From 26a3bf5c726199d7664d5878ef1f73592e55caa7 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Thu, 28 Nov 2013 00:48:57 -0800 Subject: [PATCH] i965: Stop doing our optimization on a copy of the GLSL IR. The original intent was that we'd keep a driver-private copy, and there would be the normal copy for swrast to make use of without the tuning (or anything more invasive we might do) specific to i965. Only, we don't generate swrast code any more, because swrast can't render current shaders anyway. Thus, our private copy is rather a waste, and we can just do our backend-specific operations on the linked shader. Reviewed-by: Ian Romanick --- src/mesa/drivers/dri/i965/brw_context.h | 3 - src/mesa/drivers/dri/i965/brw_fs.cpp | 4 +- src/mesa/drivers/dri/i965/brw_shader.cpp | 55 ++++++++----------- src/mesa/drivers/dri/i965/brw_vec4.cpp | 4 +- .../drivers/dri/i965/brw_vec4_gs_visitor.cpp | 2 +- 5 files changed, 28 insertions(+), 40 deletions(-) diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index e19ffc07152..9c516469cf6 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -320,9 +320,6 @@ struct brw_shader { struct gl_shader base; bool compiled_once; - - /** Shader IR transformed for native compile, at link time. */ - struct exec_list *ir; }; /* Note: If adding fields that need anything besides a normal memcmp() for diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp index baf9220125c..3536cbeecc5 100644 --- a/src/mesa/drivers/dri/i965/brw_fs.cpp +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp @@ -3156,7 +3156,7 @@ fs_visitor::run() * functions called "main"). */ if (shader) { - foreach_list(node, &*shader->ir) { + foreach_list(node, &*shader->base.ir) { ir_instruction *ir = (ir_instruction *)node; base_ir = ir; this->result = reg_undef; @@ -3305,7 +3305,7 @@ brw_wm_fs_emit(struct brw_context *brw, struct brw_wm_compile *c, if (unlikely(INTEL_DEBUG & DEBUG_WM)) { if (prog) { printf("GLSL IR for native fragment shader %d:\n", prog->Name); - _mesa_print_ir(shader->ir, NULL); + _mesa_print_ir(shader->base.ir, NULL); printf("\n\n"); } else { printf("ARB_fragment_program %d ir for native fragment shader\n", diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp index cf9ca4bbf91..5752348be65 100644 --- a/src/mesa/drivers/dri/i965/brw_shader.cpp +++ b/src/mesa/drivers/dri/i965/brw_shader.cpp @@ -135,24 +135,18 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg) _mesa_copy_linked_program_data((gl_shader_stage) stage, shProg, prog); - void *mem_ctx = ralloc_context(NULL); bool progress; - if (shader->ir) - ralloc_free(shader->ir); - shader->ir = new(shader) exec_list; - clone_ir_list(mem_ctx, shader->ir, shader->base.ir); - /* lower_packing_builtins() inserts arithmetic instructions, so it * must precede lower_instructions(). */ - brw_lower_packing_builtins(brw, (gl_shader_stage) stage, shader->ir); - do_mat_op_to_vec(shader->ir); + brw_lower_packing_builtins(brw, (gl_shader_stage) stage, shader->base.ir); + do_mat_op_to_vec(shader->base.ir); const int bitfield_insert = brw->gen >= 7 ? BITFIELD_INSERT_TO_BFM_BFI : 0; const int lrp_to_arith = brw->gen < 6 ? LRP_TO_ARITH : 0; - lower_instructions(shader->ir, + lower_instructions(shader->base.ir, MOD_TO_FRACT | DIV_TO_MUL_RCP | SUB_TO_ADD_NEG | @@ -166,17 +160,17 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg) * if-statements need to be flattened. */ if (brw->gen < 6) - lower_if_to_cond_assign(shader->ir, 16); - - do_lower_texture_projection(shader->ir); - brw_lower_texture_gradients(brw, shader->ir); - do_vec_index_to_cond_assign(shader->ir); - lower_vector_insert(shader->ir, true); - brw_do_cubemap_normalize(shader->ir); - brw_do_lower_offset_arrays(shader->ir); - brw_do_lower_unnormalized_offset(shader->ir); - lower_noise(shader->ir); - lower_quadop_vector(shader->ir, false); + lower_if_to_cond_assign(shader->base.ir, 16); + + do_lower_texture_projection(shader->base.ir); + brw_lower_texture_gradients(brw, shader->base.ir); + do_vec_index_to_cond_assign(shader->base.ir); + lower_vector_insert(shader->base.ir, true); + brw_do_cubemap_normalize(shader->base.ir); + brw_do_lower_offset_arrays(shader->base.ir); + brw_do_lower_unnormalized_offset(shader->base.ir); + lower_noise(shader->base.ir); + lower_quadop_vector(shader->base.ir, false); bool input = true; bool output = stage == MESA_SHADER_FRAGMENT; @@ -184,7 +178,7 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg) bool uniform = false; bool lowered_variable_indexing = - lower_variable_index_to_cond_assign(shader->ir, + lower_variable_index_to_cond_assign(shader->base.ir, input, output, temp, uniform); if (unlikely(brw->perf_debug && lowered_variable_indexing)) { @@ -193,23 +187,23 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg) } /* FINISHME: Do this before the variable index lowering. */ - lower_ubo_reference(&shader->base, shader->ir); + lower_ubo_reference(&shader->base, shader->base.ir); do { progress = false; if (stage == MESA_SHADER_FRAGMENT) { - brw_do_channel_expressions(shader->ir); - brw_do_vector_splitting(shader->ir); + brw_do_channel_expressions(shader->base.ir); + brw_do_vector_splitting(shader->base.ir); } - progress = do_lower_jumps(shader->ir, true, true, + progress = do_lower_jumps(shader->base.ir, true, true, true, /* main return */ false, /* continue */ false /* loops */ ) || progress; - progress = do_common_optimization(shader->ir, true, true, 32, + progress = do_common_optimization(shader->base.ir, true, true, 32, &ctx->ShaderCompilerOptions[stage]) || progress; } while (progress); @@ -221,7 +215,7 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg) * too late. At that point, the values for the built-in uniforms won't * get sent to the shader. */ - foreach_list(node, shader->ir) { + foreach_list(node, shader->base.ir) { ir_variable *var = ((ir_instruction *) node)->as_variable(); if ((var == NULL) || (var->data.mode != ir_var_uniform) @@ -237,12 +231,9 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg) } } - validate_ir_tree(shader->ir); - - reparent_ir(shader->ir, shader->ir); - ralloc_free(mem_ctx); + validate_ir_tree(shader->base.ir); - do_set_program_inouts(shader->ir, prog, shader->base.Stage); + do_set_program_inouts(shader->base.ir, prog, shader->base.Stage); prog->SamplersUsed = shader->base.active_samplers; _mesa_update_shader_textures_used(shProg, prog); diff --git a/src/mesa/drivers/dri/i965/brw_vec4.cpp b/src/mesa/drivers/dri/i965/brw_vec4.cpp index fb57707513b..9d3735a1504 100644 --- a/src/mesa/drivers/dri/i965/brw_vec4.cpp +++ b/src/mesa/drivers/dri/i965/brw_vec4.cpp @@ -1540,7 +1540,7 @@ vec4_visitor::run() * functions called "main"). */ if (shader) { - visit_instructions(shader->ir); + visit_instructions(shader->base.ir); } else { emit_program_code(); } @@ -1654,7 +1654,7 @@ brw_vs_emit(struct brw_context *brw, if (unlikely(INTEL_DEBUG & DEBUG_VS)) { if (prog) { printf("GLSL IR for native vertex shader %d:\n", prog->Name); - _mesa_print_ir(shader->ir, NULL); + _mesa_print_ir(shader->base.ir, NULL); printf("\n\n"); } else { printf("ARB_vertex_program %d for native vertex shader\n", diff --git a/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp index c93459d28a7..5d5b1690477 100644 --- a/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp +++ b/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp @@ -546,7 +546,7 @@ brw_gs_emit(struct brw_context *brw, if (unlikely(INTEL_DEBUG & DEBUG_GS)) { printf("GLSL IR for native geometry shader %d:\n", prog->Name); - _mesa_print_ir(shader->ir, NULL); + _mesa_print_ir(shader->base.ir, NULL); printf("\n\n"); } -- 2.30.2