encounter
/
dawn-cmake
mirror of https://github.com/encounter/dawn-cmake.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Cache WGSL -> DXBC/DXIL compilation 

Bug: dawn:1480
Change-Id: I858111f62be457c2e7cd5017bbf4c10e76395e83
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/95340
Reviewed-by: Loko Kung <lokokung@google.com>
Commit-Queue: Austin Eng <enga@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
This commit is contained in:
Austin Eng 2022-08-05 20:09:07 +00:00 committed by Dawn LUCI CQ
parent 9443cebd53
commit a0df1384f2
14 changed files with 529 additions and 452 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
src/dawn/native/Blob.cpp
View File

@ -17,6 +17,7 @@
#include "dawn/common/Assert.h"
#include "dawn/common/Math.h"
#include "dawn/native/Blob.h"
#include "dawn/native/stream/Stream.h"
namespace dawn::native {
@ -99,4 +100,29 @@ void Blob::AlignTo(size_t alignment) {
*this = std::move(blob);
}
template <>
void stream::Stream<Blob>::Write(stream::Sink* s, const Blob& b) {
size_t size = b.Size();
StreamIn(s, size);
if (size > 0) {
void* ptr = s->GetSpace(size);
memcpy(ptr, b.Data(), size);
}
}
template <>
MaybeError stream::Stream<Blob>::Read(stream::Source* s, Blob* b) {
size_t size;
DAWN_TRY(StreamOut(s, &size));
if (size > 0) {
const void* ptr;
DAWN_TRY(s->Read(&ptr, size));
*b = CreateBlob(size);
memcpy(b->Data(), ptr, size);
} else {
*b = Blob();
}
return {};
}
} // namespace dawn::native

20
src/dawn/native/StreamImplTint.cpp
View File

@ -14,7 +14,8 @@
#include "dawn/native/stream/Stream.h"
#include "tint/tint.h"
#include "dawn/native/TintUtils.h"
#include "tint/writer/array_length_from_uniform_options.h"
namespace dawn::native {
@ -96,4 +97,21 @@ void stream::Stream<tint::transform::VertexAttributeDescriptor>::Write(
StreamIn(sink, attrib.format, attrib.offset, attrib.shader_location);
}
// static
template <>
void stream::Stream<tint::writer::ArrayLengthFromUniformOptions>::Write(
stream::Sink* sink,
const tint::writer::ArrayLengthFromUniformOptions& o) {
static_assert(offsetof(tint::writer::ArrayLengthFromUniformOptions, ubo_binding) == 0,
"Please update serialization for tint::writer::ArrayLengthFromUniformOptions");
static_assert(
offsetof(tint::writer::ArrayLengthFromUniformOptions, bindpoint_to_size_index) == 8,
"Please update serialization for tint::writer::ArrayLengthFromUniformOptions");
static_assert(
sizeof(tint::writer::ArrayLengthFromUniformOptions) ==
8 + sizeof(tint::writer::ArrayLengthFromUniformOptions::bindpoint_to_size_index),
"Please update serialization for tint::writer::ArrayLengthFromUniformOptions");
StreamIn(sink, o.ubo_binding, o.bindpoint_to_size_index);
}
} // namespace dawn::native

7
src/dawn/native/TintUtils.cpp
View File

@ -185,7 +185,10 @@ tint::transform::VertexPulling::Config BuildVertexPullingTransformConfig(
} // namespace dawn::native
bool std::less<tint::sem::BindingPoint>::operator()(const tint::sem::BindingPoint& a,
const tint::sem::BindingPoint& b) const {
namespace tint::sem {
bool operator<(const BindingPoint& a, const BindingPoint& b) {
return std::tie(a.group, a.binding) < std::tie(b.group, b.binding);
}
} // namespace tint::sem

11
src/dawn/native/TintUtils.h
View File

@ -49,10 +49,11 @@ tint::transform::VertexPulling::Config BuildVertexPullingTransformConfig(
} // namespace dawn::native
// std::less operator for std::map containing BindingPoint
template <>
struct std::less<tint::sem::BindingPoint> {
bool operator()(const tint::sem::BindingPoint& a, const tint::sem::BindingPoint& b) const;
};
namespace tint::sem {
// Defin operator< for std::map containing BindingPoint
bool operator<(const BindingPoint& a, const BindingPoint& b);
} // namespace tint::sem
#endif // SRC_DAWN_NATIVE_TINTUTILS_H_

12
src/dawn/native/d3d12/BlobD3D12.cpp
View File

@ -28,4 +28,16 @@ Blob CreateBlob(ComPtr<ID3DBlob> blob) {
});
}
Blob CreateBlob(ComPtr<IDxcBlob> blob) {
// Detach so the deleter callback can "own" the reference
IDxcBlob* ptr = blob.Detach();
return Blob::UnsafeCreateWithDeleter(reinterpret_cast<uint8_t*>(ptr->GetBufferPointer()),
ptr->GetBufferSize(), [=]() {
// Reattach and drop to delete it.
ComPtr<IDxcBlob> b;
b.Attach(ptr);
b = nullptr;
});
}
} // namespace dawn::native

1
src/dawn/native/d3d12/BlobD3D12.h
View File

@ -18,5 +18,6 @@
namespace dawn::native {
Blob CreateBlob(ComPtr<ID3DBlob> blob);
Blob CreateBlob(ComPtr<IDxcBlob> blob);
} // namespace dawn::native

786
src/dawn/native/d3d12/ShaderModuleD3D12.cpp
View File

@ -29,79 +29,41 @@
#include "dawn/common/Log.h"
#include "dawn/common/WindowsUtils.h"
#include "dawn/native/CacheKey.h"
#include "dawn/native/CacheRequest.h"
#include "dawn/native/Pipeline.h"
#include "dawn/native/TintUtils.h"
#include "dawn/native/d3d12/AdapterD3D12.h"
#include "dawn/native/d3d12/BackendD3D12.h"
#include "dawn/native/d3d12/BindGroupLayoutD3D12.h"
#include "dawn/native/d3d12/BlobD3D12.h"
#include "dawn/native/d3d12/D3D12Error.h"
#include "dawn/native/d3d12/DeviceD3D12.h"
#include "dawn/native/d3d12/PipelineLayoutD3D12.h"
#include "dawn/native/d3d12/PlatformFunctions.h"
#include "dawn/native/d3d12/UtilsD3D12.h"
#include "dawn/native/stream/BlobSource.h"
#include "dawn/native/stream/ByteVectorSink.h"
#include "dawn/platform/DawnPlatform.h"
#include "dawn/platform/tracing/TraceEvent.h"
#include "tint/tint.h"
namespace dawn::native::stream {
// Define no-op serializations for pD3DCompile, IDxcLibrary, and IDxcCompiler.
// These are output-only interfaces used to generate bytecode.
template <>
void Stream<IDxcLibrary*>::Write(Sink*, IDxcLibrary* const&) {}
template <>
void Stream<IDxcCompiler*>::Write(Sink*, IDxcCompiler* const&) {}
template <>
void Stream<pD3DCompile>::Write(Sink*, pD3DCompile const&) {}
} // namespace dawn::native::stream
namespace dawn::native::d3d12 {
namespace {
uint64_t GetD3DCompilerVersion() {
return D3D_COMPILER_VERSION;
}
struct CompareBindingPoint {
constexpr bool operator()(const tint::transform::BindingPoint& lhs,
const tint::transform::BindingPoint& rhs) const {
if (lhs.group != rhs.group) {
return lhs.group < rhs.group;
} else {
return lhs.binding < rhs.binding;
}
}
};
void StreamIn(std::stringstream& output, const tint::ast::Access& access) {
output << access;
}
void StreamIn(std::stringstream& output, const tint::transform::BindingPoint& binding_point) {
output << "(BindingPoint";
output << " group=" << binding_point.group;
output << " binding=" << binding_point.binding;
output << ")";
}
template <typename T, typename = typename std::enable_if<std::is_fundamental<T>::value>::type>
void StreamIn(std::stringstream& output, const T& val) {
output << val;
}
template <typename T>
void StreamIn(std::stringstream& output,
const std::unordered_map<tint::transform::BindingPoint, T>& map) {
output << "(map";
std::map<tint::transform::BindingPoint, T, CompareBindingPoint> sorted(map.begin(), map.end());
for (auto& [bindingPoint, value] : sorted) {
output << " ";
StreamIn(output, bindingPoint);
output << "=";
StreamIn(output, value);
}
output << ")";
}
void StreamIn(std::stringstream& output,
const tint::writer::ArrayLengthFromUniformOptions& arrayLengthFromUniform) {
output << "(ArrayLengthFromUniformOptions";
output << " ubo_binding=";
StreamIn(output, arrayLengthFromUniform.ubo_binding);
output << " bindpoint_to_size_index=";
StreamIn(output, arrayLengthFromUniform.bindpoint_to_size_index);
output << ")";
}
// 32 bit float has 7 decimal digits of precision so setting n to 8 should be enough
std::string FloatToStringWithPrecision(float v, std::streamsize n = 8) {
@ -130,253 +92,113 @@ std::string GetHLSLValueString(EntryPointMetadata::Override::Type dawnType,
constexpr char kSpecConstantPrefix[] = "WGSL_SPEC_CONSTANT_";
void GetOverridableConstantsDefines(
std::vector<std::pair<std::string, std::string>>* defineStrings,
const PipelineConstantEntries* pipelineConstantEntries,
const EntryPointMetadata::OverridesMap* shaderEntryPointConstants) {
using DefineStrings = std::vector<std::pair<std::string, std::string>>;
DefineStrings GetOverridableConstantsDefines(
const PipelineConstantEntries& pipelineConstantEntries,
const EntryPointMetadata::OverridesMap& shaderEntryPointConstants) {
DefineStrings defineStrings;
std::unordered_set<std::string> overriddenConstants;
// Set pipeline overridden values
for (const auto& [name, value] : *pipelineConstantEntries) {
for (const auto& [name, value] : pipelineConstantEntries) {
overriddenConstants.insert(name);
// This is already validated so `name` must exist
const auto& moduleConstant = shaderEntryPointConstants->at(name);
const auto& moduleConstant = shaderEntryPointConstants.at(name);
defineStrings->emplace_back(
defineStrings.emplace_back(
kSpecConstantPrefix + std::to_string(static_cast<int32_t>(moduleConstant.id)),
GetHLSLValueString(moduleConstant.type, nullptr, value));
}
// Set shader initialized default values
for (const auto& iter : *shaderEntryPointConstants) {
for (const auto& iter : shaderEntryPointConstants) {
const std::string& name = iter.first;
if (overriddenConstants.count(name) != 0) {
// This constant already has overridden value
continue;
}
const auto& moduleConstant = shaderEntryPointConstants->at(name);
const auto& moduleConstant = shaderEntryPointConstants.at(name);
// Uninitialized default values are okay since they ar only defined to pass
// compilation but not used
defineStrings->emplace_back(
defineStrings.emplace_back(
kSpecConstantPrefix + std::to_string(static_cast<int32_t>(moduleConstant.id)),
GetHLSLValueString(moduleConstant.type, &moduleConstant.defaultValue));
}
return defineStrings;
}
// The inputs to a shader compilation. These have been intentionally isolated from the
// device to help ensure that the pipeline cache key contains all inputs for compilation.
struct ShaderCompilationRequest {
enum Compiler { FXC, DXC };
enum class Compiler { FXC, DXC };
// Common inputs
Compiler compiler;
const tint::Program* program;
const char* entryPointName;
SingleShaderStage stage;
uint32_t compileFlags;
bool disableSymbolRenaming;
tint::transform::BindingRemapper::BindingPoints remappedBindingPoints;
tint::transform::BindingRemapper::AccessControls remappedAccessControls;
bool isRobustnessEnabled;
bool usesNumWorkgroups;
uint32_t numWorkgroupsRegisterSpace;
uint32_t numWorkgroupsShaderRegister;
tint::writer::ArrayLengthFromUniformOptions arrayLengthFromUniform;
std::vector<std::pair<std::string, std::string>> defineStrings;
#define HLSL_COMPILATION_REQUEST_MEMBERS(X) \
X(const tint::Program*, inputProgram) \
X(std::string_view, entryPointName) \
X(SingleShaderStage, stage) \
X(uint32_t, shaderModel) \
X(uint32_t, compileFlags) \
X(Compiler, compiler) \
X(uint64_t, compilerVersion) \
X(std::wstring_view, dxcShaderProfile) \
X(std::string_view, fxcShaderProfile) \
X(pD3DCompile, d3dCompile) \
X(IDxcLibrary*, dxcLibrary) \
X(IDxcCompiler*, dxcCompiler) \
X(uint32_t, firstIndexOffsetShaderRegister) \
X(uint32_t, firstIndexOffsetRegisterSpace) \
X(bool, usesNumWorkgroups) \
X(uint32_t, numWorkgroupsShaderRegister) \
X(uint32_t, numWorkgroupsRegisterSpace) \
X(DefineStrings, defineStrings) \
X(tint::transform::MultiplanarExternalTexture::BindingsMap, newBindingsMap) \
X(tint::writer::ArrayLengthFromUniformOptions, arrayLengthFromUniform) \
X(tint::transform::BindingRemapper::BindingPoints, remappedBindingPoints) \
X(tint::transform::BindingRemapper::AccessControls, remappedAccessControls) \
X(bool, disableSymbolRenaming) \
X(bool, isRobustnessEnabled) \
X(bool, disableWorkgroupInit) \
X(bool, dumpShaders)
// FXC/DXC common inputs
bool disableWorkgroupInit;
#define D3D_BYTECODE_COMPILATION_REQUEST_MEMBERS(X) \
X(bool, hasShaderFloat16Feature) \
X(uint32_t, compileFlags) \
X(Compiler, compiler) \
X(uint64_t, compilerVersion) \
X(std::wstring_view, dxcShaderProfile) \
X(std::string_view, fxcShaderProfile) \
X(pD3DCompile, d3dCompile) \
X(IDxcLibrary*, dxcLibrary) \
X(IDxcCompiler*, dxcCompiler) \
X(DefineStrings, defineStrings)
// FXC inputs
uint64_t fxcVersion;
struct HlslCompilationRequest {
DAWN_VISITABLE_MEMBERS(HLSL_COMPILATION_REQUEST_MEMBERS)
// DXC inputs
uint64_t dxcVersion;
const D3D12DeviceInfo* deviceInfo;
bool hasShaderFloat16Feature;
static ResultOrError<ShaderCompilationRequest> Create(
const char* entryPointName,
SingleShaderStage stage,
const PipelineLayout* layout,
uint32_t compileFlags,
const Device* device,
const tint::Program* program,
const EntryPointMetadata& entryPoint,
const ProgrammableStage& programmableStage) {
Compiler compiler;
uint64_t dxcVersion = 0;
if (device->IsToggleEnabled(Toggle::UseDXC)) {
compiler = Compiler::DXC;
DAWN_TRY_ASSIGN(dxcVersion,
ToBackend(device->GetAdapter())->GetBackend()->GetDXCompilerVersion());
} else {
compiler = Compiler::FXC;
}
using tint::transform::BindingPoint;
using tint::transform::BindingRemapper;
BindingRemapper::BindingPoints remappedBindingPoints;
BindingRemapper::AccessControls remappedAccessControls;
tint::writer::ArrayLengthFromUniformOptions arrayLengthFromUniform;
arrayLengthFromUniform.ubo_binding = {
layout->GetDynamicStorageBufferLengthsRegisterSpace(),
layout->GetDynamicStorageBufferLengthsShaderRegister()};
const BindingInfoArray& moduleBindingInfo = entryPoint.bindings;
for (BindGroupIndex group : IterateBitSet(layout->GetBindGroupLayoutsMask())) {
const BindGroupLayout* bgl = ToBackend(layout->GetBindGroupLayout(group));
const auto& groupBindingInfo = moduleBindingInfo[group];
// d3d12::BindGroupLayout packs the bindings per HLSL register-space. We modify
// the Tint AST to make the "bindings" decoration match the offset chosen by
// d3d12::BindGroupLayout so that Tint produces HLSL with the correct registers
// assigned to each interface variable.
for (const auto& [binding, bindingInfo] : groupBindingInfo) {
BindingIndex bindingIndex = bgl->GetBindingIndex(binding);
BindingPoint srcBindingPoint{static_cast<uint32_t>(group),
static_cast<uint32_t>(binding)};
BindingPoint dstBindingPoint{static_cast<uint32_t>(group),
bgl->GetShaderRegister(bindingIndex)};
if (srcBindingPoint != dstBindingPoint) {
remappedBindingPoints.emplace(srcBindingPoint, dstBindingPoint);
}
// Declaring a read-only storage buffer in HLSL but specifying a storage
// buffer in the BGL produces the wrong output. Force read-only storage
// buffer bindings to be treated as UAV instead of SRV. Internal storage
// buffer is a storage buffer used in the internal pipeline.
const bool forceStorageBufferAsUAV =
(bindingInfo.buffer.type == wgpu::BufferBindingType::ReadOnlyStorage &&
(bgl->GetBindingInfo(bindingIndex).buffer.type ==
wgpu::BufferBindingType::Storage ||
bgl->GetBindingInfo(bindingIndex).buffer.type ==
kInternalStorageBufferBinding));
if (forceStorageBufferAsUAV) {
remappedAccessControls.emplace(srcBindingPoint, tint::ast::Access::kReadWrite);
}
}
// Add arrayLengthFromUniform options
{
for (const auto& bindingAndRegisterOffset :
layout->GetDynamicStorageBufferLengthInfo()[group].bindingAndRegisterOffsets) {
BindingNumber binding = bindingAndRegisterOffset.binding;
uint32_t registerOffset = bindingAndRegisterOffset.registerOffset;
BindingPoint bindingPoint{static_cast<uint32_t>(group),
static_cast<uint32_t>(binding)};
// Get the renamed binding point if it was remapped.
auto it = remappedBindingPoints.find(bindingPoint);
if (it != remappedBindingPoints.end()) {
bindingPoint = it->second;
}
arrayLengthFromUniform.bindpoint_to_size_index.emplace(bindingPoint,
registerOffset);
}
}
}
ShaderCompilationRequest request;
request.compiler = compiler;
request.program = program;
request.entryPointName = entryPointName;
request.stage = stage;
request.compileFlags = compileFlags;
request.disableSymbolRenaming = device->IsToggleEnabled(Toggle::DisableSymbolRenaming);
request.remappedBindingPoints = std::move(remappedBindingPoints);
request.remappedAccessControls = std::move(remappedAccessControls);
request.isRobustnessEnabled = device->IsRobustnessEnabled();
request.disableWorkgroupInit = device->IsToggleEnabled(Toggle::DisableWorkgroupInit);
request.usesNumWorkgroups = entryPoint.usesNumWorkgroups;
request.numWorkgroupsShaderRegister = layout->GetNumWorkgroupsShaderRegister();
request.numWorkgroupsRegisterSpace = layout->GetNumWorkgroupsRegisterSpace();
request.arrayLengthFromUniform = std::move(arrayLengthFromUniform);
request.fxcVersion = compiler == Compiler::FXC ? GetD3DCompilerVersion() : 0;
request.dxcVersion = compiler == Compiler::DXC ? dxcVersion : 0;
request.deviceInfo = &device->GetDeviceInfo();
request.hasShaderFloat16Feature = device->IsFeatureEnabled(Feature::ShaderFloat16);
GetOverridableConstantsDefines(
&request.defineStrings, &programmableStage.constants,
&programmableStage.module->GetEntryPoint(programmableStage.entryPoint).overrides);
return std::move(request);
}
// TODO(dawn:1341): Move to use CacheKey instead of the vector.
ResultOrError<std::vector<uint8_t>> CreateCacheKey() const {
// Generate the WGSL from the Tint program so it's normalized.
// TODO(tint:1180): Consider using a binary serialization of the tint AST for a more
// compact representation.
auto result = tint::writer::wgsl::Generate(program, tint::writer::wgsl::Options{});
if (!result.success) {
std::ostringstream errorStream;
errorStream << "Tint WGSL failure:" << std::endl;
errorStream << "Generator: " << result.error << std::endl;
return DAWN_INTERNAL_ERROR(errorStream.str().c_str());
}
std::stringstream stream;
// Prefix the key with the type to avoid collisions from another type that could
// have the same key.
stream << static_cast<uint32_t>(CacheKey::Type::Shader);
stream << "\n";
stream << result.wgsl.length();
stream << "\n";
stream << result.wgsl;
stream << "\n";
stream << "(ShaderCompilationRequest";
stream << " compiler=" << compiler;
stream << " entryPointName=" << entryPointName;
stream << " stage=" << uint32_t(stage);
stream << " compileFlags=" << compileFlags;
stream << " disableSymbolRenaming=" << disableSymbolRenaming;
stream << " remappedBindingPoints=";
StreamIn(stream, remappedBindingPoints);
stream << " remappedAccessControls=";
StreamIn(stream, remappedAccessControls);
stream << " useNumWorkgroups=" << usesNumWorkgroups;
stream << " numWorkgroupsRegisterSpace=" << numWorkgroupsRegisterSpace;
stream << " numWorkgroupsShaderRegister=" << numWorkgroupsShaderRegister;
stream << " arrayLengthFromUniform=";
StreamIn(stream, arrayLengthFromUniform);
stream << " shaderModel=" << deviceInfo->shaderModel;
stream << " disableWorkgroupInit=" << disableWorkgroupInit;
stream << " isRobustnessEnabled=" << isRobustnessEnabled;
stream << " fxcVersion=" << fxcVersion;
stream << " dxcVersion=" << dxcVersion;
stream << " hasShaderFloat16Feature=" << hasShaderFloat16Feature;
stream << " defines={";
for (const auto& [name, value] : defineStrings) {
stream << " <" << name << "," << value << ">";
}
stream << " }";
stream << ")";
stream << "\n";
return std::vector<uint8_t>(std::istreambuf_iterator<char>{stream},
std::istreambuf_iterator<char>{});
friend void StreamIn(stream::Sink* sink, const HlslCompilationRequest& r) {
r.VisitAll([&](const auto&... members) { StreamIn(sink, members...); });
}
};
struct D3DBytecodeCompilationRequest {
DAWN_VISITABLE_MEMBERS(D3D_BYTECODE_COMPILATION_REQUEST_MEMBERS)
friend void StreamIn(stream::Sink* sink, const D3DBytecodeCompilationRequest& r) {
r.VisitAll([&](const auto&... members) { StreamIn(sink, members...); });
}
};
#define D3D_COMPILATION_REQUEST_MEMBERS(X) \
X(HlslCompilationRequest, hlsl) \
X(D3DBytecodeCompilationRequest, bytecode) \
X(CacheKey::UnsafeUnkeyedValue<dawn::platform::Platform*>, tracePlatform)
DAWN_MAKE_CACHE_REQUEST(D3DCompilationRequest, D3D_COMPILATION_REQUEST_MEMBERS);
#undef HLSL_COMPILATION_REQUEST_MEMBERS
#undef D3D_BYTECODE_COMPILATION_REQUEST_MEMBERS
#undef D3D_COMPILATION_REQUEST_MEMBERS
std::vector<const wchar_t*> GetDXCArguments(uint32_t compileFlags, bool enable16BitTypes) {
std::vector<const wchar_t*> arguments;
if (compileFlags & D3DCOMPILE_ENABLE_BACKWARDS_COMPATIBILITY) {
@ -429,25 +251,24 @@ std::vector<const wchar_t*> GetDXCArguments(uint32_t compileFlags, bool enable16
return arguments;
}
ResultOrError<ComPtr<IDxcBlob>> CompileShaderDXC(IDxcLibrary* dxcLibrary,
IDxcCompiler* dxcCompiler,
const ShaderCompilationRequest& request,
ResultOrError<ComPtr<IDxcBlob>> CompileShaderDXC(const D3DBytecodeCompilationRequest& r,
const std::string& entryPointName,
const std::string& hlslSource) {
ComPtr<IDxcBlobEncoding> sourceBlob;
DAWN_TRY(CheckHRESULT(dxcLibrary->CreateBlobWithEncodingOnHeapCopy(
DAWN_TRY(CheckHRESULT(r.dxcLibrary->CreateBlobWithEncodingFromPinned(
hlslSource.c_str(), hlslSource.length(), CP_UTF8, &sourceBlob),
"DXC create blob"));
std::wstring entryPointW;
DAWN_TRY_ASSIGN(entryPointW, ConvertStringToWstring(request.entryPointName));
DAWN_TRY_ASSIGN(entryPointW, ConvertStringToWstring(entryPointName));
std::vector<const wchar_t*> arguments =
GetDXCArguments(request.compileFlags, request.hasShaderFloat16Feature);
GetDXCArguments(r.compileFlags, r.hasShaderFloat16Feature);
// Build defines for overridable constants
std::vector<std::pair<std::wstring, std::wstring>> defineStrings;
defineStrings.reserve(request.defineStrings.size());
for (const auto& [name, value] : request.defineStrings) {
defineStrings.reserve(r.defineStrings.size());
for (const auto& [name, value] : r.defineStrings) {
defineStrings.emplace_back(UTF8ToWStr(name.c_str()), UTF8ToWStr(value.c_str()));
}
@ -458,11 +279,10 @@ ResultOrError<ComPtr<IDxcBlob>> CompileShaderDXC(IDxcLibrary* dxcLibrary,
}
ComPtr<IDxcOperationResult> result;
DAWN_TRY(
CheckHRESULT(dxcCompiler->Compile(sourceBlob.Get(), nullptr, entryPointW.c_str(),
request.deviceInfo->shaderProfiles[request.stage].c_str(),
arguments.data(), arguments.size(), dxcDefines.data(),
dxcDefines.size(), nullptr, &result),
DAWN_TRY(CheckHRESULT(
r.dxcCompiler->Compile(sourceBlob.Get(), nullptr, entryPointW.c_str(),
r.dxcShaderProfile.data(), arguments.data(), arguments.size(),
dxcDefines.data(), dxcDefines.size(), nullptr, &result),
"DXC compile"));
HRESULT hr;
@ -543,31 +363,18 @@ std::string CompileFlagsToStringFXC(uint32_t compileFlags) {
return result;
}
ResultOrError<ComPtr<ID3DBlob>> CompileShaderFXC(const PlatformFunctions* functions,
const ShaderCompilationRequest& request,
ResultOrError<ComPtr<ID3DBlob>> CompileShaderFXC(const D3DBytecodeCompilationRequest& r,
const std::string& entryPointName,
const std::string& hlslSource) {
const char* targetProfile = nullptr;
switch (request.stage) {
case SingleShaderStage::Vertex:
targetProfile = "vs_5_1";
break;
case SingleShaderStage::Fragment:
targetProfile = "ps_5_1";
break;
case SingleShaderStage::Compute:
targetProfile = "cs_5_1";
break;
}
ComPtr<ID3DBlob> compiledShader;
ComPtr<ID3DBlob> errors;
// Build defines for overridable constants
const D3D_SHADER_MACRO* pDefines = nullptr;
std::vector<D3D_SHADER_MACRO> fxcDefines;
if (request.defineStrings.size() > 0) {
fxcDefines.reserve(request.defineStrings.size() + 1);
for (const auto& [name, value] : request.defineStrings) {
if (r.defineStrings.size() > 0) {
fxcDefines.reserve(r.defineStrings.size() + 1);
for (const auto& [name, value] : r.defineStrings) {
fxcDefines.push_back({name.c_str(), value.c_str()});
}
// d3dCompile D3D_SHADER_MACRO* pDefines is a nullptr terminated array
@ -575,36 +382,49 @@ ResultOrError<ComPtr<ID3DBlob>> CompileShaderFXC(const PlatformFunctions* functi
pDefines = fxcDefines.data();
}
DAWN_INVALID_IF(
FAILED(functions->d3dCompile(hlslSource.c_str(), hlslSource.length(), nullptr, pDefines,
nullptr, request.entryPointName, targetProfile,
request.compileFlags, 0, &compiledShader, &errors)),
DAWN_INVALID_IF(FAILED(r.d3dCompile(hlslSource.c_str(), hlslSource.length(), nullptr, pDefines,
nullptr, entryPointName.c_str(), r.fxcShaderProfile.data(),
r.compileFlags, 0, &compiledShader, &errors)),
"D3D compile failed with: %s", static_cast<char*>(errors->GetBufferPointer()));
return std::move(compiledShader);
}
ResultOrError<std::string> TranslateToHLSL(dawn::platform::Platform* platform,
const ShaderCompilationRequest& request,
std::string* remappedEntryPointName) {
ResultOrError<std::string> TranslateToHLSL(
HlslCompilationRequest r,
CacheKey::UnsafeUnkeyedValue<dawn::platform::Platform*> tracePlatform,
std::string* remappedEntryPointName,
bool* usesVertexOrInstanceIndex) {
std::ostringstream errorStream;
errorStream << "Tint HLSL failure:" << std::endl;
tint::transform::Manager transformManager;
tint::transform::DataMap transformInputs;
if (request.isRobustnessEnabled) {
if (!r.newBindingsMap.empty()) {
transformManager.Add<tint::transform::MultiplanarExternalTexture>();
transformInputs.Add<tint::transform::MultiplanarExternalTexture::NewBindingPoints>(
std::move(r.newBindingsMap));
}
if (r.stage == SingleShaderStage::Vertex) {
transformManager.Add<tint::transform::FirstIndexOffset>();
transformInputs.Add<tint::transform::FirstIndexOffset::BindingPoint>(
r.firstIndexOffsetShaderRegister, r.firstIndexOffsetRegisterSpace);
}
if (r.isRobustnessEnabled) {
transformManager.Add<tint::transform::Robustness>();
}
transformManager.Add<tint::transform::BindingRemapper>();
transformManager.Add<tint::transform::SingleEntryPoint>();
transformInputs.Add<tint::transform::SingleEntryPoint::Config>(request.entryPointName);
transformInputs.Add<tint::transform::SingleEntryPoint::Config>(r.entryPointName.data());
transformManager.Add<tint::transform::Renamer>();
if (request.disableSymbolRenaming) {
if (r.disableSymbolRenaming) {
// We still need to rename HLSL reserved keywords
transformInputs.Add<tint::transform::Renamer::Config>(
tint::transform::Renamer::Target::kHlslKeywords);
@ -615,104 +435,92 @@ ResultOrError<std::string> TranslateToHLSL(dawn::platform::Platform* platform,
// different types.
const bool mayCollide = true;
transformInputs.Add<tint::transform::BindingRemapper::Remappings>(
std::move(request.remappedBindingPoints), std::move(request.remappedAccessControls),
mayCollide);
std::move(r.remappedBindingPoints), std::move(r.remappedAccessControls), mayCollide);
tint::Program transformedProgram;
tint::transform::DataMap transformOutputs;
{
TRACE_EVENT0(platform, General, "RunTransforms");
TRACE_EVENT0(tracePlatform.UnsafeGetValue(), General, "RunTransforms");
DAWN_TRY_ASSIGN(transformedProgram,
RunTransforms(&transformManager, request.program, transformInputs,
RunTransforms(&transformManager, r.inputProgram, transformInputs,
&transformOutputs, nullptr));
}
if (auto* data = transformOutputs.Get<tint::transform::Renamer::Data>()) {
auto it = data->remappings.find(request.entryPointName);
auto it = data->remappings.find(r.entryPointName.data());
if (it != data->remappings.end()) {
*remappedEntryPointName = it->second;
} else {
DAWN_INVALID_IF(!request.disableSymbolRenaming,
DAWN_INVALID_IF(!r.disableSymbolRenaming,
"Could not find remapped name for entry point.");
*remappedEntryPointName = request.entryPointName;
*remappedEntryPointName = r.entryPointName;
}
} else {
return DAWN_FORMAT_VALIDATION_ERROR("Transform output missing renamer data.");
}
if (r.stage == SingleShaderStage::Vertex) {
if (auto* data = transformOutputs.Get<tint::transform::FirstIndexOffset::Data>()) {
*usesVertexOrInstanceIndex = data->has_vertex_or_instance_index;
} else {
return DAWN_FORMAT_VALIDATION_ERROR(
"Transform output missing first index offset data.");
}
}
tint::writer::hlsl::Options options;
options.disable_workgroup_init = request.disableWorkgroupInit;
if (request.usesNumWorkgroups) {
options.root_constant_binding_point = tint::sem::BindingPoint{
request.numWorkgroupsRegisterSpace, request.numWorkgroupsShaderRegister};
options.disable_workgroup_init = r.disableWorkgroupInit;
if (r.usesNumWorkgroups) {
options.root_constant_binding_point =
tint::sem::BindingPoint{r.numWorkgroupsRegisterSpace, r.numWorkgroupsShaderRegister};
}
// TODO(dawn:549): HLSL generation outputs the indices into the
// array_length_from_uniform buffer that were actually used. When the blob cache can
// store more than compiled shaders, we should reflect these used indices and store
// them as well. This would allow us to only upload root constants that are actually
// read by the shader.
options.array_length_from_uniform = request.arrayLengthFromUniform;
TRACE_EVENT0(platform, General, "tint::writer::hlsl::Generate");
options.array_length_from_uniform = r.arrayLengthFromUniform;
TRACE_EVENT0(tracePlatform.UnsafeGetValue(), General, "tint::writer::hlsl::Generate");
auto result = tint::writer::hlsl::Generate(&transformedProgram, options);
DAWN_INVALID_IF(!result.success, "An error occured while generating HLSL: %s", result.error);
return std::move(result.hlsl);
}
template <typename F>
MaybeError CompileShader(dawn::platform::Platform* platform,
const PlatformFunctions* functions,
IDxcLibrary* dxcLibrary,
IDxcCompiler* dxcCompiler,
ShaderCompilationRequest&& request,
bool dumpShaders,
F&& DumpShadersEmitLog,
CompiledShader* compiledShader) {
ResultOrError<CompiledShader> CompileShader(D3DCompilationRequest r) {
CompiledShader compiledShader;
// Compile the source shader to HLSL.
std::string hlslSource;
std::string remappedEntryPoint;
DAWN_TRY_ASSIGN(hlslSource, TranslateToHLSL(platform, request, &remappedEntryPoint));
if (dumpShaders) {
std::ostringstream dumpedMsg;
dumpedMsg << "/* Dumped generated HLSL */" << std::endl << hlslSource;
DumpShadersEmitLog(WGPULoggingType_Info, dumpedMsg.str().c_str());
}
request.entryPointName = remappedEntryPoint.c_str();
switch (request.compiler) {
case ShaderCompilationRequest::Compiler::DXC: {
TRACE_EVENT0(platform, General, "CompileShaderDXC");
DAWN_TRY_ASSIGN(compiledShader->compiledDXCShader,
CompileShaderDXC(dxcLibrary, dxcCompiler, request, hlslSource));
DAWN_TRY_ASSIGN(compiledShader.hlslSource,
TranslateToHLSL(std::move(r.hlsl), r.tracePlatform, &remappedEntryPoint,
&compiledShader.usesVertexOrInstanceIndex));
switch (r.bytecode.compiler) {
case Compiler::DXC: {
TRACE_EVENT0(r.tracePlatform.UnsafeGetValue(), General, "CompileShaderDXC");
ComPtr<IDxcBlob> compiledDXCShader;
DAWN_TRY_ASSIGN(compiledDXCShader, CompileShaderDXC(r.bytecode, remappedEntryPoint,
compiledShader.hlslSource));
compiledShader.shaderBlob = CreateBlob(std::move(compiledDXCShader));
break;
}
case ShaderCompilationRequest::Compiler::FXC: {
TRACE_EVENT0(platform, General, "CompileShaderFXC");
DAWN_TRY_ASSIGN(compiledShader->compiledFXCShader,
CompileShaderFXC(functions, request, hlslSource));
case Compiler::FXC: {
TRACE_EVENT0(r.tracePlatform.UnsafeGetValue(), General, "CompileShaderFXC");
ComPtr<ID3DBlob> compiledFXCShader;
DAWN_TRY_ASSIGN(compiledFXCShader, CompileShaderFXC(r.bytecode, remappedEntryPoint,
compiledShader.hlslSource));
compiledShader.shaderBlob = CreateBlob(std::move(compiledFXCShader));
break;
}
}
if (dumpShaders && request.compiler == ShaderCompilationRequest::Compiler::FXC) {
std::ostringstream dumpedMsg;
dumpedMsg << "/* FXC compile flags */ " << std::endl
<< CompileFlagsToStringFXC(request.compileFlags) << std::endl;
dumpedMsg << "/* Dumped disassembled DXBC */" << std::endl;
ComPtr<ID3DBlob> disassembly;
if (FAILED(functions->d3dDisassemble(compiledShader->compiledFXCShader->GetBufferPointer(),
compiledShader->compiledFXCShader->GetBufferSize(), 0,
nullptr, &disassembly))) {
dumpedMsg << "D3D disassemble failed" << std::endl;
} else {
dumpedMsg << reinterpret_cast<const char*>(disassembly->GetBufferPointer());
// If dumpShaders is false, we don't need the HLSL for logging. Clear the contents so it
// isn't stored into the cache.
if (!r.hlsl.dumpShaders) {
compiledShader.hlslSource = "";
}
DumpShadersEmitLog(WGPULoggingType_Info, dumpedMsg.str().c_str());
}
return {};
return compiledShader;
}
} // anonymous namespace
@ -741,74 +549,202 @@ ResultOrError<CompiledShader> ShaderModule::Compile(const ProgrammableStage& pro
SingleShaderStage stage,
const PipelineLayout* layout,
uint32_t compileFlags) {
TRACE_EVENT0(GetDevice()->GetPlatform(), General, "ShaderModuleD3D12::Compile");
Device* device = ToBackend(GetDevice());
TRACE_EVENT0(device->GetPlatform(), General, "ShaderModuleD3D12::Compile");
ASSERT(!IsError());
ScopedTintICEHandler scopedICEHandler(GetDevice());
ScopedTintICEHandler scopedICEHandler(device);
const EntryPointMetadata& entryPoint = GetEntryPoint(programmableStage.entryPoint);
Device* device = ToBackend(GetDevice());
D3DCompilationRequest req = {};
req.tracePlatform = UnsafeUnkeyedValue(device->GetPlatform());
req.hlsl.shaderModel = device->GetDeviceInfo().shaderModel;
req.hlsl.disableSymbolRenaming = device->IsToggleEnabled(Toggle::DisableSymbolRenaming);
req.hlsl.isRobustnessEnabled = device->IsRobustnessEnabled();
req.hlsl.disableWorkgroupInit = device->IsToggleEnabled(Toggle::DisableWorkgroupInit);
req.hlsl.dumpShaders = device->IsToggleEnabled(Toggle::DumpShaders);
CompiledShader compiledShader = {};
tint::transform::Manager transformManager;
tint::transform::DataMap transformInputs;
const tint::Program* program = GetTintProgram();
tint::Program programAsValue;
auto externalTextureBindings = BuildExternalTextureTransformBindings(layout);
if (!externalTextureBindings.empty()) {
transformManager.Add<tint::transform::MultiplanarExternalTexture>();
transformInputs.Add<tint::transform::MultiplanarExternalTexture::NewBindingPoints>(
std::move(externalTextureBindings));
}
if (stage == SingleShaderStage::Vertex) {
transformManager.Add<tint::transform::FirstIndexOffset>();
transformInputs.Add<tint::transform::FirstIndexOffset::BindingPoint>(
layout->GetFirstIndexOffsetShaderRegister(),
layout->GetFirstIndexOffsetRegisterSpace());
}
tint::transform::DataMap transformOutputs;
DAWN_TRY_ASSIGN(programAsValue, RunTransforms(&transformManager, program, transformInputs,
&transformOutputs, nullptr));
program = &programAsValue;
if (stage == SingleShaderStage::Vertex) {
if (auto* data = transformOutputs.Get<tint::transform::FirstIndexOffset::Data>()) {
// TODO(dawn:549): Consider adding this information to the pipeline cache once we
// can store more than the shader blob in it.
compiledShader.usesVertexOrInstanceIndex = data->has_vertex_or_instance_index;
req.bytecode.hasShaderFloat16Feature = device->IsFeatureEnabled(Feature::ShaderFloat16);
req.bytecode.compileFlags = compileFlags;
req.bytecode.defineStrings =
GetOverridableConstantsDefines(programmableStage.constants, entryPoint.overrides);
if (device->IsToggleEnabled(Toggle::UseDXC)) {
req.bytecode.compiler = Compiler::DXC;
req.bytecode.dxcLibrary = device->GetDxcLibrary().Get();
req.bytecode.dxcCompiler = device->GetDxcCompiler().Get();
DAWN_TRY_ASSIGN(req.bytecode.compilerVersion,
ToBackend(device->GetAdapter())->GetBackend()->GetDXCompilerVersion());
req.bytecode.dxcShaderProfile = device->GetDeviceInfo().shaderProfiles[stage];
} else {
req.bytecode.compiler = Compiler::FXC;
req.bytecode.d3dCompile = device->GetFunctions()->d3dCompile;
req.bytecode.compilerVersion = D3D_COMPILER_VERSION;
switch (stage) {
case SingleShaderStage::Vertex:
req.bytecode.fxcShaderProfile = "vs_5_1";
break;
case SingleShaderStage::Fragment:
req.bytecode.fxcShaderProfile = "ps_5_1";
break;
case SingleShaderStage::Compute:
req.bytecode.fxcShaderProfile = "cs_5_1";
break;
}
}
ShaderCompilationRequest request;
DAWN_TRY_ASSIGN(request,
ShaderCompilationRequest::Create(
programmableStage.entryPoint.c_str(), stage, layout, compileFlags, device,
program, GetEntryPoint(programmableStage.entryPoint), programmableStage));
using tint::transform::BindingPoint;
using tint::transform::BindingRemapper;
// TODO(dawn:1341): Add shader cache key generation and caching for the compiled shader.
DAWN_TRY(CompileShader(
device->GetPlatform(), device->GetFunctions(),
device->IsToggleEnabled(Toggle::UseDXC) ? device->GetDxcLibrary().Get() : nullptr,
device->IsToggleEnabled(Toggle::UseDXC) ? device->GetDxcCompiler().Get() : nullptr,
std::move(request), device->IsToggleEnabled(Toggle::DumpShaders),
[&](WGPULoggingType loggingType, const char* message) {
GetDevice()->EmitLog(loggingType, message);
},
&compiledShader));
return std::move(compiledShader);
BindingRemapper::BindingPoints remappedBindingPoints;
BindingRemapper::AccessControls remappedAccessControls;
tint::writer::ArrayLengthFromUniformOptions arrayLengthFromUniform;
arrayLengthFromUniform.ubo_binding = {layout->GetDynamicStorageBufferLengthsRegisterSpace(),
layout->GetDynamicStorageBufferLengthsShaderRegister()};
const BindingInfoArray& moduleBindingInfo = entryPoint.bindings;
for (BindGroupIndex group : IterateBitSet(layout->GetBindGroupLayoutsMask())) {
const BindGroupLayout* bgl = ToBackend(layout->GetBindGroupLayout(group));
const auto& groupBindingInfo = moduleBindingInfo[group];
// d3d12::BindGroupLayout packs the bindings per HLSL register-space. We modify
// the Tint AST to make the "bindings" decoration match the offset chosen by
// d3d12::BindGroupLayout so that Tint produces HLSL with the correct registers
// assigned to each interface variable.
for (const auto& [binding, bindingInfo] : groupBindingInfo) {
BindingIndex bindingIndex = bgl->GetBindingIndex(binding);
BindingPoint srcBindingPoint{static_cast<uint32_t>(group),
static_cast<uint32_t>(binding)};
BindingPoint dstBindingPoint{static_cast<uint32_t>(group),
bgl->GetShaderRegister(bindingIndex)};
if (srcBindingPoint != dstBindingPoint) {
remappedBindingPoints.emplace(srcBindingPoint, dstBindingPoint);
}
// Declaring a read-only storage buffer in HLSL but specifying a storage
// buffer in the BGL produces the wrong output. Force read-only storage
// buffer bindings to be treated as UAV instead of SRV. Internal storage
// buffer is a storage buffer used in the internal pipeline.
const bool forceStorageBufferAsUAV =
(bindingInfo.buffer.type == wgpu::BufferBindingType::ReadOnlyStorage &&
(bgl->GetBindingInfo(bindingIndex).buffer.type ==
wgpu::BufferBindingType::Storage ||
bgl->GetBindingInfo(bindingIndex).buffer.type == kInternalStorageBufferBinding));
if (forceStorageBufferAsUAV) {
remappedAccessControls.emplace(srcBindingPoint, tint::ast::Access::kReadWrite);
}
}
// Add arrayLengthFromUniform options
{
for (const auto& bindingAndRegisterOffset :
layout->GetDynamicStorageBufferLengthInfo()[group].bindingAndRegisterOffsets) {
BindingNumber binding = bindingAndRegisterOffset.binding;
uint32_t registerOffset = bindingAndRegisterOffset.registerOffset;
BindingPoint bindingPoint{static_cast<uint32_t>(group),
static_cast<uint32_t>(binding)};
// Get the renamed binding point if it was remapped.
auto it = remappedBindingPoints.find(bindingPoint);
if (it != remappedBindingPoints.end()) {
bindingPoint = it->second;
}
arrayLengthFromUniform.bindpoint_to_size_index.emplace(bindingPoint,
registerOffset);
}
}
}
req.hlsl.inputProgram = GetTintProgram();
req.hlsl.entryPointName = programmableStage.entryPoint.c_str();
req.hlsl.stage = stage;
req.hlsl.firstIndexOffsetShaderRegister = layout->GetFirstIndexOffsetShaderRegister();
req.hlsl.firstIndexOffsetRegisterSpace = layout->GetFirstIndexOffsetRegisterSpace();
req.hlsl.usesNumWorkgroups = entryPoint.usesNumWorkgroups;
req.hlsl.numWorkgroupsShaderRegister = layout->GetNumWorkgroupsShaderRegister();
req.hlsl.numWorkgroupsRegisterSpace = layout->GetNumWorkgroupsRegisterSpace();
req.hlsl.remappedBindingPoints = std::move(remappedBindingPoints);
req.hlsl.remappedAccessControls = std::move(remappedAccessControls);
req.hlsl.newBindingsMap = BuildExternalTextureTransformBindings(layout);
req.hlsl.arrayLengthFromUniform = std::move(arrayLengthFromUniform);
CacheResult<CompiledShader> compiledShader;
DAWN_TRY_LOAD_OR_RUN(compiledShader, device, std::move(req), CompiledShader::FromBlob,
CompileShader);
if (device->IsToggleEnabled(Toggle::DumpShaders)) {
std::ostringstream dumpedMsg;
dumpedMsg << "/* Dumped generated HLSL */" << std::endl
<< compiledShader->hlslSource << std::endl;
device->EmitLog(WGPULoggingType_Info, dumpedMsg.str().c_str());
if (device->IsToggleEnabled(Toggle::UseDXC)) {
dumpedMsg << "/* Dumped disassembled DXIL */" << std::endl;
D3D12_SHADER_BYTECODE code = compiledShader->GetD3D12ShaderBytecode();
ComPtr<IDxcBlobEncoding> dxcBlob;
ComPtr<IDxcBlobEncoding> disassembly;
if (FAILED(device->GetDxcLibrary()->CreateBlobWithEncodingFromPinned(
code.pShaderBytecode, code.BytecodeLength, 0, &dxcBlob)) ||
FAILED(device->GetDxcCompiler()->Disassemble(dxcBlob.Get(), &disassembly))) {
dumpedMsg << "DXC disassemble failed" << std::endl;
} else {
dumpedMsg << std::string_view(
static_cast<const char*>(disassembly->GetBufferPointer()),
disassembly->GetBufferSize());
}
} else {
dumpedMsg << "/* FXC compile flags */ " << std::endl
<< CompileFlagsToStringFXC(compileFlags) << std::endl;
dumpedMsg << "/* Dumped disassembled DXBC */" << std::endl;
ComPtr<ID3DBlob> disassembly;
D3D12_SHADER_BYTECODE code = compiledShader->GetD3D12ShaderBytecode();
if (FAILED(device->GetFunctions()->d3dDisassemble(
code.pShaderBytecode, code.BytecodeLength, 0, nullptr, &disassembly))) {
dumpedMsg << "D3D disassemble failed" << std::endl;
} else {
dumpedMsg << std::string_view(
static_cast<const char*>(disassembly->GetBufferPointer()),
disassembly->GetBufferSize());
}
}
device->EmitLog(WGPULoggingType_Info, dumpedMsg.str().c_str());
}
if (BlobCache* cache = device->GetBlobCache()) {
cache->EnsureStored(compiledShader);
}
// Clear the hlslSource. It is only used for logging and should not be used
// outside of the compilation.
CompiledShader result = compiledShader.Acquire();
result.hlslSource = "";
return result;
}
D3D12_SHADER_BYTECODE CompiledShader::GetD3D12ShaderBytecode() const {
if (compiledFXCShader != nullptr) {
return {compiledFXCShader->GetBufferPointer(), compiledFXCShader->GetBufferSize()};
} else if (compiledDXCShader != nullptr) {
return {compiledDXCShader->GetBufferPointer(), compiledDXCShader->GetBufferSize()};
}
UNREACHABLE();
return {};
return {shaderBlob.Data(), shaderBlob.Size()};
}
} // namespace dawn::native::d3d12
namespace dawn::native {
// Define the implementation to store d3d12::CompiledShader into the BlobCache.
template <>
void BlobCache::Store<d3d12::CompiledShader>(const CacheKey& key, const d3d12::CompiledShader& c) {
stream::ByteVectorSink sink;
c.VisitAll([&](const auto&... members) { StreamIn(&sink, members...); });
Store(key, CreateBlob(std::move(sink)));
}
// Define the implementation to load d3d12::CompiledShader from a Blob.
// static
ResultOrError<d3d12::CompiledShader> d3d12::CompiledShader::FromBlob(Blob blob) {
stream::BlobSource source(std::move(blob));
d3d12::CompiledShader c;
DAWN_TRY(c.VisitAll([&](auto&... members) { return StreamOut(&source, &members...); }));
return c;
}
} // namespace dawn::native

23
src/dawn/native/d3d12/ShaderModuleD3D12.h
View File

@ -15,8 +15,11 @@
#ifndef SRC_DAWN_NATIVE_D3D12_SHADERMODULED3D12_H_
#define SRC_DAWN_NATIVE_D3D12_SHADERMODULED3D12_H_
#include "dawn/native/ShaderModule.h"
#include <string>
#include "dawn/native/Blob.h"
#include "dawn/native/ShaderModule.h"
#include "dawn/native/VisitableMembers.h"
#include "dawn/native/d3d12/d3d12_platform.h"
namespace dawn::native {
@ -28,14 +31,22 @@ namespace dawn::native::d3d12 {
class Device;
class PipelineLayout;
// Manages a ref to one of the various representations of shader blobs and information used to
// emulate vertex/instance index starts
#define COMPILED_SHADER_MEMBERS(X) \
X(Blob, shaderBlob) \
X(std::string, hlslSource) \
X(bool, usesVertexOrInstanceIndex)
// `CompiledShader` holds a ref to one of the various representations of shader blobs and
// information used to emulate vertex/instance index starts. It also holds the `hlslSource` for the
// shader compilation, which is only transiently available during Compile, and cleared before it
// returns. It is not written to or loaded from the cache unless Toggle dump_shaders is true.
struct CompiledShader {
ComPtr<ID3DBlob> compiledFXCShader;
ComPtr<IDxcBlob> compiledDXCShader;
static ResultOrError<CompiledShader> FromBlob(Blob blob);
D3D12_SHADER_BYTECODE GetD3D12ShaderBytecode() const;
bool usesVertexOrInstanceIndex;
DAWN_VISITABLE_MEMBERS(COMPILED_SHADER_MEMBERS)
#undef COMPILED_SHADER_MEMBERS
};
class ShaderModule final : public ShaderModuleBase {

8
src/dawn/native/d3d12/UtilsD3D12.cpp
View File

@ -81,19 +81,19 @@ bool NeedBufferSizeWorkaroundForBufferTextureCopyOnD3D12(const BufferCopy& buffe
} // anonymous namespace
ResultOrError<std::wstring> ConvertStringToWstring(const char* str) {
size_t len = strlen(str);
ResultOrError<std::wstring> ConvertStringToWstring(std::string_view s) {
size_t len = s.length();
if (len == 0) {
return std::wstring();
}
int numChars = MultiByteToWideChar(CP_UTF8, MB_ERR_INVALID_CHARS, str, len, nullptr, 0);
int numChars = MultiByteToWideChar(CP_UTF8, MB_ERR_INVALID_CHARS, s.data(), len, nullptr, 0);
if (numChars == 0) {
return DAWN_INTERNAL_ERROR("Failed to convert string to wide string");
}
std::wstring result;
result.resize(numChars);
int numConvertedChars =
MultiByteToWideChar(CP_UTF8, MB_ERR_INVALID_CHARS, str, len, &result[0], numChars);
MultiByteToWideChar(CP_UTF8, MB_ERR_INVALID_CHARS, s.data(), len, &result[0], numChars);
if (numConvertedChars != numChars) {
return DAWN_INTERNAL_ERROR("Failed to convert string to wide string");
}

2
src/dawn/native/d3d12/UtilsD3D12.h
View File

@ -26,7 +26,7 @@
namespace dawn::native::d3d12 {
ResultOrError<std::wstring> ConvertStringToWstring(const char* str);
ResultOrError<std::wstring> ConvertStringToWstring(std::string_view s);
D3D12_COMPARISON_FUNC ToD3D12ComparisonFunc(wgpu::CompareFunction func);

10
src/dawn/native/stream/Stream.cpp
View File

@ -48,4 +48,14 @@ void Stream<std::string_view>::Write(Sink* s, const std::string_view& t) {
}
}
template <>
void Stream<std::wstring_view>::Write(Sink* s, const std::wstring_view& t) {
StreamIn(s, t.length());
size_t size = t.length() * sizeof(wchar_t);
if (size > 0) {
void* ptr = s->GetSpace(size);
memcpy(ptr, t.data(), size);
}
}
} // namespace dawn::native::stream

7
src/dawn/native/stream/Stream.h
View File

@ -297,10 +297,9 @@ class Stream<std::unordered_map<K, V>> {
public:
static void Write(stream::Sink* sink, const std::unordered_map<K, V>& m) {
std::vector<std::pair<K, V>> ordered(m.begin(), m.end());
std::sort(ordered.begin(), ordered.end(),
[](const std::pair<K, V>& a, const std::pair<K, V>& b) {
return std::less<K>{}(a.first, b.first);
});
std::sort(
ordered.begin(), ordered.end(),
[](const std::pair<K, V>& a, const std::pair<K, V>& b) { return a.first < b.first; });
StreamIn(sink, ordered);
}
};

5
src/dawn/tests/end2end/PipelineCachingTests.cpp
View File

@ -108,8 +108,8 @@ class PipelineCachingTests : public DawnTest {
const EntryCounts counts = {
// pipeline caching is only implemented on D3D12/Vulkan
IsD3D12() || IsVulkan() ? 1u : 0u,
// shader module caching is only implemented on Vulkan/Metal
IsVulkan() || IsMetal() ? 1u : 0u,
// shader module caching is only implemented on Vulkan/D3D12/Metal
IsVulkan() || IsMetal() || IsD3D12() ? 1u : 0u,
};
NiceMock<CachingInterfaceMock> mMockCache;
};
@ -646,6 +646,7 @@ TEST_P(SinglePipelineCachingTests, RenderPipelineBlobCacheIsolationKey) {
DAWN_INSTANTIATE_TEST(SinglePipelineCachingTests,
D3D12Backend({"enable_blob_cache"}),
D3D12Backend({"enable_blob_cache", "use_dxc"}),
MetalBackend({"enable_blob_cache"}),
OpenGLBackend({"enable_blob_cache"}),
OpenGLESBackend({"enable_blob_cache"}),

59
src/dawn/tests/unittests/native/StreamTests.cpp
View File

@ -174,6 +174,30 @@ TEST(SerializeTests, StdStringViews) {
EXPECT_CACHE_KEY_EQ(str, expected);
}
// Test that ByteVectorSink serializes std::wstring_views as expected.
TEST(SerializeTests, StdWStringViews) {
static constexpr std::wstring_view str(L"Hello world!");
ByteVectorSink expected;
StreamIn(&expected, size_t(str.length()));
size_t bytes = str.length() * sizeof(wchar_t);
memcpy(expected.GetSpace(bytes), str.data(), bytes);
EXPECT_CACHE_KEY_EQ(str, expected);
}
// Test that ByteVectorSink serializes Blobs as expected.
TEST(SerializeTests, Blob) {
uint8_t data[] = "dawn native Blob";
Blob blob = Blob::UnsafeCreateWithDeleter(data, sizeof(data), []() {});
ByteVectorSink expected;
StreamIn(&expected, sizeof(data));
expected.insert(expected.end(), data, data + sizeof(data));
EXPECT_CACHE_KEY_EQ(blob, expected);
}
// Test that ByteVectorSink serializes other ByteVectorSinks as expected.
TEST(SerializeTests, ByteVectorSinks) {
ByteVectorSink data = {'d', 'a', 't', 'a'};
@ -309,6 +333,41 @@ TEST(StreamTests, SerializeDeserializeVisitableMembers) {
}
}
// Test that serializing then deserializing a Blob yields the same data.
// Tested here instead of in the type-parameterized tests since Blobs are not copyable.
TEST(StreamTests, SerializeDeserializeBlobs) {
// Test an empty blob
{
Blob blob;
EXPECT_EQ(blob.Size(), 0u);
ByteVectorSink sink;
StreamIn(&sink, blob);
BlobSource src(CreateBlob(sink));
Blob out;
auto err = StreamOut(&src, &out);
EXPECT_FALSE(err.IsError());
EXPECT_EQ(blob.Size(), out.Size());
EXPECT_EQ(memcmp(blob.Data(), out.Data(), blob.Size()), 0);
}
// Test a blob with some data
{
Blob blob = CreateBlob(std::vector<double>{6.24, 3.12222});
ByteVectorSink sink;
StreamIn(&sink, blob);
BlobSource src(CreateBlob(sink));
Blob out;
auto err = StreamOut(&src, &out);
EXPECT_FALSE(err.IsError());
EXPECT_EQ(blob.Size(), out.Size());
EXPECT_EQ(memcmp(blob.Data(), out.Data(), blob.Size()), 0);
}
}
template <size_t N>
std::bitset<N - 1> BitsetFromBitString(const char (&str)[N]) {
// N - 1 because the last character is the null terminator.