From 537bbe6dec780f6f85838fe7e6036579c509f8a6 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Tue, 6 May 2008 10:51:08 -0700 Subject: [intel-GEM] Add tiling support to swrast. Accessing tiled surfaces without using the fence registers requires that software deal with the address swizzling itself. --- src/mesa/drivers/dri/intel/intel_context.c | 3 + src/mesa/drivers/dri/intel/intel_fbo.c | 16 +- src/mesa/drivers/dri/intel/intel_fbo.h | 3 +- src/mesa/drivers/dri/intel/intel_screen.c | 18 +- src/mesa/drivers/dri/intel/intel_screen.h | 2 + src/mesa/drivers/dri/intel/intel_span.c | 327 +++++++++++++++++++++++++++-- src/mesa/drivers/dri/intel/intel_span.h | 6 +- 7 files changed, 350 insertions(+), 25 deletions(-) (limited to 'src') diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c index 4a1e1a9ac0..d258e669c0 100644 --- a/src/mesa/drivers/dri/intel/intel_context.c +++ b/src/mesa/drivers/dri/intel/intel_context.c @@ -498,6 +498,9 @@ intel_init_bufmgr(struct intel_context *intel) intel); } + /* XXX bufmgr should be per-screen, not per-context */ + intelScreen->ttm = intel->ttm; + return GL_TRUE; } diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c index b3f6610546..bc0b579429 100644 --- a/src/mesa/drivers/dri/intel/intel_fbo.c +++ b/src/mesa/drivers/dri/intel/intel_fbo.c @@ -295,7 +295,8 @@ intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb, rb->Height = height; /* This sets the Get/PutRow/Value functions */ - intel_set_span_functions(&irb->Base); + /* XXX can we choose a different tile here? */ + intel_set_span_functions(&irb->Base, INTEL_TILE_NONE); return GL_TRUE; } @@ -375,7 +376,7 @@ intel_renderbuffer_set_region(struct intel_renderbuffer *rb, * not a user-created renderbuffer. */ struct intel_renderbuffer * -intel_create_renderbuffer(GLenum intFormat) +intel_create_renderbuffer(GLenum intFormat, int tiling) { GET_CURRENT_CONTEXT(ctx); @@ -442,12 +443,14 @@ intel_create_renderbuffer(GLenum intFormat) irb->Base.InternalFormat = intFormat; + irb->tiling = tiling; + /* intel-specific methods */ irb->Base.Delete = intel_delete_renderbuffer; irb->Base.AllocStorage = intel_alloc_window_storage; irb->Base.GetPointer = intel_get_pointer; /* This sets the Get/PutRow/Value functions */ - intel_set_span_functions(&irb->Base); + intel_set_span_functions(&irb->Base, tiling); return irb; } @@ -519,7 +522,7 @@ intel_framebuffer_renderbuffer(GLcontext * ctx, static GLboolean intel_update_wrapper(GLcontext *ctx, struct intel_renderbuffer *irb, - struct gl_texture_image *texImage) + struct gl_texture_image *texImage) { if (texImage->TexFormat == &_mesa_texformat_argb8888) { irb->Base._ActualFormat = GL_RGBA8; @@ -558,7 +561,7 @@ intel_update_wrapper(GLcontext *ctx, struct intel_renderbuffer *irb, irb->Base.Delete = intel_delete_renderbuffer; irb->Base.AllocStorage = intel_nop_alloc_storage; - intel_set_span_functions(&irb->Base); + intel_set_span_functions(&irb->Base, irb->tiling); irb->RenderToTexture = GL_TRUE; @@ -586,6 +589,9 @@ intel_wrap_texture(GLcontext * ctx, struct gl_texture_image *texImage) _mesa_init_renderbuffer(&irb->Base, name); irb->Base.ClassID = INTEL_RB_CLASS; + /* XXX can we fix this? */ + irb->tiling = INTEL_TILE_NONE; + if (!intel_update_wrapper(ctx, irb, texImage)) { _mesa_free(irb); return NULL; diff --git a/src/mesa/drivers/dri/intel/intel_fbo.h b/src/mesa/drivers/dri/intel/intel_fbo.h index c90c84b48c..9e085a1992 100644 --- a/src/mesa/drivers/dri/intel/intel_fbo.h +++ b/src/mesa/drivers/dri/intel/intel_fbo.h @@ -72,6 +72,7 @@ struct intel_renderbuffer struct intel_region *region; void *pfMap; /* possibly paged flipped map pointer */ GLuint pfPitch; /* possibly paged flipped pitch */ + int tiling; GLboolean RenderToTexture; /* RTT? */ GLuint PairedDepth; /**< only used if this is a depth renderbuffer */ @@ -90,7 +91,7 @@ intel_renderbuffer_set_region(struct intel_renderbuffer *irb, struct intel_region *region); extern struct intel_renderbuffer * -intel_create_renderbuffer(GLenum intFormat); +intel_create_renderbuffer(GLenum intFormat, int tiling); extern void intel_fbo_init(struct intel_context *intel); diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c index 356e50e726..a243324a39 100644 --- a/src/mesa/drivers/dri/intel/intel_screen.c +++ b/src/mesa/drivers/dri/intel/intel_screen.c @@ -524,20 +524,23 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv, /* setup the hardware-based renderbuffers */ { - intel_fb->color_rb[0] = intel_create_renderbuffer(rgbFormat); + intel_fb->color_rb[0] = intel_create_renderbuffer(rgbFormat, + screen->ttm ? screen->front.tiled : INTEL_TILE_NONE); _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT, &intel_fb->color_rb[0]->Base); } if (mesaVis->doubleBufferMode) { - intel_fb->color_rb[1] = intel_create_renderbuffer(rgbFormat); + intel_fb->color_rb[1] = intel_create_renderbuffer(rgbFormat, + screen->ttm ? screen->back.tiled : INTEL_TILE_NONE); _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT, &intel_fb->color_rb[1]->Base); if (screen->third.handle) { struct gl_renderbuffer *tmp_rb = NULL; - intel_fb->color_rb[2] = intel_create_renderbuffer(rgbFormat); + intel_fb->color_rb[2] = intel_create_renderbuffer(rgbFormat, + screen->ttm ? screen->third.tiled : INTEL_TILE_NONE); _mesa_reference_renderbuffer(&tmp_rb, &intel_fb->color_rb[2]->Base); } } @@ -546,7 +549,8 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv, if (mesaVis->stencilBits == 8) { /* combined depth/stencil buffer */ struct intel_renderbuffer *depthStencilRb - = intel_create_renderbuffer(GL_DEPTH24_STENCIL8_EXT); + = intel_create_renderbuffer(GL_DEPTH24_STENCIL8_EXT, + screen->ttm ? screen->depth.tiled : INTEL_TILE_NONE); /* note: bind RB to two attachment points */ _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH, &depthStencilRb->Base); @@ -554,7 +558,8 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv, &depthStencilRb->Base); } else { struct intel_renderbuffer *depthRb - = intel_create_renderbuffer(GL_DEPTH_COMPONENT24); + = intel_create_renderbuffer(GL_DEPTH_COMPONENT24, + screen->ttm ? screen->depth.tiled : INTEL_TILE_NONE); _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH, &depthRb->Base); } @@ -562,7 +567,8 @@ intelCreateBuffer(__DRIscreenPrivate * driScrnPriv, else if (mesaVis->depthBits == 16) { /* just 16-bit depth buffer, no hw stencil */ struct intel_renderbuffer *depthRb - = intel_create_renderbuffer(GL_DEPTH_COMPONENT16); + = intel_create_renderbuffer(GL_DEPTH_COMPONENT16, + screen->ttm ? screen->depth.tiled : INTEL_TILE_NONE); _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_DEPTH, &depthRb->Base); } diff --git a/src/mesa/drivers/dri/intel/intel_screen.h b/src/mesa/drivers/dri/intel/intel_screen.h index e62b2d7c89..9a73b13951 100644 --- a/src/mesa/drivers/dri/intel/intel_screen.h +++ b/src/mesa/drivers/dri/intel/intel_screen.h @@ -74,6 +74,8 @@ typedef struct int irq_active; int allow_batchbuffer; + int ttm; + /** * Configuration cache with default values for all contexts */ diff --git a/src/mesa/drivers/dri/intel/intel_span.c b/src/mesa/drivers/dri/intel/intel_span.c index df4f5927a0..149b581d88 100644 --- a/src/mesa/drivers/dri/intel/intel_span.c +++ b/src/mesa/drivers/dri/intel/intel_span.c @@ -39,6 +39,111 @@ #include "swrast/swrast.h" +/* + * Deal with tiled surfaces + */ + +#if 0 +/* These are pre-965 tile swizzling functions -- power of two widths */ +static uintptr_t x_tile_swizzle_pow2 (uintptr_t addr, int n) +{ + uintptr_t a = addr; + uintptr_t base_mask = (((~0) << (n + 4)) | 0xff); + uintptr_t x_mask = ((~0) << 12) & ~base_mask; + + a = ((a & base_mask) | + ((a >> (n-8)) & 0x7) | + ((a << 3) & x_mask)); + _mesa_printf ("x_swizzle %08x (base %x yrow %x tile#x %x xsword %x byte %x) %08x\n", + addr, + addr >> (n + 4), + (addr >> (n + 1)) & 0x7, + (addr >> 9) & ((1 << (n-8)) - 1), + (addr >> 5) & 0xf, + (addr & 0x1f), + a); + return a; +} + +static uintptr_t y_tile_swizzle_pow2 (uintptr_t addr, int n) +{ + uintptr_t a = (uintptr_t) addr; + uintptr_t base_mask = (((~0) << (n + 6)) | 0xf); + uintptr_t x_mask = ((~0) << 9) & ~base_mask; + + a = ((a & base_mask) | + ((a >> (n-3)) & 0x1f) | + ((a << 5) & x_mask)); + _mesa_printf ("y_swizzle %08x (base %x yrow %x tile#x %x xoword %x byte %x) %08x\n", + addr, + addr >> (n + 6), + (addr >> (n + 1)) & 0x01f, + (addr >> 7) & ((1 << (n-6)) - 1), + (addr >> 4) & 0x7, + (addr & 0xf), + a); + return a; +} +#endif + +static GLubyte *x_tile_swizzle(struct intel_renderbuffer *irb, struct intel_context *intel, + int x, int y) +{ + GLubyte *buf = (GLubyte *) irb->pfMap; + int tile_stride; + int xbyte; + int x_tile_off, y_tile_off; + int x_tile_number, y_tile_number; + int tile_off, tile_base; + + tile_stride = (irb->pfPitch * irb->region->cpp) << 3; + + x += intel->drawX; + y += intel->drawY; + + xbyte = x * irb->region->cpp; + + x_tile_off = xbyte & 0x1ff; + y_tile_off = y & 7; + + x_tile_number = xbyte >> 9; + y_tile_number = y >> 3; + + tile_off = (y_tile_off << 9) + x_tile_off; + tile_base = (x_tile_number << 12) + y_tile_number * tile_stride; + + return buf + tile_base + tile_off; +} + +static GLubyte *y_tile_swizzle(struct intel_renderbuffer *irb, struct intel_context *intel, + int x, int y) +{ + GLubyte *buf = (GLubyte *) irb->pfMap; + int tile_stride; + int xbyte; + int x_tile_off, y_tile_off; + int x_tile_number, y_tile_number; + int tile_off, tile_base; + + tile_stride = (irb->pfPitch * irb->region->cpp) << 3; + + x += intel->drawX; + y += intel->drawY; + + xbyte = x * irb->region->cpp; + + x_tile_off = xbyte & 0x7f; + y_tile_off = y & 0x1f; + + x_tile_number = xbyte >> 7; + y_tile_number = y >> 5; + + tile_off = ((x_tile_off & ~0xf) << 5) + (y_tile_off << 4) + (x_tile_off & 0xf); + tile_base = (x_tile_number << 12) + y_tile_number * tile_stride; + + return buf + tile_base + tile_off; +} + /* break intelWriteRGBASpan_ARGB8888 */ @@ -55,7 +160,7 @@ + (intel->drawY * irb->pfPitch + intel->drawX) * irb->region->cpp;\ GLuint p; \ assert(irb->pfMap);\ - (void) p; + (void) p; (void) buf; /* XXX FBO: this is identical to the macro in spantmp2.h except we get * the cliprect info from the context, not the driDrawable. @@ -69,12 +174,14 @@ int miny = intel->pClipRects[_nc].y1 - intel->drawY; \ int maxx = intel->pClipRects[_nc].x2 - intel->drawX; \ int maxy = intel->pClipRects[_nc].y2 - intel->drawY; - - - + +#if 0 + }} +#endif #define Y_FLIP(_y) ((_y) * yScale + yBias) +/* XXX with GEM, these need to tell the kernel */ #define HW_LOCK() #define HW_UNLOCK() @@ -99,6 +206,43 @@ #define GET_PTR(X,Y) (buf + ((Y) * irb->pfPitch + (X)) * 4) #include "spantmp2.h" +/* 16 bit RGB565 color tile spanline and pixel functions + */ + +#define SPANTMP_PIXEL_FMT GL_RGB +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5 + +#define TAG(x) intel_XTile_##x##_RGB565 +#define TAG2(x,y) intel_XTile_##x##_RGB565##y +#define GET_PTR(X,Y) x_tile_swizzle(irb, intel, X, Y) +#include "spantmp2.h" + +#define SPANTMP_PIXEL_FMT GL_RGB +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5 + +#define TAG(x) intel_YTile_##x##_RGB565 +#define TAG2(x,y) intel_YTile_##x##_RGB565##y +#define GET_PTR(X,Y) y_tile_swizzle(irb, intel, X, Y) +#include "spantmp2.h" + +/* 32 bit ARGB888 color tile spanline and pixel functions + */ + +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) intel_XTile_##x##_ARGB8888 +#define TAG2(x,y) intel_XTile_##x##_ARGB8888##y +#define GET_PTR(X,Y) x_tile_swizzle(irb, intel, X, Y) +#include "spantmp2.h" + +#define SPANTMP_PIXEL_FMT GL_BGRA +#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV + +#define TAG(x) intel_YTile_##x##_ARGB8888 +#define TAG2(x,y) intel_YTile_##x##_ARGB8888##y +#define GET_PTR(X,Y) y_tile_swizzle(irb, intel, X, Y) +#include "spantmp2.h" #define LOCAL_DEPTH_VARS \ struct intel_context *intel = intel_context(ctx); \ @@ -107,7 +251,7 @@ const GLint yScale = irb->RenderToTexture ? 1 : -1; \ const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1; \ char *buf = (char *) irb->pfMap/*XXX use region->map*/ + \ - (intel->drawY * pitch + intel->drawX) * irb->region->cpp; + (intel->drawY * pitch + intel->drawX) * irb->region->cpp; (void) buf; #define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS @@ -126,6 +270,33 @@ #include "depthtmp.h" +/** + ** 16-bit x tile depthbuffer functions. + **/ +#define WRITE_DEPTH( _x, _y, d ) \ + (*((GLushort *)x_tile_swizzle (irb, intel, _x, _y)) = d) + +#define READ_DEPTH( d, _x, _y ) \ + d = *((GLushort *)x_tile_swizzle (irb, intel, _x, _y)) + + +#define TAG(x) intel_XTile_##x##_z16 +#include "depthtmp.h" + +/** + ** 16-bit y tile depthbuffer functions. + **/ +#define WRITE_DEPTH( _x, _y, d ) \ + (*((GLushort *)y_tile_swizzle (irb, intel, _x, _y)) = d) + +#define READ_DEPTH( d, _x, _y ) \ + (d = *((GLushort *)y_tile_swizzle (irb, intel, _x, _y))) + + +#define TAG(x) intel_YTile_##x##_z16 +#include "depthtmp.h" + + /** ** 24/8-bit interleaved depth/stencil functions ** Note: we're actually reading back combined depth+stencil values. @@ -148,6 +319,49 @@ #include "depthtmp.h" +/** + ** 24/8-bit x-tile interleaved depth/stencil functions + ** Note: we're actually reading back combined depth+stencil values. + ** The wrappers in main/depthstencil.c are used to extract the depth + ** and stencil values. + **/ +/* Change ZZZS -> SZZZ */ +#define WRITE_DEPTH( _x, _y, d ) { \ + GLuint tmp = ((d) >> 8) | ((d) << 24); \ + *((GLuint *)x_tile_swizzle (irb, intel, _x, _y)) = tmp; \ +} + +/* Change SZZZ -> ZZZS */ +#define READ_DEPTH( d, _x, _y ) { \ + GLuint tmp = *((GLuint *)x_tile_swizzle (irb, intel, _x, _y)); \ + d = (tmp << 8) | (tmp >> 24); \ +} + +#define TAG(x) intel_XTile_##x##_z24_s8 +#include "depthtmp.h" + +/** + ** 24/8-bit y-tile interleaved depth/stencil functions + ** Note: we're actually reading back combined depth+stencil values. + ** The wrappers in main/depthstencil.c are used to extract the depth + ** and stencil values. + **/ +/* Change ZZZS -> SZZZ */ +#define WRITE_DEPTH( _x, _y, d ) { \ + GLuint tmp = ((d) >> 8) | ((d) << 24); \ + *((GLuint *)y_tile_swizzle (irb, intel, _x, _y)) = tmp; \ +} + +/* Change SZZZ -> ZZZS */ +#define READ_DEPTH( d, _x, _y ) { \ + GLuint tmp = *((GLuint *)y_tile_swizzle (irb, intel, _x, _y)); \ + d = (tmp << 8) | (tmp >> 24); \ +} + +#define TAG(x) intel_YTile_##x##_z24_s8 +#include "depthtmp.h" + + /** ** 8-bit stencil function (XXX FBO: This is obsolete) **/ @@ -164,6 +378,40 @@ #define TAG(x) intel##x##_z24_s8 #include "stenciltmp.h" +/** + ** 8-bit x-tile stencil function (XXX FBO: This is obsolete) + **/ +#define WRITE_STENCIL( _x, _y, d ) { \ + GLuint *a = (GLuint *) x_tile_swizzle (irb, intel, _x, _y); \ + GLuint tmp = *a; \ + tmp &= 0xffffff; \ + tmp |= ((d) << 24); \ + *a = tmp; \ +} + +#define READ_STENCIL( d, _x, _y ) \ + (d = *((GLuint*) x_tile_swizzle (irb, intel, _x, _y)) >> 24) + +#define TAG(x) intel_XTile_##x##_z24_s8 +#include "stenciltmp.h" + +/** + ** 8-bit y-tile stencil function (XXX FBO: This is obsolete) + **/ +#define WRITE_STENCIL( _x, _y, d ) { \ + GLuint *a = (GLuint *) y_tile_swizzle (irb, intel, _x, _y); \ + GLuint tmp = *a; \ + tmp &= 0xffffff; \ + tmp |= ((d) << 24); \ + *a = tmp; \ +} + +#define READ_STENCIL( d, _x, _y ) \ + (d = *((GLuint*) y_tile_swizzle (irb, intel, _x, _y)) >> 24) + +#define TAG(x) intel_YTile_##x##_z24_s8 +#include "stenciltmp.h" + /** @@ -379,25 +627,80 @@ intelInitSpanFuncs(GLcontext * ctx) * These are used for the software fallbacks. */ void -intel_set_span_functions(struct gl_renderbuffer *rb) +intel_set_span_functions(struct gl_renderbuffer *rb, int tiling) { if (rb->_ActualFormat == GL_RGB5) { /* 565 RGB */ - intelInitPointers_RGB565(rb); + switch (tiling) { + case INTEL_TILE_NONE: + default: + intelInitPointers_RGB565(rb); + break; + case INTEL_TILE_X: + intel_XTile_InitPointers_RGB565(rb); + break; + case INTEL_TILE_Y: + intel_YTile_InitPointers_RGB565(rb); + break; + } } else if (rb->_ActualFormat == GL_RGBA8) { /* 8888 RGBA */ - intelInitPointers_ARGB8888(rb); + switch (tiling) { + case INTEL_TILE_NONE: + default: + intelInitPointers_ARGB8888(rb); + break; + case INTEL_TILE_X: + intel_XTile_InitPointers_ARGB8888(rb); + break; + case INTEL_TILE_Y: + intel_YTile_InitPointers_ARGB8888(rb); + break; + } } else if (rb->_ActualFormat == GL_DEPTH_COMPONENT16) { - intelInitDepthPointers_z16(rb); + switch (tiling) { + case INTEL_TILE_NONE: + default: + intelInitDepthPointers_z16(rb); + break; + case INTEL_TILE_X: + intel_XTile_InitDepthPointers_z16(rb); + break; + case INTEL_TILE_Y: + intel_YTile_InitDepthPointers_z16(rb); + break; + } } else if (rb->_ActualFormat == GL_DEPTH_COMPONENT24 || /* XXX FBO remove */ rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT) { - intelInitDepthPointers_z24_s8(rb); + switch (tiling) { + case INTEL_TILE_NONE: + default: + intelInitDepthPointers_z24_s8(rb); + break; + case INTEL_TILE_X: + intel_XTile_InitDepthPointers_z24_s8(rb); + break; + case INTEL_TILE_Y: + intel_YTile_InitDepthPointers_z24_s8(rb); + break; + } } - else if (rb->_ActualFormat == GL_STENCIL_INDEX8_EXT) { /* XXX FBO remove */ - intelInitStencilPointers_z24_s8(rb); + else if (rb->_ActualFormat == GL_STENCIL_INDEX8_EXT) { + switch (tiling) { + case INTEL_TILE_NONE: + default: + intelInitStencilPointers_z24_s8(rb); + break; + case INTEL_TILE_X: + intel_XTile_InitStencilPointers_z24_s8(rb); + break; + case INTEL_TILE_Y: + intel_YTile_InitStencilPointers_z24_s8(rb); + break; + } } else { _mesa_problem(NULL, diff --git a/src/mesa/drivers/dri/intel/intel_span.h b/src/mesa/drivers/dri/intel/intel_span.h index 5201f6d6c6..c56e5e1611 100644 --- a/src/mesa/drivers/dri/intel/intel_span.h +++ b/src/mesa/drivers/dri/intel/intel_span.h @@ -33,6 +33,10 @@ extern void intelInitSpanFuncs(GLcontext * ctx); extern void intelSpanRenderFinish(GLcontext * ctx); extern void intelSpanRenderStart(GLcontext * ctx); -extern void intel_set_span_functions(struct gl_renderbuffer *rb); +extern void intel_set_span_functions(struct gl_renderbuffer *rb, int tiling); + +#define INTEL_TILE_NONE 0 +#define INTEL_TILE_X 1 +#define INTEL_TILE_Y 2 #endif -- cgit v1.2.3