X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fauxiliary%2Fdraw%2Fdraw_pt_fetch_shade_pipeline_llvm.c;h=a53a768d02990e4d4d4b6ac787f39ae2478dd913;hb=1d6f3543a063ab9e740fd0c149dcce26c282d773;hp=f71271bd915645e876914a21f5d9bef6d97e5d51;hpb=ccd13da0fc1f1813b55fc0d2181a6cb0d3b42b0d;p=mesa.git diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c index f71271bd915..a53a768d029 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c @@ -1,6 +1,6 @@ /************************************************************************** * - * Copyright 2010 VMWare, Inc. + * Copyright 2010 VMware, Inc. * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a @@ -28,11 +28,11 @@ #include "util/u_math.h" #include "util/u_memory.h" #include "draw/draw_context.h" +#include "draw/draw_gs.h" #include "draw/draw_vbuf.h" #include "draw/draw_vertex.h" #include "draw/draw_pt.h" #include "draw/draw_vs.h" -#include "draw/draw_gs.h" #include "draw/draw_llvm.h" @@ -41,53 +41,59 @@ struct llvm_middle_end { struct draw_context *draw; struct pt_emit *emit; + struct pt_so_emit *so_emit; struct pt_fetch *fetch; struct pt_post_vs *post_vs; unsigned vertex_data_offset; unsigned vertex_size; - unsigned prim; + unsigned input_prim; unsigned opt; struct draw_llvm *llvm; - struct draw_llvm_variant *variants; struct draw_llvm_variant *current_variant; - int nr_variants; }; static void llvm_middle_end_prepare( struct draw_pt_middle_end *middle, - unsigned prim, + unsigned in_prim, unsigned opt, unsigned *max_vertices ) { struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle; struct draw_context *draw = fpme->draw; - struct draw_vertex_shader *vs = draw->vs.vertex_shader; - struct draw_geometry_shader *gs = draw->gs.geometry_shader; - struct draw_llvm_variant_key key; + struct llvm_vertex_shader *shader = + llvm_vertex_shader(draw->vs.vertex_shader); + char store[DRAW_LLVM_MAX_VARIANT_KEY_SIZE]; + struct draw_llvm_variant_key *key; struct draw_llvm_variant *variant = NULL; + struct draw_llvm_variant_list_item *li; unsigned i; unsigned instance_id_index = ~0; + + unsigned out_prim = (draw->gs.geometry_shader ? + draw->gs.geometry_shader->output_primitive : + in_prim); + /* Add one to num_outputs because the pipeline occasionally tags on * an additional texcoord, eg for AA lines. */ - unsigned nr = MAX2( vs->info.num_inputs, - vs->info.num_outputs + 1 ); + unsigned nr = MAX2( shader->base.info.num_inputs, + shader->base.info.num_outputs + 1 ); /* Scan for instanceID system value. */ - for (i = 0; i < vs->info.num_inputs; i++) { - if (vs->info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) { + for (i = 0; i < shader->base.info.num_inputs; i++) { + if (shader->base.info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) { instance_id_index = i; break; } } - fpme->prim = prim; + fpme->input_prim = in_prim; fpme->opt = opt; /* Always leave room for the vertex header whether we need it or @@ -97,57 +103,71 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle, fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float); - - draw_pt_fetch_prepare( fpme->fetch, - vs->info.num_inputs, - fpme->vertex_size, - instance_id_index ); - if (opt & PT_SHADE) { - vs->prepare(vs, draw); - draw_geometry_shader_prepare(gs, draw); - } - - /* XXX: it's not really gl rasterization rules we care about here, * but gl vs dx9 clip spaces. */ draw_pt_post_vs_prepare( fpme->post_vs, - (boolean)draw->bypass_clipping, - (boolean)(draw->identity_viewport), + draw->clip_xy, + draw->clip_z, + draw->clip_user, + draw->identity_viewport, (boolean)draw->rasterizer->gl_rasterization_rules, - (draw->vs.edgeflag_output ? true : false) ); + (draw->vs.edgeflag_output ? TRUE : FALSE) ); + + draw_pt_so_emit_prepare( fpme->so_emit ); if (!(opt & PT_PIPELINE)) { draw_pt_emit_prepare( fpme->emit, - prim, + out_prim, max_vertices ); - *max_vertices = MAX2( *max_vertices, - DRAW_PIPE_MAX_VERTICES ); + *max_vertices = MAX2( *max_vertices, 4096 ); } else { - *max_vertices = DRAW_PIPE_MAX_VERTICES; + /* limit max fetches by limiting max_vertices */ + *max_vertices = 4096; } /* return even number */ *max_vertices = *max_vertices & ~1; + + key = draw_llvm_make_variant_key(fpme->llvm, store); - draw_llvm_make_variant_key(fpme->llvm, &key); - - variant = fpme->variants; - while(variant) { - if(memcmp(&variant->key, &key, sizeof key) == 0) + li = first_elem(&shader->variants); + while(!at_end(&shader->variants, li)) { + if(memcmp(&li->base->key, key, shader->variant_key_size) == 0) { + variant = li->base; break; + } + li = next_elem(li); + } - variant = variant->next; + if (variant) { + move_to_head(&fpme->llvm->vs_variants_list, &variant->list_item_global); } + else { + unsigned i; + if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) { + /* + * XXX: should we flush here ? + */ + for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) { + struct draw_llvm_variant_list_item *item = + last_elem(&fpme->llvm->vs_variants_list); + draw_llvm_destroy_variant(item->base); + } + } + + variant = draw_llvm_create_variant(fpme->llvm, nr, key); - if (!variant) { - variant = draw_llvm_prepare(fpme->llvm, nr); - variant->next = fpme->variants; - fpme->variants = variant; - ++fpme->nr_variants; + if (variant) { + insert_at_head(&shader->variants, &variant->list_item_local); + insert_at_head(&fpme->llvm->vs_variants_list, &variant->list_item_global); + fpme->llvm->nr_variants++; + shader->variants_cached++; + } } + fpme->current_variant = variant; /*XXX we only support one constant buffer */ @@ -155,152 +175,186 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle, draw->pt.user.vs_constants[0]; fpme->llvm->jit_context.gs_constants = draw->pt.user.gs_constants[0]; + fpme->llvm->jit_context.planes = + (float (*) [12][4]) draw->pt.user.planes[0]; + fpme->llvm->jit_context.viewport = + (float *)draw->viewport.scale; + } +static void pipeline(struct llvm_middle_end *llvm, + const struct draw_vertex_info *vert_info, + const struct draw_prim_info *prim_info) +{ + if (prim_info->linear) + draw_pipeline_run_linear( llvm->draw, + vert_info, + prim_info); + else + draw_pipeline_run( llvm->draw, + vert_info, + prim_info ); +} -static void llvm_middle_end_run( struct draw_pt_middle_end *middle, - const unsigned *fetch_elts, - unsigned fetch_count, - const ushort *draw_elts, - unsigned draw_count ) +static void emit(struct pt_emit *emit, + const struct draw_vertex_info *vert_info, + const struct draw_prim_info *prim_info) +{ + if (prim_info->linear) { + draw_pt_emit_linear(emit, vert_info, prim_info); + } + else { + draw_pt_emit(emit, vert_info, prim_info); + } +} + +static void +llvm_pipeline_generic( struct draw_pt_middle_end *middle, + const struct draw_fetch_info *fetch_info, + const struct draw_prim_info *prim_info ) { struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle; struct draw_context *draw = fpme->draw; - struct draw_vertex_shader *vshader = draw->vs.vertex_shader; struct draw_geometry_shader *gshader = draw->gs.geometry_shader; + struct draw_prim_info gs_prim_info; + struct draw_vertex_info llvm_vert_info; + struct draw_vertex_info gs_vert_info; + struct draw_vertex_info *vert_info; unsigned opt = fpme->opt; - unsigned alloc_count = align( fetch_count, 4 ); - - struct vertex_header *pipeline_verts = - (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count); - - if (!pipeline_verts) { - /* Not much we can do here - just skip the rendering. - */ + unsigned clipped = 0; + + llvm_vert_info.count = fetch_info->count; + llvm_vert_info.vertex_size = fpme->vertex_size; + llvm_vert_info.stride = fpme->vertex_size; + llvm_vert_info.verts = + (struct vertex_header *)MALLOC(fpme->vertex_size * + align(fetch_info->count, 4)); + if (!llvm_vert_info.verts) { assert(0); return; } - /* Fetch into our vertex buffer + if (fetch_info->linear) + clipped = fpme->current_variant->jit_func( &fpme->llvm->jit_context, + llvm_vert_info.verts, + (const char **)draw->pt.user.vbuffer, + fetch_info->start, + fetch_info->count, + fpme->vertex_size, + draw->pt.vertex_buffer, + draw->instance_id); + else + clipped = fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context, + llvm_vert_info.verts, + (const char **)draw->pt.user.vbuffer, + fetch_info->elts, + fetch_info->count, + fpme->vertex_size, + draw->pt.vertex_buffer, + draw->instance_id); + + /* Finished with fetch and vs: */ - draw_pt_fetch_run( fpme->fetch, - fetch_elts, - fetch_count, - (char *)pipeline_verts ); - - /* Run the shader, note that this overwrites the data[] parts of - * the pipeline verts. If there is no shader, eg if - * bypass_vs_clip_and_viewport, then the inputs == outputs, and are - * already in the correct place.*/ - if (opt & PT_SHADE) - { - vshader->run_linear(vshader, - (const float (*)[4])pipeline_verts->data, - ( float (*)[4])pipeline_verts->data, - draw->pt.user.vs_constants, - fetch_count, - fpme->vertex_size, - fpme->vertex_size); - if (gshader) - draw_geometry_shader_run(gshader, - (const float (*)[4])pipeline_verts->data, - ( float (*)[4])pipeline_verts->data, - draw->pt.user.gs_constants, - fetch_count, - fpme->vertex_size, - fpme->vertex_size); + fetch_info = NULL; + vert_info = &llvm_vert_info; + + + if ((opt & PT_SHADE) && gshader) { + draw_geometry_shader_run(gshader, + draw->pt.user.gs_constants, + draw->pt.user.gs_constants_size, + vert_info, + prim_info, + &gs_vert_info, + &gs_prim_info); + + FREE(vert_info->verts); + vert_info = &gs_vert_info; + prim_info = &gs_prim_info; + + clipped = draw_pt_post_vs_run( fpme->post_vs, vert_info ); + } - if (draw_pt_post_vs_run( fpme->post_vs, - pipeline_verts, - fetch_count, - fpme->vertex_size )) - { + /* stream output needs to be done before clipping */ + draw_pt_so_emit( fpme->so_emit, + vert_info, + prim_info ); + + if (clipped) { opt |= PT_PIPELINE; } - /* Do we need to run the pipeline? + /* Do we need to run the pipeline? Now will come here if clipped */ if (opt & PT_PIPELINE) { - draw_pipeline_run( fpme->draw, - fpme->prim, - pipeline_verts, - fetch_count, - fpme->vertex_size, - draw_elts, - draw_count ); + pipeline( fpme, + vert_info, + prim_info ); } else { - draw_pt_emit( fpme->emit, - (const float (*)[4])pipeline_verts->data, - fetch_count, - fpme->vertex_size, - draw_elts, - draw_count ); + emit( fpme->emit, + vert_info, + prim_info ); } + FREE(vert_info->verts); +} - FREE(pipeline_verts); +static void llvm_middle_end_run( struct draw_pt_middle_end *middle, + const unsigned *fetch_elts, + unsigned fetch_count, + const ushort *draw_elts, + unsigned draw_count, + unsigned prim_flags ) +{ + struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle; + struct draw_fetch_info fetch_info; + struct draw_prim_info prim_info; + + fetch_info.linear = FALSE; + fetch_info.start = 0; + fetch_info.elts = fetch_elts; + fetch_info.count = fetch_count; + + prim_info.linear = FALSE; + prim_info.start = 0; + prim_info.count = draw_count; + prim_info.elts = draw_elts; + prim_info.prim = fpme->input_prim; + prim_info.flags = prim_flags; + prim_info.primitive_count = 1; + prim_info.primitive_lengths = &draw_count; + + llvm_pipeline_generic( middle, &fetch_info, &prim_info ); } static void llvm_middle_end_linear_run( struct draw_pt_middle_end *middle, unsigned start, - unsigned count) + unsigned count, + unsigned prim_flags) { struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle; - struct draw_context *draw = fpme->draw; - unsigned opt = fpme->opt; - unsigned alloc_count = align( count, 4 ); - - struct vertex_header *pipeline_verts = - (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count); - - if (!pipeline_verts) { - /* Not much we can do here - just skip the rendering. - */ - assert(0); - return; - } - -#if 0 - debug_printf("#### Pipeline = %p (data = %p)\n", - pipeline_verts, pipeline_verts->data); -#endif - fpme->current_variant->jit_func( &fpme->llvm->jit_context, - pipeline_verts, - (const char **)draw->pt.user.vbuffer, - start, - count, - fpme->vertex_size, - draw->pt.vertex_buffer ); - - if (draw_pt_post_vs_run( fpme->post_vs, - pipeline_verts, - count, - fpme->vertex_size )) - { - opt |= PT_PIPELINE; - } - - /* Do we need to run the pipeline? - */ - if (opt & PT_PIPELINE) { - draw_pipeline_run_linear( fpme->draw, - fpme->prim, - pipeline_verts, - count, - fpme->vertex_size); - } - else { - draw_pt_emit_linear( fpme->emit, - (const float (*)[4])pipeline_verts->data, - fpme->vertex_size, - count ); - } - - FREE(pipeline_verts); + struct draw_fetch_info fetch_info; + struct draw_prim_info prim_info; + + fetch_info.linear = TRUE; + fetch_info.start = start; + fetch_info.count = count; + fetch_info.elts = NULL; + + prim_info.linear = TRUE; + prim_info.start = 0; + prim_info.count = count; + prim_info.elts = NULL; + prim_info.prim = fpme->input_prim; + prim_info.flags = prim_flags; + prim_info.primitive_count = 1; + prim_info.primitive_lengths = &count; + + llvm_pipeline_generic( middle, &fetch_info, &prim_info ); } @@ -310,56 +364,29 @@ llvm_middle_end_linear_run_elts( struct draw_pt_middle_end *middle, unsigned start, unsigned count, const ushort *draw_elts, - unsigned draw_count ) + unsigned draw_count, + unsigned prim_flags ) { struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle; - struct draw_context *draw = fpme->draw; - unsigned opt = fpme->opt; - unsigned alloc_count = align( count, 4 ); - - struct vertex_header *pipeline_verts = - (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count); - - if (!pipeline_verts) - return FALSE; - - fpme->current_variant->jit_func( &fpme->llvm->jit_context, - pipeline_verts, - (const char **)draw->pt.user.vbuffer, - start, - count, - fpme->vertex_size, - draw->pt.vertex_buffer ); - - if (draw_pt_post_vs_run( fpme->post_vs, - pipeline_verts, - count, - fpme->vertex_size )) - { - opt |= PT_PIPELINE; - } + struct draw_fetch_info fetch_info; + struct draw_prim_info prim_info; - /* Do we need to run the pipeline? - */ - if (opt & PT_PIPELINE) { - draw_pipeline_run( fpme->draw, - fpme->prim, - pipeline_verts, - count, - fpme->vertex_size, - draw_elts, - draw_count ); - } - else { - draw_pt_emit( fpme->emit, - (const float (*)[4])pipeline_verts->data, - count, - fpme->vertex_size, - draw_elts, - draw_count ); - } + fetch_info.linear = TRUE; + fetch_info.start = start; + fetch_info.count = count; + fetch_info.elts = NULL; + + prim_info.linear = FALSE; + prim_info.start = 0; + prim_info.count = draw_count; + prim_info.elts = draw_elts; + prim_info.prim = fpme->input_prim; + prim_info.flags = prim_flags; + prim_info.primitive_count = 1; + prim_info.primitive_lengths = &draw_count; + + llvm_pipeline_generic( middle, &fetch_info, &prim_info ); - FREE(pipeline_verts); return TRUE; } @@ -380,17 +407,18 @@ static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle ) if (fpme->emit) draw_pt_emit_destroy( fpme->emit ); + if (fpme->so_emit) + draw_pt_so_emit_destroy( fpme->so_emit ); + if (fpme->post_vs) draw_pt_post_vs_destroy( fpme->post_vs ); - if (fpme->llvm) - draw_llvm_destroy( fpme->llvm ); - FREE(middle); } -struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit_llvm( struct draw_context *draw ) +struct draw_pt_middle_end * +draw_pt_fetch_pipeline_or_emit_llvm(struct draw_context *draw) { struct llvm_middle_end *fpme = 0; @@ -422,13 +450,15 @@ struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit_llvm( struct draw_cont if (!fpme->emit) goto fail; - fpme->llvm = draw_llvm_create(draw); + fpme->so_emit = draw_pt_so_emit_create( draw ); + if (!fpme->so_emit) + goto fail; + + fpme->llvm = draw->llvm; if (!fpme->llvm) goto fail; - fpme->variants = NULL; fpme->current_variant = NULL; - fpme->nr_variants = 0; return &fpme->base;