radv/ac: add initial pre-pass for shader info gathering
authorDave Airlie <airlied@redhat.com>
Mon, 17 Apr 2017 03:01:52 +0000 (13:01 +1000)
committerDave Airlie <airlied@redhat.com>
Tue, 18 Apr 2017 23:00:42 +0000 (09:00 +1000)
There is some radv specific info we need to gather from shaders
before we get into converting nir->llvm, so we can make
better decisions especially around user sgpr allocation.

This is just an initial placeholder to gather if sample positions
are required in the frag shader.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
src/amd/Makefile.sources
src/amd/common/ac_nir_to_llvm.c
src/amd/common/ac_nir_to_llvm.h
src/amd/common/ac_shader_info.c [new file with mode: 0644]
src/amd/common/ac_shader_info.h [new file with mode: 0644]
src/amd/vulkan/radv_cmd_buffer.c

index 816e7e4c9931b3f6f4a6cbb84ab4be1f16ce77c3..38dd48b8d3cfa16b4647fa01c556f342df30c6d9 100644 (file)
@@ -45,6 +45,8 @@ AMD_COMPILER_FILES = \
        common/ac_llvm_build.c \
        common/ac_llvm_build.h \
        common/ac_llvm_helper.cpp \
        common/ac_llvm_build.c \
        common/ac_llvm_build.h \
        common/ac_llvm_helper.cpp \
+       common/ac_shader_info.c \
+       common/ac_shader_info.h \
        common/ac_llvm_util.c \
        common/ac_llvm_util.h
 
        common/ac_llvm_util.c \
        common/ac_llvm_util.h
 
index 9bcd5f6db4620c28d53393c33ad4d32ca2e4332e..dbb3b67b511f92e713c7a8afdac1ab1722818e97 100644 (file)
@@ -30,7 +30,7 @@
 #include "../vulkan/radv_descriptor_set.h"
 #include "util/bitscan.h"
 #include <llvm-c/Transforms/Scalar.h>
 #include "../vulkan/radv_descriptor_set.h"
 #include "util/bitscan.h"
 #include <llvm-c/Transforms/Scalar.h>
-
+#include "ac_shader_info.h"
 enum radeon_llvm_calling_convention {
        RADEON_LLVM_AMDGPU_VS = 87,
        RADEON_LLVM_AMDGPU_GS = 88,
 enum radeon_llvm_calling_convention {
        RADEON_LLVM_AMDGPU_VS = 87,
        RADEON_LLVM_AMDGPU_GS = 88,
@@ -680,7 +680,8 @@ static void create_function(struct nir_to_llvm_context *ctx)
                arg_types[arg_idx++] = ctx->i32; // GS instance id
                break;
        case MESA_SHADER_FRAGMENT:
                arg_types[arg_idx++] = ctx->i32; // GS instance id
                break;
        case MESA_SHADER_FRAGMENT:
-               arg_types[arg_idx++] = ctx->i32; /* sample position offset */
+               if (ctx->shader_info->info.ps.needs_sample_positions)
+                       arg_types[arg_idx++] = ctx->i32; /* sample position offset */
                user_sgpr_count = arg_idx;
                arg_types[arg_idx++] = ctx->i32; /* prim mask */
                sgpr_count = arg_idx;
                user_sgpr_count = arg_idx;
                arg_types[arg_idx++] = ctx->i32; /* prim mask */
                sgpr_count = arg_idx;
@@ -845,9 +846,11 @@ static void create_function(struct nir_to_llvm_context *ctx)
                ctx->gs_invocation_id = LLVMGetParam(ctx->main_function, arg_idx++);
                break;
        case MESA_SHADER_FRAGMENT:
                ctx->gs_invocation_id = LLVMGetParam(ctx->main_function, arg_idx++);
                break;
        case MESA_SHADER_FRAGMENT:
-               set_userdata_location_shader(ctx, AC_UD_PS_SAMPLE_POS_OFFSET, user_sgpr_idx, 1);
-               user_sgpr_idx += 1;
-               ctx->sample_pos_offset = LLVMGetParam(ctx->main_function, arg_idx++);
+               if (ctx->shader_info->info.ps.needs_sample_positions) {
+                       set_userdata_location_shader(ctx, AC_UD_PS_SAMPLE_POS_OFFSET, user_sgpr_idx, 1);
+                       user_sgpr_idx += 1;
+                       ctx->sample_pos_offset = LLVMGetParam(ctx->main_function, arg_idx++);
+               }
                ctx->prim_mask = LLVMGetParam(ctx->main_function, arg_idx++);
                ctx->persp_sample = LLVMGetParam(ctx->main_function, arg_idx++);
                ctx->persp_center = LLVMGetParam(ctx->main_function, arg_idx++);
                ctx->prim_mask = LLVMGetParam(ctx->main_function, arg_idx++);
                ctx->persp_sample = LLVMGetParam(ctx->main_function, arg_idx++);
                ctx->persp_center = LLVMGetParam(ctx->main_function, arg_idx++);
@@ -3528,7 +3531,6 @@ static LLVMValueRef load_sample_position(struct nir_to_llvm_context *ctx,
        sample_id = LLVMBuildAdd(ctx->builder, sample_id, ctx->sample_pos_offset, "");
        result = ac_build_indexed_load(&ctx->ac, ptr, sample_id, false);
 
        sample_id = LLVMBuildAdd(ctx->builder, sample_id, ctx->sample_pos_offset, "");
        result = ac_build_indexed_load(&ctx->ac, ptr, sample_id, false);
 
-       ctx->shader_info->fs.uses_sample_positions = true;
        return result;
 }
 
        return result;
 }
 
@@ -5703,6 +5705,8 @@ LLVMModuleRef ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
 
        memset(shader_info, 0, sizeof(*shader_info));
 
 
        memset(shader_info, 0, sizeof(*shader_info));
 
+       ac_nir_shader_info_pass(nir, options, &shader_info->info);
+               
        LLVMSetTarget(ctx.module, options->supports_spill ? "amdgcn-mesa-mesa3d" : "amdgcn--");
 
        LLVMTargetDataRef data_layout = LLVMCreateTargetDataLayout(tm);
        LLVMSetTarget(ctx.module, options->supports_spill ? "amdgcn-mesa-mesa3d" : "amdgcn--");
 
        LLVMTargetDataRef data_layout = LLVMCreateTargetDataLayout(tm);
index 3d0b456f9f563da87af623dec695552ba1d3090d..7a4065ac5fabcf1b48f50a08052f1d7186410b40 100644 (file)
@@ -29,7 +29,7 @@
 #include "llvm-c/TargetMachine.h"
 #include "amd_family.h"
 #include "../vulkan/radv_descriptor_set.h"
 #include "llvm-c/TargetMachine.h"
 #include "amd_family.h"
 #include "../vulkan/radv_descriptor_set.h"
-
+#include "ac_shader_info.h"
 #include "shader_enums.h"
 struct ac_shader_binary;
 struct ac_shader_config;
 #include "shader_enums.h"
 struct ac_shader_binary;
 struct ac_shader_config;
@@ -138,6 +138,7 @@ struct ac_es_output_info {
 
 struct ac_shader_variant_info {
        struct ac_userdata_locations user_sgprs_locs;
 
 struct ac_shader_variant_info {
        struct ac_userdata_locations user_sgprs_locs;
+       struct ac_shader_info info;
        unsigned num_user_sgprs;
        unsigned num_input_sgprs;
        unsigned num_input_vgprs;
        unsigned num_user_sgprs;
        unsigned num_input_sgprs;
        unsigned num_input_vgprs;
@@ -166,7 +167,6 @@ struct ac_shader_variant_info {
                        bool force_persample;
                        bool prim_id_input;
                        bool layer_input;
                        bool force_persample;
                        bool prim_id_input;
                        bool layer_input;
-                       bool uses_sample_positions;
                } fs;
                struct {
                        unsigned block_size[3];
                } fs;
                struct {
                        unsigned block_size[3];
diff --git a/src/amd/common/ac_shader_info.c b/src/amd/common/ac_shader_info.c
new file mode 100644 (file)
index 0000000..85252fe
--- /dev/null
@@ -0,0 +1,61 @@
+/*
+ * Copyright © 2017 Red Hat
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 "nir.h"
+#include "ac_shader_info.h"
+#include "ac_nir_to_llvm.h"
+static void
+gather_intrinsic_info(nir_intrinsic_instr *instr, struct ac_shader_info *info)
+{
+       switch (instr->intrinsic) {
+       case nir_intrinsic_interp_var_at_sample:
+               info->ps.needs_sample_positions = true;
+               break;
+       default:
+               break;
+       }
+}
+
+static void
+gather_info_block(nir_block *block, struct ac_shader_info *info)
+{
+       nir_foreach_instr(instr, block) {
+               switch (instr->type) {
+               case nir_instr_type_intrinsic:
+                       gather_intrinsic_info(nir_instr_as_intrinsic(instr), info);
+                       break;
+               default:
+                       break;
+               }
+       }
+}
+
+void
+ac_nir_shader_info_pass(struct nir_shader *nir,
+                       const struct ac_nir_compiler_options *options,
+                       struct ac_shader_info *info)
+{
+       struct nir_function *func = (struct nir_function *)exec_list_get_head(&nir->functions);
+       nir_foreach_block(block, func->impl) {
+               gather_info_block(block, info);
+       }
+}
diff --git a/src/amd/common/ac_shader_info.h b/src/amd/common/ac_shader_info.h
new file mode 100644 (file)
index 0000000..5576c3b
--- /dev/null
@@ -0,0 +1,40 @@
+/*
+ * Copyright © 2017 Red Hat
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 AC_SHADER_INFO_H
+#define AC_SHADER_INFO_H
+struct nir_shader;
+struct ac_nir_compiler_options;
+
+/* a NIR pass to gather all the info needed to optimise the alloction patterns for the RADV user sgprs */
+
+struct ac_shader_info {
+       struct {
+               bool needs_sample_positions;
+       } ps;
+};
+
+void
+ac_nir_shader_info_pass(struct nir_shader *nir,
+                       const struct ac_nir_compiler_options *options,
+                       struct ac_shader_info *info);
+#endif
index 31d04e535dc0f45db92ea38510b265027c75e2ec..f3e5f8292411fdd67b4dec17f02ee2ae2cae23ce 100644 (file)
@@ -454,7 +454,7 @@ radv_update_multisample_state(struct radv_cmd_buffer *cmd_buffer,
 
        radv_cayman_emit_msaa_sample_locs(cmd_buffer->cs, num_samples);
 
 
        radv_cayman_emit_msaa_sample_locs(cmd_buffer->cs, num_samples);
 
-       if (pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.uses_sample_positions) {
+       if (pipeline->shaders[MESA_SHADER_FRAGMENT]->info.info.ps.needs_sample_positions) {
                uint32_t offset;
                struct ac_userdata_info *loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_FRAGMENT, AC_UD_PS_SAMPLE_POS_OFFSET);
                uint32_t base_reg = shader_stage_to_user_data_0(MESA_SHADER_FRAGMENT, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
                uint32_t offset;
                struct ac_userdata_info *loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_FRAGMENT, AC_UD_PS_SAMPLE_POS_OFFSET);
                uint32_t base_reg = shader_stage_to_user_data_0(MESA_SHADER_FRAGMENT, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));