diff options
Diffstat (limited to 'src/gallium/auxiliary/draw')
25 files changed, 1057 insertions, 285 deletions
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c index dab95e5051..c127f74188 100644 --- a/src/gallium/auxiliary/draw/draw_context.c +++ b/src/gallium/auxiliary/draw/draw_context.c @@ -40,6 +40,7 @@ #if HAVE_LLVM #include "gallivm/lp_bld_init.h" +#include "draw_llvm.h" #endif struct draw_context *draw_create( struct pipe_context *pipe ) @@ -52,6 +53,7 @@ struct draw_context *draw_create( struct pipe_context *pipe ) lp_build_init(); assert(lp_build_engine); draw->engine = lp_build_engine; + draw->llvm = draw_llvm_create(draw); #endif if (!draw_init(draw)) @@ -132,6 +134,9 @@ void draw_destroy( struct draw_context *draw ) draw_pt_destroy( draw ); draw_vs_destroy( draw ); draw_gs_destroy( draw ); +#ifdef HAVE_LLVM + draw_llvm_destroy( draw->llvm ); +#endif FREE( draw ); } @@ -211,6 +216,7 @@ void draw_set_clip_state( struct draw_context *draw, assert(clip->nr <= PIPE_MAX_CLIP_PLANES); memcpy(&draw->plane[6], clip->ucp, clip->nr * sizeof(clip->ucp[0])); draw->nr_planes = 6 + clip->nr; + draw->depth_clamp = clip->depth_clamp; } @@ -601,3 +607,54 @@ draw_set_so_state(struct draw_context *draw, state, sizeof(struct pipe_stream_output_state)); } + +void +draw_set_sampler_views(struct draw_context *draw, + struct pipe_sampler_view **views, + unsigned num) +{ + unsigned i; + + debug_assert(num <= PIPE_MAX_VERTEX_SAMPLERS); + + for (i = 0; i < num; ++i) + draw->sampler_views[i] = views[i]; + for (i = num; i < PIPE_MAX_VERTEX_SAMPLERS; ++i) + draw->sampler_views[i] = NULL; + + draw->num_sampler_views = num; +} + +void +draw_set_samplers(struct draw_context *draw, + struct pipe_sampler_state **samplers, + unsigned num) +{ + unsigned i; + + debug_assert(num <= PIPE_MAX_VERTEX_SAMPLERS); + + for (i = 0; i < num; ++i) + draw->samplers[i] = samplers[i]; + for (i = num; i < PIPE_MAX_VERTEX_SAMPLERS; ++i) + draw->samplers[i] = NULL; + + draw->num_samplers = num; +} + +void +draw_set_mapped_texture(struct draw_context *draw, + unsigned sampler_idx, + uint32_t width, uint32_t height, uint32_t depth, + uint32_t last_level, + uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS], + uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS], + const void *data[DRAW_MAX_TEXTURE_LEVELS]) +{ +#ifdef HAVE_LLVM + draw_llvm_set_mapped_texture(draw, + sampler_idx, + width, height, depth, last_level, + row_stride, img_stride, data); +#endif +} diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h index c0122f2aca..116716af6f 100644 --- a/src/gallium/auxiliary/draw/draw_context.h +++ b/src/gallium/auxiliary/draw/draw_context.h @@ -47,11 +47,14 @@ struct draw_vertex_shader; struct draw_geometry_shader; struct tgsi_sampler; +#define DRAW_MAX_TEXTURE_LEVELS 13 /* 4K x 4K for now */ struct draw_context *draw_create( struct pipe_context *pipe ); void draw_destroy( struct draw_context *draw ); +void draw_flush(struct draw_context *draw); + void draw_set_viewport_state( struct draw_context *draw, const struct pipe_viewport_state *viewport ); @@ -101,6 +104,23 @@ draw_texture_samplers(struct draw_context *draw, uint num_samplers, struct tgsi_sampler **samplers); +void +draw_set_sampler_views(struct draw_context *draw, + struct pipe_sampler_view **views, + unsigned num); +void +draw_set_samplers(struct draw_context *draw, + struct pipe_sampler_state **samplers, + unsigned num); + +void +draw_set_mapped_texture(struct draw_context *draw, + unsigned sampler_idx, + uint32_t width, uint32_t height, uint32_t depth, + uint32_t last_level, + uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS], + uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS], + const void *data[DRAW_MAX_TEXTURE_LEVELS]); /* @@ -173,7 +193,7 @@ draw_set_so_state(struct draw_context *draw, /*********************************************************************** - * draw_prim.c + * draw_pt.c */ void draw_arrays(struct draw_context *draw, unsigned prim, @@ -187,8 +207,6 @@ draw_arrays_instanced(struct draw_context *draw, unsigned startInstance, unsigned instanceCount); -void draw_flush(struct draw_context *draw); - /******************************************************************************* * Driver backend interface diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c index 9117c1303d..19f96c37ab 100644 --- a/src/gallium/auxiliary/draw/draw_llvm.c +++ b/src/gallium/auxiliary/draw/draw_llvm.c @@ -1,3 +1,30 @@ +/************************************************************************** + * + * Copyright 2010 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + #include "draw_llvm.h" #include "draw_context.h" @@ -15,14 +42,13 @@ #include "tgsi/tgsi_dump.h" #include "util/u_cpu_detect.h" -#include "util/u_string.h" #include "util/u_pointer.h" +#include "util/u_string.h" #include <llvm-c/Transforms/Scalar.h> #define DEBUG_STORE 0 - /* generates the draw jit function */ static void draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *var); @@ -36,12 +62,19 @@ init_globals(struct draw_llvm *llvm) /* struct draw_jit_texture */ { - LLVMTypeRef elem_types[4]; + LLVMTypeRef elem_types[DRAW_JIT_TEXTURE_NUM_FIELDS]; elem_types[DRAW_JIT_TEXTURE_WIDTH] = LLVMInt32Type(); elem_types[DRAW_JIT_TEXTURE_HEIGHT] = LLVMInt32Type(); - elem_types[DRAW_JIT_TEXTURE_STRIDE] = LLVMInt32Type(); - elem_types[DRAW_JIT_TEXTURE_DATA] = LLVMPointerType(LLVMInt8Type(), 0); + elem_types[DRAW_JIT_TEXTURE_DEPTH] = LLVMInt32Type(); + elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type(); + elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] = + LLVMArrayType(LLVMInt32Type(), DRAW_MAX_TEXTURE_LEVELS); + elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] = + LLVMArrayType(LLVMInt32Type(), DRAW_MAX_TEXTURE_LEVELS); + elem_types[DRAW_JIT_TEXTURE_DATA] = + LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0), + DRAW_MAX_TEXTURE_LEVELS); texture_type = LLVMStructType(elem_types, Elements(elem_types), 0); @@ -51,9 +84,18 @@ init_globals(struct draw_llvm *llvm) LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, height, llvm->target, texture_type, DRAW_JIT_TEXTURE_HEIGHT); - LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, stride, + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, depth, + llvm->target, texture_type, + DRAW_JIT_TEXTURE_DEPTH); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, last_level, + llvm->target, texture_type, + DRAW_JIT_TEXTURE_LAST_LEVEL); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, row_stride, llvm->target, texture_type, - DRAW_JIT_TEXTURE_STRIDE); + DRAW_JIT_TEXTURE_ROW_STRIDE); + LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, img_stride, + llvm->target, texture_type, + DRAW_JIT_TEXTURE_IMG_STRIDE); LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, data, llvm->target, texture_type, DRAW_JIT_TEXTURE_DATA); @@ -71,7 +113,8 @@ init_globals(struct draw_llvm *llvm) elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */ elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */ - elem_types[2] = LLVMArrayType(texture_type, PIPE_MAX_SAMPLERS); /* textures */ + elem_types[2] = LLVMArrayType(texture_type, + PIPE_MAX_VERTEX_SAMPLERS); /* textures */ context_type = LLVMStructType(elem_types, Elements(elem_types), 0); @@ -81,7 +124,7 @@ init_globals(struct draw_llvm *llvm) llvm->target, context_type, 1); LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures, llvm->target, context_type, - DRAW_JIT_CONTEXT_TEXTURES_INDEX); + DRAW_JIT_CTX_TEXTURES); LP_CHECK_STRUCT_SIZE(struct draw_jit_context, llvm->target, context_type); @@ -195,9 +238,22 @@ draw_llvm_create(struct draw_context *draw) /* These are the passes currently listed in llvm-c/Transforms/Scalar.h, * but there are more on SVN. */ /* TODO: Add more passes */ + LLVMAddCFGSimplificationPass(llvm->pass); - LLVMAddPromoteMemoryToRegisterPass(llvm->pass); - LLVMAddConstantPropagationPass(llvm->pass); + + if (HAVE_LLVM >= 0x207 && sizeof(void*) == 4) { + /* For LLVM >= 2.7 and 32-bit build, use this order of passes to + * avoid generating bad code. + * Test with piglit glsl-vs-sqrt-zero test. + */ + LLVMAddConstantPropagationPass(llvm->pass); + LLVMAddPromoteMemoryToRegisterPass(llvm->pass); + } + else { + LLVMAddPromoteMemoryToRegisterPass(llvm->pass); + LLVMAddConstantPropagationPass(llvm->pass); + } + if(util_cpu_caps.has_sse4_1) { /* FIXME: There is a bug in this pass, whereby the combination of fptosi * and sitofp (necessary for trunc/floor/ceil/round implementation) @@ -219,6 +275,9 @@ draw_llvm_create(struct draw_context *draw) LLVMDumpModule(llvm->module); } + llvm->nr_variants = 0; + make_empty_list(&llvm->vs_variants_list); + return llvm; } @@ -231,9 +290,13 @@ draw_llvm_destroy(struct draw_llvm *llvm) } struct draw_llvm_variant * -draw_llvm_prepare(struct draw_llvm *llvm, int num_inputs) +draw_llvm_create_variant(struct draw_llvm *llvm, int num_inputs) { struct draw_llvm_variant *variant = MALLOC(sizeof(struct draw_llvm_variant)); + struct llvm_vertex_shader *shader = + llvm_vertex_shader(llvm->draw->vs.vertex_shader); + + variant->llvm = llvm; draw_llvm_make_variant_key(llvm, &variant->key); @@ -242,6 +305,12 @@ draw_llvm_prepare(struct draw_llvm *llvm, int num_inputs) draw_llvm_generate(llvm, variant); draw_llvm_generate_elts(llvm, variant); + variant->shader = shader; + variant->list_item_global.base = variant; + variant->list_item_local.base = variant; + /*variant->no = */shader->variants_created++; + variant->list_item_global.base = variant; + return variant; } @@ -250,11 +319,13 @@ generate_vs(struct draw_llvm *llvm, LLVMBuilderRef builder, LLVMValueRef (*outputs)[NUM_CHANNELS], const LLVMValueRef (*inputs)[NUM_CHANNELS], - LLVMValueRef context_ptr) + LLVMValueRef context_ptr, + struct lp_build_sampler_soa *draw_sampler) { const struct tgsi_token *tokens = llvm->draw->vs.vertex_shader->state.tokens; struct lp_type vs_type; LLVMValueRef consts_ptr = draw_jit_context_vs_constants(builder, context_ptr); + struct lp_build_sampler_soa *sampler = 0; memset(&vs_type, 0, sizeof vs_type); vs_type.floating = TRUE; /* floating point values */ @@ -270,6 +341,10 @@ generate_vs(struct draw_llvm *llvm, tgsi_dump(tokens, 0); } + if (llvm->draw->num_sampler_views && + llvm->draw->num_samplers) + sampler = draw_sampler; + lp_build_tgsi_soa(builder, tokens, vs_type, @@ -278,7 +353,7 @@ generate_vs(struct draw_llvm *llvm, NULL /*pos*/, inputs, outputs, - NULL/*sampler*/, + sampler, &llvm->draw->vs.vertex_shader->info); } @@ -306,7 +381,8 @@ generate_fetch(LLVMBuilderRef builder, LLVMValueRef *res, struct pipe_vertex_element *velem, LLVMValueRef vbuf, - LLVMValueRef index) + LLVMValueRef index, + LLVMValueRef instance_id) { LLVMValueRef indices = LLVMConstInt(LLVMInt64Type(), velem->vertex_buffer_index, 0); LLVMValueRef vbuffer_ptr = LLVMBuildGEP(builder, vbuffers_ptr, @@ -317,8 +393,15 @@ generate_fetch(LLVMBuilderRef builder, LLVMValueRef cond; LLVMValueRef stride; - cond = LLVMBuildICmp(builder, LLVMIntULE, index, vb_max_index, ""); + if (velem->instance_divisor) { + /* array index = instance_id / instance_divisor */ + index = LLVMBuildUDiv(builder, instance_id, + LLVMConstInt(LLVMInt32Type(), velem->instance_divisor, 0), + "instance_divisor"); + } + /* limit index to min(inex, vb_max_index) */ + cond = LLVMBuildICmp(builder, LLVMIntULE, index, vb_max_index, ""); index = LLVMBuildSelect(builder, cond, index, vb_max_index, ""); stride = LLVMBuildMul(builder, vb_stride, index, ""); @@ -586,13 +669,14 @@ convert_to_aos(LLVMBuilderRef builder, static void draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) { - LLVMTypeRef arg_types[7]; + LLVMTypeRef arg_types[8]; LLVMTypeRef func_type; LLVMValueRef context_ptr; LLVMBasicBlockRef block; LLVMBuilderRef builder; LLVMValueRef start, end, count, stride, step, io_itr; LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr; + LLVMValueRef instance_id; struct draw_context *draw = llvm->draw; unsigned i, j; struct lp_build_context bld; @@ -601,6 +685,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) const int max_vertices = 4; LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS]; void *code; + struct lp_build_sampler_soa *sampler = 0; arg_types[0] = llvm->context_ptr_type; /* context */ arg_types[1] = llvm->vertex_header_ptr_type; /* vertex_header */ @@ -609,6 +694,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) arg_types[4] = LLVMInt32Type(); /* count */ arg_types[5] = LLVMInt32Type(); /* stride */ arg_types[6] = llvm->vb_ptr_type; /* pipe_vertex_buffer's */ + arg_types[7] = LLVMInt32Type(); /* instance_id */ func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0); @@ -625,6 +711,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) count = LLVMGetParam(variant->function, 4); stride = LLVMGetParam(variant->function, 5); vb_ptr = LLVMGetParam(variant->function, 6); + instance_id = LLVMGetParam(variant->function, 7); lp_build_name(context_ptr, "context"); lp_build_name(io_ptr, "io"); @@ -633,6 +720,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) lp_build_name(count, "count"); lp_build_name(stride, "stride"); lp_build_name(vb_ptr, "vb"); + lp_build_name(instance_id, "instance_id"); /* * Function body @@ -648,6 +736,10 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0); + /* code generated texture sampling */ + sampler = draw_llvm_sampler_soa_create(variant->key.sampler, + context_ptr); + #if DEBUG_STORE lp_build_printf(builder, "start = %d, end = %d, step = %d\n", start, end, step); @@ -678,7 +770,8 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr, &vb_index, 1, ""); generate_fetch(builder, vbuffers_ptr, - &aos_attribs[j][i], velem, vb, true_index); + &aos_attribs[j][i], velem, vb, true_index, + instance_id); } } convert_to_soa(builder, aos_attribs, inputs, @@ -689,7 +782,8 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) builder, outputs, ptr_aos, - context_ptr); + context_ptr, + sampler); convert_to_aos(builder, io, outputs, draw->vs.vertex_shader->info.num_outputs, @@ -697,6 +791,8 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) } lp_build_loop_end_cond(builder, end, step, LLVMIntUGE, &lp_loop); + sampler->destroy(sampler); + LLVMBuildRetVoid(builder); LLVMDisposeBuilder(builder); @@ -730,13 +826,14 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant) static void draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *variant) { - LLVMTypeRef arg_types[7]; + LLVMTypeRef arg_types[8]; LLVMTypeRef func_type; LLVMValueRef context_ptr; LLVMBasicBlockRef block; LLVMBuilderRef builder; LLVMValueRef fetch_elts, fetch_count, stride, step, io_itr; LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr; + LLVMValueRef instance_id; struct draw_context *draw = llvm->draw; unsigned i, j; struct lp_build_context bld; @@ -747,6 +844,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS]; LLVMValueRef fetch_max; void *code; + struct lp_build_sampler_soa *sampler = 0; arg_types[0] = llvm->context_ptr_type; /* context */ arg_types[1] = llvm->vertex_header_ptr_type; /* vertex_header */ @@ -755,14 +853,17 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian arg_types[4] = LLVMInt32Type(); /* fetch_count */ arg_types[5] = LLVMInt32Type(); /* stride */ arg_types[6] = llvm->vb_ptr_type; /* pipe_vertex_buffer's */ + arg_types[7] = LLVMInt32Type(); /* instance_id */ func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0); - variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts", func_type); + variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts", + func_type); LLVMSetFunctionCallConv(variant->function_elts, LLVMCCallConv); for(i = 0; i < Elements(arg_types); ++i) if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind) - LLVMAddAttribute(LLVMGetParam(variant->function_elts, i), LLVMNoAliasAttribute); + LLVMAddAttribute(LLVMGetParam(variant->function_elts, i), + LLVMNoAliasAttribute); context_ptr = LLVMGetParam(variant->function_elts, 0); io_ptr = LLVMGetParam(variant->function_elts, 1); @@ -771,6 +872,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian fetch_count = LLVMGetParam(variant->function_elts, 4); stride = LLVMGetParam(variant->function_elts, 5); vb_ptr = LLVMGetParam(variant->function_elts, 6); + instance_id = LLVMGetParam(variant->function_elts, 7); lp_build_name(context_ptr, "context"); lp_build_name(io_ptr, "io"); @@ -779,6 +881,7 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian lp_build_name(fetch_count, "fetch_count"); lp_build_name(stride, "stride"); lp_build_name(vb_ptr, "vb"); + lp_build_name(instance_id, "instance_id"); /* * Function body @@ -793,6 +896,10 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0); + /* code generated texture sampling */ + sampler = draw_llvm_sampler_soa_create(variant->key.sampler, + context_ptr); + fetch_max = LLVMBuildSub(builder, fetch_count, LLVMConstInt(LLVMInt32Type(), 1, 0), "fetch_max"); @@ -833,7 +940,8 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr, &vb_index, 1, ""); generate_fetch(builder, vbuffers_ptr, - &aos_attribs[j][i], velem, vb, true_index); + &aos_attribs[j][i], velem, vb, true_index, + instance_id); } } convert_to_soa(builder, aos_attribs, inputs, @@ -844,7 +952,8 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian builder, outputs, ptr_aos, - context_ptr); + context_ptr, + sampler); convert_to_aos(builder, io, outputs, draw->vs.vertex_shader->info.num_outputs, @@ -852,6 +961,8 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian } lp_build_loop_end_cond(builder, fetch_count, step, LLVMIntUGE, &lp_loop); + sampler->destroy(sampler); + LLVMBuildRetVoid(builder); LLVMDisposeBuilder(builder); @@ -885,6 +996,8 @@ void draw_llvm_make_variant_key(struct draw_llvm *llvm, struct draw_llvm_variant_key *key) { + unsigned i; + memset(key, 0, sizeof(struct draw_llvm_variant_key)); key->nr_vertex_elements = llvm->draw->pt.nr_vertex_elements; @@ -896,4 +1009,72 @@ draw_llvm_make_variant_key(struct draw_llvm *llvm, memcpy(&key->vs, &llvm->draw->vs.vertex_shader->state, sizeof(struct pipe_shader_state)); + + /* if the driver implemented the sampling hooks then + * setup our sampling state */ + if (llvm->draw->num_sampler_views && llvm->draw->num_samplers) { + for(i = 0; i < PIPE_MAX_VERTEX_SAMPLERS; ++i) { + struct draw_vertex_shader *shader = llvm->draw->vs.vertex_shader; + if(shader->info.file_mask[TGSI_FILE_SAMPLER] & (1 << i)) + lp_sampler_static_state(&key->sampler[i], + llvm->draw->sampler_views[i], + llvm->draw->samplers[i]); + } + } +} + +void +draw_llvm_set_mapped_texture(struct draw_context *draw, + unsigned sampler_idx, + uint32_t width, uint32_t height, uint32_t depth, + uint32_t last_level, + uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS], + uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS], + const void *data[DRAW_MAX_TEXTURE_LEVELS]) +{ + unsigned j; + struct draw_jit_texture *jit_tex; + + assert(sampler_idx < PIPE_MAX_VERTEX_SAMPLERS); + + + jit_tex = &draw->llvm->jit_context.textures[sampler_idx]; + + jit_tex->width = width; + jit_tex->height = height; + jit_tex->depth = depth; + jit_tex->last_level = last_level; + + for (j = 0; j <= last_level; j++) { + jit_tex->data[j] = data[j]; + jit_tex->row_stride[j] = row_stride[j]; + jit_tex->img_stride[j] = img_stride[j]; + } +} + +void +draw_llvm_destroy_variant(struct draw_llvm_variant *variant) +{ + struct draw_llvm *llvm = variant->llvm; + struct draw_context *draw = llvm->draw; + + if (variant->function_elts) { + if (variant->function_elts) + LLVMFreeMachineCodeForFunction(draw->engine, + variant->function_elts); + LLVMDeleteFunction(variant->function_elts); + } + + if (variant->function) { + if (variant->function) + LLVMFreeMachineCodeForFunction(draw->engine, + variant->function); + LLVMDeleteFunction(variant->function); + } + + remove_from_list(&variant->list_item_local); + variant->shader->variants_cached--; + remove_from_list(&variant->list_item_global); + llvm->nr_variants--; + FREE(variant); } diff --git a/src/gallium/auxiliary/draw/draw_llvm.h b/src/gallium/auxiliary/draw/draw_llvm.h index 58fee7f9d6..4addb47d2d 100644 --- a/src/gallium/auxiliary/draw/draw_llvm.h +++ b/src/gallium/auxiliary/draw/draw_llvm.h @@ -1,28 +1,71 @@ -#ifndef HAVE_LLVM_H -#define HAVE_LLVM_H +/************************************************************************** + * + * Copyright 2010 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef DRAW_LLVM_H +#define DRAW_LLVM_H #include "draw/draw_private.h" +#include "draw/draw_vs.h" +#include "gallivm/lp_bld_sample.h" + #include "pipe/p_context.h" +#include "util/u_simple_list.h" #include <llvm-c/Core.h> #include <llvm-c/Analysis.h> #include <llvm-c/Target.h> #include <llvm-c/ExecutionEngine.h> +#define DRAW_MAX_TEXTURE_LEVELS 13 /* 4K x 4K for now */ + +struct draw_llvm; +struct llvm_vertex_shader; + struct draw_jit_texture { uint32_t width; uint32_t height; - uint32_t stride; - const void *data; + uint32_t depth; + uint32_t last_level; + uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS]; + uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS]; + const void *data[DRAW_MAX_TEXTURE_LEVELS]; }; enum { DRAW_JIT_TEXTURE_WIDTH = 0, DRAW_JIT_TEXTURE_HEIGHT, - DRAW_JIT_TEXTURE_STRIDE, - DRAW_JIT_TEXTURE_DATA + DRAW_JIT_TEXTURE_DEPTH, + DRAW_JIT_TEXTURE_LAST_LEVEL, + DRAW_JIT_TEXTURE_ROW_STRIDE, + DRAW_JIT_TEXTURE_IMG_STRIDE, + DRAW_JIT_TEXTURE_DATA, + DRAW_JIT_TEXTURE_NUM_FIELDS /* number of fields above */ }; enum { @@ -48,7 +91,7 @@ struct draw_jit_context const float *gs_constants; - struct draw_jit_texture textures[PIPE_MAX_SAMPLERS]; + struct draw_jit_texture textures[PIPE_MAX_VERTEX_SAMPLERS]; }; @@ -58,10 +101,10 @@ struct draw_jit_context #define draw_jit_context_gs_constants(_builder, _ptr) \ lp_build_struct_get(_builder, _ptr, 1, "gs_constants") -#define DRAW_JIT_CONTEXT_TEXTURES_INDEX 2 +#define DRAW_JIT_CTX_TEXTURES 2 #define draw_jit_context_textures(_builder, _ptr) \ - lp_build_struct_get_ptr(_builder, _ptr, DRAW_JIT_CONTEXT_TEXTURES_INDEX, "textures") + lp_build_struct_get_ptr(_builder, _ptr, DRAW_JIT_CTX_TEXTURES, "textures") @@ -92,7 +135,8 @@ typedef void unsigned start, unsigned count, unsigned stride, - struct pipe_vertex_buffer *vertex_buffers); + struct pipe_vertex_buffer *vertex_buffers, + unsigned instance_id); typedef void @@ -102,13 +146,54 @@ typedef void const unsigned *fetch_elts, unsigned fetch_count, unsigned stride, - struct pipe_vertex_buffer *vertex_buffers); + struct pipe_vertex_buffer *vertex_buffers, + unsigned instance_id); + +struct draw_llvm_variant_key +{ + struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS]; + unsigned nr_vertex_elements; + struct pipe_shader_state vs; + struct lp_sampler_static_state sampler[PIPE_MAX_VERTEX_SAMPLERS]; +}; + +struct draw_llvm_variant_list_item +{ + struct draw_llvm_variant *base; + struct draw_llvm_variant_list_item *next, *prev; +}; + +struct draw_llvm_variant +{ + struct draw_llvm_variant_key key; + LLVMValueRef function; + LLVMValueRef function_elts; + draw_jit_vert_func jit_func; + draw_jit_vert_func_elts jit_func_elts; + + struct llvm_vertex_shader *shader; + + struct draw_llvm *llvm; + struct draw_llvm_variant_list_item list_item_global; + struct draw_llvm_variant_list_item list_item_local; +}; + +struct llvm_vertex_shader { + struct draw_vertex_shader base; + + struct draw_llvm_variant_list_item variants; + unsigned variants_created; + unsigned variants_cached; +}; struct draw_llvm { struct draw_context *draw; struct draw_jit_context jit_context; + struct draw_llvm_variant_list_item vs_variants_list; + int nr_variants; + LLVMModuleRef module; LLVMExecutionEngineRef engine; LLVMModuleProviderRef provider; @@ -121,23 +206,12 @@ struct draw_llvm { LLVMTypeRef vb_ptr_type; }; -struct draw_llvm_variant_key +static INLINE struct llvm_vertex_shader * +llvm_vertex_shader(struct draw_vertex_shader *vs) { - struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS]; - unsigned nr_vertex_elements; - struct pipe_shader_state vs; -}; + return (struct llvm_vertex_shader *)vs; +} -struct draw_llvm_variant -{ - struct draw_llvm_variant_key key; - LLVMValueRef function; - LLVMValueRef function_elts; - draw_jit_vert_func jit_func; - draw_jit_vert_func_elts jit_func_elts; - - struct draw_llvm_variant *next; -}; struct draw_llvm * draw_llvm_create(struct draw_context *draw); @@ -146,7 +220,10 @@ void draw_llvm_destroy(struct draw_llvm *llvm); struct draw_llvm_variant * -draw_llvm_prepare(struct draw_llvm *llvm, int num_inputs); +draw_llvm_create_variant(struct draw_llvm *llvm, int num_inputs); + +void +draw_llvm_destroy_variant(struct draw_llvm_variant *variant); void draw_llvm_make_variant_key(struct draw_llvm *llvm, @@ -156,4 +233,18 @@ LLVMValueRef draw_llvm_translate_from(LLVMBuilderRef builder, LLVMValueRef vbuffer, enum pipe_format from_format); + +struct lp_build_sampler_soa * +draw_llvm_sampler_soa_create(const struct lp_sampler_static_state *static_state, + LLVMValueRef context_ptr); + +void +draw_llvm_set_mapped_texture(struct draw_context *draw, + unsigned sampler_idx, + uint32_t width, uint32_t height, uint32_t depth, + uint32_t last_level, + uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS], + uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS], + const void *data[DRAW_MAX_TEXTURE_LEVELS]); + #endif diff --git a/src/gallium/auxiliary/draw/draw_llvm_sample.c b/src/gallium/auxiliary/draw/draw_llvm_sample.c new file mode 100644 index 0000000000..e9811010db --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_llvm_sample.c @@ -0,0 +1,215 @@ +/************************************************************************** + * + * Copyright 2010 VMware, Inc. + * All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Texture sampling code generation + * @author Jose Fonseca <jfonseca@vmware.com> + */ + +#include "pipe/p_defines.h" +#include "pipe/p_shader_tokens.h" +#include "gallivm/lp_bld_debug.h" +#include "gallivm/lp_bld_type.h" +#include "gallivm/lp_bld_sample.h" +#include "gallivm/lp_bld_tgsi.h" + + +#include "util/u_debug.h" +#include "util/u_memory.h" +#include "util/u_pointer.h" +#include "util/u_string.h" + +#include "draw_llvm.h" + + +/** + * This provides the bridge between the sampler state store in + * lp_jit_context and lp_jit_texture and the sampler code + * generator. It provides the texture layout information required by + * the texture sampler code generator in terms of the state stored in + * lp_jit_context and lp_jit_texture in runtime. + */ +struct draw_llvm_sampler_dynamic_state +{ + struct lp_sampler_dynamic_state base; + + const struct lp_sampler_static_state *static_state; + + LLVMValueRef context_ptr; +}; + + +/** + * This is the bridge between our sampler and the TGSI translator. + */ +struct draw_llvm_sampler_soa +{ + struct lp_build_sampler_soa base; + + struct draw_llvm_sampler_dynamic_state dynamic_state; +}; + + +/** + * Fetch the specified member of the lp_jit_texture structure. + * \param emit_load if TRUE, emit the LLVM load instruction to actually + * fetch the field's value. Otherwise, just emit the + * GEP code to address the field. + * + * @sa http://llvm.org/docs/GetElementPtr.html + */ +static LLVMValueRef +draw_llvm_texture_member(const struct lp_sampler_dynamic_state *base, + LLVMBuilderRef builder, + unsigned unit, + unsigned member_index, + const char *member_name, + boolean emit_load) +{ + struct draw_llvm_sampler_dynamic_state *state = + (struct draw_llvm_sampler_dynamic_state *)base; + LLVMValueRef indices[4]; + LLVMValueRef ptr; + LLVMValueRef res; + + debug_assert(unit < PIPE_MAX_VERTEX_SAMPLERS); + + /* context[0] */ + indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0); + /* context[0].textures */ + indices[1] = LLVMConstInt(LLVMInt32Type(), DRAW_JIT_CTX_TEXTURES, 0); + /* context[0].textures[unit] */ + indices[2] = LLVMConstInt(LLVMInt32Type(), unit, 0); + /* context[0].textures[unit].member */ + indices[3] = LLVMConstInt(LLVMInt32Type(), member_index, 0); + + ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), ""); + + if (emit_load) + res = LLVMBuildLoad(builder, ptr, ""); + else + res = ptr; + + lp_build_name(res, "context.texture%u.%s", unit, member_name); + + return res; +} + + +/** + * Helper macro to instantiate the functions that generate the code to + * fetch the members of lp_jit_texture to fulfill the sampler code + * generator requests. + * + * This complexity is the price we have to pay to keep the texture + * sampler code generator a reusable module without dependencies to + * llvmpipe internals. + */ +#define DRAW_LLVM_TEXTURE_MEMBER(_name, _index, _emit_load) \ + static LLVMValueRef \ + draw_llvm_texture_##_name( const struct lp_sampler_dynamic_state *base, \ + LLVMBuilderRef builder, \ + unsigned unit) \ + { \ + return draw_llvm_texture_member(base, builder, unit, _index, #_name, _emit_load ); \ + } + + +DRAW_LLVM_TEXTURE_MEMBER(width, DRAW_JIT_TEXTURE_WIDTH, TRUE) +DRAW_LLVM_TEXTURE_MEMBER(height, DRAW_JIT_TEXTURE_HEIGHT, TRUE) +DRAW_LLVM_TEXTURE_MEMBER(depth, DRAW_JIT_TEXTURE_DEPTH, TRUE) +DRAW_LLVM_TEXTURE_MEMBER(last_level, DRAW_JIT_TEXTURE_LAST_LEVEL, TRUE) +DRAW_LLVM_TEXTURE_MEMBER(row_stride, DRAW_JIT_TEXTURE_ROW_STRIDE, FALSE) +DRAW_LLVM_TEXTURE_MEMBER(img_stride, DRAW_JIT_TEXTURE_IMG_STRIDE, FALSE) +DRAW_LLVM_TEXTURE_MEMBER(data_ptr, DRAW_JIT_TEXTURE_DATA, FALSE) + + +static void +draw_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler) +{ + FREE(sampler); +} + + +/** + * Fetch filtered values from texture. + * The 'texel' parameter returns four vectors corresponding to R, G, B, A. + */ +static void +draw_llvm_sampler_soa_emit_fetch_texel(const struct lp_build_sampler_soa *base, + LLVMBuilderRef builder, + struct lp_type type, + unsigned unit, + unsigned num_coords, + const LLVMValueRef *coords, + const LLVMValueRef *ddx, + const LLVMValueRef *ddy, + LLVMValueRef lod_bias, /* optional */ + LLVMValueRef explicit_lod, /* optional */ + LLVMValueRef *texel) +{ + struct draw_llvm_sampler_soa *sampler = (struct draw_llvm_sampler_soa *)base; + + assert(unit < PIPE_MAX_VERTEX_SAMPLERS); + + lp_build_sample_soa(builder, + &sampler->dynamic_state.static_state[unit], + &sampler->dynamic_state.base, + type, + unit, + num_coords, coords, + ddx, ddy, + lod_bias, explicit_lod, + texel); +} + + +struct lp_build_sampler_soa * +draw_llvm_sampler_soa_create(const struct lp_sampler_static_state *static_state, + LLVMValueRef context_ptr) +{ + struct draw_llvm_sampler_soa *sampler; + + sampler = CALLOC_STRUCT(draw_llvm_sampler_soa); + if(!sampler) + return NULL; + + sampler->base.destroy = draw_llvm_sampler_soa_destroy; + sampler->base.emit_fetch_texel = draw_llvm_sampler_soa_emit_fetch_texel; + sampler->dynamic_state.base.width = draw_llvm_texture_width; + sampler->dynamic_state.base.height = draw_llvm_texture_height; + sampler->dynamic_state.base.depth = draw_llvm_texture_depth; + sampler->dynamic_state.base.last_level = draw_llvm_texture_last_level; + sampler->dynamic_state.base.row_stride = draw_llvm_texture_row_stride; + sampler->dynamic_state.base.img_stride = draw_llvm_texture_img_stride; + sampler->dynamic_state.base.data_ptr = draw_llvm_texture_data_ptr; + sampler->dynamic_state.static_state = static_state; + sampler->dynamic_state.context_ptr = context_ptr; + + return &sampler->base; +} + diff --git a/src/gallium/auxiliary/draw/draw_llvm_translate.c b/src/gallium/auxiliary/draw/draw_llvm_translate.c index d7da7ed357..6ebe1f7de4 100644 --- a/src/gallium/auxiliary/draw/draw_llvm_translate.c +++ b/src/gallium/auxiliary/draw/draw_llvm_translate.c @@ -7,6 +7,7 @@ #include "gallivm/lp_bld_struct.h" #include "gallivm/lp_bld_format.h" #include "gallivm/lp_bld_debug.h" +#include "gallivm/lp_bld_type.h" #include "util/u_memory.h" #include "util/u_format.h" @@ -466,6 +467,7 @@ draw_llvm_translate_from(LLVMBuilderRef builder, const struct util_format_description *format_desc; LLVMValueRef zero; int i; + struct lp_type type = lp_float32_vec4_type(); /* * The above can only cope with straight arrays: no bitfields, @@ -493,5 +495,5 @@ draw_llvm_translate_from(LLVMBuilderRef builder, format_desc = util_format_description(from_format); zero = LLVMConstNull(LLVMInt32Type()); - return lp_build_fetch_rgba_aos(builder, format_desc, vbuffer, zero, zero); + return lp_build_fetch_rgba_aos(builder, format_desc, type, vbuffer, zero, zero, zero); } diff --git a/src/gallium/auxiliary/draw/draw_pipe.c b/src/gallium/auxiliary/draw/draw_pipe.c index 83556f10a8..8cd75ecf9a 100644 --- a/src/gallium/auxiliary/draw/draw_pipe.c +++ b/src/gallium/auxiliary/draw/draw_pipe.c @@ -177,15 +177,15 @@ static void do_triangle( struct draw_context *draw, ( DRAW_PIPE_RESET_STIPPLE | \ DRAW_PIPE_EDGE_FLAG_0 | \ DRAW_PIPE_EDGE_FLAG_1 ), \ - verts + stride * elts[i0], \ - verts + stride * elts[i1], \ - verts + stride * elts[i2]); \ + verts + stride * (elts[i0] & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (elts[i1] & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (elts[i2] & ~DRAW_PIPE_FLAG_MASK)); \ do_triangle( draw, \ ( DRAW_PIPE_EDGE_FLAG_1 | \ DRAW_PIPE_EDGE_FLAG_2 ), \ - verts + stride * elts[i0], \ - verts + stride * elts[i2], \ - verts + stride * elts[i3]) + verts + stride * (elts[i0] & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (elts[i2] & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (elts[i3] & ~DRAW_PIPE_FLAG_MASK)) /* emit last quad vertex as last vertex in triangles */ #define QUAD_LAST_PV(i0,i1,i2,i3) \ @@ -193,15 +193,15 @@ static void do_triangle( struct draw_context *draw, ( DRAW_PIPE_RESET_STIPPLE | \ DRAW_PIPE_EDGE_FLAG_0 | \ DRAW_PIPE_EDGE_FLAG_2 ), \ - verts + stride * elts[i0], \ - verts + stride * elts[i1], \ - verts + stride * elts[i3]); \ + verts + stride * (elts[i0] & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (elts[i1] & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (elts[i3] & ~DRAW_PIPE_FLAG_MASK)); \ do_triangle( draw, \ ( DRAW_PIPE_EDGE_FLAG_0 | \ DRAW_PIPE_EDGE_FLAG_1 ), \ - verts + stride * elts[i1], \ - verts + stride * elts[i2], \ - verts + stride * elts[i3]) + verts + stride * (elts[i1] & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (elts[i2] & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * (elts[i3] & ~DRAW_PIPE_FLAG_MASK)) #define TRIANGLE(flags,i0,i1,i2) \ do_triangle( draw, \ @@ -218,7 +218,7 @@ static void do_triangle( struct draw_context *draw, #define POINT(i0) \ do_point( draw, \ - verts + stride * elts[i0] ) + verts + stride * (elts[i0] & ~DRAW_PIPE_FLAG_MASK) ) #define FUNC pipe_run #define ARGS \ @@ -260,7 +260,7 @@ void draw_pipeline_run( struct draw_context *draw, const struct draw_prim_info *prim_info) { unsigned i, start; - + draw->pipeline.verts = (char *)vert_info->verts; draw->pipeline.vertex_stride = vert_info->stride; draw->pipeline.vertex_count = vert_info->count; @@ -296,14 +296,14 @@ void draw_pipeline_run( struct draw_context *draw, DRAW_PIPE_EDGE_FLAG_0 | \ DRAW_PIPE_EDGE_FLAG_1 ), \ verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i1), \ - verts + stride * (i2)); \ + verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * ((i2) & ~DRAW_PIPE_FLAG_MASK)); \ do_triangle( draw, \ ( DRAW_PIPE_EDGE_FLAG_1 | \ DRAW_PIPE_EDGE_FLAG_2 ), \ verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i2), \ - verts + stride * (i3)) + verts + stride * ((i2) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * ((i3) & ~DRAW_PIPE_FLAG_MASK)) /* emit last quad vertex as last vertex in triangles */ #define QUAD_LAST_PV(i0,i1,i2,i3) \ @@ -312,31 +312,31 @@ void draw_pipeline_run( struct draw_context *draw, DRAW_PIPE_EDGE_FLAG_0 | \ DRAW_PIPE_EDGE_FLAG_2 ), \ verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i1), \ - verts + stride * (i3)); \ + verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * ((i3) & ~DRAW_PIPE_FLAG_MASK)); \ do_triangle( draw, \ ( DRAW_PIPE_EDGE_FLAG_0 | \ DRAW_PIPE_EDGE_FLAG_1 ), \ verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i2), \ - verts + stride * (i3)) + verts + stride * ((i2) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * ((i3) & ~DRAW_PIPE_FLAG_MASK)) #define TRIANGLE(flags,i0,i1,i2) \ do_triangle( draw, \ flags, /* flags */ \ verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i1), \ - verts + stride * (i2)) + verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK), \ + verts + stride * ((i2) & ~DRAW_PIPE_FLAG_MASK)) #define LINE(flags,i0,i1) \ do_line( draw, \ flags, \ verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \ - verts + stride * (i1)) + verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK)) #define POINT(i0) \ do_point( draw, \ - verts + stride * i0 ) + verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK) ) #define FUNC pipe_run_linear #define ARGS \ diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c index debd17fd74..c0135f5bb7 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c +++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c @@ -425,7 +425,8 @@ aaline_create_texture(struct aaline_stage *aaline) /* Fill in mipmap images. * Basically each level is solid opaque, except for the outermost - * texels which are zero. Special case the 1x1 and 2x2 levels. + * texels which are zero. Special case the 1x1 and 2x2 levels + * (though, those levels shouldn't be used - see the max_lod setting). */ for (level = 0; level <= MAX_TEXTURE_LEVEL; level++) { struct pipe_transfer *transfer; @@ -497,7 +498,8 @@ aaline_create_sampler(struct aaline_stage *aaline) sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR; sampler.normalized_coords = 1; sampler.min_lod = 0.0f; - sampler.max_lod = MAX_TEXTURE_LEVEL; + /* avoid using the 1x1 and 2x2 mipmap levels */ + sampler.max_lod = MAX_TEXTURE_LEVEL - 2; aaline->sampler_cso = pipe->create_sampler_state(pipe, &sampler); if (aaline->sampler_cso == NULL) @@ -669,8 +671,8 @@ aaline_first_line(struct draw_stage *stage, struct prim_header *header) assert(draw->rasterizer->line_smooth); - if (draw->rasterizer->line_width <= 3.0) - aaline->half_line_width = 1.5f; + if (draw->rasterizer->line_width <= 2.2) + aaline->half_line_width = 1.1f; else aaline->half_line_width = 0.5f * draw->rasterizer->line_width; diff --git a/src/gallium/auxiliary/draw/draw_pipe_clip.c b/src/gallium/auxiliary/draw/draw_pipe_clip.c index 122b1c7968..1cf6ee7a7f 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_clip.c +++ b/src/gallium/auxiliary/draw/draw_pipe_clip.c @@ -262,6 +262,7 @@ do_clip_tri( struct draw_stage *stage, clipmask &= ~(1<<plane_idx); + assert(n < MAX_CLIPPED_VERTICES); inlist[n] = inlist[0]; /* prevent rotation of vertices */ for (i = 1; i <= n; i++) { @@ -270,11 +271,17 @@ do_clip_tri( struct draw_stage *stage, float dp = dot4( vert->clip, plane ); if (!IS_NEGATIVE(dp_prev)) { + assert(outcount < MAX_CLIPPED_VERTICES); outlist[outcount++] = vert_prev; } if (DIFFERENT_SIGNS(dp, dp_prev)) { - struct vertex_header *new_vert = clipper->stage.tmp[tmpnr++]; + struct vertex_header *new_vert; + + assert(tmpnr < MAX_CLIPPED_VERTICES+1); + new_vert = clipper->stage.tmp[tmpnr++]; + + assert(outcount < MAX_CLIPPED_VERTICES); outlist[outcount++] = new_vert; if (IS_NEGATIVE(dp)) { @@ -317,12 +324,14 @@ do_clip_tri( struct draw_stage *stage, if (clipper->flat) { if (stage->draw->rasterizer->flatshade_first) { if (inlist[0] != header->v[0]) { + assert(tmpnr < MAX_CLIPPED_VERTICES + 1); inlist[0] = dup_vert(stage, inlist[0], tmpnr++); copy_colors(stage, inlist[0], header->v[0]); } } else { if (inlist[0] != header->v[2]) { + assert(tmpnr < MAX_CLIPPED_VERTICES + 1); inlist[0] = dup_vert(stage, inlist[0], tmpnr++); copy_colors(stage, inlist[0], header->v[2]); } diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c index fff960c7eb..ed9a53e154 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c +++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c @@ -363,8 +363,12 @@ generate_pstip_fs(struct pstip_stage *pstip) assert(pstip->fs->sampler_unit < PIPE_MAX_SAMPLERS); pstip->fs->pstip_fs = pstip->driver_create_fs_state(pstip->pipe, &pstip_fs); - + FREE((void *)pstip_fs.tokens); + + if (!pstip->fs->pstip_fs) + return FALSE; + return TRUE; } @@ -603,13 +607,16 @@ pstip_destroy(struct draw_stage *stage) } +/** Create a new polygon stipple drawing stage object */ static struct pstip_stage * -draw_pstip_stage(struct draw_context *draw) +draw_pstip_stage(struct draw_context *draw, struct pipe_context *pipe) { struct pstip_stage *pstip = CALLOC_STRUCT(pstip_stage); if (pstip == NULL) goto fail; + pstip->pipe = pipe; + pstip->stage.draw = draw; pstip->stage.name = "pstip"; pstip->stage.next = NULL; @@ -765,14 +772,12 @@ draw_install_pstipple_stage(struct draw_context *draw, /* * Create / install pgon stipple drawing / prim stage */ - pstip = draw_pstip_stage( draw ); + pstip = draw_pstip_stage( draw, pipe ); if (pstip == NULL) goto fail; draw->pipeline.pstipple = &pstip->stage; - pstip->pipe = pipe; - /* create special texture, sampler state */ if (!pstip_create_texture(pstip)) goto fail; diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c index 3e6e538995..ee2945c7c9 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c +++ b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c @@ -226,6 +226,7 @@ static void widepoint_first_point( struct draw_stage *stage, if (rast->gl_rasterization_rules) { wide->xbias = 0.125; + wide->ybias = -0.125; } /* Disable triangle culling, stippling, unfilled mode etc. */ diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index 4584033bc2..058aeedc17 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -48,6 +48,7 @@ #ifdef HAVE_LLVM #include <llvm-c/ExecutionEngine.h> +struct draw_llvm; #endif @@ -81,6 +82,9 @@ struct vertex_header { #define UNDEFINED_VERTEX_ID 0xffff +/* maximum number of shader variants we can cache */ +#define DRAW_MAX_SHADER_VARIANTS 1024 + /** * Private context for the drawing module. */ @@ -245,6 +249,7 @@ struct draw_context */ float plane[12][4]; unsigned nr_planes; + boolean depth_clamp; /* If a prim stage introduces new vertex attributes, they'll be stored here */ @@ -259,9 +264,15 @@ struct draw_context unsigned instance_id; #ifdef HAVE_LLVM + struct draw_llvm *llvm; LLVMExecutionEngineRef engine; #endif + struct pipe_sampler_view *sampler_views[PIPE_MAX_VERTEX_SAMPLERS]; + unsigned num_sampler_views; + const struct pipe_sampler_state *samplers[PIPE_MAX_VERTEX_SAMPLERS]; + unsigned num_samplers; + void *driver_private; }; diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index 02c97fec81..92d4113b4c 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -34,9 +34,11 @@ #include "draw/draw_gs.h" #include "draw/draw_private.h" #include "draw/draw_pt.h" +#include "draw/draw_vs.h" #include "tgsi/tgsi_dump.h" #include "util/u_math.h" #include "util/u_prim.h" +#include "util/u_format.h" DEBUG_GET_ONCE_BOOL_OPTION(draw_fse, "DRAW_FSE", FALSE) @@ -69,7 +71,6 @@ draw_pt_arrays(struct draw_context *draw, struct draw_pt_front_end *frontend = NULL; struct draw_pt_middle_end *middle = NULL; unsigned opt = 0; - unsigned out_prim = prim; /* Sanitize primitive length: */ @@ -80,18 +81,19 @@ draw_pt_arrays(struct draw_context *draw, if (count < first) return TRUE; } - if (draw->gs.geometry_shader) { - out_prim = draw->gs.geometry_shader->output_primitive; - } if (!draw->force_passthrough) { + unsigned gs_out_prim = (draw->gs.geometry_shader ? + draw->gs.geometry_shader->output_primitive : + prim); + if (!draw->render) { opt |= PT_PIPELINE; } if (draw_need_pipeline(draw, draw->rasterizer, - out_prim)) { + gs_out_prim)) { opt |= PT_PIPELINE; } @@ -102,7 +104,7 @@ draw_pt_arrays(struct draw_context *draw, opt |= PT_SHADE; } - if (draw->pt.middle.llvm && !draw->gs.geometry_shader) { + if (draw->pt.middle.llvm) { middle = draw->pt.middle.llvm; } else { if (opt == 0) @@ -122,7 +124,7 @@ draw_pt_arrays(struct draw_context *draw, frontend = draw->pt.front.varray; } - frontend->prepare( frontend, prim, out_prim, middle, opt ); + frontend->prepare( frontend, prim, middle, opt ); frontend->run(frontend, draw_pt_elt_func(draw), @@ -265,31 +267,38 @@ draw_print_arrays(struct draw_context *draw, uint prim, int start, uint count) case PIPE_FORMAT_R32_FLOAT: { float *v = (float *) ptr; - debug_printf("%f @ %p\n", v[0], (void *) v); + debug_printf("R %f @ %p\n", v[0], (void *) v); } break; case PIPE_FORMAT_R32G32_FLOAT: { float *v = (float *) ptr; - debug_printf("%f %f @ %p\n", v[0], v[1], (void *) v); + debug_printf("RG %f %f @ %p\n", v[0], v[1], (void *) v); } break; case PIPE_FORMAT_R32G32B32_FLOAT: { float *v = (float *) ptr; - debug_printf("%f %f %f @ %p\n", v[0], v[1], v[2], (void *) v); + debug_printf("RGB %f %f %f @ %p\n", v[0], v[1], v[2], (void *) v); } break; case PIPE_FORMAT_R32G32B32A32_FLOAT: { float *v = (float *) ptr; - debug_printf("%f %f %f %f @ %p\n", v[0], v[1], v[2], v[3], + debug_printf("RGBA %f %f %f %f @ %p\n", v[0], v[1], v[2], v[3], (void *) v); } break; + case PIPE_FORMAT_B8G8R8A8_UNORM: + { + ubyte *u = (ubyte *) ptr; + debug_printf("BGRA %d %d %d %d @ %p\n", u[0], u[1], u[2], u[3], + (void *) u); + } + break; default: - debug_printf("other format (fix me)\n"); - ; + debug_printf("other format %s (fix me)\n", + util_format_name(draw->pt.vertex_element[j].src_format)); } } } @@ -297,11 +306,8 @@ draw_print_arrays(struct draw_context *draw, uint prim, int start, uint count) /** - * Draw vertex arrays - * This is the main entrypoint into the drawing module. - * \param prim one of PIPE_PRIM_x - * \param start index of first vertex to draw - * \param count number of vertices to draw + * Non-instanced drawing. + * \sa draw_arrays_instanced */ void draw_arrays(struct draw_context *draw, unsigned prim, @@ -310,6 +316,20 @@ draw_arrays(struct draw_context *draw, unsigned prim, draw_arrays_instanced(draw, prim, start, count, 0, 1); } + +/** + * Draw vertex arrays. + * This is the main entrypoint into the drawing module. + * If drawing an indexed primitive, the draw_set_mapped_element_buffer_range() + * function should have already been called to specify the element/index buffer + * information. + * + * \param prim one of PIPE_PRIM_x + * \param start index of first vertex to draw + * \param count number of vertices to draw + * \param startInstance number for the first primitive instance (usually 0). + * \param instanceCount number of instances to draw (1=non-instanced) + */ void draw_arrays_instanced(struct draw_context *draw, unsigned mode, @@ -329,26 +349,30 @@ draw_arrays_instanced(struct draw_context *draw, if (0) draw_print_arrays(draw, mode, start, MIN2(count, 20)); -#if 0 - { - int i; + if (0) { + unsigned int i; debug_printf("draw_arrays(mode=%u start=%u count=%u):\n", mode, start, count); tgsi_dump(draw->vs.vertex_shader->state.tokens, 0); debug_printf("Elements:\n"); for (i = 0; i < draw->pt.nr_vertex_elements; i++) { - debug_printf(" format=%s\n", + debug_printf(" %u: src_offset=%u inst_div=%u vbuf=%u format=%s\n", + i, + draw->pt.vertex_element[i].src_offset, + draw->pt.vertex_element[i].instance_divisor, + draw->pt.vertex_element[i].vertex_buffer_index, util_format_name(draw->pt.vertex_element[i].src_format)); } debug_printf("Buffers:\n"); for (i = 0; i < draw->pt.nr_vertex_buffers; i++) { - debug_printf(" stride=%u offset=%u ptr=%p\n", + debug_printf(" %u: stride=%u maxindex=%u offset=%u ptr=%p\n", + i, draw->pt.vertex_buffer[i].stride, + draw->pt.vertex_buffer[i].max_index, draw->pt.vertex_buffer[i].buffer_offset, draw->pt.user.vbuffer[i]); } } -#endif for (instance = 0; instance < instanceCount; instance++) { draw->instance_id = instance + startInstance; diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h index b6741ca83c..44356fba4c 100644 --- a/src/gallium/auxiliary/draw/draw_pt.h +++ b/src/gallium/auxiliary/draw/draw_pt.h @@ -62,8 +62,7 @@ struct draw_vertex_info; */ struct draw_pt_front_end { void (*prepare)( struct draw_pt_front_end *, - unsigned input_prim, - unsigned output_prim, + unsigned prim, struct draw_pt_middle_end *, unsigned opt ); @@ -87,8 +86,7 @@ struct draw_pt_front_end { */ struct draw_pt_middle_end { void (*prepare)( struct draw_pt_middle_end *, - unsigned input_prim, - unsigned output_prim, + unsigned prim, unsigned opt, unsigned *max_vertices ); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c index bf799db352..ae12ee24bd 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c @@ -68,31 +68,12 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch, fetch->vertex_size = vertex_size; - /* Always emit/leave space for a vertex header. - * - * It's worth considering whether the vertex headers should contain - * a pointer to the 'data', rather than having it inline. - * Something to look at after we've fully switched over to the pt - * paths. + /* Leave the clipmask/edgeflags/pad/vertex_id untouched */ - { - /* Need to set header->vertex_id = 0xffff somehow. - */ - key.element[nr].type = TRANSLATE_ELEMENT_NORMAL; - key.element[nr].input_format = PIPE_FORMAT_R32_FLOAT; - key.element[nr].input_buffer = draw->pt.nr_vertex_buffers; - key.element[nr].input_offset = 0; - key.element[nr].instance_divisor = 0; - key.element[nr].output_format = PIPE_FORMAT_R32_FLOAT; - key.element[nr].output_offset = dst_offset; - dst_offset += 1 * sizeof(float); - nr++; - - - /* Just leave the clip[] array untouched. - */ - dst_offset += 4 * sizeof(float); - } + dst_offset += 1 * sizeof(float); + /* Just leave the clip[] array untouched. + */ + dst_offset += 4 * sizeof(float); if (instance_id_index != ~0) { num_extra_inputs++; @@ -131,26 +112,11 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch, key.nr_elements = nr; key.output_stride = vertex_size; - if (!fetch->translate || translate_key_compare(&fetch->translate->key, &key) != 0) { translate_key_sanitize(&key); fetch->translate = translate_cache_find(fetch->cache, &key); - - { - static struct vertex_header vh = { 0, - 1, - 0, - UNDEFINED_VERTEX_ID, - { .0f, .0f, .0f, .0f } }; - - fetch->translate->set_buffer(fetch->translate, - draw->pt.nr_vertex_buffers, - &vh, - 0, - ~0); - } } } diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c index c629d55563..5c8af17c8e 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c @@ -36,6 +36,7 @@ #include "draw/draw_vbuf.h" #include "draw/draw_vertex.h" #include "draw/draw_pt.h" +#include "draw/draw_gs.h" #include "translate/translate.h" #include "translate/translate_cache.h" @@ -90,7 +91,6 @@ struct fetch_emit_middle_end { static void fetch_emit_prepare( struct draw_pt_middle_end *middle, unsigned prim, - unsigned out_prim, unsigned opt, unsigned *max_vertices ) { @@ -101,9 +101,14 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle, boolean ok; struct translate_key key; + unsigned gs_out_prim = (draw->gs.geometry_shader ? + draw->gs.geometry_shader->output_primitive : + prim); + + ok = draw->render->set_primitive( draw->render, - out_prim ); + gs_out_prim ); if (!ok) { assert(0); return; diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 5483a25f1d..b8270280b6 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -68,8 +68,7 @@ struct fetch_shade_emit { static void fse_prepare( struct draw_pt_middle_end *middle, - unsigned in_prim, - unsigned out_prim, + unsigned prim, unsigned opt, unsigned *max_vertices ) { @@ -80,9 +79,12 @@ static void fse_prepare( struct draw_pt_middle_end *middle, unsigned i; unsigned nr_vbs = 0; + /* Can't support geometry shader on this path. + */ + assert(!draw->gs.geometry_shader); if (!draw->render->set_primitive( draw->render, - out_prim )) { + prim )) { assert(0); return; } diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c index 43b08a030c..121dfc414a 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c @@ -48,13 +48,11 @@ struct fetch_pipeline_middle_end { unsigned vertex_data_offset; unsigned vertex_size; unsigned input_prim; - unsigned output_prim; unsigned opt; }; static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle, - unsigned in_prim, - unsigned out_prim, + unsigned prim, unsigned opt, unsigned *max_vertices ) { @@ -64,6 +62,10 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle, unsigned i; unsigned instance_id_index = ~0; + unsigned gs_out_prim = (draw->gs.geometry_shader ? + draw->gs.geometry_shader->output_primitive : + prim); + /* Add one to num_outputs because the pipeline occasionally tags on * an additional texcoord, eg for AA lines. */ @@ -79,8 +81,7 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle, } } - fpme->input_prim = in_prim; - fpme->output_prim = out_prim; + fpme->input_prim = prim; fpme->opt = opt; /* Always leave room for the vertex header whether we need it or @@ -102,13 +103,13 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle, (boolean)draw->bypass_clipping, (boolean)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, - out_prim, + gs_out_prim, max_vertices ); *max_vertices = MAX2( *max_vertices, @@ -183,7 +184,7 @@ static void draw_vertex_shader_run(struct draw_vertex_shader *vshader, output_verts->count = input_verts->count; output_verts->verts = (struct vertex_header *)MALLOC(output_verts->vertex_size * - output_verts->count); + align(output_verts->count, 4)); vshader->run_linear(vshader, (const float (*)[4])input_verts->verts->data, 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 7d2de58e73..bc074df8c2 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 @@ -49,48 +49,50 @@ struct llvm_middle_end { unsigned vertex_data_offset; unsigned vertex_size; unsigned input_prim; - unsigned output_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 in_prim, - unsigned out_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 llvm_vertex_shader *shader = + llvm_vertex_shader(draw->vs.vertex_shader); 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->input_prim = in_prim; - fpme->output_prim = out_prim; fpme->opt = opt; /* Always leave room for the vertex header whether we need it or @@ -107,7 +109,7 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle, (boolean)draw->bypass_clipping, (boolean)(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 ); @@ -128,20 +130,41 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle, 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, sizeof key) == 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); - 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 */ @@ -210,7 +233,8 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle, fetch_info->start, fetch_info->count, fpme->vertex_size, - draw->pt.vertex_buffer ); + draw->pt.vertex_buffer, + draw->instance_id); else fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context, llvm_vert_info.verts, @@ -218,7 +242,8 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle, fetch_info->elts, fetch_info->count, fpme->vertex_size, - draw->pt.vertex_buffer); + draw->pt.vertex_buffer, + draw->instance_id); /* Finished with fetch and vs: */ @@ -356,31 +381,7 @@ static void llvm_middle_end_finish( struct draw_pt_middle_end *middle ) static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle ) { struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle; - struct draw_context *draw = fpme->draw; - struct draw_llvm_variant *variant = NULL; - - variant = fpme->variants; - while(variant) { - struct draw_llvm_variant *next = variant->next; - if (variant->function_elts) { - if (variant->function_elts) - LLVMFreeMachineCodeForFunction(draw->engine, - variant->function_elts); - LLVMDeleteFunction(variant->function_elts); - } - - if (variant->function) { - if (variant->function) - LLVMFreeMachineCodeForFunction(draw->engine, - variant->function); - LLVMDeleteFunction(variant->function); - } - - FREE(variant); - - variant = next; - } if (fpme->fetch) draw_pt_fetch_destroy( fpme->fetch ); @@ -393,14 +394,12 @@ static void llvm_middle_end_destroy( struct draw_pt_middle_end *middle ) 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; @@ -436,13 +435,11 @@ struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit_llvm( struct draw_cont if (!fpme->so_emit) goto fail; - fpme->llvm = draw_llvm_create(draw); + fpme->llvm = draw->llvm; if (!fpme->llvm) goto fail; - fpme->variants = NULL; fpme->current_variant = NULL; - fpme->nr_variants = 0; return &fpme->base; diff --git a/src/gallium/auxiliary/draw/draw_pt_post_vs.c b/src/gallium/auxiliary/draw/draw_pt_post_vs.c index 112be50f9a..308f927b77 100644 --- a/src/gallium/auxiliary/draw/draw_pt_post_vs.c +++ b/src/gallium/auxiliary/draw/draw_pt_post_vs.c @@ -38,7 +38,14 @@ struct pt_post_vs { struct draw_vertex_info *info ); }; - +static INLINE void +initialize_vertex_header(struct vertex_header *header) +{ + header->clipmask = 0; + header->edgeflag = 1; + header->pad = 0; + header->vertex_id = UNDEFINED_VERTEX_ID; +} static INLINE float dot4(const float *a, const float *b) @@ -49,10 +56,9 @@ dot4(const float *a, const float *b) a[3]*b[3]); } - - static INLINE unsigned -compute_clipmask_gl(const float *clip, /*const*/ float plane[][4], unsigned nr) +compute_clipmask_gl(const float *clip, /*const*/ float plane[][4], unsigned nr, + boolean clip_depth) { unsigned mask = 0x0; unsigned i; @@ -69,8 +75,10 @@ compute_clipmask_gl(const float *clip, /*const*/ float plane[][4], unsigned nr) if ( clip[0] + clip[3] < 0) mask |= (1<<1); if (-clip[1] + clip[3] < 0) mask |= (1<<2); if ( clip[1] + clip[3] < 0) mask |= (1<<3); - if ( clip[2] + clip[3] < 0) mask |= (1<<4); /* match mesa clipplane numbering - for now */ - if (-clip[2] + clip[3] < 0) mask |= (1<<5); /* match mesa clipplane numbering - for now */ + if (clip_depth) { + if ( clip[2] + clip[3] < 0) mask |= (1<<4); /* match mesa clipplane numbering - for now */ + if (-clip[2] + clip[3] < 0) mask |= (1<<5); /* match mesa clipplane numbering - for now */ + } /* Followed by any remaining ones: */ @@ -103,6 +111,7 @@ static boolean post_vs_cliptest_viewport_gl( struct pt_post_vs *pvs, for (j = 0; j < info->count; j++) { float *position = out->data[pos]; + initialize_vertex_header(out); #if 0 debug_printf("%d) io = %p, data = %p = [%f, %f, %f, %f]\n", j, out, position, position[0], position[1], position[2], position[3]); @@ -114,9 +123,11 @@ static boolean post_vs_cliptest_viewport_gl( struct pt_post_vs *pvs, out->clip[3] = position[3]; out->vertex_id = 0xffff; + /* Disable depth clipping if depth clamping is enabled. */ out->clipmask = compute_clipmask_gl(out->clip, pvs->draw->plane, - pvs->draw->nr_planes); + pvs->draw->nr_planes, + !pvs->draw->depth_clamp); clipped += out->clipmask; if (out->clipmask == 0) @@ -192,6 +203,7 @@ static boolean post_vs_viewport( struct pt_post_vs *pvs, for (j = 0; j < info->count; j++) { float *position = out->data[pos]; + initialize_vertex_header(out); /* Viewport mapping only, no cliptest/rhw divide */ position[0] = position[0] * scale[0] + trans[0]; @@ -211,7 +223,16 @@ static boolean post_vs_viewport( struct pt_post_vs *pvs, static boolean post_vs_none( struct pt_post_vs *pvs, struct draw_vertex_info *info ) { + struct vertex_header *out = info->verts; + unsigned j; + if (0) debug_printf("%s\n", __FUNCTION__); + /* just initialize the vertex_id in all headers */ + for (j = 0; j < info->count; j++) { + initialize_vertex_header(out); + + out = (struct vertex_header *)((char *)out + info->stride); + } return FALSE; } diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c index 5ea833032f..cd7bb7bf25 100644 --- a/src/gallium/auxiliary/draw/draw_pt_varray.c +++ b/src/gallium/auxiliary/draw/draw_pt_varray.c @@ -120,24 +120,27 @@ static void varray_fan_segment(struct varray_frontend *varray, #define FUNC varray_run #include "draw_pt_varray_tmp_linear.h" -static unsigned decompose_prim[PIPE_PRIM_POLYGON + 1] = { +static unsigned decompose_prim[PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY + 1] = { PIPE_PRIM_POINTS, PIPE_PRIM_LINES, PIPE_PRIM_LINE_STRIP, /* decomposed LINELOOP */ PIPE_PRIM_LINE_STRIP, PIPE_PRIM_TRIANGLES, PIPE_PRIM_TRIANGLE_STRIP, - PIPE_PRIM_TRIANGLE_FAN, + PIPE_PRIM_TRIANGLE_FAN, PIPE_PRIM_QUADS, PIPE_PRIM_QUAD_STRIP, - PIPE_PRIM_POLYGON + PIPE_PRIM_POLYGON, + PIPE_PRIM_LINES_ADJACENCY, + PIPE_PRIM_LINE_STRIP_ADJACENCY, + PIPE_PRIM_TRIANGLES_ADJACENCY, + PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY }; static void varray_prepare(struct draw_pt_front_end *frontend, unsigned in_prim, - unsigned out_prim, struct draw_pt_middle_end *middle, unsigned opt) { @@ -146,11 +149,13 @@ static void varray_prepare(struct draw_pt_front_end *frontend, varray->base.run = varray_run; varray->input_prim = in_prim; - varray->output_prim = decompose_prim[out_prim]; + assert(in_prim < Elements(decompose_prim)); + varray->output_prim = decompose_prim[in_prim]; varray->middle = middle; - middle->prepare(middle, varray->input_prim, - varray->output_prim, opt, &varray->driver_fetch_max ); + middle->prepare(middle, + varray->output_prim, + opt, &varray->driver_fetch_max ); /* check that the max is even */ assert((varray->driver_fetch_max & 1) == 0); diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c index 914c87a9dc..8ef94c3163 100644 --- a/src/gallium/auxiliary/draw/draw_pt_vcache.c +++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c @@ -41,6 +41,7 @@ #define FETCH_MAX 256 #define DRAW_MAX (16*1024) + struct vcache_frontend { struct draw_pt_front_end base; struct draw_context *draw; @@ -64,13 +65,13 @@ struct vcache_frontend { unsigned opt; }; + static INLINE void vcache_flush( struct vcache_frontend *vcache ) { if (vcache->middle_prim != vcache->output_prim) { vcache->middle_prim = vcache->output_prim; vcache->middle->prepare( vcache->middle, - vcache->input_prim, vcache->middle_prim, vcache->opt, &vcache->fetch_max ); @@ -89,12 +90,12 @@ vcache_flush( struct vcache_frontend *vcache ) vcache->draw_count = 0; } + static INLINE void vcache_check_flush( struct vcache_frontend *vcache ) { - if ( vcache->draw_count + 6 >= DRAW_MAX || - vcache->fetch_count + 4 >= FETCH_MAX ) - { + if (vcache->draw_count + 6 >= DRAW_MAX || + vcache->fetch_count + 4 >= FETCH_MAX) { vcache_flush( vcache ); } } @@ -146,6 +147,7 @@ vcache_triangle_flags( struct vcache_frontend *vcache, vcache_check_flush(vcache); } + static INLINE void vcache_line( struct vcache_frontend *vcache, unsigned i0, @@ -177,6 +179,7 @@ vcache_point( struct vcache_frontend *vcache, vcache_check_flush(vcache); } + static INLINE void vcache_quad( struct vcache_frontend *vcache, unsigned i0, @@ -196,6 +199,7 @@ vcache_quad( struct vcache_frontend *vcache, } } + static INLINE void vcache_ef_quad( struct vcache_frontend *vcache, unsigned i0, @@ -231,6 +235,7 @@ vcache_ef_quad( struct vcache_frontend *vcache, } } + /* At least for now, we're back to using a template include file for * this. The two paths aren't too different though - it may be * possible to reunify them. @@ -256,23 +261,23 @@ rebase_uint_elts( const unsigned *src, ushort *dest ) { unsigned i; - for (i = 0; i < count; i++) dest[i] = (ushort)(src[i] + delta); } + static INLINE void rebase_ushort_elts( const ushort *src, unsigned count, int delta, - ushort *dest ) + ushort *dest ) { unsigned i; - for (i = 0; i < count; i++) dest[i] = (ushort)(src[i] + delta); } + static INLINE void rebase_ubyte_elts( const ubyte *src, unsigned count, @@ -280,42 +285,39 @@ rebase_ubyte_elts( const ubyte *src, ushort *dest ) { unsigned i; - for (i = 0; i < count; i++) dest[i] = (ushort)(src[i] + delta); } - static INLINE void translate_uint_elts( const unsigned *src, unsigned count, ushort *dest ) { unsigned i; - for (i = 0; i < count; i++) dest[i] = (ushort)(src[i]); } + static INLINE void translate_ushort_elts( const ushort *src, unsigned count, ushort *dest ) { unsigned i; - for (i = 0; i < count; i++) dest[i] = (ushort)(src[i]); } + static INLINE void translate_ubyte_elts( const ubyte *src, unsigned count, ushort *dest ) { unsigned i; - for (i = 0; i < count; i++) dest[i] = (ushort)(src[i]); } @@ -336,6 +338,7 @@ format_from_get_elt( pt_elt_func get_elt ) } #endif + static INLINE void vcache_check_run( struct draw_pt_front_end *frontend, pt_elt_func get_elt, @@ -345,18 +348,46 @@ vcache_check_run( struct draw_pt_front_end *frontend, { struct vcache_frontend *vcache = (struct vcache_frontend *)frontend; struct draw_context *draw = vcache->draw; - unsigned min_index = draw->pt.user.min_index; - unsigned max_index = draw->pt.user.max_index; - unsigned index_size = draw->pt.user.eltSize; - unsigned fetch_count = max_index + 1 - min_index; + const unsigned min_index = draw->pt.user.min_index; + const unsigned max_index = draw->pt.user.max_index; + const unsigned index_size = draw->pt.user.eltSize; + unsigned fetch_count; const ushort *transformed_elts; ushort *storage = NULL; boolean ok = FALSE; + /* debug: verify indexes are in range [min_index, max_index] */ + if (0) { + unsigned i; + for (i = 0; i < draw_count; i++) { + if (index_size == 1) { + assert( ((const ubyte *) elts)[i] >= min_index); + assert( ((const ubyte *) elts)[i] <= max_index); + } + else if (index_size == 2) { + assert( ((const ushort *) elts)[i] >= min_index); + assert( ((const ushort *) elts)[i] <= max_index); + } + else { + assert(index_size == 4); + assert( ((const uint *) elts)[i] >= min_index); + assert( ((const uint *) elts)[i] <= max_index); + } + } + } + + /* Note: max_index is frequently 0xffffffff so we have to be sure + * that any arithmetic involving max_index doesn't overflow! + */ + if (max_index >= (unsigned) DRAW_PIPE_MAX_VERTICES) + goto fail; - if (0) debug_printf("fetch_count %d fetch_max %d draw_count %d\n", fetch_count, - vcache->fetch_max, - draw_count); + fetch_count = max_index + 1 - min_index; + + if (0) + debug_printf("fetch_count %d fetch_max %d draw_count %d\n", fetch_count, + vcache->fetch_max, + draw_count); if (elt_bias + max_index >= DRAW_PIPE_MAX_VERTICES || fetch_count >= UNDEFINED_VERTEX_ID || @@ -368,23 +399,19 @@ vcache_check_run( struct draw_pt_front_end *frontend, if (vcache->middle_prim != vcache->input_prim) { vcache->middle_prim = vcache->input_prim; vcache->middle->prepare( vcache->middle, - vcache->input_prim, vcache->middle_prim, vcache->opt, &vcache->fetch_max ); } - assert((elt_bias >= 0 && min_index + elt_bias >= min_index) || (elt_bias < 0 && min_index + elt_bias < min_index)); if (min_index == 0 && - index_size == 2) - { + index_size == 2) { transformed_elts = (const ushort *)elts; } - else - { + else { storage = MALLOC( draw_count * sizeof(ushort) ); if (!storage) goto fail; @@ -419,23 +446,23 @@ vcache_check_run( struct draw_pt_front_end *frontend, switch(index_size) { case 1: rebase_ubyte_elts( (const ubyte *)elts, - draw_count, - 0 - (int)min_index, - storage ); + draw_count, + 0 - (int)min_index, + storage ); break; case 2: rebase_ushort_elts( (const ushort *)elts, - draw_count, - 0 - (int)min_index, - storage ); + draw_count, + 0 - (int)min_index, + storage ); break; case 4: rebase_uint_elts( (const uint *)elts, - draw_count, - 0 - (int)min_index, - storage ); + draw_count, + 0 - (int)min_index, + storage ); break; default: @@ -462,7 +489,7 @@ vcache_check_run( struct draw_pt_front_end *frontend, debug_printf("failed to execute atomic draw elts for %d/%d, splitting up\n", fetch_count, draw_count); - fail: +fail: vcache_run( frontend, get_elt, elts, elt_bias, draw_count ); } @@ -472,23 +499,26 @@ vcache_check_run( struct draw_pt_front_end *frontend, static void vcache_prepare( struct draw_pt_front_end *frontend, unsigned in_prim, - unsigned out_prim, struct draw_pt_middle_end *middle, unsigned opt ) { struct vcache_frontend *vcache = (struct vcache_frontend *)frontend; - if (opt & PT_PIPELINE) - { + if (opt & PT_PIPELINE) { vcache->base.run = vcache_run_extras; } - else - { + else { vcache->base.run = vcache_check_run; } + /* VCache will always emit the reduced version of its input + * primitive, ie STRIP/FANS become TRIS, etc. + * + * This is not to be confused with what the GS might be up to, + * which is a separate issue. + */ vcache->input_prim = in_prim; - vcache->output_prim = u_reduced_prim(out_prim); + vcache->output_prim = u_reduced_prim(in_prim); vcache->middle = middle; vcache->opt = opt; @@ -496,12 +526,13 @@ vcache_prepare( struct draw_pt_front_end *frontend, /* Have to run prepare here, but try and guess a good prim for * doing so: */ - vcache->middle_prim = (opt & PT_PIPELINE) ? vcache->output_prim : vcache->input_prim; - middle->prepare( middle, vcache->input_prim, - vcache->middle_prim, opt, &vcache->fetch_max ); -} - + vcache->middle_prim = (opt & PT_PIPELINE) + ? vcache->output_prim : vcache->input_prim; + middle->prepare( middle, + vcache->middle_prim, + opt, &vcache->fetch_max ); +} static void @@ -512,6 +543,7 @@ vcache_finish( struct draw_pt_front_end *frontend ) vcache->middle = NULL; } + static void vcache_destroy( struct draw_pt_front_end *frontend ) { diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c index b9db886a24..57ea63fc06 100644 --- a/src/gallium/auxiliary/draw/draw_vs.c +++ b/src/gallium/auxiliary/draw/draw_vs.c @@ -98,6 +98,11 @@ draw_create_vertex_shader(struct draw_context *draw, vs = draw_create_vs_ppc( draw, shader ); #endif } +#if HAVE_LLVM + else { + vs = draw_create_vs_llvm(draw, shader); + } +#endif if (!vs) { vs = draw_create_vs_exec( draw, shader ); diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index 6c7e94db43..a731994523 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -165,7 +165,6 @@ draw_create_vs_ppc(struct draw_context *draw, const struct pipe_shader_state *templ); - struct draw_vs_varient_key; struct draw_vertex_shader; @@ -173,6 +172,11 @@ struct draw_vs_varient * draw_vs_create_varient_aos_sse( struct draw_vertex_shader *vs, const struct draw_vs_varient_key *key ); +#if HAVE_LLVM +struct draw_vertex_shader * +draw_create_vs_llvm(struct draw_context *draw, + const struct pipe_shader_state *state); +#endif /******************************************************************************** diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c new file mode 100644 index 0000000000..6c13df7913 --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c @@ -0,0 +1,120 @@ +/************************************************************************** + * + * Copyright 2010 VMware, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "util/u_math.h" +#include "util/u_memory.h" +#include "pipe/p_shader_tokens.h" + +#include "draw_private.h" +#include "draw_context.h" +#include "draw_vs.h" +#include "draw_llvm.h" + +#include "tgsi/tgsi_parse.h" +#include "tgsi/tgsi_scan.h" + +static void +vs_llvm_prepare(struct draw_vertex_shader *shader, + struct draw_context *draw) +{ + /*struct llvm_vertex_shader *evs = llvm_vertex_shader(shader);*/ +} + +static void +vs_llvm_run_linear( struct draw_vertex_shader *shader, + const float (*input)[4], + float (*output)[4], + const void *constants[PIPE_MAX_CONSTANT_BUFFERS], + unsigned count, + unsigned input_stride, + unsigned output_stride ) +{ + /* we should never get here since the entire pipeline is + * generated in draw_pt_fetch_shade_pipeline_llvm.c */ + debug_assert(0); +} + + +static void +vs_llvm_delete( struct draw_vertex_shader *dvs ) +{ + struct llvm_vertex_shader *shader = llvm_vertex_shader(dvs); + struct pipe_fence_handle *fence = NULL; + struct draw_llvm_variant_list_item *li; + struct pipe_context *pipe = dvs->draw->pipe; + + /* + * XXX: This might be not neccessary at all. + */ + pipe->flush(pipe, 0, &fence); + if (fence) { + pipe->screen->fence_finish(pipe->screen, fence, 0); + pipe->screen->fence_reference(pipe->screen, &fence, NULL); + } + + + li = first_elem(&shader->variants); + while(!at_end(&shader->variants, li)) { + struct draw_llvm_variant_list_item *next = next_elem(li); + draw_llvm_destroy_variant(li->base); + li = next; + } + + assert(shader->variants_cached == 0); + FREE((void*) dvs->state.tokens); + FREE( dvs ); +} + + +struct draw_vertex_shader * +draw_create_vs_llvm(struct draw_context *draw, + const struct pipe_shader_state *state) +{ + struct llvm_vertex_shader *vs = CALLOC_STRUCT( llvm_vertex_shader ); + + if (vs == NULL) + return NULL; + + /* we make a private copy of the tokens */ + vs->base.state.tokens = tgsi_dup_tokens(state->tokens); + if (!vs->base.state.tokens) { + FREE(vs); + return NULL; + } + + tgsi_scan_shader(state->tokens, &vs->base.info); + + vs->base.draw = draw; + vs->base.prepare = vs_llvm_prepare; + vs->base.run_linear = vs_llvm_run_linear; + vs->base.delete = vs_llvm_delete; + vs->base.create_varient = draw_vs_create_varient_generic; + + make_empty_list(&vs->variants); + + return &vs->base; +} |