octo/yuzu
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

shader: Ignore global memory ops on devices lacking int64 support

Browse Source
This commit is contained in:
ameerj
2021-07-04 00:34:53 -04:00
parent 55233c2861
commit 11f04f1022
8 changed files with 79 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/shader_recompiler/backend/glsl/emit_context.cpp
View File

@@ -378,7 +378,7 @@ void EmitContext::SetupExtensions() {
if (info.uses_shadow_lod && profile.support_gl_texture_shadow_lod) { if (info.uses_shadow_lod && profile.support_gl_texture_shadow_lod) {
header += "#extension GL_EXT_texture_shadow_lod : enable\n"; header += "#extension GL_EXT_texture_shadow_lod : enable\n";
} }
if (info.uses_int64) { if (info.uses_int64 && profile.support_int64) {
header += "#extension GL_ARB_gpu_shader_int64 : enable\n"; header += "#extension GL_ARB_gpu_shader_int64 : enable\n";
} }
if (info.uses_int64_bit_atomics) { if (info.uses_int64_bit_atomics) {
@@ -402,7 +402,7 @@ void EmitContext::SetupExtensions() {
info.uses_subgroup_shuffles || info.uses_fswzadd) { info.uses_subgroup_shuffles || info.uses_fswzadd) {
header += "#extension GL_ARB_shader_ballot : enable\n" header += "#extension GL_ARB_shader_ballot : enable\n"
"#extension GL_ARB_shader_group_vote : enable\n"; "#extension GL_ARB_shader_group_vote : enable\n";
if (!info.uses_int64) { if (!info.uses_int64 && profile.support_int64) {
header += "#extension GL_ARB_gpu_shader_int64 : enable\n"; header += "#extension GL_ARB_gpu_shader_int64 : enable\n";
} }
if (profile.support_gl_warp_intrinsics) { if (profile.support_gl_warp_intrinsics) {
@@ -539,7 +539,7 @@ void EmitContext::DefineHelperFunctions() {
if (info.uses_atomic_s32_max) { if (info.uses_atomic_s32_max) {
header += "uint CasMaxS32(uint op_a,uint op_b){return uint(max(int(op_a),int(op_b)));}"; header += "uint CasMaxS32(uint op_a,uint op_b){return uint(max(int(op_a),int(op_b)));}";
} }
if (info.uses_global_memory) { if (info.uses_global_memory && profile.support_int64) {
header += DefineGlobalMemoryFunctions(); header += DefineGlobalMemoryFunctions();
} }
if (info.loads_indexed_attributes) { if (info.loads_indexed_attributes) {

34
src/shader_recompiler/backend/glsl/emit_glsl_memory.cpp
View File

@@ -7,6 +7,7 @@
#include "shader_recompiler/backend/glsl/emit_context.h" #include "shader_recompiler/backend/glsl/emit_context.h"
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h" #include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/frontend/ir/value.h" #include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::GLSL { namespace Shader::Backend::GLSL {
namespace { namespace {
@@ -38,15 +39,27 @@ void EmitLoadGlobalS16(EmitContext&) {
} }
void EmitLoadGlobal32(EmitContext& ctx, IR::Inst& inst, std::string_view address) { void EmitLoadGlobal32(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
ctx.AddU32("{}=LoadGlobal32({});", inst, address); if (ctx.profile.support_int64) {
return ctx.AddU32("{}=LoadGlobal32({});", inst, address);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32("{}=0u;", inst);
} }
void EmitLoadGlobal64(EmitContext& ctx, IR::Inst& inst, std::string_view address) { void EmitLoadGlobal64(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
ctx.AddU32x2("{}=LoadGlobal64({});", inst, address); if (ctx.profile.support_int64) {
return ctx.AddU32x2("{}=LoadGlobal64({});", inst, address);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32x2("{}=uvec2(0);", inst);
} }
void EmitLoadGlobal128(EmitContext& ctx, IR::Inst& inst, std::string_view address) { void EmitLoadGlobal128(EmitContext& ctx, IR::Inst& inst, std::string_view address) {
ctx.AddU32x4("{}=LoadGlobal128({});", inst, address); if (ctx.profile.support_int64) {
return ctx.AddU32x4("{}=LoadGlobal128({});", inst, address);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
ctx.AddU32x4("{}=uvec4(0);", inst);
} }
void EmitWriteGlobalU8(EmitContext&) { void EmitWriteGlobalU8(EmitContext&) {
@@ -66,15 +79,24 @@ void EmitWriteGlobalS16(EmitContext&) {
} }
void EmitWriteGlobal32(EmitContext& ctx, std::string_view address, std::string_view value) { void EmitWriteGlobal32(EmitContext& ctx, std::string_view address, std::string_view value) {
ctx.Add("WriteGlobal32({},{});", address, value); if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal32({},{});", address, value);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
} }
void EmitWriteGlobal64(EmitContext& ctx, std::string_view address, std::string_view value) { void EmitWriteGlobal64(EmitContext& ctx, std::string_view address, std::string_view value) {
ctx.Add("WriteGlobal64({},{});", address, value); if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal64({},{});", address, value);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
} }
void EmitWriteGlobal128(EmitContext& ctx, std::string_view address, std::string_view value) { void EmitWriteGlobal128(EmitContext& ctx, std::string_view address, std::string_view value) {
ctx.Add("WriteGlobal128({},{});", address, value); if (ctx.profile.support_int64) {
return ctx.Add("WriteGlobal128({},{});", address, value);
}
LOG_WARNING(Shader_GLSL, "Int64 not supported, ignoring memory operation");
} }
void EmitLoadStorageU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, void EmitLoadStorageU8(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding,

2
src/shader_recompiler/backend/spirv/emit_context.cpp
View File

@@ -830,7 +830,7 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
} }
void EmitContext::DefineGlobalMemoryFunctions(const Info& info) { void EmitContext::DefineGlobalMemoryFunctions(const Info& info) {
if (!info.uses_global_memory) { if (!info.uses_global_memory || !profile.support_int64) {
return; return;
} }
using DefPtr = Id StorageDefinitions::*; using DefPtr = Id StorageDefinitions::*;

24
src/shader_recompiler/backend/spirv/emit_spirv_memory.cpp
View File

@@ -84,16 +84,28 @@ void EmitLoadGlobalS16(EmitContext&) {
} }
Id EmitLoadGlobal32(EmitContext& ctx, Id address) { Id EmitLoadGlobal32(EmitContext& ctx, Id address) {
if (ctx.profile.support_int64) {
return ctx.OpFunctionCall(ctx.U32[1], ctx.load_global_func_u32, address); return ctx.OpFunctionCall(ctx.U32[1], ctx.load_global_func_u32, address);
} }
LOG_WARNING(Shader_SPIRV, "Int64 not supported, ignoring memory operation");
return ctx.Const(0u);
}
Id EmitLoadGlobal64(EmitContext& ctx, Id address) { Id EmitLoadGlobal64(EmitContext& ctx, Id address) {
if (ctx.profile.support_int64) {
return ctx.OpFunctionCall(ctx.U32[2], ctx.load_global_func_u32x2, address); return ctx.OpFunctionCall(ctx.U32[2], ctx.load_global_func_u32x2, address);
} }
LOG_WARNING(Shader_SPIRV, "Int64 not supported, ignoring memory operation");
return ctx.Const(0u, 0u);
}
Id EmitLoadGlobal128(EmitContext& ctx, Id address) { Id EmitLoadGlobal128(EmitContext& ctx, Id address) {
if (ctx.profile.support_int64) {
return ctx.OpFunctionCall(ctx.U32[4], ctx.load_global_func_u32x4, address); return ctx.OpFunctionCall(ctx.U32[4], ctx.load_global_func_u32x4, address);
} }
LOG_WARNING(Shader_SPIRV, "Int64 not supported, ignoring memory operation");
return ctx.Const(0u, 0u, 0u, 0u);
}
void EmitWriteGlobalU8(EmitContext&) { void EmitWriteGlobalU8(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction"); throw NotImplementedException("SPIR-V Instruction");
@@ -112,15 +124,27 @@ void EmitWriteGlobalS16(EmitContext&) {
} }
void EmitWriteGlobal32(EmitContext& ctx, Id address, Id value) { void EmitWriteGlobal32(EmitContext& ctx, Id address, Id value) {
if (ctx.profile.support_int64) {
ctx.OpFunctionCall(ctx.void_id, ctx.write_global_func_u32, address, value); ctx.OpFunctionCall(ctx.void_id, ctx.write_global_func_u32, address, value);
return;
}
LOG_WARNING(Shader_SPIRV, "Int64 not supported, ignoring memory operation");
} }
void EmitWriteGlobal64(EmitContext& ctx, Id address, Id value) { void EmitWriteGlobal64(EmitContext& ctx, Id address, Id value) {
if (ctx.profile.support_int64) {
ctx.OpFunctionCall(ctx.void_id, ctx.write_global_func_u32x2, address, value); ctx.OpFunctionCall(ctx.void_id, ctx.write_global_func_u32x2, address, value);
return;
}
LOG_WARNING(Shader_SPIRV, "Int64 not supported, ignoring memory operation");
} }
void EmitWriteGlobal128(EmitContext& ctx, Id address, Id value) { void EmitWriteGlobal128(EmitContext& ctx, Id address, Id value) {
if (ctx.profile.support_int64) {
ctx.OpFunctionCall(ctx.void_id, ctx.write_global_func_u32x4, address, value); ctx.OpFunctionCall(ctx.void_id, ctx.write_global_func_u32x4, address, value);
return;
}
LOG_WARNING(Shader_SPIRV, "Int64 not supported, ignoring memory operation");
} }
Id EmitLoadStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) { Id EmitLoadStorageU8(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) {

28
src/shader_recompiler/frontend/ir/opcodes.inc
View File

@@ -71,20 +71,20 @@ OPCODE(UndefU32, U32,
OPCODE(UndefU64, U64, ) OPCODE(UndefU64, U64, )
// Memory operations // Memory operations
OPCODE(LoadGlobalU8, U32, U64, ) OPCODE(LoadGlobalU8, U32, Opaque, )
OPCODE(LoadGlobalS8, U32, U64, ) OPCODE(LoadGlobalS8, U32, Opaque, )
OPCODE(LoadGlobalU16, U32, U64, ) OPCODE(LoadGlobalU16, U32, Opaque, )
OPCODE(LoadGlobalS16, U32, U64, ) OPCODE(LoadGlobalS16, U32, Opaque, )
OPCODE(LoadGlobal32, U32, U64, ) OPCODE(LoadGlobal32, U32, Opaque, )
OPCODE(LoadGlobal64, U32x2, U64, ) OPCODE(LoadGlobal64, U32x2, Opaque, )
OPCODE(LoadGlobal128, U32x4, U64, ) OPCODE(LoadGlobal128, U32x4, Opaque, )
OPCODE(WriteGlobalU8, Void, U64, U32, ) OPCODE(WriteGlobalU8, Void, Opaque, U32, )
OPCODE(WriteGlobalS8, Void, U64, U32, ) OPCODE(WriteGlobalS8, Void, Opaque, U32, )
OPCODE(WriteGlobalU16, Void, U64, U32, ) OPCODE(WriteGlobalU16, Void, Opaque, U32, )
OPCODE(WriteGlobalS16, Void, U64, U32, ) OPCODE(WriteGlobalS16, Void, Opaque, U32, )
OPCODE(WriteGlobal32, Void, U64, U32, ) OPCODE(WriteGlobal32, Void, Opaque, U32, )
OPCODE(WriteGlobal64, Void, U64, U32x2, ) OPCODE(WriteGlobal64, Void, Opaque, U32x2, )
OPCODE(WriteGlobal128, Void, U64, U32x4, ) OPCODE(WriteGlobal128, Void, Opaque, U32x4, )
// Storage buffer operations // Storage buffer operations
OPCODE(LoadStorageU8, U32, U32, U32, ) OPCODE(LoadStorageU8, U32, U32, U32, )

1
src/shader_recompiler/profile.h
View File

@@ -15,6 +15,7 @@ struct Profile {
bool support_descriptor_aliasing{}; bool support_descriptor_aliasing{};
bool support_int8{}; bool support_int8{};
bool support_int16{}; bool support_int16{};
bool support_int64{};
bool support_vertex_instance_id{}; bool support_vertex_instance_id{};
bool support_float_controls{}; bool support_float_controls{};
bool support_separate_denorm_behavior{}; bool support_separate_denorm_behavior{};

1
src/video_core/renderer_opengl/gl_shader_cache.cpp
View File

@@ -168,6 +168,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
.support_descriptor_aliasing = false, .support_descriptor_aliasing = false,
.support_int8 = false, .support_int8 = false,
.support_int16 = false, .support_int16 = false,
.support_int64 = device.HasShaderInt64(),
.support_vertex_instance_id = true, .support_vertex_instance_id = true,
.support_float_controls = false, .support_float_controls = false,
.support_separate_denorm_behavior = false, .support_separate_denorm_behavior = false,

1
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View File

@@ -280,6 +280,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw
.support_descriptor_aliasing = true, .support_descriptor_aliasing = true,
.support_int8 = true, .support_int8 = true,
.support_int16 = device.IsShaderInt16Supported(), .support_int16 = device.IsShaderInt16Supported(),
.support_int64 = device.IsShaderInt64Supported(),
.support_vertex_instance_id = false, .support_vertex_instance_id = false,
.support_float_controls = true, .support_float_controls = true,
.support_separate_denorm_behavior = float_control.denormBehaviorIndependence == .support_separate_denorm_behavior = float_control.denormBehaviorIndependence ==