diff options
-rw-r--r-- | src/gallium/drivers/radeonsi/si_compute.c | 8 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_pipe.c | 52 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_pipe.h | 6 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader.c | 66 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader.h | 13 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader_internal.h | 6 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c | 13 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_state_shaders.c | 28 |
8 files changed, 101 insertions, 91 deletions
diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c index 69c3dce0124..e95e79c7b46 100644 --- a/src/gallium/drivers/radeonsi/si_compute.c +++ b/src/gallium/drivers/radeonsi/si_compute.c @@ -86,13 +86,13 @@ static void si_create_compute_state_async(void *job, int thread_index) struct si_compute *program = (struct si_compute *)job; struct si_shader *shader = &program->shader; struct si_shader_selector sel; - LLVMTargetMachineRef tm; + struct si_compiler *compiler; struct pipe_debug_callback *debug = &program->compiler_ctx_state.debug; assert(!debug->debug_message || debug->async); assert(thread_index >= 0); - assert(thread_index < ARRAY_SIZE(program->screen->tm)); - tm = program->screen->tm[thread_index]; + assert(thread_index < ARRAY_SIZE(program->screen->compiler)); + compiler = &program->screen->compiler[thread_index]; memset(&sel, 0, sizeof(sel)); @@ -123,7 +123,7 @@ static void si_create_compute_state_async(void *job, int thread_index) program->uses_bindless_samplers = sel.info.uses_bindless_samplers; program->uses_bindless_images = sel.info.uses_bindless_images; - if (si_shader_create(program->screen, tm, &program->shader, debug)) { + if (si_shader_create(program->screen, compiler, &program->shader, debug)) { program->shader.compilation_failed = true; } else { bool scratch_enabled = shader->config.scratch_bytes_per_wave > 0; diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c index c5ee9f4b7b2..60bc189e4f1 100644 --- a/src/gallium/drivers/radeonsi/si_pipe.c +++ b/src/gallium/drivers/radeonsi/si_pipe.c @@ -102,6 +102,24 @@ static const struct debug_named_value debug_options[] = { DEBUG_NAMED_VALUE_END /* must be last */ }; +static void si_init_compiler(struct si_screen *sscreen, + struct si_compiler *compiler) +{ + enum ac_target_machine_options tm_options = + (sscreen->debug_flags & DBG(SI_SCHED) ? AC_TM_SISCHED : 0) | + (sscreen->info.chip_class >= GFX9 ? AC_TM_FORCE_ENABLE_XNACK : 0) | + (sscreen->info.chip_class < GFX9 ? AC_TM_FORCE_DISABLE_XNACK : 0) | + (!sscreen->llvm_has_working_vgpr_indexing ? AC_TM_PROMOTE_ALLOCA_TO_SCRATCH : 0); + + compiler->tm = ac_create_target_machine(sscreen->info.family, tm_options); +} + +static void si_destroy_compiler(struct si_compiler *compiler) +{ + if (compiler->tm) + LLVMDisposeTargetMachine(compiler->tm); +} + /* * pipe_context */ @@ -200,7 +218,7 @@ static void si_destroy_context(struct pipe_context *context) sctx->ws->fence_reference(&sctx->last_sdma_fence, NULL); r600_resource_reference(&sctx->eop_bug_scratch, NULL); - LLVMDisposeTargetMachine(sctx->tm); + si_destroy_compiler(&sctx->compiler); si_saved_cs_reference(&sctx->current_saved_cs, NULL); @@ -285,18 +303,6 @@ static void si_emit_string_marker(struct pipe_context *ctx, u_log_printf(sctx->log, "\nString marker: %*s\n", len, string); } -static LLVMTargetMachineRef -si_create_llvm_target_machine(struct si_screen *sscreen) -{ - enum ac_target_machine_options tm_options = - (sscreen->debug_flags & DBG(SI_SCHED) ? AC_TM_SISCHED : 0) | - (sscreen->info.chip_class >= GFX9 ? AC_TM_FORCE_ENABLE_XNACK : 0) | - (sscreen->info.chip_class < GFX9 ? AC_TM_FORCE_DISABLE_XNACK : 0) | - (!sscreen->llvm_has_working_vgpr_indexing ? AC_TM_PROMOTE_ALLOCA_TO_SCRATCH : 0); - - return ac_create_target_machine(sscreen->info.family, tm_options); -} - static void si_set_debug_callback(struct pipe_context *ctx, const struct pipe_debug_callback *cb) { @@ -549,7 +555,7 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen, sctx->scratch_waves = MAX2(32 * sscreen->info.num_good_compute_units, max_threads_per_block / 64); - sctx->tm = si_create_llvm_target_machine(sscreen); + si_init_compiler(sscreen, &sctx->compiler); /* Bindless handles. */ sctx->tex_handles = _mesa_hash_table_create(NULL, _mesa_hash_pointer, @@ -624,13 +630,11 @@ static void si_destroy_screen(struct pipe_screen* pscreen) util_queue_destroy(&sscreen->shader_compiler_queue); util_queue_destroy(&sscreen->shader_compiler_queue_low_priority); - for (i = 0; i < ARRAY_SIZE(sscreen->tm); i++) - if (sscreen->tm[i]) - LLVMDisposeTargetMachine(sscreen->tm[i]); + for (i = 0; i < ARRAY_SIZE(sscreen->compiler); i++) + si_destroy_compiler(&sscreen->compiler[i]); - for (i = 0; i < ARRAY_SIZE(sscreen->tm_low_priority); i++) - if (sscreen->tm_low_priority[i]) - LLVMDisposeTargetMachine(sscreen->tm_low_priority[i]); + for (i = 0; i < ARRAY_SIZE(sscreen->compiler_lowp); i++) + si_destroy_compiler(&sscreen->compiler_lowp[i]); /* Free shader parts. */ for (i = 0; i < ARRAY_SIZE(parts); i++) { @@ -837,9 +841,9 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws, */ num_threads = sysconf(_SC_NPROCESSORS_ONLN); num_threads = MAX2(1, num_threads - 1); - num_compiler_threads = MIN2(num_threads, ARRAY_SIZE(sscreen->tm)); + num_compiler_threads = MIN2(num_threads, ARRAY_SIZE(sscreen->compiler)); num_compiler_threads_lowprio = - MIN2(num_threads, ARRAY_SIZE(sscreen->tm_low_priority)); + MIN2(num_threads, ARRAY_SIZE(sscreen->compiler_lowp)); if (!util_queue_init(&sscreen->shader_compiler_queue, "si_shader", 32, num_compiler_threads, @@ -1003,9 +1007,9 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws, sscreen->debug_flags |= DBG_ALL_SHADERS; for (i = 0; i < num_compiler_threads; i++) - sscreen->tm[i] = si_create_llvm_target_machine(sscreen); + si_init_compiler(sscreen, &sscreen->compiler[i]); for (i = 0; i < num_compiler_threads_lowprio; i++) - sscreen->tm_low_priority[i] = si_create_llvm_target_machine(sscreen); + si_init_compiler(sscreen, &sscreen->compiler_lowp[i]); /* Create the auxiliary context. This must be done last. */ sscreen->aux_context = si_create_context(&sscreen->b, 0); diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h index 823509524d4..54c9b725fcb 100644 --- a/src/gallium/drivers/radeonsi/si_pipe.h +++ b/src/gallium/drivers/radeonsi/si_pipe.h @@ -530,12 +530,12 @@ struct si_screen { /* Use at most 3 normal compiler threads on quadcore and better. * Hyperthreaded CPUs report the number of threads, but we want * the number of cores. */ - LLVMTargetMachineRef tm[3]; /* used by the queue only */ + struct si_compiler compiler[3]; /* used by the queue only */ struct util_queue shader_compiler_queue_low_priority; /* Use at most 2 low priority threads on quadcore and better. * We want to minimize the impact on multithreaded Mesa. */ - LLVMTargetMachineRef tm_low_priority[2]; /* at most 2 threads */ + struct si_compiler compiler_lowp[2]; /* at most 2 threads */ }; struct si_blend_color { @@ -776,7 +776,7 @@ struct si_context { void *vs_blit_texcoord; struct si_screen *screen; struct pipe_debug_callback debug; - LLVMTargetMachineRef tm; /* only non-threaded compilation */ + struct si_compiler compiler; /* only non-threaded compilation */ struct si_shader_ctx_state fixed_func_tcs_shader; struct r600_resource *wait_mem_scratch; unsigned wait_mem_number; diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c index 9c27c5a0b3e..e8d08cd8e7f 100644 --- a/src/gallium/drivers/radeonsi/si_shader.c +++ b/src/gallium/drivers/radeonsi/si_shader.c @@ -77,7 +77,7 @@ enum si_arg_regfile { static void si_init_shader_ctx(struct si_shader_context *ctx, struct si_screen *sscreen, - LLVMTargetMachineRef tm); + struct si_compiler *compiler); static void si_llvm_emit_barrier(const struct lp_build_tgsi_action *action, struct lp_build_tgsi_context *bld_base, @@ -5640,7 +5640,7 @@ void si_shader_dump(struct si_screen *sscreen, const struct si_shader *shader, static int si_compile_llvm(struct si_screen *sscreen, struct ac_shader_binary *binary, struct si_shader_config *conf, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, LLVMModuleRef mod, struct pipe_debug_callback *debug, unsigned processor, @@ -5666,7 +5666,7 @@ static int si_compile_llvm(struct si_screen *sscreen, } if (!si_replace_shader(count, binary)) { - r = si_llvm_compile(mod, binary, tm, debug); + r = si_llvm_compile(mod, binary, compiler, debug); if (r) return r; } @@ -5718,7 +5718,7 @@ static void si_llvm_build_ret(struct si_shader_context *ctx, LLVMValueRef ret) /* Generate code for the hardware VS shader stage to go with a geometry shader */ struct si_shader * si_generate_gs_copy_shader(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader_selector *gs_selector, struct pipe_debug_callback *debug) { @@ -5749,7 +5749,7 @@ si_generate_gs_copy_shader(struct si_screen *sscreen, shader->selector = gs_selector; shader->is_gs_copy_shader = true; - si_init_shader_ctx(&ctx, sscreen, tm); + si_init_shader_ctx(&ctx, sscreen, compiler); ctx.shader = shader; ctx.type = PIPE_SHADER_VERTEX; @@ -5844,7 +5844,7 @@ si_generate_gs_copy_shader(struct si_screen *sscreen, si_llvm_optimize_module(&ctx); r = si_compile_llvm(sscreen, &ctx.shader->binary, - &ctx.shader->config, ctx.tm, + &ctx.shader->config, ctx.compiler, ctx.gallivm.module, debug, PIPE_SHADER_GEOMETRY, "GS Copy Shader"); @@ -5966,11 +5966,11 @@ static void si_dump_shader_key(unsigned processor, const struct si_shader *shade static void si_init_shader_ctx(struct si_shader_context *ctx, struct si_screen *sscreen, - LLVMTargetMachineRef tm) + struct si_compiler *compiler) { struct lp_build_tgsi_context *bld_base; - si_llvm_context_init(ctx, sscreen, tm); + si_llvm_context_init(ctx, sscreen, compiler); bld_base = &ctx->bld_base; bld_base->emit_fetch_funcs[TGSI_FILE_CONSTANT] = fetch_constant; @@ -6748,7 +6748,7 @@ static void si_build_wrapper_function(struct si_shader_context *ctx, } int si_compile_tgsi_shader(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader *shader, bool is_monolithic, struct pipe_debug_callback *debug) @@ -6768,7 +6768,7 @@ int si_compile_tgsi_shader(struct si_screen *sscreen, si_dump_streamout(&sel->so); } - si_init_shader_ctx(&ctx, sscreen, tm); + si_init_shader_ctx(&ctx, sscreen, compiler); si_llvm_context_set_tgsi(&ctx, shader); ctx.separate_prolog = !is_monolithic; @@ -6978,7 +6978,7 @@ int si_compile_tgsi_shader(struct si_screen *sscreen, LLVMPointerTypeKind); /* Compile to bytecode. */ - r = si_compile_llvm(sscreen, &shader->binary, &shader->config, tm, + r = si_compile_llvm(sscreen, &shader->binary, &shader->config, compiler, ctx.gallivm.module, debug, ctx.type, "TGSI shader"); si_llvm_dispose(&ctx); if (r) { @@ -7089,7 +7089,7 @@ si_get_shader_part(struct si_screen *sscreen, enum pipe_shader_type type, bool prolog, union si_shader_part_key *key, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct pipe_debug_callback *debug, void (*build)(struct si_shader_context *, union si_shader_part_key *), @@ -7114,7 +7114,7 @@ si_get_shader_part(struct si_screen *sscreen, struct si_shader shader = {}; struct si_shader_context ctx; - si_init_shader_ctx(&ctx, sscreen, tm); + si_init_shader_ctx(&ctx, sscreen, compiler); ctx.shader = &shader; ctx.type = type; @@ -7145,7 +7145,7 @@ si_get_shader_part(struct si_screen *sscreen, /* Compile. */ si_llvm_optimize_module(&ctx); - if (si_compile_llvm(sscreen, &result->binary, &result->config, tm, + if (si_compile_llvm(sscreen, &result->binary, &result->config, compiler, ctx.ac.module, debug, ctx.type, name)) { FREE(result); result = NULL; @@ -7335,7 +7335,7 @@ static void si_build_vs_prolog_function(struct si_shader_context *ctx, } static bool si_get_vs_prolog(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader *shader, struct pipe_debug_callback *debug, struct si_shader *main_part, @@ -7353,7 +7353,7 @@ static bool si_get_vs_prolog(struct si_screen *sscreen, shader->prolog = si_get_shader_part(sscreen, &sscreen->vs_prologs, - PIPE_SHADER_VERTEX, true, &prolog_key, tm, + PIPE_SHADER_VERTEX, true, &prolog_key, compiler, debug, si_build_vs_prolog_function, "Vertex Shader Prolog"); return shader->prolog != NULL; @@ -7363,11 +7363,11 @@ static bool si_get_vs_prolog(struct si_screen *sscreen, * Select and compile (or reuse) vertex shader parts (prolog & epilog). */ static bool si_shader_select_vs_parts(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader *shader, struct pipe_debug_callback *debug) { - return si_get_vs_prolog(sscreen, tm, shader, debug, shader, + return si_get_vs_prolog(sscreen, compiler, shader, debug, shader, &shader->key.part.vs.prolog); } @@ -7452,7 +7452,7 @@ static void si_build_tcs_epilog_function(struct si_shader_context *ctx, * Select and compile (or reuse) TCS parts (epilog). */ static bool si_shader_select_tcs_parts(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader *shader, struct pipe_debug_callback *debug) { @@ -7460,7 +7460,7 @@ static bool si_shader_select_tcs_parts(struct si_screen *sscreen, struct si_shader *ls_main_part = shader->key.part.tcs.ls->main_shader_part_ls; - if (!si_get_vs_prolog(sscreen, tm, shader, debug, ls_main_part, + if (!si_get_vs_prolog(sscreen, compiler, shader, debug, ls_main_part, &shader->key.part.tcs.ls_prolog)) return false; @@ -7474,7 +7474,7 @@ static bool si_shader_select_tcs_parts(struct si_screen *sscreen, shader->epilog = si_get_shader_part(sscreen, &sscreen->tcs_epilogs, PIPE_SHADER_TESS_CTRL, false, - &epilog_key, tm, debug, + &epilog_key, compiler, debug, si_build_tcs_epilog_function, "Tessellation Control Shader Epilog"); return shader->epilog != NULL; @@ -7484,7 +7484,7 @@ static bool si_shader_select_tcs_parts(struct si_screen *sscreen, * Select and compile (or reuse) GS parts (prolog). */ static bool si_shader_select_gs_parts(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader *shader, struct pipe_debug_callback *debug) { @@ -7493,7 +7493,7 @@ static bool si_shader_select_gs_parts(struct si_screen *sscreen, shader->key.part.gs.es->main_shader_part_es; if (shader->key.part.gs.es->type == PIPE_SHADER_VERTEX && - !si_get_vs_prolog(sscreen, tm, shader, debug, es_main_part, + !si_get_vs_prolog(sscreen, compiler, shader, debug, es_main_part, &shader->key.part.gs.vs_prolog)) return false; @@ -7509,7 +7509,7 @@ static bool si_shader_select_gs_parts(struct si_screen *sscreen, shader->prolog2 = si_get_shader_part(sscreen, &sscreen->gs_prologs, PIPE_SHADER_GEOMETRY, true, - &prolog_key, tm, debug, + &prolog_key, compiler, debug, si_build_gs_prolog_function, "Geometry Shader Prolog"); return shader->prolog2 != NULL; @@ -7897,7 +7897,7 @@ static void si_build_ps_epilog_function(struct si_shader_context *ctx, * Select and compile (or reuse) pixel shader parts (prolog & epilog). */ static bool si_shader_select_ps_parts(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader *shader, struct pipe_debug_callback *debug) { @@ -7912,7 +7912,7 @@ static bool si_shader_select_ps_parts(struct si_screen *sscreen, shader->prolog = si_get_shader_part(sscreen, &sscreen->ps_prologs, PIPE_SHADER_FRAGMENT, true, - &prolog_key, tm, debug, + &prolog_key, compiler, debug, si_build_ps_prolog_function, "Fragment Shader Prolog"); if (!shader->prolog) @@ -7925,7 +7925,7 @@ static bool si_shader_select_ps_parts(struct si_screen *sscreen, shader->epilog = si_get_shader_part(sscreen, &sscreen->ps_epilogs, PIPE_SHADER_FRAGMENT, false, - &epilog_key, tm, debug, + &epilog_key, compiler, debug, si_build_ps_epilog_function, "Fragment Shader Epilog"); if (!shader->epilog) @@ -8028,7 +8028,7 @@ static void si_fix_resource_usage(struct si_screen *sscreen, } } -int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm, +int si_shader_create(struct si_screen *sscreen, struct si_compiler *compiler, struct si_shader *shader, struct pipe_debug_callback *debug) { @@ -8046,7 +8046,7 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm, /* Monolithic shader (compiled as a whole, has many variants, * may take a long time to compile). */ - r = si_compile_tgsi_shader(sscreen, tm, shader, true, debug); + r = si_compile_tgsi_shader(sscreen, compiler, shader, true, debug); if (r) return r; } else { @@ -8086,21 +8086,21 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm, /* Select prologs and/or epilogs. */ switch (sel->type) { case PIPE_SHADER_VERTEX: - if (!si_shader_select_vs_parts(sscreen, tm, shader, debug)) + if (!si_shader_select_vs_parts(sscreen, compiler, shader, debug)) return -1; break; case PIPE_SHADER_TESS_CTRL: - if (!si_shader_select_tcs_parts(sscreen, tm, shader, debug)) + if (!si_shader_select_tcs_parts(sscreen, compiler, shader, debug)) return -1; break; case PIPE_SHADER_TESS_EVAL: break; case PIPE_SHADER_GEOMETRY: - if (!si_shader_select_gs_parts(sscreen, tm, shader, debug)) + if (!si_shader_select_gs_parts(sscreen, compiler, shader, debug)) return -1; break; case PIPE_SHADER_FRAGMENT: - if (!si_shader_select_ps_parts(sscreen, tm, shader, debug)) + if (!si_shader_select_ps_parts(sscreen, compiler, shader, debug)) return -1; /* Make sure we have at least as many VGPRs as there diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h index c26ccafdd69..8c479d638a9 100644 --- a/src/gallium/drivers/radeonsi/si_shader.h +++ b/src/gallium/drivers/radeonsi/si_shader.h @@ -311,11 +311,16 @@ enum { struct si_shader; +/* Per-thread persistent LLVM objects. */ +struct si_compiler { + LLVMTargetMachineRef tm; +}; + /* State of the context creating the shader object. */ struct si_compiler_ctx_state { /* Should only be used by si_init_shader_selector_async and * si_build_shader_variant if thread_index == -1 (non-threaded). */ - LLVMTargetMachineRef tm; + struct si_compiler *compiler; /* Used if thread_index == -1 or if debug.async is true. */ struct pipe_debug_callback debug; @@ -646,15 +651,15 @@ struct si_shader_part { /* si_shader.c */ struct si_shader * si_generate_gs_copy_shader(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader_selector *gs_selector, struct pipe_debug_callback *debug); int si_compile_tgsi_shader(struct si_screen *sscreen, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct si_shader *shader, bool is_monolithic, struct pipe_debug_callback *debug); -int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm, +int si_shader_create(struct si_screen *sscreen, struct si_compiler *compiler, struct si_shader *shader, struct pipe_debug_callback *debug); void si_shader_destroy(struct si_shader *shader); diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h index afcc14e38cc..0a347172d62 100644 --- a/src/gallium/drivers/radeonsi/si_shader_internal.h +++ b/src/gallium/drivers/radeonsi/si_shader_internal.h @@ -179,7 +179,7 @@ struct si_shader_context { /* CS */ int param_block_size; - LLVMTargetMachineRef tm; + struct si_compiler *compiler; /* Preloaded descriptors. */ LLVMValueRef esgs_ring; @@ -221,7 +221,7 @@ si_shader_context_from_abi(struct ac_shader_abi *abi) } unsigned si_llvm_compile(LLVMModuleRef M, struct ac_shader_binary *binary, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct pipe_debug_callback *debug); LLVMTypeRef tgsi2llvmtype(struct lp_build_tgsi_context *bld_base, @@ -236,7 +236,7 @@ LLVMValueRef si_llvm_bound_index(struct si_shader_context *ctx, void si_llvm_context_init(struct si_shader_context *ctx, struct si_screen *sscreen, - LLVMTargetMachineRef tm); + struct si_compiler *compiler); void si_llvm_context_set_tgsi(struct si_shader_context *ctx, struct si_shader *shader); diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c index acd7e0b2b2f..d0332a6078c 100644 --- a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c +++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c @@ -99,7 +99,7 @@ static void si_diagnostic_handler(LLVMDiagnosticInfoRef di, void *context) * @returns 0 for success, 1 for failure */ unsigned si_llvm_compile(LLVMModuleRef M, struct ac_shader_binary *binary, - LLVMTargetMachineRef tm, + struct si_compiler *compiler, struct pipe_debug_callback *debug) { struct si_llvm_diagnostics diag; @@ -119,8 +119,9 @@ unsigned si_llvm_compile(LLVMModuleRef M, struct ac_shader_binary *binary, LLVMContextSetDiagnosticHandler(llvm_ctx, si_diagnostic_handler, &diag); /* Compile IR*/ - mem_err = LLVMTargetMachineEmitToMemoryBuffer(tm, M, LLVMObjectFile, &err, - &out_buffer); + mem_err = LLVMTargetMachineEmitToMemoryBuffer(compiler->tm, M, + LLVMObjectFile, &err, + &out_buffer); /* Process Errors/Warnings */ if (mem_err) { @@ -992,7 +993,7 @@ static void emit_immediate(struct lp_build_tgsi_context *bld_base, void si_llvm_context_init(struct si_shader_context *ctx, struct si_screen *sscreen, - LLVMTargetMachineRef tm) + struct si_compiler *compiler) { struct lp_type type; @@ -1003,14 +1004,14 @@ void si_llvm_context_init(struct si_shader_context *ctx, */ memset(ctx, 0, sizeof(*ctx)); ctx->screen = sscreen; - ctx->tm = tm; + ctx->compiler = compiler; ctx->gallivm.context = LLVMContextCreate(); ctx->gallivm.module = LLVMModuleCreateWithNameInContext("tgsi", ctx->gallivm.context); LLVMSetTarget(ctx->gallivm.module, "amdgcn--"); - LLVMTargetDataRef data_layout = LLVMCreateTargetDataLayout(tm); + LLVMTargetDataRef data_layout = LLVMCreateTargetDataLayout(compiler->tm); char *data_layout_str = LLVMCopyStringRepOfTargetData(data_layout); LLVMSetDataLayout(ctx->gallivm.module, data_layout_str); LLVMDisposeTargetData(data_layout); diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c index f23ce098208..2e215b91ce8 100644 --- a/src/gallium/drivers/radeonsi/si_state_shaders.c +++ b/src/gallium/drivers/radeonsi/si_state_shaders.c @@ -1488,26 +1488,26 @@ static void si_build_shader_variant(struct si_shader *shader, { struct si_shader_selector *sel = shader->selector; struct si_screen *sscreen = sel->screen; - LLVMTargetMachineRef tm; + struct si_compiler *compiler; struct pipe_debug_callback *debug = &shader->compiler_ctx_state.debug; int r; if (thread_index >= 0) { if (low_priority) { - assert(thread_index < ARRAY_SIZE(sscreen->tm_low_priority)); - tm = sscreen->tm_low_priority[thread_index]; + assert(thread_index < ARRAY_SIZE(sscreen->compiler_lowp)); + compiler = &sscreen->compiler_lowp[thread_index]; } else { - assert(thread_index < ARRAY_SIZE(sscreen->tm)); - tm = sscreen->tm[thread_index]; + assert(thread_index < ARRAY_SIZE(sscreen->compiler)); + compiler = &sscreen->compiler[thread_index]; } if (!debug->async) debug = NULL; } else { assert(!low_priority); - tm = shader->compiler_ctx_state.tm; + compiler = shader->compiler_ctx_state.compiler; } - r = si_shader_create(sscreen, tm, shader, debug); + r = si_shader_create(sscreen, compiler, shader, debug); if (unlikely(r)) { PRINT_ERR("Failed to build shader variant (type=%u) %d\n", sel->type, r); @@ -1560,7 +1560,7 @@ static bool si_check_missing_main_part(struct si_screen *sscreen, main_part->key.as_es = key->as_es; main_part->key.as_ls = key->as_ls; - if (si_compile_tgsi_shader(sscreen, compiler_state->tm, + if (si_compile_tgsi_shader(sscreen, compiler_state->compiler, main_part, false, &compiler_state->debug) != 0) { FREE(main_part); @@ -1835,13 +1835,13 @@ static void si_init_shader_selector_async(void *job, int thread_index) { struct si_shader_selector *sel = (struct si_shader_selector *)job; struct si_screen *sscreen = sel->screen; - LLVMTargetMachineRef tm; + struct si_compiler *compiler; struct pipe_debug_callback *debug = &sel->compiler_ctx_state.debug; assert(!debug->debug_message || debug->async); assert(thread_index >= 0); - assert(thread_index < ARRAY_SIZE(sscreen->tm)); - tm = sscreen->tm[thread_index]; + assert(thread_index < ARRAY_SIZE(sscreen->compiler)); + compiler = &sscreen->compiler[thread_index]; /* Compile the main shader part for use with a prolog and/or epilog. * If this fails, the driver will try to compile a monolithic shader @@ -1879,7 +1879,7 @@ static void si_init_shader_selector_async(void *job, int thread_index) mtx_unlock(&sscreen->shader_cache_mutex); /* Compile the shader if it hasn't been loaded from the cache. */ - if (si_compile_tgsi_shader(sscreen, tm, shader, false, + if (si_compile_tgsi_shader(sscreen, compiler, shader, false, debug) != 0) { FREE(shader); FREE(ir_binary); @@ -1942,7 +1942,7 @@ static void si_init_shader_selector_async(void *job, int thread_index) /* The GS copy shader is always pre-compiled. */ if (sel->type == PIPE_SHADER_GEOMETRY) { - sel->gs_copy_shader = si_generate_gs_copy_shader(sscreen, tm, sel, debug); + sel->gs_copy_shader = si_generate_gs_copy_shader(sscreen, compiler, sel, debug); if (!sel->gs_copy_shader) { fprintf(stderr, "radeonsi: can't create GS copy shader\n"); return; @@ -3134,7 +3134,7 @@ bool si_update_shaders(struct si_context *sctx) old_ps ? old_ps->key.part.ps.epilog.spi_shader_col_format : 0; int r; - compiler_state.tm = sctx->tm; + compiler_state.compiler = &sctx->compiler; compiler_state.debug = sctx->debug; compiler_state.is_debug_context = sctx->is_debug; |