diff options
Diffstat (limited to 'src/mesa/drivers/dri/r200/r200_texstate.c')
| -rw-r--r-- | src/mesa/drivers/dri/r200/r200_texstate.c | 859 |
1 files changed, 317 insertions, 542 deletions
diff --git a/src/mesa/drivers/dri/r200/r200_texstate.c b/src/mesa/drivers/dri/r200/r200_texstate.c index 3f9a2f4ac1..c94834752e 100644 --- a/src/mesa/drivers/dri/r200/r200_texstate.c +++ b/src/mesa/drivers/dri/r200/r200_texstate.c @@ -37,9 +37,12 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/context.h" #include "main/macros.h" #include "main/texformat.h" +#include "main/teximage.h" #include "main/texobj.h" #include "main/enums.h" +#include "radeon_common.h" +#include "radeon_mipmap_tree.h" #include "r200_context.h" #include "r200_state.h" #include "r200_ioctl.h" @@ -139,257 +142,6 @@ static const struct tx_table tx_table_le[] = #undef _ALPHA #undef _INVALID -/** - * This function computes the number of bytes of storage needed for - * the given texture object (all mipmap levels, all cube faces). - * The \c image[face][level].x/y/width/height parameters for upload/blitting - * are computed here. \c pp_txfilter, \c pp_txformat, etc. will be set here - * too. - * - * \param rmesa Context pointer - * \param tObj GL texture object whose images are to be posted to - * hardware state. - */ -static void r200SetTexImages( r200ContextPtr rmesa, - struct gl_texture_object *tObj ) -{ - r200TexObjPtr t = (r200TexObjPtr)tObj->DriverData; - const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel]; - GLint curOffset, blitWidth; - GLint i, texelBytes; - GLint numLevels; - GLint log2Width, log2Height, log2Depth; - - /* Set the hardware texture format - */ - if ( !t->image_override ) { - if ( VALID_FORMAT( baseImage->TexFormat->MesaFormat ) ) { - const struct tx_table *table = _mesa_little_endian() ? tx_table_le : - tx_table_be; - - t->pp_txformat &= ~(R200_TXFORMAT_FORMAT_MASK | - R200_TXFORMAT_ALPHA_IN_MAP); - t->pp_txfilter &= ~R200_YUV_TO_RGB; - - t->pp_txformat |= table[ baseImage->TexFormat->MesaFormat ].format; - t->pp_txfilter |= table[ baseImage->TexFormat->MesaFormat ].filter; - } - else { - _mesa_problem(NULL, "unexpected texture format in %s", __FUNCTION__); - return; - } - } - - texelBytes = baseImage->TexFormat->TexelBytes; - - /* Compute which mipmap levels we really want to send to the hardware. - */ - - driCalculateTextureFirstLastLevel( (driTextureObject *) t ); - log2Width = tObj->Image[0][t->base.firstLevel]->WidthLog2; - log2Height = tObj->Image[0][t->base.firstLevel]->HeightLog2; - log2Depth = tObj->Image[0][t->base.firstLevel]->DepthLog2; - - numLevels = t->base.lastLevel - t->base.firstLevel + 1; - - assert(numLevels <= RADEON_MAX_TEXTURE_LEVELS); - - /* Calculate mipmap offsets and dimensions for blitting (uploading) - * The idea is that we lay out the mipmap levels within a block of - * memory organized as a rectangle of width BLIT_WIDTH_BYTES. - */ - curOffset = 0; - blitWidth = BLIT_WIDTH_BYTES; - t->tile_bits = 0; - - /* figure out if this texture is suitable for tiling. */ - if (texelBytes) { - if (rmesa->texmicrotile && (tObj->Target != GL_TEXTURE_RECTANGLE_NV) && - /* texrect might be able to use micro tiling too in theory? */ - (baseImage->Height > 1)) { - /* allow 32 (bytes) x 1 mip (which will use two times the space - the non-tiled version would use) max if base texture is large enough */ - if ((numLevels == 1) || - (((baseImage->Width * texelBytes / baseImage->Height) <= 32) && - (baseImage->Width * texelBytes > 64)) || - ((baseImage->Width * texelBytes / baseImage->Height) <= 16)) { - t->tile_bits |= R200_TXO_MICRO_TILE; - } - } - if (tObj->Target != GL_TEXTURE_RECTANGLE_NV) { - /* we can set macro tiling even for small textures, they will be untiled anyway */ - t->tile_bits |= R200_TXO_MACRO_TILE; - } - } - - for (i = 0; i < numLevels; i++) { - const struct gl_texture_image *texImage; - GLuint size; - - texImage = tObj->Image[0][i + t->base.firstLevel]; - if ( !texImage ) - break; - - /* find image size in bytes */ - if (texImage->IsCompressed) { - /* need to calculate the size AFTER padding even though the texture is - submitted without padding. - Only handle pot textures currently - don't know if npot is even possible, - size calculation would certainly need (trivial) adjustments. - Align (and later pad) to 32byte, not sure what that 64byte blit width is - good for? */ - if ((t->pp_txformat & R200_TXFORMAT_FORMAT_MASK) == R200_TXFORMAT_DXT1) { - /* RGB_DXT1/RGBA_DXT1, 8 bytes per block */ - if ((texImage->Width + 3) < 8) /* width one block */ - size = texImage->CompressedSize * 4; - else if ((texImage->Width + 3) < 16) - size = texImage->CompressedSize * 2; - else size = texImage->CompressedSize; - } - else /* DXT3/5, 16 bytes per block */ - if ((texImage->Width + 3) < 8) - size = texImage->CompressedSize * 2; - else size = texImage->CompressedSize; - } - else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) { - size = ((texImage->Width * texelBytes + 63) & ~63) * texImage->Height; - } - else if (t->tile_bits & R200_TXO_MICRO_TILE) { - /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned, - though the actual offset may be different (if texture is less than - 32 bytes width) to the untiled case */ - int w = (texImage->Width * texelBytes * 2 + 31) & ~31; - size = (w * ((texImage->Height + 1) / 2)) * texImage->Depth; - blitWidth = MAX2(texImage->Width, 64 / texelBytes); - } - else { - int w = (texImage->Width * texelBytes + 31) & ~31; - size = w * texImage->Height * texImage->Depth; - blitWidth = MAX2(texImage->Width, 64 / texelBytes); - } - assert(size > 0); - - /* Align to 32-byte offset. It is faster to do this unconditionally - * (no branch penalty). - */ - - curOffset = (curOffset + 0x1f) & ~0x1f; - - if (texelBytes) { - t->image[0][i].x = curOffset; /* fix x and y coords up later together with offset */ - t->image[0][i].y = 0; - t->image[0][i].width = MIN2(size / texelBytes, blitWidth); - t->image[0][i].height = (size / texelBytes) / t->image[0][i].width; - } - else { - t->image[0][i].x = curOffset % BLIT_WIDTH_BYTES; - t->image[0][i].y = curOffset / BLIT_WIDTH_BYTES; - t->image[0][i].width = MIN2(size, BLIT_WIDTH_BYTES); - t->image[0][i].height = size / t->image[0][i].width; - } - -#if 0 - /* for debugging only and only applicable to non-rectangle targets */ - assert(size % t->image[0][i].width == 0); - assert(t->image[0][i].x == 0 - || (size < BLIT_WIDTH_BYTES && t->image[0][i].height == 1)); -#endif - - if (0) - fprintf(stderr, - "level %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n", - i, texImage->Width, texImage->Height, - t->image[0][i].x, t->image[0][i].y, - t->image[0][i].width, t->image[0][i].height, size, curOffset); - - curOffset += size; - - } - - /* Align the total size of texture memory block. - */ - t->base.totalSize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK; - - /* Setup remaining cube face blits, if needed */ - if (tObj->Target == GL_TEXTURE_CUBE_MAP) { - const GLuint faceSize = t->base.totalSize; - GLuint face; - /* reuse face 0 x/y/width/height - just update the offset when uploading */ - for (face = 1; face < 6; face++) { - for (i = 0; i < numLevels; i++) { - t->image[face][i].x = t->image[0][i].x; - t->image[face][i].y = t->image[0][i].y; - t->image[face][i].width = t->image[0][i].width; - t->image[face][i].height = t->image[0][i].height; - } - } - t->base.totalSize = 6 * faceSize; /* total texmem needed */ - } - - - /* Hardware state: - */ - t->pp_txfilter &= ~R200_MAX_MIP_LEVEL_MASK; - t->pp_txfilter |= (numLevels - 1) << R200_MAX_MIP_LEVEL_SHIFT; - - t->pp_txformat &= ~(R200_TXFORMAT_WIDTH_MASK | - R200_TXFORMAT_HEIGHT_MASK | - R200_TXFORMAT_CUBIC_MAP_ENABLE | - R200_TXFORMAT_F5_WIDTH_MASK | - R200_TXFORMAT_F5_HEIGHT_MASK); - t->pp_txformat |= ((log2Width << R200_TXFORMAT_WIDTH_SHIFT) | - (log2Height << R200_TXFORMAT_HEIGHT_SHIFT)); - - t->pp_txformat_x &= ~(R200_DEPTH_LOG2_MASK | R200_TEXCOORD_MASK); - if (tObj->Target == GL_TEXTURE_3D) { - t->pp_txformat_x |= (log2Depth << R200_DEPTH_LOG2_SHIFT); - t->pp_txformat_x |= R200_TEXCOORD_VOLUME; - } - else if (tObj->Target == GL_TEXTURE_CUBE_MAP) { - ASSERT(log2Width == log2Height); - t->pp_txformat |= ((log2Width << R200_TXFORMAT_F5_WIDTH_SHIFT) | - (log2Height << R200_TXFORMAT_F5_HEIGHT_SHIFT) | -/* don't think we need this bit, if it exists at all - fglrx does not set it */ - (R200_TXFORMAT_CUBIC_MAP_ENABLE)); - t->pp_txformat_x |= R200_TEXCOORD_CUBIC_ENV; - t->pp_cubic_faces = ((log2Width << R200_FACE_WIDTH_1_SHIFT) | - (log2Height << R200_FACE_HEIGHT_1_SHIFT) | - (log2Width << R200_FACE_WIDTH_2_SHIFT) | - (log2Height << R200_FACE_HEIGHT_2_SHIFT) | - (log2Width << R200_FACE_WIDTH_3_SHIFT) | - (log2Height << R200_FACE_HEIGHT_3_SHIFT) | - (log2Width << R200_FACE_WIDTH_4_SHIFT) | - (log2Height << R200_FACE_HEIGHT_4_SHIFT)); - } - else { - /* If we don't in fact send enough texture coordinates, q will be 1, - * making TEXCOORD_PROJ act like TEXCOORD_NONPROJ (Right?) - */ - t->pp_txformat_x |= R200_TEXCOORD_PROJ; - } - - t->pp_txsize = (((tObj->Image[0][t->base.firstLevel]->Width - 1) << 0) | - ((tObj->Image[0][t->base.firstLevel]->Height - 1) << 16)); - - /* Only need to round to nearest 32 for textures, but the blitter - * requires 64-byte aligned pitches, and we may/may not need the - * blitter. NPOT only! - */ - if ( !t->image_override ) { - if (baseImage->IsCompressed) - t->pp_txpitch = (tObj->Image[0][t->base.firstLevel]->Width + 63) & ~(63); - else - t->pp_txpitch = ((tObj->Image[0][t->base.firstLevel]->Width * texelBytes) + 63) & ~(63); - t->pp_txpitch -= 32; - } - - t->dirty_state = TEX_ALL; - - /* FYI: r200UploadTexImages( rmesa, t ) used to be called here */ -} - - - /* ================================================================ * Texture combine functions */ @@ -569,7 +321,7 @@ static GLboolean r200UpdateTextureEnv( GLcontext *ctx, int unit, int slot, GLuin assert( (texUnit->_ReallyEnabled == 0) || (texUnit->_Current != NULL) ); - if ( R200_DEBUG & DEBUG_TEXTURE ) { + if ( R200_DEBUG & RADEON_TEXTURE ) { fprintf( stderr, "%s( %p, %d )\n", __FUNCTION__, (void *)ctx, unit ); } @@ -981,20 +733,19 @@ void r200SetTexOffset(__DRIcontext * pDRICtx, GLint texname, { r200ContextPtr rmesa = pDRICtx->driverPrivate; struct gl_texture_object *tObj = - _mesa_lookup_texture(rmesa->glCtx, texname); - r200TexObjPtr t; + _mesa_lookup_texture(rmesa->radeon.glCtx, texname); + radeonTexObjPtr t = radeon_tex_obj(tObj); if (!tObj) return; - t = (r200TexObjPtr) tObj->DriverData; - t->image_override = GL_TRUE; if (!offset) return; - t->pp_txoffset = offset; + t->bo = NULL; + t->override_offset = offset; t->pp_txpitch = pitch - 32; switch (depth) { @@ -1014,6 +765,125 @@ void r200SetTexOffset(__DRIcontext * pDRICtx, GLint texname, } } +void r200SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format, + __DRIdrawable *dPriv) +{ + struct gl_texture_unit *texUnit; + struct gl_texture_object *texObj; + struct gl_texture_image *texImage; + struct radeon_renderbuffer *rb; + radeon_texture_image *rImage; + radeonContextPtr radeon; + r200ContextPtr rmesa; + struct radeon_framebuffer *rfb; + radeonTexObjPtr t; + uint32_t pitch_val; + uint32_t internalFormat, type, format; + + type = GL_BGRA; + format = GL_UNSIGNED_BYTE; + internalFormat = (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT ? 3 : 4); + + radeon = pDRICtx->driverPrivate; + rmesa = pDRICtx->driverPrivate; + + rfb = dPriv->driverPrivate; + texUnit = &radeon->glCtx->Texture.Unit[radeon->glCtx->Texture.CurrentUnit]; + texObj = _mesa_select_tex_object(radeon->glCtx, texUnit, target); + texImage = _mesa_get_tex_image(radeon->glCtx, texObj, target, 0); + + rImage = get_radeon_texture_image(texImage); + t = radeon_tex_obj(texObj); + if (t == NULL) { + return; + } + + radeon_update_renderbuffers(pDRICtx, dPriv); + /* back & depth buffer are useless free them right away */ + rb = (void*)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer; + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; + } + rb = (void*)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer; + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; + } + rb = rfb->color_rb[0]; + if (rb->bo == NULL) { + /* Failed to BO for the buffer */ + return; + } + + _mesa_lock_texture(radeon->glCtx, texObj); + if (t->bo) { + radeon_bo_unref(t->bo); + t->bo = NULL; + } + if (rImage->bo) { + radeon_bo_unref(rImage->bo); + rImage->bo = NULL; + } + if (t->mt) { + radeon_miptree_unreference(t->mt); + t->mt = NULL; + } + if (rImage->mt) { + radeon_miptree_unreference(rImage->mt); + rImage->mt = NULL; + } + _mesa_init_teximage_fields(radeon->glCtx, target, texImage, + rb->base.Width, rb->base.Height, 1, 0, rb->cpp); + texImage->RowStride = rb->pitch / rb->cpp; + texImage->TexFormat = radeonChooseTextureFormat(radeon->glCtx, + internalFormat, + type, format, 0); + rImage->bo = rb->bo; + radeon_bo_ref(rImage->bo); + t->bo = rb->bo; + radeon_bo_ref(t->bo); + t->tile_bits = 0; + t->image_override = GL_TRUE; + t->override_offset = 0; + t->pp_txpitch &= (1 << 13) -1; + pitch_val = rb->pitch; + switch (rb->cpp) { + case 4: + if (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT) + t->pp_txformat = tx_table_le[MESA_FORMAT_RGB888].format; + else + t->pp_txformat = tx_table_le[MESA_FORMAT_ARGB8888].format; + t->pp_txfilter |= tx_table_le[MESA_FORMAT_ARGB8888].filter; + break; + case 3: + default: + t->pp_txformat = tx_table_le[MESA_FORMAT_RGB888].format; + t->pp_txfilter |= tx_table_le[MESA_FORMAT_RGB888].filter; + break; + case 2: + t->pp_txformat = tx_table_le[MESA_FORMAT_RGB565].format; + t->pp_txfilter |= tx_table_le[MESA_FORMAT_RGB565].filter; + break; + } + t->pp_txsize = ((rb->base.Width - 1) << RADEON_TEX_USIZE_SHIFT) + | ((rb->base.Height - 1) << RADEON_TEX_VSIZE_SHIFT); + t->pp_txformat |= R200_TXFORMAT_NON_POWER2; + t->pp_txpitch = pitch_val; + t->pp_txpitch -= 32; + + t->validated = GL_TRUE; + _mesa_unlock_texture(radeon->glCtx, texObj); + return; +} + + +void r200SetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv) +{ + r200SetTexBuffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv); +} + + #define REF_COLOR 1 #define REF_ALPHA 2 @@ -1207,12 +1077,43 @@ static GLboolean r200UpdateAllTexEnv( GLcontext *ctx ) R200_VOLUME_FILTER_MASK) +static void disable_tex_obj_state( r200ContextPtr rmesa, + int unit ) +{ + + R200_STATECHANGE( rmesa, vtx ); + rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3)); + + R200_STATECHANGE( rmesa, ctx ); + rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~(R200_TEX_0_ENABLE << unit); + if (rmesa->radeon.TclFallback & (R200_TCL_FALLBACK_TEXGEN_0<<unit)) { + TCL_FALLBACK( rmesa->radeon.glCtx, (R200_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE); + } + + /* Actually want to keep all units less than max active texture + * enabled, right? Fix this for >2 texunits. + */ + + { + GLuint tmp = rmesa->TexGenEnabled; + + rmesa->TexGenEnabled &= ~(R200_TEXGEN_TEXMAT_0_ENABLE<<unit); + rmesa->TexGenEnabled &= ~(R200_TEXMAT_0_ENABLE<<unit); + rmesa->TexGenNeedNormals[unit] = GL_FALSE; + rmesa->TexGenCompSel &= ~(R200_OUTPUT_TEX_0 << unit); + + if (tmp != rmesa->TexGenEnabled) { + rmesa->recheck_texgen[unit] = GL_TRUE; + rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX; + } + } +} static void import_tex_obj_state( r200ContextPtr rmesa, int unit, - r200TexObjPtr texobj ) + radeonTexObjPtr texobj ) { /* do not use RADEON_DB_STATE to avoid stale texture caches */ - int *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0]; + GLuint *cmd = &rmesa->hw.tex[unit].cmd[TEX_CMD_0]; R200_STATECHANGE( rmesa, tex[unit] ); @@ -1225,36 +1126,21 @@ static void import_tex_obj_state( r200ContextPtr rmesa, cmd[TEX_PP_TXSIZE] = texobj->pp_txsize; /* NPOT only! */ cmd[TEX_PP_TXPITCH] = texobj->pp_txpitch; /* NPOT only! */ cmd[TEX_PP_BORDER_COLOR] = texobj->pp_border_color; - if (rmesa->r200Screen->drmSupportsFragShader) { - cmd[TEX_PP_TXOFFSET_NEWDRM] = texobj->pp_txoffset; - } - else { - cmd[TEX_PP_TXOFFSET_OLDDRM] = texobj->pp_txoffset; - } - if (texobj->base.tObj->Target == GL_TEXTURE_CUBE_MAP) { - int *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0]; - GLuint bytesPerFace = texobj->base.totalSize / 6; - ASSERT(texobj->base.totalSize % 6 == 0); + if (texobj->base.Target == GL_TEXTURE_CUBE_MAP) { + GLuint *cube_cmd = &rmesa->hw.cube[unit].cmd[CUBE_CMD_0]; R200_STATECHANGE( rmesa, cube[unit] ); cube_cmd[CUBE_PP_CUBIC_FACES] = texobj->pp_cubic_faces; - if (rmesa->r200Screen->drmSupportsFragShader) { + if (rmesa->radeon.radeonScreen->drmSupportsFragShader) { /* that value is submitted twice. could change cube atom to not include that command when new drm is used */ cmd[TEX_PP_CUBIC_FACES] = texobj->pp_cubic_faces; } - cube_cmd[CUBE_PP_CUBIC_OFFSET_F1] = texobj->pp_txoffset + 1 * bytesPerFace; - cube_cmd[CUBE_PP_CUBIC_OFFSET_F2] = texobj->pp_txoffset + 2 * bytesPerFace; - cube_cmd[CUBE_PP_CUBIC_OFFSET_F3] = texobj->pp_txoffset + 3 * bytesPerFace; - cube_cmd[CUBE_PP_CUBIC_OFFSET_F4] = texobj->pp_txoffset + 4 * bytesPerFace; - cube_cmd[CUBE_PP_CUBIC_OFFSET_F5] = texobj->pp_txoffset + 5 * bytesPerFace; } - texobj->dirty_state &= ~(1<<unit); } - static void set_texgen_matrix( r200ContextPtr rmesa, GLuint unit, const GLfloat *s_plane, @@ -1366,37 +1252,36 @@ static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit ) fprintf(stderr, "%s unit %d\n", __FUNCTION__, unit); if (texUnit->TexGenEnabled & S_BIT) { - mode = texUnit->GenModeS; + mode = texUnit->GenS.Mode; } else { tgcm |= R200_TEXGEN_COMP_S << (unit * 4); } if (texUnit->TexGenEnabled & T_BIT) { - if (texUnit->GenModeT != mode) + if (texUnit->GenT.Mode != mode) mixed_fallback = GL_TRUE; } else { tgcm |= R200_TEXGEN_COMP_T << (unit * 4); } - if (texUnit->TexGenEnabled & R_BIT) { - if (texUnit->GenModeR != mode) + if (texUnit->GenR.Mode != mode) mixed_fallback = GL_TRUE; } else { tgcm |= R200_TEXGEN_COMP_R << (unit * 4); } if (texUnit->TexGenEnabled & Q_BIT) { - if (texUnit->GenModeQ != mode) + if (texUnit->GenQ.Mode != mode) mixed_fallback = GL_TRUE; } else { tgcm |= R200_TEXGEN_COMP_Q << (unit * 4); } if (mixed_fallback) { - if (R200_DEBUG & DEBUG_FALLBACKS) + if (R200_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "fallback mixed texgen, 0x%x (0x%x 0x%x 0x%x 0x%x)\n", - texUnit->TexGenEnabled, texUnit->GenModeS, texUnit->GenModeT, - texUnit->GenModeR, texUnit->GenModeQ); + texUnit->TexGenEnabled, texUnit->GenS.Mode, texUnit->GenT.Mode, + texUnit->GenR.Mode, texUnit->GenQ.Mode); return GL_FALSE; } @@ -1414,10 +1299,12 @@ static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit ) switch (mode) { case GL_OBJECT_LINEAR: { GLuint needtgenable = r200_need_dis_texgen( texUnit->TexGenEnabled, - texUnit->ObjectPlaneS, texUnit->ObjectPlaneT, - texUnit->ObjectPlaneR, texUnit->ObjectPlaneQ ); + texUnit->GenS.ObjectPlane, + texUnit->GenT.ObjectPlane, + texUnit->GenR.ObjectPlane, + texUnit->GenQ.ObjectPlane ); if (needtgenable & (S_BIT | T_BIT)) { - if (R200_DEBUG & DEBUG_FALLBACKS) + if (R200_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "fallback mixed texgen / obj plane, 0x%x\n", texUnit->TexGenEnabled); return GL_FALSE; @@ -1431,19 +1318,21 @@ static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit ) tgi |= R200_TEXGEN_INPUT_OBJ << inputshift; set_texgen_matrix( rmesa, unit, - (texUnit->TexGenEnabled & S_BIT) ? texUnit->ObjectPlaneS : I, - (texUnit->TexGenEnabled & T_BIT) ? texUnit->ObjectPlaneT : I + 4, - (texUnit->TexGenEnabled & R_BIT) ? texUnit->ObjectPlaneR : I + 8, - (texUnit->TexGenEnabled & Q_BIT) ? texUnit->ObjectPlaneQ : I + 12); + (texUnit->TexGenEnabled & S_BIT) ? texUnit->GenS.ObjectPlane : I, + (texUnit->TexGenEnabled & T_BIT) ? texUnit->GenT.ObjectPlane : I + 4, + (texUnit->TexGenEnabled & R_BIT) ? texUnit->GenR.ObjectPlane : I + 8, + (texUnit->TexGenEnabled & Q_BIT) ? texUnit->GenQ.ObjectPlane : I + 12); } break; case GL_EYE_LINEAR: { GLuint needtgenable = r200_need_dis_texgen( texUnit->TexGenEnabled, - texUnit->EyePlaneS, texUnit->EyePlaneT, - texUnit->EyePlaneR, texUnit->EyePlaneQ ); + texUnit->GenS.EyePlane, + texUnit->GenT.EyePlane, + texUnit->GenR.EyePlane, + texUnit->GenQ.EyePlane ); if (needtgenable & (S_BIT | T_BIT)) { - if (R200_DEBUG & DEBUG_FALLBACKS) + if (R200_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "fallback mixed texgen / eye plane, 0x%x\n", texUnit->TexGenEnabled); return GL_FALSE; @@ -1456,10 +1345,10 @@ static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit ) } tgi |= R200_TEXGEN_INPUT_EYE << inputshift; set_texgen_matrix( rmesa, unit, - (texUnit->TexGenEnabled & S_BIT) ? texUnit->EyePlaneS : I, - (texUnit->TexGenEnabled & T_BIT) ? texUnit->EyePlaneT : I + 4, - (texUnit->TexGenEnabled & R_BIT) ? texUnit->EyePlaneR : I + 8, - (texUnit->TexGenEnabled & Q_BIT) ? texUnit->EyePlaneQ : I + 12); + (texUnit->TexGenEnabled & S_BIT) ? texUnit->GenS.EyePlane : I, + (texUnit->TexGenEnabled & T_BIT) ? texUnit->GenT.EyePlane : I + 4, + (texUnit->TexGenEnabled & R_BIT) ? texUnit->GenR.EyePlane : I + 8, + (texUnit->TexGenEnabled & Q_BIT) ? texUnit->GenQ.EyePlane : I + 12); } break; @@ -1493,9 +1382,9 @@ static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit ) default: /* Unsupported mode, fallback: */ - if (R200_DEBUG & DEBUG_FALLBACKS) + if (R200_DEBUG & RADEON_FALLBACKS) fprintf(stderr, "fallback unsupported texgen, %d\n", - texUnit->GenModeS); + texUnit->GenS.Mode); return GL_FALSE; } @@ -1513,52 +1402,6 @@ static GLboolean r200_validate_texgen( GLcontext *ctx, GLuint unit ) return GL_TRUE; } - -static void disable_tex( GLcontext *ctx, int unit ) -{ - r200ContextPtr rmesa = R200_CONTEXT(ctx); - - if (rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_0_ENABLE<<unit)) { - /* Texture unit disabled */ - if ( rmesa->state.texture.unit[unit].texobj != NULL ) { - /* The old texture is no longer bound to this texture unit. - * Mark it as such. - */ - - rmesa->state.texture.unit[unit].texobj->base.bound &= ~(1UL << unit); - rmesa->state.texture.unit[unit].texobj = NULL; - } - - R200_STATECHANGE( rmesa, ctx ); - rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~(R200_TEX_0_ENABLE << unit); - - R200_STATECHANGE( rmesa, vtx ); - rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3)); - - if (rmesa->TclFallback & (R200_TCL_FALLBACK_TEXGEN_0<<unit)) { - TCL_FALLBACK( ctx, (R200_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE); - } - - /* Actually want to keep all units less than max active texture - * enabled, right? Fix this for >2 texunits. - */ - - { - GLuint tmp = rmesa->TexGenEnabled; - - rmesa->TexGenEnabled &= ~(R200_TEXGEN_TEXMAT_0_ENABLE<<unit); - rmesa->TexGenEnabled &= ~(R200_TEXMAT_0_ENABLE<<unit); - rmesa->TexGenNeedNormals[unit] = GL_FALSE; - rmesa->TexGenCompSel &= ~(R200_OUTPUT_TEX_0 << unit); - - if (tmp != rmesa->TexGenEnabled) { - rmesa->recheck_texgen[unit] = GL_TRUE; - rmesa->NewGLState |= _NEW_TEXTURE_MATRIX; - } - } - } -} - void set_re_cntl_d3d( GLcontext *ctx, int unit, GLboolean use_d3d ) { r200ContextPtr rmesa = R200_CONTEXT(ctx); @@ -1575,237 +1418,169 @@ void set_re_cntl_d3d( GLcontext *ctx, int unit, GLboolean use_d3d ) } } -static GLboolean enable_tex_2d( GLcontext *ctx, int unit ) +/** + * Compute the cached hardware register values for the given texture object. + * + * \param rmesa Context pointer + * \param t the r300 texture object + */ +static void setup_hardware_state(r200ContextPtr rmesa, radeonTexObj *t) { - r200ContextPtr rmesa = R200_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData; - - /* Need to load the 2d images associated with this unit. - */ - if (t->pp_txformat & R200_TXFORMAT_NON_POWER2) { - t->pp_txformat &= ~R200_TXFORMAT_NON_POWER2; - t->base.dirty_images[0] = ~0; + int firstlevel = t->mt ? t->mt->firstLevel : 0; + const struct gl_texture_image *firstImage = t->base.Image[0][firstlevel]; + GLint log2Width, log2Height, log2Depth, texelBytes; + + if ( t->bo ) { + return; } - ASSERT(tObj->Target == GL_TEXTURE_2D || tObj->Target == GL_TEXTURE_1D); + log2Width = firstImage->WidthLog2; + log2Height = firstImage->HeightLog2; + log2Depth = firstImage->DepthLog2; + texelBytes = firstImage->TexFormat->TexelBytes; - if ( t->base.dirty_images[0] ) { - R200_FIREVERTICES( rmesa ); - r200SetTexImages( rmesa, tObj ); - r200UploadTexImages( rmesa, (r200TexObjPtr) tObj->DriverData, 0 ); - if ( !t->base.memBlock && !t->image_override ) - return GL_FALSE; - } - set_re_cntl_d3d( ctx, unit, GL_FALSE ); - - return GL_TRUE; -} - -#if ENABLE_HW_3D_TEXTURE -static GLboolean enable_tex_3d( GLcontext *ctx, int unit ) -{ - r200ContextPtr rmesa = R200_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData; - - /* Need to load the 3d images associated with this unit. - */ - if (t->pp_txformat & R200_TXFORMAT_NON_POWER2) { - t->pp_txformat &= ~R200_TXFORMAT_NON_POWER2; - t->base.dirty_images[0] = ~0; + if (!t->image_override) { + if (VALID_FORMAT(firstImage->TexFormat->MesaFormat)) { + const struct tx_table *table = _mesa_little_endian() ? tx_table_le : + tx_table_be; + + t->pp_txformat &= ~(R200_TXFORMAT_FORMAT_MASK | + R200_TXFORMAT_ALPHA_IN_MAP); + t->pp_txfilter &= ~R200_YUV_TO_RGB; + + t->pp_txformat |= table[ firstImage->TexFormat->MesaFormat ].format; + t->pp_txfilter |= table[ firstImage->TexFormat->MesaFormat ].filter; + } else { + _mesa_problem(NULL, "unexpected texture format in %s", + __FUNCTION__); + return; + } } + + t->pp_txfilter &= ~R200_MAX_MIP_LEVEL_MASK; + t->pp_txfilter |= (t->mt->lastLevel - t->mt->firstLevel) << R200_MAX_MIP_LEVEL_SHIFT; + + t->pp_txformat &= ~(R200_TXFORMAT_WIDTH_MASK | + R200_TXFORMAT_HEIGHT_MASK | + R200_TXFORMAT_CUBIC_MAP_ENABLE | + R200_TXFORMAT_F5_WIDTH_MASK | + R200_TXFORMAT_F5_HEIGHT_MASK); + t->pp_txformat |= ((log2Width << R200_TXFORMAT_WIDTH_SHIFT) | + (log2Height << R200_TXFORMAT_HEIGHT_SHIFT)); + + t->tile_bits = 0; + + t->pp_txformat_x &= ~(R200_DEPTH_LOG2_MASK | R200_TEXCOORD_MASK); + if (t->base.Target == GL_TEXTURE_3D) { + t->pp_txformat_x |= (log2Depth << R200_DEPTH_LOG2_SHIFT); + t->pp_txformat_x |= R200_TEXCOORD_VOLUME; - ASSERT(tObj->Target == GL_TEXTURE_3D); - - /* R100 & R200 do not support mipmaps for 3D textures. - */ - if ( (tObj->MinFilter != GL_NEAREST) && (tObj->MinFilter != GL_LINEAR) ) { - return GL_FALSE; } - - if ( t->base.dirty_images[0] ) { - R200_FIREVERTICES( rmesa ); - r200SetTexImages( rmesa, tObj ); - r200UploadTexImages( rmesa, (r200TexObjPtr) tObj->DriverData, 0 ); - if ( !t->base.memBlock ) - return GL_FALSE; + else if (t->base.Target == GL_TEXTURE_CUBE_MAP) { + ASSERT(log2Width == log2Height); + t->pp_txformat |= ((log2Width << R200_TXFORMAT_F5_WIDTH_SHIFT) | + (log2Height << R200_TXFORMAT_F5_HEIGHT_SHIFT) | + /* don't think we need this bit, if it exists at all - fglrx does not set it */ + (R200_TXFORMAT_CUBIC_MAP_ENABLE)); + t->pp_txformat_x |= R200_TEXCOORD_CUBIC_ENV; + t->pp_cubic_faces = ((log2Width << R200_FACE_WIDTH_1_SHIFT) | + (log2Height << R200_FACE_HEIGHT_1_SHIFT) | + (log2Width << R200_FACE_WIDTH_2_SHIFT) | + (log2Height << R200_FACE_HEIGHT_2_SHIFT) | + (log2Width << R200_FACE_WIDTH_3_SHIFT) | + (log2Height << R200_FACE_HEIGHT_3_SHIFT) | + (log2Width << R200_FACE_WIDTH_4_SHIFT) | + (log2Height << R200_FACE_HEIGHT_4_SHIFT)); } - - set_re_cntl_d3d( ctx, unit, GL_TRUE ); - - return GL_TRUE; -} -#endif - -static GLboolean enable_tex_cube( GLcontext *ctx, int unit ) -{ - r200ContextPtr rmesa = R200_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData; - GLuint face; - - /* Need to load the 2d images associated with this unit. - */ - if (t->pp_txformat & R200_TXFORMAT_NON_POWER2) { - t->pp_txformat &= ~R200_TXFORMAT_NON_POWER2; - for (face = 0; face < 6; face++) - t->base.dirty_images[face] = ~0; + else { + /* If we don't in fact send enough texture coordinates, q will be 1, + * making TEXCOORD_PROJ act like TEXCOORD_NONPROJ (Right?) + */ + t->pp_txformat_x |= R200_TEXCOORD_PROJ; } - ASSERT(tObj->Target == GL_TEXTURE_CUBE_MAP); + t->pp_txsize = (((firstImage->Width - 1) << R200_PP_TX_WIDTHMASK_SHIFT) + | ((firstImage->Height - 1) << R200_PP_TX_HEIGHTMASK_SHIFT)); - if ( t->base.dirty_images[0] || t->base.dirty_images[1] || - t->base.dirty_images[2] || t->base.dirty_images[3] || - t->base.dirty_images[4] || t->base.dirty_images[5] ) { - /* flush */ - R200_FIREVERTICES( rmesa ); - /* layout memory space, once for all faces */ - r200SetTexImages( rmesa, tObj ); - } - - /* upload (per face) */ - for (face = 0; face < 6; face++) { - if (t->base.dirty_images[face]) { - r200UploadTexImages( rmesa, (r200TexObjPtr) tObj->DriverData, face ); - } - } - - if ( !t->base.memBlock ) { - /* texmem alloc failed, use s/w fallback */ - return GL_FALSE; + if ( !t->image_override ) { + if (firstImage->IsCompressed) + t->pp_txpitch = (firstImage->Width + 63) & ~(63); + else + t->pp_txpitch = ((firstImage->Width * texelBytes) + 63) & ~(63); + t->pp_txpitch -= 32; } - set_re_cntl_d3d( ctx, unit, GL_TRUE ); - - return GL_TRUE; -} - -static GLboolean enable_tex_rect( GLcontext *ctx, int unit ) -{ - r200ContextPtr rmesa = R200_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData; - - if (!(t->pp_txformat & R200_TXFORMAT_NON_POWER2)) { + if (t->base.Target == GL_TEXTURE_RECTANGLE_NV) { t->pp_txformat |= R200_TXFORMAT_NON_POWER2; - t->base.dirty_images[0] = ~0; - } - - ASSERT(tObj->Target == GL_TEXTURE_RECTANGLE_NV); - - if ( t->base.dirty_images[0] ) { - R200_FIREVERTICES( rmesa ); - r200SetTexImages( rmesa, tObj ); - r200UploadTexImages( rmesa, (r200TexObjPtr) tObj->DriverData, 0 ); - if ( !t->base.memBlock && - !t->image_override && - !rmesa->prefer_gart_client_texturing ) - return GL_FALSE; } - set_re_cntl_d3d( ctx, unit, GL_FALSE ); - - return GL_TRUE; } - -static GLboolean update_tex_common( GLcontext *ctx, int unit ) +static GLboolean r200_validate_texture(GLcontext *ctx, struct gl_texture_object *texObj, int unit) { r200ContextPtr rmesa = R200_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r200TexObjPtr t = (r200TexObjPtr) tObj->DriverData; - - /* Fallback if there's a texture border */ - if ( tObj->Image[0][tObj->BaseLevel]->Border > 0 ) - return GL_FALSE; - - /* Update state if this is a different texture object to last - * time. - */ - if ( rmesa->state.texture.unit[unit].texobj != t ) { - if ( rmesa->state.texture.unit[unit].texobj != NULL ) { - /* The old texture is no longer bound to this texture unit. - * Mark it as such. - */ + radeonTexObj *t = radeon_tex_obj(texObj); - rmesa->state.texture.unit[unit].texobj->base.bound &= - ~(1UL << unit); - } - - rmesa->state.texture.unit[unit].texobj = t; - t->base.bound |= (1UL << unit); - t->dirty_state |= 1<<unit; - driUpdateTextureLRU( (driTextureObject *) t ); /* XXX: should be locked! */ - } - - - /* Newly enabled? - */ - if ( 1|| !(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (R200_TEX_0_ENABLE<<unit))) { - R200_STATECHANGE( rmesa, ctx ); - rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= R200_TEX_0_ENABLE << unit; - - R200_STATECHANGE( rmesa, vtx ); - rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3)); - rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] |= 4 << (unit * 3); + if (!radeon_validate_texture_miptree(ctx, texObj)) + return GL_FALSE; - rmesa->recheck_texgen[unit] = GL_TRUE; - } + r200_validate_texgen(ctx, unit); + /* Configure the hardware registers (more precisely, the cached version + * of the hardware registers). */ + setup_hardware_state(rmesa, t); + + if (texObj->Target == GL_TEXTURE_RECTANGLE_NV || + texObj->Target == GL_TEXTURE_2D || + texObj->Target == GL_TEXTURE_1D) + set_re_cntl_d3d( ctx, unit, GL_FALSE ); + else + set_re_cntl_d3d( ctx, unit, GL_TRUE ); + R200_STATECHANGE( rmesa, ctx ); + rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= R200_TEX_0_ENABLE << unit; + + R200_STATECHANGE( rmesa, vtx ); + rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] &= ~(7 << (unit * 3)); + rmesa->hw.vtx.cmd[VTX_TCL_OUTPUT_VTXFMT_1] |= 4 << (unit * 3); - if (t->dirty_state & (1<<unit)) { - import_tex_obj_state( rmesa, unit, t ); - } + rmesa->recheck_texgen[unit] = GL_TRUE; + import_tex_obj_state( rmesa, unit, t ); if (rmesa->recheck_texgen[unit]) { GLboolean fallback = !r200_validate_texgen( ctx, unit ); TCL_FALLBACK( ctx, (R200_TCL_FALLBACK_TEXGEN_0<<unit), fallback); rmesa->recheck_texgen[unit] = 0; - rmesa->NewGLState |= _NEW_TEXTURE_MATRIX; + rmesa->radeon.NewGLState |= _NEW_TEXTURE_MATRIX; } - FALLBACK( rmesa, R200_FALLBACK_BORDER_MODE, t->border_fallback ); - return !t->border_fallback; -} + t->validated = GL_TRUE; + FALLBACK( rmesa, RADEON_FALLBACK_BORDER_MODE, t->border_fallback ); + return !t->border_fallback; +} -static GLboolean r200UpdateTextureUnit( GLcontext *ctx, int unit ) +static GLboolean r200UpdateTextureUnit(GLcontext *ctx, int unit) { r200ContextPtr rmesa = R200_CONTEXT(ctx); GLuint unitneeded = rmesa->state.texture.unit[unit].unitneeded; - if ( unitneeded & (TEXTURE_RECT_BIT) ) { - return (enable_tex_rect( ctx, unit ) && - update_tex_common( ctx, unit )); - } - else if ( unitneeded & (TEXTURE_1D_BIT | TEXTURE_2D_BIT) ) { - return (enable_tex_2d( ctx, unit ) && - update_tex_common( ctx, unit )); - } -#if ENABLE_HW_3D_TEXTURE - else if ( unitneeded & (TEXTURE_3D_BIT) ) { - return (enable_tex_3d( ctx, unit ) && - update_tex_common( ctx, unit )); - } -#endif - else if ( unitneeded & (TEXTURE_CUBE_BIT) ) { - return (enable_tex_cube( ctx, unit ) && - update_tex_common( ctx, unit )); - } - else if ( unitneeded ) { - return GL_FALSE; - } - else { - disable_tex( ctx, unit ); - return GL_TRUE; + if (!unitneeded) { + /* disable the unit */ + disable_tex_obj_state(rmesa, unit); + return GL_TRUE; } + + if (!r200_validate_texture(ctx, ctx->Texture.Unit[unit]._Current, unit)) { + _mesa_warning(ctx, + "failed to validate texture for unit %d.\n", + unit); + rmesa->state.texture.unit[unit].texobj = NULL; + return GL_FALSE; + } + + rmesa->state.texture.unit[unit].texobj = radeon_tex_obj(ctx->Texture.Unit[unit]._Current); + return GL_TRUE; } @@ -1846,11 +1621,11 @@ void r200UpdateTextureState( GLcontext *ctx ) FALLBACK( rmesa, R200_FALLBACK_TEXTURE, !ok ); - if (rmesa->TclFallback) + if (rmesa->radeon.TclFallback) r200ChooseVertexState( ctx ); - if (rmesa->r200Screen->chip_family == CHIP_FAMILY_R200) { + if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R200) { /* * T0 hang workaround ------------- @@ -1863,7 +1638,7 @@ void r200UpdateTextureState( GLcontext *ctx ) R200_STATECHANGE(rmesa, tex[1]); rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= R200_TEX_1_ENABLE; if (!(rmesa->hw.cst.cmd[CST_PP_CNTL_X] & R200_PPX_TEX_1_ENABLE)) - rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK; + rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK; rmesa->hw.tex[1].cmd[TEX_PP_TXFORMAT] |= R200_TXFORMAT_LOOKUP_DISABLE; } else if (!ctx->ATIFragmentShader._Enabled) { |