/** * Translate a Mesa fragment shader into a TGSI shader using extra info in * the key. * \return new fragment program variant */ static struct st_fp_variant * st_translate_fragment_program(struct st_context *st, struct st_fragment_program *stfp, const struct st_fp_variant_key *key) { struct pipe_context *pipe = st->pipe; struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant); GLboolean deleteFP = GL_FALSE; GLuint outputMapping[FRAG_RESULT_MAX]; GLuint inputMapping[VARYING_SLOT_MAX]; GLuint interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */ GLuint attr; GLbitfield64 inputsRead; struct ureg_program *ureg; GLboolean write_all = GL_FALSE; ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS]; GLboolean is_centroid[PIPE_MAX_SHADER_INPUTS]; uint fs_num_inputs = 0; ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS]; uint fs_num_outputs = 0; if (!variant) return NULL; assert(!(key->bitmap && key->drawpixels)); if (key->bitmap) { /* glBitmap drawing */ struct gl_fragment_program *fp; /* we free this temp program below */ st_make_bitmap_fragment_program(st, &stfp->Base, &fp, &variant->bitmap_sampler); variant->parameters = _mesa_clone_parameter_list(fp->Base.Parameters); stfp = st_fragment_program(fp); deleteFP = GL_TRUE; } else if (key->drawpixels) { /* glDrawPixels drawing */ struct gl_fragment_program *fp; /* we free this temp program below */ if (key->drawpixels_z || key->drawpixels_stencil) { fp = st_make_drawpix_z_stencil_program(st, key->drawpixels_z, key->drawpixels_stencil); } else { /* RGBA */ st_make_drawpix_fragment_program(st, &stfp->Base, &fp); variant->parameters = _mesa_clone_parameter_list(fp->Base.Parameters); deleteFP = GL_TRUE; } stfp = st_fragment_program(fp); } if (!stfp->glsl_to_tgsi) _mesa_remove_output_reads(&stfp->Base.Base, PROGRAM_OUTPUT); /* * Convert Mesa program inputs to TGSI input register semantics. */ inputsRead = stfp->Base.Base.InputsRead; for (attr = 0; attr < VARYING_SLOT_MAX; attr++) { if ((inputsRead & BITFIELD64_BIT(attr)) != 0) { const GLuint slot = fs_num_inputs++; inputMapping[attr] = slot; is_centroid[slot] = (stfp->Base.IsCentroid & BITFIELD64_BIT(attr)) != 0; switch (attr) { case VARYING_SLOT_POS: input_semantic_name[slot] = TGSI_SEMANTIC_POSITION; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_LINEAR; break; case VARYING_SLOT_COL0: input_semantic_name[slot] = TGSI_SEMANTIC_COLOR; input_semantic_index[slot] = 0; interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr], TRUE); break; case VARYING_SLOT_COL1: input_semantic_name[slot] = TGSI_SEMANTIC_COLOR; input_semantic_index[slot] = 1; interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr], TRUE); break; case VARYING_SLOT_FOGC: input_semantic_name[slot] = TGSI_SEMANTIC_FOG; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; break; case VARYING_SLOT_FACE: input_semantic_name[slot] = TGSI_SEMANTIC_FACE; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_CONSTANT; break; case VARYING_SLOT_CLIP_DIST0: input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; break; case VARYING_SLOT_CLIP_DIST1: input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST; input_semantic_index[slot] = 1; interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; break; /* In most cases, there is nothing special about these * inputs, so adopt a convention to use the generic * semantic name and the mesa VARYING_SLOT_ number as the * index. * * All that is required is that the vertex shader labels * its own outputs similarly, and that the vertex shader * generates at least every output required by the * fragment shader plus fixed-function hardware (such as * BFC). * * There is no requirement that semantic indexes start at * zero or be restricted to a particular range -- nobody * should be building tables based on semantic index. */ case VARYING_SLOT_PNTC: case VARYING_SLOT_TEX0: case VARYING_SLOT_TEX1: case VARYING_SLOT_TEX2: case VARYING_SLOT_TEX3: case VARYING_SLOT_TEX4: case VARYING_SLOT_TEX5: case VARYING_SLOT_TEX6: case VARYING_SLOT_TEX7: case VARYING_SLOT_VAR0: default: /* Actually, let's try and zero-base this just for * readability of the generated TGSI. */ assert(attr >= VARYING_SLOT_TEX0); input_semantic_index[slot] = (attr - VARYING_SLOT_TEX0); input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; if (attr == VARYING_SLOT_PNTC) interpMode[slot] = TGSI_INTERPOLATE_LINEAR; else interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr], FALSE); break; } } else { inputMapping[attr] = -1; } } /* * Semantics and mapping for outputs */ { uint numColors = 0; GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten; /* if z is written, emit that first */ if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) { fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION; fs_output_semantic_index[fs_num_outputs] = 0; outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs; fs_num_outputs++; outputsWritten &= ~(1 << FRAG_RESULT_DEPTH); } if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) { fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL; fs_output_semantic_index[fs_num_outputs] = 0; outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs; fs_num_outputs++; outputsWritten &= ~(1 << FRAG_RESULT_STENCIL); } /* handle remaining outputs (color) */ for (attr = 0; attr < FRAG_RESULT_MAX; attr++) { if (outputsWritten & BITFIELD64_BIT(attr)) { switch (attr) { case FRAG_RESULT_DEPTH: case FRAG_RESULT_STENCIL: /* handled above */ assert(0); break; case FRAG_RESULT_COLOR: write_all = GL_TRUE; /* fallthrough */ default: assert(attr == FRAG_RESULT_COLOR || (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX)); fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR; fs_output_semantic_index[fs_num_outputs] = numColors; outputMapping[attr] = fs_num_outputs; numColors++; break; } fs_num_outputs++; } } } ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT ); if (ureg == NULL) { free(variant); return NULL; } if (ST_DEBUG & DEBUG_MESA) { _mesa_print_program(&stfp->Base.Base); _mesa_print_program_parameters(st->ctx, &stfp->Base.Base); debug_printf("\n"); } if (write_all == GL_TRUE) ureg_property_fs_color0_writes_all_cbufs(ureg, 1); if (stfp->Base.FragDepthLayout != FRAG_DEPTH_LAYOUT_NONE) { switch (stfp->Base.FragDepthLayout) { case FRAG_DEPTH_LAYOUT_ANY: ureg_property_fs_depth_layout(ureg, TGSI_FS_DEPTH_LAYOUT_ANY); break; case FRAG_DEPTH_LAYOUT_GREATER: ureg_property_fs_depth_layout(ureg, TGSI_FS_DEPTH_LAYOUT_GREATER); break; case FRAG_DEPTH_LAYOUT_LESS: ureg_property_fs_depth_layout(ureg, TGSI_FS_DEPTH_LAYOUT_LESS); break; case FRAG_DEPTH_LAYOUT_UNCHANGED: ureg_property_fs_depth_layout(ureg, TGSI_FS_DEPTH_LAYOUT_UNCHANGED); break; default: assert(0); } } if (stfp->glsl_to_tgsi) st_translate_program(st->ctx, TGSI_PROCESSOR_FRAGMENT, ureg, stfp->glsl_to_tgsi, &stfp->Base.Base, /* inputs */ fs_num_inputs, inputMapping, input_semantic_name, input_semantic_index, interpMode, is_centroid, /* outputs */ fs_num_outputs, outputMapping, fs_output_semantic_name, fs_output_semantic_index, FALSE, key->clamp_color ); else st_translate_mesa_program(st->ctx, TGSI_PROCESSOR_FRAGMENT, ureg, &stfp->Base.Base, /* inputs */ fs_num_inputs, inputMapping, input_semantic_name, input_semantic_index, interpMode, /* outputs */ fs_num_outputs, outputMapping, fs_output_semantic_name, fs_output_semantic_index, FALSE, key->clamp_color); variant->tgsi.tokens = ureg_get_tokens( ureg, NULL ); ureg_destroy( ureg ); /* fill in variant */ variant->driver_shader = pipe->create_fs_state(pipe, &variant->tgsi); variant->key = *key; if (ST_DEBUG & DEBUG_TGSI) { tgsi_dump( variant->tgsi.tokens, 0/*TGSI_DUMP_VERBOSE*/ ); debug_printf("\n"); } if (deleteFP) { /* Free the temporary program made above */ struct gl_fragment_program *fp = &stfp->Base; _mesa_reference_fragprog(st->ctx, &fp, NULL); } return variant; }
/** * Translate a Mesa fragment shader into a TGSI shader using extra info in * the key. * \return new fragment program variant */ static struct st_fp_variant * st_translate_fragment_program(struct st_context *st, struct st_fragment_program *stfp, const struct st_fp_variant_key *key) { struct pipe_context *pipe = st->pipe; struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant); GLboolean deleteFP = GL_FALSE; GLuint outputMapping[FRAG_RESULT_MAX]; GLuint inputMapping[VARYING_SLOT_MAX]; GLuint interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */ GLuint interpLocation[PIPE_MAX_SHADER_INPUTS]; GLuint attr; GLbitfield64 inputsRead; struct ureg_program *ureg; GLboolean write_all = GL_FALSE; ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS]; uint fs_num_inputs = 0; ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS]; uint fs_num_outputs = 0; if (!variant) return NULL; assert(!(key->bitmap && key->drawpixels)); if (key->bitmap) { /* glBitmap drawing */ struct gl_fragment_program *fp; /* we free this temp program below */ st_make_bitmap_fragment_program(st, &stfp->Base, &fp, &variant->bitmap_sampler); variant->parameters = _mesa_clone_parameter_list(fp->Base.Parameters); stfp = st_fragment_program(fp); deleteFP = GL_TRUE; } else if (key->drawpixels) { /* glDrawPixels drawing */ struct gl_fragment_program *fp; /* we free this temp program below */ if (key->drawpixels_z || key->drawpixels_stencil) { fp = st_make_drawpix_z_stencil_program(st, key->drawpixels_z, key->drawpixels_stencil); } else { /* RGBA */ st_make_drawpix_fragment_program(st, &stfp->Base, &fp); variant->parameters = _mesa_clone_parameter_list(fp->Base.Parameters); deleteFP = GL_TRUE; } stfp = st_fragment_program(fp); } if (!stfp->glsl_to_tgsi) _mesa_remove_output_reads(&stfp->Base.Base, PROGRAM_OUTPUT); /* * Convert Mesa program inputs to TGSI input register semantics. */ inputsRead = stfp->Base.Base.InputsRead; for (attr = 0; attr < VARYING_SLOT_MAX; attr++) { if ((inputsRead & BITFIELD64_BIT(attr)) != 0) { const GLuint slot = fs_num_inputs++; inputMapping[attr] = slot; if (stfp->Base.IsCentroid & BITFIELD64_BIT(attr)) interpLocation[slot] = TGSI_INTERPOLATE_LOC_CENTROID; else if (stfp->Base.IsSample & BITFIELD64_BIT(attr)) interpLocation[slot] = TGSI_INTERPOLATE_LOC_SAMPLE; else interpLocation[slot] = TGSI_INTERPOLATE_LOC_CENTER; if (key->persample_shading) interpLocation[slot] = TGSI_INTERPOLATE_LOC_SAMPLE; switch (attr) { case VARYING_SLOT_POS: input_semantic_name[slot] = TGSI_SEMANTIC_POSITION; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_LINEAR; break; case VARYING_SLOT_COL0: input_semantic_name[slot] = TGSI_SEMANTIC_COLOR; input_semantic_index[slot] = 0; interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr], TRUE); break; case VARYING_SLOT_COL1: input_semantic_name[slot] = TGSI_SEMANTIC_COLOR; input_semantic_index[slot] = 1; interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr], TRUE); break; case VARYING_SLOT_FOGC: input_semantic_name[slot] = TGSI_SEMANTIC_FOG; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; break; case VARYING_SLOT_FACE: input_semantic_name[slot] = TGSI_SEMANTIC_FACE; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_CONSTANT; break; case VARYING_SLOT_PRIMITIVE_ID: input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_CONSTANT; break; case VARYING_SLOT_LAYER: input_semantic_name[slot] = TGSI_SEMANTIC_LAYER; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_CONSTANT; break; case VARYING_SLOT_VIEWPORT: input_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_CONSTANT; break; case VARYING_SLOT_CLIP_DIST0: input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; break; case VARYING_SLOT_CLIP_DIST1: input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST; input_semantic_index[slot] = 1; interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; break; /* In most cases, there is nothing special about these * inputs, so adopt a convention to use the generic * semantic name and the mesa VARYING_SLOT_ number as the * index. * * All that is required is that the vertex shader labels * its own outputs similarly, and that the vertex shader * generates at least every output required by the * fragment shader plus fixed-function hardware (such as * BFC). * * However, some drivers may need us to identify the PNTC and TEXi * varyings if, for example, their capability to replace them with * sprite coordinates is limited. */ case VARYING_SLOT_PNTC: if (st->needs_texcoord_semantic) { input_semantic_name[slot] = TGSI_SEMANTIC_PCOORD; input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_LINEAR; break; } /* fall through */ case VARYING_SLOT_TEX0: case VARYING_SLOT_TEX1: case VARYING_SLOT_TEX2: case VARYING_SLOT_TEX3: case VARYING_SLOT_TEX4: case VARYING_SLOT_TEX5: case VARYING_SLOT_TEX6: case VARYING_SLOT_TEX7: if (st->needs_texcoord_semantic) { input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD; input_semantic_index[slot] = attr - VARYING_SLOT_TEX0; interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr], FALSE); break; } /* fall through */ case VARYING_SLOT_VAR0: default: /* Semantic indices should be zero-based because drivers may choose * to assign a fixed slot determined by that index. * This is useful because ARB_separate_shader_objects uses location * qualifiers for linkage, and if the semantic index corresponds to * these locations, linkage passes in the driver become unecessary. * * If needs_texcoord_semantic is true, no semantic indices will be * consumed for the TEXi varyings, and we can base the locations of * the user varyings on VAR0. Otherwise, we use TEX0 as base index. */ assert(attr >= VARYING_SLOT_TEX0); input_semantic_index[slot] = st->needs_texcoord_semantic ? (attr - VARYING_SLOT_VAR0) : (attr - VARYING_SLOT_TEX0); input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; if (attr == VARYING_SLOT_PNTC) interpMode[slot] = TGSI_INTERPOLATE_LINEAR; else interpMode[slot] = st_translate_interp(stfp->Base.InterpQualifier[attr], FALSE); break; } } else { inputMapping[attr] = -1; } } /* * Semantics and mapping for outputs */ { uint numColors = 0; GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten; /* if z is written, emit that first */ if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) { fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION; fs_output_semantic_index[fs_num_outputs] = 0; outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs; fs_num_outputs++; outputsWritten &= ~(1 << FRAG_RESULT_DEPTH); } if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) { fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL; fs_output_semantic_index[fs_num_outputs] = 0; outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs; fs_num_outputs++; outputsWritten &= ~(1 << FRAG_RESULT_STENCIL); } if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)) { fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_SAMPLEMASK; fs_output_semantic_index[fs_num_outputs] = 0; outputMapping[FRAG_RESULT_SAMPLE_MASK] = fs_num_outputs; fs_num_outputs++; outputsWritten &= ~(1 << FRAG_RESULT_SAMPLE_MASK); } /* handle remaining outputs (color) */ for (attr = 0; attr < FRAG_RESULT_MAX; attr++) { if (outputsWritten & BITFIELD64_BIT(attr)) { switch (attr) { case FRAG_RESULT_DEPTH: case FRAG_RESULT_STENCIL: case FRAG_RESULT_SAMPLE_MASK: /* handled above */ assert(0); break; case FRAG_RESULT_COLOR: write_all = GL_TRUE; /* fallthrough */ default: assert(attr == FRAG_RESULT_COLOR || (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX)); fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR; fs_output_semantic_index[fs_num_outputs] = numColors; outputMapping[attr] = fs_num_outputs; numColors++; break; } fs_num_outputs++; } } } ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT ); if (ureg == NULL) { free(variant); return NULL; } if (ST_DEBUG & DEBUG_MESA) { _mesa_print_program(&stfp->Base.Base); _mesa_print_program_parameters(st->ctx, &stfp->Base.Base); debug_printf("\n"); } if (write_all == GL_TRUE) ureg_property_fs_color0_writes_all_cbufs(ureg, 1); if (stfp->Base.FragDepthLayout != FRAG_DEPTH_LAYOUT_NONE) { switch (stfp->Base.FragDepthLayout) { case FRAG_DEPTH_LAYOUT_ANY: ureg_property_fs_depth_layout(ureg, TGSI_FS_DEPTH_LAYOUT_ANY); break; case FRAG_DEPTH_LAYOUT_GREATER: ureg_property_fs_depth_layout(ureg, TGSI_FS_DEPTH_LAYOUT_GREATER); break; case FRAG_DEPTH_LAYOUT_LESS: ureg_property_fs_depth_layout(ureg, TGSI_FS_DEPTH_LAYOUT_LESS); break; case FRAG_DEPTH_LAYOUT_UNCHANGED: ureg_property_fs_depth_layout(ureg, TGSI_FS_DEPTH_LAYOUT_UNCHANGED); break; default: assert(0); } } if (stfp->glsl_to_tgsi) st_translate_program(st->ctx, TGSI_PROCESSOR_FRAGMENT, ureg, stfp->glsl_to_tgsi, &stfp->Base.Base, /* inputs */ fs_num_inputs, inputMapping, input_semantic_name, input_semantic_index, interpMode, interpLocation, /* outputs */ fs_num_outputs, outputMapping, fs_output_semantic_name, fs_output_semantic_index, FALSE, key->clamp_color ); else st_translate_mesa_program(st->ctx, TGSI_PROCESSOR_FRAGMENT, ureg, &stfp->Base.Base, /* inputs */ fs_num_inputs, inputMapping, input_semantic_name, input_semantic_index, interpMode, /* outputs */ fs_num_outputs, outputMapping, fs_output_semantic_name, fs_output_semantic_index, FALSE, key->clamp_color); variant->tgsi.tokens = ureg_get_tokens( ureg, NULL ); ureg_destroy( ureg ); /* fill in variant */ variant->driver_shader = pipe->create_fs_state(pipe, &variant->tgsi); variant->key = *key; if (ST_DEBUG & DEBUG_TGSI) { tgsi_dump( variant->tgsi.tokens, 0/*TGSI_DUMP_VERBOSE*/ ); debug_printf("\n"); } if (deleteFP) { /* Free the temporary program made above */ struct gl_fragment_program *fp = &stfp->Base; _mesa_reference_fragprog(st->ctx, &fp, NULL); } return variant; }