_destroy(struct radeon_compiler * c); + +void rc_debug(struct radeon_compiler * c, const char * fmt, ...); +void rc_error(struct radeon_compiler * c, const char * fmt, ...); + +void rc_mesa_to_rc_program(struct radeon_compiler * c, struct gl_program * program); + +void rc_calculate_inputs_outputs(struct radeon_compiler * c); + +void rc_move_input(struct radeon_compiler * c, unsigned input, struct prog_src_register new_input); +void rc_move_output(struct radeon_compiler * c, unsigned output, unsigned new_output, unsigned writemask); +void rc_copy_output(struct radeon_compiler * c, unsigned output, unsigned dup_output); +void rc_transform_fragment_wpos(struct radeon_compiler * c, unsigned wpos, unsigned new_input); + +struct r300_fragment_program_compiler { + struct radeon_compiler Base; + struct rX00_fragment_program_code *code; + struct r300_fragment_program_external_state state; + unsigned is_r500; + unsigned OutputDepth; + unsigned OutputColor; + + void * UserData; + void (*AllocateHwInputs)( + struct r300_fragment_program_compiler * c, + void (*allocate)(void * data, unsigned input, unsigned hwreg), + void * mydata); +}; + +void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c); + + +struct r300_vertex_program_compiler { + struct radeon_compiler Base; + struct r300_vertex_program_code *code; + GLbitfield RequiredOutputs; + + void * UserData; + void (*SetHwInputOutput)(struct r300_vertex_program_compiler * c); +}; + +void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* c); + +#endif /* RADEON_COMPILER_H */ diff --git a/src/mesa/drivers/dri/r300/radeon_nqssadce.c b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c index 840c9733b1..aaaa50ad1f 100644 --- a/src/mesa/drivers/dri/r300/radeon_nqssadce.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c @@ -36,6 +36,8 @@ #include "radeon_nqssadce.h" +#include "radeon_compiler.h" + /** * Return the @ref register_state for the given register (or 0 for untracked @@ -76,9 +78,10 @@ struct prog_src_register lmul_swizzle(GLuint swizzle, struct prog_src_register s } -static struct prog_instruction* track_used_srcreg(struct nqssadce_state* s, - struct prog_instruction *inst, GLint src, GLuint sourced) +static void track_used_srcreg(struct nqssadce_state* s, + GLint src, GLuint sourced) { + struct prog_instruction * inst = &s->IP->I; int i; GLuint deswz_source = 0; @@ -95,12 +98,11 @@ static struct prog_instruction* track_used_srcreg(struct nqssadce_state* s, if (!s->Descr->IsNativeSwizzle(inst->Opcode, inst->SrcReg[src])) { struct prog_dst_register dstreg = inst->DstReg; dstreg.File = PROGRAM_TEMPORARY; - dstreg.Index = _mesa_find_free_register(s->Program, PROGRAM_TEMPORARY); + dstreg.Index = rc_find_free_temporary(s->Compiler); dstreg.WriteMask = sourced; s->Descr->BuildSwizzle(s, dstreg, inst->SrcReg[src]); - inst = s->Program->Instructions + s->IP; inst->SrcReg[src].File = PROGRAM_TEMPORARY; inst->SrcReg[src].Index = dstreg.Index; inst->SrcReg[src].Swizzle = 0; @@ -117,37 +119,36 @@ static struct prog_instruction* track_used_srcreg(struct nqssadce_state* s, struct register_state *regstate; - if (inst->SrcReg[src].RelAddr) + if (inst->SrcReg[src].RelAddr) { regstate = get_reg_state(s, PROGRAM_ADDRESS, 0); - else + if (regstate) + regstate->Sourced |= WRITEMASK_X; + } else { regstate = get_reg_state(s, inst->SrcReg[src].File, inst->SrcReg[src].Index); - - if (regstate) - regstate->Sourced |= deswz_source & 0xf; - - return inst; + if (regstate) + regstate->Sourced |= deswz_source & 0xf; + } } -static void unalias_srcregs(struct prog_instruction *inst, GLuint oldindex, GLuint newindex) +static void unalias_srcregs(struct rc_instruction *inst, GLuint oldindex, GLuint newindex) { - int nsrc = _mesa_num_inst_src_regs(inst->Opcode); + int nsrc = _mesa_num_inst_src_regs(inst->I.Opcode); int i; for(i = 0; i < nsrc; ++i) - if (inst->SrcReg[i].File == PROGRAM_TEMPORARY && inst->SrcReg[i].Index == oldindex) - inst->SrcReg[i].Index = newindex; + if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY && inst->I.SrcReg[i].Index == oldindex) + inst->I.SrcReg[i].Index = newindex; } static void unalias_temporary(struct nqssadce_state* s, GLuint oldindex) { - GLuint newindex = _mesa_find_free_register(s->Program, PROGRAM_TEMPORARY); - int ip; - for(ip = 0; ip < s->IP; ++ip) { - struct prog_instruction* inst = s->Program->Instructions + ip; - if (inst->DstReg.File == PROGRAM_TEMPORARY && inst->DstReg.Index == oldindex) - inst->DstReg.Index = newindex; + GLuint newindex = rc_find_free_temporary(s->Compiler); + struct rc_instruction * inst; + for(inst = s->Compiler->Program.Instructions.Next; inst != s->IP; inst = inst->Next) { + if (inst->I.DstReg.File == PROGRAM_TEMPORARY && inst->I.DstReg.Index == oldindex) + inst->I.DstReg.Index = newindex; unalias_srcregs(inst, oldindex, newindex); } - unalias_srcregs(s->Program->Instructions + s->IP, oldindex, newindex); + unalias_srcregs(s->IP, oldindex, newindex); } @@ -156,7 +157,8 @@ static void unalias_temporary(struct nqssadce_state* s, GLuint oldindex) */ static void process_instruction(struct nqssadce_state* s) { - struct prog_instruction *inst = s->Program->Instructions + s->IP; + struct prog_instruction *inst = &s->IP->I; + GLuint WriteMask; if (inst->Opcode == OPCODE_END) return; @@ -164,7 +166,7 @@ static void process_instruction(struct nqssadce_state* s) if (inst->Opcode != OPCODE_KIL) { struct register_state *regstate = get_reg_state(s, inst->DstReg.File, inst->DstReg.Index); if (!regstate) { - _mesa_problem(s->Ctx, "NqssaDce: bad destination register (%i[%i])\n", + rc_error(s->Compiler, "NqssaDce: bad destination register (%i[%i])\n", inst->DstReg.File, inst->DstReg.Index); return; } @@ -173,7 +175,9 @@ static void process_instruction(struct nqssadce_state* s) regstate->Sourced &= ~inst->DstReg.WriteMask; if (inst->DstReg.WriteMask == 0) { - _mesa_delete_instructions(s->Program, s->IP, 1); + struct rc_instruction * inst_remove = s->IP; + s->IP = s->IP->Prev; + rc_remove_instruction(inst_remove); return; } @@ -181,16 +185,15 @@ static void process_instruction(struct nqssadce_state* s) unalias_temporary(s, inst->DstReg.Index); } - /* Attention: Due to swizzle emulation code, the following - * might change the instruction stream under us, so we have - * to be careful with the inst pointer. */ + WriteMask = inst->DstReg.WriteMask; + switch (inst->Opcode) { case OPCODE_ARL: case OPCODE_DDX: case OPCODE_DDY: case OPCODE_FRC: case OPCODE_MOV: - inst = track_used_srcreg(s, inst, 0, inst->DstReg.WriteMask); + track_used_srcreg(s, 0, WriteMask); break; case OPCODE_ADD: case OPCODE_MAX: @@ -198,14 +201,14 @@ static void process_instruction(struct nqssadce_state* s) case OPCODE_MUL: case OPCODE_SGE: case OPCODE_SLT: - inst = track_used_srcreg(s, inst, 0, inst->DstReg.WriteMask); - inst = track_used_srcreg(s, inst, 1, inst->DstReg.WriteMask); + track_used_srcreg(s, 0, WriteMask); + track_used_srcreg(s, 1, WriteMask); break; case OPCODE_CMP: case OPCODE_MAD: - inst = track_used_srcreg(s, inst, 0, inst->DstReg.WriteMask); - inst = track_used_srcreg(s, inst, 1, inst->DstReg.WriteMask); - inst = track_used_srcreg(s, inst, 2, inst->DstReg.WriteMask); + track_used_srcreg(s, 0, WriteMask); + track_used_srcreg(s, 1, WriteMask); + track_used_srcreg(s, 2, WriteMask); break; case OPCODE_COS: case OPCODE_EX2: @@ -213,83 +216,79 @@ static void process_instruction(struct nqssadce_state* s) case OPCODE_RCP: case OPCODE_RSQ: case OPCODE_SIN: - inst = track_used_srcreg(s, inst, 0, 0x1); + track_used_srcreg(s, 0, 0x1); break; case OPCODE_DP3: - inst = track_used_srcreg(s, inst, 0, 0x7); - inst = track_used_srcreg(s, inst, 1, 0x7); + track_used_srcreg(s, 0, 0x7); + track_used_srcreg(s, 1, 0x7); break; case OPCODE_DP4: - inst = track_used_srcreg(s, inst, 0, 0xf); - inst = track_used_srcreg(s, inst, 1, 0xf); + track_used_srcreg(s, 0, 0xf); + track_used_srcreg(s, 1, 0xf); break; case OPCODE_KIL: case OPCODE_TEX: case OPCODE_TXB: case OPCODE_TXP: - inst = track_used_srcreg(s, inst, 0, 0xf); + track_used_srcreg(s, 0, 0xf); break; case OPCODE_DST: - inst = track_used_srcreg(s, inst, 0, 0x6); - inst = track_used_srcreg(s, inst, 1, 0xa); + track_used_srcreg(s, 0, 0x6); + track_used_srcreg(s, 1, 0xa); break; case OPCODE_EXP: case OPCODE_LOG: case OPCODE_POW: - inst = track_used_srcreg(s, inst, 0, 0x3); + track_used_srcreg(s, 0, 0x3); break; case OPCODE_LIT: - inst = track_used_srcreg(s, inst, 0, 0xb); + track_used_srcreg(s, 0, 0xb); break; default: - _mesa_problem(s->Ctx, "NqssaDce: Unknown opcode %d\n", inst->Opcode); + rc_error(s->Compiler, "NqssaDce: Unknown opcode %d\n", inst->Opcode); return; } + + s->IP = s->IP->Prev; } -static void calculateInputsOutputs(struct gl_program *p) +void rc_calculate_inputs_outputs(struct radeon_compiler * c) { - struct prog_instruction *inst; - GLuint InputsRead, OutputsWritten; + struct rc_instruction *inst; - inst = p->Instructions; - InputsRead = 0; - OutputsWritten = 0; - while (inst->Opcode != OPCODE_END) + c->Program.InputsRead = 0; + c->Program.OutputsWritten = 0; + + for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) { - int i, num_src_regs; + int i; + int num_src_regs = _mesa_num_inst_src_regs(inst->I.Opcode); - num_src_regs = _mesa_num_inst_src_regs(inst->Opcode); for (i = 0; i < num_src_regs; ++i) { - if (inst->SrcReg[i].File == PROGRAM_INPUT) - InputsRead |= 1 << inst->SrcReg[i].Index; + if (inst->I.SrcReg[i].File == PROGRAM_INPUT) + c->Program.InputsRead |= 1 << inst->I.SrcReg[i].Index; } - if (inst->DstReg.File == PROGRAM_OUTPUT) - OutputsWritten |= 1 << inst->DstReg.Index; - - ++inst; + if (_mesa_num_inst_dst_regs(inst->I.Opcode)) { + if (inst->I.DstReg.File == PROGRAM_OUTPUT) + c->Program.OutputsWritten |= 1 << inst->I.DstReg.Index; + } } - - p->InputsRead = InputsRead; - p->OutputsWritten = OutputsWritten; } -void radeonNqssaDce(GLcontext *ctx, struct gl_program *p, struct radeon_nqssadce_descr* descr) +void radeonNqssaDce(struct radeon_compiler * c, struct radeon_nqssadce_descr* descr, void * data) { struct nqssadce_state s; _mesa_bzero(&s, sizeof(s)); - s.Ctx = ctx; - s.Program = p; + s.Compiler = c; s.Descr = descr; + s.UserData = data; s.Descr->Init(&s); - s.IP = p->NumInstructions; + s.IP = c->Program.Instructions.Prev; - while(s.IP > 0) { - s.IP--; + while(s.IP != &c->Program.Instructions && !c->Error) process_instruction(&s); - } - calculateInputsOutputs(p); + rc_calculate_inputs_outputs(c); } diff --git a/src/mesa/drivers/dri/r300/radeon_nqssadce.h b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.h index 8626f21c25..b3fc77a35a 100644 --- a/src/mesa/drivers/dri/r300/radeon_nqssadce.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.h @@ -30,7 +30,6 @@ #include "radeon_program.h" - struct register_state { /** * Bitmask indicating which components of the register are sourced @@ -44,14 +43,13 @@ struct register_state { * read from, etc. */ struct nqssadce_state { - GLcontext *Ctx; - struct gl_program *Program; + struct radeon_compiler *Compiler; struct radeon_nqssadce_descr *Descr; /** * All instructions after this instruction pointer have been dealt with. */ - int IP; + struct rc_instruction * IP; /** * Which registers are read by subsequent instructions? @@ -59,6 +57,8 @@ struct nqssadce_state { struct register_state Temps[MAX_PROGRAM_TEMPS]; struct register_state Outputs[VERT_RESULT_MAX]; struct register_state Address; + + void * UserData; }; @@ -83,11 +83,9 @@ struct radeon_nqssadce_descr { * The transformation will work recursively on the emitted instruction(s). */ void (*BuildSwizzle)(struct nqssadce_state*, struct prog_dst_register dst, struct prog_src_register src); - - void *Data; }; -void radeonNqssaDce(GLcontext *ctx, struct gl_program *p, struct radeon_nqssadce_descr* descr); +void radeonNqssaDce(struct radeon_compiler * c, struct radeon_nqssadce_descr* descr, void * data); struct prog_src_register lmul_swizzle(GLuint swizzle, struct prog_src_register srcreg); #endif /* __RADEON_PROGRAM_NQSSADCE_H_ */ diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program.c b/src/mesa/drivers/dri/r300/compiler/radeon_program.c new file mode 100644 index 0000000000..208d3b90c8 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.c @@ -0,0 +1,181 @@ +/* + * Copyright (C) 2008 Nicolai Haehnle. + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "radeon_program.h" + +#include "radeon_compiler.h" +#include "shader/prog_parameter.h" +#include "shader/prog_print.h" + + +/** + * Transform the given clause in the following way: + * 1. Replace it with an empty clause + * 2. For every instruction in the original clause, try the given + * transformations in order. + * 3. If one of the transformations returns GL_TRUE, assume that it + * has emitted the appropriate instruction(s) into the new clause; + * otherwise, copy the instruction verbatim. + * + * \note The transformation is currently not recursive; in other words, + * instructions emitted by transformations are not transformed. + * + * \note The transform is called 'local' because it can only look at + * one instruction at a time. + */ +void radeonLocalTransform( + struct radeon_compiler * c, + int num_transformations, + struct radeon_program_transformation* transformations) +{ + struct rc_instruction * inst = c->Program.Instructions.Next; + + while(inst != &c->Program.Instructions) { + struct rc_instruction * current = inst; + int i; + + inst = inst->Next; + + for(i = 0; i < num_transformations; ++i) { + struct radeon_program_transformation* t = transformations + i; + + if (t->function(c, current, t->userData)) + break; + } + } +} + + +GLint rc_find_free_temporary(struct radeon_compiler * c) +{ + GLboolean used[MAX_PROGRAM_TEMPS]; + GLuint i; + + memset(used, 0, sizeof(used)); + + for (struct rc_instruction * rcinst = c->Program.Instructions.Next; rcinst != &c->Program.Instructions; rcinst = rcinst->Next) { + const struct prog_instruction *inst = &rcinst->I; + const GLuint n = _mesa_num_inst_src_regs(inst->Opcode); + GLuint k; + + for (k = 0; k < n; k++) { + if (inst->SrcReg[k].File == PROGRAM_TEMPORARY) + used[inst->SrcReg[k].Index] = GL_TRUE; + } + } + + for (i = 0; i < MAX_PROGRAM_TEMPS; i++) { + if (!used[i]) + return i; + } + + return -1; +} + + +struct rc_instruction *rc_alloc_instruction(struct radeon_compiler * c) +{ + struct rc_instruction * inst = memory_pool_malloc(&c->Pool, sizeof(struct rc_instruction)); + + inst->Prev = 0; + inst->Next = 0; + + _mesa_init_instructions(&inst->I, 1); + + return inst; +} + + +struct rc_instruction *rc_insert_new_instruction(struct radeon_compiler * c, struct rc_instruction * after) +{ + struct rc_instruction * inst = rc_alloc_instruction(c); + + inst->Prev = after; + inst->Next = after->Next; + + inst->Prev->Next = inst; + inst->Next->Prev = inst; + + return inst; +} + +void rc_remove_instruction(struct rc_instruction * inst) +{ + inst->Prev->Next = inst->Next; + inst->Next->Prev = inst->Prev; +} + + +void rc_mesa_to_rc_program(struct radeon_compiler * c, struct gl_program * program) +{ + struct prog_instruction *source; + unsigned int i; + + for(source = program->Instructions; source->Opcode != OPCODE_END; ++source) { + struct rc_instruction * dest = rc_insert_new_instruction(c, c->Program.Instructions.Prev); + dest->I = *source; + } + + c->Program.ShadowSamplers = program->ShadowSamplers; + c->Program.InputsRead = program->InputsRead; + c->Program.OutputsWritten = program->OutputsWritten; + + for(i = 0; i < program->Parameters->NumParameters; ++i) { + struct rc_constant constant; + + constant.Type = RC_CONSTANT_EXTERNAL; + constant.Size = 4; + constant.u.External = i; + + rc_constants_add(&c->Program.Constants, &constant); + } +} + + +/** + * Print program to stderr, default options. + */ +void rc_print_program(const struct rc_program *prog) +{ + GLuint indent = 0; + GLuint linenum = 1; + struct rc_instruction *inst; + + fprintf(stderr, "# Radeon Compiler Program\n"); + + for(inst = prog->Instructions.Next; inst != &prog->Instructions; inst = inst->Next) { + fprintf(stderr, "%3d: ", linenum); + + /* Massive hack: We rely on the fact that the printers do not actually + * use the gl_program argument (last argument) in debug mode */ + indent = _mesa_fprint_instruction_opt( + stderr, &inst->I, + indent, PROG_PRINT_DEBUG, 0); + + linenum++; + } +} diff --git a/src/mesa/drivers/dri/r300/radeon_program.h b/src/mesa/drivers/dri/r300/compiler/radeon_program.h index 88474d43a2..561958608c 100644 --- a/src/mesa/drivers/dri/r300/radeon_program.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.h @@ -34,12 +34,9 @@ #include "shader/program.h" #include "shader/prog_instruction.h" - -enum { - CLAUSE_MIXED = 0, - CLAUSE_ALU, - CLAUSE_TEX -}; +struct radeon_compiler; +struct rc_instruction; +struct rc_program; enum { PROGRAM_BUILTIN = PROGRAM_FILE_MAX /**< not a real register, but a special swizzle constant */ @@ -83,19 +80,12 @@ static inline GLuint combine_swizzles(GLuint src, GLuint swz) return ret; } +static INLINE void reset_srcreg(struct prog_src_register* reg) +{ + _mesa_bzero(reg, sizeof(*reg)); + reg->Swizzle = SWIZZLE_NOOP; +} -/** - * Transformation context that is passed to local transformations. - * - * Care must be taken with some operations during transformation, - * e.g. finding new temporary registers must use @ref radeonFindFreeTemporary - */ -struct radeon_transform_context { - GLcontext *Ctx; - struct gl_program *Program; - struct prog_instruction *OldInstructions; - GLuint OldNumInstructions; -}; /** * A transformation that can be passed to \ref radeonLocalTransform. @@ -109,23 +99,23 @@ struct radeon_transform_context { */ struct radeon_program_transformation { GLboolean (*function)( - struct radeon_transform_context*, - struct prog_instruction*, + struct radeon_compiler*, + struct rc_instruction*, void*); void *userData; }; void radeonLocalTransform( - GLcontext* ctx, - struct gl_program *program, + struct radeon_compiler *c, int num_transformations, struct radeon_program_transformation* transformations); -/** - * Find a usable free temporary register during program transformation - */ -GLint radeonFindFreeTemporary(struct radeon_transform_context *ctx); +GLint rc_find_free_temporary(struct radeon_compiler * c); + +struct rc_instruction *rc_alloc_instruction(struct radeon_compiler * c); +struct rc_instruction *rc_insert_new_instruction(struct radeon_compiler * c, struct rc_instruction * after); +void rc_remove_instruction(struct rc_instruction * inst); -struct prog_instruction *radeonAppendInstructions(struct gl_program *program, int count); +void rc_print_program(const struct rc_program *prog); #endif diff --git a/src/mesa/drivers/dri/r300/radeon_program_alu.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c index 8283723bad..609e510ff2 100644 --- a/src/mesa/drivers/dri/r300/radeon_program_alu.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c @@ -35,49 +35,52 @@ #include "radeon_program_alu.h" -#include "shader/prog_parameter.h" +#include "radeon_compiler.h" -static struct prog_instruction *emit1(struct gl_program* p, +static struct rc_instruction *emit1( + struct radeon_compiler * c, struct rc_instruction * after, gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg, struct prog_src_register SrcReg) { - struct prog_instruction *fpi = radeonAppendInstructions(p, 1); + struct rc_instruction *fpi = rc_insert_new_instruction(c, after); - fpi->Opcode = Opcode; - fpi->SaturateMode = Saturate; - fpi->DstReg = DstReg; - fpi->SrcReg[0] = SrcReg; + fpi->I.Opcode = Opcode; + fpi->I.SaturateMode = Saturate; + fpi->I.DstReg = DstReg; + fpi->I.SrcReg[0] = SrcReg; return fpi; } -static struct prog_instruction *emit2(struct gl_program* p, +static struct rc_instruction *emit2( + struct radeon_compiler * c, struct rc_instruction * after, gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg, struct prog_src_register SrcReg0, struct prog_src_register SrcReg1) { - struct prog_instruction *fpi = radeonAppendInstructions(p, 1); + struct rc_instruction *fpi = rc_insert_new_instruction(c, after); - fpi->Opcode = Opcode; - fpi->SaturateMode = Saturate; - fpi->DstReg = DstReg; - fpi->SrcReg[0] = SrcReg0; - fpi->SrcReg[1] = SrcReg1; + fpi->I.Opcode = Opcode; + fpi->I.SaturateMode = Saturate; + fpi->I.DstReg = DstReg; + fpi->I.SrcReg[0] = SrcReg0; + fpi->I.SrcReg[1] = SrcReg1; return fpi; } -static struct prog_instruction *emit3(struct gl_program* p, +static struct rc_instruction *emit3( + struct radeon_compiler * c, struct rc_instruction * after, gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg, struct prog_src_register SrcReg0, struct prog_src_register SrcReg1, struct prog_src_register SrcReg2) { - struct prog_instruction *fpi = radeonAppendInstructions(p, 1); - - fpi->Opcode = Opcode; - fpi->SaturateMode = Saturate; - fpi->DstReg = DstReg; - fpi->SrcReg[0] = SrcReg0; - fpi->SrcReg[1] = SrcReg1; - fpi->SrcReg[2] = SrcReg2; + struct rc_instruction *fpi = rc_insert_new_instruction(c, after); + + fpi->I.Opcode = Opcode; + fpi->I.SaturateMode = Saturate; + fpi->I.DstReg = DstReg; + fpi->I.SrcReg[0] = SrcReg0; + fpi->I.SrcReg[1] = SrcReg1; + fpi->I.SrcReg[2] = SrcReg2; return fpi; } @@ -171,44 +174,63 @@ static struct prog_src_register scalar(struct prog_src_register reg) return swizzle(reg, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X); } -static void transform_ABS(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_ABS(struct radeon_compiler* c, + struct rc_instruction* inst) { - struct prog_src_register src = inst->SrcReg[0]; + struct prog_src_register src = inst->I.SrcReg[0]; src.Abs = 1; src.Negate = NEGATE_NONE; - emit1(t->Program, OPCODE_MOV, inst->SaturateMode, inst->DstReg, src); + emit1(c, inst->Prev, OPCODE_MOV, inst->I.SaturateMode, inst->I.DstReg, src); + rc_remove_instruction(inst); } -static void transform_DPH(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_DP3(struct radeon_compiler* c, + struct rc_instruction* inst) { - struct prog_src_register src0 = inst->SrcReg[0]; + struct prog_src_register src0 = inst->I.SrcReg[0]; + struct prog_src_register src1 = inst->I.SrcReg[1]; + src0.Negate &= ~NEGATE_W; + src0.Swizzle &= ~(7 << (3 * 3)); + src0.Swizzle |= SWIZZLE_ZERO << (3 * 3); + src1.Negate &= ~NEGATE_W; + src1.Swizzle &= ~(7 << (3 * 3)); + src1.Swizzle |= SWIZZLE_ZERO << (3 * 3); + emit2(c, inst->Prev, OPCODE_DP4, inst->I.SaturateMode, inst->I.DstReg, src0, src1); + rc_remove_instruction(inst); +} + +static void transform_DPH(struct radeon_compiler* c, + struct rc_instruction* inst) +{ + struct prog_src_register src0 = inst->I.SrcReg[0]; src0.Negate &= ~NEGATE_W; src0.Swizzle &= ~(7 << (3 * 3)); src0.Swizzle |= SWIZZLE_ONE << (3 * 3); - emit2(t->Program, OPCODE_DP4, inst->SaturateMode, inst->DstReg, src0, inst->SrcReg[1]); + emit2(c, inst->Prev, OPCODE_DP4, inst->I.SaturateMode, inst->I.DstReg, src0, inst->I.SrcReg[1]); + rc_remove_instruction(inst); } /** * [1, src0.y*src1.y, src0.z, src1.w] * So basically MUL with lotsa swizzling. */ -static void transform_DST(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_DST(struct radeon_compiler* c, + struct rc_instruction* inst) { - emit2(t->Program, OPCODE_MUL, inst->SaturateMode, inst->DstReg, - swizzle(inst->SrcReg[0], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE), - swizzle(inst->SrcReg[1], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_ONE, SWIZZLE_W)); + emit2(c, inst->Prev, OPCODE_MUL, inst->I.SaturateMode, inst->I.DstReg, + swizzle(inst->I.SrcReg[0], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE), + swizzle(inst->I.SrcReg[1], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_ONE, SWIZZLE_W)); + rc_remove_instruction(inst); } -static void transform_FLR(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_FLR(struct radeon_compiler* c, + struct rc_instruction* inst) { - int tempreg = radeonFindFreeTemporary(t); - emit1(t->Program, OPCODE_FRC, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->SrcReg[0]); - emit2(t->Program, OPCODE_ADD, inst->SaturateMode, inst->DstReg, - inst->SrcReg[0], negate(srcreg(PROGRAM_TEMPORARY, tempreg))); + int tempreg = rc_find_free_temporary(c); + emit1(c, inst->Prev, OPCODE_FRC, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0]); + emit2(c, inst->Prev, OPCODE_ADD, inst->I.SaturateMode, inst->I.DstReg, + inst->I.SrcReg[0], negate(srcreg(PROGRAM_TEMPORARY, tempreg))); + rc_remove_instruction(inst); } /** @@ -229,152 +251,159 @@ static void transform_FLR(struct radeon_transform_context* t, * 5 slots, if the subsequent optimization passes are clever enough * to pair instructions correctly. */ -static void transform_LIT(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_LIT(struct radeon_compiler* c, + struct rc_instruction* inst) { - static const GLfloat LitConst[4] = { -127.999999 }; - GLuint constant; GLuint constant_swizzle; GLuint temp; - int needTemporary = 0; struct prog_src_register srctemp; - constant = _mesa_add_unnamed_constant(t->Program->Parameters, LitConst, 1, &constant_swizzle); + constant = rc_constants_add_immediate_scalar(&c->Program.Constants, -127.999999, &constant_swizzle); - if (inst->DstReg.WriteMask != WRITEMASK_XYZW) { - needTemporary = 1; - } else if (inst->DstReg.File != PROGRAM_TEMPORARY) { - // LIT is typically followed by DP3/DP4, so there's no point - // in creating special code for this case - needTemporary = 1; - } + if (inst->I.DstReg.WriteMask != WRITEMASK_XYZW || inst->I.DstReg.File != PROGRAM_TEMPORARY) { + struct rc_instruction * inst_mov; - if (needTemporary) { - temp = radeonFindFreeTemporary(t); - } else { - temp = inst->DstReg.Index; + inst_mov = emit1(c, inst, + OPCODE_MOV, 0, inst->I.DstReg, + srcreg(PROGRAM_TEMPORARY, rc_find_free_temporary(c))); + + inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index; + inst->I.DstReg.WriteMask = WRITEMASK_XYZW; } + + temp = inst->I.DstReg.Index; srctemp = srcreg(PROGRAM_TEMPORARY, temp); // tmp.x = max(0.0, Src.x); // tmp.y = max(0.0, Src.y); // tmp.w = clamp(Src.z, -128+eps, 128-eps); - emit2(t->Program, OPCODE_MAX, 0, + emit2(c, inst->Prev, OPCODE_MAX, 0, dstregtmpmask(temp, WRITEMASK_XYW), - inst->SrcReg[0], + inst->I.SrcReg[0], swizzle(srcreg(PROGRAM_CONSTANT, constant), SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO, constant_swizzle&3)); - emit2(t->Program, OPCODE_MIN, 0, + emit2(c, inst->Prev, OPCODE_MIN, 0, dstregtmpmask(temp, WRITEMASK_Z), swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), negate(srcregswz(PROGRAM_CONSTANT, constant, constant_swizzle))); // tmp.w = Pow(tmp.y, tmp.w) - emit1(t->Program, OPCODE_LG2, 0, + emit1(c, inst->Prev, OPCODE_LG2, 0, dstregtmpmask(temp, WRITEMASK_W), swizzle(srctemp, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y)); - emit2(t->Program, OPCODE_MUL, 0, + emit2(c, inst->Prev, OPCODE_MUL, 0, dstregtmpmask(temp, WRITEMASK_W), swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), swizzle(srctemp, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)); - emit1(t->Program, OPCODE_EX2, 0, + emit1(c, inst->Prev, OPCODE_EX2, 0, dstregtmpmask(temp, WRITEMASK_W), swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W)); // tmp.z = (tmp.x > 0) ? tmp.w : 0.0 - emit3(t->Program, OPCODE_CMP, inst->SaturateMode, + emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode, dstregtmpmask(temp, WRITEMASK_Z), negate(swizzle(srctemp, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)), swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), builtin_zero); // tmp.x, tmp.y, tmp.w = 1.0, tmp.x, 1.0 - emit1(t->Program, OPCODE_MOV, inst->SaturateMode, + emit1(c, inst->Prev, OPCODE_MOV, inst->I.SaturateMode, dstregtmpmask(temp, WRITEMASK_XYW), swizzle(srctemp, SWIZZLE_ONE, SWIZZLE_X, SWIZZLE_ONE, SWIZZLE_ONE)); - if (needTemporary) - emit1(t->Program, OPCODE_MOV, 0, inst->DstReg, srctemp); + rc_remove_instruction(inst); } -static void transform_LRP(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_LRP(struct radeon_compiler* c, + struct rc_instruction* inst) { - int tempreg = radeonFindFreeTemporary(t); + int tempreg = rc_find_free_temporary(c); - emit2(t->Program, OPCODE_ADD, 0, + emit2(c, inst->Prev, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), - inst->SrcReg[1], negate(inst->SrcReg[2])); - emit3(t->Program, OPCODE_MAD, inst->SaturateMode, - inst->DstReg, - inst->SrcReg[0], srcreg(PROGRAM_TEMPORARY, tempreg), inst->SrcReg[2]); + inst->I.SrcReg[1], negate(inst->I.SrcReg[2])); + emit3(c, inst->Prev, OPCODE_MAD, inst->I.SaturateMode, + inst->I.DstReg, + inst->I.SrcReg[0], srcreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[2]); + + rc_remove_instruction(inst); } -static void transform_POW(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_POW(struct radeon_compiler* c, + struct rc_instruction* inst) { - int tempreg = radeonFindFreeTemporary(t); + int tempreg = rc_find_free_temporary(c); struct prog_dst_register tempdst = dstreg(PROGRAM_TEMPORARY, tempreg); struct prog_src_register tempsrc = srcreg(PROGRAM_TEMPORARY, tempreg); tempdst.WriteMask = WRITEMASK_W; tempsrc.Swizzle = SWIZZLE_WWWW; - emit1(t->Program, OPCODE_LG2, 0, tempdst, scalar(inst->SrcReg[0])); - emit2(t->Program, OPCODE_MUL, 0, tempdst, tempsrc, scalar(inst->SrcReg[1])); - emit1(t->Program, OPCODE_EX2, inst->SaturateMode, inst->DstReg, tempsrc); + emit1(c, inst->Prev, OPCODE_LG2, 0, tempdst, scalar(inst->I.SrcReg[0])); + emit2(c, inst->Prev, OPCODE_MUL, 0, tempdst, tempsrc, scalar(inst->I.SrcReg[1])); + emit1(c, inst->Prev, OPCODE_EX2, inst->I.SaturateMode, inst->I.DstReg, tempsrc); + + rc_remove_instruction(inst); } -static void transform_RSQ(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_RSQ(struct radeon_compiler* c, + struct rc_instruction* inst) { - emit1(t->Program, OPCODE_RSQ, inst->SaturateMode, inst->DstReg, absolute(inst->SrcReg[0])); + inst->I.SrcReg[0] = absolute(inst->I.SrcReg[0]); } -static void transform_SGE(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_SGE(struct radeon_compiler* c, + struct rc_instruction* inst) { - int tempreg = radeonFindFreeTemporary(t); + int tempreg = rc_find_free_temporary(c); - emit2(t->Program, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->SrcReg[0], negate(inst->SrcReg[1])); - emit3(t->Program, OPCODE_CMP, inst->SaturateMode, inst->DstReg, + emit2(c, inst->Prev, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0], negate(inst->I.SrcReg[1])); + emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode, inst->I.DstReg, srcreg(PROGRAM_TEMPORARY, tempreg), builtin_zero, builtin_one); + + rc_remove_instruction(inst); } -static void transform_SLT(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_SLT(struct radeon_compiler* c, + struct rc_instruction* inst) { - int tempreg = radeonFindFreeTemporary(t); + int tempreg = rc_find_free_temporary(c); - emit2(t->Program, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->SrcReg[0], negate(inst->SrcReg[1])); - emit3(t->Program, OPCODE_CMP, inst->SaturateMode, inst->DstReg, + emit2(c, inst->Prev, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0], negate(inst->I.SrcReg[1])); + emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode, inst->I.DstReg, srcreg(PROGRAM_TEMPORARY, tempreg), builtin_one, builtin_zero); + + rc_remove_instruction(inst); } -static void transform_SUB(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_SUB(struct radeon_compiler* c, + struct rc_instruction* inst) { - emit2(t->Program, OPCODE_ADD, inst->SaturateMode, inst->DstReg, inst->SrcReg[0], negate(inst->SrcReg[1])); + inst->I.Opcode = OPCODE_ADD; + inst->I.SrcReg[1] = negate(inst->I.SrcReg[1]); } -static void transform_SWZ(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_SWZ(struct radeon_compiler* c, + struct rc_instruction* inst) { - emit1(t->Program, OPCODE_MOV, inst->SaturateMode, inst->DstReg, inst->SrcReg[0]); + inst->I.Opcode = OPCODE_MOV; } -static void transform_XPD(struct radeon_transform_context* t, - struct prog_instruction* inst) +static void transform_XPD(struct radeon_compiler* c, + struct rc_instruction* inst) { - int tempreg = radeonFindFreeTemporary(t); - - emit2(t->Program, OPCODE_MUL, 0, dstreg(PROGRAM_TEMPORARY, tempreg), - swizzle(inst->SrcReg[0], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W), - swizzle(inst->SrcReg[1], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W)); - emit3(t->Program, OPCODE_MAD, inst->SaturateMode, inst->DstReg, - swizzle(inst->SrcReg[0], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W), - swizzle(inst->SrcReg[1], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W), + int tempreg = rc_find_free_temporary(c); + + emit2(c, inst->Prev, OPCODE_MUL, 0, dstreg(PROGRAM_TEMPORARY, tempreg), + swizzle(inst->I.SrcReg[0], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W), + swizzle(inst->I.SrcReg[1], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W)); + emit3(c, inst->Prev, OPCODE_MAD, inst->I.SaturateMode, inst->I.DstReg, + swizzle(inst->I.SrcReg[0], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W), + swizzle(inst->I.SrcReg[1], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W), negate(srcreg(PROGRAM_TEMPORARY, tempreg))); + + rc_remove_instruction(inst); } @@ -392,31 +421,64 @@ static void transform_XPD(struct radeon_transform_context* t, * * @note should be applicable to R300 and R500 fragment programs. */ -GLboolean radeonTransformALU(struct radeon_transform_context* t, - struct prog_instruction* inst, +GLboolean radeonTransformALU( + struct radeon_compiler * c, + struct rc_instruction* inst, void* unused) { - switch(inst->Opcode) { - case OPCODE_ABS: transform_ABS(t, inst); return GL_TRUE; - case OPCODE_DPH: transform_DPH(t, inst); return GL_TRUE; - case OPCODE_DST: transform_DST(t, inst); return GL_TRUE; - case OPCODE_FLR: transform_FLR(t, inst); return GL_TRUE; - case OPCODE_LIT: transform_LIT(t, inst); return GL_TRUE; - case OPCODE_LRP: transform_LRP(t, inst); return GL_TRUE; - case OPCODE_POW: transform_POW(t, inst); return GL_TRUE; - case OPCODE_RSQ: transform_RSQ(t, inst); return GL_TRUE; - case OPCODE_SGE: transform_SGE(t, inst); return GL_TRUE; - case OPCODE_SLT: transform_SLT(t, inst); return GL_TRUE; - case OPCODE_SUB: transform_SUB(t, inst); return GL_TRUE; - case OPCODE_SWZ: transform_SWZ(t, inst); return GL_TRUE; - case OPCODE_XPD: transform_XPD(t, inst); return GL_TRUE; + switch(inst->I.Opcode) { + case OPCODE_ABS: transform_ABS(c, inst); return GL_TRUE; + case OPCODE_DPH: transform_DPH(c, inst); return GL_TRUE; + case OPCODE_DST: transform_DST(c, inst); return GL_TRUE; + case OPCODE_FLR: transform_FLR(c, inst); return GL_TRUE; + case OPCODE_LIT: transform_LIT(c, inst); return GL_TRUE; + case OPCODE_LRP: transform_LRP(c, inst); return GL_TRUE; + case OPCODE_POW: transform_POW(c, inst); return GL_TRUE; + case OPCODE_RSQ: transform_RSQ(c, inst); return GL_TRUE; + case OPCODE_SGE: transform_SGE(c, inst); return GL_TRUE; + case OPCODE_SLT: transform_SLT(c, inst); return GL_TRUE; + case OPCODE_SUB: transform_SUB(c, inst); return GL_TRUE; + case OPCODE_SWZ: transform_SWZ(c, inst); return GL_TRUE; + case OPCODE_XPD: transform_XPD(c, inst); return GL_TRUE; default: return GL_FALSE; } } -static void sincos_constants(struct radeon_transform_context* t, GLuint *constants) +static void transform_r300_vertex_ABS(struct radeon_compiler* c, + struct rc_instruction* inst) +{ + /* Note: r500 can take absolute values, but r300 cannot. */ + inst->I.Opcode = OPCODE_MAX; + inst->I.SrcReg[1] = inst->I.SrcReg[0]; + inst->I.SrcReg[1].Negate ^= NEGATE_XYZW; +} + +/** + * For use with radeonLocalTransform, this transforms non-native ALU + * instructions of the r300 up to r500 vertex engine. + */ +GLboolean r300_transform_vertex_alu( + struct radeon_compiler * c, + struct rc_instruction* inst, + void* unused) +{ + switch(inst->I.Opcode) { + case OPCODE_ABS: transform_r300_vertex_ABS(c, inst); return GL_TRUE; + case OPCODE_DP3: transform_DP3(c, inst); return GL_TRUE; + case OPCODE_DPH: transform_DPH(c, inst); return GL_TRUE; + case OPCODE_FLR: transform_FLR(c, inst); return GL_TRUE; + case OPCODE_LRP: transform_LRP(c, inst); return GL_TRUE; + case OPCODE_SUB: transform_SUB(c, inst); return GL_TRUE; + case OPCODE_SWZ: transform_SWZ(c, inst); return GL_TRUE; + case OPCODE_XPD: transform_XPD(c, inst); return GL_TRUE; + default: + return GL_FALSE; + } +} + +static void sincos_constants(struct radeon_compiler* c, GLuint *constants) { static const GLfloat SinCosConsts[2][4] = { { @@ -434,11 +496,8 @@ static void sincos_constants(struct radeon_transform_context* t, GLuint *constan }; int i; - for(i = 0; i < 2; ++i) { - GLuint swz; - constants[i] = _mesa_add_unnamed_constant(t->Program->Parameters, SinCosConsts[i], 4, &swz); - ASSERT(swz == SWIZZLE_NOOP); - } + for(i = 0; i < 2; ++i) + constants[i] = rc_constants_add_immediate_vec4(&c->Program.Constants, SinCosConsts[i]); } /** @@ -449,23 +508,24 @@ static void sincos_constants(struct radeon_transform_context* t, GLuint *constan * MAD tmp.y, tmp.x, |tmp.x|, -tmp.x * MAD dest, tmp.y, weight, tmp.x */ -static void sin_approx(struct radeon_transform_context* t, +static void sin_approx( + struct radeon_compiler* c, struct rc_instruction * after, struct prog_dst_register dst, struct prog_src_register src, const GLuint* constants) { - GLuint tempreg = radeonFindFreeTemporary(t); + GLuint tempreg = rc_find_free_temporary(c); - emit2(t->Program, OPCODE_MUL, 0, dstregtmpmask(tempreg, WRITEMASK_XY), + emit2(c, after->Prev, OPCODE_MUL, 0, dstregtmpmask(tempreg, WRITEMASK_XY), swizzle(src, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), srcreg(PROGRAM_CONSTANT, constants[0])); - emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_X), + emit3(c, after->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_X), swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y), absolute(swizzle(src, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)), swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)); - emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_Y), + emit3(c, after->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_Y), swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), absolute(swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)), negate(swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X))); - emit3(t->Program, OPCODE_MAD, 0, dst, + emit3(c, after->Prev, OPCODE_MAD, 0, dst, swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y), swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)); @@ -476,78 +536,80 @@ static void sin_approx(struct radeon_transform_context* t, * using only the basic instructions * MOV, ADD, MUL, MAD, FRC */ -GLboolean radeonTransformTrigSimple(struct radeon_transform_context* t, - struct prog_instruction* inst, +GLboolean radeonTransformTrigSimple(struct radeon_compiler* c, + struct rc_instruction* inst, void* unused) { - if (inst->Opcode != OPCODE_COS && - inst->Opcode != OPCODE_SIN && - inst->Opcode != OPCODE_SCS) + if (inst->I.Opcode != OPCODE_COS && + inst->I.Opcode != OPCODE_SIN && + inst->I.Opcode != OPCODE_SCS) return GL_FALSE; GLuint constants[2]; - GLuint tempreg = radeonFindFreeTemporary(t); + GLuint tempreg = rc_find_free_temporary(c); - sincos_constants(t, constants); + sincos_constants(c, constants); - if (inst->Opcode == OPCODE_COS) { + if (inst->I.Opcode == OPCODE_COS) { // MAD tmp.x, src, 1/(2*PI), 0.75 // FRC tmp.x, tmp.x // MAD tmp.z, tmp.x, 2*PI, -PI - emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W), - swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), + emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W), + swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)); - emit1(t->Program, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W), + emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W), swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W)); - emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W), + emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W), swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z))); - sin_approx(t, inst->DstReg, + sin_approx(c, inst->Prev, inst->I.DstReg, swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), constants); - } else if (inst->Opcode == OPCODE_SIN) { - emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W), - swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), + } else if (inst->I.Opcode == OPCODE_SIN) { + emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W), + swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y)); - emit1(t->Program, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W), + emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W), swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W)); - emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W), + emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W), swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z))); - sin_approx(t, inst->DstReg, + sin_approx(c, inst->Prev, inst->I.DstReg, swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), constants); } else { - emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY), - swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), + emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY), + swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W)); - emit1(t->Program, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_XY), + emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_XY), srcreg(PROGRAM_TEMPORARY, tempreg)); - emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY), + emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY), srcreg(PROGRAM_TEMPORARY, tempreg), swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z))); - struct prog_dst_register dst = inst->DstReg; + struct prog_dst_register dst = inst->I.DstReg; - dst.WriteMask = inst->DstReg.WriteMask & WRITEMASK_X; - sin_approx(t, dst, + dst.WriteMask = inst->I.DstReg.WriteMask & WRITEMASK_X; + sin_approx(c, inst->Prev, dst, swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), constants); - dst.WriteMask = inst->DstReg.WriteMask & WRITEMASK_Y; - sin_approx(t, dst, + dst.WriteMask = inst->I.DstReg.WriteMask & WRITEMASK_Y; + sin_approx(c, inst->Prev, dst, swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y), constants); } + rc_remove_instruction(inst); + return GL_TRUE; } @@ -560,50 +622,52 @@ GLboolean radeonTransformTrigSimple(struct radeon_transform_context* t, * * @warning This transformation implicitly changes the semantics of SIN and COS! */ -GLboolean radeonTransformTrigScale(struct radeon_transform_context* t, - struct prog_instruction* inst, +GLboolean radeonTransformTrigScale(struct radeon_compiler* c, + struct rc_instruction* inst, void* unused) { - if (inst->Opcode != OPCODE_COS && - inst->Opcode != OPCODE_SIN && - inst->Opcode != OPCODE_SCS) + if (inst->I.Opcode != OPCODE_COS && + inst->I.Opcode != OPCODE_SIN && + inst->I.Opcode != OPCODE_SCS) return GL_FALSE; - static const GLfloat RCP_2PI[] = { 0.15915494309189535 }; + static const GLfloat RCP_2PI = 0.15915494309189535; GLuint temp; GLuint constant; GLuint constant_swizzle; - temp = radeonFindFreeTemporary(t); - constant = _mesa_add_unnamed_constant(t->Program->Parameters, RCP_2PI, 1, &constant_swizzle); + temp = rc_find_free_temporary(c); + constant = rc_constants_add_immediate_scalar(&c->Program.Constants, RCP_2PI, &constant_swizzle); - emit2(t->Program, OPCODE_MUL, 0, dstregtmpmask(temp, WRITEMASK_W), - swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), + emit2(c, inst->Prev, OPCODE_MUL, 0, dstregtmpmask(temp, WRITEMASK_W), + swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), srcregswz(PROGRAM_CONSTANT, constant, constant_swizzle)); - emit1(t->Program, OPCODE_FRC, 0, dstregtmpmask(temp, WRITEMASK_W), + emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(temp, WRITEMASK_W), srcreg(PROGRAM_TEMPORARY, temp)); - if (inst->Opcode == OPCODE_COS) { - emit1(t->Program, OPCODE_COS, inst->SaturateMode, inst->DstReg, + if (inst->I.Opcode == OPCODE_COS) { + emit1(c, inst->Prev, OPCODE_COS, inst->I.SaturateMode, inst->I.DstReg, srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW)); - } else if (inst->Opcode == OPCODE_SIN) { - emit1(t->Program, OPCODE_SIN, inst->SaturateMode, - inst->DstReg, srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW)); - } else if (inst->Opcode == OPCODE_SCS) { - struct prog_dst_register moddst = inst->DstReg; + } else if (inst->I.Opcode == OPCODE_SIN) { + emit1(c, inst->Prev, OPCODE_SIN, inst->I.SaturateMode, + inst->I.DstReg, srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW)); + } else if (inst->I.Opcode == OPCODE_SCS) { + struct prog_dst_register moddst = inst->I.DstReg; - if (inst->DstReg.WriteMask & WRITEMASK_X) { + if (inst->I.DstReg.WriteMask & WRITEMASK_X) { moddst.WriteMask = WRITEMASK_X; - emit1(t->Program, OPCODE_COS, inst->SaturateMode, moddst, + emit1(c, inst->Prev, OPCODE_COS, inst->I.SaturateMode, moddst, srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW)); } - if (inst->DstReg.WriteMask & WRITEMASK_Y) { + if (inst->I.DstReg.WriteMask & WRITEMASK_Y) { moddst.WriteMask = WRITEMASK_Y; - emit1(t->Program, OPCODE_SIN, inst->SaturateMode, moddst, + emit1(c, inst->Prev, OPCODE_SIN, inst->I.SaturateMode, moddst, srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW)); } } + rc_remove_instruction(inst); + return GL_TRUE; } @@ -615,21 +679,15 @@ GLboolean radeonTransformTrigScale(struct radeon_transform_context* t, * @warning This explicitly changes the form of DDX and DDY! */ -GLboolean radeonTransformDeriv(struct radeon_transform_context* t, - struct prog_instruction* inst, +GLboolean radeonTransformDeriv(struct radeon_compiler* c, + struct rc_instruction* inst, void* unused) { - if (inst->Opcode != OPCODE_DDX && inst->Opcode != OPCODE_DDY) + if (inst->I.Opcode != OPCODE_DDX && inst->I.Opcode != OPCODE_DDY) return GL_FALSE; - struct prog_src_register B = inst->SrcReg[1]; - - B.Swizzle = MAKE_SWIZZLE4(SWIZZLE_ONE, SWIZZLE_ONE, - SWIZZLE_ONE, SWIZZLE_ONE); - B.Negate = NEGATE_XYZW; - - emit2(t->Program, inst->Opcode, inst->SaturateMode, inst->DstReg, - inst->SrcReg[0], B); + inst->I.SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE); + inst->I.SrcReg[1].Negate = NEGATE_XYZW; return GL_TRUE; } diff --git a/src/mesa/drivers/dri/r300/radeon_program_alu.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h index b45958115c..147efec6fc 100644 --- a/src/mesa/drivers/dri/r300/radeon_program_alu.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h @@ -31,23 +31,28 @@ #include "radeon_program.h" GLboolean radeonTransformALU( - struct radeon_transform_context *t, - struct prog_instruction*, + struct radeon_compiler * c, + struct rc_instruction * inst, + void*); + +GLboolean r300_transform_vertex_alu( + struct radeon_compiler * c, + struct rc_instruction * inst, void*); GLboolean radeonTransformTrigSimple( - struct radeon_transform_context *t, - struct prog_instruction*, + struct radeon_compiler * c, + struct rc_instruction * inst, void*); GLboolean radeonTransformTrigScale( - struct radeon_transform_context *t, - struct prog_instruction*, + struct radeon_compiler * c, + struct rc_instruction * inst, void*); GLboolean radeonTransformDeriv( - struct radeon_transform_context *t, - struct prog_instruction*, + struct radeon_compiler * c, + struct rc_instruction * inst, void*); #endif /* __RADEON_PROGRAM_ALU_H_ */ diff --git a/src/mesa/drivers/dri/r300/radeon_program_pair.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.c index d6fb474cf2..48616ac461 100644 --- a/src/mesa/drivers/dri/r300/radeon_program_pair.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.c @@ -35,17 +35,19 @@ #include "radeon_program_pair.h" -#include "radeon_common.h" - +#include "memory_pool.h" +#include "radeon_compiler.h" #include "shader/prog_print.h" #define error(fmt, args...) do { \ - _mesa_problem(s->Ctx, "%s::%s(): " fmt "\n", \ + rc_error(&s->Compiler->Base, "%s::%s(): " fmt "\n", \ __FILE__, __FUNCTION__, ##args); \ - s->Error = GL_TRUE; \ } while(0) struct pair_state_instruction { + struct prog_instruction Instruction; + GLuint IP; /**< Position of this instruction in original program */ + GLuint IsTex:1; /**< Is a texture instruction */ GLuint NeedRGB:1; /**< Needs the RGB ALU */ GLuint NeedAlpha:1; /**< Needs the Alpha ALU */ @@ -73,7 +75,7 @@ struct pair_state_instruction { * Used to keep track of which instructions read a value. */ struct reg_value_reader { - GLuint IP; /**< IP of the instruction that performs this access */ + struct pair_state_instruction *Reader; struct reg_value_reader *Next; }; @@ -82,7 +84,7 @@ struct reg_value_reader { * PROGRAM_TEMPORARY. */ struct reg_value { - GLuint IP; /**< IP of the instruction that writes this value */ + struct pair_state_instruction *Writer; struct reg_value *Next; /**< Pointer to the next value to be written to the same PROGRAM_TEMPORARY component */ /** @@ -116,11 +118,8 @@ struct pair_register_translation { }; struct pair_state { - GLcontext *Ctx; - struct gl_program *Program; + struct r300_fragment_program_compiler * Compiler; const struct radeon_pair_handler *Handler; - GLboolean Error; - GLboolean Debug; GLboolean Verbose; void *UserData; @@ -130,11 +129,6 @@ struct pair_state { struct pair_register_translation Inputs[FRAG_ATTRIB_MAX]; struct pair_register_translation Temps[MAX_PROGRAM_TEMPS]; - /** - * Derived information about program instructions. - */ - struct pair_state_instruction *Instructions; - struct { GLuint RefCount; /**< # of times this occurs in an unscheduled SrcReg or DstReg */ } HwTemps[128]; @@ -147,14 +141,6 @@ struct pair_state { struct pair_state_instruction *ReadyRGB; struct pair_state_instruction *ReadyAlpha; struct pair_state_instruction *ReadyTEX; - - /** - * Pool of @ref reg_value structures for fast allocation. - */ - struct reg_value *ValuePool; - GLuint ValuePoolUsed; - struct reg_value_reader *ReaderPool; - GLuint ReaderPoolUsed; }; @@ -183,7 +169,7 @@ static GLuint get_hw_reg(struct pair_state *s, GLuint file, GLuint index) struct pair_register_translation *t = get_register(s, file, index); if (!t) { - _mesa_problem(s->Ctx, "get_hw_reg: %i[%i]\n", file, index); + error("get_hw_reg: %i[%i]\n", file, index); return 0; } @@ -221,15 +207,13 @@ static void add_pairinst_to_list(struct pair_state_instruction **list, struct pa } /** - * The instruction at the given IP has become ready. Link it into the ready + * The given instruction has become ready. Link it into the ready * instructions. */ -static void instruction_ready(struct pair_state *s, int ip) +static void instruction_ready(struct pair_state *s, struct pair_state_instruction *pairinst) { - struct pair_state_instruction *pairinst = s->Instructions + ip; - if (s->Verbose) - _mesa_printf("instruction_ready(%i)\n", ip); + _mesa_printf("instruction_ready(%i)\n", pairinst->IP); if (pairinst->IsTex) add_pairinst_to_list(&s->ReadyTEX, pairinst); @@ -296,12 +280,12 @@ static void final_rewrite(struct pair_state *s, struct prog_instruction *inst) * Classify an instruction according to which ALUs etc. it needs */ static void classify_instruction(struct pair_state *s, - struct prog_instruction *inst, struct pair_state_instruction *pairinst) + struct pair_state_instruction *psi) { - pairinst->NeedRGB = (inst->DstReg.WriteMask & WRITEMASK_XYZ) ? 1 : 0; - pairinst->NeedAlpha = (inst->DstReg.WriteMask & WRITEMASK_W) ? 1 : 0; + psi->NeedRGB = (psi->Instruction.DstReg.WriteMask & WRITEMASK_XYZ) ? 1 : 0; + psi->NeedAlpha = (psi->Instruction.DstReg.WriteMask & WRITEMASK_W) ? 1 : 0; - switch(inst->Opcode) { + switch(psi->Instruction.Opcode) { case OPCODE_ADD: case OPCODE_CMP: case OPCODE_DDX: @@ -319,24 +303,24 @@ static void classify_instruction(struct pair_state *s, case OPCODE_RCP: case OPCODE_RSQ: case OPCODE_SIN: - pairinst->IsTranscendent = 1; - pairinst->NeedAlpha = 1; + psi->IsTranscendent = 1; + psi->NeedAlpha = 1; break; case OPCODE_DP4: - pairinst->NeedAlpha = 1; + psi->NeedAlpha = 1; /* fall through */ case OPCODE_DP3: - pairinst->NeedRGB = 1; + psi->NeedRGB = 1; break; case OPCODE_KIL: case OPCODE_TEX: case OPCODE_TXB: case OPCODE_TXP: case OPCODE_END: - pairinst->IsTex = 1; + psi->IsTex = 1; break; default: - error("Unknown opcode %d\n", inst->Opcode); + error("Unknown opcode %d\n", psi->Instruction.Opcode); break; } } @@ -348,30 +332,34 @@ static void classify_instruction(struct pair_state *s, */ static void scan_instructions(struct pair_state *s) { - struct prog_instruction *inst; - struct pair_state_instruction *pairinst; + struct rc_instruction *source; GLuint ip; - for(inst = s->Program->Instructions, pairinst = s->Instructions, ip = 0; - inst->Opcode != OPCODE_END; - ++inst, ++pairinst, ++ip) { - final_rewrite(s, inst); - classify_instruction(s, inst, pairinst); + for(source = s->Compiler->Base.Program.Instructions.Next, ip = 0; + source != &s->Compiler->Base.Program.Instructions; + source = source->Next, ++ip) { + struct pair_state_instruction *pairinst = memory_pool_malloc(&s->Compiler->Base.Pool, sizeof(*pairinst)); + memset(pairinst, 0, sizeof(struct pair_state_instruction)); + + pairinst->Instruction = source->I; + pairinst->IP = ip; + final_rewrite(s, &pairinst->Instruction); + classify_instruction(s, pairinst); - int nsrc = _mesa_num_inst_src_regs(inst->Opcode); + int nsrc = _mesa_num_inst_src_regs(pairinst->Instruction.Opcode); int j; for(j = 0; j < nsrc; j++) { struct pair_register_translation *t = - get_register(s, inst->SrcReg[j].File, inst->SrcReg[j].Index); + get_register(s, pairinst->Instruction.SrcReg[j].File, pairinst->Instruction.SrcReg[j].Index); if (!t) continue; t->RefCount++; - if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) { + if (pairinst->Instruction.SrcReg[j].File == PROGRAM_TEMPORARY) { int i; for(i = 0; i < 4; ++i) { - GLuint swz = GET_SWZ(inst->SrcReg[j].Swizzle, i); + GLuint swz = GET_SWZ(pairinst->Instruction.SrcReg[j].Swizzle, i); if (swz >= 4) continue; /* constant or NIL swizzle */ if (!t->Value[swz]) @@ -381,36 +369,37 @@ static void scan_instructions(struct pair_state *s) * also rewrites the value. The code below adds * a dependency for the DstReg, which is a superset * of the SrcReg dependency. */ - if (inst->DstReg.File == PROGRAM_TEMPORARY && - inst->DstReg.Index == inst->SrcReg[j].Index && - GET_BIT(inst->DstReg.WriteMask, swz)) + if (pairinst->Instruction.DstReg.File == PROGRAM_TEMPORARY && + pairinst->Instruction.DstReg.Index == pairinst->Instruction.SrcReg[j].Index && + GET_BIT(pairinst->Instruction.DstReg.WriteMask, swz)) continue; - struct reg_value_reader* r = &s->ReaderPool[s->ReaderPoolUsed++]; + struct reg_value_reader* r = memory_pool_malloc(&s->Compiler->Base.Pool, sizeof(*r)); pairinst->NumDependencies++; t->Value[swz]->NumReaders++; - r->IP = ip; + r->Reader = pairinst; r->Next = t->Value[swz]->Readers; t->Value[swz]->Readers = r; } } } - int ndst = _mesa_num_inst_dst_regs(inst->Opcode); + int ndst = _mesa_num_inst_dst_regs(pairinst->Instruction.Opcode); if (ndst) { struct pair_register_translation *t = - get_register(s, inst->DstReg.File, inst->DstReg.Index); + get_register(s, pairinst->Instruction.DstReg.File, pairinst->Instruction.DstReg.Index); if (t) { t->RefCount++; - if (inst->DstReg.File == PROGRAM_TEMPORARY) { + if (pairinst->Instruction.DstReg.File == PROGRAM_TEMPORARY) { int j; for(j = 0; j < 4; ++j) { - if (!GET_BIT(inst->DstReg.WriteMask, j)) + if (!GET_BIT(pairinst->Instruction.DstReg.WriteMask, j)) continue; - struct reg_value* v = &s->ValuePool[s->ValuePoolUsed++]; - v->IP = ip; + struct reg_value* v = memory_pool_malloc(&s->Compiler->Base.Pool, sizeof(*v)); + memset(v, 0, sizeof(struct reg_value)); + v->Writer = pairinst; if (t->Value[j]) { pairinst->NumDependencies++; t->Value[j]->Next = v; @@ -426,7 +415,7 @@ static void scan_instructions(struct pair_state *s) _mesa_printf("scan(%i): NumDeps = %i\n", ip, pairinst->NumDependencies); if (!pairinst->NumDependencies) - instruction_ready(s, ip); + instruction_ready(s, pairinst); } /* Clear the PROGRAM_TEMPORARY state */ @@ -438,70 +427,23 @@ static void scan_instructions(struct pair_state *s) } -/** - * Reserve hardware temporary registers for the program inputs. - * - * @note This allocation is performed explicitly, because the order of inputs - * is determined by the RS hardware. - */ -static void allocate_input_registers(struct pair_state *s) -{ - GLuint InputsRead = s->Program->InputsRead; - int i; - GLuint hwindex = 0; - - /* Primary colour */ - if (InputsRead & FRAG_BIT_COL0) - alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_COL0, hwindex++); - InputsRead &= ~FRAG_BIT_COL0; - - /* Secondary color */ - if (InputsRead & FRAG_BIT_COL1) - alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_COL1, hwindex++); - InputsRead &= ~FRAG_BIT_COL1; - - /* Texcoords */ - for (i = 0; i < s->Ctx->Const.MaxTextureUnits; i++) { - if (InputsRead & (FRAG_BIT_TEX0 << i)) - alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_TEX0+i, hwindex++); - } - InputsRead &= ~FRAG_BITS_TEX_ANY; - - /* Fogcoords treated as a texcoord */ - if (InputsRead & FRAG_BIT_FOGC) - alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_FOGC, hwindex++); - InputsRead &= ~FRAG_BIT_FOGC; - - /* fragment position treated as a texcoord */ - if (InputsRead & FRAG_BIT_WPOS) - alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_WPOS, hwindex++); - InputsRead &= ~FRAG_BIT_WPOS; - - /* Anything else */ - if (InputsRead) - error("Don't know how to handle inputs 0x%x\n", InputsRead); -} - - -static void decrement_dependencies(struct pair_state *s, int ip) +static void decrement_dependencies(struct pair_state *s, struct pair_state_instruction *pairinst) { - struct pair_state_instruction *pairinst = s->Instructions + ip; ASSERT(pairinst->NumDependencies > 0); if (!--pairinst->NumDependencies) - instruction_ready(s, ip); + instruction_ready(s, pairinst); } /** * Update the dependency tracking state based on what the instruction * at the given IP does. */ -static void commit_instruction(struct pair_state *s, int ip) +static void commit_instruction(struct pair_state *s, struct pair_state_instruction *pairinst) { - struct prog_instruction *inst = s->Program->Instructions + ip; - struct pair_state_instruction *pairinst = s->Instructions + ip; + struct prog_instruction *inst = &pairinst->Instruction; if (s->Verbose) - _mesa_printf("commit_instruction(%i)\n", ip); + _mesa_printf("commit_instruction(%i)\n", pairinst->IP); if (inst->DstReg.File == PROGRAM_TEMPORARY) { struct pair_register_translation *t = &s->Temps[inst->DstReg.Index]; @@ -516,11 +458,11 @@ static void commit_instruction(struct pair_state *s, int ip) if (t->Value[i]->NumReaders) { struct reg_value_reader *r; for(r = pairinst->Values[i]->Readers; r; r = r->Next) - decrement_dependencies(s, r->IP); + decrement_dependencies(s, r->Reader); } else if (t->Value[i]->Next) { /* This happens when the only reader writes * the register at the same time */ - decrement_dependencies(s, t->Value[i]->Next->IP); + decrement_dependencies(s, t->Value[i]->Next->Writer); } } } @@ -554,7 +496,7 @@ static void commit_instruction(struct pair_state *s, int ip) if (!--t->Value[swz]->NumReaders) { if (t->Value[swz]->Next) - decrement_dependencies(s, t->Value[swz]->Next->IP); + decrement_dependencies(s, t->Value[swz]->Next->Writer); } } } @@ -585,36 +527,52 @@ static void emit_all_tex(struct pair_state *s) // Allocate destination hardware registers in one block to avoid conflicts. for(pairinst = readytex; pairinst; pairinst = pairinst->NextReady) { - int ip = pairinst - s->Instructions; - struct prog_instruction *inst = s->Program->Instructions + ip; + struct prog_instruction *inst = &pairinst->Instruction; if (inst->Opcode != OPCODE_KIL) get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index); } - if (s->Debug) + if (s->Compiler->Base.Debug) _mesa_printf(" BEGIN_TEX\n"); if (s->Handler->BeginTexBlock) - s->Error = s->Error || !s->Handler->BeginTexBlock(s->UserData); + s->Compiler->Base.Error = s->Compiler->Base.Error || !s->Handler->BeginTexBlock(s->UserData); for(pairinst = readytex; pairinst; pairinst = pairinst->NextReady) { - int ip = pairinst - s->Instructions; - struct prog_instruction *inst = s->Program->Instructions + ip; - commit_instruction(s, ip); + struct prog_instruction *inst = &pairinst->Instruction; + commit_instruction(s, pairinst); if (inst->Opcode != OPCODE_KIL) inst->DstReg.Index = get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index); inst->SrcReg[0].Index = get_hw_reg(s, inst->SrcReg[0].File, inst->SrcReg[0].Index); - if (s->Debug) { + if (s->Compiler->Base.Debug) { _mesa_printf(" "); _mesa_print_instruction(inst); fflush(stdout); } - s->Error = s->Error || !s->Handler->EmitTex(s->UserData, inst); + + struct radeon_pair_texture_instruction rpti; + + switch(inst->Opcode) { + case OPCODE_TEX: rpti.Opcode = RADEON_OPCODE_TEX; break; + case OPCODE_TXB: rpti.Opcode = RADEON_OPCODE_TXB; break; + case OPCODE_TXP: rpti.Opcode = RADEON_OPCODE_TXP; break; + default: + case OPCODE_KIL: rpti.Opcode = RADEON_OPCODE_KIL; break; + } + + rpti.DestIndex = inst->DstReg.Index; + rpti.WriteMask = inst->DstReg.WriteMask; + rpti.TexSrcUnit = inst->TexSrcUnit; + rpti.TexSrcTarget = inst->TexSrcTarget; + rpti.SrcIndex = inst->SrcReg[0].Index; + rpti.SrcSwizzle = inst->SrcReg[0].Swizzle; + + s->Compiler->Base.Error = s->Compiler->Base.Error || !s->Handler->EmitTex(s->UserData, &rpti); } - if (s->Debug) + if (s->Compiler->Base.Debug) _mesa_printf(" END_TEX\n"); } @@ -637,7 +595,7 @@ static int alloc_pair_source(struct pair_state *s, struct radeon_pair_instructio index = get_hw_reg(s, src.File, src.Index); } else { constant = 1; - s->Error |= !s->Handler->EmitConst(s->UserData, src.File, src.Index, &index); + index = src.Index; } for(i = 0; i < 3; ++i) { @@ -684,10 +642,12 @@ static int alloc_pair_source(struct pair_state *s, struct radeon_pair_instructio * Fill the given ALU instruction's opcodes and source operands into the given pair, * if possible. */ -static GLboolean fill_instruction_into_pair(struct pair_state *s, struct radeon_pair_instruction *pair, int ip) +static GLboolean fill_instruction_into_pair( + struct pair_state *s, + struct radeon_pair_instruction *pair, + struct pair_state_instruction *pairinst) { - struct pair_state_instruction *pairinst = s->Instructions + ip; - struct prog_instruction *inst = s->Program->Instructions + ip; + struct prog_instruction *inst = &pairinst->Instruction; ASSERT(!pairinst->NeedRGB || pair->RGB.Opcode == OPCODE_NOP); ASSERT(!pairinst->NeedAlpha || pair->Alpha.Opcode == OPCODE_NOP); @@ -768,16 +728,18 @@ static GLboolean fill_instruction_into_pair(struct pair_state *s, struct radeon_ * we are absolutely certain that we're going to emit a certain * instruction pairing. */ -static void fill_dest_into_pair(struct pair_state *s, struct radeon_pair_instruction *pair, int ip) +static void fill_dest_into_pair( + struct pair_state *s, + struct radeon_pair_instruction *pair, + struct pair_state_instruction *pairinst) { - struct pair_state_instruction *pairinst = s->Instructions + ip; - struct prog_instruction *inst = s->Program->Instructions + ip; + struct prog_instruction *inst = &pairinst->Instruction; if (inst->DstReg.File == PROGRAM_OUTPUT) { - if (inst->DstReg.Index == FRAG_RESULT_COLOR) { + if (inst->DstReg.Index == s->Compiler->OutputColor) { pair->RGB.OutputWriteMask |= inst->DstReg.WriteMask & WRITEMASK_XYZ; pair->Alpha.OutputWriteMask |= GET_BIT(inst->DstReg.WriteMask, 3); - } else if (inst->DstReg.Index == FRAG_RESULT_DEPTH) { + } else if (inst->DstReg.Index == s->Compiler->OutputDepth) { pair->Alpha.DepthWriteMask |= GET_BIT(inst->DstReg.WriteMask, 3); } } else { @@ -804,24 +766,24 @@ static void fill_dest_into_pair(struct pair_state *s, struct radeon_pair_instruc static void emit_alu(struct pair_state *s) { struct radeon_pair_instruction pair; + struct pair_state_instruction *psi; if (s->ReadyFullALU || !(s->ReadyRGB && s->ReadyAlpha)) { - int ip; if (s->ReadyFullALU) { - ip = s->ReadyFullALU - s->Instructions; + psi = s->ReadyFullALU; s->ReadyFullALU = s->ReadyFullALU->NextReady; } else if (s->ReadyRGB) { - ip = s->ReadyRGB - s->Instructions; + psi = s->ReadyRGB; s->ReadyRGB = s->ReadyRGB->NextReady; } else { - ip = s->ReadyAlpha - s->Instructions; + psi = s->ReadyAlpha; s->ReadyAlpha = s->ReadyAlpha->NextReady; } _mesa_bzero(&pair, sizeof(pair)); - fill_instruction_into_pair(s, &pair, ip); - fill_dest_into_pair(s, &pair, ip); - commit_instruction(s, ip); + fill_instruction_into_pair(s, &pair, psi); + fill_dest_into_pair(s, &pair, psi); + commit_instruction(s, psi); } else { struct pair_state_instruction **prgb; struct pair_state_instruction **palpha; @@ -830,65 +792,65 @@ static void emit_alu(struct pair_state *s) * many source slots; try all possible pairings if necessary */ for(prgb = &s->ReadyRGB; *prgb; prgb = &(*prgb)->NextReady) { for(palpha = &s->ReadyAlpha; *palpha; palpha = &(*palpha)->NextReady) { - int rgbip = *prgb - s->Instructions; - int alphaip = *palpha - s->Instructions; + struct pair_state_instruction * psirgb = *prgb; + struct pair_state_instruction * psialpha = *palpha; _mesa_bzero(&pair, sizeof(pair)); - fill_instruction_into_pair(s, &pair, rgbip); - if (!fill_instruction_into_pair(s, &pair, alphaip)) + fill_instruction_into_pair(s, &pair, psirgb); + if (!fill_instruction_into_pair(s, &pair, psialpha)) continue; *prgb = (*prgb)->NextReady; *palpha = (*palpha)->NextReady; - fill_dest_into_pair(s, &pair, rgbip); - fill_dest_into_pair(s, &pair, alphaip); - commit_instruction(s, rgbip); - commit_instruction(s, alphaip); + fill_dest_into_pair(s, &pair, psirgb); + fill_dest_into_pair(s, &pair, psialpha); + commit_instruction(s, psirgb); + commit_instruction(s, psialpha); goto success; } } /* No success in pairing; just take the first RGB instruction */ - int ip = s->ReadyRGB - s->Instructions; + psi = s->ReadyRGB; s->ReadyRGB = s->ReadyRGB->NextReady; + _mesa_bzero(&pair, sizeof(pair)); - fill_instruction_into_pair(s, &pair, ip); - fill_dest_into_pair(s, &pair, ip); - commit_instruction(s, ip); + fill_instruction_into_pair(s, &pair, psi); + fill_dest_into_pair(s, &pair, psi); + commit_instruction(s, psi); success: ; } - if (s->Debug) + if (s->Compiler->Base.Debug) radeonPrintPairInstruction(&pair); - s->Error = s->Error || !s->Handler->EmitPaired(s->UserData, &pair); + s->Compiler->Base.Error = s->Compiler->Base.Error || !s->Handler->EmitPaired(s->UserData, &pair); } +/* Callback function for assigning input registers to hardware registers */ +static void alloc_helper(void * data, unsigned input, unsigned hwreg) +{ + struct pair_state * s = data; + alloc_hw_reg(s, PROGRAM_INPUT, input, hwreg); +} -GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program, +void radeonPairProgram( + struct r300_fragment_program_compiler * compiler, const struct radeon_pair_handler* handler, void *userdata) { struct pair_state s; _mesa_bzero(&s, sizeof(s)); - s.Ctx = ctx; - s.Program = _mesa_clone_program(ctx, program); + s.Compiler = compiler; s.Handler = handler; s.UserData = userdata; - s.Debug = (RADEON_DEBUG & DEBUG_PIXEL) ? GL_TRUE : GL_FALSE; - s.Verbose = GL_FALSE && s.Debug; + s.Verbose = GL_FALSE && s.Compiler->Base.Debug; - s.Instructions = (struct pair_state_instruction*)_mesa_calloc( - sizeof(struct pair_state_instruction)*s.Program->NumInstructions); - s.ValuePool = (struct reg_value*)_mesa_calloc(sizeof(struct reg_value)*s.Program->NumInstructions*4); - s.ReaderPool = (struct reg_value_reader*)_mesa_calloc( - sizeof(struct reg_value_reader)*s.Program->NumInstructions*12); - - if (s.Debug) + if (s.Compiler->Base.Debug) _mesa_printf("Emit paired program\n"); scan_instructions(&s); - allocate_input_registers(&s); + s.Compiler->AllocateHwInputs(s.Compiler, &alloc_helper, &s); - while(!s.Error && + while(!s.Compiler->Base.Error && (s.ReadyTEX || s.ReadyRGB || s.ReadyAlpha || s.ReadyFullALU)) { if (s.ReadyTEX) emit_all_tex(&s); @@ -897,16 +859,8 @@ GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program, emit_alu(&s); } - if (s.Debug) + if (s.Compiler->Base.Debug) _mesa_printf(" END\n"); - - _mesa_free(s.Instructions); - _mesa_free(s.ValuePool); - _mesa_free(s.ReaderPool); - - _mesa_reference_program(ctx, &s.Program, NULL); - - return !s.Error; } diff --git a/src/mesa/drivers/dri/r300/radeon_program_pair.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h index 4624a24629..ff76178551 100644 --- a/src/mesa/drivers/dri/r300/radeon_program_pair.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h @@ -30,6 +30,8 @@ #include "radeon_program.h" +struct r300_fragment_program_compiler; + /** * Represents a paired instruction, as found in R300 and R500 @@ -82,18 +84,32 @@ struct radeon_pair_instruction { }; +enum { + RADEON_OPCODE_TEX = 0, + RADEON_OPCODE_TXB, + RADEON_OPCODE_TXP, + RADEON_OPCODE_KIL +}; + +struct radeon_pair_texture_instruction { + GLuint Opcode:2; /**< one of RADEON_OPCODE_xxx */ + + GLuint DestIndex:8; + GLuint WriteMask:4; + + GLuint TexSrcUnit:5; + GLuint TexSrcTarget:3; + + GLuint SrcIndex:8; + GLuint SrcSwizzle:12; +}; + + /** * */ struct radeon_pair_handler { /** - * Fill in the proper hardware index for the given constant register. - * - * @return GL_FALSE on error. - */ - GLboolean (*EmitConst)(void*, GLuint file, GLuint index, GLuint *hwindex); - - /** * Write a paired instruction to the hardware. * * @return GL_FALSE on error. @@ -107,7 +123,7 @@ struct radeon_pair_handler { * * @return GL_FALSE on error. */ - GLboolean (*EmitTex)(void*, struct prog_instruction*); + GLboolean (*EmitTex)(void*, struct radeon_pair_texture_instruction*); /** * Called before a block of contiguous, independent texture @@ -115,10 +131,11 @@ struct radeon_pair_handler { */ GLboolean (*BeginTexBlock)(void*); - GLuint MaxHwTemps; + unsigned MaxHwTemps; }; -GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program, +void radeonPairProgram( + struct r300_fragment_program_compiler * compiler, const struct radeon_pair_handler*, void *userdata); void radeonPrintPairInstruction(struct radeon_pair_instruction *inst); diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c index af535037d0..bd46f9acf2 100644 --- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c +++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c @@ -164,47 +164,46 @@ static void emit_tex_offsets(GLcontext *ctx, struct radeon_state_atom * atom) r300ContextPtr r300 = R300_CONTEXT(ctx); BATCH_LOCALS(&r300->radeon); int numtmus = packet0_count(r300, r300->hw.tex.offset.cmd); - int notexture = 0; - - if (numtmus) { - int i; - - for(i = 0; i < numtmus; ++i) { - radeonTexObj *t = r300->hw.textures[i]; - - if (!t) - notexture = 1; - } - - if (r300->radeon.radeonScreen->kernel_mm && notexture) { - return; - } - for(i = 0; i < numtmus; ++i) { - radeonTexObj *t = r300->hw.textures[i]; - if (t && !t->image_override) { - BEGIN_BATCH_NO_AUTOSTATE(4); - OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); - OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0, - RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); - END_BATCH(); - } else if (!t) { - /* Texture unit hasn't a texture bound nothings to do */ - } else { /* override cases */ - if (t->bo) { - BEGIN_BATCH_NO_AUTOSTATE(4); - OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); - OUT_BATCH_RELOC(t->tile_bits, t->bo, 0, - RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); - END_BATCH(); - } else if (!r300->radeon.radeonScreen->kernel_mm) { - BEGIN_BATCH_NO_AUTOSTATE(2); - OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); - OUT_BATCH(t->override_offset); - END_BATCH(); - } else { - /* Texture unit hasn't a texture bound nothings to do */ - } - } + int i; + + for(i = 0; i < numtmus; ++i) { + radeonTexObj *t = r300->hw.textures[i]; + if (t && !t->image_override) { + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); + OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + END_BATCH(); + } else if (!t) { + /* Texture unit hasn't a texture bound. + * We assign the current color buffer as a fakery to make + * KIL work on KMS (without it, the CS checker will complain). + */ + if (r300->radeon.radeonScreen->kernel_mm) { + struct radeon_renderbuffer *rrb = radeon_get_colorbuffer(&r300->radeon); + if (rrb && rrb->bo) { + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); + OUT_BATCH_RELOC(0, rrb->bo, 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + END_BATCH(); + } + } + } else { /* override cases */ + if (t->bo) { + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); + OUT_BATCH_RELOC(t->tile_bits, t->bo, 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + END_BATCH(); + } else if (!r300->radeon.radeonScreen->kernel_mm) { + BEGIN_BATCH_NO_AUTOSTATE(2); + OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); + OUT_BATCH(t->override_offset); + END_BATCH(); + } else { + /* Texture unit hasn't a texture bound nothings to do */ + } } } } diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c index 6f3aab986d..91fa77a169 100644 --- a/src/mesa/drivers/dri/r300/r300_context.c +++ b/src/mesa/drivers/dri/r300/r300_context.c @@ -64,9 +64,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_ioctl.h" #include "r300_tex.h" #include "r300_emit.h" -#include "r300_render.h" +#include "r300_queryobj.h" #include "r300_swtcl.h" #include "radeon_bocs_wrapper.h" +#include "radeon_buffer_objects.h" #include "vblank.h" @@ -74,6 +75,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "xmlpool.h" /* for symbolic values of enum-type options */ #define need_GL_VERSION_2_0 +#define need_GL_ARB_occlusion_query #define need_GL_ARB_point_parameters #define need_GL_ARB_vertex_program #define need_GL_EXT_blend_equation_separate @@ -94,6 +96,7 @@ const struct dri_extension card_extensions[] = { /* *INDENT-OFF* */ {"GL_ARB_depth_texture", NULL}, {"GL_ARB_fragment_program", NULL}, + {"GL_ARB_occlusion_query", GL_ARB_occlusion_query_functions}, {"GL_ARB_multitexture", NULL}, {"GL_ARB_point_parameters", GL_ARB_point_parameters_functions}, {"GL_ARB_shadow", NULL}, @@ -154,7 +157,6 @@ const struct dri_extension gl_20_extension[] = { }; static const struct tnl_pipeline_stage *r300_pipeline[] = { - /* Catch any t&l fallbacks */ &_tnl_vertex_transform_stage, @@ -165,21 +167,7 @@ static const struct tnl_pipeline_stage *r300_pipeline[] = { &_tnl_texture_transform_stage, &_tnl_point_attenuation_stage, &_tnl_vertex_program_stage, - - /* Try again to go to tcl? - * - no good for asymmetric-twoside (do with multipass) - * - no good for asymmetric-unfilled (do with multipass) - * - good for material - * - good for texgen - * - need to manipulate a bit of state - * - * - worth it/not worth it? - */ - - /* Else do them here. - */ - &_r300_render_stage, - &_tnl_render_stage, /* FALLBACK */ + &_tnl_render_stage, 0, }; @@ -325,6 +313,11 @@ static void r300InitConstValues(GLcontext *ctx, radeonScreenPtr screen) ctx->Const.FragmentProgram.MaxNativeTexIndirections = R300_PFS_MAX_TEX_INDIRECT; ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0; } + + if (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV530) + r300->num_z_pipes = 2; + else + r300->num_z_pipes = r300->radeon.radeonScreen->num_gb_pipes; } static void r300ParseOptions(r300ContextPtr r300, radeonScreenPtr screen) @@ -367,6 +360,11 @@ static void r300InitGLExtensions(GLcontext *ctx) } else if (r300->options.s3tc_force_disabled) { _mesa_disable_extension(ctx, "GL_EXT_texture_compression_s3tc"); } + + if (!r300->radeon.radeonScreen->drmSupportsOcclusionQueries || + !r300->options.hw_tcl_enabled) { + _mesa_disable_extension(ctx, "GL_ARB_occlusion_query"); + } } /* Create the device specific rendering context. @@ -398,6 +396,8 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual, r300InitStateFuncs(&functions); r300InitTextureFuncs(&functions); r300InitShaderFuncs(&functions); + r300InitQueryObjFunctions(&functions); + radeonInitBufferObjectFuncs(&functions); if (!radeonInitContext(&r300->radeon, &functions, glVisual, driContextPriv, @@ -451,13 +451,10 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual, r300InitState(r300); r300InitShaderFunctions(r300); - if (screen->chip_family == CHIP_FAMILY_RS600 || screen->chip_family == CHIP_FAMILY_RS690 || - screen->chip_family == CHIP_FAMILY_RS740) { - r300->radeon.texture_row_align = 64; - } - r300InitGLExtensions(ctx); + make_empty_list(&r300->query.not_flushed_head); + return GL_TRUE; } diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h index f7af7d4e57..3ba3426608 100644 --- a/src/mesa/drivers/dri/r300/r300_context.h +++ b/src/mesa/drivers/dri/r300/r300_context.h @@ -44,6 +44,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/mtypes.h" #include "shader/prog_instruction.h" +#include "compiler/radeon_code.h" struct r300_context; typedef struct r300_context r300ContextRec; @@ -389,46 +390,25 @@ struct r300_hw_state { /* Vertex shader state */ -/* Perhaps more if we store programs in vmem? */ -/* drm_r300_cmd_header_t->vpu->count is unsigned char */ -#define VSF_MAX_FRAGMENT_LENGTH (255*4) - -/* Can be tested with colormat currently. */ -#define VSF_MAX_FRAGMENT_TEMPS (14) - -#define STATE_R300_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0) -#define STATE_R300_TEXRECT_FACTOR (STATE_INTERNAL_DRIVER+1) - #define COLOR_IS_RGBA #define TAG(x) r300##x #include "tnl_dd/t_dd_vertex.h" #undef TAG +struct r300_vertex_program_key { + GLbitfield FpReads; + GLuint FogAttr; + GLuint WPosAttr; +}; + struct r300_vertex_program { struct gl_vertex_program *Base; struct r300_vertex_program *next; - struct r300_vertex_program_key { - GLuint FpReads; - GLuint FogAttr; - GLuint WPosAttr; - } key; - - struct r300_vertex_shader_hw_code { - int length; - union { - GLuint d[VSF_MAX_FRAGMENT_LENGTH]; - float f[VSF_MAX_FRAGMENT_LENGTH]; - } body; - } hw_code; - - GLboolean translated; - GLboolean error; + struct r300_vertex_program_key key; + struct r300_vertex_program_code code; - int pos_end; - int num_temporaries; /* Number of temp vars used by program */ - int inputs[VERT_ATTRIB_MAX]; - int outputs[VERT_RESULT_MAX]; + GLboolean error; }; struct r300_vertex_program_cont { @@ -441,131 +421,18 @@ struct r300_vertex_program_cont { struct r300_vertex_program *progs; }; -#define R300_PFS_MAX_ALU_INST 64 -#define R300_PFS_MAX_TEX_INST 32 -#define R300_PFS_MAX_TEX_INDIRECT 4 -#define R300_PFS_NUM_TEMP_REGS 32 -#define R300_PFS_NUM_CONST_REGS 32 - -#define R500_PFS_MAX_INST 512 -#define R500_PFS_NUM_TEMP_REGS 128 -#define R500_PFS_NUM_CONST_REGS 256 - -struct r300_pfs_compile_state; -struct r500_pfs_compile_state; - -/** - * Stores state that influences the compilation of a fragment program. - */ -struct r300_fragment_program_external_state { - struct { - /** - * If the sampler is used as a shadow sampler, - * this field is: - * 0 - GL_LUMINANCE - * 1 - GL_INTENSITY - * 2 - GL_ALPHA - * depending on the depth texture mode. - */ - GLuint depth_texture_mode : 2; - - /** - * If the sampler is used as a shadow sampler, - * this field is (texture_compare_func - GL_NEVER). - * [e.g. if compare function is GL_LEQUAL, this field is 3] - * - * Otherwise, this field is 0. - */ - GLuint texture_compare_func : 3; - } unit[16]; -}; - - -struct r300_fragment_program_node { - int tex_offset; /**< first tex instruction */ - int tex_end; /**< last tex instruction, relative to tex_offset */ - int alu_offset; /**< first ALU instruction */ - int alu_end; /**< last ALU instruction, relative to alu_offset */ - int flags; -}; - -/** - * Stores an R300 fragment program in its compiled-to-hardware form. - */ -struct r300_fragment_program_code { - struct { - int length; /**< total # of texture instructions used */ - GLuint inst[R300_PFS_MAX_TEX_INST]; - } tex; - - struct { - int length; /**< total # of ALU instructions used */ - struct { - GLuint inst0; - GLuint inst1; - GLuint inst2; - GLuint inst3; - } inst[R300_PFS_MAX_ALU_INST]; - } alu; - - struct r300_fragment_program_node node[4]; - int cur_node; - int first_node_has_tex; - - /** - * Remember which program register a given hardware constant - * belongs to. - */ - struct prog_src_register constant[R300_PFS_NUM_CONST_REGS]; - int const_nr; - - int max_temp_idx; -}; - - -struct r500_fragment_program_code { - struct { - GLuint inst0; - GLuint inst1; - GLuint inst2; - GLuint inst3; - GLuint inst4; - GLuint inst5; - } inst[R500_PFS_MAX_INST]; - - int inst_offset; - int inst_end; - - /** - * Remember which program register a given hardware constant - * belongs to. - */ - struct prog_src_register constant[R500_PFS_NUM_CONST_REGS]; - int const_nr; - - int max_temp_idx; -}; /** * Store everything about a fragment program that is needed * to render with that program. */ struct r300_fragment_program { - struct gl_program *Base; - - GLboolean translated; GLboolean error; - + struct r300_fragment_program *next; struct r300_fragment_program_external_state state; - union rX00_fragment_program_code { - struct r300_fragment_program_code r300; - struct r500_fragment_program_code r500; - } code; - GLboolean writes_depth; - GLuint optimization; - - struct r300_fragment_program *next; + struct rX00_fragment_program_code code; + GLbitfield InputsRead; /* attribute that we are sending the WPOS in */ gl_frag_attrib wpos_attr; @@ -583,12 +450,6 @@ struct r300_fragment_program_cont { struct r300_fragment_program *progs; }; -struct r300_fragment_program_compiler { - r300ContextPtr r300; - struct r300_fragment_program *fp; - union rX00_fragment_program_code *code; - struct gl_program *program; -}; #define R300_MAX_AOS_ARRAYS 16 @@ -610,8 +471,6 @@ struct r300_swtcl_info { struct r300_vtable { void (* SetupRSUnit)(GLcontext *ctx); void (* SetupFragmentShaderTextures)(GLcontext *ctx, int *tmu_mappings); - GLboolean (* BuildFragmentProgramHwCode)(struct r300_fragment_program_compiler *compiler); - void (* FragmentProgramDump)(union rX00_fragment_program_code *code); void (* SetupPixelShader)(GLcontext *ctx); }; @@ -619,11 +478,12 @@ struct r300_vertex_buffer { struct vertex_attribute { /* generic */ GLubyte element; - GLvoid *data; - GLboolean free_needed; GLuint stride; GLuint dwords; GLubyte size; /* number of components */ + GLboolean is_named_bo; + struct radeon_bo *bo; + GLint bo_offset; /* hw specific */ uint32_t data_type:4; @@ -638,12 +498,23 @@ struct r300_vertex_buffer { }; struct r300_index_buffer { - GLvoid *ptr; + struct radeon_bo *bo; + int bo_offset; + GLboolean is_32bit; - GLboolean free_needed; GLuint count; }; +struct r300_query_object { + struct gl_query_object Base; + struct radeon_bo *bo; + int curr_offset; + GLboolean emitted_begin; + + /* Double linked list of not flushed query objects */ + struct r300_query_object *prev, *next; +}; + /** * \brief R300 context structure. */ @@ -669,7 +540,7 @@ struct r300_context { uint32_t s3tc_force_disabled:1; uint32_t stencil_two_side_disabled:1; } options; - + struct r300_swtcl_info swtcl; struct r300_vertex_buffer vbuf; struct r300_index_buffer ind_buf; @@ -678,6 +549,13 @@ struct r300_context { uint32_t fallback; DECLARE_RENDERINPUTS(render_inputs_bitset); + + struct { + struct r300_query_object *current; + struct r300_query_object not_flushed_head; + } query; + + int num_z_pipes; }; #define R300_CONTEXT(ctx) ((r300ContextPtr)(ctx->DriverCtx)) diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c index 9769ff5399..d524d60299 100644 --- a/src/mesa/drivers/dri/r300/r300_draw.c +++ b/src/mesa/drivers/dri/r300/r300_draw.c @@ -36,41 +36,67 @@ #include "r300_context.h" #include "r300_emit.h" #include "r300_render.h" +#include "r300_queryobj.h" #include "r300_state.h" #include "r300_tex.h" +#include "radeon_buffer_objects.h" + #include "tnl/tnl.h" #include "tnl/t_vp_build.h" #include "vbo/vbo_context.h" #include "swrast/swrast.h" #include "swrast_setup/swrast_setup.h" -static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *mesa_ind_buf, struct gl_buffer_object **bo, GLuint *nr_bo) + +static int getTypeSize(GLenum type) +{ + switch (type) { + case GL_DOUBLE: + return sizeof(GLdouble); + case GL_FLOAT: + return sizeof(GLfloat); + case GL_INT: + return sizeof(GLint); + case GL_UNSIGNED_INT: + return sizeof(GLuint); + case GL_SHORT: + return sizeof(GLshort); + case GL_UNSIGNED_SHORT: + return sizeof(GLushort); + case GL_BYTE: + return sizeof(GLbyte); + case GL_UNSIGNED_BYTE: + return sizeof(GLubyte); + default: + assert(0); + return 0; + } +} + +static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *mesa_ind_buf) { r300ContextPtr r300 = R300_CONTEXT(ctx); - struct r300_index_buffer *ind_buf = &r300->ind_buf; GLvoid *src_ptr; + GLuint *out; + int i; + GLboolean mapped_named_bo = GL_FALSE; - if (!mesa_ind_buf) { - ind_buf->ptr = NULL; - return; - } - - ind_buf->count = mesa_ind_buf->count; if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) { - bo[*nr_bo] = mesa_ind_buf->obj; - (*nr_bo)++; ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY_ARB, mesa_ind_buf->obj); + mapped_named_bo = GL_TRUE; assert(mesa_ind_buf->obj->Pointer != NULL); } src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr); if (mesa_ind_buf->type == GL_UNSIGNED_BYTE) { + GLuint size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1); GLubyte *in = (GLubyte *)src_ptr; - GLuint *out = _mesa_malloc(sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1)); - int i; - ind_buf->ptr = out; + radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4); + + assert(r300->ind_buf.bo->ptr != NULL); + out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset); for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) { *out++ = in[i] | in[i + 1] << 16; @@ -80,16 +106,16 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *out++ = in[i]; } - ind_buf->free_needed = GL_TRUE; - ind_buf->is_32bit = GL_FALSE; - } else if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) { #if MESA_BIG_ENDIAN + } else { /* if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) */ + GLuint size; GLushort *in = (GLushort *)src_ptr; - GLuint *out = _mesa_malloc(sizeof(GLushort) * - ((mesa_ind_buf->count + 1) & ~1)); - int i; + size = sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1); + + radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offet, size, 4); - ind_buf->ptr = out; + assert(r300->ind_buf.bo->ptr != NULL) + out = (GLuint *)ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset); for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) { *out++ = in[i] | in[i + 1] << 16; @@ -98,42 +124,60 @@ static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer if (i < mesa_ind_buf->count) { *out++ = in[i]; } - - ind_buf->free_needed = GL_TRUE; -#else - ind_buf->ptr = src_ptr; - ind_buf->free_needed = GL_FALSE; #endif - ind_buf->is_32bit = GL_FALSE; - } else { - ind_buf->ptr = src_ptr; - ind_buf->free_needed = GL_FALSE; - ind_buf->is_32bit = GL_TRUE; + } + + r300->ind_buf.is_32bit = GL_FALSE; + r300->ind_buf.count = mesa_ind_buf->count; + + if (mapped_named_bo) { + ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, mesa_ind_buf->obj); } } -static int getTypeSize(GLenum type) + +static void r300SetupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *mesa_ind_buf) { - switch (type) { - case GL_DOUBLE: - return sizeof(GLdouble); - case GL_FLOAT: - return sizeof(GLfloat); - case GL_INT: - return sizeof(GLint); - case GL_UNSIGNED_INT: - return sizeof(GLuint); - case GL_SHORT: - return sizeof(GLshort); - case GL_UNSIGNED_SHORT: - return sizeof(GLushort); - case GL_BYTE: - return sizeof(GLbyte); - case GL_UNSIGNED_BYTE: - return sizeof(GLubyte); - default: - assert(0); - return 0; + r300ContextPtr r300 = R300_CONTEXT(ctx); + + if (!mesa_ind_buf) { + r300->ind_buf.bo = NULL; + return; + } + +#if MESA_BIG_ENDIAN + if (mesa_ind_buf->type == GL_UNSIGNED_INT) { +#else + if (mesa_ind_buf->type != GL_UNSIGNED_BYTE) { +#endif + const GLvoid *src_ptr; + GLvoid *dst_ptr; + GLboolean mapped_named_bo = GL_FALSE; + + if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) { + ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY_ARB, mesa_ind_buf->obj); + assert(mesa_ind_buf->obj->Pointer != NULL); + mapped_named_bo = GL_TRUE; + } + + src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr); + + const GLuint size = mesa_ind_buf->count * getTypeSize(mesa_ind_buf->type); + + radeonAllocDmaRegion(&r300->radeon, &r300->ind_buf.bo, &r300->ind_buf.bo_offset, size, 4); + + assert(r300->ind_buf.bo->ptr != NULL); + dst_ptr = ADD_POINTERS(r300->ind_buf.bo->ptr, r300->ind_buf.bo_offset); + _mesa_memcpy(dst_ptr, src_ptr, size); + + r300->ind_buf.is_32bit = (mesa_ind_buf->type == GL_UNSIGNED_INT); + r300->ind_buf.count = mesa_ind_buf->count; + + if (mapped_named_bo) { + ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, mesa_ind_buf->obj); + } + } else { + r300FixupIndexBuffer(ctx, mesa_ind_buf); } } @@ -161,26 +205,118 @@ static int getTypeSize(GLenum type) } \ } while (0) -static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const struct gl_client_array *input, struct gl_buffer_object **bo, GLuint *nr_bo) +/** + * Convert attribute data type to float + * If the attribute uses named buffer object replace the bo with newly allocated bo + */ +static void r300ConvertAttrib(GLcontext *ctx, int count, const struct gl_client_array *input, struct vertex_attribute *attr) { r300ContextPtr r300 = R300_CONTEXT(ctx); - struct r300_vertex_buffer *vbuf = &r300->vbuf; - struct vertex_attribute r300_attr; - const void *src_ptr; - GLenum type; + const GLvoid *src_ptr; + GLboolean mapped_named_bo = GL_FALSE; + GLfloat *dst_ptr; GLuint stride; + stride = (input->StrideB == 0) ? getTypeSize(input->Type) * input->Size : input->StrideB; + + /* Convert value for first element only */ + if (input->StrideB == 0) + count = 1; + if (input->BufferObj->Name) { if (!input->BufferObj->Pointer) { - bo[*nr_bo] = input->BufferObj; - (*nr_bo)++; ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY_ARB, input->BufferObj); - assert(input->BufferObj->Pointer != NULL); + mapped_named_bo = GL_TRUE; } src_ptr = ADD_POINTERS(input->BufferObj->Pointer, input->Ptr); - } else + } else { src_ptr = input->Ptr; + } + + radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, sizeof(GLfloat) * input->Size * count, 32); + dst_ptr = (GLfloat *)ADD_POINTERS(attr->bo->ptr, attr->bo_offset); + + if (RADEON_DEBUG & DEBUG_FALLBACKS) { + fprintf(stderr, "%s: Converting vertex attributes, attribute data format %x,", __FUNCTION__, input->Type); + fprintf(stderr, "stride %d, components %d\n", stride, input->Size); + } + + assert(src_ptr != NULL); + + switch (input->Type) { + case GL_DOUBLE: + CONVERT(GLdouble, (GLfloat)); + break; + case GL_UNSIGNED_INT: + CONVERT(GLuint, UINT_TO_FLOAT); + break; + case GL_INT: + CONVERT(GLint, INT_TO_FLOAT); + break; + case GL_UNSIGNED_SHORT: + CONVERT(GLushort, USHORT_TO_FLOAT); + break; + case GL_SHORT: + CONVERT(GLshort, SHORT_TO_FLOAT); + break; + case GL_UNSIGNED_BYTE: + assert(input->Format != GL_BGRA); + CONVERT(GLubyte, UBYTE_TO_FLOAT); + break; + case GL_BYTE: + CONVERT(GLbyte, BYTE_TO_FLOAT); + break; + default: + assert(0); + break; + } + + if (mapped_named_bo) { + ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj); + } +} + +static void r300AlignDataToDword(GLcontext *ctx, const struct gl_client_array *input, int count, struct vertex_attribute *attr) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + const int dst_stride = (input->StrideB + 3) & ~3; + const int size = getTypeSize(input->Type) * input->Size * count; + GLboolean mapped_named_bo = GL_FALSE; + + radeonAllocDmaRegion(&r300->radeon, &attr->bo, &attr->bo_offset, size, 32); + + if (!input->BufferObj->Pointer) { + ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY_ARB, input->BufferObj); + mapped_named_bo = GL_TRUE; + } + + { + GLvoid *src_ptr = ADD_POINTERS(input->BufferObj->Pointer, input->Ptr); + GLvoid *dst_ptr = ADD_POINTERS(attr->bo->ptr, attr->bo_offset); + int i; + + for (i = 0; i < count; ++i) { + _mesa_memcpy(dst_ptr, src_ptr, input->StrideB); + src_ptr += input->StrideB; + dst_ptr += dst_stride; + } + } + + if (mapped_named_bo) { + ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, input->BufferObj); + } + + attr->stride = dst_stride; +} + +static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const struct gl_client_array *input) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_vertex_buffer *vbuf = &r300->vbuf; + struct vertex_attribute r300_attr; + GLenum type; + GLuint stride; stride = (input->StrideB == 0) ? getTypeSize(input->Type) * input->Size : input->StrideB; @@ -189,62 +325,57 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st getTypeSize(input->Type) != 4 || #endif stride < 4) { - if (RADEON_DEBUG & DEBUG_FALLBACKS) { - fprintf(stderr, "%s: Converting vertex attributes, attribute data format %x,", __FUNCTION__, input->Type); - fprintf(stderr, "stride %d, components %d\n", stride, input->Size); - } - - GLfloat *dst_ptr, *tmp; - - /* Convert value for first element only */ - if (input->StrideB == 0) - count = 1; - - tmp = dst_ptr = _mesa_malloc(sizeof(GLfloat) * input->Size * count); - - switch (input->Type) { - case GL_DOUBLE: - CONVERT(GLdouble, (GLfloat)); - break; - case GL_UNSIGNED_INT: - CONVERT(GLuint, UINT_TO_FLOAT); - break; - case GL_INT: - CONVERT(GLint, INT_TO_FLOAT); - break; - case GL_UNSIGNED_SHORT: - CONVERT(GLushort, USHORT_TO_FLOAT); - break; - case GL_SHORT: - CONVERT(GLshort, SHORT_TO_FLOAT); - break; - case GL_UNSIGNED_BYTE: - assert(input->Format != GL_BGRA); - CONVERT(GLubyte, UBYTE_TO_FLOAT); - break; - case GL_BYTE: - CONVERT(GLbyte, BYTE_TO_FLOAT); - break; - default: - assert(0); - break; - } type = GL_FLOAT; - r300_attr.free_needed = GL_TRUE; - r300_attr.data = tmp; + + r300ConvertAttrib(ctx, count, input, &r300_attr); if (input->StrideB == 0) { r300_attr.stride = 0; } else { r300_attr.stride = sizeof(GLfloat) * input->Size; } r300_attr.dwords = input->Size; + r300_attr.is_named_bo = GL_FALSE; } else { type = input->Type; - r300_attr.free_needed = GL_FALSE; - r300_attr.data = (GLvoid *)src_ptr; - r300_attr.stride = input->StrideB; - r300_attr.dwords = (getTypeSize(type) * input->Size + 3)/ 4; + r300_attr.dwords = (getTypeSize(type) * input->Size + 3)/ 4; + if (input->BufferObj->Name) { + if (stride % 4 != 0) { + assert(((int) input->Ptr) % input->StrideB == 0); + r300AlignDataToDword(ctx, input, count, &r300_attr); + r300_attr.is_named_bo = GL_FALSE; + } else { + r300_attr.stride = input->StrideB; + r300_attr.bo_offset = (GLuint) input->Ptr; + r300_attr.bo = get_radeon_buffer_object(input->BufferObj)->bo; + r300_attr.is_named_bo = GL_TRUE; + } + } else { + int size; + uint32_t *dst; + + if (input->StrideB == 0) { + size = getTypeSize(input->Type) * input->Size; + count = 1; + r300_attr.stride = 0; + } else { + size = getTypeSize(input->Type) * input->Size * count; + r300_attr.stride = (getTypeSize(type) * input->Size + 3) & ~3; + } + + radeonAllocDmaRegion(&r300->radeon, &r300_attr.bo, &r300_attr.bo_offset, size, 32); + assert(r300_attr.bo->ptr != NULL); + dst = (uint32_t *)ADD_POINTERS(r300_attr.bo->ptr, r300_attr.bo_offset); + switch (r300_attr.dwords) { + case 1: radeonEmitVec4(dst, input->Ptr, input->StrideB, count); break; + case 2: radeonEmitVec8(dst, input->Ptr, input->StrideB, count); break; + case 3: radeonEmitVec12(dst, input->Ptr, input->StrideB, count); break; + case 4: radeonEmitVec16(dst, input->Ptr, input->StrideB, count); break; + default: assert(0); break; + } + + r300_attr.is_named_bo = GL_FALSE; + } } r300_attr.size = input->Size; @@ -333,15 +464,15 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st ++vbuf->num_attribs; } -static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *arrays[], int count, struct gl_buffer_object **bo, GLuint *nr_bo) +static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *arrays[], int count) { r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_vertex_buffer *vbuf = &r300->vbuf; - + int ret; { int i, tmp; - tmp = r300->selected_vp->Base->Base.InputsRead; + tmp = r300->selected_vp->code.InputsRead; i = 0; vbuf->num_attribs = 0; while (tmp) { @@ -351,7 +482,7 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar ++i; } - r300TranslateAttrib(ctx, i, count, arrays[i], bo, nr_bo); + r300TranslateAttrib(ctx, i, count, arrays[i]); tmp >>= 1; ++i; @@ -366,38 +497,47 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar int i; for (i = 0; i < vbuf->num_attribs; i++) { - rcommon_emit_vector(ctx, &r300->radeon.tcl.aos[i], - vbuf->attribs[i].data, vbuf->attribs[i].dwords, - vbuf->attribs[i].stride, count); + struct radeon_aos *aos = &r300->radeon.tcl.aos[i]; + + aos->count = vbuf->attribs[i].stride == 0 ? 1 : count; + aos->stride = vbuf->attribs[i].stride / sizeof(float); + aos->offset = vbuf->attribs[i].bo_offset; + aos->components = vbuf->attribs[i].dwords; + aos->bo = vbuf->attribs[i].bo; + + if (vbuf->attribs[i].is_named_bo) { + radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, r300->vbuf.attribs[i].bo, RADEON_GEM_DOMAIN_GTT, 0); + } } r300->radeon.tcl.aos_count = vbuf->num_attribs; + ret = radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, r300->radeon.dma.current, RADEON_GEM_DOMAIN_GTT, 0); + if (ret) + r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, GL_TRUE); } } -static void r300FreeData(GLcontext *ctx, struct gl_buffer_object **bo, GLuint nr_bo) +static void r300FreeData(GLcontext *ctx) { + /* Need to zero tcl.aos[n].bo and tcl.elt_dma_bo + * to prevent double unref in radeonReleaseArrays + * called during context destroy + */ + r300ContextPtr r300 = R300_CONTEXT(ctx); { - struct r300_vertex_buffer *vbuf = &R300_CONTEXT(ctx)->vbuf; int i; - for (i = 0; i < vbuf->num_attribs; i++) { - if (vbuf->attribs[i].free_needed) - _mesa_free(vbuf->attribs[i].data); + for (i = 0; i < r300->vbuf.num_attribs; i++) { + if (!r300->vbuf.attribs[i].is_named_bo) { + radeon_bo_unref(r300->vbuf.attribs[i].bo); + } + r300->radeon.tcl.aos[i].bo = NULL; } } { - struct r300_index_buffer *ind_buf = &R300_CONTEXT(ctx)->ind_buf; - if (ind_buf->free_needed) - _mesa_free(ind_buf->ptr); - } - - { - int i; - - for (i = 0; i < nr_bo; ++i) { - ctx->Driver.UnmapBuffer(ctx, 0, bo[i]); + if (r300->ind_buf.bo != NULL) { + radeon_bo_unref(r300->ind_buf.bo); } } } @@ -411,8 +551,7 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx, GLuint max_index ) { struct r300_context *r300 = R300_CONTEXT(ctx); - struct gl_buffer_object *bo[VERT_ATTRIB_MAX+1]; - GLuint i, nr_bo = 0; + GLuint i; if (ctx->NewState) _mesa_update_state( ctx ); @@ -424,35 +563,37 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx, r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, !r300ValidateBuffers(ctx)); - r300FixupIndexBuffer(ctx, ib, bo, &nr_bo); - /* ensure we have the cmd buf space in advance to cover * the state + DMA AOS pointers */ rcommonEnsureCmdBufSpace(&r300->radeon, - r300->radeon.hw.max_state_size + (50*sizeof(int)), - __FUNCTION__); + r300->radeon.hw.max_state_size + (60*sizeof(int)), + __FUNCTION__); + + r300SetupIndexBuffer(ctx, ib); - r300SetVertexFormat(ctx, arrays, max_index + 1, bo, &nr_bo); + r300SetVertexFormat(ctx, arrays, max_index + 1); if (r300->fallback) return GL_FALSE; - r300SetupVAP(ctx, r300->selected_vp->Base->Base.InputsRead, r300->selected_vp->Base->Base.OutputsWritten); + r300SetupVAP(ctx, r300->selected_vp->code.InputsRead, r300->selected_vp->code.OutputsWritten); r300UpdateShaderStates(r300); r300EmitCacheFlush(r300); radeonEmitState(&r300->radeon); + r300EmitQueryBegin(ctx); + for (i = 0; i < nr_prims; ++i) { r300RunRenderPrimitive(ctx, prim[i].start, prim[i].start + prim[i].count, prim[i].mode); } r300EmitCacheFlush(r300); - radeonReleaseArrays(ctx, ~0); + r300EmitQueryEnd(ctx); - r300FreeData(ctx, bo, nr_bo); + r300FreeData(ctx); return GL_TRUE; } @@ -462,28 +603,23 @@ static void r300DrawPrims(GLcontext *ctx, const struct _mesa_prim *prim, GLuint nr_prims, const struct _mesa_index_buffer *ib, + GLboolean index_bounds_valid, GLuint min_index, GLuint max_index) { - struct split_limits limits; GLboolean retval; - if (ib) - limits.max_verts = 0xffffffff; - else - limits.max_verts = 65535; - - limits.max_indices = 65535; - limits.max_vb_size = 1024*1024; + /* This check should get folded into just the places that + * min/max index are really needed. + */ + if (!index_bounds_valid) { + vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index); + } if (min_index) { vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib, min_index, max_index, r300DrawPrims ); return; } - if ((ib && ib->count > 65535)) { - vbo_split_prims (ctx, arrays, prim, nr_prims, ib, min_index, max_index, r300DrawPrims, &limits); - return; - } /* Make an attempt at drawing */ retval = r300TryDrawPrims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); diff --git a/src/mesa/drivers/dri/r300/r300_emit.c b/src/mesa/drivers/dri/r300/r300_emit.c index feb3370f37..07e6223087 100644 --- a/src/mesa/drivers/dri/r300/r300_emit.c +++ b/src/mesa/drivers/dri/r300/r300_emit.c @@ -124,41 +124,6 @@ GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes) return ret; } -GLboolean r300EmitArrays(GLcontext * ctx) -{ - r300ContextPtr r300 = R300_CONTEXT(ctx); - struct r300_vertex_buffer *vbuf = &r300->vbuf; - GLuint InputsRead, OutputsWritten; - - r300ChooseSwtclVertexFormat(ctx, &InputsRead, &OutputsWritten); - - r300SwitchFallback(ctx, R300_FALLBACK_AOS_LIMIT, vbuf->num_attribs > R300_MAX_AOS_ARRAYS); - if (r300->fallback & R300_RASTER_FALLBACK_MASK) - return GL_FALSE; - - { - struct vertex_buffer *mesa_vb = &TNL_CONTEXT(ctx)->vb; - GLuint attr, i; - - for (i = 0; i < vbuf->num_attribs; i++) { - attr = vbuf->attribs[i].element; - rcommon_emit_vector(ctx, &r300->radeon.tcl.aos[i], mesa_vb->AttribPtr[attr]->data, - mesa_vb->AttribPtr[attr]->size, mesa_vb->AttribPtr[attr]->stride, mesa_vb->Count); - } - - r300->radeon.tcl.aos_count = vbuf->num_attribs; - - /* Fill index buffer info */ - r300->ind_buf.ptr = mesa_vb->Elts; - r300->ind_buf.is_32bit = GL_TRUE; - r300->ind_buf.free_needed = GL_FALSE; - } - - r300SetupVAP(ctx, InputsRead, OutputsWritten); - - return GL_TRUE; -} - void r300EmitCacheFlush(r300ContextPtr rmesa) { BATCH_LOCALS(&rmesa->radeon); diff --git a/src/mesa/drivers/dri/r300/r300_emit.h b/src/mesa/drivers/dri/r300/r300_emit.h index 3f8c60ffae..8e57e354d1 100644 --- a/src/mesa/drivers/dri/r300/r300_emit.h +++ b/src/mesa/drivers/dri/r300/r300_emit.h @@ -104,7 +104,7 @@ static INLINE uint32_t cmdpacket3(struct radeon_screen *rscrn, int packet) return cmd.u; } -static INLINE uint32_t cmdcpdelay(struct radeon_screen *rscrn, +static INLINE uint32_t cmdcpdelay(struct radeon_screen *rscrn, unsigned short count) { drm_r300_cmd_header_t cmd; @@ -216,8 +216,6 @@ void static INLINE cp_wait(radeonContextPtr radeon, unsigned char flags) } } -extern GLboolean r300EmitArrays(GLcontext * ctx); - extern int r300PrimitiveType(r300ContextPtr rmesa, int prim); extern int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim); diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c deleted file mode 100644 index 55c1cfe631..0000000000 --- a/src/mesa/drivers/dri/r300/r300_fragprog.c +++ /dev/null @@ -1,451 +0,0 @@ -/* - * Copyright (C) 2005 Ben Skeggs. - * - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining - * a copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sublicense, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial - * portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, - * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. - * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE - * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION - * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION - * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#include "r300_fragprog.h" - -#include "shader/prog_parameter.h" - -#include "r300_context.h" -#include "r300_fragprog_swizzle.h" - -static void reset_srcreg(struct prog_src_register* reg) -{ - _mesa_bzero(reg, sizeof(*reg)); - reg->Swizzle = SWIZZLE_NOOP; -} - -static struct prog_src_register shadow_ambient(struct gl_program *program, int tmu) -{ - gl_state_index fail_value_tokens[STATE_LENGTH] = { - STATE_INTERNAL, STATE_SHADOW_AMBIENT, 0, 0, 0 - }; - struct prog_src_register reg = { 0, }; - - fail_value_tokens[2] = tmu; - reg.File = PROGRAM_STATE_VAR; - reg.Index = _mesa_add_state_reference(program->Parameters, fail_value_tokens); - reg.Swizzle = SWIZZLE_WWWW; - return reg; -} - -/** - * Transform TEX, TXP, TXB, and KIL instructions in the following way: - * - premultiply texture coordinates for RECT - * - extract operand swizzles - * - introduce a temporary register when write masks are needed - * - * \todo If/when r5xx uses the radeon_program architecture, this can probably - * be reused. - */ -GLboolean r300_transform_TEX( - struct radeon_transform_context *t, - struct prog_instruction* orig_inst, void* data) -{ - struct r300_fragment_program_compiler *compiler = - (struct r300_fragment_program_compiler*)data; - struct prog_instruction inst = *orig_inst; - struct prog_instruction* tgt; - GLboolean destredirect = GL_FALSE; - - if (inst.Opcode != OPCODE_TEX && - inst.Opcode != OPCODE_TXB && - inst.Opcode != OPCODE_TXP && - inst.Opcode != OPCODE_KIL) - return GL_FALSE; - - if (inst.Opcode != OPCODE_KIL && - t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) { - GLuint comparefunc = GL_NEVER + compiler->fp->state.unit[inst.TexSrcUnit].texture_compare_func; - - if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) { - tgt = radeonAppendInstructions(t->Program, 1); - - tgt->Opcode = OPCODE_MOV; - tgt->DstReg = inst.DstReg; - if (comparefunc == GL_ALWAYS) { - tgt->SrcReg[0].File = PROGRAM_BUILTIN; - tgt->SrcReg[0].Swizzle = SWIZZLE_1111; - } else { - tgt->SrcReg[0] = shadow_ambient(t->Program, inst.TexSrcUnit); - } - return GL_TRUE; - } - - inst.DstReg.File = PROGRAM_TEMPORARY; - inst.DstReg.Index = radeonFindFreeTemporary(t); - inst.DstReg.WriteMask = WRITEMASK_XYZW; - } - - - /* Hardware uses [0..1]x[0..1] range for rectangle textures - * instead of [0..Width]x[0..Height]. - * Add a scaling instruction. - */ - if (inst.Opcode != OPCODE_KIL && inst.TexSrcTarget == TEXTURE_RECT_INDEX) { - gl_state_index tokens[STATE_LENGTH] = { - STATE_INTERNAL, STATE_R300_TEXRECT_FACTOR, 0, 0, - 0 - }; - - int tempreg = radeonFindFreeTemporary(t); - int factor_index; - - tokens[2] = inst.TexSrcUnit; - factor_index = _mesa_add_state_reference(t->Program->Parameters, tokens); - - tgt = radeonAppendInstructions(t->Program, 1); - - tgt->Opcode = OPCODE_MUL; - tgt->DstReg.File = PROGRAM_TEMPORARY; - tgt->DstReg.Index = tempreg; - tgt->SrcReg[0] = inst.SrcReg[0]; - tgt->SrcReg[1].File = PROGRAM_STATE_VAR; - tgt->SrcReg[1].Index = factor_index; - - reset_srcreg(&inst.SrcReg[0]); - inst.SrcReg[0].File = PROGRAM_TEMPORARY; - inst.SrcReg[0].Index = tempreg; - } - - if (inst.Opcode != OPCODE_KIL) { - if (inst.DstReg.File != PROGRAM_TEMPORARY || - inst.DstReg.WriteMask != WRITEMASK_XYZW) { - int tempreg = radeonFindFreeTemporary(t); - - inst.DstReg.File = PROGRAM_TEMPORARY; - inst.DstReg.Index = tempreg; - inst.DstReg.WriteMask = WRITEMASK_XYZW; - destredirect = GL_TRUE; - } else if (inst.SaturateMode) { - destredirect = GL_TRUE; - } - } - - if (inst.SrcReg[0].File != PROGRAM_TEMPORARY && inst.SrcReg[0].File != PROGRAM_INPUT) { - int tmpreg = radeonFindFreeTemporary(t); - tgt = radeonAppendInstructions(t->Program, 1); - tgt->Opcode = OPCODE_MOV; - tgt->DstReg.File = PROGRAM_TEMPORARY; - tgt->DstReg.Index = tmpreg; - tgt->SrcReg[0] = inst.SrcReg[0]; - - reset_srcreg(&inst.SrcReg[0]); - inst.SrcReg[0].File = PROGRAM_TEMPORARY; - inst.SrcReg[0].Index = tmpreg; - } - - tgt = radeonAppendInstructions(t->Program, 1); - _mesa_copy_instructions(tgt, &inst, 1); - - if (inst.Opcode != OPCODE_KIL && - t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) { - GLuint comparefunc = GL_NEVER + compiler->fp->state.unit[inst.TexSrcUnit].texture_compare_func; - GLuint depthmode = compiler->fp->state.unit[inst.TexSrcUnit].depth_texture_mode; - int rcptemp = radeonFindFreeTemporary(t); - int pass, fail; - - tgt = radeonAppendInstructions(t->Program, 3); - - tgt[0].Opcode = OPCODE_RCP; - tgt[0].DstReg.File = PROGRAM_TEMPORARY; - tgt[0].DstReg.Index = rcptemp; - tgt[0].DstReg.WriteMask = WRITEMASK_W; - tgt[0].SrcReg[0] = inst.SrcReg[0]; - tgt[0].SrcReg[0].Swizzle = SWIZZLE_WWWW; - - tgt[1].Opcode = OPCODE_MAD; - tgt[1].DstReg = inst.DstReg; - tgt[1].DstReg.WriteMask = orig_inst->DstReg.WriteMask; - tgt[1].SrcReg[0] = inst.SrcReg[0]; - tgt[1].SrcReg[0].Swizzle = SWIZZLE_ZZZZ; - tgt[1].SrcReg[1].File = PROGRAM_TEMPORARY; - tgt[1].SrcReg[1].Index = rcptemp; - tgt[1].SrcReg[1].Swizzle = SWIZZLE_WWWW; - tgt[1].SrcReg[2].File = PROGRAM_TEMPORARY; - tgt[1].SrcReg[2].Index = inst.DstReg.Index; - if (depthmode == 0) /* GL_LUMINANCE */ - tgt[1].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z); - else if (depthmode == 2) /* GL_ALPHA */ - tgt[1].SrcReg[2].Swizzle = SWIZZLE_WWWW; - - /* Recall that SrcReg[0] is tex, SrcReg[2] is r and: - * r < tex <=> -tex+r < 0 - * r >= tex <=> not (-tex+r < 0 */ - if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL) - tgt[1].SrcReg[2].Negate = tgt[0].SrcReg[2].Negate ^ NEGATE_XYZW; - else - tgt[1].SrcReg[0].Negate = tgt[0].SrcReg[0].Negate ^ NEGATE_XYZW; - - tgt[2].Opcode = OPCODE_CMP; - tgt[2].DstReg = orig_inst->DstReg; - tgt[2].SrcReg[0].File = PROGRAM_TEMPORARY; - tgt[2].SrcReg[0].Index = tgt[1].DstReg.Index; - - if (comparefunc == GL_LESS || comparefunc == GL_GREATER) { - pass = 1; - fail = 2; - } else { - pass = 2; - fail = 1; - } - - tgt[2].SrcReg[pass].File = PROGRAM_BUILTIN; - tgt[2].SrcReg[pass].Swizzle = SWIZZLE_1111; - tgt[2].SrcReg[fail] = shadow_ambient(t->Program, inst.TexSrcUnit); - } else if (destredirect) { - tgt = radeonAppendInstructions(t->Program, 1); - - tgt->Opcode = OPCODE_MOV; - tgt->DstReg = orig_inst->DstReg; - tgt->SaturateMode = inst.SaturateMode; - tgt->SrcReg[0].File = PROGRAM_TEMPORARY; - tgt->SrcReg[0].Index = inst.DstReg.Index; - } - - return GL_TRUE; -} - -/* just some random things... */ -void r300FragmentProgramDump(union rX00_fragment_program_code *c) -{ - struct r300_fragment_program_code *code = &c->r300; - int n, i, j; - static int pc = 0; - - fprintf(stderr, "pc=%d*************************************\n", pc++); - - fprintf(stderr, "Hardware program\n"); - fprintf(stderr, "----------------\n"); - - for (n = 0; n < (code->cur_node + 1); n++) { - fprintf(stderr, "NODE %d: alu_offset: %d, tex_offset: %d, " - "alu_end: %d, tex_end: %d, flags: %08x\n", n, - code->node[n].alu_offset, - code->node[n].tex_offset, - code->node[n].alu_end, code->node[n].tex_end, - code->node[n].flags); - - if (n > 0 || code->first_node_has_tex) { - fprintf(stderr, " TEX:\n"); - for (i = code->node[n].tex_offset; - i <= code->node[n].tex_offset + code->node[n].tex_end; - ++i) { - const char *instr; - - switch ((code->tex. - inst[i] >> R300_TEX_INST_SHIFT) & - 15) { - case R300_TEX_OP_LD: - instr = "TEX"; - break; - case R300_TEX_OP_KIL: - instr = "KIL"; - break; - case R300_TEX_OP_TXP: - instr = "TXP"; - break; - case R300_TEX_OP_TXB: - instr = "TXB"; - break; - default: - instr = "UNKNOWN"; - } - - fprintf(stderr, - " %s t%i, %c%i, texture[%i] (%08x)\n", - instr, - (code->tex. - inst[i] >> R300_DST_ADDR_SHIFT) & 31, - 't', - (code->tex. - inst[i] >> R300_SRC_ADDR_SHIFT) & 31, - (code->tex. - inst[i] & R300_TEX_ID_MASK) >> - R300_TEX_ID_SHIFT, - code->tex.inst[i]); - } - } - - for (i = code->node[n].alu_offset; - i <= code->node[n].alu_offset + code->node[n].alu_end; ++i) { - char srcc[3][10], dstc[20]; - char srca[3][10], dsta[20]; - char argc[3][20]; - char arga[3][20]; - char flags[5], tmp[10]; - - for (j = 0; j < 3; ++j) { - int regc = code->alu.inst[i].inst1 >> (j * 6); - int rega = code->alu.inst[i].inst3 >> (j * 6); - - sprintf(srcc[j], "%c%i", - (regc & 32) ? 'c' : 't', regc & 31); - sprintf(srca[j], "%c%i", - (rega & 32) ? 'c' : 't', rega & 31); - } - - dstc[0] = 0; - sprintf(flags, "%s%s%s", - (code->alu.inst[i]. - inst1 & R300_ALU_DSTC_REG_X) ? "x" : "", - (code->alu.inst[i]. - inst1 & R300_ALU_DSTC_REG_Y) ? "y" : "", - (code->alu.inst[i]. - inst1 & R300_ALU_DSTC_REG_Z) ? "z" : ""); - if (flags[0] != 0) { - sprintf(dstc, "t%i.%s ", - (code->alu.inst[i]. - inst1 >> R300_ALU_DSTC_SHIFT) & 31, - flags); - } - sprintf(flags, "%s%s%s", - (code->alu.inst[i]. - inst1 & R300_ALU_DSTC_OUTPUT_X) ? "x" : "", - (code->alu.inst[i]. - inst1 & R300_ALU_DSTC_OUTPUT_Y) ? "y" : "", - (code->alu.inst[i]. - inst1 & R300_ALU_DSTC_OUTPUT_Z) ? "z" : ""); - if (flags[0] != 0) { - sprintf(tmp, "o%i.%s", - (code->alu.inst[i]. - inst1 >> R300_ALU_DSTC_SHIFT) & 31, - flags); - strcat(dstc, tmp); - } - - dsta[0] = 0; - if (code->alu.inst[i].inst3 & R300_ALU_DSTA_REG) { - sprintf(dsta, "t%i.w ", - (code->alu.inst[i]. - inst3 >> R300_ALU_DSTA_SHIFT) & 31); - } - if (code->alu.inst[i].inst3 & R300_ALU_DSTA_OUTPUT) { - sprintf(tmp, "o%i.w ", - (code->alu.inst[i]. - inst3 >> R300_ALU_DSTA_SHIFT) & 31); - strcat(dsta, tmp); - } - if (code->alu.inst[i].inst3 & R300_ALU_DSTA_DEPTH) { - strcat(dsta, "Z"); - } - - fprintf(stderr, - "%3i: xyz: %3s %3s %3s -> %-20s (%08x)\n" - " w: %3s %3s %3s -> %-20s (%08x)\n", i, - srcc[0], srcc[1], srcc[2], dstc, - code->alu.inst[i].inst1, srca[0], srca[1], - srca[2], dsta, code->alu.inst[i].inst3); - - for (j = 0; j < 3; ++j) { - int regc = code->alu.inst[i].inst0 >> (j * 7); - int rega = code->alu.inst[i].inst2 >> (j * 7); - int d; - char buf[20]; - - d = regc & 31; - if (d < 12) { - switch (d % 4) { - case R300_ALU_ARGC_SRC0C_XYZ: - sprintf(buf, "%s.xyz", - srcc[d / 4]); - break; - case R300_ALU_ARGC_SRC0C_XXX: - sprintf(buf, "%s.xxx", - srcc[d / 4]); - break; - case R300_ALU_ARGC_SRC0C_YYY: - sprintf(buf, "%s.yyy", - srcc[d / 4]); - break; - case R300_ALU_ARGC_SRC0C_ZZZ: - sprintf(buf, "%s.zzz", - srcc[d / 4]); - break; - } - } else if (d < 15) { - sprintf(buf, "%s.www", srca[d - 12]); - } else if (d == 20) { - sprintf(buf, "0.0"); - } else if (d == 21) { - sprintf(buf, "1.0"); - } else if (d == 22) { - sprintf(buf, "0.5"); - } else if (d >= 23 && d < 32) { - d -= 23; - switch (d / 3) { - case 0: - sprintf(buf, "%s.yzx", - srcc[d % 3]); - break; - case 1: - sprintf(buf, "%s.zxy", - srcc[d % 3]); - break; - case 2: - sprintf(buf, "%s.Wzy", - srcc[d % 3]); - break; - } - } else { - sprintf(buf, "%i", d); - } - - sprintf(argc[j], "%s%s%s%s", - (regc & 32) ? "-" : "", - (regc & 64) ? "|" : "", - buf, (regc & 64) ? "|" : ""); - - d = rega & 31; - if (d < 9) { - sprintf(buf, "%s.%c", srcc[d / 3], - 'x' + (char)(d % 3)); - } else if (d < 12) { - sprintf(buf, "%s.w", srca[d - 9]); - } else if (d == 16) { - sprintf(buf, "0.0"); - } else if (d == 17) { - sprintf(buf, "1.0"); - } else if (d == 18) { - sprintf(buf, "0.5"); - } else { - sprintf(buf, "%i", d); - } - - sprintf(arga[j], "%s%s%s%s", - (rega & 32) ? "-" : "", - (rega & 64) ? "|" : "", - buf, (rega & 64) ? "|" : ""); - } - - fprintf(stderr, " xyz: %8s %8s %8s op: %08x\n" - " w: %8s %8s %8s op: %08x\n", - argc[0], argc[1], argc[2], - code->alu.inst[i].inst0, arga[0], arga[1], - arga[2], code->alu.inst[i].inst2); - } - } -} diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.h b/src/mesa/drivers/dri/r300/r300_fragprog.h deleted file mode 100644 index 5ce6f33cee..0000000000 --- a/src/mesa/drivers/dri/r300/r300_fragprog.h +++ /dev/null @@ -1,111 +0,0 @@ -/* - * Copyright (C) 2005 Ben Skeggs. - * - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining - * a copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sublicense, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial - * portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, - * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. - * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE - * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION - * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION - * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - */ - -/* - * Authors: - * Ben Skeggs <darktama@iinet.net.au> - * Jerome Glisse <j.glisse@gmail.com> - */ -#ifndef __R300_FRAGPROG_H_ -#define __R300_FRAGPROG_H_ - -#include "shader/program.h" -#include "shader/prog_instruction.h" - -#include "r300_context.h" -#include "radeon_program.h" - -#define DRI_CONF_FP_OPTIMIZATION_SPEED 0 -#define DRI_CONF_FP_OPTIMIZATION_QUALITY 1 - -#if 1 - -/** - * Fragment program helper macros - */ - -/* Produce unshifted source selectors */ -#define FP_TMP(idx) (idx) -#define FP_CONST(idx) ((idx) | (1 << 5)) - -/* Produce source/dest selector dword */ -#define FP_SELC_MASK_NO 0 -#define FP_SELC_MASK_X 1 -#define FP_SELC_MASK_Y 2 -#define FP_SELC_MASK_XY 3 -#define FP_SELC_MASK_Z 4 -#define FP_SELC_MASK_XZ 5 -#define FP_SELC_MASK_YZ 6 -#define FP_SELC_MASK_XYZ 7 - -#define FP_SELC(destidx,regmask,outmask,src0,src1,src2) \ - (((destidx) << R300_ALU_DSTC_SHIFT) | \ - (FP_SELC_MASK_##regmask << 23) | \ - (FP_SELC_MASK_##outmask << 26) | \ - ((src0) << R300_ALU_SRC0C_SHIFT) | \ - ((src1) << R300_ALU_SRC1C_SHIFT) | \ - ((src2) << R300_ALU_SRC2C_SHIFT)) - -#define FP_SELA_MASK_NO 0 -#define FP_SELA_MASK_W 1 - -#define FP_SELA(destidx,regmask,outmask,src0,src1,src2) \ - (((destidx) << R300_ALU_DSTA_SHIFT) | \ - (FP_SELA_MASK_##regmask << 23) | \ - (FP_SELA_MASK_##outmask << 24) | \ - ((src0) << R300_ALU_SRC0A_SHIFT) | \ - ((src1) << R300_ALU_SRC1A_SHIFT) | \ - ((src2) << R300_ALU_SRC2A_SHIFT)) - -/* Produce unshifted argument selectors */ -#define FP_ARGC(source) R300_ALU_ARGC_##source -#define FP_ARGA(source) R300_ALU_ARGA_##source -#define FP_ABS(arg) ((arg) | (1 << 6)) -#define FP_NEG(arg) ((arg) ^ (1 << 5)) - -/* Produce instruction dword */ -#define FP_INSTRC(opcode,arg0,arg1,arg2) \ - (R300_ALU_OUTC_##opcode | \ - ((arg0) << R300_ALU_ARG0C_SHIFT) | \ - ((arg1) << R300_ALU_ARG1C_SHIFT) | \ - ((arg2) << R300_ALU_ARG2C_SHIFT)) - -#define FP_INSTRA(opcode,arg0,arg1,arg2) \ - (R300_ALU_OUTA_##opcode | \ - ((arg0) << R300_ALU_ARG0A_SHIFT) | \ - ((arg1) << R300_ALU_ARG1A_SHIFT) | \ - ((arg2) << R300_ALU_ARG2A_SHIFT)) - -#endif - -extern GLboolean r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler); - -extern void r300FragmentProgramDump(union rX00_fragment_program_code *c); - -extern GLboolean r300_transform_TEX(struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data); - -#endif diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.c b/src/mesa/drivers/dri/r300/r300_fragprog_common.c index f5c4c0f4a0..6674efc5bc 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog_common.c +++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.c @@ -42,21 +42,51 @@ #include "shader/prog_parameter.h" #include "shader/prog_print.h" +#include "compiler/radeon_compiler.h" + #include "r300_state.h" -#include "r300_fragprog.h" -#include "r300_fragprog_swizzle.h" -#include "r500_fragprog.h" -#include "radeon_program.h" -#include "radeon_program_alu.h" -static void nqssadce_init(struct nqssadce_state* s) +static GLuint build_dtm(GLuint depthmode) +{ + switch(depthmode) { + default: + case GL_LUMINANCE: return 0; + case GL_INTENSITY: return 1; + case GL_ALPHA: return 2; + } +} + +static GLuint build_func(GLuint comparefunc) { - s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW; - s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W; + return comparefunc - GL_NEVER; } /** + * Collect all external state that is relevant for compiling the given + * fragment program. + */ +static void build_state( + r300ContextPtr r300, + struct gl_fragment_program *fp, + struct r300_fragment_program_external_state *state) +{ + int unit; + + _mesa_bzero(state, sizeof(*state)); + + for(unit = 0; unit < 16; ++unit) { + if (fp->Base.ShadowSamplers & (1 << unit)) { + struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current; + + state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode); + state->unit[unit].texture_compare_func = build_func(tex->CompareFunc); + } + } +} + + +/** * Transform the program to support fragment.position. * * Introduce a small fragment at the start of the program that will be @@ -65,104 +95,26 @@ static void nqssadce_init(struct nqssadce_state* s) * to read from a newly allocated temporary. * */ -static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler) +static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler, struct r300_fragment_program * fp) { - GLuint InputsRead = compiler->fp->Base->InputsRead; + int i; - if (!(InputsRead & FRAG_BIT_WPOS)) { - compiler->fp->wpos_attr = FRAG_ATTRIB_MAX; + if (!(compiler->Base.Program.InputsRead & FRAG_BIT_WPOS)) { + fp->wpos_attr = FRAG_ATTRIB_MAX; return; } - static gl_state_index tokens[STATE_LENGTH] = { - STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0 - }; - struct prog_instruction *fpi; - GLuint window_index; - int i = 0; - for (i = FRAG_ATTRIB_TEX0; i <= FRAG_ATTRIB_TEX7; ++i) { - if (!(InputsRead & (1 << i))) { - InputsRead &= ~(1 << FRAG_ATTRIB_WPOS); - InputsRead |= 1 << i; - compiler->fp->Base->InputsRead = InputsRead; - compiler->fp->wpos_attr = i; + if (!(compiler->Base.Program.InputsRead & (1 << i))) { + fp->wpos_attr = i; break; } } - GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY); - - _mesa_insert_instructions(compiler->program, 0, 3); - fpi = compiler->program->Instructions; - i = 0; - - /* perspective divide */ - fpi[i].Opcode = OPCODE_RCP; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_W; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_INPUT; - fpi[i].SrcReg[0].Index = compiler->fp->wpos_attr; - fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW; - i++; - - fpi[i].Opcode = OPCODE_MUL; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_INPUT; - fpi[i].SrcReg[0].Index = compiler->fp->wpos_attr; - fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW; - - fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[1].Index = tempregi; - fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW; - i++; - - /* viewport transformation */ - window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens); - - fpi[i].Opcode = OPCODE_MAD; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[0].Index = tempregi; - fpi[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR; - fpi[i].SrcReg[1].Index = window_index; - fpi[i].SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR; - fpi[i].SrcReg[2].Index = window_index; - fpi[i].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - i++; - - for (; i < compiler->program->NumInstructions; ++i) { - int reg; - for (reg = 0; reg < 3; reg++) { - if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT && - fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) { - fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[reg].Index = tempregi; - } - } - } + rc_transform_fragment_wpos(&compiler->Base, FRAG_ATTRIB_WPOS, fp->wpos_attr); } - /** * Rewrite fragment.fogcoord to use a texture coordinate slot. * Note that fogcoord is forced into an X001 pattern, and this enforcement @@ -170,205 +122,117 @@ static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler) * * See also the counterpart rewriting for vertex programs. */ -static void rewriteFog(struct r300_fragment_program_compiler *compiler) +static void rewriteFog(struct r300_fragment_program_compiler *compiler, struct r300_fragment_program * fp) { - struct r300_fragment_program *fp = compiler->fp; - GLuint InputsRead; + struct prog_src_register src; int i; - InputsRead = fp->Base->InputsRead; - - if (!(InputsRead & FRAG_BIT_FOGC)) { + if (!(compiler->Base.Program.InputsRead & FRAG_BIT_FOGC)) { fp->fog_attr = FRAG_ATTRIB_MAX; return; } for (i = FRAG_ATTRIB_TEX0; i <= FRAG_ATTRIB_TEX7; ++i) { - if (!(InputsRead & (1 << i))) { - InputsRead &= ~(1 << FRAG_ATTRIB_FOGC); - InputsRead |= 1 << i; - fp->Base->InputsRead = InputsRead; + if (!(compiler->Base.Program.InputsRead & (1 << i))) { fp->fog_attr = i; break; } } - { - struct prog_instruction *inst; - - inst = compiler->program->Instructions; - while (inst->Opcode != OPCODE_END) { - const int src_regs = _mesa_num_inst_src_regs(inst->Opcode); - for (i = 0; i < src_regs; ++i) { - if (inst->SrcReg[i].File == PROGRAM_INPUT && inst->SrcReg[i].Index == FRAG_ATTRIB_FOGC) { - inst->SrcReg[i].Index = fp->fog_attr; - inst->SrcReg[i].Swizzle = combine_swizzles( - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE), - inst->SrcReg[i].Swizzle); - } - } - ++inst; - } - } -} - -static GLuint build_dtm(GLuint depthmode) -{ - switch(depthmode) { - default: - case GL_LUMINANCE: return 0; - case GL_INTENSITY: return 1; - case GL_ALPHA: return 2; - } + memset(&src, 0, sizeof(src)); + src.File = PROGRAM_INPUT; + src.Index = fp->fog_attr; + src.Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE); + rc_move_input(&compiler->Base, FRAG_ATTRIB_FOGC, src); } -static GLuint build_func(GLuint comparefunc) -{ - return comparefunc - GL_NEVER; -} /** - * Collect all external state that is relevant for compiling the given - * fragment program. + * Reserve hardware temporary registers for the program inputs. + * + * @note This allocation is performed explicitly, because the order of inputs + * is determined by the RS hardware. */ -static void build_state( - r300ContextPtr r300, - struct gl_fragment_program *fp, - struct r300_fragment_program_external_state *state) +static void allocate_hw_inputs( + struct r300_fragment_program_compiler * c, + void (*allocate)(void * data, unsigned input, unsigned hwreg), + void * mydata) { - int unit; - - _mesa_bzero(state, sizeof(*state)); - - for(unit = 0; unit < 16; ++unit) { - if (fp->Base.ShadowSamplers & (1 << unit)) { - struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current; - - state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode); - state->unit[unit].texture_compare_func = build_func(tex->CompareFunc); - } + GLuint InputsRead = c->Base.Program.InputsRead; + int i; + GLuint hwindex = 0; + + /* Primary colour */ + if (InputsRead & FRAG_BIT_COL0) + allocate(mydata, FRAG_ATTRIB_COL0, hwindex++); + InputsRead &= ~FRAG_BIT_COL0; + + /* Secondary color */ + if (InputsRead & FRAG_BIT_COL1) + allocate(mydata, FRAG_ATTRIB_COL1, hwindex++); + InputsRead &= ~FRAG_BIT_COL1; + + /* Texcoords */ + for (i = 0; i < 8; i++) { + if (InputsRead & (FRAG_BIT_TEX0 << i)) + allocate(mydata, FRAG_ATTRIB_TEX0+i, hwindex++); } -} + InputsRead &= ~FRAG_BITS_TEX_ANY; -static void rewrite_depth_out(struct gl_program *prog) -{ - struct prog_instruction *inst; + /* Fogcoords treated as a texcoord */ + if (InputsRead & FRAG_BIT_FOGC) + allocate(mydata, FRAG_ATTRIB_FOGC, hwindex++); + InputsRead &= ~FRAG_BIT_FOGC; - for (inst = prog->Instructions; inst->Opcode != OPCODE_END; ++inst) { - if (inst->DstReg.File != PROGRAM_OUTPUT || inst->DstReg.Index != FRAG_RESULT_DEPTH) - continue; + /* fragment position treated as a texcoord */ + if (InputsRead & FRAG_BIT_WPOS) + allocate(mydata, FRAG_ATTRIB_WPOS, hwindex++); + InputsRead &= ~FRAG_BIT_WPOS; - if (inst->DstReg.WriteMask & WRITEMASK_Z) { - inst->DstReg.WriteMask = WRITEMASK_W; - } else { - inst->DstReg.WriteMask = 0; - continue; - } - - switch (inst->Opcode) { - case OPCODE_FRC: - case OPCODE_MOV: - inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); - break; - case OPCODE_ADD: - case OPCODE_MAX: - case OPCODE_MIN: - case OPCODE_MUL: - inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); - inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]); - break; - case OPCODE_CMP: - case OPCODE_MAD: - inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]); - inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]); - inst->SrcReg[2] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[2]); - break; - default: - // Scalar instructions needn't be reswizzled - break; - } - } + /* Anything else */ + if (InputsRead) + rc_error(&c->Base, "Don't know how to handle inputs 0x%x\n", InputsRead); } -void r300TranslateFragmentShader(GLcontext *ctx, struct r300_fragment_program *fp) + +static void translate_fragment_program(GLcontext *ctx, struct r300_fragment_program_cont *cont, struct r300_fragment_program *fp) { r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_fragment_program_compiler compiler; - compiler.r300 = r300; - compiler.fp = fp; + rc_init(&compiler.Base); + compiler.Base.Debug = (RADEON_DEBUG & DEBUG_PIXEL) ? GL_TRUE : GL_FALSE; + compiler.code = &fp->code; - compiler.program = fp->Base; + compiler.state = fp->state; + compiler.is_r500 = (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) ? GL_TRUE : GL_FALSE; + compiler.OutputDepth = FRAG_RESULT_DEPTH; + compiler.OutputColor = FRAG_RESULT_COLOR; + compiler.AllocateHwInputs = &allocate_hw_inputs; - if (RADEON_DEBUG & DEBUG_PIXEL) { + if (compiler.Base.Debug) { fflush(stdout); _mesa_printf("Fragment Program: Initial program:\n"); - _mesa_print_program(compiler.program); + _mesa_print_program(&cont->Base.Base); fflush(stdout); } - insert_WPOS_trailer(&compiler); - - rewriteFog(&compiler); - - rewrite_depth_out(compiler.program); - - if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - struct radeon_program_transformation transformations[] = { - { &r500_transform_TEX, &compiler }, - { &radeonTransformALU, 0 }, - { &radeonTransformDeriv, 0 }, - { &radeonTransformTrigScale, 0 } - }; - radeonLocalTransform(ctx, compiler.program, 4, transformations); - } else { - struct radeon_program_transformation transformations[] = { - { &r300_transform_TEX, &compiler }, - { &radeonTransformALU, 0 }, - { &radeonTransformTrigSimple, 0 } - }; - radeonLocalTransform(ctx, compiler.program, 3, transformations); - } + rc_mesa_to_rc_program(&compiler.Base, &cont->Base.Base); - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Fragment Program: After native rewrite:\n"); - _mesa_print_program(compiler.program); - fflush(stdout); - } - - if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - struct radeon_nqssadce_descr nqssadce = { - .Init = &nqssadce_init, - .IsNativeSwizzle = &r500FPIsNativeSwizzle, - .BuildSwizzle = &r500FPBuildSwizzle - }; - radeonNqssaDce(ctx, compiler.program, &nqssadce); - } else { - struct radeon_nqssadce_descr nqssadce = { - .Init = &nqssadce_init, - .IsNativeSwizzle = &r300FPIsNativeSwizzle, - .BuildSwizzle = &r300FPBuildSwizzle - }; - radeonNqssaDce(ctx, compiler.program, &nqssadce); - } + insert_WPOS_trailer(&compiler, fp); - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Compiler: after NqSSA-DCE:\n"); - _mesa_print_program(compiler.program); - fflush(stdout); - } + rewriteFog(&compiler, fp); - if (!r300->vtbl.BuildFragmentProgramHwCode(&compiler)) - fp->error = GL_TRUE; + r3xx_compile_fragment_program(&compiler); + fp->error = compiler.Base.Error; - fp->translated = GL_TRUE; + fp->InputsRead = compiler.Base.Program.InputsRead; - if (fp->error || (RADEON_DEBUG & DEBUG_PIXEL)) - r300->vtbl.FragmentProgramDump(&fp->code); + rc_destroy(&compiler.Base); } -struct r300_fragment_program *r300SelectFragmentShader(GLcontext *ctx) +struct r300_fragment_program *r300SelectAndTranslateFragmentShader(GLcontext *ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_fragment_program_cont *fp_list; @@ -389,11 +253,11 @@ struct r300_fragment_program *r300SelectFragmentShader(GLcontext *ctx) fp = _mesa_calloc(sizeof(struct r300_fragment_program)); fp->state = state; - fp->translated = GL_FALSE; - fp->Base = _mesa_clone_program(ctx, &ctx->FragmentProgram._Current->Base); fp->next = fp_list->progs; fp_list->progs = fp; + translate_fragment_program(ctx, fp_list, fp); + return r300->selected_fp = fp; } diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.h b/src/mesa/drivers/dri/r300/r300_fragprog_common.h index 5e103ee408..3d64c08cee 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog_common.h +++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.h @@ -32,8 +32,6 @@ #include "r300_context.h" -extern void r300TranslateFragmentShader(GLcontext *ctx, struct r300_fragment_program *fp); - -struct r300_fragment_program *r300SelectFragmentShader(GLcontext *ctx); +struct r300_fragment_program *r300SelectAndTranslateFragmentShader(GLcontext *ctx); #endif diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c index ddabd53992..da801f42e4 100644 --- a/src/mesa/drivers/dri/r300/r300_ioctl.c +++ b/src/mesa/drivers/dri/r300/r300_ioctl.c @@ -44,6 +44,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/imports.h" #include "main/macros.h" #include "main/context.h" +#include "main/simple_list.h" #include "swrast/swrast.h" #include "radeon_common.h" @@ -55,8 +56,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_vertprog.h" #include "radeon_reg.h" #include "r300_emit.h" -#include "r300_fragprog.h" #include "r300_context.h" +#include "r300_queryobj.h" #include "vblank.h" @@ -66,6 +67,66 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define CLEARBUFFER_DEPTH 0x2 #define CLEARBUFFER_STENCIL 0x4 +#if 1 + +/** + * Fragment program helper macros + */ + +/* Produce unshifted source selectors */ +#define FP_TMP(idx) (idx) +#define FP_CONST(idx) ((idx) | (1 << 5)) + +/* Produce source/dest selector dword */ +#define FP_SELC_MASK_NO 0 +#define FP_SELC_MASK_X 1 +#define FP_SELC_MASK_Y 2 +#define FP_SELC_MASK_XY 3 +#define FP_SELC_MASK_Z 4 +#define FP_SELC_MASK_XZ 5 +#define FP_SELC_MASK_YZ 6 +#define FP_SELC_MASK_XYZ 7 + +#define FP_SELC(destidx,regmask,outmask,src0,src1,src2) \ + (((destidx) << R300_ALU_DSTC_SHIFT) | \ + (FP_SELC_MASK_##regmask << 23) | \ + (FP_SELC_MASK_##outmask << 26) | \ + ((src0) << R300_ALU_SRC0C_SHIFT) | \ + ((src1) << R300_ALU_SRC1C_SHIFT) | \ + ((src2) << R300_ALU_SRC2C_SHIFT)) + +#define FP_SELA_MASK_NO 0 +#define FP_SELA_MASK_W 1 + +#define FP_SELA(destidx,regmask,outmask,src0,src1,src2) \ + (((destidx) << R300_ALU_DSTA_SHIFT) | \ + (FP_SELA_MASK_##regmask << 23) | \ + (FP_SELA_MASK_##outmask << 24) | \ + ((src0) << R300_ALU_SRC0A_SHIFT) | \ + ((src1) << R300_ALU_SRC1A_SHIFT) | \ + ((src2) << R300_ALU_SRC2A_SHIFT)) + +/* Produce unshifted argument selectors */ +#define FP_ARGC(source) R300_ALU_ARGC_##source +#define FP_ARGA(source) R300_ALU_ARGA_##source +#define FP_ABS(arg) ((arg) | (1 << 6)) +#define FP_NEG(arg) ((arg) ^ (1 << 5)) + +/* Produce instruction dword */ +#define FP_INSTRC(opcode,arg0,arg1,arg2) \ + (R300_ALU_OUTC_##opcode | \ + ((arg0) << R300_ALU_ARG0C_SHIFT) | \ + ((arg1) << R300_ALU_ARG1C_SHIFT) | \ + ((arg2) << R300_ALU_ARG2C_SHIFT)) + +#define FP_INSTRA(opcode,arg0,arg1,arg2) \ + (R300_ALU_OUTA_##opcode | \ + ((arg0) << R300_ALU_ARG0A_SHIFT) | \ + ((arg1) << R300_ALU_ARG1A_SHIFT) | \ + ((arg2) << R300_ALU_ARG2A_SHIFT)) + +#endif + static void r300EmitClearState(GLcontext * ctx); static void r300ClearBuffer(r300ContextPtr r300, int flags, @@ -109,18 +170,21 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, } #if 1 if (flags & (CLEARBUFFER_DEPTH | CLEARBUFFER_STENCIL)) { - assert(rrbd != 0); - cbpitch = (rrbd->pitch / rrbd->cpp); + uint32_t zbpitch = (rrbd->pitch / rrbd->cpp); if (rrbd->bo->flags & RADEON_BO_FLAGS_MACRO_TILE){ - cbpitch |= R300_DEPTHMACROTILE_ENABLE; + zbpitch |= R300_DEPTHMACROTILE_ENABLE; } if (rrbd->bo->flags & RADEON_BO_FLAGS_MICRO_TILE){ - cbpitch |= R300_DEPTHMICROTILE_TILED; + zbpitch |= R300_DEPTHMICROTILE_TILED; } BEGIN_BATCH_NO_AUTOSTATE(6); OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1); OUT_BATCH_RELOC(0, rrbd->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); - OUT_BATCH_REGVAL(R300_ZB_DEPTHPITCH, cbpitch); + OUT_BATCH_REGSEQ(R300_ZB_DEPTHPITCH, 1); + if (!r300->radeon.radeonScreen->kernel_mm) + OUT_BATCH(zbpitch); + else + OUT_BATCH_RELOC(zbpitch, rrbd->bo, zbpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0); END_BATCH(); } #endif @@ -508,12 +572,12 @@ static void r300EmitClearState(GLcontext * ctx) 0, 0xf, PVS_DST_REG_OUT); vpu.cmd[2] = PVS_SRC_OPERAND(0, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, - PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + PVS_SRC_REG_INPUT, NEGATE_NONE); vpu.cmd[3] = PVS_SRC_OPERAND(0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, - PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + PVS_SRC_REG_INPUT, NEGATE_NONE); vpu.cmd[4] = 0x0; vpu.cmd[5] = PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, 1, 0xf, @@ -521,13 +585,12 @@ static void r300EmitClearState(GLcontext * ctx) vpu.cmd[6] = PVS_SRC_OPERAND(1, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, - - VSF_FLAG_NONE); + NEGATE_NONE); vpu.cmd[7] = PVS_SRC_OPERAND(1, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, - PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + PVS_SRC_REG_INPUT, NEGATE_NONE); vpu.cmd[8] = 0x0; r300->vap_flush_needed = GL_TRUE; @@ -546,7 +609,7 @@ static int r300KernelClear(GLcontext *ctx, GLuint flags) /* Make sure it fits there. */ radeon_cs_space_reset_bos(r300->radeon.cmdbuf.cs); - + if (flags & BUFFER_BIT_COLOR0) { rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_COLOR0); radeon_cs_space_add_persistent_bo(r300->radeon.cmdbuf.cs, @@ -692,10 +755,19 @@ static void r300Clear(GLcontext * ctx, GLbitfield mask) } } +static void r300Flush(GLcontext *ctx) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + + radeonFlush(ctx); + + make_empty_list(&r300->query.not_flushed_head); +} + void r300InitIoctlFuncs(struct dd_function_table *functions) { functions->Clear = r300Clear; functions->Finish = radeonFinish; - functions->Flush = radeonFlush; + functions->Flush = r300Flush; } diff --git a/src/mesa/drivers/dri/r300/r300_queryobj.c b/src/mesa/drivers/dri/r300/r300_queryobj.c new file mode 100644 index 0000000000..df1fb32ee7 --- /dev/null +++ b/src/mesa/drivers/dri/r300/r300_queryobj.c @@ -0,0 +1,250 @@ +/* + * Copyright © 2008-2009 Maciej Cencora <m.cencora@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Maciej Cencora <m.cencora@gmail.com> + * + */ + +#include "r300_queryobj.h" +#include "r300_emit.h" + +#include "main/imports.h" +#include "main/simple_list.h" + +#define DDEBUG 0 + +#define PAGE_SIZE 4096 + +static void r300QueryGetResult(GLcontext *ctx, struct gl_query_object *q) +{ + struct r300_query_object *query = (struct r300_query_object *)q; + uint32_t *result; + int i; + + if (DDEBUG) fprintf(stderr, "%s: query id %d, result %d\n", __FUNCTION__, query->Base.Id, (int) query->Base.Result); + + radeon_bo_map(query->bo, GL_FALSE); + + result = query->bo->ptr; + + query->Base.Result = 0; + for (i = 0; i < query->curr_offset/sizeof(uint32_t); ++i) { + query->Base.Result += result[i]; + if (DDEBUG) fprintf(stderr, "result[%d] = %d\n", i, result[i]); + } + + radeon_bo_unmap(query->bo); +} + +static struct gl_query_object * r300NewQueryObject(GLcontext *ctx, GLuint id) +{ + struct r300_query_object *query; + + query = _mesa_calloc(sizeof(struct r300_query_object)); + + query->Base.Id = id; + query->Base.Result = 0; + query->Base.Active = GL_FALSE; + query->Base.Ready = GL_TRUE; + + if (DDEBUG) fprintf(stderr, "%s: query id %d\n", __FUNCTION__, query->Base.Id); + + return &query->Base; +} + +static void r300DeleteQuery(GLcontext *ctx, struct gl_query_object *q) +{ + struct r300_query_object *query = (struct r300_query_object *)q; + + if (DDEBUG) fprintf(stderr, "%s: query id %d\n", __FUNCTION__, q->Id); + + if (query->bo) { + radeon_bo_unref(query->bo); + } + + _mesa_free(query); +} + +static void r300BeginQuery(GLcontext *ctx, struct gl_query_object *q) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_query_object *query = (struct r300_query_object *)q; + + if (DDEBUG) fprintf(stderr, "%s: query id %d\n", __FUNCTION__, q->Id); + + assert(r300->query.current == NULL); + + if (!query->bo) { + query->bo = radeon_bo_open(r300->radeon.radeonScreen->bom, 0, PAGE_SIZE, PAGE_SIZE, RADEON_GEM_DOMAIN_GTT, 0); + } + query->curr_offset = 0; + + r300->query.current = query; + insert_at_tail(&r300->query.not_flushed_head, query); +} + +static void r300EndQuery(GLcontext *ctx, struct gl_query_object *q) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + + if (DDEBUG) fprintf(stderr, "%s: query id %d\n", __FUNCTION__, q->Id); + + r300EmitQueryEnd(ctx); + + r300->query.current = NULL; +} + +static void r300WaitQuery(GLcontext *ctx, struct gl_query_object *q) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_query_object *tmp, *query = (struct r300_query_object *)q; + + /* If the cmdbuf with packets for this query hasn't been flushed yet, do it now */ + { + GLboolean found = GL_FALSE; + foreach(tmp, &r300->query.not_flushed_head) { + if (tmp == query) { + found = GL_TRUE; + break; + } + } + + if (found) + ctx->Driver.Flush(ctx); + } + + if (DDEBUG) fprintf(stderr, "%s: query id %d, bo %p, offset %d\n", __FUNCTION__, q->Id, query->bo, query->curr_offset); + + r300QueryGetResult(ctx, q); + + query->Base.Ready = GL_TRUE; +} + + +/** + * TODO: + * should check if bo is idle, bo there's no interface to do it + * just wait for result now + */ +static void r300CheckQuery(GLcontext *ctx, struct gl_query_object *q) +{ + if (DDEBUG) fprintf(stderr, "%s: query id %d\n", __FUNCTION__, q->Id); + + r300WaitQuery(ctx, q); +} + +void r300EmitQueryBegin(GLcontext *ctx) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_query_object *query = r300->query.current; + BATCH_LOCALS(&r300->radeon); + + if (!query || query->emitted_begin) + return; + + if (DDEBUG) fprintf(stderr, "%s: query id %d\n", __FUNCTION__, query->Base.Id); + + if (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV530) { + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH_REGVAL(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); + OUT_BATCH_REGVAL(R300_ZB_ZPASS_DATA, 0); + END_BATCH(); + } else { + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH_REGVAL(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL); + OUT_BATCH_REGVAL(R300_ZB_ZPASS_DATA, 0); + END_BATCH(); + } + + query->emitted_begin = GL_TRUE; +} + +void r300EmitQueryEnd(GLcontext *ctx) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_query_object *query = r300->query.current; + BATCH_LOCALS(&r300->radeon); + + if (!query || !query->emitted_begin) + return; + + if (DDEBUG) fprintf(stderr, "%s: query id %d, bo %p, offset %d\n", __FUNCTION__, query->Base.Id, query->bo, query->curr_offset); + + radeon_cs_space_check_with_bo(r300->radeon.cmdbuf.cs, + query->bo, + 0, RADEON_GEM_DOMAIN_GTT); + + if (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV530) { + BEGIN_BATCH_NO_AUTOSTATE(14); + OUT_BATCH_REGVAL(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0); + OUT_BATCH_REGSEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_BATCH_RELOC(0, query->bo, query->curr_offset, 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_BATCH_REGVAL(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1); + OUT_BATCH_REGSEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_BATCH_RELOC(0, query->bo, query->curr_offset + sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_BATCH_REGVAL(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); + END_BATCH(); + } else { + BEGIN_BATCH_NO_AUTOSTATE(3 * 2 *r300->num_z_pipes + 2); + switch (r300->num_z_pipes) { + case 4: + OUT_BATCH_REGVAL(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_3); + OUT_BATCH_REGSEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_BATCH_RELOC(0, query->bo, query->curr_offset+3*sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT, 0); + case 3: + OUT_BATCH_REGVAL(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_2); + OUT_BATCH_REGSEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_BATCH_RELOC(0, query->bo, query->curr_offset+2*sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT, 0); + case 2: + if (r300->radeon.radeonScreen->chip_family <= CHIP_FAMILY_RV380) { + OUT_BATCH_REGVAL(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_3); + } else { + OUT_BATCH_REGVAL(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_1); + } + OUT_BATCH_REGSEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_BATCH_RELOC(0, query->bo, query->curr_offset+1*sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT, 0); + case 1: + default: + OUT_BATCH_REGVAL(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_0); + OUT_BATCH_REGSEQ(R300_ZB_ZPASS_ADDR, 1); + OUT_BATCH_RELOC(0, query->bo, query->curr_offset, 0, RADEON_GEM_DOMAIN_GTT, 0); + break; + } + OUT_BATCH_REGVAL(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL); + END_BATCH(); + } + + query->curr_offset += r300->num_z_pipes * sizeof(uint32_t); + assert(query->curr_offset < PAGE_SIZE); + query->emitted_begin = GL_FALSE; +} + +void r300InitQueryObjFunctions(struct dd_function_table *functions) +{ + functions->NewQueryObject = r300NewQueryObject; + functions->DeleteQuery = r300DeleteQuery; + functions->BeginQuery = r300BeginQuery; + functions->EndQuery = r300EndQuery; + functions->CheckQuery = r300CheckQuery; + functions->WaitQuery = r300WaitQuery; +} diff --git a/src/mesa/drivers/dri/r300/r300_queryobj.h b/src/mesa/drivers/dri/r300/r300_queryobj.h new file mode 100644 index 0000000000..f301f0b113 --- /dev/null +++ b/src/mesa/drivers/dri/r300/r300_queryobj.h @@ -0,0 +1,34 @@ +/* + * Copyright © 2008 Maciej Cencora <m.cencora@gmail.com> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Maciej Cencora <m.cencora@gmail.com> + * + */ + +#include "main/imports.h" +#include "r300_context.h" + +extern void r300EmitQueryBegin(GLcontext *ctx); +extern void r300EmitQueryEnd(GLcontext *ctx); + +extern void r300InitQueryObjFunctions(struct dd_function_table *functions); diff --git a/src/mesa/drivers/dri/r300/r300_reg.h b/src/mesa/drivers/dri/r300/r300_reg.h index 357c600af9..39b4b61a10 100644 --- a/src/mesa/drivers/dri/r300/r300_reg.h +++ b/src/mesa/drivers/dri/r300/r300_reg.h @@ -1128,6 +1128,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. /* SU Depth Offset value */ #define R300_SU_DEPTH_OFFSET 0x42c4 +#define R300_SU_REG_DEST 0x42c8 +# define R300_RASTER_PIPE_SELECT_0 (1 << 0) +# define R300_RASTER_PIPE_SELECT_1 (1 << 1) +# define R300_RASTER_PIPE_SELECT_2 (1 << 2) +# define R300_RASTER_PIPE_SELECT_3 (1 << 3) +# define R300_RASTER_PIPE_SELECT_ALL 0xf + /* BEGIN: Rasterization / Interpolators - many guesses */ @@ -2014,6 +2021,11 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #define R500_FG_ALPHA_VALUE 0x4be0 # define R500_FG_ALPHA_VALUE_MASK 0x0000ffff +#define RV530_FG_ZBREG_DEST 0x4be8 +# define RV530_FG_ZBREG_DEST_PIPE_SELECT_0 (1 << 0) +# define RV530_FG_ZBREG_DEST_PIPE_SELECT_1 (1 << 1) +# define RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL (3 << 0) + /* gap */ /* Fragment program parameters in 7.16 floating point */ @@ -2667,6 +2679,24 @@ enum { PVS_SRC_ADDR_MODE_1_SHIFT = 32, }; + +#define PVS_OP_DST_OPERAND(opcode, math_inst, macro_inst, reg_index, reg_writemask, reg_class) \ + (((opcode & PVS_DST_OPCODE_MASK) << PVS_DST_OPCODE_SHIFT) \ + | ((math_inst & PVS_DST_MATH_INST_MASK) << PVS_DST_MATH_INST_SHIFT) \ + | ((macro_inst & PVS_DST_MACRO_INST_MASK) << PVS_DST_MACRO_INST_SHIFT) \ + | ((reg_index & PVS_DST_OFFSET_MASK) << PVS_DST_OFFSET_SHIFT) \ + | ((reg_writemask & 0xf) << PVS_DST_WE_X_SHIFT) /* X Y Z W */ \ + | ((reg_class & PVS_DST_REG_TYPE_MASK) << PVS_DST_REG_TYPE_SHIFT)) + +#define PVS_SRC_OPERAND(in_reg_index, comp_x, comp_y, comp_z, comp_w, reg_class, negate) \ + (((in_reg_index & PVS_SRC_OFFSET_MASK) << PVS_SRC_OFFSET_SHIFT) \ + | ((comp_x & PVS_SRC_SWIZZLE_X_MASK) << PVS_SRC_SWIZZLE_X_SHIFT) \ + | ((comp_y & PVS_SRC_SWIZZLE_Y_MASK) << PVS_SRC_SWIZZLE_Y_SHIFT) \ + | ((comp_z & PVS_SRC_SWIZZLE_Z_MASK) << PVS_SRC_SWIZZLE_Z_SHIFT) \ + | ((comp_w & PVS_SRC_SWIZZLE_W_MASK) << PVS_SRC_SWIZZLE_W_SHIFT) \ + | ((negate & 0xf) << PVS_SRC_MODIFIER_X_SHIFT) /* X Y Z W */ \ + | ((reg_class & PVS_SRC_REG_TYPE_MASK) << PVS_SRC_REG_TYPE_SHIFT)) + /*\}*/ /* BEGIN: Packet 3 commands */ diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c index 08d67b73ed..45330cda3c 100644 --- a/src/mesa/drivers/dri/r300/r300_render.c +++ b/src/mesa/drivers/dri/r300/r300_render.c @@ -64,6 +64,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "swrast/swrast.h" #include "swrast_setup/swrast_setup.h" #include "vbo/vbo.h" +#include "vbo/vbo_split.h" #include "tnl/tnl.h" #include "tnl/t_vp_build.h" #include "radeon_reg.h" @@ -75,6 +76,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_tex.h" #include "r300_emit.h" #include "r300_fragprog_common.h" +#include "r300_queryobj.h" #include "r300_swtcl.h" /** @@ -172,64 +174,45 @@ int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim) return num_verts - verts_off; } -static void r300EmitElts(GLcontext * ctx, unsigned long n_elts) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - void *out; - GLuint size; - - size = ((rmesa->ind_buf.is_32bit ? 4 : 2) * n_elts + 3) & ~3; - - radeonAllocDmaRegion(&rmesa->radeon, &rmesa->radeon.tcl.elt_dma_bo, - &rmesa->radeon.tcl.elt_dma_offset, size, 4); - radeon_bo_map(rmesa->radeon.tcl.elt_dma_bo, 1); - out = rmesa->radeon.tcl.elt_dma_bo->ptr + rmesa->radeon.tcl.elt_dma_offset; - memcpy(out, rmesa->ind_buf.ptr, size); - radeon_bo_unmap(rmesa->radeon.tcl.elt_dma_bo); -} - -static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type) +static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type, int offset) { BATCH_LOCALS(&rmesa->radeon); + int size; + + /* offset is in indices */ + BEGIN_BATCH(10); + OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0); + if (rmesa->ind_buf.is_32bit) { + /* convert to bytes */ + offset *= 4; + size = vertex_count; + OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | + (vertex_count << 16) | type | + R300_VAP_VF_CNTL__INDEX_SIZE_32bit); + } else { + /* convert to bytes */ + offset *= 2; + size = (vertex_count + 1) >> 1; + OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | + (vertex_count << 16) | type); + } - r300_emit_scissor(rmesa->radeon.glCtx); - if (vertex_count > 0) { - int size; - - BEGIN_BATCH(10); - OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0); - if (rmesa->ind_buf.is_32bit) { - size = vertex_count; - OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | - ((vertex_count + 0) << 16) | type | - R300_VAP_VF_CNTL__INDEX_SIZE_32bit); - } else { - size = (vertex_count + 1) >> 1; - OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | - ((vertex_count + 0) << 16) | type); - } - - if (!rmesa->radeon.radeonScreen->kernel_mm) { - OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2); - OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) | - (R300_VAP_PORT_IDX0 >> 2)); - OUT_BATCH_RELOC(rmesa->radeon.tcl.elt_dma_offset, - rmesa->radeon.tcl.elt_dma_bo, - rmesa->radeon.tcl.elt_dma_offset, - RADEON_GEM_DOMAIN_GTT, 0, 0); - OUT_BATCH(size); - } else { - OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2); - OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) | - (R300_VAP_PORT_IDX0 >> 2)); - OUT_BATCH(rmesa->radeon.tcl.elt_dma_offset); - OUT_BATCH(size); - radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, - rmesa->radeon.tcl.elt_dma_bo, - RADEON_GEM_DOMAIN_GTT, 0, 0); - } - END_BATCH(); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2); + OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) | + (R300_VAP_PORT_IDX0 >> 2)); + OUT_BATCH_RELOC(0, rmesa->ind_buf.bo, rmesa->ind_buf.bo_offset + offset, RADEON_GEM_DOMAIN_GTT, 0, 0); + OUT_BATCH(size); + } else { + OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2); + OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) | + (R300_VAP_PORT_IDX0 >> 2)); + OUT_BATCH(rmesa->ind_buf.bo_offset + offset); + OUT_BATCH(size); + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->ind_buf.bo, RADEON_GEM_DOMAIN_GTT, 0, 0); } + END_BATCH(); } static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset) @@ -340,7 +323,7 @@ static void r300FireAOS(r300ContextPtr rmesa, int vertex_count, int type) { BATCH_LOCALS(&rmesa->radeon); - r300_emit_scissor(rmesa->radeon.glCtx); + r300_emit_scissor(rmesa->radeon.glCtx); BEGIN_BATCH(3); OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0); OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (vertex_count << 16) | type); @@ -365,8 +348,16 @@ void r300RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim) */ rcommonEnsureCmdBufSpace(&rmesa->radeon, 128, __FUNCTION__); - if (rmesa->ind_buf.ptr) { - r300EmitElts(ctx, num_verts); + if (rmesa->ind_buf.bo) { + GLuint first, incr, offset = 0; + + if (!split_prim_inplace(prim & PRIM_MODE_MASK, &first, &incr) && + num_verts > 65500) { + WARN_ONCE("Fixme: can't handle spliting prim %d\n", prim); + return; + } + + r300EmitAOS(rmesa, rmesa->radeon.tcl.aos_count, 0); if (rmesa->radeon.radeonScreen->kernel_mm) { BEGIN_BATCH_NO_AUTOSTATE(2); @@ -374,45 +365,56 @@ void r300RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim) OUT_BATCH(rmesa->radeon.tcl.aos[0].count); END_BATCH(); } - r300FireEB(rmesa, num_verts, type); - } else { - r300EmitAOS(rmesa, rmesa->radeon.tcl.aos_count, start); - r300FireAOS(rmesa, num_verts, type); - } - COMMIT_BATCH(); -} - -static void r300RunRender(GLcontext * ctx, struct tnl_pipeline_stage *stage) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - int i; - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *vb = &tnl->vb; - - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s\n", __FUNCTION__); - r300UpdateShaders(rmesa); - r300EmitArrays(ctx); + r300_emit_scissor(rmesa->radeon.glCtx); + while (num_verts > 0) { + int nr; + int align; + + nr = MIN2(num_verts, 65535); + nr -= (nr - first) % incr; + + /* get alignment for IB correct */ + if (nr != num_verts) { + do { + align = nr * (rmesa->ind_buf.is_32bit ? 4 : 2); + if (align % 4) + nr -= incr; + } while(align % 4); + if (nr <= 0) { + WARN_ONCE("did the impossible happen? we never aligned nr to dword\n"); + return; + } + + } + r300FireEB(rmesa, nr, type, offset); - r300UpdateShaderStates(rmesa); + num_verts -= nr; + offset += nr; + } - r300EmitCacheFlush(rmesa); - radeonEmitState(&rmesa->radeon); + } else { + GLuint first, incr, offset = 0; - for (i = 0; i < vb->PrimitiveCount; i++) { - GLuint prim = _tnl_translate_prim(&vb->Primitive[i]); - GLuint start = vb->Primitive[i].start; - GLuint end = vb->Primitive[i].start + vb->Primitive[i].count; - r300RunRenderPrimitive(ctx, start, end, prim); + if (!split_prim_inplace(prim & PRIM_MODE_MASK, &first, &incr) && + num_verts > 65500) { + WARN_ONCE("Fixme: can't handle spliting prim %d\n", prim); + return; + } + r300_emit_scissor(rmesa->radeon.glCtx); + while (num_verts > 0) { + int nr; + nr = MIN2(num_verts, 65535); + nr -= (nr - first) % incr; + r300EmitAOS(rmesa, rmesa->radeon.tcl.aos_count, start + offset); + r300FireAOS(rmesa, nr, type); + num_verts -= nr; + offset += nr; + } } - - r300EmitCacheFlush(rmesa); - - radeonReleaseArrays(ctx, ~0); + COMMIT_BATCH(); } - static const char *getFallbackString(uint32_t bit) { switch (bit) { @@ -449,7 +451,7 @@ void r300SwitchFallback(GLcontext *ctx, uint32_t bit, GLboolean mode) r300ContextPtr rmesa = R300_CONTEXT(ctx); uint32_t old_fallback = rmesa->fallback; static uint32_t fallback_warn = 0; - + if (mode) { if ((fallback_warn & bit) == 0) { if (RADEON_DEBUG & DEBUG_FALLBACKS) @@ -470,7 +472,7 @@ void r300SwitchFallback(GLcontext *ctx, uint32_t bit, GLboolean mode) /* update only if we change from no raster fallbacks to some raster fallbacks */ if (((old_fallback & R300_RASTER_FALLBACK_MASK) == 0) && ((bit & R300_RASTER_FALLBACK_MASK) > 0)) { - + radeon_firevertices(&rmesa->radeon); rmesa->radeon.swtcl.RenderIndex = ~0; _swsetup_Wakeup( ctx ); @@ -489,7 +491,7 @@ void r300SwitchFallback(GLcontext *ctx, uint32_t bit, GLboolean mode) /* update only if we have disabled all raster fallbacks */ if ((old_fallback & R300_RASTER_FALLBACK_MASK) == bit) { _swrast_flush( ctx ); - + tnl->Driver.Render.Start = r300RenderStart; tnl->Driver.Render.Finish = r300RenderFinish; tnl->Driver.Render.PrimitiveNotify = r300RenderPrimitive; @@ -497,38 +499,10 @@ void r300SwitchFallback(GLcontext *ctx, uint32_t bit, GLboolean mode) tnl->Driver.Render.BuildVertices = _tnl_build_vertices; tnl->Driver.Render.CopyPV = _tnl_copy_pv; tnl->Driver.Render.Interp = _tnl_interp; - + _tnl_invalidate_vertex_state( ctx, ~0 ); _tnl_invalidate_vertices( ctx, ~0 ); } } - -} -static GLboolean r300RunNonTCLRender(GLcontext * ctx, - struct tnl_pipeline_stage *stage) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (rmesa->fallback & R300_RASTER_FALLBACK_MASK) - return GL_TRUE; - - if (rmesa->options.hw_tcl_enabled == GL_FALSE) - return GL_TRUE; - - r300RunRender(ctx, stage); - - return GL_FALSE; } - -const struct tnl_pipeline_stage _r300_render_stage = { - "r300 Hardware Rasterization", - NULL, - NULL, - NULL, - NULL, - r300RunNonTCLRender -}; diff --git a/src/mesa/drivers/dri/r300/r300_shader.c b/src/mesa/drivers/dri/r300/r300_shader.c index 62228a3786..a4f9db13ec 100644 --- a/src/mesa/drivers/dri/r300/r300_shader.c +++ b/src/mesa/drivers/dri/r300/r300_shader.c @@ -38,7 +38,7 @@ static void freeFragProgCache(GLcontext *ctx, struct r300_fragment_program_cont while (fp) { tmp = fp->next; - _mesa_reference_program(ctx, &fp->Base, NULL); + rc_constants_destroy(&fp->code.constants); _mesa_free(fp); fp = tmp; } @@ -50,6 +50,7 @@ static void freeVertProgCache(GLcontext *ctx, struct r300_vertex_program_cont *c while (vp) { tmp = vp->next; + rc_constants_destroy(&vp->code.constants); _mesa_reference_vertprog(ctx, &vp->Base, NULL); _mesa_free(vp); vp = tmp; @@ -122,15 +123,11 @@ static GLboolean r300IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog) { if (target == GL_FRAGMENT_PROGRAM_ARB) { - struct r300_fragment_program *fp = r300SelectFragmentShader(ctx); - if (!fp->translated) - r300TranslateFragmentShader(ctx, fp); + struct r300_fragment_program *fp = r300SelectAndTranslateFragmentShader(ctx); return !fp->error; } else { - struct r300_vertex_program *vp = r300SelectVertexShader(ctx); - if (!vp->translated) - r300TranslateVertexShader(vp); + struct r300_vertex_program *vp = r300SelectAndTranslateVertexShader(ctx); return !vp->error; } diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c index 12fbf281d9..6081c33786 100644 --- a/src/mesa/drivers/dri/r300/r300_state.c +++ b/src/mesa/drivers/dri/r300/r300_state.c @@ -62,8 +62,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_emit.h" #include "r300_tex.h" #include "r300_fragprog_common.h" -#include "r300_fragprog.h" -#include "r500_fragprog.h" #include "r300_render.h" #include "r300_vertprog.h" @@ -458,7 +456,7 @@ static GLboolean current_fragment_program_writes_depth(GLcontext* ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); - return ctx->FragmentProgram._Current && r300->selected_fp->writes_depth; + return ctx->FragmentProgram._Current && r300->selected_fp->code.writes_depth; } static void r300SetEarlyZState(GLcontext * ctx) @@ -473,6 +471,8 @@ static void r300SetEarlyZState(GLcontext * ctx) topZ = R300_ZTOP_DISABLE; else if (ctx->FragmentProgram._Current && ctx->FragmentProgram._Current->UsesKill) topZ = R300_ZTOP_DISABLE; + else if (r300->query.current) + topZ = R300_ZTOP_DISABLE; if (topZ != r300->hw.zstencil_format.cmd[2]) { /* Note: This completely reemits the stencil format. @@ -1046,53 +1046,6 @@ void r300UpdateViewportOffset(GLcontext * ctx) radeonUpdateScissor(ctx); } -static void -r300FetchStateParameter(GLcontext * ctx, - const gl_state_index state[STATE_LENGTH], - GLfloat * value) -{ - r300ContextPtr r300 = R300_CONTEXT(ctx); - - switch (state[0]) { - case STATE_INTERNAL: - switch (state[1]) { - case STATE_R300_WINDOW_DIMENSION: { - __DRIdrawablePrivate * drawable = radeon_get_drawable(&r300->radeon); - value[0] = drawable->w * 0.5f; /* width*0.5 */ - value[1] = drawable->h * 0.5f; /* height*0.5 */ - value[2] = 0.5F; /* for moving range [-1 1] -> [0 1] */ - value[3] = 1.0F; /* not used */ - break; - } - - case STATE_R300_TEXRECT_FACTOR:{ - struct gl_texture_object *t = - ctx->Texture.Unit[state[2]].CurrentTex[TEXTURE_RECT_INDEX]; - - if (t && t->Image[0][t->BaseLevel]) { - struct gl_texture_image *image = - t->Image[0][t->BaseLevel]; - value[0] = 1.0 / image->Width2; - value[1] = 1.0 / image->Height2; - } else { - value[0] = 1.0; - value[1] = 1.0; - } - value[2] = 1.0; - value[3] = 1.0; - break; - } - - default: - break; - } - break; - - default: - break; - } -} - /** * Update R300's own internal state parameters. * For now just STATE_R300_WINDOW_DIMENSION @@ -1101,7 +1054,6 @@ static void r300UpdateStateParameters(GLcontext * ctx, GLuint new_state) { r300ContextPtr rmesa = R300_CONTEXT(ctx); struct gl_program_parameter_list *paramList; - GLuint i; if (!(new_state & (_NEW_BUFFERS | _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS))) return; @@ -1109,21 +1061,12 @@ static void r300UpdateStateParameters(GLcontext * ctx, GLuint new_state) if (!ctx->FragmentProgram._Current || !rmesa->selected_fp) return; - paramList = rmesa->selected_fp->Base->Parameters; + paramList = ctx->FragmentProgram._Current->Base.Parameters; if (!paramList) return; _mesa_load_state_parameters(ctx, paramList); - - for (i = 0; i < paramList->NumParameters; i++) { - if (paramList->Parameters[i].Type == PROGRAM_STATE_VAR) { - r300FetchStateParameter(ctx, - paramList->Parameters[i]. - StateIndexes, - paramList->ParameterValues[i]); - } - } } /* ============================================================= @@ -1230,7 +1173,7 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings) { r300ContextPtr r300 = R300_CONTEXT(ctx); int i; - struct r300_fragment_program_code *code = &r300->selected_fp->code.r300; + struct r300_fragment_program_code *code = &r300->selected_fp->code.code.r300; R300_STATECHANGE(r300, fpt); @@ -1272,7 +1215,7 @@ static void r500SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings) { r300ContextPtr r300 = R300_CONTEXT(ctx); int i; - struct r500_fragment_program_code *code = &r300->selected_fp->code.r500; + struct r500_fragment_program_code *code = &r300->selected_fp->code.code.r500; /* find all the texture instructions and relocate the texture units */ for (i = 0; i < code->inst_end + 1; i++) { @@ -1312,6 +1255,7 @@ static GLuint translate_lod_bias(GLfloat bias) return (((GLuint)b) << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK; } + static void r300SetupTextures(GLcontext * ctx) { int i, mtu; @@ -1404,6 +1348,28 @@ static void r300SetupTextures(GLcontext * ctx) } } + /* R3xx and R4xx chips require that the texture unit corresponding to + * KIL instructions is really enabled. + * + * We do some fakery here and in the state atom emit logic to enable + * the texture without tripping up the CS checker in the kernel. + */ + if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) { + if (ctx->FragmentProgram._Current->UsesKill && last_hw_tmu < 0) { + last_hw_tmu++; + + r300->hw.txe.cmd[R300_TXE_ENABLE] |= 1; + + r300->hw.tex.border_color.cmd[R300_TEX_VALUE_0] = 0; + r300->hw.tex.chroma_key.cmd[R300_TEX_VALUE_0] = 0; + r300->hw.tex.filter.cmd[R300_TEX_VALUE_0] = 0; + r300->hw.tex.filter_1.cmd[R300_TEX_VALUE_0] = 0; + r300->hw.tex.size.cmd[R300_TEX_VALUE_0] = 0; /* 1x1 texture */ + r300->hw.tex.format.cmd[R300_TEX_VALUE_0] = 0; /* A8 format */ + r300->hw.tex.pitch.cmd[R300_TEX_VALUE_0] = 0; + } + } + r300->hw.tex.filter.cmd[R300_TEX_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER0_0, last_hw_tmu + 1); r300->hw.tex.filter_1.cmd[R300_TEX_CMD_0] = @@ -1421,16 +1387,6 @@ static void r300SetupTextures(GLcontext * ctx) r300->hw.tex.border_color.cmd[R300_TEX_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_BORDER_COLOR_0, last_hw_tmu + 1); - if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) { - if (ctx->FragmentProgram._Current->UsesKill && last_hw_tmu < 0) { - // The KILL operation requires the first texture unit - // to be enabled. - r300->hw.txe.cmd[R300_TXE_ENABLE] |= 1; - r300->hw.tex.filter.cmd[R300_TEX_VALUE_0] = 0; - r300->hw.tex.filter.cmd[R300_TEX_CMD_0] = - cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER0_0, 1); - } - } r300->vtbl.SetupFragmentShaderTextures(ctx, tmu_mappings); if (RADEON_DEBUG & DEBUG_STATE) @@ -1460,11 +1416,11 @@ static void r300SetupRSUnit(GLcontext * ctx) hw_tcl_on = r300->options.hw_tcl_enabled; if (hw_tcl_on) - OutputsWritten.vp_outputs = r300->selected_vp->Base->Base.OutputsWritten; + OutputsWritten.vp_outputs = r300->selected_vp->code.OutputsWritten; else RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->render_inputs_bitset); - InputsRead = r300->selected_fp->Base->InputsRead; + InputsRead = r300->selected_fp->InputsRead; R300_STATECHANGE(r300, ri); R300_STATECHANGE(r300, rc); @@ -1554,11 +1510,11 @@ static void r500SetupRSUnit(GLcontext * ctx) hw_tcl_on = r300->options.hw_tcl_enabled; if (hw_tcl_on) - OutputsWritten.vp_outputs = r300->selected_vp->Base->Base.OutputsWritten; + OutputsWritten.vp_outputs = r300->selected_vp->code.OutputsWritten; else RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->render_inputs_bitset); - InputsRead = r300->selected_fp->Base->InputsRead; + InputsRead = r300->selected_fp->InputsRead; R300_STATECHANGE(r300, ri); R300_STATECHANGE(r300, rc); @@ -1997,9 +1953,7 @@ void r300UpdateShaders(r300ContextPtr rmesa) { struct r300_fragment_program *fp; - fp = r300SelectFragmentShader(ctx); - if (!fp->translated) - r300TranslateFragmentShader(ctx, fp); + fp = r300SelectAndTranslateFragmentShader(ctx); r300SwitchFallback(ctx, R300_FALLBACK_FRAGMENT_PROGRAM, fp->error); } @@ -2024,9 +1978,7 @@ void r300UpdateShaders(r300ContextPtr rmesa) } } - vp = r300SelectVertexShader(ctx); - if (!vp->translated) - r300TranslateVertexShader(vp); + vp = r300SelectAndTranslateVertexShader(ctx); r300SwitchFallback(ctx, R300_FALLBACK_VERTEX_PROGRAM, vp->error); } @@ -2035,24 +1987,61 @@ void r300UpdateShaders(r300ContextPtr rmesa) rmesa->radeon.NewGLState = 0; } -static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx, - struct gl_program *program, struct prog_src_register srcreg) +static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx, GLuint index, GLfloat * buffer) { static const GLfloat dummy[4] = { 0, 0, 0, 0 }; + r300ContextPtr rmesa = R300_CONTEXT(ctx); + struct rc_constant * rcc = &rmesa->selected_fp->code.constants.Constants[index]; + + switch(rcc->Type) { + case RC_CONSTANT_EXTERNAL: + return ctx->FragmentProgram._Current->Base.Parameters->ParameterValues[rcc->u.External]; + case RC_CONSTANT_IMMEDIATE: + return rcc->u.Immediate; + case RC_CONSTANT_STATE: + switch(rcc->u.State[0]) { + case RC_STATE_SHADOW_AMBIENT: { + const int unit = (int) rcc->u.State[1]; + const struct gl_texture_object *texObj = ctx->Texture.Unit[unit]._Current; + if (texObj) { + buffer[0] = + buffer[1] = + buffer[2] = + buffer[3] = texObj->CompareFailValue; + } + return buffer; + } - switch(srcreg.File) { - case PROGRAM_LOCAL_PARAM: - return program->LocalParams[srcreg.Index]; - case PROGRAM_ENV_PARAM: - return ctx->FragmentProgram.Parameters[srcreg.Index]; - case PROGRAM_STATE_VAR: - case PROGRAM_NAMED_PARAM: - case PROGRAM_CONSTANT: - return program->Parameters->ParameterValues[srcreg.Index]; - default: - _mesa_problem(ctx, "get_fragmentprogram_constant: Unknown\n"); - return dummy; + case RC_STATE_R300_WINDOW_DIMENSION: { + __DRIdrawablePrivate * drawable = radeon_get_drawable(&rmesa->radeon); + buffer[0] = drawable->w * 0.5f; /* width*0.5 */ + buffer[1] = drawable->h * 0.5f; /* height*0.5 */ + buffer[2] = 0.5F; /* for moving range [-1 1] -> [0 1] */ + buffer[3] = 1.0F; /* not used */ + return buffer; + } + + case RC_STATE_R300_TEXRECT_FACTOR: { + struct gl_texture_object *t = + ctx->Texture.Unit[rcc->u.State[1]].CurrentTex[TEXTURE_RECT_INDEX]; + + if (t && t->Image[0][t->BaseLevel]) { + struct gl_texture_image *image = + t->Image[0][t->BaseLevel]; + buffer[0] = 1.0 / image->Width2; + buffer[1] = 1.0 / image->Height2; + } else { + buffer[0] = 1.0; + buffer[1] = 1.0; + } + buffer[2] = 1.0; + buffer[3] = 1.0; + return buffer; + } + } } + + return dummy; } @@ -2061,9 +2050,9 @@ static void r300SetupPixelShader(GLcontext *ctx) r300ContextPtr rmesa = R300_CONTEXT(ctx); struct r300_fragment_program *fp = rmesa->selected_fp; struct r300_fragment_program_code *code; - int i, k; + int i; - code = &fp->code.r300; + code = &fp->code.code.r300; R300_STATECHANGE(rmesa, fpi[0]); R300_STATECHANGE(rmesa, fpi[1]); @@ -2074,38 +2063,24 @@ static void r300SetupPixelShader(GLcontext *ctx) rmesa->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_ALPHA_INST_0, code->alu.length); rmesa->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_ALPHA_ADDR_0, code->alu.length); for (i = 0; i < code->alu.length; i++) { - rmesa->hw.fpi[0].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst0; - rmesa->hw.fpi[1].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst1; - rmesa->hw.fpi[2].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst2; - rmesa->hw.fpi[3].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst3; + rmesa->hw.fpi[0].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].rgb_inst; + rmesa->hw.fpi[1].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].rgb_addr; + rmesa->hw.fpi[2].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].alpha_inst; + rmesa->hw.fpi[3].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].alpha_addr; } R300_STATECHANGE(rmesa, fp); - rmesa->hw.fp.cmd[R300_FP_CNTL0] = code->cur_node | (code->first_node_has_tex << 3); - rmesa->hw.fp.cmd[R300_FP_CNTL1] = code->max_temp_idx; - rmesa->hw.fp.cmd[R300_FP_CNTL2] = - (0 << R300_PFS_CNTL_ALU_OFFSET_SHIFT) | - ((code->alu.length-1) << R300_PFS_CNTL_ALU_END_SHIFT) | - (0 << R300_PFS_CNTL_TEX_OFFSET_SHIFT) | - ((code->tex.length ? code->tex.length-1 : 0) << R300_PFS_CNTL_TEX_END_SHIFT); - /* I just want to say, the way these nodes are stored.. weird.. */ - for (i = 0, k = (4 - (code->cur_node + 1)); i < 4; i++, k++) { - if (i < (code->cur_node + 1)) { - rmesa->hw.fp.cmd[R300_FP_NODE0 + k] = - (code->node[i].alu_offset << R300_ALU_START_SHIFT) | - (code->node[i].alu_end << R300_ALU_SIZE_SHIFT) | - (code->node[i].tex_offset << R300_TEX_START_SHIFT) | - (code->node[i].tex_end << R300_TEX_SIZE_SHIFT) | - code->node[i].flags; - } else { - rmesa->hw.fp.cmd[R300_FP_NODE0 + (3 - i)] = 0; - } - } + rmesa->hw.fp.cmd[R300_FP_CNTL0] = code->config; + rmesa->hw.fp.cmd[R300_FP_CNTL1] = code->pixsize; + rmesa->hw.fp.cmd[R300_FP_CNTL2] = code->code_offset; + for (i = 0; i < 4; i++) + rmesa->hw.fp.cmd[R300_FP_NODE0 + i] = code->code_addr[i]; R300_STATECHANGE(rmesa, fpp); - rmesa->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_PFS_PARAM_0_X, code->const_nr * 4); - for (i = 0; i < code->const_nr; i++) { - const GLfloat *constant = get_fragmentprogram_constant(ctx, fp->Base, code->constant[i]); + rmesa->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_PFS_PARAM_0_X, fp->code.constants.Count * 4); + for (i = 0; i < fp->code.constants.Count; i++) { + GLfloat buffer[4]; + const GLfloat *constant = get_fragmentprogram_constant(ctx, i, buffer); rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 0] = r300PackFloat24(constant[0]); rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 1] = r300PackFloat24(constant[1]); rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 2] = r300PackFloat24(constant[2]); @@ -2137,19 +2112,19 @@ static void r500SetupPixelShader(GLcontext *ctx) ((drm_r300_cmd_header_t *) rmesa->hw.r500fp.cmd)->r500fp.count = 0; ((drm_r300_cmd_header_t *) rmesa->hw.r500fp_const.cmd)->r500fp.count = 0; - code = &fp->code.r500; + code = &fp->code.code.r500; R300_STATECHANGE(rmesa, fp); rmesa->hw.fp.cmd[R500_FP_PIXSIZE] = code->max_temp_idx; rmesa->hw.fp.cmd[R500_FP_CODE_ADDR] = - R500_US_CODE_START_ADDR(code->inst_offset) | + R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end); rmesa->hw.fp.cmd[R500_FP_CODE_RANGE] = - R500_US_CODE_RANGE_ADDR(code->inst_offset) | + R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end); rmesa->hw.fp.cmd[R500_FP_CODE_OFFSET] = - R500_US_CODE_OFFSET_ADDR(0); /* FIXME when we add flow control */ + R500_US_CODE_OFFSET_ADDR(0); R300_STATECHANGE(rmesa, r500fp); /* Emit our shader... */ @@ -2165,14 +2140,15 @@ static void r500SetupPixelShader(GLcontext *ctx) bump_r500fp_count(rmesa->hw.r500fp.cmd, (code->inst_end + 1) * 6); R300_STATECHANGE(rmesa, r500fp_const); - for (i = 0; i < code->const_nr; i++) { - const GLfloat *constant = get_fragmentprogram_constant(ctx, fp->Base, code->constant[i]); + for (i = 0; i < fp->code.constants.Count; i++) { + GLfloat buffer[4]; + const GLfloat *constant = get_fragmentprogram_constant(ctx, i, buffer); rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 0] = r300PackFloat32(constant[0]); rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 1] = r300PackFloat32(constant[1]); rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 2] = r300PackFloat32(constant[2]); rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 3] = r300PackFloat32(constant[3]); } - bump_r500fp_const_count(rmesa->hw.r500fp_const.cmd, code->const_nr * 4); + bump_r500fp_const_count(rmesa->hw.r500fp_const.cmd, fp->code.constants.Count * 4); } void r300SetupVAP(GLcontext *ctx, GLuint InputsRead, GLuint OutputsWritten) @@ -2273,6 +2249,7 @@ static void r300InvalidateState(GLcontext * ctx, GLuint new_state) _mesa_update_draw_buffer_bounds(ctx); R300_STATECHANGE(r300, cb); + R300_STATECHANGE(r300, zb); } r300->radeon.NewGLState |= new_state; @@ -2345,13 +2322,9 @@ void r300InitShaderFunctions(r300ContextPtr r300) r300->vtbl.SetupRSUnit = r500SetupRSUnit; r300->vtbl.SetupPixelShader = r500SetupPixelShader; r300->vtbl.SetupFragmentShaderTextures = r500SetupFragmentShaderTextures; - r300->vtbl.BuildFragmentProgramHwCode = r500BuildFragmentProgramHwCode; - r300->vtbl.FragmentProgramDump = r500FragmentProgramDump; } else { r300->vtbl.SetupRSUnit = r300SetupRSUnit; r300->vtbl.SetupPixelShader = r300SetupPixelShader; r300->vtbl.SetupFragmentShaderTextures = r300SetupFragmentShaderTextures; - r300->vtbl.BuildFragmentProgramHwCode = r300BuildFragmentProgramHwCode; - r300->vtbl.FragmentProgramDump = r300FragmentProgramDump; } } diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.c b/src/mesa/drivers/dri/r300/r300_swtcl.c index 56ed519cf4..a634cb5192 100644 --- a/src/mesa/drivers/dri/r300/r300_swtcl.c +++ b/src/mesa/drivers/dri/r300/r300_swtcl.c @@ -76,7 +76,7 @@ void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *_InputsRead, GLuint *_ GLuint InputsRead = 0; GLuint OutputsWritten = 0; int num_attrs = 0; - GLuint fp_reads = rmesa->selected_fp->Base->InputsRead; + GLuint fp_reads = rmesa->selected_fp->InputsRead; struct vertex_attribute *attrs = rmesa->vbuf.attribs; rmesa->swtcl.coloroffset = rmesa->swtcl.specoffset = 0; diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c index de32013032..c5edbd0052 100644 --- a/src/mesa/drivers/dri/r300/r300_vertprog.c +++ b/src/mesa/drivers/dri/r300/r300_vertprog.c @@ -40,927 +40,124 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "shader/prog_statevars.h" #include "tnl/tnl.h" -#include "radeon_nqssadce.h" +#include "compiler/radeon_compiler.h" +#include "compiler/radeon_nqssadce.h" #include "r300_context.h" #include "r300_state.h" -/* TODO: Get rid of t_src_class call */ -#define CMP_SRCS(a, b) ((a.RelAddr != b.RelAddr) || (a.Index != b.Index && \ - ((t_src_class(a.File) == PVS_SRC_REG_CONSTANT && \ - t_src_class(b.File) == PVS_SRC_REG_CONSTANT) || \ - (t_src_class(a.File) == PVS_SRC_REG_INPUT && \ - t_src_class(b.File) == PVS_SRC_REG_INPUT)))) \ - -/* - * Take an already-setup and valid source then swizzle it appropriately to - * obtain a constant ZERO or ONE source. +/** + * Write parameter array for the given vertex program into dst. + * Return the total number of components written. */ -#define __CONST(x, y) \ - (PVS_SRC_OPERAND(t_src_index(vp, &src[x]), \ - t_swizzle(y), \ - t_swizzle(y), \ - t_swizzle(y), \ - t_swizzle(y), \ - t_src_class(src[x].File), \ - VSF_FLAG_NONE) | (src[x].RelAddr << 4)) - -#define FREE_TEMPS() \ - do { \ - int u_temp_used = (VSF_MAX_FRAGMENT_TEMPS - 1) - u_temp_i; \ - if((vp->num_temporaries + u_temp_used) > VSF_MAX_FRAGMENT_TEMPS) { \ - WARN_ONCE("Ran out of temps, num temps %d, us %d\n", vp->num_temporaries, u_temp_used); \ - vp->error = GL_TRUE; \ - } \ - u_temp_i=VSF_MAX_FRAGMENT_TEMPS-1; \ - } while (0) - -static int r300VertexProgUpdateParams(GLcontext * ctx, struct gl_vertex_program *vp, float *dst) +static int r300VertexProgUpdateParams(GLcontext * ctx, struct r300_vertex_program *vp, float *dst) { - int pi; - float *dst_o = dst; - struct gl_program_parameter_list *paramList; + int i; - if (vp->IsNVProgram) { + if (vp->Base->IsNVProgram) { _mesa_load_tracked_matrices(ctx); - - for (pi = 0; pi < MAX_NV_VERTEX_PROGRAM_PARAMS; pi++) { - *dst++ = ctx->VertexProgram.Parameters[pi][0]; - *dst++ = ctx->VertexProgram.Parameters[pi][1]; - *dst++ = ctx->VertexProgram.Parameters[pi][2]; - *dst++ = ctx->VertexProgram.Parameters[pi][3]; + } else { + if (vp->Base->Base.Parameters) { + _mesa_load_state_parameters(ctx, vp->Base->Base.Parameters); } - return dst - dst_o; } - if (!vp->Base.Parameters) - return 0; - - _mesa_load_state_parameters(ctx, vp->Base.Parameters); - - if (vp->Base.Parameters->NumParameters * 4 > - VSF_MAX_FRAGMENT_LENGTH) { + if (vp->code.constants.Count * 4 > VSF_MAX_FRAGMENT_LENGTH) { + /* Should have checked this earlier... */ fprintf(stderr, "%s:Params exhausted\n", __FUNCTION__); _mesa_exit(-1); } - paramList = vp->Base.Parameters; - for (pi = 0; pi < paramList->NumParameters; pi++) { - switch (paramList->Parameters[pi].Type) { - case PROGRAM_STATE_VAR: - case PROGRAM_NAMED_PARAM: - //fprintf(stderr, "%s", vp->Parameters->Parameters[pi].Name); - case PROGRAM_CONSTANT: - *dst++ = paramList->ParameterValues[pi][0]; - *dst++ = paramList->ParameterValues[pi][1]; - *dst++ = paramList->ParameterValues[pi][2]; - *dst++ = paramList->ParameterValues[pi][3]; - break; - default: - _mesa_problem(NULL, "Bad param type in %s", - __FUNCTION__); - } - - } - - return dst - dst_o; -} - -static unsigned long t_dst_mask(GLuint mask) -{ - /* WRITEMASK_* is equivalent to VSF_FLAG_* */ - return mask & VSF_FLAG_ALL; -} - -static unsigned long t_dst_class(gl_register_file file) -{ - - switch (file) { - case PROGRAM_TEMPORARY: - return PVS_DST_REG_TEMPORARY; - case PROGRAM_OUTPUT: - return PVS_DST_REG_OUT; - case PROGRAM_ADDRESS: - return PVS_DST_REG_A0; - /* - case PROGRAM_INPUT: - case PROGRAM_LOCAL_PARAM: - case PROGRAM_ENV_PARAM: - case PROGRAM_NAMED_PARAM: - case PROGRAM_STATE_VAR: - case PROGRAM_WRITE_ONLY: - case PROGRAM_ADDRESS: - */ - default: - fprintf(stderr, "problem in %s", __FUNCTION__); - _mesa_exit(-1); - return -1; - } -} + for(i = 0; i < vp->code.constants.Count; ++i) { + const float * src = 0; + const struct rc_constant * constant = &vp->code.constants.Constants[i]; -static unsigned long t_dst_index(struct r300_vertex_program *vp, - struct prog_dst_register *dst) -{ - if (dst->File == PROGRAM_OUTPUT) - return vp->outputs[dst->Index]; + switch(constant->Type) { + case RC_CONSTANT_EXTERNAL: + if (vp->Base->IsNVProgram) { + src = ctx->VertexProgram.Parameters[constant->u.External]; + } else { + src = vp->Base->Base.Parameters->ParameterValues[constant->u.External]; + } + break; - return dst->Index; -} + case RC_CONSTANT_IMMEDIATE: + src = constant->u.Immediate; + break; + } -static unsigned long t_src_class(gl_register_file file) -{ - switch (file) { - case PROGRAM_TEMPORARY: - return PVS_SRC_REG_TEMPORARY; - case PROGRAM_INPUT: - return PVS_SRC_REG_INPUT; - case PROGRAM_LOCAL_PARAM: - case PROGRAM_ENV_PARAM: - case PROGRAM_NAMED_PARAM: - case PROGRAM_CONSTANT: - case PROGRAM_STATE_VAR: - return PVS_SRC_REG_CONSTANT; - /* - case PROGRAM_OUTPUT: - case PROGRAM_WRITE_ONLY: - case PROGRAM_ADDRESS: - */ - default: - fprintf(stderr, "problem in %s", __FUNCTION__); - _mesa_exit(-1); - return -1; + dst[4*i] = src[0]; + dst[4*i + 1] = src[1]; + dst[4*i + 2] = src[2]; + dst[4*i + 3] = src[3]; } -} -static INLINE unsigned long t_swizzle(GLubyte swizzle) -{ -/* this is in fact a NOP as the Mesa SWIZZLE_* are all identical to VSF_IN_COMPONENT_* */ - return swizzle; + return 4 * vp->code.constants.Count; } -#if 0 -static void vp_dump_inputs(struct r300_vertex_program *vp, char *caller) +static GLbitfield compute_required_outputs(struct gl_vertex_program * vp, GLbitfield fpreads) { + GLbitfield outputs = 0; int i; - if (vp == NULL) { - fprintf(stderr, "vp null in call to %s from %s\n", __FUNCTION__, - caller); - return; - } - - fprintf(stderr, "%s:<", caller); - for (i = 0; i < VERT_ATTRIB_MAX; i++) - fprintf(stderr, "%d ", vp->inputs[i]); - fprintf(stderr, ">\n"); - -} -#endif - -static unsigned long t_src_index(struct r300_vertex_program *vp, - struct prog_src_register *src) -{ - if (src->File == PROGRAM_INPUT) { - assert(vp->inputs[src->Index] != -1); - return vp->inputs[src->Index]; - } else { - if (src->Index < 0) { - fprintf(stderr, - "negative offsets for indirect addressing do not work.\n"); - return 0; - } - return src->Index; - } -} - -/* these two functions should probably be merged... */ - -static unsigned long t_src(struct r300_vertex_program *vp, - struct prog_src_register *src) -{ - /* src->Negate uses the NEGATE_ flags from program_instruction.h, - * which equal our VSF_FLAGS_ values, so it's safe to just pass it here. - */ - return PVS_SRC_OPERAND(t_src_index(vp, src), - t_swizzle(GET_SWZ(src->Swizzle, 0)), - t_swizzle(GET_SWZ(src->Swizzle, 1)), - t_swizzle(GET_SWZ(src->Swizzle, 2)), - t_swizzle(GET_SWZ(src->Swizzle, 3)), - t_src_class(src->File), - src->Negate) | (src->RelAddr << 4); -} +#define ADD_OUTPUT(fp_attr, vp_result) \ + do { \ + if (fpreads & (1 << (fp_attr))) \ + outputs |= (1 << (vp_result)); \ + } while (0) -static unsigned long t_src_scalar(struct r300_vertex_program *vp, - struct prog_src_register *src) -{ - /* src->Negate uses the NEGATE_ flags from program_instruction.h, - * which equal our VSF_FLAGS_ values, so it's safe to just pass it here. - */ - return PVS_SRC_OPERAND(t_src_index(vp, src), - t_swizzle(GET_SWZ(src->Swizzle, 0)), - t_swizzle(GET_SWZ(src->Swizzle, 0)), - t_swizzle(GET_SWZ(src->Swizzle, 0)), - t_swizzle(GET_SWZ(src->Swizzle, 0)), - t_src_class(src->File), - src->Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src->RelAddr << 4); -} + ADD_OUTPUT(FRAG_ATTRIB_COL0, VERT_RESULT_COL0); + ADD_OUTPUT(FRAG_ATTRIB_COL1, VERT_RESULT_COL1); -static GLboolean valid_dst(struct r300_vertex_program *vp, - struct prog_dst_register *dst) -{ - if (dst->File == PROGRAM_OUTPUT && vp->outputs[dst->Index] == -1) { - return GL_FALSE; - } else if (dst->File == PROGRAM_ADDRESS) { - assert(dst->Index == 0); + for (i = 0; i <= 7; ++i) { + ADD_OUTPUT(FRAG_ATTRIB_TEX0 + i, VERT_RESULT_TEX0 + i); } - return GL_TRUE; -} - -static GLuint *r300TranslateOpcodeABS(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - //MAX RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W - - inst[0] = PVS_OP_DST_OPERAND(VE_MAXIMUM, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), - t_swizzle(GET_SWZ(src[0].Swizzle, 1)), - t_swizzle(GET_SWZ(src[0].Swizzle, 2)), - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), - t_src_class(src[0].File), - (!src[0]. - Negate) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - inst[3] = 0; - - return inst; -} - -static GLuint *r300TranslateOpcodeADD(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_ADD, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeARL(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_FLT2FIX_DX, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeDP3(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - //DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ZERO} PARAM 0{} {X Y Z ZERO} - - inst[0] = PVS_OP_DST_OPERAND(VE_DOT_PRODUCT, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), - t_swizzle(GET_SWZ(src[0].Swizzle, 1)), - t_swizzle(GET_SWZ(src[0].Swizzle, 2)), - SWIZZLE_ZERO, - t_src_class(src[0].File), - src[0].Negate ? VSF_FLAG_XYZ : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - inst[2] = - PVS_SRC_OPERAND(t_src_index(vp, &src[1]), - t_swizzle(GET_SWZ(src[1].Swizzle, 0)), - t_swizzle(GET_SWZ(src[1].Swizzle, 1)), - t_swizzle(GET_SWZ(src[1].Swizzle, 2)), SWIZZLE_ZERO, - t_src_class(src[1].File), - src[1].Negate ? VSF_FLAG_XYZ : VSF_FLAG_NONE) | - (src[1].RelAddr << 4); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeDP4(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_DOT_PRODUCT, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeDPH(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - //DOT RESULT 1.X Y Z W PARAM 0{} {X Y Z ONE} PARAM 0{} {X Y Z W} - inst[0] = PVS_OP_DST_OPERAND(VE_DOT_PRODUCT, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), - t_swizzle(GET_SWZ(src[0].Swizzle, 1)), - t_swizzle(GET_SWZ(src[0].Swizzle, 2)), - PVS_SRC_SELECT_FORCE_1, - t_src_class(src[0].File), - src[0].Negate ? VSF_FLAG_XYZ : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeDST(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_DISTANCE_VECTOR, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeEX2(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(ME_EXP_BASE2_FULL_DX, - GL_TRUE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src_scalar(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeEXP(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(ME_EXP_BASE2_DX, - GL_TRUE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src_scalar(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeFLR(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3], - int *u_temp_i) -{ - /* FRC TMP 0.X Y Z W PARAM 0{} {X Y Z W} - ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} TMP 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W */ - - inst[0] = PVS_OP_DST_OPERAND(VE_FRACTION, - GL_FALSE, - GL_FALSE, - *u_temp_i, - t_dst_mask(vpi->DstReg.WriteMask), - PVS_DST_REG_TEMPORARY); - inst[1] = t_src(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - inst += 4; - - inst[0] = PVS_OP_DST_OPERAND(VE_ADD, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = PVS_SRC_OPERAND(*u_temp_i, - PVS_SRC_SELECT_X, - PVS_SRC_SELECT_Y, - PVS_SRC_SELECT_Z, - PVS_SRC_SELECT_W, PVS_SRC_REG_TEMPORARY, - /* Not 100% sure about this */ - (!src[0]. - Negate) ? VSF_FLAG_ALL : VSF_FLAG_NONE - /*VSF_FLAG_ALL */ ); - inst[3] = __CONST(0, SWIZZLE_ZERO); - (*u_temp_i)--; - - return inst; -} - -static GLuint *r300TranslateOpcodeFRC(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_FRACTION, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeLG2(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - // LG2 RESULT 1.X Y Z W PARAM 0{} {X X X X} - - inst[0] = PVS_OP_DST_OPERAND(ME_LOG_BASE2_FULL_DX, - GL_TRUE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), - t_src_class(src[0].File), - src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeLIT(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - //LIT TMP 1.Y Z TMP 1{} {X W Z Y} TMP 1{} {Y W Z X} TMP 1{} {Y X Z W} - - inst[0] = PVS_OP_DST_OPERAND(ME_LIGHT_COEFF_DX, - GL_TRUE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - /* NOTE: Users swizzling might not work. */ - inst[1] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // X - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // W - PVS_SRC_SELECT_FORCE_0, // Z - t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // Y - t_src_class(src[0].File), - src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - inst[2] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // Y - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // W - PVS_SRC_SELECT_FORCE_0, // Z - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // X - t_src_class(src[0].File), - src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - inst[3] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // Y - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // X - PVS_SRC_SELECT_FORCE_0, // Z - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // W - t_src_class(src[0].File), - src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - - return inst; -} - -static GLuint *r300TranslateOpcodeLOG(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(ME_LOG_BASE2_DX, - GL_TRUE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src_scalar(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeMAD(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(PVS_MACRO_OP_2CLK_MADD, - GL_FALSE, - GL_TRUE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = t_src(vp, &src[2]); - - return inst; -} - -static GLuint *r300TranslateOpcodeMAX(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_MAXIMUM, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeMIN(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_MINIMUM, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeMOV(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - //ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{} {ZERO ZERO ZERO ZERO} - - inst[0] = PVS_OP_DST_OPERAND(VE_ADD, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodeMUL(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_MULTIPLY, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} - -static GLuint *r300TranslateOpcodePOW(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(ME_POWER_FUNC_FF, - GL_TRUE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src_scalar(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = t_src_scalar(vp, &src[1]); - - return inst; -} - -static GLuint *r300TranslateOpcodeRCP(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(ME_RECIP_DX, - GL_TRUE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src_scalar(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} +#undef ADD_OUTPUT -static GLuint *r300TranslateOpcodeRSQ(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(ME_RECIP_SQRT_DX, - GL_TRUE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src_scalar(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} + if ((fpreads & (1 << FRAG_ATTRIB_COL0)) && + (vp->Base.OutputsWritten & (1 << VERT_RESULT_BFC0))) + outputs |= 1 << VERT_RESULT_BFC0; + if ((fpreads & (1 << FRAG_ATTRIB_COL1)) && + (vp->Base.OutputsWritten & (1 << VERT_RESULT_BFC1))) + outputs |= 1 << VERT_RESULT_BFC1; -static GLuint *r300TranslateOpcodeSGE(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_SET_GREATER_THAN_EQUAL, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; -} + outputs |= 1 << VERT_RESULT_HPOS; + if (vp->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) + outputs |= 1 << VERT_RESULT_PSIZ; -static GLuint *r300TranslateOpcodeSLT(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - inst[0] = PVS_OP_DST_OPERAND(VE_SET_LESS_THAN, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = t_src(vp, &src[1]); - inst[3] = __CONST(1, SWIZZLE_ZERO); - - return inst; + return outputs; } -static GLuint *r300TranslateOpcodeSUB(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - //ADD RESULT 1.X Y Z W TMP 0{} {X Y Z W} PARAM 1{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W - -#if 0 - inst[0] = PVS_OP_DST_OPERAND(VE_ADD, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = PVS_SRC_OPERAND(t_src_index(vp, &src[1]), - t_swizzle(GET_SWZ(src[1].Swizzle, 0)), - t_swizzle(GET_SWZ(src[1].Swizzle, 1)), - t_swizzle(GET_SWZ(src[1].Swizzle, 2)), - t_swizzle(GET_SWZ(src[1].Swizzle, 3)), - t_src_class(src[1].File), - (!src[1]. - Negate) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[1].RelAddr << 4); - inst[3] = 0; -#else - inst[0] = - PVS_OP_DST_OPERAND(VE_MULTIPLY_ADD, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ONE); - inst[3] = PVS_SRC_OPERAND(t_src_index(vp, &src[1]), - t_swizzle(GET_SWZ(src[1].Swizzle, 0)), - t_swizzle(GET_SWZ(src[1].Swizzle, 1)), - t_swizzle(GET_SWZ(src[1].Swizzle, 2)), - t_swizzle(GET_SWZ(src[1].Swizzle, 3)), - t_src_class(src[1].File), - (!src[1]. - Negate) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[1].RelAddr << 4); -#endif - - return inst; -} - -static GLuint *r300TranslateOpcodeSWZ(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3]) -{ - //ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{} {ZERO ZERO ZERO ZERO} - - inst[0] = PVS_OP_DST_OPERAND(VE_ADD, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = t_src(vp, &src[0]); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - - return inst; -} -static GLuint *r300TranslateOpcodeXPD(struct r300_vertex_program *vp, - struct prog_instruction *vpi, - GLuint * inst, - struct prog_src_register src[3], - int *u_temp_i) -{ - /* mul r0, r1.yzxw, r2.zxyw - mad r0, -r2.yzxw, r1.zxyw, r0 - */ - - inst[0] = PVS_OP_DST_OPERAND(VE_MULTIPLY_ADD, - GL_FALSE, - GL_FALSE, - *u_temp_i, - t_dst_mask(vpi->DstReg.WriteMask), - PVS_DST_REG_TEMPORARY); - inst[1] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // Y - t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // Z - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // X - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // W - t_src_class(src[0].File), - src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - inst[2] = PVS_SRC_OPERAND(t_src_index(vp, &src[1]), t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // Z - t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // X - t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // Y - t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // W - t_src_class(src[1].File), - src[1].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[1].RelAddr << 4); - inst[3] = __CONST(1, SWIZZLE_ZERO); - inst += 4; - - inst[0] = PVS_OP_DST_OPERAND(VE_MULTIPLY_ADD, - GL_FALSE, - GL_FALSE, - t_dst_index(vp, &vpi->DstReg), - t_dst_mask(vpi->DstReg.WriteMask), - t_dst_class(vpi->DstReg.File)); - inst[1] = PVS_SRC_OPERAND(t_src_index(vp, &src[1]), t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // Y - t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // Z - t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // X - t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // W - t_src_class(src[1].File), - (!src[1]. - Negate) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[1].RelAddr << 4); - inst[2] = PVS_SRC_OPERAND(t_src_index(vp, &src[0]), t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // Z - t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // X - t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // Y - t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // W - t_src_class(src[0].File), - src[0].Negate ? VSF_FLAG_ALL : VSF_FLAG_NONE) | - (src[0].RelAddr << 4); - inst[3] = - PVS_SRC_OPERAND(*u_temp_i, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, - PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, - PVS_SRC_REG_TEMPORARY, VSF_FLAG_NONE); - - (*u_temp_i)--; - - return inst; -} - -static void t_inputs_outputs(struct r300_vertex_program *vp) +static void t_inputs_outputs(struct r300_vertex_program_compiler * c) { int i; int cur_reg; GLuint OutputsWritten, InputsRead; - OutputsWritten = vp->Base->Base.OutputsWritten; - InputsRead = vp->Base->Base.InputsRead; + OutputsWritten = c->Base.Program.OutputsWritten; + InputsRead = c->Base.Program.InputsRead; cur_reg = -1; for (i = 0; i < VERT_ATTRIB_MAX; i++) { if (InputsRead & (1 << i)) - vp->inputs[i] = ++cur_reg; + c->code->inputs[i] = ++cur_reg; else - vp->inputs[i] = -1; + c->code->inputs[i] = -1; } cur_reg = 0; for (i = 0; i < VERT_RESULT_MAX; i++) - vp->outputs[i] = -1; + c->code->outputs[i] = -1; assert(OutputsWritten & (1 << VERT_RESULT_HPOS)); if (OutputsWritten & (1 << VERT_RESULT_HPOS)) { - vp->outputs[VERT_RESULT_HPOS] = cur_reg++; + c->code->outputs[VERT_RESULT_HPOS] = cur_reg++; } if (OutputsWritten & (1 << VERT_RESULT_PSIZ)) { - vp->outputs[VERT_RESULT_PSIZ] = cur_reg++; + c->code->outputs[VERT_RESULT_PSIZ] = cur_reg++; } /* If we're writing back facing colors we need to send @@ -970,668 +167,99 @@ static void t_inputs_outputs(struct r300_vertex_program *vp) * get written into appropriate output vectors. */ if (OutputsWritten & (1 << VERT_RESULT_COL0)) { - vp->outputs[VERT_RESULT_COL0] = cur_reg++; + c->code->outputs[VERT_RESULT_COL0] = cur_reg++; } else if (OutputsWritten & (1 << VERT_RESULT_BFC0) || OutputsWritten & (1 << VERT_RESULT_BFC1)) { cur_reg++; } if (OutputsWritten & (1 << VERT_RESULT_COL1)) { - vp->outputs[VERT_RESULT_COL1] = cur_reg++; + c->code->outputs[VERT_RESULT_COL1] = cur_reg++; } else if (OutputsWritten & (1 << VERT_RESULT_BFC0) || OutputsWritten & (1 << VERT_RESULT_BFC1)) { cur_reg++; } if (OutputsWritten & (1 << VERT_RESULT_BFC0)) { - vp->outputs[VERT_RESULT_BFC0] = cur_reg++; + c->code->outputs[VERT_RESULT_BFC0] = cur_reg++; } else if (OutputsWritten & (1 << VERT_RESULT_BFC1)) { cur_reg++; } if (OutputsWritten & (1 << VERT_RESULT_BFC1)) { - vp->outputs[VERT_RESULT_BFC1] = cur_reg++; + c->code->outputs[VERT_RESULT_BFC1] = cur_reg++; } else if (OutputsWritten & (1 << VERT_RESULT_BFC0)) { cur_reg++; } for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) { if (OutputsWritten & (1 << i)) { - vp->outputs[i] = cur_reg++; + c->code->outputs[i] = cur_reg++; } } if (OutputsWritten & (1 << VERT_RESULT_FOGC)) { - vp->outputs[VERT_RESULT_FOGC] = cur_reg++; + c->code->outputs[VERT_RESULT_FOGC] = cur_reg++; } } -void r300TranslateVertexShader(struct r300_vertex_program *vp) -{ - struct prog_instruction *vpi = vp->Base->Base.Instructions; - int i; - GLuint *inst; - unsigned long num_operands; - /* Initial value should be last tmp reg that hw supports. - Strangely enough r300 doesnt mind even though these would be out of range. - Smart enough to realize that it doesnt need it? */ - int u_temp_i = VSF_MAX_FRAGMENT_TEMPS - 1; - struct prog_src_register src[3]; - - vp->pos_end = 0; /* Not supported yet */ - vp->hw_code.length = 0; - vp->translated = GL_TRUE; - vp->error = GL_FALSE; - - t_inputs_outputs(vp); - - for (inst = vp->hw_code.body.d; vpi->Opcode != OPCODE_END; - vpi++, inst += 4) { - - FREE_TEMPS(); - - if (!valid_dst(vp, &vpi->DstReg)) { - /* redirect result to unused temp */ - vpi->DstReg.File = PROGRAM_TEMPORARY; - vpi->DstReg.Index = u_temp_i; - } - - num_operands = _mesa_num_inst_src_regs(vpi->Opcode); - - /* copy the sources (src) from mesa into a local variable... is this needed? */ - for (i = 0; i < num_operands; i++) { - src[i] = vpi->SrcReg[i]; - } - - if (num_operands == 3) { /* TODO: scalars */ - if (CMP_SRCS(src[1], src[2]) - || CMP_SRCS(src[0], src[2])) { - inst[0] = PVS_OP_DST_OPERAND(VE_ADD, - GL_FALSE, - GL_FALSE, - u_temp_i, - VSF_FLAG_ALL, - PVS_DST_REG_TEMPORARY); - inst[1] = - PVS_SRC_OPERAND(t_src_index(vp, &src[2]), - SWIZZLE_X, - SWIZZLE_Y, - SWIZZLE_Z, - SWIZZLE_W, - t_src_class(src[2].File), - VSF_FLAG_NONE) | (src[2]. - RelAddr << - 4); - inst[2] = __CONST(2, SWIZZLE_ZERO); - inst[3] = __CONST(2, SWIZZLE_ZERO); - inst += 4; - - src[2].File = PROGRAM_TEMPORARY; - src[2].Index = u_temp_i; - src[2].RelAddr = 0; - u_temp_i--; - } - } - - if (num_operands >= 2) { - if (CMP_SRCS(src[1], src[0])) { - inst[0] = PVS_OP_DST_OPERAND(VE_ADD, - GL_FALSE, - GL_FALSE, - u_temp_i, - VSF_FLAG_ALL, - PVS_DST_REG_TEMPORARY); - inst[1] = - PVS_SRC_OPERAND(t_src_index(vp, &src[0]), - SWIZZLE_X, - SWIZZLE_Y, - SWIZZLE_Z, - SWIZZLE_W, - t_src_class(src[0].File), - VSF_FLAG_NONE) | (src[0]. - RelAddr << - 4); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - inst += 4; - - src[0].File = PROGRAM_TEMPORARY; - src[0].Index = u_temp_i; - src[0].RelAddr = 0; - u_temp_i--; - } - } - - switch (vpi->Opcode) { - case OPCODE_ABS: - inst = r300TranslateOpcodeABS(vp, vpi, inst, src); - break; - case OPCODE_ADD: - inst = r300TranslateOpcodeADD(vp, vpi, inst, src); - break; - case OPCODE_ARL: - inst = r300TranslateOpcodeARL(vp, vpi, inst, src); - break; - case OPCODE_DP3: - inst = r300TranslateOpcodeDP3(vp, vpi, inst, src); - break; - case OPCODE_DP4: - inst = r300TranslateOpcodeDP4(vp, vpi, inst, src); - break; - case OPCODE_DPH: - inst = r300TranslateOpcodeDPH(vp, vpi, inst, src); - break; - case OPCODE_DST: - inst = r300TranslateOpcodeDST(vp, vpi, inst, src); - break; - case OPCODE_EX2: - inst = r300TranslateOpcodeEX2(vp, vpi, inst, src); - break; - case OPCODE_EXP: - inst = r300TranslateOpcodeEXP(vp, vpi, inst, src); - break; - case OPCODE_FLR: - inst = r300TranslateOpcodeFLR(vp, vpi, inst, src, /* FIXME */ - &u_temp_i); - break; - case OPCODE_FRC: - inst = r300TranslateOpcodeFRC(vp, vpi, inst, src); - break; - case OPCODE_LG2: - inst = r300TranslateOpcodeLG2(vp, vpi, inst, src); - break; - case OPCODE_LIT: - inst = r300TranslateOpcodeLIT(vp, vpi, inst, src); - break; - case OPCODE_LOG: - inst = r300TranslateOpcodeLOG(vp, vpi, inst, src); - break; - case OPCODE_MAD: - inst = r300TranslateOpcodeMAD(vp, vpi, inst, src); - break; - case OPCODE_MAX: - inst = r300TranslateOpcodeMAX(vp, vpi, inst, src); - break; - case OPCODE_MIN: - inst = r300TranslateOpcodeMIN(vp, vpi, inst, src); - break; - case OPCODE_MOV: - inst = r300TranslateOpcodeMOV(vp, vpi, inst, src); - break; - case OPCODE_MUL: - inst = r300TranslateOpcodeMUL(vp, vpi, inst, src); - break; - case OPCODE_POW: - inst = r300TranslateOpcodePOW(vp, vpi, inst, src); - break; - case OPCODE_RCP: - inst = r300TranslateOpcodeRCP(vp, vpi, inst, src); - break; - case OPCODE_RSQ: - inst = r300TranslateOpcodeRSQ(vp, vpi, inst, src); - break; - case OPCODE_SGE: - inst = r300TranslateOpcodeSGE(vp, vpi, inst, src); - break; - case OPCODE_SLT: - inst = r300TranslateOpcodeSLT(vp, vpi, inst, src); - break; - case OPCODE_SUB: - inst = r300TranslateOpcodeSUB(vp, vpi, inst, src); - break; - case OPCODE_SWZ: - inst = r300TranslateOpcodeSWZ(vp, vpi, inst, src); - break; - case OPCODE_XPD: - inst = r300TranslateOpcodeXPD(vp, vpi, inst, src, /* FIXME */ - &u_temp_i); - break; - default: - vp->error = GL_TRUE; - break; - } - } - - vp->hw_code.length = (inst - vp->hw_code.body.d); - if (vp->hw_code.length >= VSF_MAX_FRAGMENT_LENGTH) { - vp->error = GL_TRUE; - } -} - -static void insert_wpos(struct gl_program *prog, GLuint temp_index, int tex_id) -{ - struct prog_instruction *vpi; - - _mesa_insert_instructions(prog, prog->NumInstructions - 1, 2); - - vpi = &prog->Instructions[prog->NumInstructions - 3]; - - vpi->Opcode = OPCODE_MOV; - - vpi->DstReg.File = PROGRAM_OUTPUT; - vpi->DstReg.Index = VERT_RESULT_HPOS; - vpi->DstReg.WriteMask = WRITEMASK_XYZW; - vpi->DstReg.CondMask = COND_TR; - - vpi->SrcReg[0].File = PROGRAM_TEMPORARY; - vpi->SrcReg[0].Index = temp_index; - vpi->SrcReg[0].Swizzle = SWIZZLE_XYZW; - - ++vpi; - - vpi->Opcode = OPCODE_MOV; - - vpi->DstReg.File = PROGRAM_OUTPUT; - vpi->DstReg.Index = VERT_RESULT_TEX0 + tex_id; - vpi->DstReg.WriteMask = WRITEMASK_XYZW; - vpi->DstReg.CondMask = COND_TR; - - vpi->SrcReg[0].File = PROGRAM_TEMPORARY; - vpi->SrcReg[0].Index = temp_index; - vpi->SrcReg[0].Swizzle = SWIZZLE_XYZW; - - ++vpi; - - vpi->Opcode = OPCODE_END; -} - -static void pos_as_texcoord(struct gl_program *prog, int tex_id) -{ - struct prog_instruction *vpi; - GLuint tempregi = prog->NumTemporaries; - - prog->NumTemporaries++; - - for (vpi = prog->Instructions; vpi->Opcode != OPCODE_END; vpi++) { - if (vpi->DstReg.File == PROGRAM_OUTPUT && vpi->DstReg.Index == VERT_RESULT_HPOS) { - vpi->DstReg.File = PROGRAM_TEMPORARY; - vpi->DstReg.Index = tempregi; - } - } - - insert_wpos(prog, tempregi, tex_id); - - prog->OutputsWritten |= 1 << (VERT_RESULT_TEX0 + tex_id); -} - -/** - * The fogcoord attribute is special in that only the first component - * is relevant, and the remaining components are always fixed (when read - * from by the fragment program) to yield an X001 pattern. - * - * We need to enforce this either in the vertex program or in the fragment - * program, and this code chooses not to enforce it in the vertex program. - * This is slightly cheaper, as long as the fragment program does not use - * weird swizzles. - * - * And it seems that usually, weird swizzles are not used, so... - * - * See also the counterpart rewriting for fragment programs. - */ -static void fog_as_texcoord(struct gl_program *prog, int tex_id) -{ - struct prog_instruction *vpi; - - vpi = prog->Instructions; - while (vpi->Opcode != OPCODE_END) { - if (vpi->DstReg.File == PROGRAM_OUTPUT && vpi->DstReg.Index == VERT_RESULT_FOGC) { - vpi->DstReg.Index = VERT_RESULT_TEX0 + tex_id; - vpi->DstReg.WriteMask = WRITEMASK_X; - } - - ++vpi; - } - - prog->OutputsWritten &= ~(1 << VERT_RESULT_FOGC); - prog->OutputsWritten |= 1 << (VERT_RESULT_TEX0 + tex_id); -} - -static int translateABS(struct gl_program *prog, int pos) -{ - struct prog_instruction *inst; - - inst = &prog->Instructions[pos]; - - inst->Opcode = OPCODE_MAX; - inst->SrcReg[1] = inst->SrcReg[0]; - inst->SrcReg[1].Negate ^= NEGATE_XYZW; - - return 0; -} - -static int translateDP3(struct gl_program *prog, int pos) -{ - struct prog_instruction *inst; - - inst = &prog->Instructions[pos]; - - inst->Opcode = OPCODE_DP4; - inst->SrcReg[0].Swizzle = combine_swizzles4(inst->SrcReg[0].Swizzle, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - return 0; -} - -static int translateDPH(struct gl_program *prog, int pos) -{ - struct prog_instruction *inst; - - inst = &prog->Instructions[pos]; - - inst->Opcode = OPCODE_DP4; - inst->SrcReg[0].Swizzle = combine_swizzles4(inst->SrcReg[0].Swizzle, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE); - - return 0; -} - -static int translateFLR(struct gl_program *prog, int pos) -{ - struct prog_instruction *inst; - struct prog_dst_register dst; - int tmp_idx; - - tmp_idx = prog->NumTemporaries++; - - _mesa_insert_instructions(prog, pos + 1, 1); - - inst = &prog->Instructions[pos]; - dst = inst->DstReg; - - inst->Opcode = OPCODE_FRC; - inst->DstReg.File = PROGRAM_TEMPORARY; - inst->DstReg.Index = tmp_idx; - ++inst; - - inst->Opcode = OPCODE_ADD; - inst->DstReg = dst; - inst->SrcReg[0] = (inst-1)->SrcReg[0]; - inst->SrcReg[1].File = PROGRAM_TEMPORARY; - inst->SrcReg[1].Index = tmp_idx; - inst->SrcReg[1].Negate = NEGATE_XYZW; - - return 1; -} - -static int translateSUB(struct gl_program *prog, int pos) -{ - struct prog_instruction *inst; - - inst = &prog->Instructions[pos]; - - inst->Opcode = OPCODE_ADD; - inst->SrcReg[1].Negate ^= NEGATE_XYZW; - - return 0; -} - -static int translateSWZ(struct gl_program *prog, int pos) -{ - prog->Instructions[pos].Opcode = OPCODE_MOV; - - return 0; -} - -static int translateXPD(struct gl_program *prog, int pos) -{ - struct prog_instruction *inst; - int tmp_idx; - - tmp_idx = prog->NumTemporaries++; - - _mesa_insert_instructions(prog, pos + 1, 1); - - inst = &prog->Instructions[pos]; - - *(inst+1) = *inst; - - inst->Opcode = OPCODE_MUL; - inst->DstReg.File = PROGRAM_TEMPORARY; - inst->DstReg.Index = tmp_idx; - inst->SrcReg[0].Swizzle = combine_swizzles4(inst->SrcReg[0].Swizzle, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W); - inst->SrcReg[1].Swizzle = combine_swizzles4(inst->SrcReg[1].Swizzle, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W); - ++inst; - - inst->Opcode = OPCODE_MAD; - inst->SrcReg[0].Swizzle = combine_swizzles4(inst->SrcReg[0].Swizzle, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W); - inst->SrcReg[1].Swizzle = combine_swizzles4(inst->SrcReg[1].Swizzle, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W); - inst->SrcReg[1].Negate ^= NEGATE_XYZW; - inst->SrcReg[2].File = PROGRAM_TEMPORARY; - inst->SrcReg[2].Index = tmp_idx; - - return 1; -} - -static void translateInsts(struct gl_program *prog) -{ - struct prog_instruction *inst; - int i; - - for (i = 0; i < prog->NumInstructions; ++i) { - inst = &prog->Instructions[i]; - - switch (inst->Opcode) { - case OPCODE_ABS: - i += translateABS(prog, i); - break; - case OPCODE_DP3: - i += translateDP3(prog, i); - break; - case OPCODE_DPH: - i += translateDPH(prog, i); - break; - case OPCODE_FLR: - i += translateFLR(prog, i); - break; - case OPCODE_SUB: - i += translateSUB(prog, i); - break; - case OPCODE_SWZ: - i += translateSWZ(prog, i); - break; - case OPCODE_XPD: - i += translateXPD(prog, i); - break; - default: - break; - } - } -} - -#define ADD_OUTPUT(fp_attr, vp_result) \ - do { \ - if ((FpReads & (1 << (fp_attr))) && !(prog->OutputsWritten & (1 << (vp_result)))) { \ - OutputsAdded |= 1 << (vp_result); \ - count++; \ - } \ - } while (0) - -static void addArtificialOutputs(GLcontext *ctx, struct gl_program *prog) -{ - r300ContextPtr r300 = R300_CONTEXT(ctx); - GLuint OutputsAdded, FpReads; - int i, count; - - OutputsAdded = 0; - count = 0; - FpReads = r300->selected_fp->Base->InputsRead; - - ADD_OUTPUT(FRAG_ATTRIB_COL0, VERT_RESULT_COL0); - ADD_OUTPUT(FRAG_ATTRIB_COL1, VERT_RESULT_COL1); - - for (i = 0; i < 7; ++i) { - ADD_OUTPUT(FRAG_ATTRIB_TEX0 + i, VERT_RESULT_TEX0 + i); - } - - /* Some outputs may be artificially added, to match the inputs of the fragment program. - * Issue 16 of vertex program spec says that all vertex attributes that are unwritten by - * vertex program are undefined, so just use MOV [vertex_result], CONST[0] - */ - if (count > 0) { - struct prog_instruction *inst; - - _mesa_insert_instructions(prog, prog->NumInstructions - 1, count); - inst = &prog->Instructions[prog->NumInstructions - 1 - count]; - - for (i = 0; i < VERT_RESULT_MAX; ++i) { - if (OutputsAdded & (1 << i)) { - inst->Opcode = OPCODE_MOV; - - inst->DstReg.File = PROGRAM_OUTPUT; - inst->DstReg.Index = i; - inst->DstReg.WriteMask = WRITEMASK_XYZW; - inst->DstReg.CondMask = COND_TR; - - inst->SrcReg[0].File = PROGRAM_CONSTANT; - inst->SrcReg[0].Index = 0; - inst->SrcReg[0].Swizzle = SWIZZLE_XYZW; - - ++inst; - } - } - - prog->OutputsWritten |= OutputsAdded; - } -} - -#undef ADD_OUTPUT - -static void nqssadceInit(struct nqssadce_state* s) -{ - r300ContextPtr r300 = R300_CONTEXT(s->Ctx); - GLuint fp_reads; - - fp_reads = r300->selected_fp->Base->InputsRead; - { - if (fp_reads & FRAG_BIT_COL0) { - s->Outputs[VERT_RESULT_COL0].Sourced = WRITEMASK_XYZW; - s->Outputs[VERT_RESULT_BFC0].Sourced = WRITEMASK_XYZW; - } - - if (fp_reads & FRAG_BIT_COL1) { - s->Outputs[VERT_RESULT_COL1].Sourced = WRITEMASK_XYZW; - s->Outputs[VERT_RESULT_BFC1].Sourced = WRITEMASK_XYZW; - } - } - - { - int i; - for (i = 0; i < 8; ++i) { - if (fp_reads & FRAG_BIT_TEX(i)) { - s->Outputs[VERT_RESULT_TEX0 + i].Sourced = WRITEMASK_XYZW; - } - } - } - - s->Outputs[VERT_RESULT_HPOS].Sourced = WRITEMASK_XYZW; - if (s->Program->OutputsWritten & (1 << VERT_RESULT_PSIZ)) - s->Outputs[VERT_RESULT_PSIZ].Sourced = WRITEMASK_X; -} - -static GLboolean swizzleIsNative(GLuint opcode, struct prog_src_register reg) -{ - (void) opcode; - (void) reg; - - return GL_TRUE; -} static struct r300_vertex_program *build_program(GLcontext *ctx, struct r300_vertex_program_key *wanted_key, const struct gl_vertex_program *mesa_vp) { - r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_vertex_program *vp; - struct gl_program *prog; + struct r300_vertex_program_compiler compiler; vp = _mesa_calloc(sizeof(*vp)); vp->Base = (struct gl_vertex_program *) _mesa_clone_program(ctx, &mesa_vp->Base); _mesa_memcpy(&vp->key, wanted_key, sizeof(vp->key)); - prog = &vp->Base->Base; + rc_init(&compiler.Base); + compiler.Base.Debug = (RADEON_DEBUG & DEBUG_VERTS) ? GL_TRUE : GL_FALSE; + + compiler.code = &vp->code; + compiler.RequiredOutputs = compute_required_outputs(vp->Base, vp->key.FpReads); + compiler.SetHwInputOutput = &t_inputs_outputs; - if (RADEON_DEBUG & DEBUG_VERTS) { + if (compiler.Base.Debug) { fprintf(stderr, "Initial vertex program:\n"); - _mesa_print_program(prog); + _mesa_print_program(&vp->Base->Base); fflush(stdout); } - if (vp->Base->IsPositionInvariant) { + if (mesa_vp->IsPositionInvariant) { _mesa_insert_mvp_code(ctx, vp->Base); } - if (r300->selected_fp->wpos_attr != FRAG_ATTRIB_MAX) { - pos_as_texcoord(&vp->Base->Base, r300->selected_fp->wpos_attr - FRAG_ATTRIB_TEX0); - } - - if (r300->selected_fp->fog_attr != FRAG_ATTRIB_MAX) { - fog_as_texcoord(&vp->Base->Base, r300->selected_fp->fog_attr - FRAG_ATTRIB_TEX0); - } + rc_mesa_to_rc_program(&compiler.Base, &vp->Base->Base); - addArtificialOutputs(ctx, prog); + rc_move_output(&compiler.Base, VERT_RESULT_PSIZ, VERT_RESULT_PSIZ, WRITEMASK_X); - translateInsts(prog); - - if (RADEON_DEBUG & DEBUG_VERTS) { - fprintf(stderr, "Vertex program after native rewrite:\n"); - _mesa_print_program(prog); - fflush(stdout); + if (vp->key.WPosAttr != FRAG_ATTRIB_MAX) { + rc_copy_output(&compiler.Base, + VERT_RESULT_HPOS, + vp->key.WPosAttr - FRAG_ATTRIB_TEX0 + VERT_RESULT_TEX0); } - { - struct radeon_nqssadce_descr nqssadce = { - .Init = &nqssadceInit, - .IsNativeSwizzle = &swizzleIsNative, - .BuildSwizzle = NULL - }; - radeonNqssaDce(ctx, prog, &nqssadce); - - /* We need this step for reusing temporary registers */ - _mesa_optimize_program(ctx, prog); - - if (RADEON_DEBUG & DEBUG_VERTS) { - fprintf(stderr, "Vertex program after NQSSADCE:\n"); - _mesa_print_program(prog); - fflush(stdout); - } + if (vp->key.FogAttr != FRAG_ATTRIB_MAX) { + rc_move_output(&compiler.Base, + VERT_RESULT_FOGC, + vp->key.FogAttr - FRAG_ATTRIB_TEX0 + VERT_RESULT_TEX0, WRITEMASK_X); } - assert(prog->NumInstructions); - { - struct prog_instruction *inst; - int max, i, tmp; - - inst = prog->Instructions; - max = -1; - while (inst->Opcode != OPCODE_END) { - tmp = _mesa_num_inst_src_regs(inst->Opcode); - for (i = 0; i < tmp; ++i) { - if (inst->SrcReg[i].File == PROGRAM_TEMPORARY) { - if ((int) inst->SrcReg[i].Index > max) { - max = inst->SrcReg[i].Index; - } - } - } + r3xx_compile_vertex_program(&compiler); + vp->error = compiler.Base.Error; - if (_mesa_num_inst_dst_regs(inst->Opcode)) { - if (inst->DstReg.File == PROGRAM_TEMPORARY) { - if ((int) inst->DstReg.Index > max) { - max = inst->DstReg.Index; - } - } - } - ++inst; - } + vp->Base->Base.InputsRead = vp->code.InputsRead; + vp->Base->Base.OutputsWritten = vp->code.OutputsWritten; - /* We actually want highest index of used temporary register, - * not the number of temporaries used. - * These values aren't always the same. - */ - vp->num_temporaries = max + 1; - } + rc_destroy(&compiler.Base); return vp; } -struct r300_vertex_program * r300SelectVertexShader(GLcontext *ctx) +struct r300_vertex_program * r300SelectAndTranslateVertexShader(GLcontext *ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); struct r300_vertex_program_key wanted_key = { 0 }; @@ -1639,7 +267,7 @@ struct r300_vertex_program * r300SelectVertexShader(GLcontext *ctx) struct r300_vertex_program *vp; vpc = (struct r300_vertex_program_cont *)ctx->VertexProgram._Current; - wanted_key.FpReads = r300->selected_fp->Base->InputsRead; + wanted_key.FpReads = r300->selected_fp->InputsRead; wanted_key.FogAttr = r300->selected_fp->fog_attr; wanted_key.WPosAttr = r300->selected_fp->wpos_attr; @@ -1664,7 +292,7 @@ struct r300_vertex_program * r300SelectVertexShader(GLcontext *ctx) if(_nc>_p->vpu.count)_p->vpu.count=_nc; \ } while(0) -static void r300EmitVertexProgram(r300ContextPtr r300, int dest, struct r300_vertex_shader_hw_code *code) +static void r300EmitVertexProgram(r300ContextPtr r300, int dest, struct r300_vertex_program_code *code) { int i; @@ -1708,15 +336,15 @@ void r300SetupVertexProgram(r300ContextPtr rmesa) ((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0; R300_STATECHANGE(rmesa, vpp); - param_count = r300VertexProgUpdateParams(ctx, prog->Base, (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]); + param_count = r300VertexProgUpdateParams(ctx, prog, (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]); bump_vpu_count(rmesa->hw.vpp.cmd, param_count); param_count /= 4; - r300EmitVertexProgram(rmesa, R300_PVS_CODE_START, &(prog->hw_code)); - inst_count = (prog->hw_code.length / 4) - 1; + r300EmitVertexProgram(rmesa, R300_PVS_CODE_START, &(prog->code)); + inst_count = (prog->code.length / 4) - 1; - r300VapCntl(rmesa, _mesa_bitcount(prog->Base->Base.InputsRead), - _mesa_bitcount(prog->Base->Base.OutputsWritten), prog->num_temporaries); + r300VapCntl(rmesa, _mesa_bitcount(prog->code.InputsRead), + _mesa_bitcount(prog->code.OutputsWritten), prog->code.num_temporaries); R300_STATECHANGE(rmesa, pvs); rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] = (0 << R300_PVS_FIRST_INST_SHIFT) | (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) | diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.h b/src/mesa/drivers/dri/r300/r300_vertprog.h index 2dab11c337..ccec896be4 100644 --- a/src/mesa/drivers/dri/r300/r300_vertprog.h +++ b/src/mesa/drivers/dri/r300/r300_vertprog.h @@ -3,39 +3,9 @@ #include "r300_reg.h" -#define PVS_OP_DST_OPERAND(opcode, math_inst, macro_inst, reg_index, reg_writemask, reg_class) \ - (((opcode & PVS_DST_OPCODE_MASK) << PVS_DST_OPCODE_SHIFT) \ - | ((math_inst & PVS_DST_MATH_INST_MASK) << PVS_DST_MATH_INST_SHIFT) \ - | ((macro_inst & PVS_DST_MACRO_INST_MASK) << PVS_DST_MACRO_INST_SHIFT) \ - | ((reg_index & PVS_DST_OFFSET_MASK) << PVS_DST_OFFSET_SHIFT) \ - | ((reg_writemask & 0xf) << PVS_DST_WE_X_SHIFT) /* X Y Z W */ \ - | ((reg_class & PVS_DST_REG_TYPE_MASK) << PVS_DST_REG_TYPE_SHIFT)) - -#define PVS_SRC_OPERAND(in_reg_index, comp_x, comp_y, comp_z, comp_w, reg_class, negate) \ - (((in_reg_index & PVS_SRC_OFFSET_MASK) << PVS_SRC_OFFSET_SHIFT) \ - | ((comp_x & PVS_SRC_SWIZZLE_X_MASK) << PVS_SRC_SWIZZLE_X_SHIFT) \ - | ((comp_y & PVS_SRC_SWIZZLE_Y_MASK) << PVS_SRC_SWIZZLE_Y_SHIFT) \ - | ((comp_z & PVS_SRC_SWIZZLE_Z_MASK) << PVS_SRC_SWIZZLE_Z_SHIFT) \ - | ((comp_w & PVS_SRC_SWIZZLE_W_MASK) << PVS_SRC_SWIZZLE_W_SHIFT) \ - | ((negate & 0xf) << PVS_SRC_MODIFIER_X_SHIFT) /* X Y Z W */ \ - | ((reg_class & PVS_SRC_REG_TYPE_MASK) << PVS_SRC_REG_TYPE_SHIFT)) - -#if 1 - -#define VSF_FLAG_X 1 -#define VSF_FLAG_Y 2 -#define VSF_FLAG_Z 4 -#define VSF_FLAG_W 8 -#define VSF_FLAG_XYZ (VSF_FLAG_X | VSF_FLAG_Y | VSF_FLAG_Z) -#define VSF_FLAG_ALL 0xf -#define VSF_FLAG_NONE 0 - -#endif void r300SetupVertexProgram(r300ContextPtr rmesa); -struct r300_vertex_program * r300SelectVertexShader(GLcontext *ctx); - -void r300TranslateVertexShader(struct r300_vertex_program *vp); +struct r300_vertex_program * r300SelectAndTranslateVertexShader(GLcontext *ctx); #endif diff --git a/src/mesa/drivers/dri/r300/radeon_bo_legacy.c b/src/mesa/drivers/dri/r300/radeon_bo_legacy.c new file mode 120000 index 0000000000..79ad050e6b --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_bo_legacy.c @@ -0,0 +1 @@ +../radeon/radeon_bo_legacy.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_bo_legacy.h b/src/mesa/drivers/dri/r300/radeon_bo_legacy.h new file mode 120000 index 0000000000..83b0f7ffab --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_bo_legacy.h @@ -0,0 +1 @@ +../radeon/radeon_bo_legacy.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_bocs_wrapper.h b/src/mesa/drivers/dri/r300/radeon_bocs_wrapper.h new file mode 120000 index 0000000000..ca894b2443 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_bocs_wrapper.h @@ -0,0 +1 @@ +../radeon/radeon_bocs_wrapper.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_buffer_objects.c b/src/mesa/drivers/dri/r300/radeon_buffer_objects.c new file mode 120000 index 0000000000..f6a5f66470 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_buffer_objects.c @@ -0,0 +1 @@ +../radeon/radeon_buffer_objects.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_buffer_objects.h b/src/mesa/drivers/dri/r300/radeon_buffer_objects.h new file mode 120000 index 0000000000..2f134fd17b --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_buffer_objects.h @@ -0,0 +1 @@ +../radeon/radeon_buffer_objects.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_chipset.h b/src/mesa/drivers/dri/r300/radeon_chipset.h new file mode 120000 index 0000000000..eba99001ff --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_chipset.h @@ -0,0 +1 @@ +../radeon/radeon_chipset.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_cmdbuf.h b/src/mesa/drivers/dri/r300/radeon_cmdbuf.h new file mode 120000 index 0000000000..a799e1dc6d --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_cmdbuf.h @@ -0,0 +1 @@ +../radeon/radeon_cmdbuf.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_common.c b/src/mesa/drivers/dri/r300/radeon_common.c new file mode 120000 index 0000000000..67b19ba940 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_common.c @@ -0,0 +1 @@ +../radeon/radeon_common.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_common.h b/src/mesa/drivers/dri/r300/radeon_common.h new file mode 120000 index 0000000000..5bcb696a9f --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_common.h @@ -0,0 +1 @@ +../radeon/radeon_common.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_common_context.c b/src/mesa/drivers/dri/r300/radeon_common_context.c new file mode 120000 index 0000000000..86800f3819 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_common_context.c @@ -0,0 +1 @@ +../radeon/radeon_common_context.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_common_context.h b/src/mesa/drivers/dri/r300/radeon_common_context.h new file mode 120000 index 0000000000..4d66312550 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_common_context.h @@ -0,0 +1 @@ +../radeon/radeon_common_context.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_cs_legacy.c b/src/mesa/drivers/dri/r300/radeon_cs_legacy.c new file mode 120000 index 0000000000..006720f8a4 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_cs_legacy.c @@ -0,0 +1 @@ +../radeon/radeon_cs_legacy.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_cs_legacy.h b/src/mesa/drivers/dri/r300/radeon_cs_legacy.h new file mode 120000 index 0000000000..a5f95e0a3d --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_cs_legacy.h @@ -0,0 +1 @@ +../radeon/radeon_cs_legacy.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_cs_space_drm.c b/src/mesa/drivers/dri/r300/radeon_cs_space_drm.c new file mode 120000 index 0000000000..c248ea7d1a --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_cs_space_drm.c @@ -0,0 +1 @@ +../radeon/radeon_cs_space_drm.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_dma.c b/src/mesa/drivers/dri/r300/radeon_dma.c new file mode 120000 index 0000000000..43be000625 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_dma.c @@ -0,0 +1 @@ +../radeon/radeon_dma.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_dma.h b/src/mesa/drivers/dri/r300/radeon_dma.h new file mode 120000 index 0000000000..82e50634e3 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_dma.h @@ -0,0 +1 @@ +../radeon/radeon_dma.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_fbo.c b/src/mesa/drivers/dri/r300/radeon_fbo.c new file mode 120000 index 0000000000..0d738d8d78 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_fbo.c @@ -0,0 +1 @@ +../radeon/radeon_fbo.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_lock.c b/src/mesa/drivers/dri/r300/radeon_lock.c new file mode 120000 index 0000000000..af4108a8e3 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_lock.c @@ -0,0 +1 @@ +../radeon/radeon_lock.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_lock.h b/src/mesa/drivers/dri/r300/radeon_lock.h new file mode 120000 index 0000000000..64bdf94ee7 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_lock.h @@ -0,0 +1 @@ +../radeon/radeon_lock.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_mipmap_tree.c b/src/mesa/drivers/dri/r300/radeon_mipmap_tree.c new file mode 120000 index 0000000000..31c0cfbe94 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_mipmap_tree.c @@ -0,0 +1 @@ +../radeon/radeon_mipmap_tree.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_mipmap_tree.h b/src/mesa/drivers/dri/r300/radeon_mipmap_tree.h new file mode 120000 index 0000000000..254d50cf8c --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_mipmap_tree.h @@ -0,0 +1 @@ +../radeon/radeon_mipmap_tree.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_program.c b/src/mesa/drivers/dri/r300/radeon_program.c deleted file mode 100644 index da5e7aefce..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_program.c +++ /dev/null @@ -1,128 +0,0 @@ -/* - * Copyright (C) 2008 Nicolai Haehnle. - * - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining - * a copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sublicense, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial - * portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, - * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. - * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE - * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION - * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION - * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#include "radeon_program.h" - -#include "shader/prog_print.h" - - -/** - * Transform the given clause in the following way: - * 1. Replace it with an empty clause - * 2. For every instruction in the original clause, try the given - * transformations in order. - * 3. If one of the transformations returns GL_TRUE, assume that it - * has emitted the appropriate instruction(s) into the new clause; - * otherwise, copy the instruction verbatim. - * - * \note The transformation is currently not recursive; in other words, - * instructions emitted by transformations are not transformed. - * - * \note The transform is called 'local' because it can only look at - * one instruction at a time. - */ -void radeonLocalTransform( - GLcontext *Ctx, - struct gl_program *program, - int num_transformations, - struct radeon_program_transformation* transformations) -{ - struct radeon_transform_context ctx; - int ip; - - ctx.Ctx = Ctx; - ctx.Program = program; - ctx.OldInstructions = program->Instructions; - ctx.OldNumInstructions = program->NumInstructions; - - program->Instructions = 0; - program->NumInstructions = 0; - - for(ip = 0; ip < ctx.OldNumInstructions; ++ip) { - struct prog_instruction *instr = ctx.OldInstructions + ip; - int i; - - for(i = 0; i < num_transformations; ++i) { - struct radeon_program_transformation* t = transformations + i; - - if (t->function(&ctx, instr, t->userData)) - break; - } - - if (i >= num_transformations) { - struct prog_instruction* dest = radeonAppendInstructions(program, 1); - _mesa_copy_instructions(dest, instr, 1); - } - } - - _mesa_free_instructions(ctx.OldInstructions, ctx.OldNumInstructions); -} - - -static void scan_instructions(GLboolean* used, const struct prog_instruction* insts, GLuint count) -{ - GLuint i; - for (i = 0; i < count; i++) { - const struct prog_instruction *inst = insts + i; - const GLuint n = _mesa_num_inst_src_regs(inst->Opcode); - GLuint k; - - for (k = 0; k < n; k++) { - if (inst->SrcReg[k].File == PROGRAM_TEMPORARY) - used[inst->SrcReg[k].Index] = GL_TRUE; - } - } -} - -GLint radeonFindFreeTemporary(struct radeon_transform_context *t) -{ - GLboolean used[MAX_PROGRAM_TEMPS]; - GLuint i; - - _mesa_memset(used, 0, sizeof(used)); - scan_instructions(used, t->Program->Instructions, t->Program->NumInstructions); - scan_instructions(used, t->OldInstructions, t->OldNumInstructions); - - for (i = 0; i < MAX_PROGRAM_TEMPS; i++) { - if (!used[i]) - return i; - } - - return -1; -} - - -/** - * Append the given number of instructions to the program and return a - * pointer to the first new instruction. - */ -struct prog_instruction *radeonAppendInstructions(struct gl_program *program, int count) -{ - int oldnum = program->NumInstructions; - _mesa_insert_instructions(program, oldnum, count); - return program->Instructions + oldnum; -} diff --git a/src/mesa/drivers/dri/r300/radeon_screen.c b/src/mesa/drivers/dri/r300/radeon_screen.c new file mode 120000 index 0000000000..86161118dd --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_screen.c @@ -0,0 +1 @@ +../radeon/radeon_screen.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_screen.h b/src/mesa/drivers/dri/r300/radeon_screen.h new file mode 120000 index 0000000000..23bb6bd459 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_screen.h @@ -0,0 +1 @@ +../radeon/radeon_screen.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_span.c b/src/mesa/drivers/dri/r300/radeon_span.c new file mode 120000 index 0000000000..232868c4c9 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_span.c @@ -0,0 +1 @@ +../radeon/radeon_span.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_span.h b/src/mesa/drivers/dri/r300/radeon_span.h new file mode 120000 index 0000000000..f9d634508c --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_span.h @@ -0,0 +1 @@ +../radeon/radeon_span.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_texture.c b/src/mesa/drivers/dri/r300/radeon_texture.c new file mode 120000 index 0000000000..a822710915 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_texture.c @@ -0,0 +1 @@ +../radeon/radeon_texture.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/radeon_texture.h b/src/mesa/drivers/dri/r300/radeon_texture.h new file mode 120000 index 0000000000..17fac3d5ea --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_texture.h @@ -0,0 +1 @@ +../radeon/radeon_texture.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/server/radeon.h b/src/mesa/drivers/dri/r300/server/radeon.h new file mode 120000 index 0000000000..81274a54f1 --- /dev/null +++ b/src/mesa/drivers/dri/r300/server/radeon.h @@ -0,0 +1 @@ +../../radeon/server/radeon.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/server/radeon_dri.c b/src/mesa/drivers/dri/r300/server/radeon_dri.c new file mode 120000 index 0000000000..d05847d650 --- /dev/null +++ b/src/mesa/drivers/dri/r300/server/radeon_dri.c @@ -0,0 +1 @@ +../../radeon/server/radeon_dri.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/server/radeon_dri.h b/src/mesa/drivers/dri/r300/server/radeon_dri.h new file mode 120000 index 0000000000..27c591d3c9 --- /dev/null +++ b/src/mesa/drivers/dri/r300/server/radeon_dri.h @@ -0,0 +1 @@ +../../radeon/server/radeon_dri.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/server/radeon_egl.c b/src/mesa/drivers/dri/r300/server/radeon_egl.c new file mode 120000 index 0000000000..d7735a7643 --- /dev/null +++ b/src/mesa/drivers/dri/r300/server/radeon_egl.c @@ -0,0 +1 @@ +../../radeon/server/radeon_egl.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/server/radeon_macros.h b/src/mesa/drivers/dri/r300/server/radeon_macros.h new file mode 120000 index 0000000000..c56cd735b8 --- /dev/null +++ b/src/mesa/drivers/dri/r300/server/radeon_macros.h @@ -0,0 +1 @@ +../../radeon/server/radeon_macros.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r300/server/radeon_reg.h b/src/mesa/drivers/dri/r300/server/radeon_reg.h new file mode 120000 index 0000000000..e2349dcb68 --- /dev/null +++ b/src/mesa/drivers/dri/r300/server/radeon_reg.h @@ -0,0 +1 @@ +../../radeon/server/radeon_reg.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/Makefile b/src/mesa/drivers/dri/r600/Makefile index 2f225f7cb5..5bdc1afbf0 100644 --- a/src/mesa/drivers/dri/r600/Makefile +++ b/src/mesa/drivers/dri/r600/Makefile @@ -68,54 +68,10 @@ DRIVER_DEFINES = -DCOMPILE_R600 -DR200_MERGED=0 \ # -DRADEON_BO_TRACK \ -Wall -SYMLINKS = \ - server/radeon_dri.c \ - server/radeon_dri.h \ - server/radeon.h \ - server/radeon_macros.h \ - server/radeon_reg.h \ - server/radeon_egl.c - -COMMON_SYMLINKS = \ - radeon_chipset.h \ - radeon_screen.c \ - radeon_screen.h \ - radeon_span.h \ - radeon_span.c \ - radeon_bo_legacy.c \ - radeon_cs_legacy.c \ - radeon_bo_legacy.h \ - radeon_cs_legacy.h \ - radeon_bocs_wrapper.h \ - radeon_lock.c \ - radeon_lock.h \ - radeon_common.c \ - radeon_common.h \ - radeon_common_context.c \ - radeon_common_context.h \ - radeon_cmdbuf.h \ - radeon_dma.c \ - radeon_dma.h \ - radeon_mipmap_tree.c \ - radeon_mipmap_tree.h \ - radeon_texture.c \ - radeon_texture.h \ - radeon_fbo.c \ - $(CS_SOURCES) - DRI_LIB_DEPS += $(RADEON_LDFLAGS) ##### TARGETS ##### include ../Makefile.template -server: - mkdir -p server - -$(SYMLINKS): server - @[ -e $@ ] || ln -sf ../../radeon/$@ server/ - -$(COMMON_SYMLINKS): - @[ -e $@ ] || ln -sf ../radeon/$@ ./ - -symlinks: $(SYMLINKS) $(COMMON_SYMLINKS) +symlinks: diff --git a/src/mesa/drivers/dri/r600/r600_cmdbuf.c b/src/mesa/drivers/dri/r600/r600_cmdbuf.c index 74fec02584..dc2fb0144a 100644 --- a/src/mesa/drivers/dri/r600/r600_cmdbuf.c +++ b/src/mesa/drivers/dri/r600/r600_cmdbuf.c @@ -78,8 +78,7 @@ int r600_cs_write_reloc(struct radeon_cs *cs, struct radeon_bo *bo, uint32_t read_domain, uint32_t write_domain, - uint32_t flags, - offset_modifiers* poffset_mod) + uint32_t flags) { struct r600_cs_reloc_legacy *relocs; int i; @@ -135,10 +134,6 @@ int r600_cs_write_reloc(struct radeon_cs *cs, cs->section_ndw += 2; cs->section_cdw += 2; - relocs[i].offset_mod.shift = poffset_mod->shift; - relocs[i].offset_mod.shiftbits = poffset_mod->shiftbits; - relocs[i].offset_mod.mask = poffset_mod->mask; - return 0; } } @@ -160,9 +155,6 @@ int r600_cs_write_reloc(struct radeon_cs *cs, { return -ENOMEM; } - relocs[cs->crelocs].offset_mod.shift = poffset_mod->shift; - relocs[cs->crelocs].offset_mod.shiftbits = poffset_mod->shiftbits; - relocs[cs->crelocs].offset_mod.mask = poffset_mod->mask; relocs[cs->crelocs].indices[0] = cs->cdw - 1; relocs[cs->crelocs].reloc_indices[0] = cs->section_cdw; @@ -255,65 +247,44 @@ static int r600_cs_process_relocs(struct radeon_cs *cs, csm = (struct r600_cs_manager_legacy*)cs->csm; relocs = (struct r600_cs_reloc_legacy *)cs->relocs; restart: - for (i = 0; i < cs->crelocs; i++) - { - for (j = 0; j < relocs[i].cindices; j++) - { + for (i = 0; i < cs->crelocs; i++) { uint32_t soffset, eoffset, asicoffset; r = radeon_bo_legacy_validate(relocs[i].base.bo, - &soffset, &eoffset); - if (r == -EAGAIN) - { - goto restart; + &soffset, &eoffset); + if (r == -EAGAIN) { + goto restart; } - if (r) - { - fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n", - relocs[i].base.bo, soffset, eoffset); - return r; + if (r) { + fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n", + relocs[i].base.bo, soffset, eoffset); + return r; } asicoffset = soffset; - if (asicoffset >= eoffset) - { - /* radeon_bo_debug(relocs[i].base.bo, 12); */ - fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n", - relocs[i].base.bo, soffset, eoffset); - fprintf(stderr, "above end: %p 0x%08X 0x%08X\n", - relocs[i].base.bo, - cs->packets[relocs[i].indices[j]], - eoffset); - exit(0); - return -EINVAL; - } - /* apply offset operator */ - switch (relocs[i].offset_mod.shift) - { - case NO_SHIFT: - asicoffset = asicoffset & relocs[i].offset_mod.mask; - break; - case LEFT_SHIFT: - asicoffset = (asicoffset << relocs[i].offset_mod.shiftbits) & relocs[i].offset_mod.mask; - break; - case RIGHT_SHIFT: - asicoffset = (asicoffset >> relocs[i].offset_mod.shiftbits) & relocs[i].offset_mod.mask; - break; - default: - break; - }; - - /* pkt3 nop header in ib chunk */ - cs->packets[relocs[i].reloc_indices[j]] = 0xC0001000; - - /* reloc index in ib chunk */ - cs->packets[relocs[i].reloc_indices[j] + 1] = offset_dw; - - /* asic offset in reloc chunk */ /* see alex drm r600_nomm_relocate */ - reloc_chunk[offset_dw] = asicoffset; - reloc_chunk[offset_dw + 3] = 0; - - offset_dw += 4; - } + + for (j = 0; j < relocs[i].cindices; j++) { + if (asicoffset >= eoffset) { + /* radeon_bo_debug(relocs[i].base.bo, 12); */ + fprintf(stderr, "validated %p [0x%08X, 0x%08X]\n", + relocs[i].base.bo, soffset, eoffset); + fprintf(stderr, "above end: %p 0x%08X 0x%08X\n", + relocs[i].base.bo, + cs->packets[relocs[i].indices[j]], + eoffset); + exit(0); + return -EINVAL; + } + /* pkt3 nop header in ib chunk */ + cs->packets[relocs[i].reloc_indices[j]] = 0xC0001000; + /* reloc index in ib chunk */ + cs->packets[relocs[i].reloc_indices[j] + 1] = offset_dw; + } + + /* asic offset in reloc chunk */ /* see alex drm r600_nomm_relocate */ + reloc_chunk[offset_dw] = asicoffset; + reloc_chunk[offset_dw + 3] = 0; + + offset_dw += 4; } *length_dw_reloc_chunk = offset_dw; @@ -351,10 +322,7 @@ static int r600_cs_emit(struct radeon_cs *cs) struct r600_cs_manager_legacy *csm = (struct r600_cs_manager_legacy*)cs->csm; struct drm_radeon_cs cs_cmd; struct drm_radeon_cs_chunk cs_chunk[2]; - drm_radeon_cmd_buffer_t cmd; - /* drm_r300_cmd_header_t age; */ uint32_t length_dw_reloc_chunk; - uint64_t ull; uint64_t chunk_ptrs[2]; uint32_t reloc_chunk[128]; int r; @@ -363,43 +331,13 @@ static int r600_cs_emit(struct radeon_cs *cs) /* TODO : put chip level things here if need. */ /* csm->ctx->vtbl.emit_cs_header(cs, csm->ctx); */ - BATCH_LOCALS(csm->ctx); - drm_radeon_getparam_t gp; - uint32_t current_scratchx_age; - - gp.param = RADEON_PARAM_LAST_CLEAR; - gp.value = (int *)¤t_scratchx_age; - r = drmCommandWriteRead(cs->csm->fd, - DRM_RADEON_GETPARAM, - &gp, - sizeof(gp)); - if (r) - { - fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, r); - exit(1); - } - - csm->pending_age = 0; csm->pending_count = 1; - current_scratchx_age++; - csm->pending_age = current_scratchx_age; - - BEGIN_BATCH_NO_AUTOSTATE(3); - R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1)); - R600_OUT_BATCH((SCRATCH_REG2 - R600_SET_CONFIG_REG_OFFSET) >> 2); - R600_OUT_BATCH(current_scratchx_age); - END_BATCH(); - COMMIT_BATCH(); - - //TODO ioctl to get back cs id assigned in drm - //csm->pending_age = cs_id_back; - r = r600_cs_process_relocs(cs, &(reloc_chunk[0]), &length_dw_reloc_chunk); if (r) { return 0; } - + /* raw ib chunk */ cs_chunk[0].chunk_id = RADEON_CHUNK_ID_IB; cs_chunk[0].length_dw = cs->cdw; @@ -429,6 +367,8 @@ static int r600_cs_emit(struct radeon_cs *cs) return r; } + csm->pending_age = cs_cmd.cs_id; + r600_cs_set_age(cs); cs->csm->read_used = 0; @@ -514,8 +454,12 @@ struct radeon_cs_manager * r600_radeon_cs_manager_legacy_ctor(struct radeon_cont void r600InitCmdBuf(context_t *r600) /* from rcommonInitCmdBuf */ { radeonContextPtr rmesa = &r600->radeon; - GLuint size; + rmesa->hw.max_state_size = 4000; /* rough estimate */ + + rmesa->hw.all_dirty = GL_TRUE; + rmesa->hw.is_dirty = GL_TRUE; + /* Initialize command buffer */ size = 256 * driQueryOptioni(&rmesa->optionCache, "command_buffer_size"); diff --git a/src/mesa/drivers/dri/r600/r600_cmdbuf.h b/src/mesa/drivers/dri/r600/r600_cmdbuf.h index bd1ed7fdff..5df0cf1ab6 100644 --- a/src/mesa/drivers/dri/r600/r600_cmdbuf.h +++ b/src/mesa/drivers/dri/r600/r600_cmdbuf.h @@ -132,15 +132,13 @@ struct r600_cs_reloc_legacy { uint32_t cindices; uint32_t *indices; uint32_t *reloc_indices; - struct offset_modifiers offset_mod; }; extern int r600_cs_write_reloc(struct radeon_cs *cs, struct radeon_bo *bo, uint32_t read_domain, uint32_t write_domain, - uint32_t flags, - offset_modifiers* poffset_mod); + uint32_t flags); static inline void r600_cs_write_dword(struct radeon_cs *cs, uint32_t dword) { @@ -171,7 +169,7 @@ struct radeon_cs_manager * r600_radeon_cs_manager_legacy_ctor(struct radeon_cont /** * Write a relocated dword to the command buffer. */ -#define R600_OUT_BATCH_RELOC(data, bo, offset, rd, wd, flags, offset_mod) \ +#define R600_OUT_BATCH_RELOC(data, bo, offset, rd, wd, flags) \ do { \ if (0 && offset) { \ fprintf(stderr, "(%s:%s:%d) offset : %d\n", \ @@ -179,7 +177,7 @@ struct radeon_cs_manager * r600_radeon_cs_manager_legacy_ctor(struct radeon_cont } \ r600_cs_write_dword(b_l_rmesa->cmdbuf.cs, offset); \ r600_cs_write_reloc(b_l_rmesa->cmdbuf.cs, \ - bo, rd, wd, flags, offset_mod); \ + bo, rd, wd, flags); \ } while(0) /* R600/R700 */ diff --git a/src/mesa/drivers/dri/r600/r600_context.c b/src/mesa/drivers/dri/r600/r600_context.c index 78bad8726b..7009374b0c 100644 --- a/src/mesa/drivers/dri/r600/r600_context.c +++ b/src/mesa/drivers/dri/r600/r600_context.c @@ -57,7 +57,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "drivers/common/driverfuncs.h" #include "r600_context.h" -#include "radeon_context.h" +#include "radeon_common_context.h" #include "radeon_span.h" #include "r600_cmdbuf.h" #include "r600_emit.h" @@ -218,7 +218,6 @@ GLboolean r600CreateContext(const __GLcontextModes * glVisual, struct dd_function_table functions; context_t *r600; GLcontext *ctx; - int tcl_mode; assert(glVisual); assert(driContextPriv); @@ -377,23 +376,12 @@ GLboolean r600CreateContext(const __GLcontextModes * glVisual, TNL_CONTEXT(ctx)->Driver.RunPipeline = r600RunPipeline; - tcl_mode = driQueryOptioni(&r600->radeon.optionCache, "tcl_mode"); if (driQueryOptionb(&r600->radeon.optionCache, "no_rast")) { fprintf(stderr, "disabling 3D acceleration\n"); #if R200_MERGED FALLBACK(&r600->radeon, RADEON_FALLBACK_DISABLE, 1); #endif } - if (tcl_mode == DRI_CONF_TCL_SW || - !(r600->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) { - if (r600->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { - r600->radeon.radeonScreen->chip_flags &= - ~RADEON_CHIPSET_TCL; - fprintf(stderr, "Disabling HW TCL support\n"); - } - TCL_FALLBACK(r600->radeon.glCtx, - RADEON_TCL_FALLBACK_TCL_DISABLE, 1); - } return GL_TRUE; } @@ -407,6 +395,8 @@ r600DestroyContext (__DRIcontextPrivate * driContextPriv) if (context) FREE(context->hw.pStateList); + + radeonDestroyContext(driContextPriv); } diff --git a/src/mesa/drivers/dri/r600/r600_context.h b/src/mesa/drivers/dri/r600/r600_context.h index bcb33e1386..30ddce682c 100644 --- a/src/mesa/drivers/dri/r600/r600_context.h +++ b/src/mesa/drivers/dri/r600/r600_context.h @@ -128,13 +128,6 @@ enum RIGHT_SHIFT = 2, }; -typedef struct offset_modifiers -{ - GLuint shift; - GLuint shiftbits; - GLuint mask; -} offset_modifiers; - /** * \brief R600 context structure. */ @@ -165,6 +158,17 @@ extern GLboolean r700InitChipObject(context_t *context); extern GLboolean r700SendContextStates(context_t *context); extern GLboolean r700SendViewportState(context_t *context, int id); extern GLboolean r700SendRenderTargetState(context_t *context, int id); +extern GLboolean r700SendTextureState(context_t *context); +extern GLboolean r700SendDepthTargetState(context_t *context); +extern GLboolean r700SendUCPState(context_t *context); +extern GLboolean r700SendFSState(context_t *context); +extern void r700EmitState(GLcontext * ctx); + +extern GLboolean r700SyncSurf(context_t *context, + struct radeon_bo *pbo, + uint32_t read_domain, + uint32_t write_domain, + uint32_t sync_type); extern int r700SetupStreams(GLcontext * ctx); extern void r700SetupVTXConstants(GLcontext * ctx, diff --git a/src/mesa/drivers/dri/r600/r600_emit.c b/src/mesa/drivers/dri/r600/r600_emit.c index 685f7fe473..b695ed9583 100644 --- a/src/mesa/drivers/dri/r600/r600_emit.c +++ b/src/mesa/drivers/dri/r600/r600_emit.c @@ -63,26 +63,15 @@ GLboolean r600EmitShader(GLcontext * ctx, struct radeon_bo * pbo; uint32_t *out; -shader_again_alloc: -#ifdef RADEON_DEBUG_BO - pbo = radeon_bo_open(radeonctx->radeonScreen->bom, - 0, - sizeinDWORD * 4, - 256, - RADEON_GEM_DOMAIN_GTT, - 0, - szShaderUsage); -#else +shader_again_alloc: pbo = radeon_bo_open(radeonctx->radeonScreen->bom, - 0, - sizeinDWORD * 4, - 256, + 0, + sizeinDWORD * 4, + 256, RADEON_GEM_DOMAIN_GTT, - 0); -#endif /* RADEON_DEBUG_BO */ + 0); - if (!pbo) - { + if (!pbo) { rcommonFlushCmdBuf(radeonctx, __FUNCTION__); goto shader_again_alloc; } diff --git a/src/mesa/drivers/dri/r600/r600_tex.c b/src/mesa/drivers/dri/r600/r600_tex.c index 43d9f641af..6d531bf0f9 100644 --- a/src/mesa/drivers/dri/r600/r600_tex.c +++ b/src/mesa/drivers/dri/r600/r600_tex.c @@ -62,10 +62,10 @@ static unsigned int translate_wrap_mode(GLenum wrapmode) case GL_CLAMP: return SQ_TEX_CLAMP_HALF_BORDER; case GL_CLAMP_TO_EDGE: return SQ_TEX_CLAMP_LAST_TEXEL; case GL_CLAMP_TO_BORDER: return SQ_TEX_CLAMP_BORDER; - case GL_MIRRORED_REPEAT: return SQ_TEX_MIRROR_ONCE_HALF_BORDER; - case GL_MIRROR_CLAMP_EXT: return SQ_TEX_MIRROR; - case GL_MIRROR_CLAMP_TO_EDGE_EXT: return SQ_TEX_MIRROR_ONCE_BORDER; - case GL_MIRROR_CLAMP_TO_BORDER_EXT: return SQ_TEX_MIRROR_ONCE_LAST_TEXEL; + case GL_MIRRORED_REPEAT: return SQ_TEX_MIRROR; + case GL_MIRROR_CLAMP_EXT: return SQ_TEX_MIRROR_ONCE_HALF_BORDER; + case GL_MIRROR_CLAMP_TO_EDGE_EXT: return SQ_TEX_MIRROR_ONCE_LAST_TEXEL; + case GL_MIRROR_CLAMP_TO_BORDER_EXT: return SQ_TEX_MIRROR_ONCE_BORDER; default: _mesa_problem(NULL, "bad wrap mode in %s", __FUNCTION__); return 0; @@ -127,10 +127,18 @@ static void r600SetTexDefaultState(radeonTexObjPtr t) SETfield(t->SQ_TEX_RESOURCE4, SQ_ENDIAN_NONE, SQ_TEX_RESOURCE_WORD4_0__ENDIAN_SWAP_shift, SQ_TEX_RESOURCE_WORD4_0__ENDIAN_SWAP_mask); SETfield(t->SQ_TEX_RESOURCE4, 1, REQUEST_SIZE_shift, REQUEST_SIZE_mask); - t->SQ_TEX_RESOURCE4 |= SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift - |SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift - |SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift - |SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift; + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); SETfield(t->SQ_TEX_RESOURCE4, 0, BASE_LEVEL_shift, BASE_LEVEL_mask); /* mip-maps */ t->SQ_TEX_RESOURCE5 = 0; @@ -141,17 +149,18 @@ static void r600SetTexDefaultState(radeonTexObjPtr t) /* Initialize sampler registers */ t->SQ_TEX_SAMPLER0 = 0; - t->SQ_TEX_SAMPLER0 |= - SQ_TEX_WRAP << SQ_TEX_SAMPLER_WORD0_0__CLAMP_X_shift - |SQ_TEX_WRAP << CLAMP_Y_shift - |SQ_TEX_WRAP << CLAMP_Z_shift - |SQ_TEX_XY_FILTER_POINT << XY_MAG_FILTER_shift - |SQ_TEX_XY_FILTER_POINT << XY_MIN_FILTER_shift - |SQ_TEX_Z_FILTER_NONE << Z_FILTER_shift - |SQ_TEX_Z_FILTER_NONE << MIP_FILTER_shift - |SQ_TEX_BORDER_COLOR_TRANS_BLACK << BORDER_COLOR_TYPE_shift; - - t->SQ_TEX_SAMPLER1 = 0x7FF << MAX_LOD_shift; + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_WRAP, SQ_TEX_SAMPLER_WORD0_0__CLAMP_X_shift, + SQ_TEX_SAMPLER_WORD0_0__CLAMP_X_mask); + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_WRAP, CLAMP_Y_shift, CLAMP_Y_mask); + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_WRAP, CLAMP_Z_shift, CLAMP_Z_mask); + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_XY_FILTER_POINT, XY_MAG_FILTER_shift, XY_MAG_FILTER_mask); + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_XY_FILTER_POINT, XY_MIN_FILTER_shift, XY_MIN_FILTER_mask); + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_Z_FILTER_NONE, Z_FILTER_shift, Z_FILTER_mask); + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_Z_FILTER_NONE, MIP_FILTER_shift, MIP_FILTER_mask); + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_BORDER_COLOR_TRANS_BLACK, BORDER_COLOR_TYPE_shift, BORDER_COLOR_TYPE_mask); + + t->SQ_TEX_SAMPLER1 = 0; + SETfield(t->SQ_TEX_SAMPLER1, 0x3ff, MAX_LOD_shift, MAX_LOD_mask); t->SQ_TEX_SAMPLER2 = 0; SETbit(t->SQ_TEX_SAMPLER2, SQ_TEX_SAMPLER_WORD2_0__TYPE_bit); @@ -260,14 +269,12 @@ static void r600SetTexFilter(radeonTexObjPtr t, GLenum minf, GLenum magf, GLfloa static void r600SetTexBorderColor(radeonTexObjPtr t, const GLfloat color[4]) { -#if 0 - GLubyte c[4]; - CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]); - CLAMPED_FLOAT_TO_UBYTE(c[1], color[1]); - CLAMPED_FLOAT_TO_UBYTE(c[2], color[2]); - CLAMPED_FLOAT_TO_UBYTE(c[3], color[3]); - t->pp_border_color = PACK_COLOR_8888(c[3], c[0], c[1], c[2]); -#endif + t->TD_PS_SAMPLER0_BORDER_ALPHA = *((uint32_t*)&(color[3])); + t->TD_PS_SAMPLER0_BORDER_RED = *((uint32_t*)&(color[2])); + t->TD_PS_SAMPLER0_BORDER_GREEN = *((uint32_t*)&(color[1])); + t->TD_PS_SAMPLER0_BORDER_BLUE = *((uint32_t*)&(color[0])); + SETfield(t->SQ_TEX_SAMPLER0, SQ_TEX_BORDER_COLOR_REGISTER, + BORDER_COLOR_TYPE_shift, BORDER_COLOR_TYPE_mask); } /** diff --git a/src/mesa/drivers/dri/r600/r600_texstate.c b/src/mesa/drivers/dri/r600/r600_texstate.c index 2466aa9595..ee9b64ee43 100644 --- a/src/mesa/drivers/dri/r600/r600_texstate.c +++ b/src/mesa/drivers/dri/r600/r600_texstate.c @@ -75,10 +75,10 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, GLuint mesa_fo { radeonTexObj *t = radeon_tex_obj(tObj); - t->SQ_TEX_RESOURCE4 &= ~( SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask - |SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask - |SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask - |SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask ); + CLEARfield(t->SQ_TEX_RESOURCE4, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + CLEARfield(t->SQ_TEX_RESOURCE4, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + CLEARfield(t->SQ_TEX_RESOURCE4, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + CLEARfield(t->SQ_TEX_RESOURCE4, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); switch (mesa_format) /* This is mesa format. */ { @@ -86,163 +86,211 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, GLuint mesa_fo SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_RGBA8888_REV: SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_ARGB8888: SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_ARGB8888_REV: SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_RGB888: SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_RGB565: SETfield(t->SQ_TEX_RESOURCE1, FMT_5_6_5, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_RGB565_REV: SETfield(t->SQ_TEX_RESOURCE1, FMT_5_6_5, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_ARGB4444: SETfield(t->SQ_TEX_RESOURCE1, FMT_4_4_4_4, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_ARGB4444_REV: SETfield(t->SQ_TEX_RESOURCE1, FMT_4_4_4_4, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_ARGB1555: SETfield(t->SQ_TEX_RESOURCE1, FMT_1_5_5_5, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_ARGB1555_REV: SETfield(t->SQ_TEX_RESOURCE1, FMT_1_5_5_5, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_AL88: case MESA_FORMAT_AL88_REV: /* TODO : Check this. */ SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_RGB332: SETfield(t->SQ_TEX_RESOURCE1, FMT_3_3_2, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_A8: /* ZERO, ZERO, ZERO, X */ SETfield(t->SQ_TEX_RESOURCE1, FMT_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_L8: /* X, X, X, ONE */ SETfield(t->SQ_TEX_RESOURCE1, FMT_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_I8: /* X, X, X, X */ case MESA_FORMAT_CI8: SETfield(t->SQ_TEX_RESOURCE1, FMT_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; /* YUV422 TODO conversion */ /* X, Y, Z, ONE, G8R8_G8B8 */ /* @@ -272,121 +320,157 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, GLuint mesa_fo SETfield(t->SQ_TEX_RESOURCE1, FMT_32_32_32_32_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_RGBA_FLOAT16: SETfield(t->SQ_TEX_RESOURCE1, FMT_16_16_16_16_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_RGB_FLOAT32: /* X, Y, Z, ONE */ SETfield(t->SQ_TEX_RESOURCE1, FMT_32_32_32_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_RGB_FLOAT16: SETfield(t->SQ_TEX_RESOURCE1, FMT_16_16_16_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_ALPHA_FLOAT32: /* ZERO, ZERO, ZERO, X */ SETfield(t->SQ_TEX_RESOURCE1, FMT_32_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_ALPHA_FLOAT16: /* ZERO, ZERO, ZERO, X */ SETfield(t->SQ_TEX_RESOURCE1, FMT_16_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_LUMINANCE_FLOAT32: /* X, X, X, ONE */ SETfield(t->SQ_TEX_RESOURCE1, FMT_32_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_LUMINANCE_FLOAT16: /* X, X, X, ONE */ SETfield(t->SQ_TEX_RESOURCE1, FMT_16_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT32: SETfield(t->SQ_TEX_RESOURCE1, FMT_32_32_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_LUMINANCE_ALPHA_FLOAT16: SETfield(t->SQ_TEX_RESOURCE1, FMT_16_16_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_INTENSITY_FLOAT32: /* X, X, X, X */ SETfield(t->SQ_TEX_RESOURCE1, FMT_32_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_INTENSITY_FLOAT16: /* X, X, X, X */ SETfield(t->SQ_TEX_RESOURCE1, FMT_16_FLOAT, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case MESA_FORMAT_Z16: case MESA_FORMAT_Z24_S8: @@ -407,25 +491,34 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, GLuint mesa_fo }; switch (tObj->DepthMode) { case GL_LUMINANCE: /* X, X, X, ONE */ - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case GL_INTENSITY: /* X, X, X, X */ - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; case GL_ALPHA: /* ZERO, ZERO, ZERO, X */ - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_0 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_0, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); break; default: return GL_FALSE; @@ -441,7 +534,6 @@ static GLboolean r600GetTexFormat(struct gl_texture_object *tObj, GLuint mesa_fo void r600SetDepthTexMode(struct gl_texture_object *tObj) { - const GLuint *format; radeonTexObjPtr t; if (!tObj) @@ -464,7 +556,7 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex radeonTexObj *t = radeon_tex_obj(texObj); const struct gl_texture_image *firstImage; int firstlevel = t->mt ? t->mt->firstLevel : 0; - GLuint uTexelPitch; + GLuint uTexelPitch, row_align; firstImage = t->base.Image[0][firstlevel]; @@ -476,9 +568,6 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex } } - if (t->image_override && t->bo) - return; - switch (texObj->Target) { case GL_TEXTURE_1D: SETfield(t->SQ_TEX_RESOURCE0, SQ_TEX_DIM_1D, DIM_shift, DIM_mask); @@ -503,15 +592,26 @@ static void setup_hardware_state(context_t *rmesa, struct gl_texture_object *tex return; } - uTexelPitch = (firstImage->Width + R700_TEXEL_PITCH_ALIGNMENT_MASK) + row_align = rmesa->radeon.texture_row_align - 1; + uTexelPitch = ((firstImage->Width * t->mt->bpp + row_align) & ~row_align) / t->mt->bpp; + uTexelPitch = (uTexelPitch + R700_TEXEL_PITCH_ALIGNMENT_MASK) & ~R700_TEXEL_PITCH_ALIGNMENT_MASK; + /* min pitch is 8 */ + if (uTexelPitch < 8) + uTexelPitch = 8; + SETfield(t->SQ_TEX_RESOURCE0, (uTexelPitch/8)-1, PITCH_shift, PITCH_mask); SETfield(t->SQ_TEX_RESOURCE0, firstImage->Width - 1, TEX_WIDTH_shift, TEX_WIDTH_mask); SETfield(t->SQ_TEX_RESOURCE1, firstImage->Height - 1, TEX_HEIGHT_shift, TEX_HEIGHT_mask); + if ((t->mt->lastLevel - t->mt->firstLevel) > 0) { + t->SQ_TEX_RESOURCE3 = t->mt->levels[0].size / 256; + SETfield(t->SQ_TEX_RESOURCE4, t->mt->firstLevel, BASE_LEVEL_shift, BASE_LEVEL_mask); + SETfield(t->SQ_TEX_RESOURCE5, t->mt->lastLevel, LAST_LEVEL_shift, LAST_LEVEL_mask); + } } /** @@ -598,7 +698,7 @@ void r600SetTexOffset(__DRIcontext * pDRICtx, GLint texname, struct gl_texture_object *tObj = _mesa_lookup_texture(rmesa->radeon.glCtx, texname); radeonTexObjPtr t = radeon_tex_obj(tObj); - uint32_t pitch_val; + uint32_t pitch_val, size; if (!tObj) return; @@ -608,7 +708,12 @@ void r600SetTexOffset(__DRIcontext * pDRICtx, GLint texname, if (!offset) return; - t->bo = NULL; + size = pitch;//h * w * (depth / 8); + if (t->bo) { + radeon_bo_unref(t->bo); + t->bo = NULL; + } + t->bo = radeon_legacy_bo_alloc_fake(rmesa->radeon.radeonScreen->bom, size, offset); t->override_offset = offset; pitch_val = pitch; switch (depth) { @@ -616,11 +721,14 @@ void r600SetTexOffset(__DRIcontext * pDRICtx, GLint texname, SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); pitch_val /= 4; break; case 24: @@ -628,28 +736,39 @@ void r600SetTexOffset(__DRIcontext * pDRICtx, GLint texname, SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); pitch_val /= 4; break; case 16: SETfield(t->SQ_TEX_RESOURCE1, FMT_5_6_5, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); pitch_val /= 2; break; } pitch_val = (pitch_val + R700_TEXEL_PITCH_ALIGNMENT_MASK) & ~R700_TEXEL_PITCH_ALIGNMENT_MASK; + + /* min pitch is 8 */ + if (pitch_val < 8) + pitch_val = 8; + SETfield(t->SQ_TEX_RESOURCE0, (pitch_val/8)-1, PITCH_shift, PITCH_mask); } @@ -739,20 +858,26 @@ void r600SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_fo SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); } else { SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); } pitch_val /= 4; break; @@ -762,22 +887,28 @@ void r600SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_fo SETfield(t->SQ_TEX_RESOURCE1, FMT_8_8_8_8, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_W << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_W, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); pitch_val /= 4; break; case 2: SETfield(t->SQ_TEX_RESOURCE1, FMT_5_6_5, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_shift, SQ_TEX_RESOURCE_WORD1_0__DATA_FORMAT_mask); - t->SQ_TEX_RESOURCE4 |= - (SQ_SEL_Z << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift) - |(SQ_SEL_Y << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift) - |(SQ_SEL_X << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift) - |(SQ_SEL_1 << SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Z, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_X_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_Y, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Y_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_X, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_Z_mask); + SETfield(t->SQ_TEX_RESOURCE4, SQ_SEL_1, + SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_shift, SQ_TEX_RESOURCE_WORD4_0__DST_SEL_W_mask); pitch_val /= 2; break; } @@ -785,6 +916,10 @@ void r600SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_fo pitch_val = (pitch_val + R700_TEXEL_PITCH_ALIGNMENT_MASK) & ~R700_TEXEL_PITCH_ALIGNMENT_MASK; + /* min pitch is 8 */ + if (pitch_val < 8) + pitch_val = 8; + SETfield(t->SQ_TEX_RESOURCE0, (pitch_val/8)-1, PITCH_shift, PITCH_mask); SETfield(t->SQ_TEX_RESOURCE0, rb->base.Width - 1, TEX_WIDTH_shift, TEX_WIDTH_mask); diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index 1d41c5cf78..eaacd06113 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -1205,7 +1205,7 @@ GLboolean tex_src(r700_AssemblerBase *pAsm) return GL_TRUE; } -GLboolean assemble_tex_instruction(r700_AssemblerBase *pAsm) +GLboolean assemble_tex_instruction(r700_AssemblerBase *pAsm, GLboolean normalized) { PVSSRC * texture_coordinate_source; PVSSRC * texture_unit_source; @@ -1227,10 +1227,18 @@ GLboolean assemble_tex_instruction(r700_AssemblerBase *pAsm) tex_instruction_ptr->m_Word0.f.resource_id = texture_unit_source->reg; tex_instruction_ptr->m_Word1.f.lod_bias = 0x0; - tex_instruction_ptr->m_Word1.f.coord_type_x = SQ_TEX_NORMALIZED; - tex_instruction_ptr->m_Word1.f.coord_type_y = SQ_TEX_NORMALIZED; - tex_instruction_ptr->m_Word1.f.coord_type_z = SQ_TEX_NORMALIZED; - tex_instruction_ptr->m_Word1.f.coord_type_w = SQ_TEX_NORMALIZED; + if (normalized) { + tex_instruction_ptr->m_Word1.f.coord_type_x = SQ_TEX_NORMALIZED; + tex_instruction_ptr->m_Word1.f.coord_type_y = SQ_TEX_NORMALIZED; + tex_instruction_ptr->m_Word1.f.coord_type_z = SQ_TEX_NORMALIZED; + tex_instruction_ptr->m_Word1.f.coord_type_w = SQ_TEX_NORMALIZED; + } else { + /* XXX: UNNORMALIZED tex coords have limited wrap modes */ + tex_instruction_ptr->m_Word1.f.coord_type_x = SQ_TEX_UNNORMALIZED; + tex_instruction_ptr->m_Word1.f.coord_type_y = SQ_TEX_UNNORMALIZED; + tex_instruction_ptr->m_Word1.f.coord_type_z = SQ_TEX_UNNORMALIZED; + tex_instruction_ptr->m_Word1.f.coord_type_w = SQ_TEX_UNNORMALIZED; + } tex_instruction_ptr->m_Word2.f.offset_x = 0x0; tex_instruction_ptr->m_Word2.f.offset_y = 0x0; @@ -2196,11 +2204,19 @@ GLboolean next_ins(r700_AssemblerBase *pAsm) if( GL_TRUE == IsTex(pILInst->Opcode) ) { - if( GL_FALSE == assemble_tex_instruction(pAsm) ) - { - r700_error(ERROR_ASM_TEXINSTRUCTION, "Error assembling TEX instruction"); - return GL_FALSE; - } + if (pILInst->TexSrcTarget == TEXTURE_RECT_INDEX) { + if( GL_FALSE == assemble_tex_instruction(pAsm, GL_FALSE) ) + { + r700_error(ERROR_ASM_TEXINSTRUCTION, "Error assembling TEX instruction"); + return GL_FALSE; + } + } else { + if( GL_FALSE == assemble_tex_instruction(pAsm, GL_TRUE) ) + { + r700_error(ERROR_ASM_TEXINSTRUCTION, "Error assembling TEX instruction"); + return GL_FALSE; + } + } } else { //ALU @@ -3823,6 +3839,9 @@ GLboolean Process_Export(r700_AssemblerBase* pAsm, if (export_count == 1) { ucWriteMask = pAsm->pucOutMask[starting_register_number - pAsm->starting_export_register_number]; + /* exports Z as a float into Red channel */ + if (GL_TRUE == is_depth_export) + ucWriteMask = 0x1; if( (ucWriteMask & 0x1) != 0) { @@ -4014,6 +4033,22 @@ GLboolean Process_Vertex_Exports(r700_AssemblerBase *pR700AsmCode, export_starting_index++; } + unBit = 1 << VERT_RESULT_FOGC; + if(OutputsWritten & unBit) + { + if( GL_FALSE == Process_Export(pR700AsmCode, + SQ_EXPORT_PARAM, + export_starting_index, + 1, + pR700AsmCode->ucVP_OutputMap[VERT_RESULT_FOGC], + GL_FALSE) ) + { + return GL_FALSE; + } + + export_starting_index++; + } + for(i=0; i<8; i++) { unBit = 1 << (VERT_RESULT_TEX0 + i); diff --git a/src/mesa/drivers/dri/r600/r700_assembler.h b/src/mesa/drivers/dri/r600/r700_assembler.h index e9b21b802e..f9c4d849c6 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.h +++ b/src/mesa/drivers/dri/r600/r700_assembler.h @@ -424,7 +424,7 @@ GLboolean assemble_src(r700_AssemblerBase *pAsm, GLboolean assemble_dst(r700_AssemblerBase *pAsm); GLboolean tex_dst(r700_AssemblerBase *pAsm); GLboolean tex_src(r700_AssemblerBase *pAsm); -GLboolean assemble_tex_instruction(r700_AssemblerBase *pAsm); +GLboolean assemble_tex_instruction(r700_AssemblerBase *pAsm, GLboolean normalized); void initialize(r700_AssemblerBase *pAsm); GLboolean assemble_alu_src(R700ALUInstruction* alu_instruction_ptr, int source_index, diff --git a/src/mesa/drivers/dri/r600/r700_chip.c b/src/mesa/drivers/dri/r600/r700_chip.c index 087d17312e..0fb355a0b6 100644 --- a/src/mesa/drivers/dri/r600/r700_chip.c +++ b/src/mesa/drivers/dri/r600/r700_chip.c @@ -38,6 +38,8 @@ #include "r700_vertprog.h" #include "r700_ioctl.h" +#include "radeon_mipmap_tree.h" + #define LINK_STATES(reg) \ do \ { \ @@ -136,9 +138,23 @@ GLboolean r700InitChipObject(context_t *context) LINK_STATES(CB_CLRCMP_MSK); LINK_STATES(CB_BLEND_CONTROL); + //DB + LINK_STATES(DB_HTILE_DATA_BASE); + LINK_STATES(DB_STENCIL_CLEAR); + LINK_STATES(DB_DEPTH_CLEAR); + LINK_STATES(DB_STENCILREFMASK); + LINK_STATES(DB_STENCILREFMASK_BF); + LINK_STATES(DB_DEPTH_CONTROL); + LINK_STATES(DB_SHADER_CONTROL); + LINK_STATES(DB_RENDER_CONTROL); + LINK_STATES(DB_RENDER_OVERRIDE); + LINK_STATES(DB_HTILE_SURFACE); + LINK_STATES(DB_ALPHA_TO_MASK); + // SX LINK_STATES(SX_MISC); LINK_STATES(SX_ALPHA_TEST_CONTROL); + LINK_STATES(SX_ALPHA_REF); // VGT LINK_STATES(VGT_MAX_VTX_INDX); @@ -159,7 +175,6 @@ GLboolean r700InitChipObject(context_t *context) LINK_STATES(VGT_GROUP_VECT_1_FMT_CNTL); LINK_STATES(VGT_GS_MODE); LINK_STATES(VGT_PRIMITIVEID_EN); - LINK_STATES(VGT_DMA_NUM_INSTANCES); LINK_STATES(VGT_MULTI_PRIM_IB_RESET_EN); LINK_STATES(VGT_INSTANCE_STEP_RATE_0); LINK_STATES(VGT_INSTANCE_STEP_RATE_1); @@ -213,39 +228,6 @@ GLboolean r700InitChipObject(context_t *context) LINK_STATES(SPI_VS_OUT_ID_8); LINK_STATES(SPI_VS_OUT_ID_9); - LINK_STATES(SPI_PS_INPUT_CNTL_0); - LINK_STATES(SPI_PS_INPUT_CNTL_1); - LINK_STATES(SPI_PS_INPUT_CNTL_2); - LINK_STATES(SPI_PS_INPUT_CNTL_3); - LINK_STATES(SPI_PS_INPUT_CNTL_4); - LINK_STATES(SPI_PS_INPUT_CNTL_5); - LINK_STATES(SPI_PS_INPUT_CNTL_6); - LINK_STATES(SPI_PS_INPUT_CNTL_7); - LINK_STATES(SPI_PS_INPUT_CNTL_8); - LINK_STATES(SPI_PS_INPUT_CNTL_9); - LINK_STATES(SPI_PS_INPUT_CNTL_10); - LINK_STATES(SPI_PS_INPUT_CNTL_11); - LINK_STATES(SPI_PS_INPUT_CNTL_12); - LINK_STATES(SPI_PS_INPUT_CNTL_13); - LINK_STATES(SPI_PS_INPUT_CNTL_14); - LINK_STATES(SPI_PS_INPUT_CNTL_15); - LINK_STATES(SPI_PS_INPUT_CNTL_16); - LINK_STATES(SPI_PS_INPUT_CNTL_17); - LINK_STATES(SPI_PS_INPUT_CNTL_18); - LINK_STATES(SPI_PS_INPUT_CNTL_19); - LINK_STATES(SPI_PS_INPUT_CNTL_20); - LINK_STATES(SPI_PS_INPUT_CNTL_21); - LINK_STATES(SPI_PS_INPUT_CNTL_22); - LINK_STATES(SPI_PS_INPUT_CNTL_23); - LINK_STATES(SPI_PS_INPUT_CNTL_24); - LINK_STATES(SPI_PS_INPUT_CNTL_25); - LINK_STATES(SPI_PS_INPUT_CNTL_26); - LINK_STATES(SPI_PS_INPUT_CNTL_27); - LINK_STATES(SPI_PS_INPUT_CNTL_28); - LINK_STATES(SPI_PS_INPUT_CNTL_29); - LINK_STATES(SPI_PS_INPUT_CNTL_30); - LINK_STATES(SPI_PS_INPUT_CNTL_31); - LINK_STATES(SPI_VS_OUT_CONFIG); LINK_STATES(SPI_THREAD_GROUPING); LINK_STATES(SPI_PS_IN_CONTROL_0); @@ -274,6 +256,67 @@ GLboolean r700InitChipObject(context_t *context) return GL_TRUE; } +GLboolean r700SendTextureState(context_t *context) +{ + unsigned int i; + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + struct radeon_bo *bo = NULL; + BATCH_LOCALS(&context->radeon); + + for (i=0; i<R700_TEXTURE_NUMBERUNITS; i++) { + radeonTexObj *t = r700->textures[i]; + if (t) { + if (!t->image_override) + bo = t->mt->bo; + else + bo = t->bo; + if (bo) { + + r700SyncSurf(context, bo, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, + 0, TC_ACTION_ENA_bit); + + BEGIN_BATCH_NO_AUTOSTATE(9); + R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_RESOURCE, 7)); + R600_OUT_BATCH(i * 7); + R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE0); + R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE1); + R600_OUT_BATCH_RELOC(r700->textures[i]->SQ_TEX_RESOURCE2, + bo, + 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + R600_OUT_BATCH_RELOC(r700->textures[i]->SQ_TEX_RESOURCE3, + bo, + r700->textures[i]->SQ_TEX_RESOURCE3, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE4); + R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE5); + R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE6); + END_BATCH(); + + BEGIN_BATCH_NO_AUTOSTATE(5); + R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_SAMPLER, 3)); + R600_OUT_BATCH(i * 3); + R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER0); + R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER1); + R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER2); + END_BATCH(); + + BEGIN_BATCH_NO_AUTOSTATE(2 + 4); + R600_OUT_BATCH_REGSEQ((TD_PS_SAMPLER0_BORDER_RED + (i * 16)), 4); + R600_OUT_BATCH(r700->textures[i]->TD_PS_SAMPLER0_BORDER_RED); + R600_OUT_BATCH(r700->textures[i]->TD_PS_SAMPLER0_BORDER_GREEN); + R600_OUT_BATCH(r700->textures[i]->TD_PS_SAMPLER0_BORDER_BLUE); + R600_OUT_BATCH(r700->textures[i]->TD_PS_SAMPLER0_BORDER_ALPHA); + END_BATCH(); + + COMMIT_BATCH(); + } + } + } + return GL_TRUE; +} + void r700SetupVTXConstants(GLcontext * ctx, unsigned int nStreamID, void * pAos, @@ -282,10 +325,7 @@ void r700SetupVTXConstants(GLcontext * ctx, unsigned int count) /* number of vectors in stream */ { context_t *context = R700_CONTEXT(ctx); - uint32_t *dest; struct radeon_aos * paos = (struct radeon_aos *)pAos; - offset_modifiers offset_mod = {NO_SHIFT, 0, 0xFFFFFFFF}; - BATCH_LOCALS(&context->radeon); unsigned int uSQ_VTX_CONSTANT_WORD0_0; @@ -294,6 +334,9 @@ void r700SetupVTXConstants(GLcontext * ctx, unsigned int uSQ_VTX_CONSTANT_WORD3_0 = 0; unsigned int uSQ_VTX_CONSTANT_WORD6_0 = 0; + if (!paos->bo) + return; + if ((context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV610) || (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV620) || (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_RS780) || @@ -305,15 +348,19 @@ void r700SetupVTXConstants(GLcontext * ctx, uSQ_VTX_CONSTANT_WORD0_0 = paos->offset; uSQ_VTX_CONSTANT_WORD1_0 = count * (size * 4) - 1; - uSQ_VTX_CONSTANT_WORD2_0 |= 0 << BASE_ADDRESS_HI_shift /* TODO */ - |stride << SQ_VTX_CONSTANT_WORD2_0__STRIDE_shift - |GetSurfaceFormat(GL_FLOAT, size, NULL) << SQ_VTX_CONSTANT_WORD2_0__DATA_FORMAT_shift /* TODO : trace back api for initial data type, not only GL_FLOAT */ - |SQ_NUM_FORMAT_SCALED << SQ_VTX_CONSTANT_WORD2_0__NUM_FORMAT_ALL_shift - |SQ_VTX_CONSTANT_WORD2_0__FORMAT_COMP_ALL_bit; - - uSQ_VTX_CONSTANT_WORD3_0 |= 1 << MEM_REQUEST_SIZE_shift; + SETfield(uSQ_VTX_CONSTANT_WORD2_0, 0, BASE_ADDRESS_HI_shift, BASE_ADDRESS_HI_mask); /* TODO */ + SETfield(uSQ_VTX_CONSTANT_WORD2_0, stride, SQ_VTX_CONSTANT_WORD2_0__STRIDE_shift, + SQ_VTX_CONSTANT_WORD2_0__STRIDE_mask); + SETfield(uSQ_VTX_CONSTANT_WORD2_0, GetSurfaceFormat(GL_FLOAT, size, NULL), + SQ_VTX_CONSTANT_WORD2_0__DATA_FORMAT_shift, + SQ_VTX_CONSTANT_WORD2_0__DATA_FORMAT_mask); /* TODO : trace back api for initial data type, not only GL_FLOAT */ + SETfield(uSQ_VTX_CONSTANT_WORD2_0, SQ_NUM_FORMAT_SCALED, + SQ_VTX_CONSTANT_WORD2_0__NUM_FORMAT_ALL_shift, SQ_VTX_CONSTANT_WORD2_0__NUM_FORMAT_ALL_mask); + SETbit(uSQ_VTX_CONSTANT_WORD2_0, SQ_VTX_CONSTANT_WORD2_0__FORMAT_COMP_ALL_bit); - uSQ_VTX_CONSTANT_WORD6_0 |= SQ_TEX_VTX_VALID_BUFFER << SQ_TEX_RESOURCE_WORD6_0__TYPE_shift; + SETfield(uSQ_VTX_CONSTANT_WORD3_0, 1, MEM_REQUEST_SIZE_shift, MEM_REQUEST_SIZE_mask); + SETfield(uSQ_VTX_CONSTANT_WORD6_0, SQ_TEX_VTX_VALID_BUFFER, + SQ_TEX_RESOURCE_WORD6_0__TYPE_shift, SQ_TEX_RESOURCE_WORD6_0__TYPE_mask); BEGIN_BATCH_NO_AUTOSTATE(9); @@ -323,7 +370,7 @@ void r700SetupVTXConstants(GLcontext * ctx, R600_OUT_BATCH_RELOC(uSQ_VTX_CONSTANT_WORD0_0, paos->bo, uSQ_VTX_CONSTANT_WORD0_0, - RADEON_GEM_DOMAIN_GTT, 0, 0, &offset_mod); + RADEON_GEM_DOMAIN_GTT, 0, 0); R600_OUT_BATCH(uSQ_VTX_CONSTANT_WORD1_0); R600_OUT_BATCH(uSQ_VTX_CONSTANT_WORD2_0); R600_OUT_BATCH(uSQ_VTX_CONSTANT_WORD3_0); @@ -339,7 +386,6 @@ void r700SetupVTXConstants(GLcontext * ctx, int r700SetupStreams(GLcontext * ctx) { context_t *context = R700_CONTEXT(ctx); - BATCH_LOCALS(&context->radeon); struct r700_vertex_program *vpc @@ -353,7 +399,7 @@ int r700SetupStreams(GLcontext * ctx) BEGIN_BATCH_NO_AUTOSTATE(6); R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CTL_CONST, 1)); - R600_OUT_BATCH(mmSQ_VTX_BASE_VTX_LOC - ASIC_CTL_CONST_BASE_INDEX); + R600_OUT_BATCH(mmSQ_VTX_BASE_VTX_LOC - ASIC_CTL_CONST_BASE_INDEX); R600_OUT_BATCH(0); R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CTL_CONST, 1)); @@ -431,17 +477,37 @@ GLboolean r700SendContextStates(context_t *context) }; END_BATCH(); }; + + /* todo: + * - split this into a separate function? + * - only emit the ones we use + */ + BEGIN_BATCH_NO_AUTOSTATE(2 + R700_MAX_SHADER_EXPORTS); + R600_OUT_BATCH_REGSEQ(SPI_PS_INPUT_CNTL_0, R700_MAX_SHADER_EXPORTS); + for(ui = 0; ui < R700_MAX_SHADER_EXPORTS; ui++) + R600_OUT_BATCH(r700->SPI_PS_INPUT_CNTL[ui].u32All); + END_BATCH(); + + if (context->radeon.radeonScreen->chip_family > CHIP_FAMILY_R600) { + for (ui = 0; ui < R700_MAX_RENDER_TARGETS; ui++) { + if (r700->render_target[ui].enabled) { + BEGIN_BATCH_NO_AUTOSTATE(3); + R600_OUT_BATCH_REGVAL(CB_BLEND0_CONTROL + (4 * ui), + r700->render_target[ui].CB_BLEND0_CONTROL.u32All); + END_BATCH(); + } + } + } + COMMIT_BATCH(); return GL_TRUE; } - -GLboolean r700SendDepthTargetState(context_t *context, int id) +GLboolean r700SendDepthTargetState(context_t *context) { R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); struct radeon_renderbuffer *rrb; - offset_modifiers offset_mod; BATCH_LOCALS(&context->radeon); rrb = radeon_get_depthbuffer(&context->radeon); @@ -450,42 +516,25 @@ GLboolean r700SendDepthTargetState(context_t *context, int id) return GL_FALSE; } - offset_mod.shift = NO_SHIFT; - offset_mod.shiftbits = 0; - offset_mod.mask = 0xFFFFFFFF; - - BEGIN_BATCH_NO_AUTOSTATE(9); + BEGIN_BATCH_NO_AUTOSTATE(8); R600_OUT_BATCH_REGSEQ(DB_DEPTH_SIZE, 2); R600_OUT_BATCH(r700->DB_DEPTH_SIZE.u32All); R600_OUT_BATCH(r700->DB_DEPTH_VIEW.u32All); - R600_OUT_BATCH_REGSEQ(DB_DEPTH_BASE, 3); + R600_OUT_BATCH_REGSEQ(DB_DEPTH_BASE, 2); R600_OUT_BATCH_RELOC(r700->DB_DEPTH_BASE.u32All, rrb->bo, r700->DB_DEPTH_BASE.u32All, - 0, RADEON_GEM_DOMAIN_VRAM, 0, &offset_mod); + 0, RADEON_GEM_DOMAIN_VRAM, 0); R600_OUT_BATCH(r700->DB_DEPTH_INFO.u32All); - R600_OUT_BATCH(r700->DB_HTILE_DATA_BASE.u32All); END_BATCH(); - BEGIN_BATCH_NO_AUTOSTATE(24); - R600_OUT_BATCH_REGSEQ(DB_STENCIL_CLEAR, 2); - R600_OUT_BATCH(r700->DB_STENCIL_CLEAR.u32All); - R600_OUT_BATCH(r700->DB_DEPTH_CLEAR.u32All); - - R600_OUT_BATCH_REGSEQ(DB_STENCILREFMASK, 2); - R600_OUT_BATCH(r700->DB_STENCILREFMASK.u32All); - R600_OUT_BATCH(r700->DB_STENCILREFMASK_BF.u32All); - - R600_OUT_BATCH_REGVAL(DB_DEPTH_CONTROL, r700->DB_DEPTH_CONTROL.u32All); - R600_OUT_BATCH_REGVAL(DB_SHADER_CONTROL, r700->DB_SHADER_CONTROL.u32All); - - R600_OUT_BATCH_REGSEQ(DB_RENDER_CONTROL, 2); - R600_OUT_BATCH(r700->DB_RENDER_CONTROL.u32All); - R600_OUT_BATCH(r700->DB_RENDER_OVERRIDE.u32All); - - R600_OUT_BATCH_REGVAL(DB_HTILE_SURFACE, r700->DB_HTILE_SURFACE.u32All); - R600_OUT_BATCH_REGVAL(DB_ALPHA_TO_MASK, r700->DB_ALPHA_TO_MASK.u32All); - END_BATCH(); + if ((context->radeon.radeonScreen->chip_family > CHIP_FAMILY_R600) && + (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770)) { + BEGIN_BATCH_NO_AUTOSTATE(2); + R600_OUT_BATCH(CP_PACKET3(R600_IT_SURFACE_BASE_UPDATE, 0)); + R600_OUT_BATCH(1 << 0); + END_BATCH(); + } COMMIT_BATCH(); @@ -499,7 +548,6 @@ GLboolean r700SendRenderTargetState(context_t *context, int id) { R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); struct radeon_renderbuffer *rrb; - offset_modifiers offset_mod; BATCH_LOCALS(&context->radeon); rrb = radeon_get_colorbuffer(&context->radeon); @@ -514,16 +562,12 @@ GLboolean r700SendRenderTargetState(context_t *context, int id) if (!r700->render_target[id].enabled) return GL_FALSE; - offset_mod.shift = NO_SHIFT; - offset_mod.shiftbits = 0; - offset_mod.mask = 0xFFFFFFFF; - BEGIN_BATCH_NO_AUTOSTATE(3); R600_OUT_BATCH_REGSEQ(CB_COLOR0_BASE + (4 * id), 1); R600_OUT_BATCH_RELOC(r700->render_target[id].CB_COLOR0_BASE.u32All, rrb->bo, r700->render_target[id].CB_COLOR0_BASE.u32All, - 0, RADEON_GEM_DOMAIN_VRAM, 0, &offset_mod); + 0, RADEON_GEM_DOMAIN_VRAM, 0); END_BATCH(); if ((context->radeon.radeonScreen->chip_family > CHIP_FAMILY_R600) && @@ -543,12 +587,6 @@ GLboolean r700SendRenderTargetState(context_t *context, int id) R600_OUT_BATCH_REGVAL(CB_COLOR0_MASK + (4 * id), r700->render_target[id].CB_COLOR0_MASK.u32All); END_BATCH(); - if (context->radeon.radeonScreen->chip_family > CHIP_FAMILY_R600) { - BEGIN_BATCH_NO_AUTOSTATE(3); - R600_OUT_BATCH_REGVAL(CB_BLEND0_CONTROL + (4 * id), r700->render_target[id].CB_BLEND0_CONTROL.u32All); - END_BATCH(); - } - COMMIT_BATCH(); r700SyncSurf(context, rrb->bo, 0, RADEON_GEM_DOMAIN_VRAM, @@ -560,16 +598,13 @@ GLboolean r700SendRenderTargetState(context_t *context, int id) GLboolean r700SendPSState(context_t *context) { R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); - struct radeon_renderbuffer *rrb; struct radeon_bo * pbo; - offset_modifiers offset_mod; BATCH_LOCALS(&context->radeon); pbo = (struct radeon_bo *)r700GetActiveFpShaderBo(GL_CONTEXT(context)); - offset_mod.shift = NO_SHIFT; - offset_mod.shiftbits = 0; - offset_mod.mask = 0xFFFFFFFF; + if (!pbo) + return GL_FALSE; r700SyncSurf(context, pbo, RADEON_GEM_DOMAIN_GTT, 0, SH_ACTION_ENA_bit); @@ -578,7 +613,7 @@ GLboolean r700SendPSState(context_t *context) R600_OUT_BATCH_RELOC(r700->ps.SQ_PGM_START_PS.u32All, pbo, r700->ps.SQ_PGM_START_PS.u32All, - RADEON_GEM_DOMAIN_GTT, 0, 0, &offset_mod); + RADEON_GEM_DOMAIN_GTT, 0, 0); END_BATCH(); BEGIN_BATCH_NO_AUTOSTATE(9); @@ -595,16 +630,13 @@ GLboolean r700SendPSState(context_t *context) GLboolean r700SendVSState(context_t *context) { R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); - struct radeon_renderbuffer *rrb; struct radeon_bo * pbo; - offset_modifiers offset_mod; BATCH_LOCALS(&context->radeon); pbo = (struct radeon_bo *)r700GetActiveVpShaderBo(GL_CONTEXT(context)); - offset_mod.shift = NO_SHIFT; - offset_mod.shiftbits = 0; - offset_mod.mask = 0xFFFFFFFF; + if (!pbo) + return GL_FALSE; r700SyncSurf(context, pbo, RADEON_GEM_DOMAIN_GTT, 0, SH_ACTION_ENA_bit); @@ -613,7 +645,7 @@ GLboolean r700SendVSState(context_t *context) R600_OUT_BATCH_RELOC(r700->vs.SQ_PGM_START_VS.u32All, pbo, r700->vs.SQ_PGM_START_VS.u32All, - RADEON_GEM_DOMAIN_GTT, 0, 0, &offset_mod); + RADEON_GEM_DOMAIN_GTT, 0, 0); END_BATCH(); BEGIN_BATCH_NO_AUTOSTATE(6); @@ -629,9 +661,7 @@ GLboolean r700SendVSState(context_t *context) GLboolean r700SendFSState(context_t *context) { R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); - struct radeon_renderbuffer *rrb; struct radeon_bo * pbo; - offset_modifiers offset_mod; BATCH_LOCALS(&context->radeon); /* XXX fixme @@ -645,9 +675,8 @@ GLboolean r700SendFSState(context_t *context) r700->fs.SQ_PGM_CF_OFFSET_FS.u32All = 0; /* XXX */ - offset_mod.shift = NO_SHIFT; - offset_mod.shiftbits = 0; - offset_mod.mask = 0xFFFFFFFF; + if (!pbo) + return GL_FALSE; r700SyncSurf(context, pbo, RADEON_GEM_DOMAIN_GTT, 0, SH_ACTION_ENA_bit); @@ -656,7 +685,7 @@ GLboolean r700SendFSState(context_t *context) R600_OUT_BATCH_RELOC(r700->fs.SQ_PGM_START_FS.u32All, pbo, r700->fs.SQ_PGM_START_FS.u32All, - RADEON_GEM_DOMAIN_GTT, 0, 0, &offset_mod); + RADEON_GEM_DOMAIN_GTT, 0, 0); END_BATCH(); BEGIN_BATCH_NO_AUTOSTATE(6); @@ -672,8 +701,6 @@ GLboolean r700SendFSState(context_t *context) GLboolean r700SendViewportState(context_t *context, int id) { R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); - struct radeon_renderbuffer *rrb; - offset_modifiers offset_mod; BATCH_LOCALS(&context->radeon); if (id > R700_MAX_VIEWPORTS) @@ -722,3 +749,25 @@ GLboolean r700SendSQConfig(context_t *context) return GL_TRUE; } +GLboolean r700SendUCPState(context_t *context) +{ + R700_CHIP_CONTEXT *r700 = R700_CONTEXT_STATES(context); + BATCH_LOCALS(&context->radeon); + int i; + + for (i = 0; i < R700_MAX_UCP; i++) { + if (r700->ucp[i].enabled) { + BEGIN_BATCH_NO_AUTOSTATE(6); + R600_OUT_BATCH_REGSEQ(PA_CL_UCP_0_X + (16 * i), 4); + R600_OUT_BATCH(r700->ucp[i].PA_CL_UCP_0_X.u32All); + R600_OUT_BATCH(r700->ucp[i].PA_CL_UCP_0_Y.u32All); + R600_OUT_BATCH(r700->ucp[i].PA_CL_UCP_0_Z.u32All); + R600_OUT_BATCH(r700->ucp[i].PA_CL_UCP_0_W.u32All); + END_BATCH(); + COMMIT_BATCH(); + } + } + + return GL_TRUE; +} + diff --git a/src/mesa/drivers/dri/r600/r700_chip.h b/src/mesa/drivers/dri/r600/r700_chip.h index 04af4bc3b9..4e89c75f2f 100644 --- a/src/mesa/drivers/dri/r600/r700_chip.h +++ b/src/mesa/drivers/dri/r600/r700_chip.h @@ -372,6 +372,7 @@ typedef struct _R700_CHIP_CONTEXT // SX union UINT_FLOAT SX_MISC ; /* 0xA0D4 */ union UINT_FLOAT SX_ALPHA_TEST_CONTROL ; /* 0xA104 */ + union UINT_FLOAT SX_ALPHA_REF ; /* 0xA10E */ // VGT union UINT_FLOAT VGT_MAX_VTX_INDX ; /* 0xA100 */ @@ -392,7 +393,6 @@ typedef struct _R700_CHIP_CONTEXT union UINT_FLOAT VGT_GROUP_VECT_1_FMT_CNTL ; /* 0xA28F */ union UINT_FLOAT VGT_GS_MODE ; /* 0xA290 */ union UINT_FLOAT VGT_PRIMITIVEID_EN ; /* 0xA2A1 */ - union UINT_FLOAT VGT_DMA_NUM_INSTANCES ; /* 0xA2A2 */ union UINT_FLOAT VGT_MULTI_PRIM_IB_RESET_EN; /* 0xA2A5 */ union UINT_FLOAT VGT_INSTANCE_STEP_RATE_0 ; /* 0xA2A8 */ union UINT_FLOAT VGT_INSTANCE_STEP_RATE_1 ; /* 0xA2A9 */ @@ -422,7 +422,7 @@ typedef struct _R700_CHIP_CONTEXT union UINT_FLOAT SPI_FOG_FUNC_SCALE ; /* 0xA1B8 */ union UINT_FLOAT SPI_FOG_FUNC_BIAS ; /* 0xA1B9 */ - union UINT_FLOAT SQ_VTX_SEMANTIC_0 ; /* 0xA0E0 */ + union UINT_FLOAT SQ_VTX_SEMANTIC_0 ; /* 0xA0E0 */ union UINT_FLOAT SQ_VTX_SEMANTIC_1 ; /* 0xA0E1 */ union UINT_FLOAT SQ_VTX_SEMANTIC_2 ; /* 0xA0E2 */ union UINT_FLOAT SQ_VTX_SEMANTIC_3 ; /* 0xA0E3 */ @@ -455,38 +455,7 @@ typedef struct _R700_CHIP_CONTEXT union UINT_FLOAT SQ_VTX_SEMANTIC_30 ; /* 0xA0FE */ union UINT_FLOAT SQ_VTX_SEMANTIC_31 ; /* 0xA0FF */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_0 ; /* 0xA191 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_1 ; /* 0xA192 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_2 ; /* 0xA193 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_3 ; /* 0xA194 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_4 ; /* 0xA195 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_5 ; /* 0xA196 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_6 ; /* 0xA197 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_7 ; /* 0xA198 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_8 ; /* 0xA199 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_9 ; /* 0xA19A */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_10 ; /* 0xA19B */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_11 ; /* 0xA19C */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_12 ; /* 0xA19D */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_13 ; /* 0xA19E */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_14 ; /* 0xA19F */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_15 ; /* 0xA1A0 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_16 ; /* 0xA1A1 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_17 ; /* 0xA1A2 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_18 ; /* 0xA1A3 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_19 ; /* 0xA1A4 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_20 ; /* 0xA1A5 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_21 ; /* 0xA1A6 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_22 ; /* 0xA1A7 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_23 ; /* 0xA1A8 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_24 ; /* 0xA1A9 */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_25 ; /* 0xA1AA */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_26 ; /* 0xA1AB */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_27 ; /* 0xA1AC */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_28 ; /* 0xA1AD */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_29 ; /* 0xA1AE */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_30 ; /* 0xA1AF */ - union UINT_FLOAT SPI_PS_INPUT_CNTL_31 ; /* 0xA1B0 */ + union UINT_FLOAT SPI_PS_INPUT_CNTL[R700_MAX_SHADER_EXPORTS]; // shaders PS_STATE_STRUCT ps; diff --git a/src/mesa/drivers/dri/r600/r700_clear.c b/src/mesa/drivers/dri/r600/r700_clear.c index e84be38622..05d4af331e 100644 --- a/src/mesa/drivers/dri/r600/r700_clear.c +++ b/src/mesa/drivers/dri/r600/r700_clear.c @@ -31,6 +31,7 @@ #include "main/imports.h" #include "main/mtypes.h" #include "main/enums.h" +#include "swrast/swrast.h" #include "radeon_lock.h" #include "r600_context.h" diff --git a/src/mesa/drivers/dri/r600/r700_fragprog.c b/src/mesa/drivers/dri/r600/r700_fragprog.c index 44de2aebee..6249bde6f1 100644 --- a/src/mesa/drivers/dri/r600/r700_fragprog.c +++ b/src/mesa/drivers/dri/r600/r700_fragprog.c @@ -55,6 +55,12 @@ void Map_Fragment_Program(r700_AssemblerBase *pAsm, //Input mapping : mesa_fp->Base.InputsRead set the flag, set in //The flags parsed in parse_attrib_binding. FRAG_ATTRIB_COLx, FRAG_ATTRIB_TEXx, ... //MUST match order in Map_Vertex_Output + unBit = 1 << FRAG_ATTRIB_WPOS; + if(mesa_fp->Base.InputsRead & unBit) + { + pAsm->uiFP_AttributeMap[FRAG_ATTRIB_WPOS] = pAsm->number_used_registers++; + } + unBit = 1 << FRAG_ATTRIB_COL0; if(mesa_fp->Base.InputsRead & unBit) { @@ -67,6 +73,12 @@ void Map_Fragment_Program(r700_AssemblerBase *pAsm, pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL1] = pAsm->number_used_registers++; } + unBit = 1 << FRAG_ATTRIB_FOGC; + if(mesa_fp->Base.InputsRead & unBit) + { + pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FOGC] = pAsm->number_used_registers++; + } + for(i=0; i<8; i++) { unBit = 1 << (FRAG_ATTRIB_TEX0 + i); @@ -106,6 +118,7 @@ void Map_Fragment_Program(r700_AssemblerBase *pAsm, pAsm->uiFP_OutputMap[FRAG_RESULT_DEPTH] = pAsm->number_used_registers++; pAsm->number_of_exports++; pAsm->number_of_colorandz_exports++; + pAsm->pR700Shader->depthIsExported = 1; } pAsm->pucOutMask = (unsigned char*) MALLOC(pAsm->number_of_exports); @@ -255,20 +268,19 @@ void * r700GetActiveFpShaderBo(GLcontext * ctx) GLboolean r700SetupFragmentProgram(GLcontext * ctx) { - context_t *context = R700_CONTEXT(ctx); - BATCH_LOCALS(&context->radeon); - + context_t *context = R700_CONTEXT(ctx); R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); - + BATCH_LOCALS(&context->radeon); struct r700_fragment_program *fp = (struct r700_fragment_program *) (ctx->FragmentProgram._Current); - + r700_AssemblerBase *pAsm = &(fp->r700AsmCode); + struct gl_fragment_program *mesa_fp = &(fp->mesa_program); struct gl_program_parameter_list *paramList; unsigned int unNumParamData; - unsigned int ui; - + unsigned int ui, i; unsigned int unNumOfReg; - + unsigned int unBit; + if(GL_FALSE == fp->loaded) { if(fp->r700Shader.bNeedsAssembly == GL_TRUE) @@ -277,11 +289,11 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) } /* Load fp to gpu */ - r600EmitShader(ctx, - &(fp->shaderbo), + r600EmitShader(ctx, + &(fp->shaderbo), (GLvoid *)(fp->r700Shader.pProgram), fp->r700Shader.uShaderBinaryDWORDSize, - "FS"); + "FS"); fp->loaded = GL_TRUE; } @@ -299,7 +311,17 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) ui = (r700->SPI_PS_IN_CONTROL_0.u32All & NUM_INTERP_mask) / (1 << NUM_INTERP_shift); - ui = ui ? ui : unNumOfReg; + /* PS uses fragment.position */ + if (mesa_fp->Base.InputsRead & (1 << FRAG_ATTRIB_WPOS)) + { + ui += 1; + SETfield(r700->SPI_PS_IN_CONTROL_0.u32All, ui, NUM_INTERP_shift, NUM_INTERP_mask); + SETfield(r700->SPI_PS_IN_CONTROL_0.u32All, CENTERS_ONLY, BARYC_SAMPLE_CNTL_shift, BARYC_SAMPLE_CNTL_mask); + SETbit(r700->SPI_PS_IN_CONTROL_0.u32All, POSITION_ENA_bit); + SETbit(r700->SPI_INPUT_Z.u32All, PROVIDE_Z_TO_SPI_bit); + } + + ui = (unNumOfReg < ui) ? ui : unNumOfReg; SETfield(r700->ps.SQ_PGM_RESOURCES_PS.u32All, ui, NUM_GPRS_shift, NUM_GPRS_mask); @@ -333,7 +355,6 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) } /* sent out shader constants. */ - paramList = fp->mesa_program.Base.Parameters; if(NULL != paramList) @@ -343,7 +364,7 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) unNumParamData = paramList->NumParameters * 4; BEGIN_BATCH_NO_AUTOSTATE(2 + unNumParamData); - + R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_ALU_CONST, unNumParamData)); /* assembler map const from very beginning. */ @@ -362,8 +383,72 @@ GLboolean r700SetupFragmentProgram(GLcontext * ctx) COMMIT_BATCH(); } - return GL_TRUE; -} + // emit ps input map + unBit = 1 << FRAG_ATTRIB_WPOS; + if(mesa_fp->Base.InputsRead & unBit) + { + ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_WPOS]; + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); + SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui, + SEMANTIC_shift, SEMANTIC_mask); + if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit) + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + else + CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + } + + unBit = 1 << FRAG_ATTRIB_COL0; + if(mesa_fp->Base.InputsRead & unBit) + { + ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL0]; + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); + SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui, + SEMANTIC_shift, SEMANTIC_mask); + if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit) + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + else + CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + } + + unBit = 1 << FRAG_ATTRIB_COL1; + if(mesa_fp->Base.InputsRead & unBit) + { + ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL1]; + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); + SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui, + SEMANTIC_shift, SEMANTIC_mask); + if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit) + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + else + CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + } + unBit = 1 << FRAG_ATTRIB_FOGC; + if(mesa_fp->Base.InputsRead & unBit) + { + ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_FOGC]; + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); + SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui, + SEMANTIC_shift, SEMANTIC_mask); + if (r700->SPI_INTERP_CONTROL_0.u32All & FLAT_SHADE_ENA_bit) + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + else + CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + } + + for(i=0; i<8; i++) + { + unBit = 1 << (FRAG_ATTRIB_TEX0 + i); + if(mesa_fp->Base.InputsRead & unBit) + { + ui = pAsm->uiFP_AttributeMap[FRAG_ATTRIB_TEX0 + i]; + SETbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, SEL_CENTROID_bit); + SETfield(r700->SPI_PS_INPUT_CNTL[ui].u32All, ui, + SEMANTIC_shift, SEMANTIC_mask); + CLEARbit(r700->SPI_PS_INPUT_CNTL[ui].u32All, FLAT_SHADE_bit); + } + } + return GL_TRUE; +} diff --git a/src/mesa/drivers/dri/r600/r700_ioctl.c b/src/mesa/drivers/dri/r600/r700_ioctl.c index c479532001..23cc128d6d 100644 --- a/src/mesa/drivers/dri/r600/r700_ioctl.c +++ b/src/mesa/drivers/dri/r600/r700_ioctl.c @@ -43,7 +43,6 @@ static void r700Flush(GLcontext *ctx) { radeonContextPtr radeon = RADEON_CONTEXT(ctx); - context_t * context = R700_CONTEXT(ctx); if (RADEON_DEBUG & DEBUG_IOCTL) fprintf(stderr, "%s %d\n", __FUNCTION__, radeon->cmdbuf.cs->cdw); @@ -59,7 +58,7 @@ static void r700Flush(GLcontext *ctx) if (radeon->dma.flush) radeon->dma.flush( ctx ); - r700SendContextStates(context); + r700EmitState(ctx); if (radeon->cmdbuf.cs->cdw) rcommonFlushCmdBuf(radeon, __FUNCTION__); diff --git a/src/mesa/drivers/dri/r600/r700_oglprog.c b/src/mesa/drivers/dri/r600/r700_oglprog.c index 36de143b1a..c49b90c1cc 100644 --- a/src/mesa/drivers/dri/r600/r700_oglprog.c +++ b/src/mesa/drivers/dri/r600/r700_oglprog.c @@ -33,6 +33,7 @@ #include "tnl/tnl.h" #include "r600_context.h" +#include "r600_emit.h" #include "r700_oglprog.h" #include "r700_fragprog.h" @@ -87,7 +88,6 @@ static void r700DeleteProgram(GLcontext * ctx, struct gl_program *prog) { struct r700_vertex_program * vp; struct r700_fragment_program * fp; - context_t *context = R700_CONTEXT(ctx); switch (prog->Target) { diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c index f1e467a317..6985bd4ffa 100644 --- a/src/mesa/drivers/dri/r600/r700_render.c +++ b/src/mesa/drivers/dri/r600/r700_render.c @@ -44,7 +44,6 @@ #include "tnl/t_vertex.h" #include "tnl/t_pipeline.h" -#include "radeon_mipmap_tree.h" #include "r600_context.h" #include "r600_cmdbuf.h" @@ -58,7 +57,7 @@ void r700WaitForIdle(context_t *context); void r700WaitForIdleClean(context_t *context); void r700Start3D(context_t *context); GLboolean r700SendTextureState(context_t *context); -unsigned int r700PrimitiveType(int prim); +static unsigned int r700PrimitiveType(int prim); void r600UpdateTextureState(GLcontext * ctx); GLboolean r700SyncSurf(context_t *context, struct radeon_bo *pbo, @@ -138,59 +137,15 @@ static GLboolean r700SetupShaders(GLcontext * ctx) exportCount = (r700->ps.SQ_PGM_EXPORTS_PS.u32All & EXPORT_MODE_mask) / (1 << EXPORT_MODE_shift); r700->CB_SHADER_CONTROL.u32All = (1 << exportCount) - 1; - return GL_TRUE; -} + r600UpdateTextureState(ctx); -GLboolean r700SendTextureState(context_t *context) -{ - unsigned int i; - R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); - offset_modifiers offset_mod = {NO_SHIFT, 0, 0xFFFFFFFF}; - struct radeon_bo *bo = NULL; - BATCH_LOCALS(&context->radeon); + r700SendFSState(context); // FIXME just a place holder for now + r700SendPSState(context); + r700SendVSState(context); + + r700SendTextureState(context); + r700SetupStreams(ctx); - for (i=0; i<R700_TEXTURE_NUMBERUNITS; i++) { - radeonTexObj *t = r700->textures[i]; - if (t) { - if (!t->image_override) - bo = t->mt->bo; - else - bo = t->bo; - if (bo) { - - r700SyncSurf(context, bo, - RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, - 0, TC_ACTION_ENA_bit); - - BEGIN_BATCH_NO_AUTOSTATE(9); - R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_RESOURCE, 7)); - R600_OUT_BATCH(i * 7); - R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE0); - R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE1); - R600_OUT_BATCH_RELOC(r700->textures[i]->SQ_TEX_RESOURCE2, - bo, - 0, - RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0, &offset_mod); - R600_OUT_BATCH_RELOC(r700->textures[i]->SQ_TEX_RESOURCE3, - bo, - 0, - RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0, &offset_mod); - R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE4); - R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE5); - R600_OUT_BATCH(r700->textures[i]->SQ_TEX_RESOURCE6); - END_BATCH(); - - BEGIN_BATCH_NO_AUTOSTATE(5); - R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_SAMPLER, 3)); - R600_OUT_BATCH(i * 3); - R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER0); - R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER1); - R600_OUT_BATCH(r700->textures[i]->SQ_TEX_SAMPLER2); - END_BATCH(); - COMMIT_BATCH(); - } - } - } return GL_TRUE; } @@ -202,17 +157,15 @@ GLboolean r700SyncSurf(context_t *context, { BATCH_LOCALS(&context->radeon); uint32_t cp_coher_size; - offset_modifiers offset_mod; + + if (!pbo) + return GL_FALSE; if (pbo->size == 0xffffffff) cp_coher_size = 0xffffffff; else cp_coher_size = ((pbo->size + 255) >> 8); - offset_mod.shift = NO_SHIFT; - offset_mod.shiftbits = 0; - offset_mod.mask = 0xFFFFFFFF; - BEGIN_BATCH_NO_AUTOSTATE(5); R600_OUT_BATCH(CP_PACKET3(R600_IT_SURFACE_SYNC, 3)); R600_OUT_BATCH(sync_type); @@ -220,7 +173,7 @@ GLboolean r700SyncSurf(context_t *context, R600_OUT_BATCH_RELOC(0, pbo, 0, - read_domain, write_domain, 0, &offset_mod); // ??? + read_domain, write_domain, 0); // ??? R600_OUT_BATCH(10); END_BATCH(); @@ -229,7 +182,7 @@ GLboolean r700SyncSurf(context_t *context, return GL_TRUE; } -unsigned int r700PrimitiveType(int prim) +static unsigned int r700PrimitiveType(int prim) { switch (prim & PRIM_MODE_MASK) { @@ -270,119 +223,163 @@ unsigned int r700PrimitiveType(int prim) } } -static GLboolean r700RunRender(GLcontext * ctx, - struct tnl_pipeline_stage *stage) +static int r700NumVerts(int num_verts, int prim) { - context_t *context = R700_CONTEXT(ctx); - R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); - int lastIndex = 0; -#if 1 - BATCH_LOCALS(&context->radeon); - - unsigned int i, j; - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *vb = &tnl->vb; - - struct r700_fragment_program *fp = (struct r700_fragment_program *) - (ctx->FragmentProgram._Current); - if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770) - { - fp->r700AsmCode.bR6xx = 1; - } - - r700Start3D(context); /* TODO : this is too much. */ + int verts_off = 0; + + switch (prim & PRIM_MODE_MASK) { + case GL_POINTS: + verts_off = 0; + break; + case GL_LINES: + verts_off = num_verts % 2; + break; + case GL_LINE_STRIP: + if (num_verts < 2) + verts_off = num_verts; + break; + case GL_LINE_LOOP: + if (num_verts < 2) + verts_off = num_verts; + break; + case GL_TRIANGLES: + verts_off = num_verts % 3; + break; + case GL_TRIANGLE_STRIP: + if (num_verts < 3) + verts_off = num_verts; + break; + case GL_TRIANGLE_FAN: + if (num_verts < 3) + verts_off = num_verts; + break; + case GL_QUADS: + verts_off = num_verts % 4; + break; + case GL_QUAD_STRIP: + if (num_verts < 4) + verts_off = num_verts; + else + verts_off = num_verts % 2; + break; + case GL_POLYGON: + if (num_verts < 3) + verts_off = num_verts; + break; + default: + assert(0); + return -1; + break; + } + + return num_verts - verts_off; +} - r700SendSQConfig(context); +static void r700RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim) +{ + context_t *context = R700_CONTEXT(ctx); + BATCH_LOCALS(&context->radeon); + int type, i, total_emit; + int num_indices; + uint32_t vgt_draw_initiator = 0; + uint32_t vgt_index_type = 0; + uint32_t vgt_primitive_type = 0; + uint32_t vgt_num_indices = 0; + + type = r700PrimitiveType(prim); + num_indices = r700NumVerts(end - start, prim); + + if (type < 0 || num_indices <= 0) + return; + + total_emit = 3 /* VGT_PRIMITIVE_TYPE */ + + 2 /* VGT_INDEX_TYPE */ + + 2 /* NUM_INSTANCES */ + + num_indices + 3; /* DRAW_INDEX_IMMD */ + + BEGIN_BATCH_NO_AUTOSTATE(total_emit); + // prim + SETfield(vgt_primitive_type, type, + VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift, VGT_PRIMITIVE_TYPE__PRIM_TYPE_mask); + R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1)); + R600_OUT_BATCH(mmVGT_PRIMITIVE_TYPE - ASIC_CONFIG_BASE_INDEX); + R600_OUT_BATCH(vgt_primitive_type); - r700UpdateShaders(ctx); + // index type + SETfield(vgt_index_type, DI_INDEX_SIZE_32_BIT, INDEX_TYPE_shift, INDEX_TYPE_mask); + R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0)); + R600_OUT_BATCH(vgt_index_type); - r700SetScissor(context); - r700SetRenderTarget(context, 0); - r700SetDepthTarget(context); + // num instances + R600_OUT_BATCH(CP_PACKET3(R600_IT_NUM_INSTANCES, 0)); + R600_OUT_BATCH(1); - if(r700SetupStreams(ctx)) - { - return GL_TRUE; - } + // draw packet + vgt_num_indices = num_indices; + SETfield(vgt_draw_initiator, DI_SRC_SEL_IMMEDIATE, SOURCE_SELECT_shift, SOURCE_SELECT_mask); + SETfield(vgt_draw_initiator, DI_MAJOR_MODE_0, MAJOR_MODE_shift, MAJOR_MODE_mask); - r600UpdateTextureState(ctx); - r700SendTextureState(context); + R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_IMMD, (num_indices + 1))); + R600_OUT_BATCH(vgt_num_indices); + R600_OUT_BATCH(vgt_draw_initiator); - if(GL_FALSE == fp->translated) - { - if( GL_FALSE == r700TranslateFragmentShader(fp, &(fp->mesa_program)) ) - { - return GL_TRUE; + for (i = start; i < (start + num_indices); i++) { + R600_OUT_BATCH(i); } - } + END_BATCH(); + COMMIT_BATCH(); - r700SetupShaders(ctx); +} - r700SendFSState(context); // FIXME just a place holder for now - r700SendPSState(context); - r700SendVSState(context); +void r700EmitState(GLcontext * ctx) +{ + context_t *context = R700_CONTEXT(ctx); + radeonContextPtr radeon = &context->radeon; - r700SendContextStates(context); - r700SendViewportState(context, 0); - r700SendRenderTargetState(context, 0); - r700SendDepthTargetState(context); + if (radeon->cmdbuf.cs->cdw && !radeon->hw.is_dirty && !radeon->hw.all_dirty) + return; - /* richard test code */ - for (i = 0; i < vb->PrimitiveCount; i++) - { - GLuint prim = _tnl_translate_prim(&vb->Primitive[i]); - GLuint start = vb->Primitive[i].start; - GLuint end = vb->Primitive[i].start + vb->Primitive[i].count; - GLuint numIndices = vb->Primitive[i].count; - GLuint numEntires; - - unsigned int VGT_DRAW_INITIATOR = 0; - unsigned int VGT_INDEX_TYPE = 0; - unsigned int VGT_PRIMITIVE_TYPE = 0; - unsigned int VGT_NUM_INDICES = 0; - - numEntires = 2 /* VGT_INDEX_TYPE */ - + 3 /* VGT_PRIMITIVE_TYPE */ - + numIndices + 3; /* DRAW_INDEX_IMMD */ - - BEGIN_BATCH_NO_AUTOSTATE(numEntires); - - VGT_INDEX_TYPE |= DI_INDEX_SIZE_32_BIT << INDEX_TYPE_shift; + rcommonEnsureCmdBufSpace(&context->radeon, + context->radeon.hw.max_state_size, __FUNCTION__); - R600_OUT_BATCH(CP_PACKET3(R600_IT_INDEX_TYPE, 0)); - R600_OUT_BATCH(VGT_INDEX_TYPE); + r700SendSQConfig(context); - VGT_NUM_INDICES = numIndices; + r700SendUCPState(context); + r700SendContextStates(context); + r700SendViewportState(context, 0); + r700SendRenderTargetState(context, 0); + r700SendDepthTargetState(context); - VGT_PRIMITIVE_TYPE |= r700PrimitiveType(prim) << VGT_PRIMITIVE_TYPE__PRIM_TYPE_shift; - R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1)); - R600_OUT_BATCH(mmVGT_PRIMITIVE_TYPE - ASIC_CONFIG_BASE_INDEX); - R600_OUT_BATCH(VGT_PRIMITIVE_TYPE); +} - VGT_DRAW_INITIATOR |= DI_SRC_SEL_IMMEDIATE << SOURCE_SELECT_shift; - VGT_DRAW_INITIATOR |= DI_MAJOR_MODE_0 << MAJOR_MODE_shift; +static GLboolean r700RunRender(GLcontext * ctx, + struct tnl_pipeline_stage *stage) +{ + context_t *context = R700_CONTEXT(ctx); + unsigned int i; + TNLcontext *tnl = TNL_CONTEXT(ctx); + struct vertex_buffer *vb = &tnl->vb; - R600_OUT_BATCH(CP_PACKET3(R600_IT_DRAW_INDEX_IMMD, (numIndices + 1))); - R600_OUT_BATCH(VGT_NUM_INDICES); - R600_OUT_BATCH(VGT_DRAW_INITIATOR); + r700Start3D(context); - for (j = lastIndex; j < lastIndex + numIndices; j++) - { - R600_OUT_BATCH(j); - } - lastIndex += numIndices; + r700UpdateShaders(ctx); + r700SetScissor(context); + r700SetupShaders(ctx); - END_BATCH(); - COMMIT_BATCH(); + r700EmitState(ctx); + + /* richard test code */ + for (i = 0; i < vb->PrimitiveCount; i++) { + GLuint prim = _tnl_translate_prim(&vb->Primitive[i]); + GLuint start = vb->Primitive[i].start; + GLuint end = vb->Primitive[i].start + vb->Primitive[i].count; + r700RunRenderPrimitive(ctx, start, end, prim); } /* Flush render op cached for last several quads. */ r700WaitForIdleClean(context); - radeonReleaseArrays(ctx, 0); - -#endif //0 - rcommonFlushCmdBuf( &context->radeon, __FUNCTION__ ); + radeonReleaseArrays(ctx, ~0); return GL_FALSE; } @@ -410,10 +407,6 @@ static GLboolean r700RunTCLRender(GLcontext * ctx, /*----------------------*/ return GL_TRUE; } - context_t *context = R700_CONTEXT(ctx); - - r700UpdateShaders(ctx); - bRet = r700RunRender(ctx, stage); return bRet; diff --git a/src/mesa/drivers/dri/r600/r700_state.c b/src/mesa/drivers/dri/r600/r700_state.c index eda224def3..835b5e18c2 100644 --- a/src/mesa/drivers/dri/r600/r700_state.c +++ b/src/mesa/drivers/dri/r600/r700_state.c @@ -30,6 +30,7 @@ #include "main/imports.h" #include "main/enums.h" #include "main/macros.h" +#include "main/context.h" #include "main/dd.h" #include "main/simple_list.h" @@ -55,6 +56,13 @@ #include "r700_vertprog.h" +static void r700SetClipPlaneState(GLcontext * ctx, GLenum cap, GLboolean state); +static void r700UpdatePolygonMode(GLcontext * ctx); +static void r700SetPolygonOffsetState(GLcontext * ctx, GLboolean state); +static void r700SetStencilState(GLcontext * ctx, GLboolean state); +static void r700SetRenderTarget(context_t *context, int id); +static void r700SetDepthTarget(context_t *context); + void r700SetDefaultStates(context_t *context) //-------------------- { @@ -66,12 +74,28 @@ void r700UpdateShaders (GLcontext * ctx) //---------------------------------- GLvector4f dummy_attrib[_TNL_ATTRIB_MAX]; GLvector4f *temp_attrib[_TNL_ATTRIB_MAX]; - - struct r700_vertex_program *vp; - int i; + int i; + + if (ctx->FragmentProgram._Current) { + struct r700_fragment_program *fp = (struct r700_fragment_program *) + (ctx->FragmentProgram._Current); + if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770) + { + fp->r700AsmCode.bR6xx = 1; + } + + if(GL_FALSE == fp->translated) + { + if( GL_FALSE == r700TranslateFragmentShader(fp, &(fp->mesa_program)) ) + { + //return GL_TRUE; + } + } + } if (context->radeon.NewGLState) { + struct r700_vertex_program *vp; context->radeon.NewGLState = 0; for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) @@ -113,10 +137,21 @@ void r700UpdateShaders (GLcontext * ctx) //---------------------------------- */ void r700UpdateViewportOffset(GLcontext * ctx) //------------------ { + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&context->radeon); + GLfloat xoffset = (GLfloat) dPriv->x; + GLfloat yoffset = (GLfloat) dPriv->y + dPriv->h; + const GLfloat *v = ctx->Viewport._WindowMap.m; + int id = 0; - //radeonUpdateScissor(ctx); + GLfloat tx = v[MAT_TX] + xoffset; + GLfloat ty = (-v[MAT_TY]) + yoffset; - return; + r700->viewport[id].PA_CL_VPORT_XOFFSET.f32All = tx; + r700->viewport[id].PA_CL_VPORT_YOFFSET.f32All = ty; + + radeonUpdateScissor(ctx); } /** @@ -125,29 +160,16 @@ void r700UpdateViewportOffset(GLcontext * ctx) //------------------ */ void r700UpdateDrawBuffer(GLcontext * ctx) /* TODO */ //--------------------- { -#if 0 /* to be enabled */ - context_t *context = R700_CONTEXT(ctx); + context_t *context = R700_CONTEXT(ctx); - switch (ctx->DrawBuffer->_ColorDrawBufferIndexes[0]) - { - case BUFFER_FRONT_LEFT: - context->target.rt = context->screen->frontBuffer; - break; - case BUFFER_BACK_LEFT: - context->target.rt = context->screen->backBuffer; - break; - default: - memset (&context->target.rt, sizeof(context->target.rt), 0); - } -#endif /* to be enabled */ + r700SetRenderTarget(context, 0); + r700SetDepthTarget(context); } static void r700FetchStateParameter(GLcontext * ctx, const gl_state_index state[STATE_LENGTH], GLfloat * value) { - context_t *context = R700_CONTEXT(ctx); - /* TODO */ } @@ -257,40 +279,40 @@ static void r700SetDepthState(GLcontext * ctx) switch (ctx->Depth.Func) { - case GL_NEVER: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_NEVER, + case GL_NEVER: + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_NEVER, ZFUNC_shift, ZFUNC_mask); break; case GL_LESS: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_LESS, - ZFUNC_shift, ZFUNC_mask); + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_LESS, + ZFUNC_shift, ZFUNC_mask); break; case GL_EQUAL: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_EQUAL, + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_EQUAL, ZFUNC_shift, ZFUNC_mask); break; case GL_LEQUAL: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_LEQUAL, + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_LEQUAL, ZFUNC_shift, ZFUNC_mask); break; case GL_GREATER: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_GREATER, - ZFUNC_shift, ZFUNC_mask); + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_GREATER, + ZFUNC_shift, ZFUNC_mask); break; case GL_NOTEQUAL: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_NOTEQUAL, + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_NOTEQUAL, ZFUNC_shift, ZFUNC_mask); break; case GL_GEQUAL: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_GEQUAL, + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_GEQUAL, ZFUNC_shift, ZFUNC_mask); break; case GL_ALWAYS: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_ALWAYS, + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_ALWAYS, ZFUNC_shift, ZFUNC_mask); break; default: - SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_ALWAYS, + SETfield(r700->DB_DEPTH_CONTROL.u32All, FRAG_ALWAYS, ZFUNC_shift, ZFUNC_mask); break; } @@ -302,24 +324,336 @@ static void r700SetDepthState(GLcontext * ctx) } } +static void r700SetAlphaState(GLcontext * ctx) +{ + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + uint32_t alpha_func = REF_ALWAYS; + GLboolean really_enabled = ctx->Color.AlphaEnabled; + + switch (ctx->Color.AlphaFunc) { + case GL_NEVER: + alpha_func = REF_NEVER; + break; + case GL_LESS: + alpha_func = REF_LESS; + break; + case GL_EQUAL: + alpha_func = REF_EQUAL; + break; + case GL_LEQUAL: + alpha_func = REF_LEQUAL; + break; + case GL_GREATER: + alpha_func = REF_GREATER; + break; + case GL_NOTEQUAL: + alpha_func = REF_NOTEQUAL; + break; + case GL_GEQUAL: + alpha_func = REF_GEQUAL; + break; + case GL_ALWAYS: + /*alpha_func = REF_ALWAYS; */ + really_enabled = GL_FALSE; + break; + } + + if (really_enabled) { + SETfield(r700->SX_ALPHA_TEST_CONTROL.u32All, alpha_func, + ALPHA_FUNC_shift, ALPHA_FUNC_mask); + SETbit(r700->SX_ALPHA_TEST_CONTROL.u32All, ALPHA_TEST_ENABLE_bit); + r700->SX_ALPHA_REF.f32All = ctx->Color.AlphaRef; + } else { + CLEARbit(r700->SX_ALPHA_TEST_CONTROL.u32All, ALPHA_TEST_ENABLE_bit); + } + +} + static void r700AlphaFunc(GLcontext * ctx, GLenum func, GLfloat ref) //--------------- { + (void)func; + (void)ref; + r700SetAlphaState(ctx); } static void r700BlendColor(GLcontext * ctx, const GLfloat cf[4]) //---------------- { + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + + r700->CB_BLEND_RED.f32All = cf[0]; + r700->CB_BLEND_GREEN.f32All = cf[1]; + r700->CB_BLEND_BLUE.f32All = cf[2]; + r700->CB_BLEND_ALPHA.f32All = cf[3]; +} + +static int blend_factor(GLenum factor, GLboolean is_src) +{ + switch (factor) { + case GL_ZERO: + return BLEND_ZERO; + break; + case GL_ONE: + return BLEND_ONE; + break; + case GL_DST_COLOR: + return BLEND_DST_COLOR; + break; + case GL_ONE_MINUS_DST_COLOR: + return BLEND_ONE_MINUS_DST_COLOR; + break; + case GL_SRC_COLOR: + return BLEND_SRC_COLOR; + break; + case GL_ONE_MINUS_SRC_COLOR: + return BLEND_ONE_MINUS_SRC_COLOR; + break; + case GL_SRC_ALPHA: + return BLEND_SRC_ALPHA; + break; + case GL_ONE_MINUS_SRC_ALPHA: + return BLEND_ONE_MINUS_SRC_ALPHA; + break; + case GL_DST_ALPHA: + return BLEND_DST_ALPHA; + break; + case GL_ONE_MINUS_DST_ALPHA: + return BLEND_ONE_MINUS_DST_ALPHA; + break; + case GL_SRC_ALPHA_SATURATE: + return (is_src) ? BLEND_SRC_ALPHA_SATURATE : BLEND_ZERO; + break; + case GL_CONSTANT_COLOR: + return BLEND_CONSTANT_COLOR; + break; + case GL_ONE_MINUS_CONSTANT_COLOR: + return BLEND_ONE_MINUS_CONSTANT_COLOR; + break; + case GL_CONSTANT_ALPHA: + return BLEND_CONSTANT_ALPHA; + break; + case GL_ONE_MINUS_CONSTANT_ALPHA: + return BLEND_ONE_MINUS_CONSTANT_ALPHA; + break; + default: + fprintf(stderr, "unknown blend factor %x\n", factor); + return (is_src) ? BLEND_ONE : BLEND_ZERO; + break; + } +} + +static void r700SetBlendState(GLcontext * ctx) +{ + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + int id = 0; + uint32_t blend_reg = 0, eqn, eqnA; + + if (RGBA_LOGICOP_ENABLED(ctx) || !ctx->Color.BlendEnabled) { + SETfield(blend_reg, + BLEND_ONE, COLOR_SRCBLEND_shift, COLOR_SRCBLEND_mask); + SETfield(blend_reg, + BLEND_ZERO, COLOR_DESTBLEND_shift, COLOR_DESTBLEND_mask); + SETfield(blend_reg, + COMB_DST_PLUS_SRC, COLOR_COMB_FCN_shift, COLOR_COMB_FCN_mask); + SETfield(blend_reg, + BLEND_ONE, ALPHA_SRCBLEND_shift, ALPHA_SRCBLEND_mask); + SETfield(blend_reg, + BLEND_ZERO, ALPHA_DESTBLEND_shift, ALPHA_DESTBLEND_mask); + SETfield(blend_reg, + COMB_DST_PLUS_SRC, ALPHA_COMB_FCN_shift, ALPHA_COMB_FCN_mask); + if (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_R600) + r700->CB_BLEND_CONTROL.u32All = blend_reg; + else + r700->render_target[id].CB_BLEND0_CONTROL.u32All = blend_reg; + return; + } + + SETfield(blend_reg, + blend_factor(ctx->Color.BlendSrcRGB, GL_TRUE), + COLOR_SRCBLEND_shift, COLOR_SRCBLEND_mask); + SETfield(blend_reg, + blend_factor(ctx->Color.BlendDstRGB, GL_FALSE), + COLOR_DESTBLEND_shift, COLOR_DESTBLEND_mask); + + switch (ctx->Color.BlendEquationRGB) { + case GL_FUNC_ADD: + eqn = COMB_DST_PLUS_SRC; + break; + case GL_FUNC_SUBTRACT: + eqn = COMB_SRC_MINUS_DST; + break; + case GL_FUNC_REVERSE_SUBTRACT: + eqn = COMB_DST_MINUS_SRC; + break; + case GL_MIN: + eqn = COMB_MIN_DST_SRC; + SETfield(blend_reg, + BLEND_ONE, + COLOR_SRCBLEND_shift, COLOR_SRCBLEND_mask); + SETfield(blend_reg, + BLEND_ONE, + COLOR_DESTBLEND_shift, COLOR_DESTBLEND_mask); + break; + case GL_MAX: + eqn = COMB_MAX_DST_SRC; + SETfield(blend_reg, + BLEND_ONE, + COLOR_SRCBLEND_shift, COLOR_SRCBLEND_mask); + SETfield(blend_reg, + BLEND_ONE, + COLOR_DESTBLEND_shift, COLOR_DESTBLEND_mask); + break; + + default: + fprintf(stderr, + "[%s:%u] Invalid RGB blend equation (0x%04x).\n", + __FUNCTION__, __LINE__, ctx->Color.BlendEquationRGB); + return; + } + SETfield(blend_reg, + eqn, COLOR_COMB_FCN_shift, COLOR_COMB_FCN_mask); + + SETfield(blend_reg, + blend_factor(ctx->Color.BlendSrcRGB, GL_TRUE), + ALPHA_SRCBLEND_shift, ALPHA_SRCBLEND_mask); + SETfield(blend_reg, + blend_factor(ctx->Color.BlendDstRGB, GL_FALSE), + ALPHA_DESTBLEND_shift, ALPHA_DESTBLEND_mask); + + switch (ctx->Color.BlendEquationA) { + case GL_FUNC_ADD: + eqnA = COMB_DST_PLUS_SRC; + break; + case GL_FUNC_SUBTRACT: + eqnA = COMB_SRC_MINUS_DST; + break; + case GL_FUNC_REVERSE_SUBTRACT: + eqnA = COMB_DST_MINUS_SRC; + break; + case GL_MIN: + eqnA = COMB_MIN_DST_SRC; + SETfield(blend_reg, + BLEND_ONE, + ALPHA_SRCBLEND_shift, ALPHA_SRCBLEND_mask); + SETfield(blend_reg, + BLEND_ONE, + ALPHA_DESTBLEND_shift, ALPHA_DESTBLEND_mask); + break; + case GL_MAX: + eqnA = COMB_MAX_DST_SRC; + SETfield(blend_reg, + BLEND_ONE, + ALPHA_SRCBLEND_shift, ALPHA_SRCBLEND_mask); + SETfield(blend_reg, + BLEND_ONE, + ALPHA_DESTBLEND_shift, ALPHA_DESTBLEND_mask); + break; + default: + fprintf(stderr, + "[%s:%u] Invalid A blend equation (0x%04x).\n", + __FUNCTION__, __LINE__, ctx->Color.BlendEquationA); + return; + } + + SETfield(blend_reg, + eqnA, ALPHA_COMB_FCN_shift, ALPHA_COMB_FCN_mask); + + SETbit(blend_reg, SEPARATE_ALPHA_BLEND_bit); + + if (context->radeon.radeonScreen->chip_family == CHIP_FAMILY_R600) + r700->CB_BLEND_CONTROL.u32All = blend_reg; + else { + r700->render_target[id].CB_BLEND0_CONTROL.u32All = blend_reg; + SETbit(r700->CB_COLOR_CONTROL.u32All, PER_MRT_BLEND_bit); + } + SETfield(r700->CB_COLOR_CONTROL.u32All, (1 << id), + TARGET_BLEND_ENABLE_shift, TARGET_BLEND_ENABLE_mask); + } static void r700BlendEquationSeparate(GLcontext * ctx, GLenum modeRGB, GLenum modeA) //----------------- { + r700SetBlendState(ctx); } static void r700BlendFuncSeparate(GLcontext * ctx, GLenum sfactorRGB, GLenum dfactorRGB, GLenum sfactorA, GLenum dfactorA) //------------------------ { + r700SetBlendState(ctx); +} + +/** + * Translate LogicOp enums into hardware representation. + */ +static GLuint translate_logicop(GLenum logicop) +{ + switch (logicop) { + case GL_CLEAR: + return 0x00; + case GL_SET: + return 0xff; + case GL_COPY: + return 0xcc; + case GL_COPY_INVERTED: + return 0x33; + case GL_NOOP: + return 0xaa; + case GL_INVERT: + return 0x55; + case GL_AND: + return 0x88; + case GL_NAND: + return 0x77; + case GL_OR: + return 0xee; + case GL_NOR: + return 0x11; + case GL_XOR: + return 0x66; + case GL_EQUIV: + return 0xaa; + case GL_AND_REVERSE: + return 0x44; + case GL_AND_INVERTED: + return 0x22; + case GL_OR_REVERSE: + return 0xdd; + case GL_OR_INVERTED: + return 0xbb; + default: + fprintf(stderr, "unknown blend logic operation %x\n", logicop); + return 0xcc; + } +} + +/** + * Used internally to update the r300->hw hardware state to match the + * current OpenGL state. + */ +static void r700SetLogicOpState(GLcontext *ctx) +{ + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&R700_CONTEXT(ctx)->hw); + + if (RGBA_LOGICOP_ENABLED(ctx)) + SETfield(r700->CB_COLOR_CONTROL.u32All, + translate_logicop(ctx->Color.LogicOp), ROP3_shift, ROP3_mask); + else + SETfield(r700->CB_COLOR_CONTROL.u32All, 0xCC, ROP3_shift, ROP3_mask); +} + +/** + * Called by Mesa when an application program changes the LogicOp state + * via glLogicOp. + */ +static void r700LogicOpcode(GLcontext *ctx, GLenum logicop) +{ + if (RGBA_LOGICOP_ENABLED(ctx)) + r700SetLogicOpState(ctx); } static void r700UpdateCulling(GLcontext * ctx) @@ -330,11 +664,11 @@ static void r700UpdateCulling(GLcontext * ctx) CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, CULL_FRONT_bit); CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, CULL_BACK_bit); - if (ctx->Polygon.CullFlag) + if (ctx->Polygon.CullFlag) { - switch (ctx->Polygon.CullFaceMode) + switch (ctx->Polygon.CullFaceMode) { - case GL_FRONT: + case GL_FRONT: SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, CULL_FRONT_bit); CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, CULL_BACK_bit); break; @@ -353,13 +687,13 @@ static void r700UpdateCulling(GLcontext * ctx) } } - switch (ctx->Polygon.FrontFace) + switch (ctx->Polygon.FrontFace) { case GL_CW: SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, FACE_bit); break; case GL_CCW: - CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, FACE_bit); + CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, FACE_bit); break; default: CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, FACE_bit); /* default: ccw */ @@ -394,13 +728,13 @@ static void r700Enable(GLcontext * ctx, GLenum cap, GLboolean state) //--------- /* empty */ break; case GL_ALPHA_TEST: - //r700SetAlphaState(ctx); + r700SetAlphaState(ctx); break; case GL_COLOR_LOGIC_OP: - //r700SetLogicOpState(ctx); + r700SetLogicOpState(ctx); /* fall-through, because logic op overrides blending */ case GL_BLEND: - //r700SetBlendState(ctx); + r700SetBlendState(ctx); break; case GL_CLIP_PLANE0: case GL_CLIP_PLANE1: @@ -408,13 +742,13 @@ static void r700Enable(GLcontext * ctx, GLenum cap, GLboolean state) //--------- case GL_CLIP_PLANE3: case GL_CLIP_PLANE4: case GL_CLIP_PLANE5: - //r700SetClipPlaneState(ctx, cap, state); + r700SetClipPlaneState(ctx, cap, state); break; case GL_DEPTH_TEST: r700SetDepthState(ctx); break; case GL_STENCIL_TEST: - //r700SetStencilState(ctx, state); + r700SetStencilState(ctx, state); break; case GL_CULL_FACE: r700UpdateCulling(ctx); @@ -422,7 +756,7 @@ static void r700Enable(GLcontext * ctx, GLenum cap, GLboolean state) //--------- case GL_POLYGON_OFFSET_POINT: case GL_POLYGON_OFFSET_LINE: case GL_POLYGON_OFFSET_FILL: - //r700SetPolygonOffsetState(ctx, state); + r700SetPolygonOffsetState(ctx, state); break; case GL_SCISSOR_TEST: radeon_firevertices(&context->radeon); @@ -431,7 +765,7 @@ static void r700Enable(GLcontext * ctx, GLenum cap, GLboolean state) //--------- break; case GL_LINE_STIPPLE: r700UpdateLineStipple(ctx); - break; + break; default: break; } @@ -499,6 +833,7 @@ static void r700Fogfv(GLcontext * ctx, GLenum pname, const GLfloat * param) //-- static void r700FrontFace(GLcontext * ctx, GLenum mode) //------------------ { r700UpdateCulling(ctx); + r700UpdatePolygonMode(ctx); } static void r700ShadeModel(GLcontext * ctx, GLenum mode) //-------------------- @@ -519,28 +854,201 @@ static void r700ShadeModel(GLcontext * ctx, GLenum mode) //-------------------- } } +/* ============================================================= + * Point state + */ +static void r700PointSize(GLcontext * ctx, GLfloat size) +{ + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + + /* We need to clamp to user defined range here, because + * the HW clamping happens only for per vertex point size. */ + size = CLAMP(size, ctx->Point.MinSize, ctx->Point.MaxSize); + + /* same size limits for AA, non-AA points */ + size = CLAMP(size, ctx->Const.MinPointSize, ctx->Const.MaxPointSize); + + /* format is 12.4 fixed point */ + SETfield(r700->PA_SU_POINT_SIZE.u32All, (int)(size * 16), + PA_SU_POINT_SIZE__HEIGHT_shift, PA_SU_POINT_SIZE__HEIGHT_mask); + SETfield(r700->PA_SU_POINT_SIZE.u32All, (int)(size * 16), + PA_SU_POINT_SIZE__WIDTH_shift, PA_SU_POINT_SIZE__WIDTH_mask); + +} + static void r700PointParameter(GLcontext * ctx, GLenum pname, const GLfloat * param) //--------------- { + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + + /* format is 12.4 fixed point */ + switch (pname) { + case GL_POINT_SIZE_MIN: + SETfield(r700->PA_SU_POINT_MINMAX.u32All, (int)(ctx->Point.MinSize * 16.0), + MIN_SIZE_shift, MIN_SIZE_mask); + break; + case GL_POINT_SIZE_MAX: + SETfield(r700->PA_SU_POINT_MINMAX.u32All, (int)(ctx->Point.MaxSize * 16.0), + MAX_SIZE_shift, MAX_SIZE_mask); + break; + case GL_POINT_DISTANCE_ATTENUATION: + break; + case GL_POINT_FADE_THRESHOLD_SIZE: + break; + default: + break; + } +} + +static int translate_stencil_func(int func) +{ + switch (func) { + case GL_NEVER: + return REF_NEVER; + case GL_LESS: + return REF_LESS; + case GL_EQUAL: + return REF_EQUAL; + case GL_LEQUAL: + return REF_LEQUAL; + case GL_GREATER: + return REF_GREATER; + case GL_NOTEQUAL: + return REF_NOTEQUAL; + case GL_GEQUAL: + return REF_GEQUAL; + case GL_ALWAYS: + return REF_ALWAYS; + } + return 0; +} + +static int translate_stencil_op(int op) +{ + switch (op) { + case GL_KEEP: + return STENCIL_KEEP; + case GL_ZERO: + return STENCIL_ZERO; + case GL_REPLACE: + return STENCIL_REPLACE; + case GL_INCR: + return STENCIL_INCR_CLAMP; + case GL_DECR: + return STENCIL_DECR_CLAMP; + case GL_INCR_WRAP_EXT: + return STENCIL_INCR_WRAP; + case GL_DECR_WRAP_EXT: + return STENCIL_DECR_WRAP; + case GL_INVERT: + return STENCIL_INVERT; + default: + WARN_ONCE("Do not know how to translate stencil op"); + return STENCIL_KEEP; + } + return 0; +} + +static void r700SetStencilState(GLcontext * ctx, GLboolean state) +{ + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + GLboolean hw_stencil = GL_FALSE; + + //fixme + //r300CatchStencilFallback(ctx); + + if (ctx->DrawBuffer) { + struct radeon_renderbuffer *rrbStencil + = radeon_get_renderbuffer(ctx->DrawBuffer, BUFFER_STENCIL); + hw_stencil = (rrbStencil && rrbStencil->bo); + } + + if (hw_stencil) { + if (state) + SETbit(r700->DB_DEPTH_CONTROL.u32All, STENCIL_ENABLE_bit); + else + CLEARbit(r700->DB_DEPTH_CONTROL.u32All, STENCIL_ENABLE_bit); + } } static void r700StencilFuncSeparate(GLcontext * ctx, GLenum face, GLenum func, GLint ref, GLuint mask) //--------------------- { -} + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + const unsigned back = ctx->Stencil._BackFace; + + //fixme + //r300CatchStencilFallback(ctx); + //front + SETfield(r700->DB_STENCILREFMASK.u32All, ctx->Stencil.Ref[0], + STENCILREF_shift, STENCILREF_mask); + SETfield(r700->DB_STENCILREFMASK.u32All, ctx->Stencil.ValueMask[0], + STENCILMASK_shift, STENCILMASK_mask); + + SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_func(ctx->Stencil.Function[0]), + STENCILFUNC_shift, STENCILFUNC_mask); + + //back + SETfield(r700->DB_STENCILREFMASK_BF.u32All, ctx->Stencil.Ref[back], + STENCILREF_BF_shift, STENCILREF_BF_mask); + SETfield(r700->DB_STENCILREFMASK_BF.u32All, ctx->Stencil.ValueMask[back], + STENCILMASK_BF_shift, STENCILMASK_BF_mask); + + SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_func(ctx->Stencil.Function[back]), + STENCILFUNC_BF_shift, STENCILFUNC_BF_mask); + +} static void r700StencilMaskSeparate(GLcontext * ctx, GLenum face, GLuint mask) //-------------- { + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + const unsigned back = ctx->Stencil._BackFace; + + //fixme + //r300CatchStencilFallback(ctx); + + // front + SETfield(r700->DB_STENCILREFMASK.u32All, ctx->Stencil.WriteMask[0], + STENCILWRITEMASK_shift, STENCILWRITEMASK_mask); + + // back + SETfield(r700->DB_STENCILREFMASK_BF.u32All, ctx->Stencil.WriteMask[back], + STENCILWRITEMASK_BF_shift, STENCILWRITEMASK_BF_mask); + } static void r700StencilOpSeparate(GLcontext * ctx, GLenum face, GLenum fail, GLenum zfail, GLenum zpass) //-------------------- { + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + const unsigned back = ctx->Stencil._BackFace; + + //fixme + //r300CatchStencilFallback(ctx); + + SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.FailFunc[0]), + STENCILFAIL_shift, STENCILFAIL_mask); + SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.ZFailFunc[0]), + STENCILZFAIL_shift, STENCILZFAIL_mask); + SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.ZPassFunc[0]), + STENCILZPASS_shift, STENCILZPASS_mask); + + SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.FailFunc[back]), + STENCILFAIL_BF_shift, STENCILFAIL_BF_mask); + SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.ZFailFunc[back]), + STENCILZFAIL_BF_shift, STENCILZFAIL_BF_mask); + SETfield(r700->DB_DEPTH_CONTROL.u32All, translate_stencil_op(ctx->Stencil.ZPassFunc[back]), + STENCILZPASS_BF_shift, STENCILZPASS_BF_mask); } static void r700UpdateWindow(GLcontext * ctx, int id) //-------------------- { - context_t *context = R700_CONTEXT(ctx); R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); __DRIdrawablePrivate *dPriv = radeon_get_drawable(&context->radeon); @@ -599,10 +1107,6 @@ static void r700DepthRange(GLcontext * ctx, GLclampd nearval, GLclampd farval) / r700UpdateWindow(ctx, 0); } -static void r700PointSize(GLcontext * ctx, GLfloat size) //------------------- -{ -} - static void r700LineWidth(GLcontext * ctx, GLfloat widthf) //--------------- { context_t *context = R700_CONTEXT(ctx); @@ -624,21 +1128,146 @@ static void r700LineStipple(GLcontext *ctx, GLint factor, GLushort pattern) SETfield(r700->PA_SC_LINE_STIPPLE.u32All, 1, AUTO_RESET_CNTL_shift, AUTO_RESET_CNTL_mask); } +static void r700SetPolygonOffsetState(GLcontext * ctx, GLboolean state) +{ + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + + if (state) { + SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_FRONT_ENABLE_bit); + SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_BACK_ENABLE_bit); + SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_PARA_ENABLE_bit); + } else { + CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_FRONT_ENABLE_bit); + CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_BACK_ENABLE_bit); + CLEARbit(r700->PA_SU_SC_MODE_CNTL.u32All, POLY_OFFSET_PARA_ENABLE_bit); + } +} + static void r700PolygonOffset(GLcontext * ctx, GLfloat factor, GLfloat units) //-------------- { + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + GLfloat constant = units; + + switch (ctx->Visual.depthBits) { + case 16: + constant *= 4.0; + break; + case 24: + constant *= 2.0; + break; + } + + factor *= 12.0; + + r700->PA_SU_POLY_OFFSET_FRONT_SCALE.f32All = factor; + r700->PA_SU_POLY_OFFSET_FRONT_OFFSET.f32All = constant; + r700->PA_SU_POLY_OFFSET_BACK_SCALE.f32All = factor; + r700->PA_SU_POLY_OFFSET_BACK_OFFSET.f32All = constant; } +static void r700UpdatePolygonMode(GLcontext * ctx) +{ + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + + SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DISABLE_POLY_MODE, POLY_MODE_shift, POLY_MODE_mask); + + /* Only do something if a polygon mode is wanted, default is GL_FILL */ + if (ctx->Polygon.FrontMode != GL_FILL || + ctx->Polygon.BackMode != GL_FILL) { + GLenum f, b; + + /* Handle GL_CW (clock wise and GL_CCW (counter clock wise) + * correctly by selecting the correct front and back face + */ + if (ctx->Polygon.FrontFace == GL_CCW) { + f = ctx->Polygon.FrontMode; + b = ctx->Polygon.BackMode; + } else { + f = ctx->Polygon.BackMode; + b = ctx->Polygon.FrontMode; + } + + /* Enable polygon mode */ + SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DUAL_MODE, POLY_MODE_shift, POLY_MODE_mask); + + switch (f) { + case GL_LINE: + SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DRAW_LINES, + POLYMODE_FRONT_PTYPE_shift, POLYMODE_FRONT_PTYPE_mask); + break; + case GL_POINT: + SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DRAW_POINTS, + POLYMODE_FRONT_PTYPE_shift, POLYMODE_FRONT_PTYPE_mask); + break; + case GL_FILL: + SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DRAW_TRIANGLES, + POLYMODE_FRONT_PTYPE_shift, POLYMODE_FRONT_PTYPE_mask); + break; + } + + switch (b) { + case GL_LINE: + SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DRAW_LINES, + POLYMODE_BACK_PTYPE_shift, POLYMODE_BACK_PTYPE_mask); + break; + case GL_POINT: + SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DRAW_POINTS, + POLYMODE_BACK_PTYPE_shift, POLYMODE_BACK_PTYPE_mask); + break; + case GL_FILL: + SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DRAW_TRIANGLES, + POLYMODE_BACK_PTYPE_shift, POLYMODE_BACK_PTYPE_mask); + break; + } + } +} static void r700PolygonMode(GLcontext * ctx, GLenum face, GLenum mode) //------------------ { + (void)face; + (void)mode; + + r700UpdatePolygonMode(ctx); } - + static void r700RenderMode(GLcontext * ctx, GLenum mode) //--------------------- { } -static void r700ClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq ) //----------------- +static void r700ClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq ) +{ + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + GLint p; + GLint *ip; + + p = (GLint) plane - (GLint) GL_CLIP_PLANE0; + ip = (GLint *)ctx->Transform._ClipUserPlane[p]; + + r700->ucp[p].PA_CL_UCP_0_X.u32All = ip[0]; + r700->ucp[p].PA_CL_UCP_0_Y.u32All = ip[1]; + r700->ucp[p].PA_CL_UCP_0_Z.u32All = ip[2]; + r700->ucp[p].PA_CL_UCP_0_W.u32All = ip[3]; +} + +static void r700SetClipPlaneState(GLcontext * ctx, GLenum cap, GLboolean state) { + context_t *context = R700_CONTEXT(ctx); + R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + GLuint p; + + p = cap - GL_CLIP_PLANE0; + if (state) { + r700->PA_CL_CLIP_CNTL.u32All |= (UCP_ENA_0_bit << p); + r700->ucp[p].enabled = GL_TRUE; + r700ClipPlane(ctx, cap, NULL); + } else { + r700->PA_CL_CLIP_CNTL.u32All &= ~(UCP_ENA_0_bit << p); + r700->ucp[p].enabled = GL_FALSE; + } } void r700SetScissor(context_t *context) //--------------- @@ -719,22 +1348,22 @@ void r700SetScissor(context_t *context) //--------------- r700->viewport[id].enabled = GL_TRUE; } -void r700SetRenderTarget(context_t *context, int id) +static void r700SetRenderTarget(context_t *context, int id) { R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); struct radeon_renderbuffer *rrb; unsigned int nPitchInPixel; + rrb = radeon_get_colorbuffer(&context->radeon); + if (!rrb || !rrb->bo) { + fprintf(stderr, "no rrb\n"); + return; + } + /* screen/window/view */ SETfield(r700->CB_TARGET_MASK.u32All, 0xF, (4 * id), TARGET0_ENABLE_mask); - rrb = radeon_get_colorbuffer(&context->radeon); - if (!rrb || !rrb->bo) { - fprintf(stderr, "no rrb\n"); - return; - } - /* color buffer */ r700->render_target[id].CB_COLOR0_BASE.u32All = context->radeon.state.color.draw_offset; @@ -764,45 +1393,25 @@ void r700SetRenderTarget(context_t *context, int id) SETbit(r700->render_target[id].CB_COLOR0_INFO.u32All, BLEND_CLAMP_bit); SETfield(r700->render_target[id].CB_COLOR0_INFO.u32All, NUMBER_UNORM, NUMBER_TYPE_shift, NUMBER_TYPE_mask); - CLEARfield(r700->render_target[id].CB_BLEND0_CONTROL.u32All, COLOR_SRCBLEND_mask); /* no dst blend */ - CLEARfield(r700->render_target[id].CB_BLEND0_CONTROL.u32All, ALPHA_SRCBLEND_mask); /* no dst blend */ - r700->render_target[id].enabled = GL_TRUE; } -void r700SetDepthTarget(context_t *context) +static void r700SetDepthTarget(context_t *context) { R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); struct radeon_renderbuffer *rrb; unsigned int nPitchInPixel; + rrb = radeon_get_depthbuffer(&context->radeon); + if (!rrb) + return; + /* depth buf */ r700->DB_DEPTH_SIZE.u32All = 0; r700->DB_DEPTH_BASE.u32All = 0; r700->DB_DEPTH_INFO.u32All = 0; - - r700->DB_DEPTH_CLEAR.u32All = 0x3F800000; - r700->DB_DEPTH_VIEW.u32All = 0; - r700->DB_RENDER_CONTROL.u32All = 0; - SETbit(r700->DB_RENDER_CONTROL.u32All, STENCIL_COMPRESS_DISABLE_bit); - SETbit(r700->DB_RENDER_CONTROL.u32All, DEPTH_COMPRESS_DISABLE_bit); - r700->DB_RENDER_OVERRIDE.u32All = 0; - if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770) - SETbit(r700->DB_RENDER_OVERRIDE.u32All, FORCE_SHADER_Z_ORDER_bit); - SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIZ_ENABLE_shift, FORCE_HIZ_ENABLE_mask); - SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIS_ENABLE0_shift, FORCE_HIS_ENABLE0_mask); - SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIS_ENABLE1_shift, FORCE_HIS_ENABLE1_mask); - - r700->DB_ALPHA_TO_MASK.u32All = 0; - SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET0_shift, ALPHA_TO_MASK_OFFSET0_mask); - SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET1_shift, ALPHA_TO_MASK_OFFSET1_mask); - SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET2_shift, ALPHA_TO_MASK_OFFSET2_mask); - SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET3_shift, ALPHA_TO_MASK_OFFSET3_mask); - - rrb = radeon_get_depthbuffer(&context->radeon); - if (!rrb) - return; + r700->DB_DEPTH_VIEW.u32All = 0; nPitchInPixel = rrb->pitch/rrb->cpp; @@ -1033,9 +1642,10 @@ static void r700InitSQConfig(GLcontext * ctx) void r700InitState(GLcontext * ctx) //------------------- { context_t *context = R700_CONTEXT(ctx); - R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); + radeon_firevertices(&context->radeon); + r700->TA_CNTL_AUX.u32All = 0; SETfield(r700->TA_CNTL_AUX.u32All, 28, TD_FIFO_CREDIT_shift, TD_FIFO_CREDIT_mask); r700->VC_ENHANCE.u32All = 0; @@ -1064,37 +1674,14 @@ void r700InitState(GLcontext * ctx) //------------------- r700->VGT_MIN_VTX_INDX.u32All = 0; r700->VGT_INDX_OFFSET.u32All = 0; - /* Specify the number of instances */ - r700->VGT_DMA_NUM_INSTANCES.u32All = 1; - - /* not alpha blend */ - CLEARfield(r700->SX_ALPHA_TEST_CONTROL.u32All, ALPHA_FUNC_mask); - CLEARbit(r700->SX_ALPHA_TEST_CONTROL.u32All, ALPHA_TEST_ENABLE_bit); - /* default shader connections. */ r700->SPI_VS_OUT_ID_0.u32All = 0x03020100; r700->SPI_VS_OUT_ID_1.u32All = 0x07060504; - r700->SPI_PS_INPUT_CNTL_0.u32All = 0x00000800; - r700->SPI_PS_INPUT_CNTL_1.u32All = 0x00000801; - r700->SPI_PS_INPUT_CNTL_2.u32All = 0x00000802; - r700->SPI_THREAD_GROUPING.u32All = 0; if (context->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV770) SETfield(r700->SPI_THREAD_GROUPING.u32All, 1, PS_GROUPING_shift, PS_GROUPING_mask); - SETfield(r700->CB_COLOR_CONTROL.u32All, 0xCC, ROP3_shift, ROP3_mask); - CLEARbit(r700->CB_COLOR_CONTROL.u32All, PER_MRT_BLEND_bit); - - r700->DB_SHADER_CONTROL.u32All = 0; - SETbit(r700->DB_SHADER_CONTROL.u32All, DUAL_EXPORT_ENABLE_bit); - - /* Set up the culling control register */ - SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DRAW_TRIANGLES, - POLYMODE_FRONT_PTYPE_shift, POLYMODE_FRONT_PTYPE_mask); - SETfield(r700->PA_SU_SC_MODE_CNTL.u32All, X_DRAW_TRIANGLES, - POLYMODE_BACK_PTYPE_shift, POLYMODE_BACK_PTYPE_mask); - /* screen */ r700->PA_SC_SCREEN_SCISSOR_TL.u32All = 0x0; @@ -1138,21 +1725,8 @@ void r700InitState(GLcontext * ctx) //------------------- SETbit(r700->PA_CL_VTE_CNTL.u32All, VPORT_Z_SCALE_ENA_bit); SETbit(r700->PA_CL_VTE_CNTL.u32All, VPORT_Z_OFFSET_ENA_bit); - /* Set up point sizes and min/max values */ - SETfield(r700->PA_SU_POINT_SIZE.u32All, 0x8, - PA_SU_POINT_SIZE__HEIGHT_shift, PA_SU_POINT_SIZE__HEIGHT_mask); - SETfield(r700->PA_SU_POINT_SIZE.u32All, 0x8, - PA_SU_POINT_SIZE__WIDTH_shift, PA_SU_POINT_SIZE__WIDTH_mask); - CLEARfield(r700->PA_SU_POINT_MINMAX.u32All, MIN_SIZE_mask); - SETfield(r700->PA_SU_POINT_MINMAX.u32All, 0x8000, MAX_SIZE_shift, MAX_SIZE_mask); - - /* Set up line control */ - SETfield(r700->PA_SU_LINE_CNTL.u32All, 0x8, - PA_SU_LINE_CNTL__WIDTH_shift, PA_SU_LINE_CNTL__WIDTH_mask); - - r700->PA_SC_LINE_CNTL.u32All = 0; - CLEARbit(r700->PA_SC_LINE_CNTL.u32All, EXPAND_LINE_WIDTH_bit); - SETbit(r700->PA_SC_LINE_CNTL.u32All, LAST_PIXEL_bit); + /* GL uses last vtx for flat shading components */ + SETbit(r700->PA_SU_SC_MODE_CNTL.u32All, PROVOKING_VTX_LAST_bit); /* Set up vertex control */ r700->PA_SU_VTX_CNTL.u32All = 0; @@ -1167,7 +1741,84 @@ void r700InitState(GLcontext * ctx) //------------------- r700->PA_CL_GB_HORZ_CLIP_ADJ.u32All = 0x3F800000; r700->PA_CL_GB_HORZ_DISC_ADJ.u32All = 0x3F800000; + /* Enable all samples for multi-sample anti-aliasing */ + r700->PA_SC_AA_MASK.u32All = 0xFFFFFFFF; + /* Turn off AA */ + r700->PA_SC_AA_CONFIG.u32All = 0; + + r700->SX_MISC.u32All = 0; + + r700InitSQConfig(ctx); + + r700ColorMask(ctx, + ctx->Color.ColorMask[RCOMP], + ctx->Color.ColorMask[GCOMP], + ctx->Color.ColorMask[BCOMP], + ctx->Color.ColorMask[ACOMP]); + + r700Enable(ctx, GL_DEPTH_TEST, ctx->Depth.Test); + r700DepthMask(ctx, ctx->Depth.Mask); + r700DepthFunc(ctx, ctx->Depth.Func); + SETbit(r700->DB_SHADER_CONTROL.u32All, DUAL_EXPORT_ENABLE_bit); + + r700->DB_DEPTH_CLEAR.u32All = 0x3F800000; + + r700->DB_RENDER_CONTROL.u32All = 0; + SETbit(r700->DB_RENDER_CONTROL.u32All, STENCIL_COMPRESS_DISABLE_bit); + SETbit(r700->DB_RENDER_CONTROL.u32All, DEPTH_COMPRESS_DISABLE_bit); + r700->DB_RENDER_OVERRIDE.u32All = 0; + if (context->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV770) + SETbit(r700->DB_RENDER_OVERRIDE.u32All, FORCE_SHADER_Z_ORDER_bit); + SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIZ_ENABLE_shift, FORCE_HIZ_ENABLE_mask); + SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIS_ENABLE0_shift, FORCE_HIS_ENABLE0_mask); + SETfield(r700->DB_RENDER_OVERRIDE.u32All, FORCE_DISABLE, FORCE_HIS_ENABLE1_shift, FORCE_HIS_ENABLE1_mask); + + r700->DB_ALPHA_TO_MASK.u32All = 0; + SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET0_shift, ALPHA_TO_MASK_OFFSET0_mask); + SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET1_shift, ALPHA_TO_MASK_OFFSET1_mask); + SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET2_shift, ALPHA_TO_MASK_OFFSET2_mask); + SETfield(r700->DB_ALPHA_TO_MASK.u32All, 2, ALPHA_TO_MASK_OFFSET3_shift, ALPHA_TO_MASK_OFFSET3_mask); + + /* stencil */ + r700Enable(ctx, GL_STENCIL_TEST, ctx->Stencil._Enabled); + r700StencilMaskSeparate(ctx, 0, ctx->Stencil.WriteMask[0]); + r700StencilFuncSeparate(ctx, 0, ctx->Stencil.Function[0], + ctx->Stencil.Ref[0], ctx->Stencil.ValueMask[0]); + r700StencilOpSeparate(ctx, 0, ctx->Stencil.FailFunc[0], + ctx->Stencil.ZFailFunc[0], + ctx->Stencil.ZPassFunc[0]); + + r700UpdateCulling(ctx); + + r700SetBlendState(ctx); + r700SetLogicOpState(ctx); + + r700AlphaFunc(ctx, ctx->Color.AlphaFunc, ctx->Color.AlphaRef); + r700Enable(ctx, GL_ALPHA_TEST, ctx->Color.AlphaEnabled); + + r700PointSize(ctx, 1.0); + + CLEARfield(r700->PA_SU_POINT_MINMAX.u32All, MIN_SIZE_mask); + SETfield(r700->PA_SU_POINT_MINMAX.u32All, 0x8000, MAX_SIZE_shift, MAX_SIZE_mask); + + r700LineWidth(ctx, 1.0); + + r700->PA_SC_LINE_CNTL.u32All = 0; + CLEARbit(r700->PA_SC_LINE_CNTL.u32All, EXPAND_LINE_WIDTH_bit); + SETbit(r700->PA_SC_LINE_CNTL.u32All, LAST_PIXEL_bit); + + r700ShadeModel(ctx, ctx->Light.ShadeModel); + r700PolygonMode(ctx, GL_FRONT, ctx->Polygon.FrontMode); + r700PolygonMode(ctx, GL_BACK, ctx->Polygon.BackMode); + r700PolygonOffset(ctx, ctx->Polygon.OffsetFactor, + ctx->Polygon.OffsetUnits); + r700Enable(ctx, GL_POLYGON_OFFSET_POINT, ctx->Polygon.OffsetPoint); + r700Enable(ctx, GL_POLYGON_OFFSET_LINE, ctx->Polygon.OffsetLine); + r700Enable(ctx, GL_POLYGON_OFFSET_FILL, ctx->Polygon.OffsetFill); + /* CB */ + r700BlendColor(ctx, ctx->Color.BlendColor); + r700->CB_CLEAR_RED_R6XX.f32All = 1.0; //r6xx only r700->CB_CLEAR_GREEN_R6XX.f32All = 0.0; //r6xx only r700->CB_CLEAR_BLUE_R6XX.f32All = 1.0; //r6xx only @@ -1176,13 +1827,6 @@ void r700InitState(GLcontext * ctx) //------------------- r700->CB_FOG_GREEN_R6XX.u32All = 0; //r6xx only r700->CB_FOG_BLUE_R6XX.u32All = 0; //r6xx only - r700->CB_BLEND_RED.u32All = 0; - r700->CB_BLEND_GREEN.u32All = 0; - r700->CB_BLEND_BLUE.u32All = 0; - r700->CB_BLEND_ALPHA.u32All = 0; - - r700->CB_BLEND_CONTROL.u32All = 0; - /* Disable color compares */ SETfield(r700->CB_CLRCMP_CONTROL.u32All, CLRCMP_DRAW_ALWAYS, CLRCMP_FCN_SRC_shift, CLRCMP_FCN_SRC_mask); @@ -1200,17 +1844,8 @@ void r700InitState(GLcontext * ctx) //------------------- /* Set up color compare mask */ r700->CB_CLRCMP_MSK.u32All = 0xFFFFFFFF; - /* default color mask */ - SETfield(r700->CB_SHADER_MASK.u32All, 0xF, OUTPUT0_ENABLE_shift, OUTPUT0_ENABLE_mask); - - /* Enable all samples for multi-sample anti-aliasing */ - r700->PA_SC_AA_MASK.u32All = 0xFFFFFFFF; - /* Turn off AA */ - r700->PA_SC_AA_CONFIG.u32All = 0; - - r700->SX_MISC.u32All = 0; + context->radeon.hw.all_dirty = GL_TRUE; - r700InitSQConfig(ctx); } void r700InitStateFuncs(struct dd_function_table *functions) //----------------- @@ -1228,6 +1863,7 @@ void r700InitStateFuncs(struct dd_function_table *functions) //----------------- functions->Fogfv = r700Fogfv; functions->FrontFace = r700FrontFace; functions->ShadeModel = r700ShadeModel; + functions->LogicOpcode = r700LogicOpcode; /* ARB_point_parameters */ functions->PointParameterfv = r700PointParameter; diff --git a/src/mesa/drivers/dri/r600/r700_state.h b/src/mesa/drivers/dri/r600/r700_state.h index 23246367db..30eb54e8b0 100644 --- a/src/mesa/drivers/dri/r600/r700_state.h +++ b/src/mesa/drivers/dri/r600/r700_state.h @@ -42,10 +42,8 @@ extern void r700UpdateDrawBuffer (GLcontext * ctx); extern void r700InitState (GLcontext * ctx); extern void r700InitStateFuncs (struct dd_function_table *functions); -extern void r700SetRenderTarget(context_t *context, int id); extern void r700SetDefaultStates(context_t * context); -void r700SetScissor(context_t *context); -void r700SetDepthTarget(context_t *context); +extern void r700SetScissor(context_t *context); #endif /* _R600_SCREEN_H */ diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index cbfeaf071b..31e71cdfa3 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -52,7 +52,7 @@ unsigned int Map_Vertex_Output(r700_AssemblerBase *pAsm, unsigned int unBit; unsigned int unTotal = unStart; - //!!!!!!! THE ORDER MATCH FS INPUT + //!!!!!!! THE ORDER MATCH FS INPUT unBit = 1 << VERT_RESULT_HPOS; if(mesa_vp->Base.OutputsWritten & unBit) @@ -73,31 +73,31 @@ unsigned int Map_Vertex_Output(r700_AssemblerBase *pAsm, } //TODO : dealing back face. - //unBit = 1 << VERT_RESULT_BFC0; - //if(mesa_vp->Base.OutputsWritten & unBit) - //{ - // pAsm->ucVP_OutputMap[VERT_RESULT_COL0] = unTotal++; - //} - - //unBit = 1 << VERT_RESULT_BFC1; - //if(mesa_vp->Base.OutputsWritten & unBit) - //{ - // pAsm->ucVP_OutputMap[VERT_RESULT_COL1] = unTotal++; - //} + unBit = 1 << VERT_RESULT_BFC0; + if(mesa_vp->Base.OutputsWritten & unBit) + { + pAsm->ucVP_OutputMap[VERT_RESULT_BFC0] = unTotal++; + } + + unBit = 1 << VERT_RESULT_BFC1; + if(mesa_vp->Base.OutputsWritten & unBit) + { + pAsm->ucVP_OutputMap[VERT_RESULT_BFC1] = unTotal++; + } //TODO : dealing fog. - //unBit = 1 << VERT_RESULT_FOGC; - //if(mesa_vp->Base.OutputsWritten & unBit) - //{ - // pAsm->ucVP_OutputMap[VERT_RESULT_FOGC] = unTotal++; - //} + unBit = 1 << VERT_RESULT_FOGC; + if(mesa_vp->Base.OutputsWritten & unBit) + { + pAsm->ucVP_OutputMap[VERT_RESULT_FOGC] = unTotal++; + } //TODO : dealing point size. - //unBit = 1 << VERT_RESULT_PSIZ; - //if(mesa_vp->Base.OutputsWritten & unBit) - //{ - // pAsm->ucVP_OutputMap[VERT_RESULT_PSIZ] = unTotal++; - //} + unBit = 1 << VERT_RESULT_PSIZ; + if(mesa_vp->Base.OutputsWritten & unBit) + { + pAsm->ucVP_OutputMap[VERT_RESULT_PSIZ] = unTotal++; + } for(i=0; i<8; i++) { @@ -335,17 +335,13 @@ void * r700GetActiveVpShaderBo(GLcontext * ctx) GLboolean r700SetupVertexProgram(GLcontext * ctx) { context_t *context = R700_CONTEXT(ctx); - - BATCH_LOCALS(&context->radeon); - R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw); - + BATCH_LOCALS(&context->radeon); struct r700_vertex_program *vp = (struct r700_vertex_program *)ctx->VertexProgram._Current; struct gl_program_parameter_list *paramList; unsigned int unNumParamData; - unsigned int ui; if(GL_FALSE == vp->loaded) @@ -359,7 +355,7 @@ GLboolean r700SetupVertexProgram(GLcontext * ctx) r600EmitShader(ctx, &(vp->shaderbo), (GLvoid *)(vp->r700Shader.pProgram), - vp->r700Shader.uShaderBinaryDWORDSize,
+ vp->r700Shader.uShaderBinaryDWORDSize, "VS"); vp->loaded = GL_TRUE; @@ -394,7 +390,6 @@ GLboolean r700SetupVertexProgram(GLcontext * ctx) */ /* sent out shader constants. */ - paramList = vp->mesa_program.Base.Parameters; if(NULL != paramList) @@ -403,8 +398,8 @@ GLboolean r700SetupVertexProgram(GLcontext * ctx) unNumParamData = paramList->NumParameters * 4; - BEGIN_BATCH_NO_AUTOSTATE(unNumParamData + 2); - + BEGIN_BATCH_NO_AUTOSTATE(unNumParamData + 2); + R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_ALU_CONST, unNumParamData)); /* assembler map const from very beginning. */ R600_OUT_BATCH(SQ_ALU_CONSTANT_VS_OFFSET * 4); diff --git a/src/mesa/drivers/dri/r600/radeon_bo_legacy.c b/src/mesa/drivers/dri/r600/radeon_bo_legacy.c new file mode 120000 index 0000000000..79ad050e6b --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_bo_legacy.c @@ -0,0 +1 @@ +../radeon/radeon_bo_legacy.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_bo_legacy.h b/src/mesa/drivers/dri/r600/radeon_bo_legacy.h new file mode 120000 index 0000000000..83b0f7ffab --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_bo_legacy.h @@ -0,0 +1 @@ +../radeon/radeon_bo_legacy.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_bocs_wrapper.h b/src/mesa/drivers/dri/r600/radeon_bocs_wrapper.h new file mode 120000 index 0000000000..ca894b2443 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_bocs_wrapper.h @@ -0,0 +1 @@ +../radeon/radeon_bocs_wrapper.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_chipset.h b/src/mesa/drivers/dri/r600/radeon_chipset.h new file mode 120000 index 0000000000..eba99001ff --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_chipset.h @@ -0,0 +1 @@ +../radeon/radeon_chipset.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_cmdbuf.h b/src/mesa/drivers/dri/r600/radeon_cmdbuf.h new file mode 120000 index 0000000000..a799e1dc6d --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_cmdbuf.h @@ -0,0 +1 @@ +../radeon/radeon_cmdbuf.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_common.c b/src/mesa/drivers/dri/r600/radeon_common.c new file mode 120000 index 0000000000..67b19ba940 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_common.c @@ -0,0 +1 @@ +../radeon/radeon_common.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_common.h b/src/mesa/drivers/dri/r600/radeon_common.h new file mode 120000 index 0000000000..5bcb696a9f --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_common.h @@ -0,0 +1 @@ +../radeon/radeon_common.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_common_context.c b/src/mesa/drivers/dri/r600/radeon_common_context.c new file mode 120000 index 0000000000..86800f3819 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_common_context.c @@ -0,0 +1 @@ +../radeon/radeon_common_context.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_common_context.h b/src/mesa/drivers/dri/r600/radeon_common_context.h new file mode 120000 index 0000000000..4d66312550 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_common_context.h @@ -0,0 +1 @@ +../radeon/radeon_common_context.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_context.h b/src/mesa/drivers/dri/r600/radeon_context.h deleted file mode 100644 index 250570f6b8..0000000000 --- a/src/mesa/drivers/dri/r600/radeon_context.h +++ /dev/null @@ -1,76 +0,0 @@ -/************************************************************************** - -Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and - VA Linux Systems Inc., Fremont, California. -Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. - -The Weather Channel (TM) funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 license. -This notice must be preserved. - -All Rights Reserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Gareth Hughes <gareth@valinux.com> - * Keith Whitwell <keith@tungstengraphics.com> - * Kevin E. Martin <martin@valinux.com> - * Nicolai Haehnle <prefect_@gmx.net> - */ - -#ifndef __RADEON_CONTEXT_H__ -#define __RADEON_CONTEXT_H__ - -#include "main/mtypes.h" -#include "main/colormac.h" -#include "radeon_screen.h" -#include "drm.h" -#include "dri_util.h" - -#include "radeon_screen.h" - -#if R200_MERGED -extern void radeonFallback(GLcontext * ctx, GLuint bit, GLboolean mode); - -#define FALLBACK( radeon, bit, mode ) do { \ - if ( 0 ) fprintf( stderr, "FALLBACK in %s: #%d=%d\n", \ - __FUNCTION__, bit, mode ); \ - radeonFallback( (radeon)->glCtx, bit, mode ); \ -} while (0) -#else -#define FALLBACK( radeon, bit, mode ) fprintf(stderr, "%s:%s\n", __LINE__, __FILE__); -#endif - -/* TCL fallbacks */ -extern void radeonTclFallback(GLcontext * ctx, GLuint bit, GLboolean mode); - -#if R200_MERGED -#define TCL_FALLBACK( ctx, bit, mode ) radeonTclFallback( ctx, bit, mode ) -#else -#define TCL_FALLBACK( ctx, bit, mode ) ; -#endif - - -#endif /* __RADEON_CONTEXT_H__ */ diff --git a/src/mesa/drivers/dri/r600/radeon_cs_legacy.c b/src/mesa/drivers/dri/r600/radeon_cs_legacy.c new file mode 120000 index 0000000000..006720f8a4 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_cs_legacy.c @@ -0,0 +1 @@ +../radeon/radeon_cs_legacy.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_cs_legacy.h b/src/mesa/drivers/dri/r600/radeon_cs_legacy.h new file mode 120000 index 0000000000..a5f95e0a3d --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_cs_legacy.h @@ -0,0 +1 @@ +../radeon/radeon_cs_legacy.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_cs_space_drm.c b/src/mesa/drivers/dri/r600/radeon_cs_space_drm.c new file mode 120000 index 0000000000..c248ea7d1a --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_cs_space_drm.c @@ -0,0 +1 @@ +../radeon/radeon_cs_space_drm.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_dma.c b/src/mesa/drivers/dri/r600/radeon_dma.c new file mode 120000 index 0000000000..43be000625 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_dma.c @@ -0,0 +1 @@ +../radeon/radeon_dma.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_dma.h b/src/mesa/drivers/dri/r600/radeon_dma.h new file mode 120000 index 0000000000..82e50634e3 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_dma.h @@ -0,0 +1 @@ +../radeon/radeon_dma.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_fbo.c b/src/mesa/drivers/dri/r600/radeon_fbo.c new file mode 120000 index 0000000000..0d738d8d78 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_fbo.c @@ -0,0 +1 @@ +../radeon/radeon_fbo.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_lock.c b/src/mesa/drivers/dri/r600/radeon_lock.c new file mode 120000 index 0000000000..af4108a8e3 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_lock.c @@ -0,0 +1 @@ +../radeon/radeon_lock.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_lock.h b/src/mesa/drivers/dri/r600/radeon_lock.h new file mode 120000 index 0000000000..64bdf94ee7 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_lock.h @@ -0,0 +1 @@ +../radeon/radeon_lock.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_mipmap_tree.c b/src/mesa/drivers/dri/r600/radeon_mipmap_tree.c new file mode 120000 index 0000000000..31c0cfbe94 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_mipmap_tree.c @@ -0,0 +1 @@ +../radeon/radeon_mipmap_tree.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_mipmap_tree.h b/src/mesa/drivers/dri/r600/radeon_mipmap_tree.h new file mode 120000 index 0000000000..254d50cf8c --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_mipmap_tree.h @@ -0,0 +1 @@ +../radeon/radeon_mipmap_tree.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_screen.c b/src/mesa/drivers/dri/r600/radeon_screen.c new file mode 120000 index 0000000000..86161118dd --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_screen.c @@ -0,0 +1 @@ +../radeon/radeon_screen.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_screen.h b/src/mesa/drivers/dri/r600/radeon_screen.h new file mode 120000 index 0000000000..23bb6bd459 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_screen.h @@ -0,0 +1 @@ +../radeon/radeon_screen.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_span.c b/src/mesa/drivers/dri/r600/radeon_span.c new file mode 120000 index 0000000000..232868c4c9 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_span.c @@ -0,0 +1 @@ +../radeon/radeon_span.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_span.h b/src/mesa/drivers/dri/r600/radeon_span.h new file mode 120000 index 0000000000..f9d634508c --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_span.h @@ -0,0 +1 @@ +../radeon/radeon_span.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_texture.c b/src/mesa/drivers/dri/r600/radeon_texture.c new file mode 120000 index 0000000000..a822710915 --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_texture.c @@ -0,0 +1 @@ +../radeon/radeon_texture.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/radeon_texture.h b/src/mesa/drivers/dri/r600/radeon_texture.h new file mode 120000 index 0000000000..17fac3d5ea --- /dev/null +++ b/src/mesa/drivers/dri/r600/radeon_texture.h @@ -0,0 +1 @@ +../radeon/radeon_texture.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/server/radeon.h b/src/mesa/drivers/dri/r600/server/radeon.h new file mode 120000 index 0000000000..81274a54f1 --- /dev/null +++ b/src/mesa/drivers/dri/r600/server/radeon.h @@ -0,0 +1 @@ +../../radeon/server/radeon.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/server/radeon_dri.c b/src/mesa/drivers/dri/r600/server/radeon_dri.c new file mode 120000 index 0000000000..d05847d650 --- /dev/null +++ b/src/mesa/drivers/dri/r600/server/radeon_dri.c @@ -0,0 +1 @@ +../../radeon/server/radeon_dri.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/server/radeon_dri.h b/src/mesa/drivers/dri/r600/server/radeon_dri.h new file mode 120000 index 0000000000..27c591d3c9 --- /dev/null +++ b/src/mesa/drivers/dri/r600/server/radeon_dri.h @@ -0,0 +1 @@ +../../radeon/server/radeon_dri.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/server/radeon_egl.c b/src/mesa/drivers/dri/r600/server/radeon_egl.c new file mode 120000 index 0000000000..d7735a7643 --- /dev/null +++ b/src/mesa/drivers/dri/r600/server/radeon_egl.c @@ -0,0 +1 @@ +../../radeon/server/radeon_egl.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/server/radeon_macros.h b/src/mesa/drivers/dri/r600/server/radeon_macros.h new file mode 120000 index 0000000000..c56cd735b8 --- /dev/null +++ b/src/mesa/drivers/dri/r600/server/radeon_macros.h @@ -0,0 +1 @@ +../../radeon/server/radeon_macros.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/r600/server/radeon_reg.h b/src/mesa/drivers/dri/r600/server/radeon_reg.h new file mode 120000 index 0000000000..e2349dcb68 --- /dev/null +++ b/src/mesa/drivers/dri/r600/server/radeon_reg.h @@ -0,0 +1 @@ +../../radeon/server/radeon_reg.h
\ No newline at end of file diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_drm.h b/src/mesa/drivers/dri/radeon/radeon_bo_drm.h index 8eeaea1cb2..8789e3ab09 100644 --- a/src/mesa/drivers/dri/radeon/radeon_bo_drm.h +++ b/src/mesa/drivers/dri/radeon/radeon_bo_drm.h @@ -34,10 +34,6 @@ #include <stdint.h> //#include "radeon_track.h" -#ifndef RADEON_DEBUG_BO -#define RADEON_DEBUG_BO 0 -#endif - /* bo object */ #define RADEON_BO_FLAGS_MACRO_TILE 1 #define RADEON_BO_FLAGS_MICRO_TILE 2 @@ -61,28 +57,22 @@ struct radeon_bo { /* bo functions */ struct radeon_bo_funcs { -#ifdef RADEON_DEBUG_BO - struct radeon_bo *(*bo_open)(struct radeon_bo_manager *bom, - uint32_t handle, - uint32_t size, - uint32_t alignment, - uint32_t domains, - uint32_t flags, - char * szBufUsage); -#else struct radeon_bo *(*bo_open)(struct radeon_bo_manager *bom, uint32_t handle, uint32_t size, uint32_t alignment, uint32_t domains, uint32_t flags); -#endif /* RADEON_DEBUG_BO */ void (*bo_ref)(struct radeon_bo *bo); struct radeon_bo *(*bo_unref)(struct radeon_bo *bo); int (*bo_map)(struct radeon_bo *bo, int write); int (*bo_unmap)(struct radeon_bo *bo); int (*bo_wait)(struct radeon_bo *bo); int (*bo_is_static)(struct radeon_bo *bo); + int (*bo_set_tiling)(struct radeon_bo *bo, uint32_t tiling_flags, + uint32_t pitch); + int (*bo_get_tiling)(struct radeon_bo *bo, uint32_t *tiling_flags, + uint32_t *pitch); }; struct radeon_bo_manager { @@ -110,20 +100,13 @@ static inline struct radeon_bo *_radeon_bo_open(struct radeon_bo_manager *bom, uint32_t alignment, uint32_t domains, uint32_t flags, -#ifdef RADEON_DEBUG_BO - char * szBufUsage, -#endif /* RADEON_DEBUG_BO */ const char *file, const char *func, int line) { struct radeon_bo *bo; -#ifdef RADEON_DEBUG_BO - bo = bom->funcs->bo_open(bom, handle, size, alignment, domains, flags, szBufUsage); -#else bo = bom->funcs->bo_open(bom, handle, size, alignment, domains, flags); -#endif /* RADEON_DEBUG_BO */ #ifdef RADEON_BO_TRACK if (bo) { @@ -187,6 +170,18 @@ static inline int _radeon_bo_wait(struct radeon_bo *bo, return bo->bom->funcs->bo_wait(bo); } +static inline int radeon_bo_set_tiling(struct radeon_bo *bo, + uint32_t tiling_flags, uint32_t pitch) +{ + return bo->bom->funcs->bo_set_tiling(bo, tiling_flags, pitch); +} + +static inline int radeon_bo_get_tiling(struct radeon_bo *bo, + uint32_t *tiling_flags, uint32_t *pitch) +{ + return bo->bom->funcs->bo_get_tiling(bo, tiling_flags, pitch); +} + static inline int radeon_bo_is_static(struct radeon_bo *bo) { if (bo->bom->funcs->bo_is_static) @@ -194,13 +189,8 @@ static inline int radeon_bo_is_static(struct radeon_bo *bo) return 0; } -#ifdef RADEON_DEBUG_BO -#define radeon_bo_open(bom, h, s, a, d, f, u)\ - _radeon_bo_open(bom, h, s, a, d, f, u, __FILE__, __FUNCTION__, __LINE__) -#else #define radeon_bo_open(bom, h, s, a, d, f)\ _radeon_bo_open(bom, h, s, a, d, f, __FILE__, __FUNCTION__, __LINE__) -#endif /* RADEON_DEBUG_BO */ #define radeon_bo_ref(bo)\ _radeon_bo_ref(bo, __FILE__, __FUNCTION__, __LINE__) #define radeon_bo_unref(bo)\ diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c index 992eb4611b..5575da6971 100644 --- a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c +++ b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.c @@ -69,9 +69,6 @@ struct bo_legacy { void *ptr; struct bo_legacy *next, *prev; struct bo_legacy *pnext, *pprev; -#ifdef RADEON_DEBUG_BO - char szBufUsage[16]; -#endif /* RADEON_DEBUG_BO */ }; struct bo_manager_legacy { @@ -289,12 +286,7 @@ static struct bo_legacy *bo_allocate(struct bo_manager_legacy *boml, uint32_t size, uint32_t alignment, uint32_t domains, -#ifdef RADEON_DEBUG_BO - uint32_t flags, - char * szBufUsage) -#else uint32_t flags) -#endif /* RADEON_DEBUG_BO */ { struct bo_legacy *bo_legacy; static int pgsize; @@ -327,10 +319,6 @@ static struct bo_legacy *bo_allocate(struct bo_manager_legacy *boml, bo_legacy->next->prev = bo_legacy; } -#ifdef RADEON_DEBUG_BO - sprintf(bo_legacy->szBufUsage, "%s", szBufUsage); -#endif /* RADEON_DEBUG_BO */ - return bo_legacy; } @@ -429,12 +417,7 @@ static struct radeon_bo *bo_open(struct radeon_bo_manager *bom, uint32_t size, uint32_t alignment, uint32_t domains, -#ifdef RADEON_DEBUG_BO - uint32_t flags, - char * szBufUsage) -#else uint32_t flags) -#endif /* RADEON_DEBUG_BO */ { struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom; struct bo_legacy *bo_legacy; @@ -451,11 +434,7 @@ static struct radeon_bo *bo_open(struct radeon_bo_manager *bom, } return NULL; } -#ifdef RADEON_DEBUG_BO - bo_legacy = bo_allocate(boml, size, alignment, domains, flags, szBufUsage); -#else bo_legacy = bo_allocate(boml, size, alignment, domains, flags); -#endif /* RADEON_DEBUG_BO */ bo_legacy->static_bo = 0; r = legacy_new_handle(boml, &bo_legacy->base.handle); if (r) { @@ -577,6 +556,8 @@ static struct radeon_bo_funcs bo_legacy_funcs = { bo_unmap, NULL, bo_is_static, + NULL, + NULL, }; static int bo_vram_validate(struct radeon_bo *bo, @@ -622,12 +603,34 @@ static int bo_vram_validate(struct radeon_bo *bo, if (bo_legacy->dirty || bo_legacy->tobj->base.dirty_images[0]) { if (IS_R600_CLASS(boml->screen)) { - char *src = bo_legacy->ptr; - char *dst = (char *) boml->screen->driScreen->pFB + - (bo_legacy->offset - boml->fb_location); + drm_radeon_texture_t tex; + drm_radeon_tex_image_t tmp; + int ret; - /* FIXME: alignment, pitch, etc. */ - memcpy(dst, src, bo->size); + tex.offset = bo_legacy->offset; + tex.image = &tmp; + assert(!(tex.offset & 1023)); + + tmp.x = 0; + tmp.y = 0; + tmp.width = bo->size; + tmp.height = 1; + tmp.data = bo_legacy->ptr; + tex.format = RADEON_TXFORMAT_ARGB8888; + tex.width = tmp.width; + tex.height = tmp.height; + tex.pitch = bo->size; + do { + ret = drmCommandWriteRead(bo->bom->fd, + DRM_RADEON_TEXTURE, + &tex, + sizeof(drm_radeon_texture_t)); + if (ret) { + if (RADEON_DEBUG & DEBUG_IOCTL) + fprintf(stderr, "DRM_RADEON_TEXTURE: again!\n"); + usleep(1); + } + } while (ret == -EAGAIN); } else { /* Copy to VRAM using a blit. * All memory is 4K aligned. We're using 1024 pixels wide blits. @@ -689,14 +692,8 @@ int radeon_bo_legacy_validate(struct radeon_bo *bo, int retries = 0; if (bo_legacy->map_count) { -#ifdef RADEON_DEBUG_BO - fprintf(stderr, "bo(%p, %d, %s) is mapped (%d) can't valide it.\n", - bo, bo->size, bo_legacy->szBufUsage, bo_legacy->map_count); -#else fprintf(stderr, "bo(%p, %d) is mapped (%d) can't valide it.\n", bo, bo->size, bo_legacy->map_count); -#endif /* RADEON_DEBUG_BO */ - return -EINVAL; } if (bo_legacy->static_bo || bo_legacy->validated) { @@ -768,21 +765,13 @@ void radeon_bo_manager_legacy_dtor(struct radeon_bo_manager *bom) } static struct bo_legacy *radeon_legacy_bo_alloc_static(struct bo_manager_legacy *bom, - int size, -#ifdef RADEON_DEBUG_BO - uint32_t offset, - char * szBufUsage) -#else - uint32_t offset) -#endif /* RADEON_DEBUG_BO */ + int size, + uint32_t offset) { struct bo_legacy *bo; -#ifdef RADEON_DEBUG_BO - bo = bo_allocate(bom, size, 0, RADEON_GEM_DOMAIN_VRAM, 0, szBufUsage); -#else bo = bo_allocate(bom, size, 0, RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ + if (bo == NULL) return NULL; bo->static_bo = 1; @@ -843,11 +832,8 @@ struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *sc size = 4096*4096*4; /* allocate front */ -#ifdef RADEON_DEBUG_BO - bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->frontOffset, "FRONT BUF"); -#else bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->frontOffset); -#endif /* RADEON_DEBUG_BO */ + if (!bo) { radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom); return NULL; @@ -857,11 +843,8 @@ struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *sc } /* allocate back */ -#ifdef RADEON_DEBUG_BO - bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->backOffset, "BACK BUF"); -#else bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->backOffset); -#endif /* RADEON_DEBUG_BO */ + if (!bo) { radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom); return NULL; @@ -871,11 +854,8 @@ struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *sc } /* allocate depth */ -#ifdef RADEON_DEBUG_BO - bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->depthOffset, "Z BUF"); -#else bo = radeon_legacy_bo_alloc_static(bom, size, bom->screen->depthOffset); -#endif /* RADEON_DEBUG_BO */ + if (!bo) { radeon_bo_manager_legacy_dtor((struct radeon_bo_manager*)bom); return NULL; @@ -904,3 +884,29 @@ unsigned radeon_bo_legacy_relocs_size(struct radeon_bo *bo) return bo->size; } +/* + * Fake up a bo for things like texture image_override. + * bo->offset already includes fb_location + */ +struct radeon_bo *radeon_legacy_bo_alloc_fake(struct radeon_bo_manager *bom, + int size, + uint32_t offset) +{ + struct bo_manager_legacy *boml = (struct bo_manager_legacy *)bom; + struct bo_legacy *bo; + + bo = bo_allocate(boml, size, 0, RADEON_GEM_DOMAIN_VRAM, 0); + + if (bo == NULL) + return NULL; + bo->static_bo = 1; + bo->offset = offset; + bo->base.handle = bo->offset; + bo->ptr = boml->screen->driScreen->pFB + (offset - boml->fb_location); + if (bo->base.handle > boml->nhandle) { + boml->nhandle = bo->base.handle + 1; + } + radeon_bo_ref(&(bo->base)); + return &(bo->base); +} + diff --git a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h index 0db817cab0..455adebc09 100644 --- a/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h +++ b/src/mesa/drivers/dri/radeon/radeon_bo_legacy.h @@ -42,5 +42,8 @@ struct radeon_bo_manager *radeon_bo_manager_legacy_ctor(struct radeon_screen *sc void radeon_bo_manager_legacy_dtor(struct radeon_bo_manager *bom); void radeon_bo_legacy_texture_age(struct radeon_bo_manager *bom); unsigned radeon_bo_legacy_relocs_size(struct radeon_bo *bo); +struct radeon_bo *radeon_legacy_bo_alloc_fake(struct radeon_bo_manager *bom, + int size, + uint32_t offset); #endif diff --git a/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h b/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h index e0c70dd9a1..a42870f4a9 100644 --- a/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h +++ b/src/mesa/drivers/dri/radeon/radeon_bocs_wrapper.h @@ -15,6 +15,12 @@ #define RADEON_GEM_DOMAIN_GTT 0x2 // GTT or cache flushed #define RADEON_GEM_DOMAIN_VRAM 0x4 // VRAM domain +#define RADEON_TILING_MACRO 0x1 +#define RADEON_TILING_MICRO 0x2 +#define RADEON_TILING_SWAP 0x4 +#define RADEON_TILING_SURFACE 0x8 /* this object requires a surface + * when mapped - i.e. front buffer */ + /* to be used to build locally in mesa with no libdrm bits */ #include "../radeon/radeon_bo_drm.h" #include "../radeon/radeon_cs_drm.h" diff --git a/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c new file mode 100644 index 0000000000..e8ae51e6ea --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.c @@ -0,0 +1,217 @@ +/* + * Copyright 2009 Maciej Cencora <m.cencora@gmail.com> + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "radeon_buffer_objects.h" + +#include "main/imports.h" +#include "main/mtypes.h" +#include "main/bufferobj.h" + +#include "radeon_common.h" + +struct radeon_buffer_object * +get_radeon_buffer_object(struct gl_buffer_object *obj) +{ + return (struct radeon_buffer_object *) obj; +} + +static struct gl_buffer_object * +radeonNewBufferObject(GLcontext * ctx, + GLuint name, + GLenum target) +{ + struct radeon_buffer_object *obj = CALLOC_STRUCT(radeon_buffer_object); + + _mesa_initialize_buffer_object(&obj->Base, name, target); + + obj->bo = NULL; + + return &obj->Base; +} + +/** + * Called via glDeleteBuffersARB(). + */ +static void +radeonDeleteBufferObject(GLcontext * ctx, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + if (obj->Pointer) { + radeon_bo_unmap(radeon_obj->bo); + } + + if (radeon_obj->bo) { + radeon_bo_unref(radeon_obj->bo); + } + + _mesa_free(radeon_obj); +} + + +/** + * Allocate space for and store data in a buffer object. Any data that was + * previously stored in the buffer object is lost. If data is NULL, + * memory will be allocated, but no copy will occur. + * Called via glBufferDataARB(). + */ +static void +radeonBufferData(GLcontext * ctx, + GLenum target, + GLsizeiptrARB size, + const GLvoid * data, + GLenum usage, + struct gl_buffer_object *obj) +{ + radeonContextPtr radeon = RADEON_CONTEXT(ctx); + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + radeon_obj->Base.Size = size; + radeon_obj->Base.Usage = usage; + + if (radeon_obj->bo != NULL) { + radeon_bo_unref(radeon_obj->bo); + radeon_obj->bo = NULL; + } + + if (size != 0) { + radeon_obj->bo = radeon_bo_open(radeon->radeonScreen->bom, + 0, + size, + 32, + RADEON_GEM_DOMAIN_GTT, + 0); + + if (data != NULL) { + radeon_bo_map(radeon_obj->bo, GL_TRUE); + + _mesa_memcpy(radeon_obj->bo->ptr, data, size); + + radeon_bo_unmap(radeon_obj->bo); + } + } +} + +/** + * Replace data in a subrange of buffer object. If the data range + * specified by size + offset extends beyond the end of the buffer or + * if data is NULL, no copy is performed. + * Called via glBufferSubDataARB(). + */ +static void +radeonBufferSubData(GLcontext * ctx, + GLenum target, + GLintptrARB offset, + GLsizeiptrARB size, + const GLvoid * data, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + radeon_bo_map(radeon_obj->bo, GL_TRUE); + + _mesa_memcpy(radeon_obj->bo->ptr + offset, data, size); + + radeon_bo_unmap(radeon_obj->bo); +} + +/** + * Called via glGetBufferSubDataARB() + */ +static void +radeonGetBufferSubData(GLcontext * ctx, + GLenum target, + GLintptrARB offset, + GLsizeiptrARB size, + GLvoid * data, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + radeon_bo_map(radeon_obj->bo, GL_FALSE); + + _mesa_memcpy(data, radeon_obj->bo->ptr + offset, size); + + radeon_bo_unmap(radeon_obj->bo); +} + +/** + * Called via glMapBufferARB() + */ +static void * +radeonMapBuffer(GLcontext * ctx, + GLenum target, + GLenum access, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + if (access == GL_WRITE_ONLY_ARB) { + ctx->Driver.Flush(ctx); + } + + if (radeon_obj->bo == NULL) { + obj->Pointer = NULL; + return NULL; + } + + radeon_bo_map(radeon_obj->bo, access == GL_WRITE_ONLY_ARB); + + return obj->Pointer = radeon_obj->bo->ptr; +} + + +/** + * Called via glUnmapBufferARB() + */ +static GLboolean +radeonUnmapBuffer(GLcontext * ctx, + GLenum target, + struct gl_buffer_object *obj) +{ + struct radeon_buffer_object *radeon_obj = get_radeon_buffer_object(obj); + + if (radeon_obj->bo != NULL) { + radeon_bo_unmap(radeon_obj->bo); + obj->Pointer = NULL; + } + + return GL_TRUE; +} + +void +radeonInitBufferObjectFuncs(struct dd_function_table *functions) +{ + functions->NewBufferObject = radeonNewBufferObject; + functions->DeleteBuffer = radeonDeleteBufferObject; + functions->BufferData = radeonBufferData; + functions->BufferSubData = radeonBufferSubData; + functions->GetBufferSubData = radeonGetBufferSubData; + functions->MapBuffer = radeonMapBuffer; + functions->UnmapBuffer = radeonUnmapBuffer; +} diff --git a/src/mesa/drivers/dri/radeon/radeon_buffer_objects.h b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.h new file mode 100644 index 0000000000..d681960825 --- /dev/null +++ b/src/mesa/drivers/dri/radeon/radeon_buffer_objects.h @@ -0,0 +1,52 @@ +/* + * Copyright 2009 Maciej Cencora <m.cencora@gmail.com> + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef RADEON_BUFFER_OBJECTS_H +#define RADEON_BUFFER_OBJECTS_H + +#include "main/mtypes.h" + +struct radeon_bo; + +/** + * Radeon vertex/pixel buffer object, derived from Mesa's gl_buffer_object. + */ +struct radeon_buffer_object +{ + struct gl_buffer_object Base; + struct radeon_bo *bo; +}; + +struct radeon_buffer_object * +get_radeon_buffer_object(struct gl_buffer_object *obj); + +/** + * Hook the bufferobject implementation into mesa: + */ +void radeonInitBufferObjectFuncs(struct dd_function_table *functions); + +#endif diff --git a/src/mesa/drivers/dri/radeon/radeon_chipset.h b/src/mesa/drivers/dri/radeon/radeon_chipset.h index 0a6a2df35b..a275c8fb14 100644 --- a/src/mesa/drivers/dri/radeon/radeon_chipset.h +++ b/src/mesa/drivers/dri/radeon/radeon_chipset.h @@ -356,11 +356,14 @@ #define PCI_CHIP_RV770_947A 0x947A #define PCI_CHIP_RV770_947B 0x947B +#define PCI_CHIP_RV730_9480 0x9480 #define PCI_CHIP_RV730_9487 0x9487 +#define PCI_CHIP_RV730_9488 0x9488 #define PCI_CHIP_RV730_9489 0x9489 #define PCI_CHIP_RV730_948F 0x948F #define PCI_CHIP_RV730_9490 0x9490 #define PCI_CHIP_RV730_9491 0x9491 +#define PCI_CHIP_RV730_9495 0x9495 #define PCI_CHIP_RV730_9498 0x9498 #define PCI_CHIP_RV730_949C 0x949C #define PCI_CHIP_RV730_949E 0x949E @@ -374,12 +377,16 @@ #define PCI_CHIP_RV710_9552 0x9552 #define PCI_CHIP_RV710_9553 0x9553 #define PCI_CHIP_RV710_9555 0x9555 +#define PCI_CHIP_RV710_9557 0x9557 #define PCI_CHIP_RV740_94A0 0x94A0 #define PCI_CHIP_RV740_94A1 0x94A1 +#define PCI_CHIP_RV740_94A3 0x94A3 #define PCI_CHIP_RV740_94B1 0x94B1 #define PCI_CHIP_RV740_94B3 0x94B3 +#define PCI_CHIP_RV740_94B4 0x94B4 #define PCI_CHIP_RV740_94B5 0x94B5 +#define PCI_CHIP_RV740_94B9 0x94B9 enum { CHIP_FAMILY_R100, diff --git a/src/mesa/drivers/dri/radeon/radeon_common.c b/src/mesa/drivers/dri/radeon/radeon_common.c index dde615a4d9..0614c89459 100644 --- a/src/mesa/drivers/dri/radeon/radeon_common.c +++ b/src/mesa/drivers/dri/radeon/radeon_common.c @@ -481,32 +481,6 @@ void radeonCopyBuffer( __DRIdrawablePrivate *dPriv, if (!n) continue; - if (IS_R600_CLASS(rmesa->radeonScreen)) { - int cpp = rmesa->radeonScreen->cpp; - int src_pitch = rmesa->radeonScreen->backPitch * cpp; - int dst_pitch = rmesa->radeonScreen->frontPitch * cpp; - char *src = (char *)rmesa->radeonScreen->driScreen->pFB + rmesa->radeonScreen->backOffset; - char *dst = (char *)rmesa->radeonScreen->driScreen->pFB + rmesa->radeonScreen->frontOffset; - int j; - drm_clip_rect_t *pb = rmesa->sarea->boxes; - - for (j = 0; j < n; j++) { - int x = pb[j].x1; - int y = pb[j].y1; - int w = pb[j].x2 - x; - int h = pb[j].y2 - y; - - src += (y * src_pitch) + (x * cpp); - dst += (y * dst_pitch) + (x * cpp); - - while (h--) { - memcpy(dst, src, w * cpp); - src += src_pitch; - dst += dst_pitch; - } - } - } - ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP ); if ( ret ) { @@ -887,10 +861,11 @@ void radeonUpdatePageFlipping(radeonContextPtr radeon) void radeon_window_moved(radeonContextPtr radeon) { + /* Cliprects has to be updated before doing anything else */ + radeonSetCliprects(radeon); if (!radeon->radeonScreen->driScreen->dri2.enabled) { radeonUpdatePageFlipping(radeon); } - radeonSetCliprects(radeon); } void radeon_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei height) @@ -905,7 +880,7 @@ void radeon_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei he if (!radeon->meta.internal_viewport_call && ctx->DrawBuffer->Name == 0) { if (radeon->is_front_buffer_rendering) { - radeonFlush(ctx); + ctx->Driver.Flush(ctx); } radeon_update_renderbuffers(driContext, driContext->driDrawablePriv); if (driContext->driDrawablePriv != driContext->driReadablePriv) @@ -1092,7 +1067,7 @@ void radeonFlush(GLcontext *ctx) * each of N places that do rendering. This has worse performances, * but it is much easier to get correct. */ - if (radeon->is_front_buffer_rendering) { + if (!radeon->is_front_buffer_rendering) { radeon->front_buffer_dirty = GL_FALSE; } } @@ -1236,7 +1211,7 @@ void rcommonInitCmdBuf(radeonContextPtr rmesa) rmesa->cmdbuf.size = size; radeon_cs_space_set_flush(rmesa->cmdbuf.cs, - (void (*)(void *))radeonFlush, rmesa->glCtx); + (void (*)(void *))rmesa->glCtx->Driver.Flush, rmesa->glCtx); if (!rmesa->radeonScreen->kernel_mm) { radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, rmesa->radeonScreen->texSize[0]); diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.c b/src/mesa/drivers/dri/radeon/radeon_common_context.c index 285e015c92..c0abcbfa21 100644 --- a/src/mesa/drivers/dri/radeon/radeon_common_context.c +++ b/src/mesa/drivers/dri/radeon/radeon_common_context.c @@ -85,6 +85,17 @@ static const char* get_chip_family_name(int chip_family) case CHIP_FAMILY_R580: return "R580"; case CHIP_FAMILY_RV560: return "RV560"; case CHIP_FAMILY_RV570: return "RV570"; + case CHIP_FAMILY_R600: return "R600"; + case CHIP_FAMILY_RV610: return "RV610"; + case CHIP_FAMILY_RV630: return "RV630"; + case CHIP_FAMILY_RV670: return "RV670"; + case CHIP_FAMILY_RV620: return "RV620"; + case CHIP_FAMILY_RV635: return "RV635"; + case CHIP_FAMILY_RS780: return "RS780"; + case CHIP_FAMILY_RV770: return "RV770"; + case CHIP_FAMILY_RV730: return "RV730"; + case CHIP_FAMILY_RV710: return "RV710"; + case CHIP_FAMILY_RV740: return "RV740"; default: return "unknown"; } } @@ -200,6 +211,7 @@ GLboolean radeonInitContext(radeonContextPtr radeon, radeon->dri.screen = sPriv; radeon->dri.hwContext = driContextPriv->hHWContext; radeon->dri.hwLock = &sPriv->pSAREA->lock; + radeon->dri.hwLockCount = 0; radeon->dri.fd = sPriv->fd; radeon->dri.drmMinor = sPriv->drm_version.minor; @@ -230,7 +242,26 @@ GLboolean radeonInitContext(radeonContextPtr radeon, radeon->texture_depth = ( glVisual->rgbBits > 16 ) ? DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16; - radeon->texture_row_align = 32; + if (IS_R600_CLASS(radeon->radeonScreen)) { + radeon->texture_row_align = 256; + radeon->texture_rect_row_align = 256; + radeon->texture_compressed_row_align = 256; + } else if (IS_R200_CLASS(radeon->radeonScreen) || + IS_R100_CLASS(radeon->radeonScreen)) { + radeon->texture_row_align = 32; + radeon->texture_rect_row_align = 64; + radeon->texture_compressed_row_align = 32; + } else { /* R300 - not sure this is all correct */ + int chip_family = radeon->radeonScreen->chip_family; + if (chip_family == CHIP_FAMILY_RS600 || + chip_family == CHIP_FAMILY_RS690 || + chip_family == CHIP_FAMILY_RS740) + radeon->texture_row_align = 64; + else + radeon->texture_row_align = 32; + radeon->texture_rect_row_align = 64; + radeon->texture_compressed_row_align = 64; + } return GL_TRUE; } @@ -264,11 +295,10 @@ void radeonDestroyContext(__DRIcontextPrivate *driContextPriv ) GET_CURRENT_CONTEXT(ctx); radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate; radeonContextPtr current = ctx ? RADEON_CONTEXT(ctx) : NULL; - - /* +r6/r7 */ - __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) /* +r6/r7 */ + __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private); - /* --------- */ +#endif if (radeon == current) { radeon_firevertices(radeon); @@ -276,16 +306,7 @@ void radeonDestroyContext(__DRIcontextPrivate *driContextPriv ) } assert(radeon); - if (radeon) - { - -#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) /* +r6/r7 */ - if (IS_R600_CLASS(screen)) - { - r600DestroyContext(driContextPriv); - } -#endif - + if (radeon) { if (radeon->dma.current) { rcommonFlushCmdBuf( radeon, __FUNCTION__ ); } @@ -356,88 +377,48 @@ radeon_make_kernel_renderbuffer_current(radeonContextPtr radeon, if ((rb = (void *)draw->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer)) { if (!rb->bo) { -#ifdef RADEON_DEBUG_BO - rb->bo = radeon_bo_open(radeon->radeonScreen->bom, - radeon->radeonScreen->frontOffset, - 0, - 0, - RADEON_GEM_DOMAIN_VRAM, - 0, - "Front Buf"); -#else rb->bo = radeon_bo_open(radeon->radeonScreen->bom, radeon->radeonScreen->frontOffset, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ } rb->cpp = radeon->radeonScreen->cpp; rb->pitch = radeon->radeonScreen->frontPitch * rb->cpp; } if ((rb = (void *)draw->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer)) { if (!rb->bo) { -#ifdef RADEON_DEBUG_BO rb->bo = radeon_bo_open(radeon->radeonScreen->bom, radeon->radeonScreen->backOffset, 0, 0, RADEON_GEM_DOMAIN_VRAM, - 0, - "Back Buf"); -#else - rb->bo = radeon_bo_open(radeon->radeonScreen->bom, - radeon->radeonScreen->backOffset, - 0, - 0, - RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ } rb->cpp = radeon->radeonScreen->cpp; rb->pitch = radeon->radeonScreen->backPitch * rb->cpp; } if ((rb = (void *)draw->base.Attachment[BUFFER_DEPTH].Renderbuffer)) { if (!rb->bo) { -#ifdef RADEON_DEBUG_BO - rb->bo = radeon_bo_open(radeon->radeonScreen->bom, - radeon->radeonScreen->depthOffset, - 0, - 0, - RADEON_GEM_DOMAIN_VRAM, - 0, - "Z Buf"); -#else rb->bo = radeon_bo_open(radeon->radeonScreen->bom, radeon->radeonScreen->depthOffset, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ } rb->cpp = radeon->radeonScreen->cpp; rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp; } if ((rb = (void *)draw->base.Attachment[BUFFER_STENCIL].Renderbuffer)) { if (!rb->bo) { -#ifdef RADEON_DEBUG_BO - rb->bo = radeon_bo_open(radeon->radeonScreen->bom, - radeon->radeonScreen->depthOffset, - 0, - 0, - RADEON_GEM_DOMAIN_VRAM, - 0, - "Stencil Buf"); -#else rb->bo = radeon_bo_open(radeon->radeonScreen->bom, radeon->radeonScreen->depthOffset, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ } rb->cpp = radeon->radeonScreen->cpp; rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp; @@ -460,16 +441,6 @@ radeon_make_renderbuffer_current(radeonContextPtr radeon, if ((rb = (void *)draw->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer)) { if (!rb->bo) { -#ifdef RADEON_DEBUG_BO - rb->bo = radeon_bo_open(radeon->radeonScreen->bom, - radeon->radeonScreen->frontOffset + - radeon->radeonScreen->fbLocation, - size, - 4096, - RADEON_GEM_DOMAIN_VRAM, - 0, - "Front Buf"); -#else rb->bo = radeon_bo_open(radeon->radeonScreen->bom, radeon->radeonScreen->frontOffset + radeon->radeonScreen->fbLocation, @@ -477,23 +448,12 @@ radeon_make_renderbuffer_current(radeonContextPtr radeon, 4096, RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ } rb->cpp = radeon->radeonScreen->cpp; rb->pitch = radeon->radeonScreen->frontPitch * rb->cpp; } if ((rb = (void *)draw->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer)) { if (!rb->bo) { -#ifdef RADEON_DEBUG_BO - rb->bo = radeon_bo_open(radeon->radeonScreen->bom, - radeon->radeonScreen->backOffset + - radeon->radeonScreen->fbLocation, - size, - 4096, - RADEON_GEM_DOMAIN_VRAM, - 0, - "Back Buf"); -#else rb->bo = radeon_bo_open(radeon->radeonScreen->bom, radeon->radeonScreen->backOffset + radeon->radeonScreen->fbLocation, @@ -501,55 +461,32 @@ radeon_make_renderbuffer_current(radeonContextPtr radeon, 4096, RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ } rb->cpp = radeon->radeonScreen->cpp; rb->pitch = radeon->radeonScreen->backPitch * rb->cpp; } if ((rb = (void *)draw->base.Attachment[BUFFER_DEPTH].Renderbuffer)) { if (!rb->bo) { -#ifdef RADEON_DEBUG_BO rb->bo = radeon_bo_open(radeon->radeonScreen->bom, radeon->radeonScreen->depthOffset + radeon->radeonScreen->fbLocation, size, 4096, RADEON_GEM_DOMAIN_VRAM, - 0, - "Z Buf"); -#else - rb->bo = radeon_bo_open(radeon->radeonScreen->bom, - radeon->radeonScreen->depthOffset + - radeon->radeonScreen->fbLocation, - size, - 4096, - RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ } rb->cpp = radeon->radeonScreen->cpp; rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp; } if ((rb = (void *)draw->base.Attachment[BUFFER_STENCIL].Renderbuffer)) { if (!rb->bo) { -#ifdef RADEON_DEBUG_BO rb->bo = radeon_bo_open(radeon->radeonScreen->bom, radeon->radeonScreen->depthOffset + radeon->radeonScreen->fbLocation, size, 4096, RADEON_GEM_DOMAIN_VRAM, - 0, - "Stencil Buf"); -#else - rb->bo = radeon_bo_open(radeon->radeonScreen->bom, - radeon->radeonScreen->depthOffset + - radeon->radeonScreen->fbLocation, - size, - 4096, - RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ } rb->cpp = radeon->radeonScreen->cpp; rb->pitch = radeon->radeonScreen->depthPitch * rb->cpp; @@ -732,28 +669,29 @@ radeon_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable) bo = depth_bo; radeon_bo_ref(bo); } else { -#ifdef RADEON_DEBUG_BO - bo = radeon_bo_open(radeon->radeonScreen->bom, - buffers[i].name, - 0, - 0, - RADEON_GEM_DOMAIN_VRAM, - buffers[i].flags, - regname); -#else + uint32_t tiling_flags = 0, pitch = 0; + int ret; + bo = radeon_bo_open(radeon->radeonScreen->bom, buffers[i].name, 0, 0, RADEON_GEM_DOMAIN_VRAM, buffers[i].flags); -#endif /* RADEON_DEBUG_BO */ + if (bo == NULL) { fprintf(stderr, "failed to attach %s %d\n", regname, buffers[i].name); } + + ret = radeon_bo_get_tiling(bo, &tiling_flags, &pitch); + if (tiling_flags & RADEON_TILING_MACRO) + bo->flags |= RADEON_BO_FLAGS_MACRO_TILE; + if (tiling_flags & RADEON_TILING_MICRO) + bo->flags |= RADEON_BO_FLAGS_MICRO_TILE; + } if (buffers[i].attachment == __DRI_BUFFER_DEPTH) { diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.h b/src/mesa/drivers/dri/radeon/radeon_common_context.h index e4a8da0596..ee46c1f81a 100644 --- a/src/mesa/drivers/dri/radeon/radeon_common_context.h +++ b/src/mesa/drivers/dri/radeon/radeon_common_context.h @@ -239,6 +239,11 @@ struct radeon_tex_obj { GLuint SQ_TEX_SAMPLER1; GLuint SQ_TEX_SAMPLER2; + GLuint TD_PS_SAMPLER0_BORDER_RED; + GLuint TD_PS_SAMPLER0_BORDER_GREEN; + GLuint TD_PS_SAMPLER0_BORDER_BLUE; + GLuint TD_PS_SAMPLER0_BORDER_ALPHA; + GLboolean border_fallback; @@ -360,6 +365,7 @@ struct radeon_dri_mirror { drm_context_t hwContext; drm_hw_lock_t *hwLock; + int hwLockCount; int fd; int drmMinor; }; @@ -424,6 +430,8 @@ struct radeon_context { int texture_depth; float initialMaxAnisotropy; uint32_t texture_row_align; + uint32_t texture_rect_row_align; + uint32_t texture_compressed_row_align; struct radeon_dma dma; struct radeon_hw_state hw; @@ -438,7 +446,6 @@ struct radeon_context { GLuint numClipRects; /* Cliprects for the draw buffer */ drm_clip_rect_t *pClipRects; unsigned int lastStamp; - GLboolean lost_context; drm_radeon_sarea_t *sarea; /* Private SAREA data */ /* Mirrors of some DRI state */ @@ -587,10 +594,4 @@ extern int RADEON_DEBUG; #define RADEON_DEBUG 0 #endif -#ifndef HAVE_LIBDRM_RADEON -#ifndef RADEON_DEBUG_BO -#define RADEON_DEBUG_BO 1 -#endif -#endif - #endif diff --git a/src/mesa/drivers/dri/radeon/radeon_cs_space_drm.c b/src/mesa/drivers/dri/radeon/radeon_cs_space_drm.c index 5a8df7bb8c..89cbbb5a6b 100644 --- a/src/mesa/drivers/dri/radeon/radeon_cs_space_drm.c +++ b/src/mesa/drivers/dri/radeon/radeon_cs_space_drm.c @@ -82,7 +82,7 @@ static inline int radeon_cs_setup_bo(struct radeon_cs_space_check *sc, struct ra if (write_domain == RADEON_GEM_DOMAIN_VRAM) { sizes->op_read -= bo->size; sizes->op_vram_write += bo->size; - } else if (write_domain == RADEON_GEM_DOMAIN_VRAM) { + } else if (write_domain == RADEON_GEM_DOMAIN_GTT) { sizes->op_read -= bo->size; sizes->op_gart_write += bo->size; } diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.c b/src/mesa/drivers/dri/radeon/radeon_dma.c index 48114a0012..5e755c51ed 100644 --- a/src/mesa/drivers/dri/radeon/radeon_dma.c +++ b/src/mesa/drivers/dri/radeon/radeon_dma.c @@ -52,7 +52,7 @@ do { \ } while (0) #endif -static void radeonEmitVec4(uint32_t *out, GLvoid * data, int stride, int count) +void radeonEmitVec4(uint32_t *out, const GLvoid * data, int stride, int count) { int i; @@ -70,7 +70,7 @@ static void radeonEmitVec4(uint32_t *out, GLvoid * data, int stride, int count) } } -void radeonEmitVec8(uint32_t *out, GLvoid * data, int stride, int count) +void radeonEmitVec8(uint32_t *out, const GLvoid * data, int stride, int count) { int i; @@ -89,7 +89,7 @@ void radeonEmitVec8(uint32_t *out, GLvoid * data, int stride, int count) } } -void radeonEmitVec12(uint32_t *out, GLvoid * data, int stride, int count) +void radeonEmitVec12(uint32_t *out, const GLvoid * data, int stride, int count) { int i; @@ -110,7 +110,7 @@ void radeonEmitVec12(uint32_t *out, GLvoid * data, int stride, int count) } } -static void radeonEmitVec16(uint32_t *out, GLvoid * data, int stride, int count) +void radeonEmitVec16(uint32_t *out, const GLvoid * data, int stride, int count) { int i; @@ -132,7 +132,7 @@ static void radeonEmitVec16(uint32_t *out, GLvoid * data, int stride, int count) } void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos, - GLvoid * data, int size, int stride, int count) + const GLvoid * data, int size, int stride, int count) { radeonContextPtr rmesa = RADEON_CONTEXT(ctx); uint32_t *out; @@ -185,15 +185,9 @@ void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size) } again_alloc: -#ifdef RADEON_DEBUG_BO - rmesa->dma.current = radeon_bo_open(rmesa->radeonScreen->bom, - 0, size, 4, RADEON_GEM_DOMAIN_GTT, - 0, "dma.current"); -#else rmesa->dma.current = radeon_bo_open(rmesa->radeonScreen->bom, 0, size, 4, RADEON_GEM_DOMAIN_GTT, 0); -#endif /* RADEON_DEBUG_BO */ if (!rmesa->dma.current) { rcommonFlushCmdBuf(rmesa, __FUNCTION__); diff --git a/src/mesa/drivers/dri/radeon/radeon_dma.h b/src/mesa/drivers/dri/radeon/radeon_dma.h index 06e388fc1d..55509ed00c 100644 --- a/src/mesa/drivers/dri/radeon/radeon_dma.h +++ b/src/mesa/drivers/dri/radeon/radeon_dma.h @@ -33,11 +33,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #ifndef RADEON_DMA_H #define RADEON_DMA_H -void radeonEmitVec8(uint32_t *out, GLvoid * data, int stride, int count); -void radeonEmitVec12(uint32_t *out, GLvoid * data, int stride, int count); +void radeonEmitVec4(uint32_t *out, const GLvoid * data, int stride, int count); +void radeonEmitVec8(uint32_t *out, const GLvoid * data, int stride, int count); +void radeonEmitVec12(uint32_t *out, const GLvoid * data, int stride, int count); +void radeonEmitVec16(uint32_t *out, const GLvoid * data, int stride, int count); void rcommon_emit_vector(GLcontext * ctx, struct radeon_aos *aos, - GLvoid * data, int size, int stride, int count); + const GLvoid * data, int size, int stride, int count); void radeonRefillCurrentDmaRegion(radeonContextPtr rmesa, int size); void radeonAllocDmaRegion(radeonContextPtr rmesa, diff --git a/src/mesa/drivers/dri/radeon/radeon_fbo.c b/src/mesa/drivers/dri/radeon/radeon_fbo.c index f28efa33e9..8303917b0b 100644 --- a/src/mesa/drivers/dri/radeon/radeon_fbo.c +++ b/src/mesa/drivers/dri/radeon/radeon_fbo.c @@ -185,22 +185,12 @@ radeon_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb, rrb->pitch = pitch * cpp; rrb->cpp = cpp; -#ifdef RADEON_DEBUG_BO - rrb->bo = radeon_bo_open(radeon->radeonScreen->bom, - 0, - size, - 0, - RADEON_GEM_DOMAIN_VRAM, - 0, - "Radeon RBO"); -#else rrb->bo = radeon_bo_open(radeon->radeonScreen->bom, 0, size, 0, RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ rb->Width = width; rb->Height = height; return GL_TRUE; @@ -291,7 +281,7 @@ radeon_create_renderbuffer(GLenum format, __DRIdrawablePrivate *driDrawPriv) rrb->base.RedBits = 8; rrb->base.GreenBits = 8; rrb->base.BlueBits = 8; - rrb->base.AlphaBits = 8; + rrb->base.AlphaBits = 0; rrb->base.DataType = GL_UNSIGNED_BYTE; break; case GL_RGBA8: @@ -407,7 +397,7 @@ restart: rrb->cpp = 2; rrb->base._ActualFormat = GL_RGB5; rrb->base._BaseFormat = GL_RGB; - rrb->base.DataType = GL_UNSIGNED_SHORT; + rrb->base.DataType = GL_UNSIGNED_BYTE; DBG("Render to RGB5 texture OK\n"); } else if (texImage->TexFormat == &_mesa_texformat_argb1555) { diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.c b/src/mesa/drivers/dri/radeon/radeon_lock.c index 5774f7ebcf..6294b7e42b 100644 --- a/src/mesa/drivers/dri/radeon/radeon_lock.c +++ b/src/mesa/drivers/dri/radeon/radeon_lock.c @@ -85,11 +85,35 @@ void radeonGetLock(radeonContextPtr rmesa, GLuint flags) } rmesa->vtbl.get_lock(rmesa); - - rmesa->lost_context = GL_TRUE; } - -void radeon_lock_hardware(radeonContextPtr radeon) +#ifndef NDEBUG +struct lock_debug { + const char* function; + const char* file; + int line; +}; + +static struct lock_debug ldebug = {0}; +#endif + +#if 0 +/** TODO: use atomic operations for reference counting **/ +/** gcc 4.2 has builtin functios for this **/ +#define ATOMIC_INC_AND_FETCH(atomic) __sync_add_and_fetch(&atomic, 1) +#define ATOMIC_DEC_AND_FETCH(atomic) __sync_sub_and_fetch(&atomic, 1) +#else +#define ATOMIC_INC_AND_FETCH(atomic) (++atomic) +#define ATOMIC_DEC_AND_FETCH(atomic) (--atomic) +#endif + + +void radeon_lock_hardware(radeonContextPtr radeon +#ifndef NDEBUG + ,const char* function + ,const char* file + ,const int line +#endif + ) { char ret = 0; struct radeon_framebuffer *rfb = NULL; @@ -104,16 +128,39 @@ void radeon_lock_hardware(radeonContextPtr radeon) } if (!radeon->radeonScreen->driScreen->dri2.enabled) { + if (ATOMIC_INC_AND_FETCH(radeon->dri.hwLockCount) > 1) + { +#ifndef NDEBUG + if ( RADEON_DEBUG & DEBUG_SANITY ) + fprintf(stderr, "*** %d times of recursive call to %s ***\n" + "Original call was from %s (file: %s line: %d)\n" + "Now call is coming from %s (file: %s line: %d)\n" + , radeon->dri.hwLockCount, __FUNCTION__ + , ldebug.function, ldebug.file, ldebug.line + , function, file, line + ); +#endif + return; + } DRM_CAS(radeon->dri.hwLock, radeon->dri.hwContext, (DRM_LOCK_HELD | radeon->dri.hwContext), ret ); if (ret) radeonGetLock(radeon, 0); +#ifndef NDEBUG + ldebug.function = function; + ldebug.file = file; + ldebug.line = line; +#endif } } void radeon_unlock_hardware(radeonContextPtr radeon) { if (!radeon->radeonScreen->driScreen->dri2.enabled) { + if (ATOMIC_DEC_AND_FETCH(radeon->dri.hwLockCount) > 0) + { + return; + } DRM_UNLOCK( radeon->dri.fd, radeon->dri.hwLock, radeon->dri.hwContext ); diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.h b/src/mesa/drivers/dri/radeon/radeon_lock.h index 2817709eed..da5a5b4371 100644 --- a/src/mesa/drivers/dri/radeon/radeon_lock.h +++ b/src/mesa/drivers/dri/radeon/radeon_lock.h @@ -48,12 +48,22 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags); -void radeon_lock_hardware(radeonContextPtr rmesa); +void radeon_lock_hardware(radeonContextPtr rmesa +#ifndef NDEBUG + ,const char* function + ,const char* file + ,const int line +#endif + ); void radeon_unlock_hardware(radeonContextPtr rmesa); /* Lock the hardware and validate our state. */ +#ifdef NDEBUG #define LOCK_HARDWARE( rmesa ) radeon_lock_hardware(rmesa) +#else +#define LOCK_HARDWARE( rmesa ) radeon_lock_hardware(rmesa, __FUNCTION__, __FILE__, __LINE__) +#endif #define UNLOCK_HARDWARE( rmesa ) radeon_unlock_hardware(rmesa) #endif diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c index f04a07fecd..eba9f5857f 100644 --- a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c +++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c @@ -90,16 +90,18 @@ static void compute_tex_image_offset(radeonContextPtr rmesa, radeon_mipmap_tree GLuint face, GLuint level, GLuint* curOffset) { radeon_mipmap_level *lvl = &mt->levels[level]; - uint32_t row_align = rmesa->texture_row_align - 1; + uint32_t row_align; /* Find image size in bytes */ if (mt->compressed) { /* TODO: Is this correct? Need test cases for compressed textures! */ - lvl->rowstride = (lvl->width * mt->bpp + 63) & ~63; + row_align = rmesa->texture_compressed_row_align - 1; + lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align; lvl->size = radeon_compressed_texture_size(mt->radeon->glCtx, lvl->width, lvl->height, lvl->depth, mt->compressed); } else if (mt->target == GL_TEXTURE_RECTANGLE_NV) { - lvl->rowstride = (lvl->width * mt->bpp + 63) & ~63; + row_align = rmesa->texture_rect_row_align - 1; + lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align; lvl->size = lvl->rowstride * lvl->height; } else if (mt->tilebits & RADEON_TXO_MICRO_TILE) { /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned, @@ -108,6 +110,7 @@ static void compute_tex_image_offset(radeonContextPtr rmesa, radeon_mipmap_tree lvl->rowstride = (lvl->width * mt->bpp * 2 + 31) & ~31; lvl->size = lvl->rowstride * ((lvl->height + 1) / 2) * lvl->depth; } else { + row_align = rmesa->texture_row_align - 1; lvl->rowstride = (lvl->width * mt->bpp + row_align) & ~row_align; lvl->size = lvl->rowstride * lvl->height * lvl->depth; } @@ -212,18 +215,10 @@ radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa, radeonTexObj * else calculate_miptree_layout_r100(rmesa, mt); -#ifdef RADEON_DEBUG_BO - mt->bo = radeon_bo_open(rmesa->radeonScreen->bom, - 0, mt->totalsize, 1024, - RADEON_GEM_DOMAIN_VRAM, - 0, - "MIPMAP TREE"); -#else mt->bo = radeon_bo_open(rmesa->radeonScreen->bom, 0, mt->totalsize, 1024, RADEON_GEM_DOMAIN_VRAM, 0); -#endif /* RADEON_DEBUG_BO */ return mt; } @@ -323,7 +318,8 @@ GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt, if (face >= mt->faces || level < mt->firstLevel || level > mt->lastLevel) return GL_FALSE; - if (texImage->IsCompressed != mt->compressed) + if ((!texImage->IsCompressed && mt->compressed) || + (texImage->IsCompressed && !mt->compressed)) return GL_FALSE; if (!texImage->IsCompressed && @@ -363,8 +359,8 @@ GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_textu mt->width0 == firstImage->Width && mt->height0 == firstImage->Height && mt->depth0 == firstImage->Depth && - mt->bpp == firstImage->TexFormat->TexelBytes && - mt->compressed == compressed); + mt->compressed == compressed && + (!mt->compressed ? (mt->bpp == firstImage->TexFormat->TexelBytes) : 1)); } diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.c b/src/mesa/drivers/dri/radeon/radeon_screen.c index 290ef2394d..c8d491621a 100644 --- a/src/mesa/drivers/dri/radeon/radeon_screen.c +++ b/src/mesa/drivers/dri/radeon/radeon_screen.c @@ -57,7 +57,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r200_tex.h" #elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) #include "r300_context.h" -#include "r300_fragprog.h" #include "r300_tex.h" #elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) #include "r600_context.h" @@ -150,6 +149,9 @@ extern const struct dri_extension point_extensions[]; #elif RADEON_COMMON && (defined(RADEON_COMMON_FOR_R300) || defined(RADEON_COMMON_FOR_R600)) +#define DRI_CONF_FP_OPTIMIZATION_SPEED 0 +#define DRI_CONF_FP_OPTIMIZATION_QUALITY 1 + /* TODO: integrate these into xmlpool.h! */ #define DRI_CONF_MAX_TEXTURE_IMAGE_UNITS(def,min,max) \ DRI_CONF_OPT_BEGIN_V(texture_image_units,int,def, # min ":" # max ) \ @@ -876,11 +878,14 @@ static int radeon_set_screen_flags(radeonScreenPtr screen, int device_id) screen->chip_flags = RADEON_CHIPSET_TCL; break; + case PCI_CHIP_RV730_9480: case PCI_CHIP_RV730_9487: + case PCI_CHIP_RV730_9488: case PCI_CHIP_RV730_9489: case PCI_CHIP_RV730_948F: case PCI_CHIP_RV730_9490: case PCI_CHIP_RV730_9491: + case PCI_CHIP_RV730_9495: case PCI_CHIP_RV730_9498: case PCI_CHIP_RV730_949C: case PCI_CHIP_RV730_949E: @@ -897,15 +902,19 @@ static int radeon_set_screen_flags(radeonScreenPtr screen, int device_id) case PCI_CHIP_RV710_9552: case PCI_CHIP_RV710_9553: case PCI_CHIP_RV710_9555: + case PCI_CHIP_RV710_9557: screen->chip_family = CHIP_FAMILY_RV710; screen->chip_flags = RADEON_CHIPSET_TCL; break; case PCI_CHIP_RV740_94A0: case PCI_CHIP_RV740_94A1: + case PCI_CHIP_RV740_94A3: case PCI_CHIP_RV740_94B1: case PCI_CHIP_RV740_94B3: + case PCI_CHIP_RV740_94B4: case PCI_CHIP_RV740_94B5: + case PCI_CHIP_RV740_94B9: screen->chip_family = CHIP_FAMILY_RV740; screen->chip_flags = RADEON_CHIPSET_TCL; break; @@ -990,6 +999,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) screen->drmSupportsPointSprites = (sPriv->drm_version.minor >= 13); screen->drmSupportsCubeMapsR100 = (sPriv->drm_version.minor >= 15); screen->drmSupportsVertexProgram = (sPriv->drm_version.minor >= 25); + screen->drmSupportsOcclusionQueries = (sPriv->drm_version.minor >= 30); } ret = radeon_set_screen_flags(screen, dri_priv->deviceID); @@ -1025,7 +1035,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) ((GLubyte *)screen->status.map + RADEON_SCRATCH_REG_OFFSET); else screen->scratch = (__volatile__ uint32_t *) - ((GLubyte *)screen->status.map + RADEON_SCRATCH_REG_OFFSET); + ((GLubyte *)screen->status.map + R600_SCRATCH_REG_OFFSET); screen->buffers = drmMapBufs( sPriv->fd ); if ( !screen->buffers ) { @@ -1085,7 +1095,7 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) /* +r6/r7 */ if(screen->chip_family >= CHIP_FAMILY_R600) { - if (ret) + if (ret) { FREE( screen ); fprintf(stderr, "Unable to get fb location need newer drm\n"); @@ -1098,18 +1108,18 @@ radeonCreateScreen( __DRIscreenPrivate *sPriv ) } else { - if (ret) + if (ret) { if (screen->chip_family < CHIP_FAMILY_RS600 && !screen->kernel_mm) screen->fbLocation = ( INREG( RADEON_MC_FB_LOCATION ) & 0xffff) << 16; - else + else { FREE( screen ); fprintf(stderr, "Unable to get fb location need newer drm\n"); return NULL; } - } - else + } + else { screen->fbLocation = (temp & 0xffff) << 16; } @@ -1289,6 +1299,7 @@ radeonCreateScreen2(__DRIscreenPrivate *sPriv) screen->drmSupportsPointSprites = 1; screen->drmSupportsCubeMapsR100 = 1; screen->drmSupportsVertexProgram = 1; + screen->drmSupportsOcclusionQueries = 1; screen->irq = 1; ret = radeonGetParam(sPriv, RADEON_PARAM_DEVICE_ID, &device_id); @@ -1577,21 +1588,11 @@ static GLboolean radeonCreateContext(const __GLcontextModes * glVisual, { __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private); -#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) - if (IS_R600_CLASS(screen)) - return r600CreateContext(glVisual, driContextPriv, sharedContextPriv); -#endif - #if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300) if (IS_R300_CLASS(screen)) return r300CreateContext(glVisual, driContextPriv, sharedContextPriv); #endif -#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) - if (IS_R200_CLASS(screen)) - return r200CreateContext(glVisual, driContextPriv, sharedContextPriv); -#endif - #if !RADEON_COMMON (void)screen; return r100CreateContext(glVisual, driContextPriv, sharedContextPriv); @@ -1791,8 +1792,16 @@ getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo ) const struct __DriverAPIRec driDriverAPI = { .InitScreen = radeonInitScreen, .DestroyScreen = radeonDestroyScreen, +#if RADEON_COMMON && defined(RADEON_COMMON_FOR_R200) + .CreateContext = r200CreateContext, + .DestroyContext = r200DestroyContext, +#elif RADEON_COMMON && defined(RADEON_COMMON_FOR_R600) + .CreateContext = r600CreateContext, + .DestroyContext = r600DestroyContext, +#else .CreateContext = radeonCreateContext, .DestroyContext = radeonDestroyContext, +#endif .CreateBuffer = radeonCreateBuffer, .DestroyBuffer = radeonDestroyBuffer, .SwapBuffers = radeonSwapBuffers, diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.h b/src/mesa/drivers/dri/radeon/radeon_screen.h index 2a2f6b1b0b..f0dd46b0b1 100644 --- a/src/mesa/drivers/dri/radeon/radeon_screen.h +++ b/src/mesa/drivers/dri/radeon/radeon_screen.h @@ -99,6 +99,7 @@ typedef struct radeon_screen { GLboolean drmSupportsPointSprites; /* need radeon kernel module >= 1.13 */ GLboolean drmSupportsCubeMapsR100; /* need radeon kernel module >= 1.15 */ GLboolean drmSupportsVertexProgram; /* need radeon kernel module >= 1.25 */ + GLboolean drmSupportsOcclusionQueries; /* need radeon kernel module >= 1.30 */ GLboolean depthHasSurface; /* Configuration cache with default values for all contexts */ diff --git a/src/mesa/drivers/dri/radeon/radeon_span.c b/src/mesa/drivers/dri/radeon/radeon_span.c index b2a468b4fd..5e4bf00d7a 100644 --- a/src/mesa/drivers/dri/radeon/radeon_span.c +++ b/src/mesa/drivers/dri/radeon/radeon_span.c @@ -51,6 +51,59 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. static void radeonSetSpanFunctions(struct radeon_renderbuffer *rrb); + +/* r200 depth buffer is always tiled - this is the formula + according to the docs unless I typo'ed in it +*/ +static GLubyte *r200_depth_2byte(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + GLint offset; + if (rrb->has_surface) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + GLuint b; + offset = 0; + b = (((y >> 4) * (rrb->pitch >> 8) + (x >> 6))); + offset += (b >> 1) << 12; + offset += (((rrb->pitch >> 8) & 0x1) ? (b & 0x1) : ((b & 0x1) ^ ((y >> 4) & 0x1))) << 11; + offset += ((y >> 2) & 0x3) << 9; + offset += ((x >> 3) & 0x1) << 8; + offset += ((x >> 4) & 0x3) << 6; + offset += ((x >> 2) & 0x1) << 5; + offset += ((y >> 1) & 0x1) << 4; + offset += ((x >> 1) & 0x1) << 3; + offset += (y & 0x1) << 2; + offset += (x & 0x1) << 1; + } + return &ptr[offset]; +} + +static GLubyte *r200_depth_4byte(const struct radeon_renderbuffer * rrb, + GLint x, GLint y) +{ + GLubyte *ptr = rrb->bo->ptr; + GLint offset; + if (rrb->has_surface) { + offset = x * rrb->cpp + y * rrb->pitch; + } else { + GLuint b; + offset = 0; + b = (((y & 0x7ff) >> 4) * (rrb->pitch >> 7) + (x >> 5)); + offset += (b >> 1) << 12; + offset += (((rrb->pitch >> 7) & 0x1) ? (b & 0x1) : ((b & 0x1) ^ ((y >> 4) & 0x1))) << 11; + offset += ((y >> 2) & 0x3) << 9; + offset += ((x >> 2) & 0x1) << 8; + offset += ((x >> 3) & 0x3) << 6; + offset += ((y >> 1) & 0x1) << 5; + offset += ((x >> 1) & 0x1) << 4; + offset += (y & 0x1) << 3; + offset += (x & 0x1) << 2; + } + return &ptr[offset]; +} + /* radeon tiling on r300-r500 has 4 states, macro-linear/micro-linear macro-linear/micro-tiled @@ -61,7 +114,6 @@ static void radeonSetSpanFunctions(struct radeon_renderbuffer *rrb); 4 byte surface 8/16 byte (unused) */ - static GLubyte *radeon_ptr_4byte(const struct radeon_renderbuffer * rrb, GLint x, GLint y) { @@ -285,11 +337,21 @@ s8z24_to_z24s8(uint32_t val) */ #define VALUE_TYPE GLushort +#if defined(RADEON_COMMON_FOR_R200) +#define WRITE_DEPTH( _x, _y, d ) \ + *(GLushort *)r200_depth_2byte(rrb, _x + x_off, _y + y_off) = d +#else #define WRITE_DEPTH( _x, _y, d ) \ *(GLushort *)radeon_ptr_2byte_8x2(rrb, _x + x_off, _y + y_off) = d +#endif +#if defined(RADEON_COMMON_FOR_R200) +#define READ_DEPTH( d, _x, _y ) \ + d = *(GLushort *)r200_depth_2byte(rrb, _x + x_off, _y + y_off) +#else #define READ_DEPTH( d, _x, _y ) \ d = *(GLushort *)radeon_ptr_2byte_8x2(rrb, _x + x_off, _y + y_off) +#endif #define TAG(x) radeon##x##_z16 #include "depthtmp.h" @@ -301,7 +363,7 @@ s8z24_to_z24s8(uint32_t val) */ #define VALUE_TYPE GLuint -#ifdef COMPILE_R300 +#if defined(COMPILE_R300) #define WRITE_DEPTH( _x, _y, d ) \ do { \ GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ @@ -310,6 +372,15 @@ do { \ tmp |= ((d << 8) & 0xffffff00); \ *_ptr = tmp; \ } while (0) +#elif defined(RADEON_COMMON_FOR_R200) +#define WRITE_DEPTH( _x, _y, d ) \ +do { \ + GLuint *_ptr = (GLuint*)r200_depth_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = *_ptr; \ + tmp &= 0xff000000; \ + tmp |= ((d) & 0x00ffffff); \ + *_ptr = tmp; \ +} while (0) #else #define WRITE_DEPTH( _x, _y, d ) \ do { \ @@ -321,19 +392,21 @@ do { \ } while (0) #endif -#ifdef COMPILE_R300 +#if defined(COMPILE_R300) #define READ_DEPTH( d, _x, _y ) \ do { \ d = (*(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off)) & 0xffffff00) >> 8; \ }while(0) +#elif defined(RADEON_COMMON_FOR_R200) +#define READ_DEPTH( d, _x, _y ) \ + do { \ + d = *(GLuint*)(r200_depth_4byte(rrb, _x + x_off, _y + y_off)) & 0x00ffffff; \ + }while(0) #else #define READ_DEPTH( d, _x, _y ) \ d = *(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off)) & 0x00ffffff; #endif -/* - fprintf(stderr, "dval(%d, %d, %d, %d)=0x%08X\n", _x, xo, _y, yo, d);\ - d = *(GLuint*)(radeon_ptr(rrb, _x, _y )) & 0x00ffffff; -*/ + #define TAG(x) radeon##x##_z24 #include "depthtmp.h" @@ -345,12 +418,19 @@ do { \ */ #define VALUE_TYPE GLuint -#ifdef COMPILE_R300 +#if defined(COMPILE_R300) #define WRITE_DEPTH( _x, _y, d ) \ do { \ GLuint *_ptr = (GLuint*)radeon_ptr_4byte( rrb, _x + x_off, _y + y_off ); \ *_ptr = d; \ } while (0) +#elif defined(RADEON_COMMON_FOR_R200) +#define WRITE_DEPTH( _x, _y, d ) \ +do { \ + GLuint *_ptr = (GLuint*)r200_depth_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = z24s8_to_s8z24(d); \ + *_ptr = tmp; \ +} while (0) #else #define WRITE_DEPTH( _x, _y, d ) \ do { \ @@ -360,20 +440,22 @@ do { \ } while (0) #endif -#ifdef COMPILE_R300 +#if defined(COMPILE_R300) #define READ_DEPTH( d, _x, _y ) \ do { \ d = (*(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off))); \ }while(0) +#elif defined(RADEON_COMMON_FOR_R200) +#define READ_DEPTH( d, _x, _y ) \ + do { \ + d = s8z24_to_z24s8(*(GLuint*)(r200_depth_4byte(rrb, _x + x_off, _y + y_off))); \ + }while(0) #else #define READ_DEPTH( d, _x, _y ) do { \ d = s8z24_to_z24s8(*(GLuint*)(radeon_ptr_4byte(rrb, _x + x_off, _y + y_off ))); \ } while (0) #endif -/* - fprintf(stderr, "dval(%d, %d, %d, %d)=0x%08X\n", _x, xo, _y, yo, d);\ - d = *(GLuint*)(radeon_ptr(rrb, _x, _y )) & 0x00ffffff; -*/ + #define TAG(x) radeon##x##_z24_s8 #include "depthtmp.h" @@ -392,6 +474,15 @@ do { \ tmp |= (d) & 0xff; \ *_ptr = tmp; \ } while (0) +#elif defined(RADEON_COMMON_FOR_R200) +#define WRITE_STENCIL( _x, _y, d ) \ +do { \ + GLuint *_ptr = (GLuint*)r200_depth_4byte(rrb, _x + x_off, _y + y_off); \ + GLuint tmp = *_ptr; \ + tmp &= 0x00ffffff; \ + tmp |= (((d) & 0xff) << 24); \ + *_ptr = tmp; \ +} while (0) #else #define WRITE_STENCIL( _x, _y, d ) \ do { \ @@ -410,6 +501,13 @@ do { \ GLuint tmp = *_ptr; \ d = tmp & 0x000000ff; \ } while (0) +#elif defined(RADEON_COMMON_FOR_R200) +#define READ_STENCIL( d, _x, _y ) \ +do { \ + GLuint *_ptr = (GLuint*)r200_depth_4byte( rrb, _x + x_off, _y + y_off ); \ + GLuint tmp = *_ptr; \ + d = (tmp & 0xff000000) >> 24; \ +} while (0) #else #define READ_STENCIL( d, _x, _y ) \ do { \ diff --git a/src/mesa/drivers/dri/radeon/radeon_state_init.c b/src/mesa/drivers/dri/radeon/radeon_state_init.c index aaa82b1d6a..57aa7f1ca4 100644 --- a/src/mesa/drivers/dri/radeon/radeon_state_init.c +++ b/src/mesa/drivers/dri/radeon/radeon_state_init.c @@ -423,7 +423,7 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) if (drb) dwords += 6; if (rrb) - dwords += 6; + dwords += 8; BEGIN_BATCH_NO_AUTOSTATE(dwords); /* In the CS case we need to split this up */ @@ -449,7 +449,7 @@ static void ctx_emit_cs(GLcontext *ctx, struct radeon_state_atom *atom) OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); OUT_BATCH(CP_PACKET0(RADEON_RB3D_COLORPITCH, 0)); - OUT_BATCH(cbpitch); + OUT_BATCH_RELOC(cbpitch, rrb->bo, cbpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0); } // if (atom->cmd_size == CTX_STATE_SIZE_NEWDRM) { diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.c b/src/mesa/drivers/dri/radeon/radeon_texture.c index ad501c454c..fa16f44c18 100644 --- a/src/mesa/drivers/dri/radeon/radeon_texture.c +++ b/src/mesa/drivers/dri/radeon/radeon_texture.c @@ -610,9 +610,17 @@ static void radeon_teximage( if (pixels) { radeon_teximage_map(image, GL_TRUE); - if (compressed) { - memcpy(texImage->Data, pixels, imageSize); + if (image->mt) { + uint32_t srcRowStride, bytesPerRow, rows; + srcRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, width); + bytesPerRow = srcRowStride; + rows = (height + 3) / 4; + copy_rows(texImage->Data, image->mt->levels[level].rowstride, + pixels, srcRowStride, rows, bytesPerRow); + } else { + memcpy(texImage->Data, pixels, imageSize); + } } else { GLuint dstRowStride; GLuint *dstImageOffsets; @@ -756,14 +764,23 @@ static void radeon_texsubimage(GLcontext* ctx, int dims, GLenum target, int leve } if (compressed) { - uint32_t srcRowStride, bytesPerRow, rows; - dstRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, texImage->Width); + uint32_t srcRowStride, bytesPerRow, rows; + GLubyte *img_start; + if (!image->mt) { + dstRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, texImage->Width); + img_start = _mesa_compressed_image_address(xoffset, yoffset, 0, + texImage->TexFormat->MesaFormat, + texImage->Width, texImage->Data); + } + else { + uint32_t blocks_x = dstRowStride / (image->mt->bpp * 4); + img_start = texImage->Data + image->mt->bpp * 4 * (blocks_x * (yoffset / 4) + xoffset / 4); + } srcRowStride = _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, width); bytesPerRow = srcRowStride; - rows = height / 4; + rows = (height + 3) / 4; - copy_rows(texImage->Data, dstRowStride, image->base.Data, srcRowStride, rows, - bytesPerRow); + copy_rows(img_start, dstRowStride, pixels, srcRowStride, rows, bytesPerRow); } else { if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat, @@ -884,8 +901,8 @@ static void migrate_image_to_miptree(radeon_mipmap_tree *mt, radeon_texture_imag uint32_t height; /* need to confirm this value is correct */ if (mt->compressed) { - height = image->base.Height / 4; - srcrowstride = image->base.RowStride * mt->bpp; + height = (image->base.Height + 3) / 4; + srcrowstride = _mesa_compressed_row_stride(image->base.TexFormat->MesaFormat, image->base.Width); } else { height = image->base.Height * image->base.Depth; srcrowstride = image->base.Width * image->base.TexFormat->TexelBytes; @@ -951,7 +968,7 @@ int radeon_validate_texture_miptree(GLcontext * ctx, struct gl_texture_object *t fprintf(stderr, " Allocate new miptree\n"); radeon_try_alloc_miptree(rmesa, t, &baseimage->base, 0, texObj->BaseLevel); if (!t->mt) { - _mesa_problem(ctx, "r300_validate_texture failed to alloc miptree"); + _mesa_problem(ctx, "radeon_validate_texture failed to alloc miptree"); return GL_FALSE; } } @@ -1000,6 +1017,8 @@ radeon_get_tex_image(GLcontext * ctx, GLenum target, GLint level, } if (compressed) { + /* FIXME: this can't work for small textures (mips) which + use different hw stride */ _mesa_get_compressed_teximage(ctx, target, level, pixels, texObj, texImage); } else { diff --git a/src/mesa/drivers/dri/swrast/swrast.c b/src/mesa/drivers/dri/swrast/swrast.c index fbfa49c99d..3aa7843b1b 100644 --- a/src/mesa/drivers/dri/swrast/swrast.c +++ b/src/mesa/drivers/dri/swrast/swrast.c @@ -63,6 +63,7 @@ /* sw extensions not associated with some GL version */ #define need_GL_ARB_shader_objects +#define need_GL_ARB_vertex_array_object #define need_GL_ARB_vertex_program #define need_GL_APPLE_vertex_array_object #define need_GL_ATI_fragment_shader @@ -94,6 +95,7 @@ const struct dri_extension card_extensions[] = { "GL_SGI_color_table", GL_SGI_color_table_functions }, { "GL_ARB_shader_objects", GL_ARB_shader_objects_functions }, + { "GL_ARB_vertex_array_object", GL_ARB_vertex_array_object_functions }, { "GL_ARB_vertex_program", GL_ARB_vertex_program_functions }, { "GL_APPLE_vertex_array_object", GL_APPLE_vertex_array_object_functions }, { "GL_ATI_fragment_shader", GL_ATI_fragment_shader_functions }, diff --git a/src/mesa/drivers/windows/gdi/mesa.def b/src/mesa/drivers/windows/gdi/mesa.def index ede43ef4c0..bd3e5b2137 100644 --- a/src/mesa/drivers/windows/gdi/mesa.def +++ b/src/mesa/drivers/windows/gdi/mesa.def @@ -902,7 +902,6 @@ EXPORTS _mesa_generate_mipmap _mesa_get_compressed_teximage _mesa_get_current_context - _mesa_get_program_register _mesa_get_teximage _mesa_init_driver_functions _mesa_init_glsl_driver_functions |