From 07fafc7c9346aa260829603bf3188596481e9e62 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 15 Nov 2009 11:15:25 -0800 Subject: [PATCH] mesa/st: refactor vertex and fragment shader translation Translate vertex shaders independently of fragment shaders. Previously tried to make fragment shader semantic indexes always start at zero and exclude holes. This was unnecessary but meant that vertex shader translation had to be adjusted to take this into account. Now use a fixed scheme for labelling special FS input semantics (color, etc), and another fixed scheme for the generics. With this, vertex shaders can be translated independently of the bound fragment shader, assuming mesa has done its own job and ensured that the vertex shader provides at least the inputs the fragment shader is looking for. The state-tracker didn't attempt to do anything about this previously, so it shouldn't be needed now. --- src/mesa/state_tracker/st_atom.c | 3 +- src/mesa/state_tracker/st_atom.h | 3 +- src/mesa/state_tracker/st_atom_shader.c | 326 ++++++---------------- src/mesa/state_tracker/st_cb_bitmap.c | 5 - src/mesa/state_tracker/st_cb_drawpixels.c | 8 +- src/mesa/state_tracker/st_cb_program.c | 43 +-- src/mesa/state_tracker/st_context.h | 2 + src/mesa/state_tracker/st_debug.c | 3 +- src/mesa/state_tracker/st_draw.c | 2 +- src/mesa/state_tracker/st_draw_feedback.c | 8 +- src/mesa/state_tracker/st_program.c | 267 ++++++++---------- src/mesa/state_tracker/st_program.h | 75 +++-- 12 files changed, 276 insertions(+), 469 deletions(-) diff --git a/src/mesa/state_tracker/st_atom.c b/src/mesa/state_tracker/st_atom.c index ca15ce1b474..dfce955fd96 100644 --- a/src/mesa/state_tracker/st_atom.c +++ b/src/mesa/state_tracker/st_atom.c @@ -46,7 +46,8 @@ static const struct st_tracked_state *atoms[] = &st_update_clip, &st_finalize_textures, - &st_update_shader, + &st_update_fp, + &st_update_vp, &st_update_rasterizer, &st_update_polygon_stipple, diff --git a/src/mesa/state_tracker/st_atom.h b/src/mesa/state_tracker/st_atom.h index c7cffd85c8a..f34b49203b2 100644 --- a/src/mesa/state_tracker/st_atom.h +++ b/src/mesa/state_tracker/st_atom.h @@ -47,7 +47,8 @@ void st_validate_state( struct st_context *st ); extern const struct st_tracked_state st_update_framebuffer; extern const struct st_tracked_state st_update_clip; extern const struct st_tracked_state st_update_depth_stencil_alpha; -extern const struct st_tracked_state st_update_shader; +extern const struct st_tracked_state st_update_fp; +extern const struct st_tracked_state st_update_vp; extern const struct st_tracked_state st_update_rasterizer; extern const struct st_tracked_state st_update_polygon_stipple; extern const struct st_tracked_state st_update_viewport; diff --git a/src/mesa/state_tracker/st_atom_shader.c b/src/mesa/state_tracker/st_atom_shader.c index ee649be885e..09baff875bc 100644 --- a/src/mesa/state_tracker/st_atom_shader.c +++ b/src/mesa/state_tracker/st_atom_shader.c @@ -56,82 +56,18 @@ #include "st_mesa_to_tgsi.h" -/** - * This represents a vertex program, especially translated to match - * the inputs of a particular fragment shader. - */ -struct translated_vertex_program -{ - struct st_vertex_program *master; - - /** The fragment shader "signature" this vertex shader is meant for: */ - GLbitfield frag_inputs; - /** Compared against master vertex program's serialNo: */ - GLuint serialNo; - /** Maps VERT_RESULT_x to slot */ - GLuint output_to_slot[VERT_RESULT_MAX]; - ubyte output_to_semantic_name[VERT_RESULT_MAX]; - ubyte output_to_semantic_index[VERT_RESULT_MAX]; - - /** Pointer to the translated vertex program */ - struct st_vertex_program *vp; - - struct translated_vertex_program *next; /**< next in linked list */ -}; - - -/** - * Given a vertex program output attribute, return the corresponding - * fragment program input attribute. - * \return -1 for vertex outputs that have no corresponding fragment input +/* + * Translate fragment program if needed. */ -static GLint -vp_out_to_fp_in(GLuint vertResult) -{ - if (vertResult >= VERT_RESULT_TEX0 && - vertResult < VERT_RESULT_TEX0 + MAX_TEXTURE_COORD_UNITS) - return FRAG_ATTRIB_TEX0 + (vertResult - VERT_RESULT_TEX0); - - if (vertResult >= VERT_RESULT_VAR0 && - vertResult < VERT_RESULT_VAR0 + MAX_VARYING) - return FRAG_ATTRIB_VAR0 + (vertResult - VERT_RESULT_VAR0); - - switch (vertResult) { - case VERT_RESULT_HPOS: - return FRAG_ATTRIB_WPOS; - case VERT_RESULT_COL0: - return FRAG_ATTRIB_COL0; - case VERT_RESULT_COL1: - return FRAG_ATTRIB_COL1; - case VERT_RESULT_FOGC: - return FRAG_ATTRIB_FOGC; - default: - /* Back-face colors, edge flags, etc */ - return -1; - } -} - - -/** - * Find a translated vertex program that corresponds to stvp and - * has outputs matched to stfp's inputs. - * This performs vertex and fragment translation (to TGSI) when needed. - */ -static struct translated_vertex_program * -find_translated_vp(struct st_context *st, - struct st_vertex_program *stvp, - struct st_fragment_program *stfp) +static void +translate_fp(struct st_context *st, + struct st_fragment_program *stfp) { - static const GLuint UNUSED = ~0; - struct translated_vertex_program *xvp; const GLbitfield fragInputsRead = stfp->Base.Base.InputsRead; - /* - * Translate fragment program if needed. - */ if (!stfp->state.tokens) { GLuint inAttr, numIn = 0; @@ -141,7 +77,7 @@ find_translated_vp(struct st_context *st, numIn++; } else { - stfp->input_to_slot[inAttr] = UNUSED; + stfp->input_to_slot[inAttr] = -1; } } @@ -151,170 +87,63 @@ find_translated_vp(struct st_context *st, st_translate_fragment_program(st, stfp, stfp->input_to_slot); } +} - /* See if we've got a translated vertex program whose outputs match - * the fragment program's inputs. - * XXX This could be a hash lookup, using InputsRead as the key. - */ - for (xvp = stfp->vertex_programs; xvp; xvp = xvp->next) { - if (xvp->master == stvp && xvp->frag_inputs == fragInputsRead) { - break; - } - } - /* No? Allocate translated vp object now */ - if (!xvp) { - xvp = ST_CALLOC_STRUCT(translated_vertex_program); - xvp->frag_inputs = fragInputsRead; - xvp->master = stvp; +/** + * Find a translated vertex program that corresponds to stvp and + * has outputs matched to stfp's inputs. + * This performs vertex and fragment translation (to TGSI) when needed. + */ +static struct st_vp_varient * +find_translated_vp(struct st_context *st, + struct st_vertex_program *stvp ) +{ + struct st_vp_varient *vpv; + struct st_vp_varient_key key; - xvp->next = stfp->vertex_programs; - stfp->vertex_programs = xvp; - } + /* Nothing in our key yet. This will change: + */ + memset(&key, 0, sizeof key); + key.dummy = 0; - /* See if we need to translate vertex program to TGSI form */ - if (xvp->serialNo != stvp->serialNo) { - GLuint outAttr; - const GLbitfield outputsWritten = stvp->Base.Base.OutputsWritten; - GLuint numVpOuts = 0; - GLboolean emitPntSize = GL_FALSE, emitBFC0 = GL_FALSE, emitBFC1 = GL_FALSE; - GLbitfield usedGenerics = 0x0; - GLbitfield usedOutputSlots = 0x0; - - /* Compute mapping of vertex program outputs to slots, which depends - * on the fragment program's input->slot mapping. + /* Do we need to throw away old translations after a change in the + * GL program string? + */ + if (stvp->serialNo != stvp->lastSerialNo) { + /* These may have changed if the program string changed. */ - for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) { - /* set defaults: */ - xvp->output_to_slot[outAttr] = UNUSED; - xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_COUNT; - xvp->output_to_semantic_index[outAttr] = 99; - - if (outAttr == VERT_RESULT_HPOS) { - /* always put xformed position into slot zero */ - GLuint slot = 0; - xvp->output_to_slot[VERT_RESULT_HPOS] = slot; - xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_POSITION; - xvp->output_to_semantic_index[outAttr] = 0; - numVpOuts++; - usedOutputSlots |= (1 << slot); - } - else if (outputsWritten & (1 << outAttr)) { - /* see if the frag prog wants this vert output */ - GLint fpInAttrib = vp_out_to_fp_in(outAttr); - if (fpInAttrib >= 0) { - GLuint fpInSlot = stfp->input_to_slot[fpInAttrib]; - if (fpInSlot != ~0) { - /* match this vp output to the fp input */ - GLuint vpOutSlot = stfp->input_map[fpInSlot]; - xvp->output_to_slot[outAttr] = vpOutSlot; - xvp->output_to_semantic_name[outAttr] = stfp->input_semantic_name[fpInSlot]; - xvp->output_to_semantic_index[outAttr] = stfp->input_semantic_index[fpInSlot]; - numVpOuts++; - usedOutputSlots |= (1 << vpOutSlot); - } - else { -#if 0 /*debug*/ - printf("VP output %d not used by FP\n", outAttr); -#endif - } - } - else if (outAttr == VERT_RESULT_PSIZ) - emitPntSize = GL_TRUE; - else if (outAttr == VERT_RESULT_BFC0) - emitBFC0 = GL_TRUE; - else if (outAttr == VERT_RESULT_BFC1) - emitBFC1 = GL_TRUE; - } -#if 0 /*debug*/ - printf("assign vp output_to_slot[%d] = %d\n", outAttr, - xvp->output_to_slot[outAttr]); -#endif - } - - /* must do these last */ - if (emitPntSize) { - GLuint slot = numVpOuts++; - xvp->output_to_slot[VERT_RESULT_PSIZ] = slot; - xvp->output_to_semantic_name[VERT_RESULT_PSIZ] = TGSI_SEMANTIC_PSIZE; - xvp->output_to_semantic_index[VERT_RESULT_PSIZ] = 0; - usedOutputSlots |= (1 << slot); - } - if (emitBFC0) { - GLuint slot = numVpOuts++; - xvp->output_to_slot[VERT_RESULT_BFC0] = slot; - xvp->output_to_semantic_name[VERT_RESULT_BFC0] = TGSI_SEMANTIC_COLOR; - xvp->output_to_semantic_index[VERT_RESULT_BFC0] = 0; - usedOutputSlots |= (1 << slot); - } - if (emitBFC1) { - GLuint slot = numVpOuts++; - xvp->output_to_slot[VERT_RESULT_BFC1] = slot; - xvp->output_to_semantic_name[VERT_RESULT_BFC1] = TGSI_SEMANTIC_COLOR; - xvp->output_to_semantic_index[VERT_RESULT_BFC1] = 1; - usedOutputSlots |= (1 << slot); - } - - /* build usedGenerics mask */ - usedGenerics = 0x0; - for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) { - if (xvp->output_to_semantic_name[outAttr] == TGSI_SEMANTIC_GENERIC) { - usedGenerics |= (1 << xvp->output_to_semantic_index[outAttr]); - } - } + st_prepare_vertex_program( st, stvp ); - /* For each vertex program output that doesn't match up to a fragment - * program input, map the vertex program output to a free slot and - * free generic attribute. + /* We are now up-to-date: */ - for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) { - if (outputsWritten & (1 << outAttr)) { - if (xvp->output_to_slot[outAttr] == UNUSED) { - GLint freeGeneric = _mesa_ffs(~usedGenerics) - 1; - GLint freeSlot = _mesa_ffs(~usedOutputSlots) - 1; - usedGenerics |= (1 << freeGeneric); - usedOutputSlots |= (1 << freeSlot); - xvp->output_to_slot[outAttr] = freeSlot; - xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_GENERIC; - xvp->output_to_semantic_index[outAttr] = freeGeneric; - } - } - -#if 0 /*debug*/ - printf("vp output_to_slot[%d] = %d\n", outAttr, - xvp->output_to_slot[outAttr]); -#endif + stvp->lastSerialNo = stvp->serialNo; + } + + /* See if we've got a translated vertex program whose outputs match + * the fragment program's inputs. + */ + for (vpv = stvp->varients; vpv; vpv = vpv->next) { + if (memcmp(&vpv->key, &key, sizeof key) == 0) { + break; } + } - assert(stvp->Base.Base.NumInstructions > 1); - - st_translate_vertex_program(st, stvp, xvp->output_to_slot, - xvp->output_to_semantic_name, - xvp->output_to_semantic_index); - - xvp->vp = stvp; - - /* translated VP is up to date now */ - xvp->serialNo = stvp->serialNo; + /* No? Perform new translation here. */ + if (!vpv) { + vpv = st_translate_vertex_program(st, stvp, &key); + if (!vpv) + return NULL; + + vpv->next = stvp->varients; + stvp->varients = vpv; } - return xvp; + return vpv; } -void -st_free_translated_vertex_programs(struct st_context *st, - struct translated_vertex_program *xvp) -{ - struct translated_vertex_program *next; - - while (xvp) { - next = xvp->next; - _mesa_free(xvp); - xvp = next; - } -} static void * @@ -328,32 +157,19 @@ get_passthrough_fs(struct st_context *st) return st->passthrough_fs; } - static void -update_linkage( struct st_context *st ) +update_fp( struct st_context *st ) { - struct st_vertex_program *stvp; struct st_fragment_program *stfp; - struct translated_vertex_program *xvp; - - /* find active shader and params -- Should be covered by - * ST_NEW_VERTEX_PROGRAM - */ - assert(st->ctx->VertexProgram._Current); - stvp = st_vertex_program(st->ctx->VertexProgram._Current); - assert(stvp->Base.Base.Target == GL_VERTEX_PROGRAM_ARB); assert(st->ctx->FragmentProgram._Current); stfp = st_fragment_program(st->ctx->FragmentProgram._Current); assert(stfp->Base.Base.Target == GL_FRAGMENT_PROGRAM_ARB); - xvp = find_translated_vp(st, stvp, stfp); + translate_fp(st, stfp); - st_reference_vertprog(st, &st->vp, stvp); st_reference_fragprog(st, &st->fp, stfp); - cso_set_vertex_shader_handle(st->cso_context, stvp->driver_shader); - if (st->missing_textures) { /* use a pass-through frag shader that uses no textures */ void *fs = get_passthrough_fs(st); @@ -362,16 +178,48 @@ update_linkage( struct st_context *st ) else { cso_set_fragment_shader_handle(st->cso_context, stfp->driver_shader); } +} + +const struct st_tracked_state st_update_fp = { + "st_update_fp", /* name */ + { /* dirty */ + 0, /* mesa */ + ST_NEW_FRAGMENT_PROGRAM /* st */ + }, + update_fp /* update */ +}; + + + + +static void +update_vp( struct st_context *st ) +{ + struct st_vertex_program *stvp; + + /* find active shader and params -- Should be covered by + * ST_NEW_VERTEX_PROGRAM + */ + assert(st->ctx->VertexProgram._Current); + stvp = st_vertex_program(st->ctx->VertexProgram._Current); + assert(stvp->Base.Base.Target == GL_VERTEX_PROGRAM_ARB); + + st->vp_varient = find_translated_vp(st, stvp); + + st_reference_vertprog(st, &st->vp, stvp); + + cso_set_vertex_shader_handle(st->cso_context, + st->vp_varient->driver_shader); - st->vertex_result_to_slot = xvp->output_to_slot; + st->vertex_result_to_slot = stvp->result_to_output; } -const struct st_tracked_state st_update_shader = { - "st_update_shader", /* name */ +const struct st_tracked_state st_update_vp = { + "st_update_vp", /* name */ { /* dirty */ 0, /* mesa */ - ST_NEW_VERTEX_PROGRAM | ST_NEW_FRAGMENT_PROGRAM /* st */ + ST_NEW_VERTEX_PROGRAM /* st */ }, - update_linkage /* update */ + update_vp /* update */ }; diff --git a/src/mesa/state_tracker/st_cb_bitmap.c b/src/mesa/state_tracker/st_cb_bitmap.c index a22fa68299b..226d1ab14cb 100644 --- a/src/mesa/state_tracker/st_cb_bitmap.c +++ b/src/mesa/state_tracker/st_cb_bitmap.c @@ -169,11 +169,6 @@ make_bitmap_fragment_program(GLcontext *ctx, GLuint samplerIndex) stfp = (struct st_fragment_program *) p; stfp->Base.UsesKill = GL_TRUE; - /* No need to send this incomplete program down to hardware: - * - * st_translate_fragment_program(ctx->st, stfp, NULL); - */ - return stfp; } diff --git a/src/mesa/state_tracker/st_cb_drawpixels.c b/src/mesa/state_tracker/st_cb_drawpixels.c index 13e454b7a7b..60394731f79 100644 --- a/src/mesa/state_tracker/st_cb_drawpixels.c +++ b/src/mesa/state_tracker/st_cb_drawpixels.c @@ -927,7 +927,7 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy, struct pipe_context *pipe = st->pipe; struct pipe_screen *screen = pipe->screen; struct st_renderbuffer *rbRead; - struct st_vertex_program *stvp; + void *driver_vp; struct st_fragment_program *stfp; struct pipe_texture *pt; GLfloat *color; @@ -976,14 +976,14 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy, rbRead = st_get_color_read_renderbuffer(ctx); color = NULL; stfp = combined_drawpix_fragment_program(ctx); - stvp = st_make_passthrough_vertex_shader(st, GL_FALSE); + driver_vp = st_make_passthrough_vertex_shader(st, GL_FALSE); } else { assert(type == GL_DEPTH); rbRead = st_renderbuffer(ctx->ReadBuffer->_DepthBuffer); color = ctx->Current.Attrib[VERT_ATTRIB_COLOR0]; stfp = make_fragment_shader_z(st); - stvp = st_make_passthrough_vertex_shader(st, GL_TRUE); + driver_vp = st_make_passthrough_vertex_shader(st, GL_TRUE); } srcFormat = rbRead->texture->format; @@ -1116,7 +1116,7 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy, draw_textured_quad(ctx, dstx, dsty, ctx->Current.RasterPos[2], width, height, ctx->Pixel.ZoomX, ctx->Pixel.ZoomY, pt, - stvp->driver_shader, + driver_vp, stfp->driver_shader, color, GL_TRUE); diff --git a/src/mesa/state_tracker/st_cb_program.c b/src/mesa/state_tracker/st_cb_program.c index b2d5c39a3a0..8c276f8128e 100644 --- a/src/mesa/state_tracker/st_cb_program.c +++ b/src/mesa/state_tracker/st_cb_program.c @@ -138,24 +138,7 @@ st_delete_program(GLcontext *ctx, struct gl_program *prog) case GL_VERTEX_PROGRAM_ARB: { struct st_vertex_program *stvp = (struct st_vertex_program *) prog; - - if (stvp->driver_shader) { - cso_delete_vertex_shader(st->cso_context, stvp->driver_shader); - stvp->driver_shader = NULL; - } - - if (stvp->draw_shader) { -#if FEATURE_feedback || FEATURE_drawpix - /* this would only have been allocated for the RasterPos path */ - draw_delete_vertex_shader(st->draw, stvp->draw_shader); - stvp->draw_shader = NULL; -#endif - } - - if (stvp->state.tokens) { - st_free_tokens(stvp->state.tokens); - stvp->state.tokens = NULL; - } + st_vp_release_varients( st, stvp ); } break; case GL_FRAGMENT_PROGRAM_ARB: @@ -177,8 +160,6 @@ st_delete_program(GLcontext *ctx, struct gl_program *prog) _mesa_reference_program(ctx, &prg, NULL); stfp->bitmap_program = NULL; } - - st_free_translated_vertex_programs(st, stfp->vertex_programs); } break; default: @@ -219,8 +200,6 @@ static void st_program_string_notify( GLcontext *ctx, stfp->state.tokens = NULL; } - stfp->param_state = stfp->Base.Base.Parameters->StateFlags; - if (st->fp == stfp) st->dirty.st |= ST_NEW_FRAGMENT_PROGRAM; } @@ -229,25 +208,7 @@ static void st_program_string_notify( GLcontext *ctx, stvp->serialNo++; - if (stvp->driver_shader) { - cso_delete_vertex_shader(st->cso_context, stvp->driver_shader); - stvp->driver_shader = NULL; - } - - if (stvp->draw_shader) { -#if FEATURE_feedback || FEATURE_drawpix - /* this would only have been allocated for the RasterPos path */ - draw_delete_vertex_shader(st->draw, stvp->draw_shader); - stvp->draw_shader = NULL; -#endif - } - - if (stvp->state.tokens) { - st_free_tokens(stvp->state.tokens); - stvp->state.tokens = NULL; - } - - stvp->param_state = stvp->Base.Base.Parameters->StateFlags; + st_vp_release_varients( st, stvp ); if (st->vp == stvp) st->dirty.st |= ST_NEW_VERTEX_PROGRAM; diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h index 18adb35e872..b7607286583 100644 --- a/src/mesa/state_tracker/st_context.h +++ b/src/mesa/state_tracker/st_context.h @@ -127,6 +127,8 @@ struct st_context struct st_vertex_program *vp; /**< Currently bound vertex program */ struct st_fragment_program *fp; /**< Currently bound fragment program */ + struct st_vp_varient *vp_varient; + struct gl_texture_object *default_texture; struct { diff --git a/src/mesa/state_tracker/st_debug.c b/src/mesa/state_tracker/st_debug.c index 3009cde9d51..6e699ca5522 100644 --- a/src/mesa/state_tracker/st_debug.c +++ b/src/mesa/state_tracker/st_debug.c @@ -86,7 +86,8 @@ st_print_current(void) } #endif - tgsi_dump( st->vp->state.tokens, 0 ); + if (st->vp->varients) + tgsi_dump( st->vp->varients[0].state.tokens, 0 ); if (st->vp->Base.Base.Parameters) _mesa_print_parameter_list(st->vp->Base.Base.Parameters); diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c index c76bff91819..337c21a6c49 100644 --- a/src/mesa/state_tracker/st_draw.c +++ b/src/mesa/state_tracker/st_draw.c @@ -563,7 +563,7 @@ st_draw_vbo(GLcontext *ctx, /* must get these after state validation! */ vp = ctx->st->vp; - vs = &ctx->st->vp->state; + vs = &ctx->st->vp_varient->state; #if 0 if (MESA_VERBOSE & VERBOSE_GLSL) { diff --git a/src/mesa/state_tracker/st_draw_feedback.c b/src/mesa/state_tracker/st_draw_feedback.c index b2d682ef640..d793f820bc3 100644 --- a/src/mesa/state_tracker/st_draw_feedback.c +++ b/src/mesa/state_tracker/st_draw_feedback.c @@ -120,10 +120,10 @@ st_feedback_draw_vbo(GLcontext *ctx, /* must get these after state validation! */ vp = ctx->st->vp; - vs = &st->vp->state; + vs = &st->vp_varient->state; - if (!st->vp->draw_shader) { - st->vp->draw_shader = draw_create_vertex_shader(draw, vs); + if (!st->vp_varient->draw_shader) { + st->vp_varient->draw_shader = draw_create_vertex_shader(draw, vs); } /* @@ -136,7 +136,7 @@ st_feedback_draw_vbo(GLcontext *ctx, draw_set_viewport_state(draw, &st->state.viewport); draw_set_clip_state(draw, &st->state.clip); draw_set_rasterizer_state(draw, &st->state.rasterizer); - draw_bind_vertex_shader(draw, st->vp->draw_shader); + draw_bind_vertex_shader(draw, st->vp_varient->draw_shader); set_feedback_vertex_format(ctx); /* loop over TGSI shader inputs to determine vertex buffer diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c index 5450b59a2d9..6a2b99cf1b4 100644 --- a/src/mesa/state_tracker/st_program.c +++ b/src/mesa/state_tracker/st_program.c @@ -49,6 +49,36 @@ #include "st_mesa_to_tgsi.h" #include "cso_cache/cso_context.h" + /* Clean out any old compilations: + */ +void +st_vp_release_varients( struct st_context *st, + struct st_vertex_program *stvp ) +{ + struct st_vp_varient *vpv; + + for (vpv = stvp->varients; vpv; ) { + struct st_vp_varient *next = vpv->next; + + if (vpv->driver_shader) + cso_delete_vertex_shader(st->cso_context, vpv->driver_shader); + + if (vpv->draw_shader) + draw_delete_vertex_shader( st->draw, vpv->draw_shader ); + + if (vpv->state.tokens) + st_free_tokens(vpv->state.tokens); + + FREE( vpv ); + + vpv = next; + } + + stvp->varients = NULL; +} + + + /** * Translate a Mesa vertex shader into a TGSI shader. @@ -58,22 +88,13 @@ * \return pointer to cached pipe_shader object. */ void -st_translate_vertex_program(struct st_context *st, - struct st_vertex_program *stvp, - const GLuint outputMapping[], - const ubyte *outputSemanticName, - const ubyte *outputSemanticIndex) +st_prepare_vertex_program(struct st_context *st, + struct st_vertex_program *stvp) { - struct pipe_context *pipe = st->pipe; - GLuint defaultOutputMapping[VERT_RESULT_MAX]; - GLuint attr, i; - GLuint num_generic = 0; - - uint vs_num_inputs = 0; + GLuint attr; - ubyte vs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; - ubyte vs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS]; - uint vs_num_outputs = 0; + stvp->num_inputs = 0; + stvp->num_outputs = 0; if (stvp->Base.IsPositionInvariant) _mesa_insert_mvp_code(st->ctx, &stvp->Base); @@ -84,92 +105,56 @@ st_translate_vertex_program(struct st_context *st, */ for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) { if (stvp->Base.Base.InputsRead & (1 << attr)) { - stvp->input_to_index[attr] = vs_num_inputs; - stvp->index_to_input[vs_num_inputs] = attr; - - vs_num_inputs++; + stvp->input_to_index[attr] = stvp->num_inputs; + stvp->index_to_input[stvp->num_inputs] = attr; + stvp->num_inputs++; } } -#if 0 - if (outputMapping && outputSemanticName) { - printf("VERT_RESULT written out_slot semantic_name semantic_index\n"); - for (attr = 0; attr < VERT_RESULT_MAX; attr++) { - printf(" %-2d %c %3d %2d %2d\n", - attr, - ((stvp->Base.Base.OutputsWritten & (1 << attr)) ? 'Y' : ' '), - outputMapping[attr], - outputSemanticName[attr], - outputSemanticIndex[attr]); - } - } -#endif - - /* initialize output semantics to defaults */ - for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) { - assert(i < Elements(vs_output_semantic_name)); - vs_output_semantic_name[i] = TGSI_SEMANTIC_GENERIC; - vs_output_semantic_index[i] = 0; - } - - num_generic = 0; - /* - * Determine number of outputs, the (default) output register - * mapping and the semantic information for each output. + /* Compute mapping of vertex program outputs to slots. */ for (attr = 0; attr < VERT_RESULT_MAX; attr++) { - if (stvp->Base.Base.OutputsWritten & (1 << attr)) { - GLuint slot; - - /* XXX - * Pass in the fragment program's input's semantic info. - * Use the generic semantic indexes from there, instead of - * guessing below. - */ - - if (outputMapping) { - slot = outputMapping[attr]; - assert(slot != ~0); - } - else { - slot = vs_num_outputs; - vs_num_outputs++; - defaultOutputMapping[attr] = slot; - } + if ((stvp->Base.Base.OutputsWritten & (1 << attr)) == 0) { + stvp->result_to_output[attr] = ~0; + } + else { + unsigned slot = stvp->num_outputs++; + + stvp->result_to_output[attr] = slot; switch (attr) { case VERT_RESULT_HPOS: - assert(slot == 0); - vs_output_semantic_name[slot] = TGSI_SEMANTIC_POSITION; - vs_output_semantic_index[slot] = 0; + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_POSITION; + stvp->output_semantic_index[slot] = 0; break; case VERT_RESULT_COL0: - vs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR; - vs_output_semantic_index[slot] = 0; + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_COLOR; + stvp->output_semantic_index[slot] = 0; break; case VERT_RESULT_COL1: - vs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR; - vs_output_semantic_index[slot] = 1; + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_COLOR; + stvp->output_semantic_index[slot] = 1; break; case VERT_RESULT_BFC0: - vs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR; - vs_output_semantic_index[slot] = 0; + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR; + stvp->output_semantic_index[slot] = 0; break; case VERT_RESULT_BFC1: - vs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR; - vs_output_semantic_index[slot] = 1; + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR; + stvp->output_semantic_index[slot] = 1; break; case VERT_RESULT_FOGC: - vs_output_semantic_name[slot] = TGSI_SEMANTIC_FOG; - vs_output_semantic_index[slot] = 0; + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_FOG; + stvp->output_semantic_index[slot] = 0; break; case VERT_RESULT_PSIZ: - vs_output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE; - vs_output_semantic_index[slot] = 0; + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE; + stvp->output_semantic_index[slot] = 0; break; case VERT_RESULT_EDGE: assert(0); break; + case VERT_RESULT_TEX0: case VERT_RESULT_TEX1: case VERT_RESULT_TEX2: @@ -178,87 +163,50 @@ st_translate_vertex_program(struct st_context *st, case VERT_RESULT_TEX5: case VERT_RESULT_TEX6: case VERT_RESULT_TEX7: - /* fall-through */ + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; + stvp->output_semantic_index[slot] = attr - VERT_RESULT_TEX0; + break; + case VERT_RESULT_VAR0: - /* fall-through */ default: - assert(slot < Elements(vs_output_semantic_name)); - if (outputSemanticName) { - /* use provided semantic into */ - assert(outputSemanticName[attr] != TGSI_SEMANTIC_COUNT); - vs_output_semantic_name[slot] = outputSemanticName[attr]; - vs_output_semantic_index[slot] = outputSemanticIndex[attr]; - } - else { - /* use default semantic info */ - vs_output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; - vs_output_semantic_index[slot] = num_generic++; - } + assert(attr < VERT_RESULT_MAX); + stvp->output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; + stvp->output_semantic_index[slot] = (FRAG_ATTRIB_VAR0 - + FRAG_ATTRIB_TEX0 + + attr - + VERT_RESULT_VAR0); + break; } } } +} - if (outputMapping) { - /* find max output slot referenced to compute vs_num_outputs */ - GLuint maxSlot = 0; - for (attr = 0; attr < VERT_RESULT_MAX; attr++) { - if (outputMapping[attr] != ~0 && outputMapping[attr] > maxSlot) - maxSlot = outputMapping[attr]; - } - vs_num_outputs = maxSlot + 1; - } - else { - outputMapping = defaultOutputMapping; - } -#if 0 /* debug */ - { - GLuint i; - printf("outputMapping? %d\n", outputMapping ? 1 : 0); - if (outputMapping) { - printf("attr -> slot\n"); - for (i = 0; i < 16; i++) { - printf(" %2d %3d\n", i, outputMapping[i]); - } - } - printf("slot sem_name sem_index\n"); - for (i = 0; i < vs_num_outputs; i++) { - printf(" %2d %d %d\n", - i, - vs_output_semantic_name[i], - vs_output_semantic_index[i]); - } - } -#endif - - /* free old shader state, if any */ - if (stvp->state.tokens) { - st_free_tokens(stvp->state.tokens); - stvp->state.tokens = NULL; - } - if (stvp->driver_shader) { - cso_delete_vertex_shader(st->cso_context, stvp->driver_shader); - stvp->driver_shader = NULL; - } +struct st_vp_varient * +st_translate_vertex_program(struct st_context *st, + struct st_vertex_program *stvp, + const struct st_vp_varient_key *key) +{ + struct st_vp_varient *vpv = CALLOC_STRUCT(st_vp_varient); + struct pipe_context *pipe = st->pipe; - stvp->state.tokens = + vpv->state.tokens = st_translate_mesa_program(st->ctx, TGSI_PROCESSOR_VERTEX, &stvp->Base.Base, /* inputs */ - vs_num_inputs, + stvp->num_inputs, stvp->input_to_index, NULL, /* input semantic name */ NULL, /* input semantic index */ NULL, /* outputs */ - vs_num_outputs, - outputMapping, - vs_output_semantic_name, - vs_output_semantic_index ); + stvp->num_outputs, + stvp->result_to_output, + stvp->output_semantic_name, + stvp->output_semantic_index ); - stvp->num_inputs = vs_num_inputs; - stvp->driver_shader = pipe->create_vs_state(pipe, &stvp->state); + vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->state); if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) { _mesa_print_program(&stvp->Base.Base); @@ -266,9 +214,11 @@ st_translate_vertex_program(struct st_context *st, } if (ST_DEBUG & DEBUG_TGSI) { - tgsi_dump( stvp->state.tokens, 0 ); + tgsi_dump( vpv->state.tokens, 0 ); debug_printf("\n"); } + + return vpv; } @@ -291,7 +241,6 @@ st_translate_fragment_program(struct st_context *st, GLuint attr; const GLbitfield inputsRead = stfp->Base.Base.InputsRead; GLuint vslot = 0; - GLuint num_generic = 0; uint fs_num_inputs = 0; @@ -341,14 +290,25 @@ st_translate_fragment_program(struct st_context *st, break; case FRAG_ATTRIB_FACE: stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FACE; - stfp->input_semantic_index[slot] = num_generic++; + stfp->input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_CONSTANT; break; - case FRAG_ATTRIB_PNTC: - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; - stfp->input_semantic_index[slot] = num_generic++; - 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 FRAG_ATTRIB_ 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 FRAG_ATTRIB_TEX0: case FRAG_ATTRIB_TEX1: case FRAG_ATTRIB_TEX2: @@ -357,16 +317,17 @@ st_translate_fragment_program(struct st_context *st, case FRAG_ATTRIB_TEX5: case FRAG_ATTRIB_TEX6: case FRAG_ATTRIB_TEX7: - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; - stfp->input_semantic_index[slot] = num_generic++; - interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; - break; + case FRAG_ATTRIB_PNTC: case FRAG_ATTRIB_VAR0: - /* fall-through */ default: + /* Actually, let's try and zero-base this just for + * readability of the generated TGSI. + */ + assert(attr >= FRAG_ATTRIB_TEX0); + stfp->input_semantic_index[slot] = (attr - FRAG_ATTRIB_TEX0); stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; - stfp->input_semantic_index[slot] = num_generic++; interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; + break; } } } diff --git a/src/mesa/state_tracker/st_program.h b/src/mesa/state_tracker/st_program.h index e2e5eddef22..88aadbd7510 100644 --- a/src/mesa/state_tracker/st_program.h +++ b/src/mesa/state_tracker/st_program.h @@ -64,41 +64,70 @@ struct st_fragment_program struct pipe_shader_state state; void *driver_shader; - GLuint param_state; - - /** List of vertex programs which have been translated such that their - * outputs match this fragment program's inputs. - */ - struct translated_vertex_program *vertex_programs; - /** Program prefixed with glBitmap prologue */ struct st_fragment_program *bitmap_program; uint bitmap_sampler; }; + +struct st_vp_varient_key +{ + char dummy; /* currently unused */ +}; + + +/** + * This represents a vertex program, especially translated to match + * the inputs of a particular fragment shader. + */ +struct st_vp_varient +{ + /* Parameters which generated this translated version of a vertex + * shader: + */ + struct st_vp_varient_key key; + + /** TGSI tokens -- why? + */ + struct pipe_shader_state state; + + /** Driver's compiled shader */ + void *driver_shader; + + /** For using our private draw module (glRasterPos) */ + struct draw_vertex_shader *draw_shader; + + /** Next in linked list */ + struct st_vp_varient *next; +}; + + + + /** * Derived from Mesa gl_fragment_program: */ struct st_vertex_program { struct gl_vertex_program Base; /**< The Mesa vertex program */ - GLuint serialNo; + GLuint serialNo, lastSerialNo; /** maps a Mesa VERT_ATTRIB_x to a packed TGSI input index */ GLuint input_to_index[VERT_ATTRIB_MAX]; /** maps a TGSI input index back to a Mesa VERT_ATTRIB_x */ GLuint index_to_input[PIPE_MAX_SHADER_INPUTS]; - GLuint num_inputs; - struct pipe_shader_state state; - void *driver_shader; + /** Maps VERT_RESULT_x to slot */ + GLuint result_to_output[VERT_RESULT_MAX]; + ubyte output_semantic_name[VERT_RESULT_MAX]; + ubyte output_semantic_index[VERT_RESULT_MAX]; + GLuint num_outputs; - /** For using our private draw module (glRasterPos) */ - struct draw_vertex_shader *draw_shader; - - GLuint param_state; + /** List of translated varients of this vertex program. + */ + struct st_vp_varient *varients; }; @@ -143,13 +172,21 @@ st_translate_fragment_program(struct st_context *st, const GLuint inputMapping[]); +/* Called after program string change, discard all previous + * compilation results. + */ extern void +st_prepare_vertex_program(struct st_context *st, + struct st_vertex_program *stvp); + +extern struct st_vp_varient * st_translate_vertex_program(struct st_context *st, - struct st_vertex_program *vp, - const GLuint vert_output_to_slot[], - const ubyte *fs_input_semantic_name, - const ubyte *fs_input_semantic_index); + struct st_vertex_program *stvp, + const struct st_vp_varient_key *key); +void +st_vp_release_varients( struct st_context *st, + struct st_vertex_program *stvp ); extern void st_print_shaders(GLcontext *ctx); -- 2.30.2