diff options
Diffstat (limited to 'src/gallium/drivers')
-rw-r--r-- | src/gallium/drivers/radeon/radeon_llvm.h | 11 | ||||
-rw-r--r-- | src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c | 66 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader.c | 6 |
3 files changed, 47 insertions, 36 deletions
diff --git a/src/gallium/drivers/radeon/radeon_llvm.h b/src/gallium/drivers/radeon/radeon_llvm.h index 13f33363147..6086dd6864a 100644 --- a/src/gallium/drivers/radeon/radeon_llvm.h +++ b/src/gallium/drivers/radeon/radeon_llvm.h @@ -50,11 +50,6 @@ struct radeon_llvm_loop { LLVMBasicBlockRef endloop_block; }; -struct radeon_llvm_array { - struct tgsi_declaration_range range; - LLVMValueRef alloca; -}; - struct radeon_llvm_context { struct lp_build_tgsi_soa_context soa; @@ -101,7 +96,8 @@ struct radeon_llvm_context { unsigned loop_depth; unsigned loop_depth_max; - struct radeon_llvm_array *arrays; + struct tgsi_array_info *temp_arrays; + LLVMValueRef *temp_array_allocas; LLVMValueRef main_fn; LLVMTypeRef return_type; @@ -124,7 +120,8 @@ void radeon_llvm_emit_prepare_cube_coords(struct lp_build_tgsi_context *bld_base LLVMValueRef *derivs_arg); void radeon_llvm_context_init(struct radeon_llvm_context *ctx, - const char *triple); + const char *triple, + const struct tgsi_shader_info *info); void radeon_llvm_create_func(struct radeon_llvm_context *ctx, LLVMTypeRef *return_types, unsigned num_return_elems, diff --git a/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c b/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c index d8ab5b0f6d4..25210236467 100644 --- a/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c +++ b/src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c @@ -115,26 +115,26 @@ static LLVMValueRef emit_swizzle(struct lp_build_tgsi_context *bld_base, * Return the description of the array covering the given temporary register * index. */ -static const struct radeon_llvm_array * -get_temp_array(struct lp_build_tgsi_context *bld_base, - unsigned reg_index, - const struct tgsi_ind_register *reg) +static unsigned +get_temp_array_id(struct lp_build_tgsi_context *bld_base, + unsigned reg_index, + const struct tgsi_ind_register *reg) { struct radeon_llvm_context *ctx = radeon_llvm_context(bld_base); unsigned num_arrays = ctx->soa.bld_base.info->array_max[TGSI_FILE_TEMPORARY]; unsigned i; if (reg && reg->ArrayID > 0 && reg->ArrayID <= num_arrays) - return &ctx->arrays[reg->ArrayID - 1]; + return reg->ArrayID; for (i = 0; i < num_arrays; i++) { - const struct radeon_llvm_array *array = &ctx->arrays[i]; + const struct tgsi_array_info *array = &ctx->temp_arrays[i]; if (reg_index >= array->range.First && reg_index <= array->range.Last) - return array; + return i + 1; } - return NULL; + return 0; } static struct tgsi_declaration_range @@ -142,13 +142,13 @@ get_array_range(struct lp_build_tgsi_context *bld_base, unsigned File, unsigned reg_index, const struct tgsi_ind_register *reg) { + struct radeon_llvm_context *ctx = radeon_llvm_context(bld_base); struct tgsi_declaration_range range; if (File == TGSI_FILE_TEMPORARY) { - const struct radeon_llvm_array *array = - get_temp_array(bld_base, reg_index, reg); - if (array) - return array->range; + unsigned array_id = get_temp_array_id(bld_base, reg_index, reg); + if (array_id) + return ctx->temp_arrays[array_id - 1].range; } range.First = 0; @@ -184,25 +184,31 @@ get_pointer_into_array(struct radeon_llvm_context *ctx, unsigned reg_index, const struct tgsi_ind_register *reg_indirect) { - const struct radeon_llvm_array *array; + unsigned array_id; struct gallivm_state *gallivm = ctx->soa.bld_base.base.gallivm; LLVMBuilderRef builder = gallivm->builder; LLVMValueRef idxs[2]; LLVMValueRef index; + LLVMValueRef alloca; if (file != TGSI_FILE_TEMPORARY) return NULL; - array = get_temp_array(&ctx->soa.bld_base, reg_index, reg_indirect); - if (!array || !array->alloca) + array_id = get_temp_array_id(&ctx->soa.bld_base, reg_index, reg_indirect); + if (!array_id) return NULL; - index = emit_array_index(&ctx->soa, reg_indirect, reg_index - array->range.First); + alloca = ctx->temp_array_allocas[array_id - 1]; + if (!alloca) + return NULL; + + index = emit_array_index(&ctx->soa, reg_indirect, + reg_index - ctx->temp_arrays[array_id - 1].range.First); index = LLVMBuildMul(builder, index, lp_build_const_int32(gallivm, TGSI_NUM_CHANNELS), ""); index = LLVMBuildAdd(builder, index, lp_build_const_int32(gallivm, swizzle), ""); idxs[0] = ctx->soa.bld_base.uint_bld.zero; idxs[1] = index; - return LLVMBuildGEP(builder, array->alloca, idxs, 2, ""); + return LLVMBuildGEP(builder, alloca, idxs, 2, ""); } LLVMValueRef @@ -478,12 +484,8 @@ static void emit_declaration(struct lp_build_tgsi_context *bld_base, decl_size = 4 * ((last - first) + 1); if (decl->Declaration.Array) { unsigned id = decl->Array.ArrayID - 1; - if (!ctx->arrays) { - int size = bld_base->info->array_max[TGSI_FILE_TEMPORARY]; - ctx->arrays = CALLOC(size, sizeof(ctx->arrays[0])); - } - ctx->arrays[id].range = decl->Range; + ctx->temp_arrays[id].range = decl->Range; /* If the array has more than 16 elements, store it * in memory using an alloca that spans the entire @@ -507,7 +509,7 @@ static void emit_declaration(struct lp_build_tgsi_context *bld_base, array_alloca = LLVMBuildAlloca(builder, LLVMArrayType(bld_base->base.vec_type, decl_size), "array"); - ctx->arrays[id].alloca = array_alloca; + ctx->temp_array_allocas[id] = array_alloca; } } @@ -1727,7 +1729,8 @@ static void emit_rsq(const struct lp_build_tgsi_action *action, bld_base->base.one, sqrt); } -void radeon_llvm_context_init(struct radeon_llvm_context *ctx, const char *triple) +void radeon_llvm_context_init(struct radeon_llvm_context *ctx, const char *triple, + const struct tgsi_shader_info *info) { struct lp_type type; @@ -1746,6 +1749,15 @@ void radeon_llvm_context_init(struct radeon_llvm_context *ctx, const char *tripl struct lp_build_tgsi_context *bld_base = &ctx->soa.bld_base; + bld_base->info = info; + + if (info && info->array_max[TGSI_FILE_TEMPORARY] > 0) { + int size = info->array_max[TGSI_FILE_TEMPORARY]; + + ctx->temp_arrays = CALLOC(size, sizeof(ctx->temp_arrays[0])); + ctx->temp_array_allocas = CALLOC(size, sizeof(ctx->temp_array_allocas[0])); + } + type.floating = true; type.fixed = false; type.sign = true; @@ -1966,8 +1978,10 @@ void radeon_llvm_dispose(struct radeon_llvm_context *ctx) { LLVMDisposeModule(ctx->soa.bld_base.base.gallivm->module); LLVMContextDispose(ctx->soa.bld_base.base.gallivm->context); - FREE(ctx->arrays); - ctx->arrays = NULL; + FREE(ctx->temp_arrays); + ctx->temp_arrays = NULL; + FREE(ctx->temp_array_allocas); + ctx->temp_array_allocas = NULL; FREE(ctx->temps); ctx->temps = NULL; ctx->temps_count = 0; diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c index 2de20cb1bd0..c595ee00482 100644 --- a/src/gallium/drivers/radeonsi/si_shader.c +++ b/src/gallium/drivers/radeonsi/si_shader.c @@ -6522,7 +6522,9 @@ static void si_init_shader_ctx(struct si_shader_context *ctx, struct lp_build_tgsi_action tmpl = {}; memset(ctx, 0, sizeof(*ctx)); - radeon_llvm_context_init(&ctx->radeon_bld, "amdgcn--"); + radeon_llvm_context_init( + &ctx->radeon_bld, "amdgcn--", + (shader && shader->selector) ? &shader->selector->info : NULL); ctx->tm = tm; ctx->screen = sscreen; if (shader && shader->selector) @@ -6545,8 +6547,6 @@ static void si_init_shader_ctx(struct si_shader_context *ctx, ctx->v8i32 = LLVMVectorType(ctx->i32, 8); bld_base = &ctx->radeon_bld.soa.bld_base; - if (shader && shader->selector) - bld_base->info = &shader->selector->info; bld_base->emit_fetch_funcs[TGSI_FILE_CONSTANT] = fetch_constant; bld_base->op_actions[TGSI_OPCODE_INTERP_CENTROID] = interp_action; |