Merge branch 'jeffbolznv-memory_scope_semantics'

This commit is contained in:
John Kessenich 2018-09-07 08:56:46 -06:00
Родитель 97068d8b30 36831c9bad
Коммит ddfb4a8022
28 изменённых файлов: 5544 добавлений и 4242 удалений

Просмотреть файл

@ -39,5 +39,6 @@ static const char* const E_SPV_KHR_16bit_storage = "SPV_KHR_16bit
static const char* const E_SPV_KHR_8bit_storage = "SPV_KHR_8bit_storage";
static const char* const E_SPV_KHR_storage_buffer_storage_class = "SPV_KHR_storage_buffer_storage_class";
static const char* const E_SPV_KHR_post_depth_coverage = "SPV_KHR_post_depth_coverage";
static const char* const E_SPV_KHR_vulkan_memory_model = "SPV_KHR_vulkan_memory_model";
#endif // #ifndef GLSLextKHR_H

Просмотреть файл

@ -129,6 +129,10 @@ protected:
spv::Decoration TranslateInterpolationDecoration(const glslang::TQualifier& qualifier);
spv::Decoration TranslateAuxiliaryStorageDecoration(const glslang::TQualifier& qualifier);
spv::Decoration TranslateNonUniformDecoration(const glslang::TQualifier& qualifier);
spv::Builder::AccessChain::CoherentFlags TranslateCoherent(const glslang::TType& type);
spv::MemoryAccessMask TranslateMemoryAccess(const spv::Builder::AccessChain::CoherentFlags &coherentFlags);
spv::ImageOperandsMask TranslateImageOperands(const spv::Builder::AccessChain::CoherentFlags &coherentFlags);
spv::Scope TranslateMemoryScope(const spv::Builder::AccessChain::CoherentFlags &coherentFlags);
spv::BuiltIn TranslateBuiltInDecoration(glslang::TBuiltInVariable, bool memberDeclaration);
spv::ImageFormat TranslateImageFormat(const glslang::TType& type);
spv::SelectionControlMask TranslateSelectionControl(const glslang::TIntermSelection&) const;
@ -327,13 +331,15 @@ spv::Decoration TranslateBlockDecoration(const glslang::TType& type, bool useSto
}
// Translate glslang type to SPIR-V memory decorations.
void TranslateMemoryDecoration(const glslang::TQualifier& qualifier, std::vector<spv::Decoration>& memory)
void TranslateMemoryDecoration(const glslang::TQualifier& qualifier, std::vector<spv::Decoration>& memory, bool useVulkanMemoryModel)
{
if (qualifier.coherent)
memory.push_back(spv::DecorationCoherent);
if (qualifier.volatil) {
memory.push_back(spv::DecorationVolatile);
memory.push_back(spv::DecorationCoherent);
if (!useVulkanMemoryModel) {
if (qualifier.coherent)
memory.push_back(spv::DecorationCoherent);
if (qualifier.volatil) {
memory.push_back(spv::DecorationVolatile);
memory.push_back(spv::DecorationCoherent);
}
}
if (qualifier.restrict)
memory.push_back(spv::DecorationRestrict);
@ -450,6 +456,105 @@ spv::Decoration TGlslangToSpvTraverser::TranslateNonUniformDecoration(const glsl
return spv::DecorationMax;
}
spv::MemoryAccessMask TGlslangToSpvTraverser::TranslateMemoryAccess(const spv::Builder::AccessChain::CoherentFlags &coherentFlags)
{
if (!glslangIntermediate->usingVulkanMemoryModel() || coherentFlags.isImage) {
return spv::MemoryAccessMaskNone;
}
spv::MemoryAccessMask mask = spv::MemoryAccessMaskNone;
if (coherentFlags.volatil ||
coherentFlags.coherent ||
coherentFlags.devicecoherent ||
coherentFlags.queuefamilycoherent ||
coherentFlags.workgroupcoherent ||
coherentFlags.subgroupcoherent) {
mask = mask | spv::MemoryAccessMakePointerAvailableKHRMask |
spv::MemoryAccessMakePointerVisibleKHRMask;
}
if (coherentFlags.nonprivate) {
mask = mask | spv::MemoryAccessNonPrivatePointerKHRMask;
}
if (coherentFlags.volatil) {
mask = mask | spv::MemoryAccessVolatileMask;
}
if (mask != spv::MemoryAccessMaskNone) {
builder.addCapability(spv::CapabilityVulkanMemoryModelKHR);
}
return mask;
}
spv::ImageOperandsMask TGlslangToSpvTraverser::TranslateImageOperands(const spv::Builder::AccessChain::CoherentFlags &coherentFlags)
{
if (!glslangIntermediate->usingVulkanMemoryModel()) {
return spv::ImageOperandsMaskNone;
}
spv::ImageOperandsMask mask = spv::ImageOperandsMaskNone;
if (coherentFlags.volatil ||
coherentFlags.coherent ||
coherentFlags.devicecoherent ||
coherentFlags.queuefamilycoherent ||
coherentFlags.workgroupcoherent ||
coherentFlags.subgroupcoherent) {
mask = mask | spv::ImageOperandsMakeTexelAvailableKHRMask |
spv::ImageOperandsMakeTexelVisibleKHRMask;
}
if (coherentFlags.nonprivate) {
mask = mask | spv::ImageOperandsNonPrivateTexelKHRMask;
}
if (coherentFlags.volatil) {
mask = mask | spv::ImageOperandsVolatileTexelKHRMask;
}
if (mask != spv::ImageOperandsMaskNone) {
builder.addCapability(spv::CapabilityVulkanMemoryModelKHR);
}
return mask;
}
spv::Builder::AccessChain::CoherentFlags TGlslangToSpvTraverser::TranslateCoherent(const glslang::TType& type)
{
spv::Builder::AccessChain::CoherentFlags flags;
flags.coherent = type.getQualifier().coherent;
flags.devicecoherent = type.getQualifier().devicecoherent;
flags.queuefamilycoherent = type.getQualifier().queuefamilycoherent;
// shared variables are implicitly workgroupcoherent in GLSL.
flags.workgroupcoherent = type.getQualifier().workgroupcoherent ||
type.getQualifier().storage == glslang::EvqShared;
flags.subgroupcoherent = type.getQualifier().subgroupcoherent;
// *coherent variables are implicitly nonprivate in GLSL
flags.nonprivate = type.getQualifier().nonprivate ||
type.getQualifier().subgroupcoherent ||
type.getQualifier().workgroupcoherent ||
type.getQualifier().queuefamilycoherent ||
type.getQualifier().devicecoherent ||
type.getQualifier().coherent;
flags.volatil = type.getQualifier().volatil;
flags.isImage = type.getBasicType() == glslang::EbtSampler;
return flags;
}
spv::Scope TGlslangToSpvTraverser::TranslateMemoryScope(const spv::Builder::AccessChain::CoherentFlags &coherentFlags)
{
spv::Scope scope;
if (coherentFlags.coherent) {
// coherent defaults to Device scope in the old model, QueueFamilyKHR scope in the new model
scope = glslangIntermediate->usingVulkanMemoryModel() ? spv::ScopeQueueFamilyKHR : spv::ScopeDevice;
} else if (coherentFlags.devicecoherent) {
scope = spv::ScopeDevice;
} else if (coherentFlags.queuefamilycoherent) {
scope = spv::ScopeQueueFamilyKHR;
} else if (coherentFlags.workgroupcoherent) {
scope = spv::ScopeWorkgroup;
} else if (coherentFlags.subgroupcoherent) {
scope = spv::ScopeSubgroup;
} else {
scope = spv::ScopeMax;
}
if (glslangIntermediate->usingVulkanMemoryModel() && scope == spv::ScopeDevice) {
builder.addCapability(spv::CapabilityVulkanMemoryModelDeviceScopeKHR);
}
return scope;
}
// Translate a glslang built-in variable to a SPIR-V built in decoration. Also generate
// associated capabilities when required. For some built-in variables, a capability
// is generated only when using the variable in an executable instruction, but not when
@ -979,6 +1084,16 @@ void InheritQualifiers(glslang::TQualifier& child, const glslang::TQualifier& pa
child.sample = true;
if (parent.coherent)
child.coherent = true;
if (parent.devicecoherent)
child.devicecoherent = true;
if (parent.queuefamilycoherent)
child.queuefamilycoherent = true;
if (parent.workgroupcoherent)
child.workgroupcoherent = true;
if (parent.subgroupcoherent)
child.subgroupcoherent = true;
if (parent.nonprivate)
child.nonprivate = true;
if (parent.volatil)
child.volatil = true;
if (parent.restrict)
@ -1045,7 +1160,12 @@ TGlslangToSpvTraverser::TGlslangToSpvTraverser(unsigned int spvVersion, const gl
builder.setSourceText(text);
}
stdBuiltins = builder.import("GLSL.std.450");
builder.setMemoryModel(spv::AddressingModelLogical, spv::MemoryModelGLSL450);
if (glslangIntermediate->usingVulkanMemoryModel()) {
builder.setMemoryModel(spv::AddressingModelLogical, spv::MemoryModelVulkanKHR);
builder.addExtension(spv::E_SPV_KHR_vulkan_memory_model);
} else {
builder.setMemoryModel(spv::AddressingModelLogical, spv::MemoryModelGLSL450);
}
shaderEntry = builder.makeEntryPoint(glslangIntermediate->getEntryPointName().c_str());
entryPoint = builder.addEntryPoint(executionModel, shaderEntry, glslangIntermediate->getEntryPointName().c_str());
@ -1351,7 +1471,7 @@ bool TGlslangToSpvTraverser::visitBinary(glslang::TVisit /* visit */, glslang::T
// store the result
builder.setAccessChain(lValue);
multiTypeStore(node->getType(), rValue);
multiTypeStore(node->getLeft()->getType(), rValue);
// assignments are expressions having an rValue after they are evaluated...
builder.clearAccessChain();
@ -1388,7 +1508,7 @@ bool TGlslangToSpvTraverser::visitBinary(glslang::TVisit /* visit */, glslang::T
}
// normal case for indexing array or structure or block
builder.accessChainPush(builder.makeIntConstant(spvIndex));
builder.accessChainPush(builder.makeIntConstant(spvIndex), TranslateCoherent(node->getLeft()->getType()));
// Add capabilities here for accessing PointSize and clip/cull distance.
// We have deferred generation of associated capabilities until now.
@ -1424,7 +1544,7 @@ bool TGlslangToSpvTraverser::visitBinary(glslang::TVisit /* visit */, glslang::T
if (! node->getLeft()->getType().isArray() && node->getLeft()->getType().isVector())
builder.accessChainPushComponent(index, convertGlslangToSpvType(node->getLeft()->getType()));
else
builder.accessChainPush(index);
builder.accessChainPush(index, TranslateCoherent(node->getLeft()->getType()));
}
return false;
case glslang::EOpVectorSwizzle:
@ -1658,11 +1778,11 @@ bool TGlslangToSpvTraverser::visitAggregate(glslang::TVisit visit, glslang::TInt
builder.setAccessChainRValue(result);
return false;
} else if (node->getOp() == glslang::EOpImageStore ||
#ifdef AMD_EXTENSIONS
} else if (node->getOp() == glslang::EOpImageStore || node->getOp() == glslang::EOpImageStoreLod) {
#else
} else if (node->getOp() == glslang::EOpImageStore) {
node->getOp() == glslang::EOpImageStoreLod ||
#endif
node->getOp() == glslang::EOpImageAtomicStore) {
// "imageStore" is a special case, which has no result
return false;
}
@ -1952,6 +2072,10 @@ bool TGlslangToSpvTraverser::visitAggregate(glslang::TVisit visit, glslang::TInt
// These all have 0 operands and will naturally finish up in the code below for 0 operands
break;
case glslang::EOpAtomicStore:
noReturnValue = true;
// fallthrough
case glslang::EOpAtomicLoad:
case glslang::EOpAtomicAdd:
case glslang::EOpAtomicMin:
case glslang::EOpAtomicMax:
@ -2050,6 +2174,8 @@ bool TGlslangToSpvTraverser::visitAggregate(glslang::TVisit visit, glslang::TInt
case glslang::EOpAtomicXor:
case glslang::EOpAtomicExchange:
case glslang::EOpAtomicCompSwap:
case glslang::EOpAtomicLoad:
case glslang::EOpAtomicStore:
case glslang::EOpAtomicCounterAdd:
case glslang::EOpAtomicCounterSubtract:
case glslang::EOpAtomicCounterMin:
@ -2876,7 +3002,7 @@ void TGlslangToSpvTraverser::decorateStructType(const glslang::TType& type,
qualifier.storage == glslang::EvqBuffer) {
// Add memory decorations only to top-level members of shader storage block
std::vector<spv::Decoration> memory;
TranslateMemoryDecoration(memberQualifier, memory);
TranslateMemoryDecoration(memberQualifier, memory, glslangIntermediate->usingVulkanMemoryModel());
for (unsigned int i = 0; i < memory.size(); ++i)
builder.addMemberDecoration(spvType, member, memory[i]);
}
@ -2987,8 +3113,15 @@ spv::Id TGlslangToSpvTraverser::makeArraySizeId(const glslang::TArraySizes& arra
spv::Id TGlslangToSpvTraverser::accessChainLoad(const glslang::TType& type)
{
spv::Id nominalTypeId = builder.accessChainGetInferredType();
spv::Builder::AccessChain::CoherentFlags coherentFlags = builder.getAccessChain().coherentFlags;
coherentFlags |= TranslateCoherent(type);
spv::Id loadedId = builder.accessChainLoad(TranslatePrecisionDecoration(type),
TranslateNonUniformDecoration(type.getQualifier()), nominalTypeId);
TranslateNonUniformDecoration(type.getQualifier()),
nominalTypeId,
spv::MemoryAccessMask(TranslateMemoryAccess(coherentFlags) & ~spv::MemoryAccessMakePointerAvailableKHRMask),
TranslateMemoryScope(coherentFlags));
// Need to convert to abstract types when necessary
if (type.getBasicType() == glslang::EbtBool) {
@ -3044,7 +3177,12 @@ void TGlslangToSpvTraverser::accessChainStore(const glslang::TType& type, spv::I
}
}
builder.accessChainStore(rvalue);
spv::Builder::AccessChain::CoherentFlags coherentFlags = builder.getAccessChain().coherentFlags;
coherentFlags |= TranslateCoherent(type);
builder.accessChainStore(rvalue,
spv::MemoryAccessMask(TranslateMemoryAccess(coherentFlags) & ~spv::MemoryAccessMakePointerVisibleKHRMask),
TranslateMemoryScope(coherentFlags));
}
// For storing when types match at the glslang level, but not might match at the
@ -3090,7 +3228,7 @@ void TGlslangToSpvTraverser::multiTypeStore(const glslang::TType& type, spv::Id
// set up the target storage
builder.clearAccessChain();
builder.setAccessChainLValue(lValue);
builder.accessChainPush(builder.makeIntConstant(index));
builder.accessChainPush(builder.makeIntConstant(index), TranslateCoherent(type));
// store the member
multiTypeStore(glslangElementType, elementRValue);
@ -3110,7 +3248,7 @@ void TGlslangToSpvTraverser::multiTypeStore(const glslang::TType& type, spv::Id
// set up the target storage
builder.clearAccessChain();
builder.setAccessChainLValue(lValue);
builder.accessChainPush(builder.makeIntConstant(m));
builder.accessChainPush(builder.makeIntConstant(m), TranslateCoherent(type));
// store the member
multiTypeStore(glslangMemberType, memberRValue);
@ -3287,11 +3425,11 @@ bool TGlslangToSpvTraverser::originalParam(glslang::TStorageQualifier qualifier,
// Make all the functions, skeletally, without actually visiting their bodies.
void TGlslangToSpvTraverser::makeFunctions(const glslang::TIntermSequence& glslFunctions)
{
const auto getParamDecorations = [](std::vector<spv::Decoration>& decorations, const glslang::TType& type) {
const auto getParamDecorations = [](std::vector<spv::Decoration>& decorations, const glslang::TType& type, bool useVulkanMemoryModel) {
spv::Decoration paramPrecision = TranslatePrecisionDecoration(type);
if (paramPrecision != spv::NoPrecision)
decorations.push_back(paramPrecision);
TranslateMemoryDecoration(type.getQualifier(), decorations);
TranslateMemoryDecoration(type.getQualifier(), decorations, useVulkanMemoryModel);
};
for (int f = 0; f < (int)glslFunctions.size(); ++f) {
@ -3330,7 +3468,7 @@ void TGlslangToSpvTraverser::makeFunctions(const glslang::TIntermSequence& glslF
typeId = builder.makePointer(spv::StorageClassFunction, typeId);
else
rValueParameters.insert(parameters[p]->getAsSymbolNode()->getId());
getParamDecorations(paramDecorations[p], paramType);
getParamDecorations(paramDecorations[p], paramType, glslangIntermediate->usingVulkanMemoryModel());
paramTypes.push_back(typeId);
}
@ -3420,6 +3558,8 @@ void TGlslangToSpvTraverser::translateArguments(const glslang::TIntermAggregate&
case glslang::EOpImageAtomicXor:
case glslang::EOpImageAtomicExchange:
case glslang::EOpImageAtomicCompSwap:
case glslang::EOpImageAtomicLoad:
case glslang::EOpImageAtomicStore:
if (i == 0)
lvalue = true;
break;
@ -3547,8 +3687,10 @@ spv::Id TGlslangToSpvTraverser::createImageTextureFunctionCall(glslang::TIntermO
builder.setLine(node->getLoc().line);
// Process a GLSL texturing op (will be SPV image)
const glslang::TSampler sampler = node->getAsAggregate() ? node->getAsAggregate()->getSequence()[0]->getAsTyped()->getType().getSampler()
: node->getAsUnaryNode()->getOperand()->getAsTyped()->getType().getSampler();
const glslang::TType &imageType = node->getAsAggregate() ? node->getAsAggregate()->getSequence()[0]->getAsTyped()->getType()
: node->getAsUnaryNode()->getOperand()->getAsTyped()->getType();
const glslang::TSampler sampler = imageType.getSampler();
#ifdef AMD_EXTENSIONS
bool f16ShadowCompare = (sampler.shadow && node->getAsAggregate())
? node->getAsAggregate()->getSequence()[1]->getAsTyped()->getType().getBasicType() == glslang::EbtFloat16
@ -3651,22 +3793,38 @@ spv::Id TGlslangToSpvTraverser::createImageTextureFunctionCall(glslang::TIntermO
#else
if (node->getOp() == glslang::EOpImageLoad) {
#endif
spv::ImageOperandsMask mask = spv::ImageOperandsMaskNone;
if (sampler.ms) {
spv::IdImmediate imageOperands = { false, spv::ImageOperandsSampleMask };
operands.push_back(imageOperands);
spv::IdImmediate imageOperand = { true, *opIt };
operands.push_back(imageOperand);
mask = mask | spv::ImageOperandsSampleMask;
}
#ifdef AMD_EXTENSIONS
} else if (cracked.lod) {
if (cracked.lod) {
builder.addExtension(spv::E_SPV_AMD_shader_image_load_store_lod);
builder.addCapability(spv::CapabilityImageReadWriteLodAMD);
spv::IdImmediate imageOperands = { false, spv::ImageOperandsLodMask };
operands.push_back(imageOperands);
spv::IdImmediate imageOperand = { true, *opIt };
operands.push_back(imageOperand);
#endif
mask = mask | spv::ImageOperandsLodMask;
}
#endif
mask = mask | TranslateImageOperands(TranslateCoherent(imageType));
mask = (spv::ImageOperandsMask)(mask & ~spv::ImageOperandsMakeTexelAvailableKHRMask);
if (mask) {
spv::IdImmediate imageOperands = { false, (unsigned int)mask };
operands.push_back(imageOperands);
}
if (mask & spv::ImageOperandsSampleMask) {
spv::IdImmediate imageOperand = { true, *opIt++ };
operands.push_back(imageOperand);
}
#ifdef AMD_EXTENSIONS
if (mask & spv::ImageOperandsLodMask) {
spv::IdImmediate imageOperand = { true, *opIt++ };
operands.push_back(imageOperand);
}
#endif
if (mask & spv::ImageOperandsMakeTexelVisibleKHRMask) {
spv::IdImmediate imageOperand = { true, builder.makeUintConstant(TranslateMemoryScope(TranslateCoherent(imageType))) };
operands.push_back(imageOperand);
}
if (builder.getImageTypeFormat(builder.getImageType(operands.front().word)) == spv::ImageFormatUnknown)
builder.addCapability(spv::CapabilityStorageImageReadWithoutFormat);
@ -3683,29 +3841,52 @@ spv::Id TGlslangToSpvTraverser::createImageTextureFunctionCall(glslang::TIntermO
#else
} else if (node->getOp() == glslang::EOpImageStore) {
#endif
if (sampler.ms) {
spv::IdImmediate texel = { true, *(opIt + 1) };
operands.push_back(texel);
spv::IdImmediate imageOperands = { false, spv::ImageOperandsSampleMask };
operands.push_back(imageOperands);
spv::IdImmediate imageOperand = { true, *opIt };
operands.push_back(imageOperand);
#ifdef AMD_EXTENSIONS
} else if (cracked.lod) {
builder.addExtension(spv::E_SPV_AMD_shader_image_load_store_lod);
builder.addCapability(spv::CapabilityImageReadWriteLodAMD);
// Push the texel value before the operands
#ifdef AMD_EXTENSIONS
if (sampler.ms || cracked.lod) {
#else
if (sampler.ms) {
#endif
spv::IdImmediate texel = { true, *(opIt + 1) };
operands.push_back(texel);
spv::IdImmediate imageOperands = { false, spv::ImageOperandsLodMask };
operands.push_back(imageOperands);
spv::IdImmediate imageOperand = { true, *opIt };
operands.push_back(imageOperand);
#endif
} else {
spv::IdImmediate texel = { true, *opIt };
operands.push_back(texel);
}
spv::ImageOperandsMask mask = spv::ImageOperandsMaskNone;
if (sampler.ms) {
mask = mask | spv::ImageOperandsSampleMask;
}
#ifdef AMD_EXTENSIONS
if (cracked.lod) {
builder.addExtension(spv::E_SPV_AMD_shader_image_load_store_lod);
builder.addCapability(spv::CapabilityImageReadWriteLodAMD);
mask = mask | spv::ImageOperandsLodMask;
}
#endif
mask = mask | TranslateImageOperands(TranslateCoherent(imageType));
mask = (spv::ImageOperandsMask)(mask & ~spv::ImageOperandsMakeTexelVisibleKHRMask);
if (mask) {
spv::IdImmediate imageOperands = { false, (unsigned int)mask };
operands.push_back(imageOperands);
}
if (mask & spv::ImageOperandsSampleMask) {
spv::IdImmediate imageOperand = { true, *opIt++ };
operands.push_back(imageOperand);
}
#ifdef AMD_EXTENSIONS
if (mask & spv::ImageOperandsLodMask) {
spv::IdImmediate imageOperand = { true, *opIt++ };
operands.push_back(imageOperand);
}
#endif
if (mask & spv::ImageOperandsMakeTexelAvailableKHRMask) {
spv::IdImmediate imageOperand = { true, builder.makeUintConstant(TranslateMemoryScope(TranslateCoherent(imageType))) };
operands.push_back(imageOperand);
}
builder.createNoResultOp(spv::OpImageWrite, operands);
if (builder.getImageTypeFormat(builder.getImageType(operands.front().word)) == spv::ImageFormatUnknown)
builder.addCapability(spv::CapabilityStorageImageWriteWithoutFormat);
@ -3719,21 +3900,37 @@ spv::Id TGlslangToSpvTraverser::createImageTextureFunctionCall(glslang::TIntermO
if (builder.getImageTypeFormat(builder.getImageType(operands.front().word)) == spv::ImageFormatUnknown)
builder.addCapability(spv::CapabilityStorageImageReadWithoutFormat);
spv::ImageOperandsMask mask = spv::ImageOperandsMaskNone;
if (sampler.ms) {
spv::IdImmediate imageOperands = { false, spv::ImageOperandsSampleMask };
operands.push_back(imageOperands);
spv::IdImmediate imageOperand = { true, *opIt++ };
operands.push_back(imageOperand);
mask = mask | spv::ImageOperandsSampleMask;
}
#ifdef AMD_EXTENSIONS
} else if (cracked.lod) {
if (cracked.lod) {
builder.addExtension(spv::E_SPV_AMD_shader_image_load_store_lod);
builder.addCapability(spv::CapabilityImageReadWriteLodAMD);
spv::IdImmediate imageOperands = { false, spv::ImageOperandsLodMask };
mask = mask | spv::ImageOperandsLodMask;
}
#endif
mask = mask | TranslateImageOperands(TranslateCoherent(imageType));
mask = (spv::ImageOperandsMask)(mask & ~spv::ImageOperandsMakeTexelAvailableKHRMask);
if (mask) {
spv::IdImmediate imageOperands = { false, (unsigned int)mask };
operands.push_back(imageOperands);
}
if (mask & spv::ImageOperandsSampleMask) {
spv::IdImmediate imageOperand = { true, *opIt++ };
operands.push_back(imageOperand);
}
#ifdef AMD_EXTENSIONS
if (mask & spv::ImageOperandsLodMask) {
spv::IdImmediate imageOperand = { true, *opIt++ };
operands.push_back(imageOperand);
}
#endif
if (mask & spv::ImageOperandsMakeTexelVisibleKHRMask) {
spv::IdImmediate imageOperand = { true, builder.makeUintConstant(TranslateMemoryScope(TranslateCoherent(imageType))) };
operands.push_back(imageOperand);
}
// Create the return type that was a special structure
@ -3756,7 +3953,14 @@ spv::Id TGlslangToSpvTraverser::createImageTextureFunctionCall(glslang::TIntermO
spv::IdImmediate sample = { true, sampler.ms ? *(opIt++) : builder.makeUintConstant(0) };
operands.push_back(sample);
spv::Id resultTypeId = builder.makePointer(spv::StorageClassImage, resultType());
spv::Id resultTypeId;
// imageAtomicStore has a void return type so base the pointer type on
// the type of the value operand.
if (node->getOp() == glslang::EOpImageAtomicStore) {
resultTypeId = builder.makePointer(spv::StorageClassImage, builder.getTypeId(operands[2].word));
} else {
resultTypeId = builder.makePointer(spv::StorageClassImage, resultType());
}
spv::Id pointer = builder.createOp(spv::OpImageTexelPointer, resultTypeId, operands);
std::vector<spv::Id> operands;
@ -3972,6 +4176,16 @@ spv::Id TGlslangToSpvTraverser::createImageTextureFunctionCall(glslang::TIntermO
}
}
// nonprivate
if (imageType.getQualifier().nonprivate) {
params.nonprivate = true;
}
// volatile
if (imageType.getQualifier().volatil) {
params.volatil = true;
}
std::vector<spv::Id> result( 1,
builder.createTextureCall(precision, resultType(), sparse, cracked.fetch, cracked.proj, cracked.gather, noImplicitLod, params)
);
@ -5340,8 +5554,14 @@ spv::Id TGlslangToSpvTraverser::createAtomicOperation(glslang::TOperator op, spv
opCode = spv::OpAtomicIDecrement;
break;
case glslang::EOpAtomicCounter:
case glslang::EOpImageAtomicLoad:
case glslang::EOpAtomicLoad:
opCode = spv::OpAtomicLoad;
break;
case glslang::EOpAtomicStore:
case glslang::EOpImageAtomicStore:
opCode = spv::OpAtomicStore;
break;
default:
assert(0);
break;
@ -5352,36 +5572,82 @@ spv::Id TGlslangToSpvTraverser::createAtomicOperation(glslang::TOperator op, spv
// Sort out the operands
// - mapping from glslang -> SPV
// - there are extra SPV operands with no glslang source
// - there are extra SPV operands that are optional in glslang
// - compare-exchange swaps the value and comparator
// - compare-exchange has an extra memory semantics
// - EOpAtomicCounterDecrement needs a post decrement
std::vector<spv::Id> spvAtomicOperands; // hold the spv operands
auto opIt = operands.begin(); // walk the glslang operands
spvAtomicOperands.push_back(*(opIt++));
spvAtomicOperands.push_back(builder.makeUintConstant(spv::ScopeDevice)); // TBD: what is the correct scope?
spvAtomicOperands.push_back(builder.makeUintConstant(spv::MemorySemanticsMaskNone)); // TBD: what are the correct memory semantics?
if (opCode == spv::OpAtomicCompareExchange) {
// There are 2 memory semantics for compare-exchange. And the operand order of "comparator" and "new value" in GLSL
// differs from that in SPIR-V. Hence, special processing is required.
spvAtomicOperands.push_back(builder.makeUintConstant(spv::MemorySemanticsMaskNone));
spvAtomicOperands.push_back(*(opIt + 1));
spvAtomicOperands.push_back(*opIt);
opIt += 2;
spv::Id pointerId = 0, compareId = 0, valueId = 0;
// scope defaults to Device in the old model, QueueFamilyKHR in the new model
spv::Id scopeId;
if (glslangIntermediate->usingVulkanMemoryModel()) {
scopeId = builder.makeUintConstant(spv::ScopeQueueFamilyKHR);
} else {
scopeId = builder.makeUintConstant(spv::ScopeDevice);
}
// semantics default to relaxed
spv::Id semanticsId = builder.makeUintConstant(spv::MemorySemanticsMaskNone);
spv::Id semanticsId2 = semanticsId;
pointerId = operands[0];
if (opCode == spv::OpAtomicIIncrement || opCode == spv::OpAtomicIDecrement) {
// no additional operands
} else if (opCode == spv::OpAtomicCompareExchange) {
compareId = operands[1];
valueId = operands[2];
if (operands.size() > 3) {
scopeId = operands[3];
semanticsId = builder.makeUintConstant(builder.getConstantScalar(operands[4]) | builder.getConstantScalar(operands[5]));
semanticsId2 = builder.makeUintConstant(builder.getConstantScalar(operands[6]) | builder.getConstantScalar(operands[7]));
}
} else if (opCode == spv::OpAtomicLoad) {
if (operands.size() > 1) {
scopeId = operands[1];
semanticsId = builder.makeUintConstant(builder.getConstantScalar(operands[2]) | builder.getConstantScalar(operands[3]));
}
} else {
// atomic store or RMW
valueId = operands[1];
if (operands.size() > 2) {
scopeId = operands[2];
semanticsId = builder.makeUintConstant(builder.getConstantScalar(operands[3]) | builder.getConstantScalar(operands[4]));
}
}
// Add the rest of the operands, skipping any that were dealt with above.
for (; opIt != operands.end(); ++opIt)
spvAtomicOperands.push_back(*opIt);
// Check for capabilities
unsigned semanticsImmediate = builder.getConstantScalar(semanticsId) | builder.getConstantScalar(semanticsId2);
if (semanticsImmediate & (spv::MemorySemanticsMakeAvailableKHRMask | spv::MemorySemanticsMakeVisibleKHRMask | spv::MemorySemanticsOutputMemoryKHRMask)) {
builder.addCapability(spv::CapabilityVulkanMemoryModelKHR);
}
spv::Id resultId = builder.createOp(opCode, typeId, spvAtomicOperands);
if (glslangIntermediate->usingVulkanMemoryModel() && builder.getConstantScalar(scopeId) == spv::ScopeDevice) {
builder.addCapability(spv::CapabilityVulkanMemoryModelDeviceScopeKHR);
}
// GLSL and HLSL atomic-counter decrement return post-decrement value,
// while SPIR-V returns pre-decrement value. Translate between these semantics.
if (op == glslang::EOpAtomicCounterDecrement)
resultId = builder.createBinOp(spv::OpISub, typeId, resultId, builder.makeIntConstant(1));
std::vector<spv::Id> spvAtomicOperands; // hold the spv operands
spvAtomicOperands.push_back(pointerId);
spvAtomicOperands.push_back(scopeId);
spvAtomicOperands.push_back(semanticsId);
if (opCode == spv::OpAtomicCompareExchange) {
spvAtomicOperands.push_back(semanticsId2);
spvAtomicOperands.push_back(valueId);
spvAtomicOperands.push_back(compareId);
} else if (opCode != spv::OpAtomicLoad && opCode != spv::OpAtomicIIncrement && opCode != spv::OpAtomicIDecrement) {
spvAtomicOperands.push_back(valueId);
}
return resultId;
if (opCode == spv::OpAtomicStore) {
builder.createNoResultOp(opCode, spvAtomicOperands);
return 0;
} else {
spv::Id resultId = builder.createOp(opCode, typeId, spvAtomicOperands);
// GLSL and HLSL atomic-counter decrement return post-decrement value,
// while SPIR-V returns pre-decrement value. Translate between these semantics.
if (op == glslang::EOpAtomicCounterDecrement)
resultId = builder.createBinOp(spv::OpISub, typeId, resultId, builder.makeIntConstant(1));
return resultId;
}
}
// Create group invocation operations.
@ -6282,7 +6548,41 @@ spv::Id TGlslangToSpvTraverser::createMiscOperation(glslang::TOperator op, spv::
libCall = spv::InterpolateAtVertexAMD;
break;
#endif
case glslang::EOpBarrier:
{
// This is for the extended controlBarrier function, with four operands.
// The unextended barrier() goes through createNoArgOperation.
assert(operands.size() == 4);
unsigned int executionScope = builder.getConstantScalar(operands[0]);
unsigned int memoryScope = builder.getConstantScalar(operands[1]);
unsigned int semantics = builder.getConstantScalar(operands[2]) | builder.getConstantScalar(operands[3]);
builder.createControlBarrier((spv::Scope)executionScope, (spv::Scope)memoryScope, (spv::MemorySemanticsMask)semantics);
if (semantics & (spv::MemorySemanticsMakeAvailableKHRMask | spv::MemorySemanticsMakeVisibleKHRMask | spv::MemorySemanticsOutputMemoryKHRMask)) {
builder.addCapability(spv::CapabilityVulkanMemoryModelKHR);
}
if (glslangIntermediate->usingVulkanMemoryModel() && (executionScope == spv::ScopeDevice || memoryScope == spv::ScopeDevice)) {
builder.addCapability(spv::CapabilityVulkanMemoryModelDeviceScopeKHR);
}
return 0;
}
break;
case glslang::EOpMemoryBarrier:
{
// This is for the extended memoryBarrier function, with three operands.
// The unextended memoryBarrier() goes through createNoArgOperation.
assert(operands.size() == 3);
unsigned int memoryScope = builder.getConstantScalar(operands[0]);
unsigned int semantics = builder.getConstantScalar(operands[1]) | builder.getConstantScalar(operands[2]);
builder.createMemoryBarrier((spv::Scope)memoryScope, (spv::MemorySemanticsMask)semantics);
if (semantics & (spv::MemorySemanticsMakeAvailableKHRMask | spv::MemorySemanticsMakeVisibleKHRMask | spv::MemorySemanticsOutputMemoryKHRMask)) {
builder.addCapability(spv::CapabilityVulkanMemoryModelKHR);
}
if (glslangIntermediate->usingVulkanMemoryModel() && memoryScope == spv::ScopeDevice) {
builder.addCapability(spv::CapabilityVulkanMemoryModelDeviceScopeKHR);
}
return 0;
}
break;
default:
return 0;
}
@ -6351,7 +6651,8 @@ spv::Id TGlslangToSpvTraverser::createMiscOperation(glslang::TOperator op, spv::
// Intrinsics with no arguments (or no return value, and no precision).
spv::Id TGlslangToSpvTraverser::createNoArgOperation(glslang::TOperator op, spv::Decoration precision, spv::Id typeId)
{
// TODO: get the barrier operands correct
// GLSL memory barriers use queuefamily scope in new model, device scope in old model
spv::Scope memoryBarrierScope = glslangIntermediate->usingVulkanMemoryModel() ? spv::ScopeQueueFamilyKHR : spv::ScopeDevice;
switch (op) {
case glslang::EOpEmitVertex:
@ -6362,11 +6663,14 @@ spv::Id TGlslangToSpvTraverser::createNoArgOperation(glslang::TOperator op, spv:
return 0;
case glslang::EOpBarrier:
if (glslangIntermediate->getStage() == EShLangTessControl) {
builder.createControlBarrier(spv::ScopeWorkgroup, spv::ScopeInvocation, spv::MemorySemanticsMaskNone);
// TODO: prefer the following, when available:
// builder.createControlBarrier(spv::ScopePatch, spv::ScopePatch,
// spv::MemorySemanticsPatchMask |
// spv::MemorySemanticsAcquireReleaseMask);
if (glslangIntermediate->usingVulkanMemoryModel()) {
builder.createControlBarrier(spv::ScopeWorkgroup, spv::ScopeWorkgroup,
spv::MemorySemanticsOutputMemoryKHRMask |
spv::MemorySemanticsAcquireReleaseMask);
builder.addCapability(spv::CapabilityVulkanMemoryModelKHR);
} else {
builder.createControlBarrier(spv::ScopeWorkgroup, spv::ScopeInvocation, spv::MemorySemanticsMaskNone);
}
} else {
builder.createControlBarrier(spv::ScopeWorkgroup, spv::ScopeWorkgroup,
spv::MemorySemanticsWorkgroupMemoryMask |
@ -6374,24 +6678,24 @@ spv::Id TGlslangToSpvTraverser::createNoArgOperation(glslang::TOperator op, spv:
}
return 0;
case glslang::EOpMemoryBarrier:
builder.createMemoryBarrier(spv::ScopeDevice, spv::MemorySemanticsAllMemory |
spv::MemorySemanticsAcquireReleaseMask);
builder.createMemoryBarrier(memoryBarrierScope, spv::MemorySemanticsAllMemory |
spv::MemorySemanticsAcquireReleaseMask);
return 0;
case glslang::EOpMemoryBarrierAtomicCounter:
builder.createMemoryBarrier(spv::ScopeDevice, spv::MemorySemanticsAtomicCounterMemoryMask |
spv::MemorySemanticsAcquireReleaseMask);
builder.createMemoryBarrier(memoryBarrierScope, spv::MemorySemanticsAtomicCounterMemoryMask |
spv::MemorySemanticsAcquireReleaseMask);
return 0;
case glslang::EOpMemoryBarrierBuffer:
builder.createMemoryBarrier(spv::ScopeDevice, spv::MemorySemanticsUniformMemoryMask |
spv::MemorySemanticsAcquireReleaseMask);
builder.createMemoryBarrier(memoryBarrierScope, spv::MemorySemanticsUniformMemoryMask |
spv::MemorySemanticsAcquireReleaseMask);
return 0;
case glslang::EOpMemoryBarrierImage:
builder.createMemoryBarrier(spv::ScopeDevice, spv::MemorySemanticsImageMemoryMask |
spv::MemorySemanticsAcquireReleaseMask);
builder.createMemoryBarrier(memoryBarrierScope, spv::MemorySemanticsImageMemoryMask |
spv::MemorySemanticsAcquireReleaseMask);
return 0;
case glslang::EOpMemoryBarrierShared:
builder.createMemoryBarrier(spv::ScopeDevice, spv::MemorySemanticsWorkgroupMemoryMask |
spv::MemorySemanticsAcquireReleaseMask);
builder.createMemoryBarrier(memoryBarrierScope, spv::MemorySemanticsWorkgroupMemoryMask |
spv::MemorySemanticsAcquireReleaseMask);
return 0;
case glslang::EOpGroupMemoryBarrier:
builder.createMemoryBarrier(spv::ScopeWorkgroup, spv::MemorySemanticsAllMemory |
@ -6520,7 +6824,7 @@ spv::Id TGlslangToSpvTraverser::getSymbolId(const glslang::TIntermSymbol* symbol
if (symbol->getType().isImage()) {
std::vector<spv::Decoration> memory;
TranslateMemoryDecoration(symbol->getType().getQualifier(), memory);
TranslateMemoryDecoration(symbol->getType().getQualifier(), memory, glslangIntermediate->usingVulkanMemoryModel());
for (unsigned int i = 0; i < memory.size(); ++i)
builder.addDecoration(id, memory[i]);
}

Просмотреть файл

@ -1216,19 +1216,35 @@ Id Builder::createUndefined(Id type)
}
// Comments in header
void Builder::createStore(Id rValue, Id lValue)
void Builder::createStore(Id rValue, Id lValue, spv::MemoryAccessMask memoryAccess, spv::Scope scope)
{
Instruction* store = new Instruction(OpStore);
store->addIdOperand(lValue);
store->addIdOperand(rValue);
if (memoryAccess != MemoryAccessMaskNone) {
store->addImmediateOperand(memoryAccess);
if (memoryAccess & spv::MemoryAccessMakePointerAvailableKHRMask) {
store->addIdOperand(makeUintConstant(scope));
}
}
buildPoint->addInstruction(std::unique_ptr<Instruction>(store));
}
// Comments in header
Id Builder::createLoad(Id lValue)
Id Builder::createLoad(Id lValue, spv::MemoryAccessMask memoryAccess, spv::Scope scope)
{
Instruction* load = new Instruction(getUniqueId(), getDerefTypeId(lValue), OpLoad);
load->addIdOperand(lValue);
if (memoryAccess != MemoryAccessMaskNone) {
load->addImmediateOperand(memoryAccess);
if (memoryAccess & spv::MemoryAccessMakePointerVisibleKHRMask) {
load->addIdOperand(makeUintConstant(scope));
}
}
buildPoint->addInstruction(std::unique_ptr<Instruction>(load));
return load->getResultId();
@ -1361,6 +1377,16 @@ void Builder::createNoResultOp(Op opCode, Id operand)
buildPoint->addInstruction(std::unique_ptr<Instruction>(op));
}
// An opcode that has multiple operands, no result id, and no type
void Builder::createNoResultOp(Op opCode, const std::vector<Id>& operands)
{
Instruction* op = new Instruction(opCode);
for (auto it = operands.cbegin(); it != operands.cend(); ++it) {
op->addIdOperand(*it);
}
buildPoint->addInstruction(std::unique_ptr<Instruction>(op));
}
// An opcode that has multiple operands, no result id, and no type
void Builder::createNoResultOp(Op opCode, const std::vector<IdImmediate>& operands)
{
@ -1679,6 +1705,12 @@ Id Builder::createTextureCall(Decoration precision, Id resultType, bool sparse,
mask = (ImageOperandsMask)(mask | ImageOperandsMinLodMask);
texArgs[numArgs++] = parameters.lodClamp;
}
if (parameters.nonprivate) {
mask = mask | ImageOperandsNonPrivateTexelKHRMask;
}
if (parameters.volatil) {
mask = mask | ImageOperandsVolatileTexelKHRMask;
}
if (mask == ImageOperandsMaskNone)
--numArgs; // undo speculative reservation for the mask argument
else
@ -2352,6 +2384,7 @@ void Builder::clearAccessChain()
accessChain.component = NoResult;
accessChain.preSwizzleBaseType = NoType;
accessChain.isRValue = false;
accessChain.coherentFlags.clear();
}
// Comments in header
@ -2378,7 +2411,7 @@ void Builder::accessChainPushSwizzle(std::vector<unsigned>& swizzle, Id preSwizz
}
// Comments in header
void Builder::accessChainStore(Id rvalue)
void Builder::accessChainStore(Id rvalue, spv::MemoryAccessMask memoryAccess, spv::Scope scope)
{
assert(accessChain.isRValue == false);
@ -2396,11 +2429,11 @@ void Builder::accessChainStore(Id rvalue)
source = createLvalueSwizzle(getTypeId(tempBaseId), tempBaseId, source, accessChain.swizzle);
}
createStore(source, base);
createStore(source, base, memoryAccess, scope);
}
// Comments in header
Id Builder::accessChainLoad(Decoration precision, Decoration nonUniform, Id resultType)
Id Builder::accessChainLoad(Decoration precision, Decoration nonUniform, Id resultType, spv::MemoryAccessMask memoryAccess, spv::Scope scope)
{
Id id;
@ -2444,7 +2477,7 @@ Id Builder::accessChainLoad(Decoration precision, Decoration nonUniform, Id resu
} else {
transferAccessChainSwizzle(true);
// load through the access chain
id = createLoad(collapseAccessChain());
id = createLoad(collapseAccessChain(), memoryAccess, scope);
setPrecision(id, precision);
addDecoration(id, nonUniform);
}

Просмотреть файл

@ -274,10 +274,10 @@ public:
Id createUndefined(Id type);
// Store into an Id and return the l-value
void createStore(Id rValue, Id lValue);
void createStore(Id rValue, Id lValue, spv::MemoryAccessMask memoryAccess = spv::MemoryAccessMaskNone, spv::Scope scope = spv::ScopeMax);
// Load from an Id and return it
Id createLoad(Id lValue);
Id createLoad(Id lValue, spv::MemoryAccessMask memoryAccess = spv::MemoryAccessMaskNone, spv::Scope scope = spv::ScopeMax);
// Create an OpAccessChain instruction
Id createAccessChain(StorageClass, Id base, const std::vector<Id>& offsets);
@ -296,6 +296,7 @@ public:
void createNoResultOp(Op);
void createNoResultOp(Op, Id operand);
void createNoResultOp(Op, const std::vector<Id>& operands);
void createNoResultOp(Op, const std::vector<IdImmediate>& operands);
void createControlBarrier(Scope execution, Scope memory, MemorySemanticsMask);
void createMemoryBarrier(unsigned executionScope, unsigned memorySemantics);
@ -365,6 +366,8 @@ public:
Id component;
Id texelOut;
Id lodClamp;
bool nonprivate;
bool volatil;
};
// Select the correct texture operation based on all inputs, and emit the correct instruction
@ -504,6 +507,43 @@ public:
Id component; // a dynamic component index, can coexist with a swizzle, done after the swizzle, NoResult if not present
Id preSwizzleBaseType; // dereferenced type, before swizzle or component is applied; NoType unless a swizzle or component is present
bool isRValue; // true if 'base' is an r-value, otherwise, base is an l-value
// Accumulate whether anything in the chain of structures has coherent decorations.
struct CoherentFlags {
unsigned coherent : 1;
unsigned devicecoherent : 1;
unsigned queuefamilycoherent : 1;
unsigned workgroupcoherent : 1;
unsigned subgroupcoherent : 1;
unsigned nonprivate : 1;
unsigned volatil : 1;
unsigned isImage : 1;
void clear() {
coherent = 0;
devicecoherent = 0;
queuefamilycoherent = 0;
workgroupcoherent = 0;
subgroupcoherent = 0;
nonprivate = 0;
volatil = 0;
isImage = 0;
}
CoherentFlags() { clear(); }
CoherentFlags operator |=(const CoherentFlags &other) {
coherent |= other.coherent;
devicecoherent |= other.devicecoherent;
queuefamilycoherent |= other.queuefamilycoherent;
workgroupcoherent |= other.workgroupcoherent;
subgroupcoherent |= other.subgroupcoherent;
nonprivate |= other.nonprivate;
volatil |= other.volatil;
isImage |= other.isImage;
return *this;
}
};
CoherentFlags coherentFlags;
};
//
@ -533,9 +573,10 @@ public:
}
// push offset onto the end of the chain
void accessChainPush(Id offset)
void accessChainPush(Id offset, AccessChain::CoherentFlags coherentFlags)
{
accessChain.indexChain.push_back(offset);
accessChain.coherentFlags |= coherentFlags;
}
// push new swizzle onto the end of any existing swizzle, merging into a single swizzle
@ -553,10 +594,10 @@ public:
}
// use accessChain and swizzle to store value
void accessChainStore(Id rvalue);
void accessChainStore(Id rvalue, spv::MemoryAccessMask memoryAccess = spv::MemoryAccessMaskNone, spv::Scope scope = spv::ScopeMax);
// use accessChain and swizzle to load an r-value
Id accessChainLoad(Decoration precision, Decoration nonUniform, Id ResultType);
Id accessChainLoad(Decoration precision, Decoration nonUniform, Id ResultType, spv::MemoryAccessMask memoryAccess = spv::MemoryAccessMaskNone, spv::Scope scope = spv::ScopeMax);
// get the direct pointer for an l-value
Id accessChainGetLValue();

Просмотреть файл

@ -535,6 +535,11 @@ void SpirvStream::disassembleInstruction(Id resultId, Id /*typeId*/, Op opCode,
case OperandLiteralString:
numOperands -= disassembleString();
break;
case OperandMemoryAccess:
outputMask(OperandMemoryAccess, stream[word++]);
--numOperands;
disassembleIds(numOperands);
return;
default:
assert(operandClass >= OperandSource && operandClass < OperandOpcode);

Просмотреть файл

@ -117,9 +117,10 @@ const char* AddressingString(int addr)
const char* MemoryString(int mem)
{
switch (mem) {
case 0: return "Simple";
case 1: return "GLSL450";
case 2: return "OpenCL";
case MemoryModelSimple: return "Simple";
case MemoryModelGLSL450: return "GLSL450";
case MemoryModelOpenCL: return "OpenCL";
case MemoryModelVulkanKHR: return "VulkanKHR";
default: return "Bad";
}
@ -499,19 +500,23 @@ const char* ImageChannelDataTypeString(int type)
}
}
const int ImageOperandsCeiling = 8;
const int ImageOperandsCeiling = 12;
const char* ImageOperandsString(int format)
{
switch (format) {
case 0: return "Bias";
case 1: return "Lod";
case 2: return "Grad";
case 3: return "ConstOffset";
case 4: return "Offset";
case 5: return "ConstOffsets";
case 6: return "Sample";
case 7: return "MinLod";
case ImageOperandsBiasShift: return "Bias";
case ImageOperandsLodShift: return "Lod";
case ImageOperandsGradShift: return "Grad";
case ImageOperandsConstOffsetShift: return "ConstOffset";
case ImageOperandsOffsetShift: return "Offset";
case ImageOperandsConstOffsetsShift: return "ConstOffsets";
case ImageOperandsSampleShift: return "Sample";
case ImageOperandsMinLodShift: return "MinLod";
case ImageOperandsMakeTexelAvailableKHRShift: return "MakeTexelAvailableKHR";
case ImageOperandsMakeTexelVisibleKHRShift: return "MakeTexelVisibleKHR";
case ImageOperandsNonPrivateTexelKHRShift: return "NonPrivateTexelKHR";
case ImageOperandsVolatileTexelKHRShift: return "VolatileTexelKHR";
case ImageOperandsCeiling:
default:
@ -645,12 +650,17 @@ const char* MemorySemanticsString(int mem)
}
}
const int MemoryAccessCeiling = 6;
const char* MemoryAccessString(int mem)
{
switch (mem) {
case 0: return "Volatile";
case 1: return "Aligned";
case 2: return "Nontemporal";
case MemoryAccessVolatileShift: return "Volatile";
case MemoryAccessAlignedShift: return "Aligned";
case MemoryAccessNontemporalShift: return "Nontemporal";
case MemoryAccessMakePointerAvailableKHRShift: return "MakePointerAvailableKHR";
case MemoryAccessMakePointerVisibleKHRShift: return "MakePointerVisibleKHR";
case MemoryAccessNonPrivatePointerKHRShift: return "NonPrivatePointerKHR";
default: return "Bad";
}
@ -833,6 +843,9 @@ const char* CapabilityString(int info)
case CapabilityUniformTexelBufferArrayNonUniformIndexingEXT: return "CapabilityUniformTexelBufferArrayNonUniformIndexingEXT";
case CapabilityStorageTexelBufferArrayNonUniformIndexingEXT: return "CapabilityStorageTexelBufferArrayNonUniformIndexingEXT";
case CapabilityVulkanMemoryModelKHR: return "CapabilityVulkanMemoryModelKHR";
case CapabilityVulkanMemoryModelDeviceScopeKHR: return "CapabilityVulkanMemoryModelDeviceScopeKHR";
default: return "Bad";
}
}
@ -1245,6 +1258,7 @@ EnumParameters DecorationParams[DecorationCeiling];
EnumParameters LoopControlParams[FunctionControlCeiling];
EnumParameters SelectionControlParams[SelectControlCeiling];
EnumParameters FunctionControlParams[FunctionControlCeiling];
EnumParameters MemoryAccessParams[MemoryAccessCeiling];
// Set up all the parameterizing descriptions of the opcodes, operands, etc.
void Parameterize()
@ -1400,7 +1414,7 @@ void Parameterize()
OperandClassParams[OperandLoop].set(LoopControlCeiling, LoopControlString, LoopControlParams, true);
OperandClassParams[OperandFunction].set(FunctionControlCeiling, FunctionControlString, FunctionControlParams, true);
OperandClassParams[OperandMemorySemantics].set(0, MemorySemanticsString, nullptr, true);
OperandClassParams[OperandMemoryAccess].set(0, MemoryAccessString, nullptr, true);
OperandClassParams[OperandMemoryAccess].set(MemoryAccessCeiling, MemoryAccessString, MemoryAccessParams, true);
OperandClassParams[OperandScope].set(0, ScopeString, nullptr);
OperandClassParams[OperandGroupOperation].set(0, GroupOperationString, nullptr);
OperandClassParams[OperandKernelEnqueueFlags].set(0, KernelEnqueueFlagsString, nullptr);
@ -1522,10 +1536,14 @@ void Parameterize()
InstructionDesc[OpLoad].operands.push(OperandId, "'Pointer'");
InstructionDesc[OpLoad].operands.push(OperandMemoryAccess, "", true);
InstructionDesc[OpLoad].operands.push(OperandLiteralNumber, "", true);
InstructionDesc[OpLoad].operands.push(OperandId, "", true);
InstructionDesc[OpStore].operands.push(OperandId, "'Pointer'");
InstructionDesc[OpStore].operands.push(OperandId, "'Object'");
InstructionDesc[OpStore].operands.push(OperandMemoryAccess, "", true);
InstructionDesc[OpStore].operands.push(OperandLiteralNumber, "", true);
InstructionDesc[OpStore].operands.push(OperandId, "", true);
InstructionDesc[OpPhi].operands.push(OperandVariableIds, "'Variable, Parent, ...'");

Просмотреть файл

@ -87,6 +87,7 @@ enum MemoryModel {
MemoryModelSimple = 0,
MemoryModelGLSL450 = 1,
MemoryModelOpenCL = 2,
MemoryModelVulkanKHR = 3,
MemoryModelMax = 0x7fffffff,
};
@ -275,6 +276,10 @@ enum ImageOperandsShift {
ImageOperandsConstOffsetsShift = 5,
ImageOperandsSampleShift = 6,
ImageOperandsMinLodShift = 7,
ImageOperandsMakeTexelAvailableKHRShift = 8,
ImageOperandsMakeTexelVisibleKHRShift = 9,
ImageOperandsNonPrivateTexelKHRShift = 10,
ImageOperandsVolatileTexelKHRShift = 11,
ImageOperandsMax = 0x7fffffff,
};
@ -288,6 +293,10 @@ enum ImageOperandsMask {
ImageOperandsConstOffsetsMask = 0x00000020,
ImageOperandsSampleMask = 0x00000040,
ImageOperandsMinLodMask = 0x00000080,
ImageOperandsMakeTexelAvailableKHRMask = 0x00000100,
ImageOperandsMakeTexelVisibleKHRMask = 0x00000200,
ImageOperandsNonPrivateTexelKHRMask = 0x00000400,
ImageOperandsVolatileTexelKHRMask = 0x00000800,
};
enum FPFastMathModeShift {
@ -528,6 +537,9 @@ enum MemorySemanticsShift {
MemorySemanticsCrossWorkgroupMemoryShift = 9,
MemorySemanticsAtomicCounterMemoryShift = 10,
MemorySemanticsImageMemoryShift = 11,
MemorySemanticsOutputMemoryKHRShift = 12,
MemorySemanticsMakeAvailableKHRShift = 13,
MemorySemanticsMakeVisibleKHRShift = 14,
MemorySemanticsMax = 0x7fffffff,
};
@ -543,12 +555,18 @@ enum MemorySemanticsMask {
MemorySemanticsCrossWorkgroupMemoryMask = 0x00000200,
MemorySemanticsAtomicCounterMemoryMask = 0x00000400,
MemorySemanticsImageMemoryMask = 0x00000800,
MemorySemanticsOutputMemoryKHRMask = 0x00001000,
MemorySemanticsMakeAvailableKHRMask = 0x00002000,
MemorySemanticsMakeVisibleKHRMask = 0x00004000,
};
enum MemoryAccessShift {
MemoryAccessVolatileShift = 0,
MemoryAccessAlignedShift = 1,
MemoryAccessNontemporalShift = 2,
MemoryAccessMakePointerAvailableKHRShift = 3,
MemoryAccessMakePointerVisibleKHRShift = 4,
MemoryAccessNonPrivatePointerKHRShift = 5,
MemoryAccessMax = 0x7fffffff,
};
@ -557,6 +575,9 @@ enum MemoryAccessMask {
MemoryAccessVolatileMask = 0x00000001,
MemoryAccessAlignedMask = 0x00000002,
MemoryAccessNontemporalMask = 0x00000004,
MemoryAccessMakePointerAvailableKHRMask = 0x00000008,
MemoryAccessMakePointerVisibleKHRMask = 0x00000010,
MemoryAccessNonPrivatePointerKHRMask = 0x00000020,
};
enum Scope {
@ -565,6 +586,7 @@ enum Scope {
ScopeWorkgroup = 2,
ScopeSubgroup = 3,
ScopeInvocation = 4,
ScopeQueueFamilyKHR = 5,
ScopeMax = 0x7fffffff,
};
@ -708,6 +730,8 @@ enum Capability {
CapabilityInputAttachmentArrayNonUniformIndexingEXT = 5310,
CapabilityUniformTexelBufferArrayNonUniformIndexingEXT = 5311,
CapabilityStorageTexelBufferArrayNonUniformIndexingEXT = 5312,
CapabilityVulkanMemoryModelKHR = 5345,
CapabilityVulkanMemoryModelDeviceScopeKHR = 5346,
CapabilitySubgroupShuffleINTEL = 5568,
CapabilitySubgroupBufferBlockIOINTEL = 5569,
CapabilitySubgroupImageBlockIOINTEL = 5570,

Просмотреть файл

@ -0,0 +1,243 @@
spv.memoryScopeSemantics.comp
error: SPIRV-Tools Validation Errors
error: Capability Int64Atomics is not allowed by Vulkan 1.0 specification (or requires extension)
OpCapability Int64Atomics
// Module Version 10000
// Generated by (magic number): 80007
// Id's are bound by 142
Capability Shader
Capability Int64
Capability Int64Atomics
Capability CapabilityVulkanMemoryModelKHR
Capability CapabilityVulkanMemoryModelDeviceScopeKHR
Extension "SPV_KHR_vulkan_memory_model"
1: ExtInstImport "GLSL.std.450"
MemoryModel Logical VulkanKHR
EntryPoint GLCompute 4 "main"
ExecutionMode 4 LocalSize 1 1 1
Source GLSL 450
SourceExtension "GL_ARB_gpu_shader_int64"
SourceExtension "GL_KHR_memory_scope_semantics"
Name 4 "main"
Name 8 "origi"
Name 10 "atomi"
Name 21 "origu"
Name 23 "atomu"
Name 24 "value"
Name 36 "imagei"
Name 45 "imageu"
Name 65 "BufferU"
MemberName 65(BufferU) 0 "x"
Name 67 "bufferu"
Name 72 "y"
Name 77 "BufferI"
MemberName 77(BufferI) 0 "x"
Name 79 "bufferi"
Name 83 "A"
MemberName 83(A) 0 "x"
Name 84 "BufferJ"
MemberName 84(BufferJ) 0 "a"
Name 87 "bufferj"
Name 98 "BufferK"
MemberName 98(BufferK) 0 "x"
Name 100 "bufferk"
Name 109 "imagej"
Name 121 "samp"
Name 132 "atomu64"
Name 137 "atomi64"
Decorate 36(imagei) DescriptorSet 0
Decorate 36(imagei) Binding 1
Decorate 45(imageu) DescriptorSet 0
Decorate 45(imageu) Binding 0
MemberDecorate 65(BufferU) 0 Offset 0
Decorate 65(BufferU) BufferBlock
Decorate 67(bufferu) DescriptorSet 0
Decorate 67(bufferu) Binding 2
MemberDecorate 77(BufferI) 0 Offset 0
Decorate 77(BufferI) BufferBlock
Decorate 79(bufferi) DescriptorSet 0
Decorate 79(bufferi) Binding 3
Decorate 82 ArrayStride 4
MemberDecorate 83(A) 0 Offset 0
MemberDecorate 84(BufferJ) 0 Offset 0
Decorate 84(BufferJ) BufferBlock
Decorate 87(bufferj) DescriptorSet 0
Decorate 87(bufferj) Binding 4
MemberDecorate 98(BufferK) 0 Offset 0
Decorate 98(BufferK) Block
Decorate 100(bufferk) DescriptorSet 0
Decorate 100(bufferk) Binding 7
Decorate 109(imagej) DescriptorSet 0
Decorate 109(imagej) Binding 5
Decorate 121(samp) DescriptorSet 0
Decorate 121(samp) Binding 6
2: TypeVoid
3: TypeFunction 2
6: TypeInt 32 1
7: TypePointer Function 6(int)
9: TypePointer Workgroup 6(int)
10(atomi): 9(ptr) Variable Workgroup
11: 6(int) Constant 3
12: 6(int) Constant 1
13: 6(int) Constant 320
14: 6(int) Constant 4
15: TypeInt 32 0
16: 15(int) Constant 5
17: 15(int) Constant 0
18: 15(int) Constant 324
20: TypePointer Function 15(int)
22: TypePointer Workgroup 15(int)
23(atomu): 22(ptr) Variable Workgroup
24(value): 22(ptr) Variable Workgroup
26: 15(int) Constant 2
28: 6(int) Constant 64
29: 6(int) Constant 2
30: 15(int) Constant 66
33: 15(int) Constant 68
34: TypeImage 6(int) 2D nonsampled format:R32i
35: TypePointer UniformConstant 34
36(imagei): 35(ptr) Variable UniformConstant
37: TypeVector 6(int) 2
38: 6(int) Constant 0
39: 37(ivec2) ConstantComposite 38 38
40: TypePointer Image 6(int)
43: TypeImage 15(int) 2D nonsampled format:R32ui
44: TypePointer UniformConstant 43
45(imageu): 44(ptr) Variable UniformConstant
46: 15(int) Constant 3
47: TypePointer Image 15(int)
50: 15(int) Constant 4
52: 15(int) Constant 7
57: 6(int) Constant 7
61: 15(int) Constant 10
63: 15(int) Constant 322
65(BufferU): TypeStruct 15(int)
66: TypePointer Uniform 65(BufferU)
67(bufferu): 66(ptr) Variable Uniform
68: TypePointer Uniform 15(int)
70: 15(int) Constant 1
77(BufferI): TypeStruct 15(int)
78: TypePointer Uniform 77(BufferI)
79(bufferi): 78(ptr) Variable Uniform
82: TypeArray 15(int) 26
83(A): TypeStruct 82
84(BufferJ): TypeStruct 83(A)
85: TypeArray 84(BufferJ) 26
86: TypePointer Uniform 85
87(bufferj): 86(ptr) Variable Uniform
94: TypePointer Uniform 83(A)
98(BufferK): TypeStruct 15(int)
99: TypePointer Uniform 98(BufferK)
100(bufferk): 99(ptr) Variable Uniform
105: TypeVector 6(int) 4
107: TypeArray 34 26
108: TypePointer UniformConstant 107
109(imagej): 108(ptr) Variable UniformConstant
115: 105(ivec4) ConstantComposite 38 38 38 38
116: TypeFloat 32
117: TypeImage 116(float) 2D sampled format:Unknown
118: TypeSampledImage 117
119: TypeArray 118 26
120: TypePointer UniformConstant 119
121(samp): 120(ptr) Variable UniformConstant
122: TypePointer UniformConstant 118
125: TypeVector 116(float) 2
126: 116(float) Constant 0
127: 125(fvec2) ConstantComposite 126 126
128: TypeVector 116(float) 4
130: TypeInt 64 0
131: TypePointer Workgroup 130(int64_t)
132(atomu64): 131(ptr) Variable Workgroup
133:130(int64_t) Constant 7 0
135: TypeInt 64 1
136: TypePointer Workgroup 135(int64_t)
137(atomi64): 136(ptr) Variable Workgroup
138:135(int64_t) Constant 10 0
4(main): 2 Function None 3
5: Label
8(origi): 7(ptr) Variable Function
21(origu): 20(ptr) Variable Function
72(y): 20(ptr) Variable Function
19: 6(int) AtomicIAdd 10(atomi) 12 18 11
Store 8(origi) 19
25: 15(int) Load 24(value) MakePointerVisibleKHR 26
27: 15(int) AtomicAnd 23(atomu) 16 17 25
Store 21(origu) 27
31: 6(int) AtomicLoad 10(atomi) 12 30
Store 8(origi) 31
32: 15(int) Load 24(value) MakePointerVisibleKHR 26
AtomicStore 23(atomu) 12 33 32
41: 40(ptr) ImageTexelPointer 36(imagei) 39 17
42: 6(int) AtomicLoad 41 12 30
Store 8(origi) 42
48: 47(ptr) ImageTexelPointer 45(imageu) 39 17
49: 15(int) AtomicIAdd 48 12 30 46
Store 21(origu) 49
51: 47(ptr) ImageTexelPointer 45(imageu) 39 17
AtomicStore 51 12 33 50
53: 15(int) AtomicOr 23(atomu) 12 17 52
Store 21(origu) 53
54: 15(int) AtomicXor 23(atomu) 12 17 52
Store 21(origu) 54
55: 15(int) Load 24(value) MakePointerVisibleKHR 26
56: 15(int) AtomicUMin 23(atomu) 12 17 55
Store 21(origu) 56
58: 6(int) AtomicSMax 10(atomi) 12 17 57
Store 8(origi) 58
59: 6(int) Load 8(origi)
60: 6(int) AtomicExchange 10(atomi) 12 17 59
Store 8(origi) 60
62: 15(int) Load 24(value) MakePointerVisibleKHR 26
64: 15(int) AtomicCompareExchange 23(atomu) 12 63 63 62 61
Store 21(origu) 64
69: 68(ptr) AccessChain 67(bufferu) 38
71: 15(int) AtomicIAdd 69 12 18 70
MemoryBarrier 26 18
ControlBarrier 26 26 63
ControlBarrier 26 26 17
73: 68(ptr) AccessChain 67(bufferu) 38
74: 15(int) Load 73 MakePointerVisibleKHR NonPrivatePointerKHR 26
Store 72(y) 74
75: 15(int) Load 72(y)
76: 68(ptr) AccessChain 67(bufferu) 38
Store 76 75 MakePointerAvailableKHR NonPrivatePointerKHR 26
80: 68(ptr) AccessChain 79(bufferi) 38
81: 15(int) Load 80 MakePointerVisibleKHR NonPrivatePointerKHR 16
Store 72(y) 81
88: 68(ptr) AccessChain 87(bufferj) 38 38 38 12
89: 15(int) Load 88 Volatile MakePointerVisibleKHR NonPrivatePointerKHR 46
Store 72(y) 89
90: 15(int) Load 72(y)
91: 68(ptr) AccessChain 79(bufferi) 38
Store 91 90 MakePointerAvailableKHR NonPrivatePointerKHR 16
92: 15(int) Load 72(y)
93: 68(ptr) AccessChain 87(bufferj) 38 38 38 12
Store 93 92 Volatile MakePointerAvailableKHR NonPrivatePointerKHR 46
95: 94(ptr) AccessChain 87(bufferj) 12 38
96: 83(A) Load 95 Volatile MakePointerVisibleKHR NonPrivatePointerKHR 46
97: 94(ptr) AccessChain 87(bufferj) 38 38
Store 97 96 Volatile MakePointerAvailableKHR NonPrivatePointerKHR 46
101: 68(ptr) AccessChain 100(bufferk) 38
102: 15(int) Load 101 NonPrivatePointerKHR
103: 68(ptr) AccessChain 79(bufferi) 38
Store 103 102 MakePointerAvailableKHR NonPrivatePointerKHR 16
104: 34 Load 36(imagei)
106: 105(ivec4) ImageRead 104 39 MakeTexelVisibleKHR NonPrivateTexelKHR VolatileTexelKHR 16
110: 35(ptr) AccessChain 109(imagej) 38
111: 34 Load 110
112: 105(ivec4) ImageRead 111 39 NonPrivateTexelKHR
113: 35(ptr) AccessChain 109(imagej) 12
114: 34 Load 113
ImageWrite 114 39 115 NonPrivateTexelKHR
123: 122(ptr) AccessChain 121(samp) 38
124: 118 Load 123
129: 128(fvec4) ImageSampleExplicitLod 124 127 Lod NonPrivateTexelKHR 126
134:130(int64_t) AtomicUMax 132(atomu64) 12 17 133
Store 132(atomu64) 134 MakePointerAvailableKHR 26
139:130(int64_t) Load 132(atomu64) MakePointerVisibleKHR 26
140:135(int64_t) Bitcast 139
141:135(int64_t) AtomicCompareExchange 137(atomi64) 12 63 63 140 138
Return
FunctionEnd

Просмотреть файл

@ -0,0 +1,17 @@
spv.memoryScopeSemantics_Error.comp
ERROR: 0:15: 'atomicStore' : gl_SemanticsAcquire must not be used with (image) atomic store
ERROR: 0:16: 'imageAtomicLoad' : gl_SemanticsRelease must not be used with (image) atomic load
ERROR: 0:17: 'atomicStore' : gl_SemanticsAcquireRelease must not be used with (image) atomic load/store
ERROR: 0:18: 'atomicStore' : Invalid semantics value
ERROR: 0:19: 'imageAtomicLoad' : Invalid storage class semantics value
ERROR: 0:20: 'memoryBarrier' : Semantics must include exactly one of gl_SemanticsRelease, gl_SemanticsAcquire, or gl_SemanticsAcquireRelease
ERROR: 0:21: 'memoryBarrier' : Storage class semantics must not be zero
ERROR: 0:22: 'memoryBarrier' : Semantics must include exactly one of gl_SemanticsRelease, gl_SemanticsAcquire, or gl_SemanticsAcquireRelease
ERROR: 0:23: 'atomicAdd' : Semantics must not include multiple of gl_SemanticsRelease, gl_SemanticsAcquire, or gl_SemanticsAcquireRelease
ERROR: 0:24: 'atomicCompSwap' : semUnequal must not be gl_SemanticsRelease or gl_SemanticsAcquireRelease
ERROR: 0:25: 'memoryBarrier' : gl_SemanticsMakeVisible requires gl_SemanticsAcquire or gl_SemanticsAcquireRelease
ERROR: 0:26: 'memoryBarrier' : gl_SemanticsMakeAvailable requires gl_SemanticsRelease or gl_SemanticsAcquireRelease
ERROR: 12 compilation errors. No code generated.
SPIR-V is not generated for failed compile or link

Просмотреть файл

@ -11,7 +11,7 @@ spv.specConstant.vert
Source GLSL 400
Name 4 "main"
Name 9 "arraySize"
Name 14 "foo(vf4[s2543];"
Name 14 "foo(vf4[s2765];"
Name 13 "p"
Name 17 "builtin_spec_constant("
Name 20 "color"
@ -102,10 +102,10 @@ spv.specConstant.vert
Store 20(color) 46
48: 10 Load 22(ucol)
Store 47(param) 48
49: 2 FunctionCall 14(foo(vf4[s2543];) 47(param)
49: 2 FunctionCall 14(foo(vf4[s2765];) 47(param)
Return
FunctionEnd
14(foo(vf4[s2543];): 2 Function None 12
14(foo(vf4[s2765];): 2 Function None 12
13(p): 11(ptr) FunctionParameter
15: Label
54: 24(ptr) AccessChain 53(dupUcol) 23

Просмотреть файл

@ -0,0 +1,61 @@
#version 450
#extension GL_KHR_memory_scope_semantics : require
#extension GL_ARB_gpu_shader_int64 : require
#pragma use_vulkan_memory_model
shared uint value;
shared int atomi;
shared uint atomu;
layout(binding = 0, r32ui) workgroupcoherent uniform uimage2D imageu;
layout(binding = 1, r32i) volatile coherent uniform iimage2D imagei;
layout(binding = 5, r32i) nonprivate uniform iimage2D imagej[2];
layout (binding = 2) buffer BufferU { workgroupcoherent uint x; } bufferu;
layout (binding = 3) coherent buffer BufferI { uint x; } bufferi;
struct A { uint x[2]; };
layout (binding = 4) volatile buffer BufferJ { subgroupcoherent A a; } bufferj[2];
layout (binding = 6) nonprivate uniform sampler2D samp[2];
layout (binding = 7) nonprivate uniform BufferK { uint x; } bufferk;
shared uint64_t atomu64;
shared int64_t atomi64;
void main()
{
int origi = atomicAdd(atomi, 3, gl_ScopeDevice, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsRelease);
uint origu = atomicAnd(atomu, value);
origi = atomicLoad(atomi, gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsAcquire);
atomicStore(atomu, value, gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsRelease);
origi = imageAtomicLoad(imagei, ivec2(0,0), gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsAcquire);
origu = imageAtomicAdd(imageu, ivec2(0,0), 3u, gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsAcquire);
imageAtomicStore(imageu, ivec2(0,0), 4u, gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsRelease);
origu = atomicOr(atomu, 7u, gl_ScopeDevice, 0, 0);
origu = atomicXor(atomu, 7u, gl_ScopeDevice, 0, 0);
origu = atomicMin(atomu, value, gl_ScopeDevice, 0, 0);
origi = atomicMax(atomi, 7, gl_ScopeDevice, 0, 0);
origi = atomicExchange(atomi, origi, gl_ScopeDevice, 0, 0);
origu = atomicCompSwap(atomu, 10u, value, gl_ScopeDevice, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsAcquire, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsAcquire);
atomicAdd(bufferu.x, 1, gl_ScopeDevice, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsRelease);
memoryBarrier(gl_ScopeWorkgroup, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsRelease);
controlBarrier(gl_ScopeWorkgroup, gl_ScopeWorkgroup, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsAcquire);
controlBarrier(gl_ScopeWorkgroup, gl_ScopeWorkgroup, 0, 0);
uint y;
y = bufferu.x;
bufferu.x = y;
y = bufferi.x;
y = bufferj[0].a.x[1];
bufferi.x = y;
bufferj[0].a.x[1] = y;
bufferj[0].a = bufferj[1].a;
bufferi.x = bufferk.x;
imageLoad(imagei, ivec2(0,0));
imageLoad(imagej[0], ivec2(0,0));
imageStore(imagej[1], ivec2(0,0), ivec4(0,0,0,0));
texture(samp[0], vec2(0,0));
atomu64 = atomicMax(atomu64, uint64_t(7), gl_ScopeDevice, 0, 0);
atomicCompSwap(atomi64, int64_t(10), int64_t(atomu64), gl_ScopeDevice, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsAcquire, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsAcquire);
}

Просмотреть файл

@ -0,0 +1,28 @@
#version 450
#extension GL_KHR_memory_scope_semantics : require
shared uint value;
shared int atomi;
shared uint atomu;
layout(binding = 0, r32ui) workgroupcoherent uniform uimage2D imageu;
layout(binding = 1, r32i) coherent uniform iimage2D imagei;
layout (binding = 2) buffer BufferU { workgroupcoherent uint x; } bufferu;
layout (binding = 3) subgroupcoherent buffer BufferI { uint x; } bufferi;
void main()
{
atomicStore(atomu, value, gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsAcquire);
int origi = imageAtomicLoad(imagei, ivec2(0,0), gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsRelease);
atomicStore(atomu, value, gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsAcquireRelease);
atomicStore(atomu, value, gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_StorageSemanticsBuffer);
origi = imageAtomicLoad(imagei, ivec2(0,0), gl_ScopeDevice, gl_SemanticsAcquire, gl_SemanticsAcquire);
memoryBarrier(gl_ScopeWorkgroup, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, 0);
memoryBarrier(gl_ScopeWorkgroup, 0, gl_SemanticsRelease);
memoryBarrier(gl_ScopeWorkgroup, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsRelease | gl_SemanticsAcquire);
atomicAdd(atomu, value, gl_ScopeDevice, gl_StorageSemanticsBuffer, gl_SemanticsRelease | gl_SemanticsAcquire);
uint origu = atomicCompSwap(atomu, 10u, value, gl_ScopeDevice, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsAcquire, gl_StorageSemanticsBuffer | gl_StorageSemanticsShared, gl_SemanticsAcquireRelease);
memoryBarrier(gl_ScopeWorkgroup, gl_StorageSemanticsBuffer, gl_SemanticsRelease | gl_SemanticsMakeVisible);
memoryBarrier(gl_ScopeWorkgroup, gl_StorageSemanticsBuffer, gl_SemanticsAcquire | gl_SemanticsMakeAvailable);
}

Просмотреть файл

@ -462,6 +462,11 @@ public:
void clearMemory()
{
coherent = false;
devicecoherent = false;
queuefamilycoherent = false;
workgroupcoherent = false;
subgroupcoherent = false;
nonprivate = false;
volatil = false;
restrict = false;
readonly = false;
@ -499,6 +504,11 @@ public:
bool patch : 1;
bool sample : 1;
bool coherent : 1;
bool devicecoherent : 1;
bool queuefamilycoherent : 1;
bool workgroupcoherent : 1;
bool subgroupcoherent : 1;
bool nonprivate : 1;
bool volatil : 1;
bool restrict : 1;
bool readonly : 1;
@ -508,7 +518,11 @@ public:
bool isMemory() const
{
return coherent || volatil || restrict || readonly || writeonly;
return subgroupcoherent || workgroupcoherent || queuefamilycoherent || devicecoherent || coherent || volatil || restrict || readonly || writeonly || nonprivate;
}
bool isMemoryQualifierImageAndSSBOOnly() const
{
return subgroupcoherent || workgroupcoherent || queuefamilycoherent || devicecoherent || coherent || volatil || restrict || readonly || writeonly;
}
bool isInterpolation() const
{
@ -1713,6 +1727,16 @@ public:
appendStr(" sample");
if (qualifier.coherent)
appendStr(" coherent");
if (qualifier.devicecoherent)
appendStr(" devicecoherent");
if (qualifier.queuefamilycoherent)
appendStr(" queuefamilycoherent");
if (qualifier.workgroupcoherent)
appendStr(" workgroupcoherent");
if (qualifier.subgroupcoherent)
appendStr(" subgroupcoherent");
if (qualifier.nonprivate)
appendStr(" nonprivate");
if (qualifier.volatil)
appendStr(" volatile");
if (qualifier.restrict)

Просмотреть файл

@ -592,6 +592,8 @@ enum TOperator {
EOpAtomicXor,
EOpAtomicExchange,
EOpAtomicCompSwap,
EOpAtomicLoad,
EOpAtomicStore,
EOpAtomicCounterIncrement, // results in pre-increment value
EOpAtomicCounterDecrement, // results in post-decrement value
@ -784,6 +786,8 @@ enum TOperator {
EOpImageAtomicXor,
EOpImageAtomicExchange,
EOpImageAtomicCompSwap,
EOpImageAtomicLoad,
EOpImageAtomicStore,
EOpSubpassLoad,
EOpSubpassLoadMS,

Просмотреть файл

@ -935,27 +935,49 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
commonBuiltins.append(
"uint atomicAdd(coherent volatile inout uint, uint);"
" int atomicAdd(coherent volatile inout int, int);"
"uint atomicAdd(coherent volatile inout uint, uint, int, int, int);"
" int atomicAdd(coherent volatile inout int, int, int, int, int);"
"uint atomicMin(coherent volatile inout uint, uint);"
" int atomicMin(coherent volatile inout int, int);"
"uint atomicMin(coherent volatile inout uint, uint, int, int, int);"
" int atomicMin(coherent volatile inout int, int, int, int, int);"
"uint atomicMax(coherent volatile inout uint, uint);"
" int atomicMax(coherent volatile inout int, int);"
"uint atomicMax(coherent volatile inout uint, uint, int, int, int);"
" int atomicMax(coherent volatile inout int, int, int, int, int);"
"uint atomicAnd(coherent volatile inout uint, uint);"
" int atomicAnd(coherent volatile inout int, int);"
"uint atomicAnd(coherent volatile inout uint, uint, int, int, int);"
" int atomicAnd(coherent volatile inout int, int, int, int, int);"
"uint atomicOr (coherent volatile inout uint, uint);"
" int atomicOr (coherent volatile inout int, int);"
"uint atomicOr (coherent volatile inout uint, uint, int, int, int);"
" int atomicOr (coherent volatile inout int, int, int, int, int);"
"uint atomicXor(coherent volatile inout uint, uint);"
" int atomicXor(coherent volatile inout int, int);"
"uint atomicXor(coherent volatile inout uint, uint, int, int, int);"
" int atomicXor(coherent volatile inout int, int, int, int, int);"
"uint atomicExchange(coherent volatile inout uint, uint);"
" int atomicExchange(coherent volatile inout int, int);"
"uint atomicExchange(coherent volatile inout uint, uint, int, int, int);"
" int atomicExchange(coherent volatile inout int, int, int, int, int);"
"uint atomicCompSwap(coherent volatile inout uint, uint, uint);"
" int atomicCompSwap(coherent volatile inout int, int, int);"
"uint atomicCompSwap(coherent volatile inout uint, uint, uint, int, int, int, int, int);"
" int atomicCompSwap(coherent volatile inout int, int, int, int, int, int, int, int);"
"uint atomicLoad(coherent volatile in uint, int, int, int);"
" int atomicLoad(coherent volatile in int, int, int, int);"
"void atomicStore(coherent volatile out uint, uint, int, int, int);"
"void atomicStore(coherent volatile out int, int, int, int, int);"
"\n");
}
@ -965,23 +987,49 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
commonBuiltins.append(
"uint64_t atomicMin(coherent volatile inout uint64_t, uint64_t);"
" int64_t atomicMin(coherent volatile inout int64_t, int64_t);"
"uint64_t atomicMin(coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicMin(coherent volatile inout int64_t, int64_t, int, int, int);"
"uint64_t atomicMax(coherent volatile inout uint64_t, uint64_t);"
" int64_t atomicMax(coherent volatile inout int64_t, int64_t);"
"uint64_t atomicMax(coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicMax(coherent volatile inout int64_t, int64_t, int, int, int);"
"uint64_t atomicAnd(coherent volatile inout uint64_t, uint64_t);"
" int64_t atomicAnd(coherent volatile inout int64_t, int64_t);"
"uint64_t atomicAnd(coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicAnd(coherent volatile inout int64_t, int64_t, int, int, int);"
"uint64_t atomicOr (coherent volatile inout uint64_t, uint64_t);"
" int64_t atomicOr (coherent volatile inout int64_t, int64_t);"
"uint64_t atomicOr (coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicOr (coherent volatile inout int64_t, int64_t, int, int, int);"
"uint64_t atomicXor(coherent volatile inout uint64_t, uint64_t);"
" int64_t atomicXor(coherent volatile inout int64_t, int64_t);"
"uint64_t atomicXor(coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicXor(coherent volatile inout int64_t, int64_t, int, int, int);"
" int64_t atomicAdd(coherent volatile inout int64_t, int64_t);"
" int64_t atomicExchange(coherent volatile inout int64_t, int64_t);"
" int64_t atomicCompSwap(coherent volatile inout int64_t, int64_t, int64_t);"
"uint64_t atomicAdd(coherent volatile inout uint64_t, uint64_t);"
" int64_t atomicAdd(coherent volatile inout int64_t, int64_t);"
"uint64_t atomicAdd(coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicAdd(coherent volatile inout int64_t, int64_t, int, int, int);"
"uint64_t atomicExchange(coherent volatile inout uint64_t, uint64_t);"
" int64_t atomicExchange(coherent volatile inout int64_t, int64_t);"
"uint64_t atomicExchange(coherent volatile inout uint64_t, uint64_t, int, int, int);"
" int64_t atomicExchange(coherent volatile inout int64_t, int64_t, int, int, int);"
"uint64_t atomicCompSwap(coherent volatile inout uint64_t, uint64_t, uint64_t);"
" int64_t atomicCompSwap(coherent volatile inout int64_t, int64_t, int64_t);"
"uint64_t atomicCompSwap(coherent volatile inout uint64_t, uint64_t, uint64_t, int, int, int, int, int);"
" int64_t atomicCompSwap(coherent volatile inout int64_t, int64_t, int64_t, int, int, int, int, int);"
"uint64_t atomicLoad(coherent volatile in uint64_t, int, int, int);"
" int64_t atomicLoad(coherent volatile in int64_t, int, int, int);"
"void atomicStore(coherent volatile out uint64_t, uint64_t, int, int, int);"
"void atomicStore(coherent volatile out int64_t, int64_t, int, int, int);"
"\n");
}
#endif
@ -4693,6 +4741,9 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
);
}
commonBuiltins.append("void controlBarrier(int, int, int, int);\n"
"void memoryBarrier(int, int, int);\n");
//============================================================================
//
// Prototypes for built-in functions seen by fragment shaders only.
@ -5807,6 +5858,28 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
"\n");
}
if ((profile != EEsProfile && version >= 420) ||
(profile == EEsProfile && version >= 310)) {
commonBuiltins.append("const int gl_ScopeDevice = 1;\n");
commonBuiltins.append("const int gl_ScopeWorkgroup = 2;\n");
commonBuiltins.append("const int gl_ScopeSubgroup = 3;\n");
commonBuiltins.append("const int gl_ScopeInvocation = 4;\n");
commonBuiltins.append("const int gl_ScopeQueueFamily = 5;\n");
commonBuiltins.append("const int gl_SemanticsRelaxed = 0x0;\n");
commonBuiltins.append("const int gl_SemanticsAcquire = 0x2;\n");
commonBuiltins.append("const int gl_SemanticsRelease = 0x4;\n");
commonBuiltins.append("const int gl_SemanticsAcquireRelease = 0x8;\n");
commonBuiltins.append("const int gl_SemanticsMakeAvailable = 0x2000;\n");
commonBuiltins.append("const int gl_SemanticsMakeVisible = 0x4000;\n");
commonBuiltins.append("const int gl_StorageSemanticsNone = 0x0;\n");
commonBuiltins.append("const int gl_StorageSemanticsBuffer = 0x40;\n");
commonBuiltins.append("const int gl_StorageSemanticsShared = 0x100;\n");
commonBuiltins.append("const int gl_StorageSemanticsImage = 0x800;\n");
commonBuiltins.append("const int gl_StorageSemanticsOutput = 0x1000;\n");
}
// printf("%s\n", commonBuiltins.c_str());
// printf("%s\n", stageBuiltins[EShLangFragment].c_str());
}
@ -6106,23 +6179,44 @@ void TBuiltIns::addImageFunctions(TSampler sampler, const TString& typeName, int
" imageAtomicExchange(volatile coherent "
};
for (size_t i = 0; i < numBuiltins; ++i) {
// Loop twice to add prototypes with/without scope/semantics
for (int j = 0; j < 2; ++j) {
for (size_t i = 0; i < numBuiltins; ++i) {
commonBuiltins.append(dataType);
commonBuiltins.append(atomicFunc[i]);
commonBuiltins.append(imageParams);
commonBuiltins.append(", ");
commonBuiltins.append(dataType);
if (j == 1) {
commonBuiltins.append(", int, int, int");
}
commonBuiltins.append(");\n");
}
commonBuiltins.append(dataType);
commonBuiltins.append(atomicFunc[i]);
commonBuiltins.append(" imageAtomicCompSwap(volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", ");
commonBuiltins.append(dataType);
commonBuiltins.append(", ");
commonBuiltins.append(dataType);
if (j == 1) {
commonBuiltins.append(", int, int, int, int, int");
}
commonBuiltins.append(");\n");
}
commonBuiltins.append(dataType);
commonBuiltins.append(" imageAtomicCompSwap(volatile coherent ");
commonBuiltins.append(" imageAtomicLoad(volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", int, int, int);\n");
commonBuiltins.append("void imageAtomicStore(volatile coherent ");
commonBuiltins.append(imageParams);
commonBuiltins.append(", ");
commonBuiltins.append(dataType);
commonBuiltins.append(", ");
commonBuiltins.append(dataType);
commonBuiltins.append(");\n");
commonBuiltins.append(", int, int, int);\n");
} else {
// not int or uint
// GL_ARB_ES3_1_compatibility
@ -7969,6 +8063,26 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.setFunctionExtensions("shadow2DEXT", 1, &E_GL_EXT_shadow_samplers);
symbolTable.setFunctionExtensions("shadow2DProjEXT", 1, &E_GL_EXT_shadow_samplers);
}
if (spvVersion.vulkan > 0) {
symbolTable.setVariableExtensions("gl_ScopeDevice", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_ScopeWorkgroup", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_ScopeSubgroup", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_ScopeInvocation", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_SemanticsRelaxed", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_SemanticsAcquire", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_SemanticsRelease", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_SemanticsAcquireRelease", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_SemanticsMakeAvailable", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_SemanticsMakeVisible", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_StorageSemanticsNone", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_StorageSemanticsBuffer", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_StorageSemanticsShared", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_StorageSemanticsImage", 1, &E_GL_KHR_memory_scope_semantics);
symbolTable.setVariableExtensions("gl_StorageSemanticsOutput", 1, &E_GL_KHR_memory_scope_semantics);
}
break;
case EShLangCompute:
@ -8003,6 +8117,8 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.setFunctionExtensions("groupMemoryBarrier", 1, &E_GL_ARB_compute_shader);
}
symbolTable.setFunctionExtensions("controlBarrier", 1, &E_GL_KHR_memory_scope_semantics);
// GL_ARB_shader_ballot
if (profile != EEsProfile) {
symbolTable.setVariableExtensions("gl_SubGroupSizeARB", 1, &E_GL_ARB_shader_ballot);
@ -8213,6 +8329,7 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.relateToOperator("all", EOpAll);
symbolTable.relateToOperator("barrier", EOpBarrier);
symbolTable.relateToOperator("controlBarrier", EOpBarrier);
symbolTable.relateToOperator("memoryBarrier", EOpMemoryBarrier);
symbolTable.relateToOperator("memoryBarrierAtomicCounter", EOpMemoryBarrierAtomicCounter);
symbolTable.relateToOperator("memoryBarrierBuffer", EOpMemoryBarrierBuffer);
@ -8226,6 +8343,8 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.relateToOperator("atomicXor", EOpAtomicXor);
symbolTable.relateToOperator("atomicExchange", EOpAtomicExchange);
symbolTable.relateToOperator("atomicCompSwap", EOpAtomicCompSwap);
symbolTable.relateToOperator("atomicLoad", EOpAtomicLoad);
symbolTable.relateToOperator("atomicStore", EOpAtomicStore);
symbolTable.relateToOperator("atomicCounterIncrement", EOpAtomicCounterIncrement);
symbolTable.relateToOperator("atomicCounterDecrement", EOpAtomicCounterDecrement);
@ -8270,6 +8389,8 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.relateToOperator("imageAtomicXor", EOpImageAtomicXor);
symbolTable.relateToOperator("imageAtomicExchange", EOpImageAtomicExchange);
symbolTable.relateToOperator("imageAtomicCompSwap", EOpImageAtomicCompSwap);
symbolTable.relateToOperator("imageAtomicLoad", EOpImageAtomicLoad);
symbolTable.relateToOperator("imageAtomicStore", EOpImageAtomicStore);
symbolTable.relateToOperator("subpassLoad", EOpSubpassLoad);
symbolTable.relateToOperator("subpassLoadMS", EOpSubpassLoadMS);

Просмотреть файл

@ -267,6 +267,10 @@ void TParseContext::handlePragma(const TSourceLoc& loc, const TVector<TString>&
if (tokens.size() != 1)
error(loc, "extra tokens", "#pragma", "");
intermediate.setUseStorageBuffer();
} else if (spvVersion.spv > 0 && tokens[0].compare("use_vulkan_memory_model") == 0) {
if (tokens.size() != 1)
error(loc, "extra tokens", "#pragma", "");
intermediate.setUseVulkanMemoryModel();
} else if (tokens[0].compare("once") == 0) {
warn(loc, "not implemented", "#pragma once", "");
} else if (tokens[0].compare("glslang_binary_double_output") == 0)
@ -1028,8 +1032,16 @@ TIntermTyped* TParseContext::handleFunctionCall(const TSourceLoc& loc, TFunction
const char* message = "argument cannot drop memory qualifier when passed to formal parameter";
if (argQualifier.volatil && ! formalQualifier.volatil)
error(arguments->getLoc(), message, "volatile", "");
if (argQualifier.coherent && ! formalQualifier.coherent)
if (argQualifier.coherent && ! (formalQualifier.devicecoherent || formalQualifier.coherent))
error(arguments->getLoc(), message, "coherent", "");
if (argQualifier.devicecoherent && ! (formalQualifier.devicecoherent || formalQualifier.coherent))
error(arguments->getLoc(), message, "devicecoherent", "");
if (argQualifier.queuefamilycoherent && ! (formalQualifier.queuefamilycoherent || formalQualifier.devicecoherent || formalQualifier.coherent))
error(arguments->getLoc(), message, "queuefamilycoherent", "");
if (argQualifier.workgroupcoherent && ! (formalQualifier.workgroupcoherent || formalQualifier.queuefamilycoherent || formalQualifier.devicecoherent || formalQualifier.coherent))
error(arguments->getLoc(), message, "workgroupcoherent", "");
if (argQualifier.subgroupcoherent && ! (formalQualifier.subgroupcoherent || formalQualifier.workgroupcoherent || formalQualifier.queuefamilycoherent || formalQualifier.devicecoherent || formalQualifier.coherent))
error(arguments->getLoc(), message, "subgroupcoherent", "");
if (argQualifier.readonly && ! formalQualifier.readonly)
error(arguments->getLoc(), message, "readonly", "");
if (argQualifier.writeonly && ! formalQualifier.writeonly)
@ -1428,6 +1440,159 @@ TIntermTyped* TParseContext::addOutputArgumentConversions(const TFunction& funct
return conversionTree;
}
void TParseContext::memorySemanticsCheck(const TSourceLoc& loc, const TFunction& fnCandidate, const TIntermOperator& callNode)
{
const TIntermSequence* argp = &callNode.getAsAggregate()->getSequence();
const int gl_SemanticsRelaxed = 0x0;
const int gl_SemanticsAcquire = 0x2;
const int gl_SemanticsRelease = 0x4;
const int gl_SemanticsAcquireRelease = 0x8;
const int gl_SemanticsMakeAvailable = 0x2000;
const int gl_SemanticsMakeVisible = 0x4000;
const int gl_StorageSemanticsNone = 0x0;
const int gl_StorageSemanticsBuffer = 0x40;
const int gl_StorageSemanticsShared = 0x100;
const int gl_StorageSemanticsImage = 0x800;
const int gl_StorageSemanticsOutput = 0x1000;
unsigned int semantics = 0, storageClassSemantics = 0;
unsigned int semantics2 = 0, storageClassSemantics2 = 0;
// Grab the semantics and storage class semantics from the operands, based on opcode
switch (callNode.getOp()) {
case EOpAtomicAdd:
case EOpAtomicMin:
case EOpAtomicMax:
case EOpAtomicAnd:
case EOpAtomicOr:
case EOpAtomicXor:
case EOpAtomicExchange:
case EOpAtomicStore:
storageClassSemantics = (*argp)[3]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics = (*argp)[4]->getAsConstantUnion()->getConstArray()[0].getIConst();
break;
case EOpAtomicLoad:
storageClassSemantics = (*argp)[2]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics = (*argp)[3]->getAsConstantUnion()->getConstArray()[0].getIConst();
break;
case EOpAtomicCompSwap:
storageClassSemantics = (*argp)[4]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics = (*argp)[5]->getAsConstantUnion()->getConstArray()[0].getIConst();
storageClassSemantics2 = (*argp)[6]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics2 = (*argp)[7]->getAsConstantUnion()->getConstArray()[0].getIConst();
break;
case EOpImageAtomicAdd:
case EOpImageAtomicMin:
case EOpImageAtomicMax:
case EOpImageAtomicAnd:
case EOpImageAtomicOr:
case EOpImageAtomicXor:
case EOpImageAtomicExchange:
case EOpImageAtomicStore:
storageClassSemantics = (*argp)[4]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics = (*argp)[5]->getAsConstantUnion()->getConstArray()[0].getIConst();
break;
case EOpImageAtomicLoad:
storageClassSemantics = (*argp)[3]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics = (*argp)[4]->getAsConstantUnion()->getConstArray()[0].getIConst();
break;
case EOpImageAtomicCompSwap:
storageClassSemantics = (*argp)[5]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics = (*argp)[6]->getAsConstantUnion()->getConstArray()[0].getIConst();
storageClassSemantics2 = (*argp)[7]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics2 = (*argp)[8]->getAsConstantUnion()->getConstArray()[0].getIConst();
break;
case EOpBarrier:
storageClassSemantics = (*argp)[2]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics = (*argp)[3]->getAsConstantUnion()->getConstArray()[0].getIConst();
break;
case EOpMemoryBarrier:
storageClassSemantics = (*argp)[1]->getAsConstantUnion()->getConstArray()[0].getIConst();
semantics = (*argp)[2]->getAsConstantUnion()->getConstArray()[0].getIConst();
break;
}
if ((semantics & gl_SemanticsAcquire) &&
(callNode.getOp() == EOpAtomicStore || callNode.getOp() == EOpImageAtomicStore)) {
error(loc, "gl_SemanticsAcquire must not be used with (image) atomic store",
fnCandidate.getName().c_str(), "");
}
if ((semantics & gl_SemanticsRelease) &&
(callNode.getOp() == EOpAtomicLoad || callNode.getOp() == EOpImageAtomicLoad)) {
error(loc, "gl_SemanticsRelease must not be used with (image) atomic load",
fnCandidate.getName().c_str(), "");
}
if ((semantics & gl_SemanticsAcquireRelease) &&
(callNode.getOp() == EOpAtomicStore || callNode.getOp() == EOpImageAtomicStore ||
callNode.getOp() == EOpAtomicLoad || callNode.getOp() == EOpImageAtomicLoad)) {
error(loc, "gl_SemanticsAcquireRelease must not be used with (image) atomic load/store",
fnCandidate.getName().c_str(), "");
}
if (((semantics | semantics2) & ~(gl_SemanticsAcquire |
gl_SemanticsRelease |
gl_SemanticsAcquireRelease |
gl_SemanticsMakeAvailable |
gl_SemanticsMakeVisible))) {
error(loc, "Invalid semantics value", fnCandidate.getName().c_str(), "");
}
if (((storageClassSemantics | storageClassSemantics2) & ~(gl_StorageSemanticsBuffer |
gl_StorageSemanticsShared |
gl_StorageSemanticsImage |
gl_StorageSemanticsOutput))) {
error(loc, "Invalid storage class semantics value", fnCandidate.getName().c_str(), "");
}
if (callNode.getOp() == EOpMemoryBarrier) {
if (!IsPow2(semantics & (gl_SemanticsAcquire | gl_SemanticsRelease | gl_SemanticsAcquireRelease))) {
error(loc, "Semantics must include exactly one of gl_SemanticsRelease, gl_SemanticsAcquire, or "
"gl_SemanticsAcquireRelease", fnCandidate.getName().c_str(), "");
}
} else {
if (semantics & (gl_SemanticsAcquire | gl_SemanticsRelease | gl_SemanticsAcquireRelease)) {
if (!IsPow2(semantics & (gl_SemanticsAcquire | gl_SemanticsRelease | gl_SemanticsAcquireRelease))) {
error(loc, "Semantics must not include multiple of gl_SemanticsRelease, gl_SemanticsAcquire, or "
"gl_SemanticsAcquireRelease", fnCandidate.getName().c_str(), "");
}
}
if (semantics2 & (gl_SemanticsAcquire | gl_SemanticsRelease | gl_SemanticsAcquireRelease)) {
if (!IsPow2(semantics2 & (gl_SemanticsAcquire | gl_SemanticsRelease | gl_SemanticsAcquireRelease))) {
error(loc, "semUnequal must not include multiple of gl_SemanticsRelease, gl_SemanticsAcquire, or "
"gl_SemanticsAcquireRelease", fnCandidate.getName().c_str(), "");
}
}
}
if (callNode.getOp() == EOpMemoryBarrier) {
if (storageClassSemantics == 0) {
error(loc, "Storage class semantics must not be zero", fnCandidate.getName().c_str(), "");
}
}
if (callNode.getOp() == EOpBarrier && semantics != 0 && storageClassSemantics == 0) {
error(loc, "Storage class semantics must not be zero", fnCandidate.getName().c_str(), "");
}
if ((callNode.getOp() == EOpAtomicCompSwap || callNode.getOp() == EOpImageAtomicCompSwap) &&
(semantics2 & (gl_SemanticsRelease | gl_SemanticsAcquireRelease))) {
error(loc, "semUnequal must not be gl_SemanticsRelease or gl_SemanticsAcquireRelease",
fnCandidate.getName().c_str(), "");
}
if ((semantics & gl_SemanticsMakeAvailable) &&
!(semantics & (gl_SemanticsRelease | gl_SemanticsAcquireRelease))) {
error(loc, "gl_SemanticsMakeAvailable requires gl_SemanticsRelease or gl_SemanticsAcquireRelease",
fnCandidate.getName().c_str(), "");
}
if ((semantics & gl_SemanticsMakeVisible) &&
!(semantics & (gl_SemanticsAcquire | gl_SemanticsAcquireRelease))) {
error(loc, "gl_SemanticsMakeVisible requires gl_SemanticsAcquire or gl_SemanticsAcquireRelease",
fnCandidate.getName().c_str(), "");
}
}
//
// Do additional checking of built-in function calls that is not caught
// by normal semantic checks on argument type, extension tagging, etc.
@ -1656,6 +1821,8 @@ void TParseContext::builtInOpCheck(const TSourceLoc& loc, const TFunction& fnCan
case EOpImageAtomicXor:
case EOpImageAtomicExchange:
case EOpImageAtomicCompSwap:
case EOpImageAtomicLoad:
case EOpImageAtomicStore:
{
// Make sure the image types have the correct layout() format and correct argument types
const TType& imageType = arg0->getType();
@ -1669,10 +1836,14 @@ void TParseContext::builtInOpCheck(const TSourceLoc& loc, const TFunction& fnCan
error(loc, "only supported on image with format r32f", fnCandidate.getName().c_str(), "");
}
if (argp->size() > 4) {
requireExtensions(loc, 1, &E_GL_KHR_memory_scope_semantics, fnCandidate.getName().c_str());
memorySemanticsCheck(loc, fnCandidate, callNode);
}
break;
}
#ifdef NV_EXTENSIONS
case EOpAtomicAdd:
case EOpAtomicMin:
case EOpAtomicMax:
@ -1681,13 +1852,19 @@ void TParseContext::builtInOpCheck(const TSourceLoc& loc, const TFunction& fnCan
case EOpAtomicXor:
case EOpAtomicExchange:
case EOpAtomicCompSwap:
case EOpAtomicLoad:
case EOpAtomicStore:
{
if (arg0->getType().getBasicType() == EbtInt64 || arg0->getType().getBasicType() == EbtUint64)
if (argp->size() > 3) {
requireExtensions(loc, 1, &E_GL_KHR_memory_scope_semantics, fnCandidate.getName().c_str());
memorySemanticsCheck(loc, fnCandidate, callNode);
}
#ifdef NV_EXTENSIONS
else if (arg0->getType().getBasicType() == EbtInt64 || arg0->getType().getBasicType() == EbtUint64)
requireExtensions(loc, 1, &E_GL_NV_shader_atomic_int64, fnCandidate.getName().c_str());
#endif
break;
}
#endif
case EOpInterpolateAtCentroid:
case EOpInterpolateAtSample:
@ -1751,6 +1928,14 @@ void TParseContext::builtInOpCheck(const TSourceLoc& loc, const TFunction& fnCan
}
break;
case EOpBarrier:
case EOpMemoryBarrier:
if (argp->size() > 0) {
requireExtensions(loc, 1, &E_GL_KHR_memory_scope_semantics, fnCandidate.getName().c_str());
memorySemanticsCheck(loc, fnCandidate, callNode);
}
break;
default:
break;
}
@ -2806,8 +2991,11 @@ void TParseContext::globalQualifierTypeCheck(const TSourceLoc& loc, const TQuali
if (! symbolTable.atGlobalLevel())
return;
if (qualifier.isMemory() && ! publicType.isImage() && publicType.qualifier.storage != EvqBuffer)
if (qualifier.isMemoryQualifierImageAndSSBOOnly() && ! publicType.isImage() && publicType.qualifier.storage != EvqBuffer) {
error(loc, "memory qualifiers cannot be used on this type", "", "");
} else if (qualifier.isMemory() && (publicType.basicType != EbtSampler) && !publicType.qualifier.isUniformOrBuffer()) {
error(loc, "memory qualifiers cannot be used on this type", "", "");
}
if (qualifier.storage == EvqBuffer && publicType.basicType != EbtBlock)
error(loc, "buffers can be declared only as blocks", "buffer", "");
@ -3020,6 +3208,13 @@ void TParseContext::mergeQualifiers(const TSourceLoc& loc, TQualifier& dst, cons
if (dst.precision == EpqNone || (force && src.precision != EpqNone))
dst.precision = src.precision;
if (!force && ((src.coherent && (dst.devicecoherent || dst.queuefamilycoherent || dst.workgroupcoherent || dst.subgroupcoherent)) ||
(src.devicecoherent && (dst.coherent || dst.queuefamilycoherent || dst.workgroupcoherent || dst.subgroupcoherent)) ||
(src.queuefamilycoherent && (dst.coherent || dst.devicecoherent || dst.workgroupcoherent || dst.subgroupcoherent)) ||
(src.workgroupcoherent && (dst.coherent || dst.devicecoherent || dst.queuefamilycoherent || dst.subgroupcoherent)) ||
(src.subgroupcoherent && (dst.coherent || dst.devicecoherent || dst.queuefamilycoherent || dst.workgroupcoherent)))) {
error(loc, "only one coherent/devicecoherent/queuefamilycoherent/workgroupcoherent/subgroupcoherent qualifier allowed", GetPrecisionQualifierString(src.precision), "");
}
// Layout qualifiers
mergeObjectLayoutQualifiers(dst, src, false);
@ -3038,6 +3233,11 @@ void TParseContext::mergeQualifiers(const TSourceLoc& loc, TQualifier& dst, cons
MERGE_SINGLETON(patch);
MERGE_SINGLETON(sample);
MERGE_SINGLETON(coherent);
MERGE_SINGLETON(devicecoherent);
MERGE_SINGLETON(queuefamilycoherent);
MERGE_SINGLETON(workgroupcoherent);
MERGE_SINGLETON(subgroupcoherent);
MERGE_SINGLETON(nonprivate);
MERGE_SINGLETON(volatil);
MERGE_SINGLETON(restrict);
MERGE_SINGLETON(readonly);
@ -3862,6 +4062,11 @@ void TParseContext::paramCheckFix(const TSourceLoc& loc, const TQualifier& quali
if (qualifier.isMemory()) {
type.getQualifier().volatil = qualifier.volatil;
type.getQualifier().coherent = qualifier.coherent;
type.getQualifier().devicecoherent = qualifier.devicecoherent ;
type.getQualifier().queuefamilycoherent = qualifier.queuefamilycoherent;
type.getQualifier().workgroupcoherent = qualifier.workgroupcoherent;
type.getQualifier().subgroupcoherent = qualifier.subgroupcoherent;
type.getQualifier().nonprivate = qualifier.nonprivate;
type.getQualifier().readonly = qualifier.readonly;
type.getQualifier().writeonly = qualifier.writeonly;
type.getQualifier().restrict = qualifier.restrict;

Просмотреть файл

@ -323,6 +323,7 @@ public:
TFunction* handleConstructorCall(const TSourceLoc&, const TPublicType&);
void handlePrecisionQualifier(const TSourceLoc&, TQualifier&, TPrecisionQualifier);
void checkPrecisionQualifier(const TSourceLoc&, TPrecisionQualifier);
void memorySemanticsCheck(const TSourceLoc&, const TFunction&, const TIntermOperator& callNode);
void assignError(const TSourceLoc&, const char* op, TString left, TString right);
void unaryOpError(const TSourceLoc&, const char* op, TString operand);

Просмотреть файл

@ -380,6 +380,11 @@ void TScanContext::fillInKeywordMap()
(*KeywordMap)["varying"] = VARYING;
(*KeywordMap)["buffer"] = BUFFER;
(*KeywordMap)["coherent"] = COHERENT;
(*KeywordMap)["devicecoherent"] = DEVICECOHERENT;
(*KeywordMap)["queuefamilycoherent"] = QUEUEFAMILYCOHERENT;
(*KeywordMap)["workgroupcoherent"] = WORKGROUPCOHERENT;
(*KeywordMap)["subgroupcoherent"] = SUBGROUPCOHERENT;
(*KeywordMap)["nonprivate"] = NONPRIVATE;
(*KeywordMap)["restrict"] = RESTRICT;
(*KeywordMap)["readonly"] = READONLY;
(*KeywordMap)["writeonly"] = WRITEONLY;
@ -937,6 +942,11 @@ int TScanContext::tokenizeIdentifier()
return es30ReservedFromGLSL(420);
case COHERENT:
case DEVICECOHERENT:
case QUEUEFAMILYCOHERENT:
case WORKGROUPCOHERENT:
case SUBGROUPCOHERENT:
case NONPRIVATE:
case RESTRICT:
case READONLY:
case WRITEONLY:

Просмотреть файл

@ -194,6 +194,7 @@ void TParseVersions::initializeExtensionBehavior()
extensionBehavior[E_GL_KHR_shader_subgroup_shuffle_relative] = EBhDisable;
extensionBehavior[E_GL_KHR_shader_subgroup_clustered] = EBhDisable;
extensionBehavior[E_GL_KHR_shader_subgroup_quad] = EBhDisable;
extensionBehavior[E_GL_KHR_memory_scope_semantics] = EBhDisable;
extensionBehavior[E_GL_EXT_shader_non_constant_global_initializers] = EBhDisable;
extensionBehavior[E_GL_EXT_shader_image_load_formatted] = EBhDisable;

Просмотреть файл

@ -148,6 +148,7 @@ const char* const E_GL_KHR_shader_subgroup_shuffle = "GL_KHR_shader_sub
const char* const E_GL_KHR_shader_subgroup_shuffle_relative = "GL_KHR_shader_subgroup_shuffle_relative";
const char* const E_GL_KHR_shader_subgroup_clustered = "GL_KHR_shader_subgroup_clustered";
const char* const E_GL_KHR_shader_subgroup_quad = "GL_KHR_shader_subgroup_quad";
const char* const E_GL_KHR_memory_scope_semantics = "GL_KHR_memory_scope_semantics";
const char* const E_GL_EXT_shader_non_constant_global_initializers = "GL_EXT_shader_non_constant_global_initializers";
const char* const E_GL_EXT_shader_image_load_formatted = "GL_EXT_shader_image_load_formatted";

Просмотреть файл

@ -141,7 +141,7 @@ extern int yylex(YYSTYPE*, TParseContext&);
%token <lex> VEC2 VEC3 VEC4
%token <lex> MAT2 MAT3 MAT4 CENTROID IN OUT INOUT
%token <lex> UNIFORM PATCH SAMPLE BUFFER SHARED NONUNIFORM
%token <lex> COHERENT VOLATILE RESTRICT READONLY WRITEONLY
%token <lex> COHERENT VOLATILE RESTRICT READONLY WRITEONLY DEVICECOHERENT QUEUEFAMILYCOHERENT WORKGROUPCOHERENT SUBGROUPCOHERENT NONPRIVATE
%token <lex> DVEC2 DVEC3 DVEC4 DMAT2 DMAT3 DMAT4
%token <lex> F16VEC2 F16VEC3 F16VEC4 F16MAT2 F16MAT3 F16MAT4
%token <lex> F32VEC2 F32VEC3 F32VEC4 F32MAT2 F32MAT3 F32MAT4
@ -1317,6 +1317,31 @@ storage_qualifier
$$.init($1.loc);
$$.qualifier.coherent = true;
}
| DEVICECOHERENT {
$$.init($1.loc);
parseContext.requireExtensions($1.loc, 1, &E_GL_KHR_memory_scope_semantics, "devicecoherent");
$$.qualifier.devicecoherent = true;
}
| QUEUEFAMILYCOHERENT {
$$.init($1.loc);
parseContext.requireExtensions($1.loc, 1, &E_GL_KHR_memory_scope_semantics, "queuefamilycoherent");
$$.qualifier.queuefamilycoherent = true;
}
| WORKGROUPCOHERENT {
$$.init($1.loc);
parseContext.requireExtensions($1.loc, 1, &E_GL_KHR_memory_scope_semantics, "workgroupcoherent");
$$.qualifier.workgroupcoherent = true;
}
| SUBGROUPCOHERENT {
$$.init($1.loc);
parseContext.requireExtensions($1.loc, 1, &E_GL_KHR_memory_scope_semantics, "subgroupcoherent");
$$.qualifier.subgroupcoherent = true;
}
| NONPRIVATE {
$$.init($1.loc);
parseContext.requireExtensions($1.loc, 1, &E_GL_KHR_memory_scope_semantics, "nonprivate");
$$.qualifier.nonprivate = true;
}
| VOLATILE {
$$.init($1.loc);
$$.qualifier.volatil = true;

Разница между файлами не показана из-за своего большого размера Загрузить разницу

Просмотреть файл

@ -1,8 +1,8 @@
/* A Bison parser, made by GNU Bison 3.0. */
/* A Bison parser, made by GNU Bison 3.0.4. */
/* Bison interface for Yacc-like parsers in C
Copyright (C) 1984, 1989-1990, 2000-2013 Free Software Foundation, Inc.
Copyright (C) 1984, 1989-1990, 2000-2015 Free Software Foundation, Inc.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@ -130,315 +130,320 @@ extern int yydebug;
RESTRICT = 340,
READONLY = 341,
WRITEONLY = 342,
DVEC2 = 343,
DVEC3 = 344,
DVEC4 = 345,
DMAT2 = 346,
DMAT3 = 347,
DMAT4 = 348,
F16VEC2 = 349,
F16VEC3 = 350,
F16VEC4 = 351,
F16MAT2 = 352,
F16MAT3 = 353,
F16MAT4 = 354,
F32VEC2 = 355,
F32VEC3 = 356,
F32VEC4 = 357,
F32MAT2 = 358,
F32MAT3 = 359,
F32MAT4 = 360,
F64VEC2 = 361,
F64VEC3 = 362,
F64VEC4 = 363,
F64MAT2 = 364,
F64MAT3 = 365,
F64MAT4 = 366,
NOPERSPECTIVE = 367,
FLAT = 368,
SMOOTH = 369,
LAYOUT = 370,
EXPLICITINTERPAMD = 371,
MAT2X2 = 372,
MAT2X3 = 373,
MAT2X4 = 374,
MAT3X2 = 375,
MAT3X3 = 376,
MAT3X4 = 377,
MAT4X2 = 378,
MAT4X3 = 379,
MAT4X4 = 380,
DMAT2X2 = 381,
DMAT2X3 = 382,
DMAT2X4 = 383,
DMAT3X2 = 384,
DMAT3X3 = 385,
DMAT3X4 = 386,
DMAT4X2 = 387,
DMAT4X3 = 388,
DMAT4X4 = 389,
F16MAT2X2 = 390,
F16MAT2X3 = 391,
F16MAT2X4 = 392,
F16MAT3X2 = 393,
F16MAT3X3 = 394,
F16MAT3X4 = 395,
F16MAT4X2 = 396,
F16MAT4X3 = 397,
F16MAT4X4 = 398,
F32MAT2X2 = 399,
F32MAT2X3 = 400,
F32MAT2X4 = 401,
F32MAT3X2 = 402,
F32MAT3X3 = 403,
F32MAT3X4 = 404,
F32MAT4X2 = 405,
F32MAT4X3 = 406,
F32MAT4X4 = 407,
F64MAT2X2 = 408,
F64MAT2X3 = 409,
F64MAT2X4 = 410,
F64MAT3X2 = 411,
F64MAT3X3 = 412,
F64MAT3X4 = 413,
F64MAT4X2 = 414,
F64MAT4X3 = 415,
F64MAT4X4 = 416,
ATOMIC_UINT = 417,
SAMPLER1D = 418,
SAMPLER2D = 419,
SAMPLER3D = 420,
SAMPLERCUBE = 421,
SAMPLER1DSHADOW = 422,
SAMPLER2DSHADOW = 423,
SAMPLERCUBESHADOW = 424,
SAMPLER1DARRAY = 425,
SAMPLER2DARRAY = 426,
SAMPLER1DARRAYSHADOW = 427,
SAMPLER2DARRAYSHADOW = 428,
ISAMPLER1D = 429,
ISAMPLER2D = 430,
ISAMPLER3D = 431,
ISAMPLERCUBE = 432,
ISAMPLER1DARRAY = 433,
ISAMPLER2DARRAY = 434,
USAMPLER1D = 435,
USAMPLER2D = 436,
USAMPLER3D = 437,
USAMPLERCUBE = 438,
USAMPLER1DARRAY = 439,
USAMPLER2DARRAY = 440,
SAMPLER2DRECT = 441,
SAMPLER2DRECTSHADOW = 442,
ISAMPLER2DRECT = 443,
USAMPLER2DRECT = 444,
SAMPLERBUFFER = 445,
ISAMPLERBUFFER = 446,
USAMPLERBUFFER = 447,
SAMPLERCUBEARRAY = 448,
SAMPLERCUBEARRAYSHADOW = 449,
ISAMPLERCUBEARRAY = 450,
USAMPLERCUBEARRAY = 451,
SAMPLER2DMS = 452,
ISAMPLER2DMS = 453,
USAMPLER2DMS = 454,
SAMPLER2DMSARRAY = 455,
ISAMPLER2DMSARRAY = 456,
USAMPLER2DMSARRAY = 457,
SAMPLEREXTERNALOES = 458,
F16SAMPLER1D = 459,
F16SAMPLER2D = 460,
F16SAMPLER3D = 461,
F16SAMPLER2DRECT = 462,
F16SAMPLERCUBE = 463,
F16SAMPLER1DARRAY = 464,
F16SAMPLER2DARRAY = 465,
F16SAMPLERCUBEARRAY = 466,
F16SAMPLERBUFFER = 467,
F16SAMPLER2DMS = 468,
F16SAMPLER2DMSARRAY = 469,
F16SAMPLER1DSHADOW = 470,
F16SAMPLER2DSHADOW = 471,
F16SAMPLER1DARRAYSHADOW = 472,
F16SAMPLER2DARRAYSHADOW = 473,
F16SAMPLER2DRECTSHADOW = 474,
F16SAMPLERCUBESHADOW = 475,
F16SAMPLERCUBEARRAYSHADOW = 476,
SAMPLER = 477,
SAMPLERSHADOW = 478,
TEXTURE1D = 479,
TEXTURE2D = 480,
TEXTURE3D = 481,
TEXTURECUBE = 482,
TEXTURE1DARRAY = 483,
TEXTURE2DARRAY = 484,
ITEXTURE1D = 485,
ITEXTURE2D = 486,
ITEXTURE3D = 487,
ITEXTURECUBE = 488,
ITEXTURE1DARRAY = 489,
ITEXTURE2DARRAY = 490,
UTEXTURE1D = 491,
UTEXTURE2D = 492,
UTEXTURE3D = 493,
UTEXTURECUBE = 494,
UTEXTURE1DARRAY = 495,
UTEXTURE2DARRAY = 496,
TEXTURE2DRECT = 497,
ITEXTURE2DRECT = 498,
UTEXTURE2DRECT = 499,
TEXTUREBUFFER = 500,
ITEXTUREBUFFER = 501,
UTEXTUREBUFFER = 502,
TEXTURECUBEARRAY = 503,
ITEXTURECUBEARRAY = 504,
UTEXTURECUBEARRAY = 505,
TEXTURE2DMS = 506,
ITEXTURE2DMS = 507,
UTEXTURE2DMS = 508,
TEXTURE2DMSARRAY = 509,
ITEXTURE2DMSARRAY = 510,
UTEXTURE2DMSARRAY = 511,
F16TEXTURE1D = 512,
F16TEXTURE2D = 513,
F16TEXTURE3D = 514,
F16TEXTURE2DRECT = 515,
F16TEXTURECUBE = 516,
F16TEXTURE1DARRAY = 517,
F16TEXTURE2DARRAY = 518,
F16TEXTURECUBEARRAY = 519,
F16TEXTUREBUFFER = 520,
F16TEXTURE2DMS = 521,
F16TEXTURE2DMSARRAY = 522,
SUBPASSINPUT = 523,
SUBPASSINPUTMS = 524,
ISUBPASSINPUT = 525,
ISUBPASSINPUTMS = 526,
USUBPASSINPUT = 527,
USUBPASSINPUTMS = 528,
F16SUBPASSINPUT = 529,
F16SUBPASSINPUTMS = 530,
IMAGE1D = 531,
IIMAGE1D = 532,
UIMAGE1D = 533,
IMAGE2D = 534,
IIMAGE2D = 535,
UIMAGE2D = 536,
IMAGE3D = 537,
IIMAGE3D = 538,
UIMAGE3D = 539,
IMAGE2DRECT = 540,
IIMAGE2DRECT = 541,
UIMAGE2DRECT = 542,
IMAGECUBE = 543,
IIMAGECUBE = 544,
UIMAGECUBE = 545,
IMAGEBUFFER = 546,
IIMAGEBUFFER = 547,
UIMAGEBUFFER = 548,
IMAGE1DARRAY = 549,
IIMAGE1DARRAY = 550,
UIMAGE1DARRAY = 551,
IMAGE2DARRAY = 552,
IIMAGE2DARRAY = 553,
UIMAGE2DARRAY = 554,
IMAGECUBEARRAY = 555,
IIMAGECUBEARRAY = 556,
UIMAGECUBEARRAY = 557,
IMAGE2DMS = 558,
IIMAGE2DMS = 559,
UIMAGE2DMS = 560,
IMAGE2DMSARRAY = 561,
IIMAGE2DMSARRAY = 562,
UIMAGE2DMSARRAY = 563,
F16IMAGE1D = 564,
F16IMAGE2D = 565,
F16IMAGE3D = 566,
F16IMAGE2DRECT = 567,
F16IMAGECUBE = 568,
F16IMAGE1DARRAY = 569,
F16IMAGE2DARRAY = 570,
F16IMAGECUBEARRAY = 571,
F16IMAGEBUFFER = 572,
F16IMAGE2DMS = 573,
F16IMAGE2DMSARRAY = 574,
STRUCT = 575,
VOID = 576,
WHILE = 577,
IDENTIFIER = 578,
TYPE_NAME = 579,
FLOATCONSTANT = 580,
DOUBLECONSTANT = 581,
INT16CONSTANT = 582,
UINT16CONSTANT = 583,
INT32CONSTANT = 584,
UINT32CONSTANT = 585,
INTCONSTANT = 586,
UINTCONSTANT = 587,
INT64CONSTANT = 588,
UINT64CONSTANT = 589,
BOOLCONSTANT = 590,
FLOAT16CONSTANT = 591,
LEFT_OP = 592,
RIGHT_OP = 593,
INC_OP = 594,
DEC_OP = 595,
LE_OP = 596,
GE_OP = 597,
EQ_OP = 598,
NE_OP = 599,
AND_OP = 600,
OR_OP = 601,
XOR_OP = 602,
MUL_ASSIGN = 603,
DIV_ASSIGN = 604,
ADD_ASSIGN = 605,
MOD_ASSIGN = 606,
LEFT_ASSIGN = 607,
RIGHT_ASSIGN = 608,
AND_ASSIGN = 609,
XOR_ASSIGN = 610,
OR_ASSIGN = 611,
SUB_ASSIGN = 612,
LEFT_PAREN = 613,
RIGHT_PAREN = 614,
LEFT_BRACKET = 615,
RIGHT_BRACKET = 616,
LEFT_BRACE = 617,
RIGHT_BRACE = 618,
DOT = 619,
COMMA = 620,
COLON = 621,
EQUAL = 622,
SEMICOLON = 623,
BANG = 624,
DASH = 625,
TILDE = 626,
PLUS = 627,
STAR = 628,
SLASH = 629,
PERCENT = 630,
LEFT_ANGLE = 631,
RIGHT_ANGLE = 632,
VERTICAL_BAR = 633,
CARET = 634,
AMPERSAND = 635,
QUESTION = 636,
INVARIANT = 637,
PRECISE = 638,
HIGH_PRECISION = 639,
MEDIUM_PRECISION = 640,
LOW_PRECISION = 641,
PRECISION = 642,
PACKED = 643,
RESOURCE = 644,
SUPERP = 645
DEVICECOHERENT = 343,
QUEUEFAMILYCOHERENT = 344,
WORKGROUPCOHERENT = 345,
SUBGROUPCOHERENT = 346,
NONPRIVATE = 347,
DVEC2 = 348,
DVEC3 = 349,
DVEC4 = 350,
DMAT2 = 351,
DMAT3 = 352,
DMAT4 = 353,
F16VEC2 = 354,
F16VEC3 = 355,
F16VEC4 = 356,
F16MAT2 = 357,
F16MAT3 = 358,
F16MAT4 = 359,
F32VEC2 = 360,
F32VEC3 = 361,
F32VEC4 = 362,
F32MAT2 = 363,
F32MAT3 = 364,
F32MAT4 = 365,
F64VEC2 = 366,
F64VEC3 = 367,
F64VEC4 = 368,
F64MAT2 = 369,
F64MAT3 = 370,
F64MAT4 = 371,
NOPERSPECTIVE = 372,
FLAT = 373,
SMOOTH = 374,
LAYOUT = 375,
EXPLICITINTERPAMD = 376,
MAT2X2 = 377,
MAT2X3 = 378,
MAT2X4 = 379,
MAT3X2 = 380,
MAT3X3 = 381,
MAT3X4 = 382,
MAT4X2 = 383,
MAT4X3 = 384,
MAT4X4 = 385,
DMAT2X2 = 386,
DMAT2X3 = 387,
DMAT2X4 = 388,
DMAT3X2 = 389,
DMAT3X3 = 390,
DMAT3X4 = 391,
DMAT4X2 = 392,
DMAT4X3 = 393,
DMAT4X4 = 394,
F16MAT2X2 = 395,
F16MAT2X3 = 396,
F16MAT2X4 = 397,
F16MAT3X2 = 398,
F16MAT3X3 = 399,
F16MAT3X4 = 400,
F16MAT4X2 = 401,
F16MAT4X3 = 402,
F16MAT4X4 = 403,
F32MAT2X2 = 404,
F32MAT2X3 = 405,
F32MAT2X4 = 406,
F32MAT3X2 = 407,
F32MAT3X3 = 408,
F32MAT3X4 = 409,
F32MAT4X2 = 410,
F32MAT4X3 = 411,
F32MAT4X4 = 412,
F64MAT2X2 = 413,
F64MAT2X3 = 414,
F64MAT2X4 = 415,
F64MAT3X2 = 416,
F64MAT3X3 = 417,
F64MAT3X4 = 418,
F64MAT4X2 = 419,
F64MAT4X3 = 420,
F64MAT4X4 = 421,
ATOMIC_UINT = 422,
SAMPLER1D = 423,
SAMPLER2D = 424,
SAMPLER3D = 425,
SAMPLERCUBE = 426,
SAMPLER1DSHADOW = 427,
SAMPLER2DSHADOW = 428,
SAMPLERCUBESHADOW = 429,
SAMPLER1DARRAY = 430,
SAMPLER2DARRAY = 431,
SAMPLER1DARRAYSHADOW = 432,
SAMPLER2DARRAYSHADOW = 433,
ISAMPLER1D = 434,
ISAMPLER2D = 435,
ISAMPLER3D = 436,
ISAMPLERCUBE = 437,
ISAMPLER1DARRAY = 438,
ISAMPLER2DARRAY = 439,
USAMPLER1D = 440,
USAMPLER2D = 441,
USAMPLER3D = 442,
USAMPLERCUBE = 443,
USAMPLER1DARRAY = 444,
USAMPLER2DARRAY = 445,
SAMPLER2DRECT = 446,
SAMPLER2DRECTSHADOW = 447,
ISAMPLER2DRECT = 448,
USAMPLER2DRECT = 449,
SAMPLERBUFFER = 450,
ISAMPLERBUFFER = 451,
USAMPLERBUFFER = 452,
SAMPLERCUBEARRAY = 453,
SAMPLERCUBEARRAYSHADOW = 454,
ISAMPLERCUBEARRAY = 455,
USAMPLERCUBEARRAY = 456,
SAMPLER2DMS = 457,
ISAMPLER2DMS = 458,
USAMPLER2DMS = 459,
SAMPLER2DMSARRAY = 460,
ISAMPLER2DMSARRAY = 461,
USAMPLER2DMSARRAY = 462,
SAMPLEREXTERNALOES = 463,
F16SAMPLER1D = 464,
F16SAMPLER2D = 465,
F16SAMPLER3D = 466,
F16SAMPLER2DRECT = 467,
F16SAMPLERCUBE = 468,
F16SAMPLER1DARRAY = 469,
F16SAMPLER2DARRAY = 470,
F16SAMPLERCUBEARRAY = 471,
F16SAMPLERBUFFER = 472,
F16SAMPLER2DMS = 473,
F16SAMPLER2DMSARRAY = 474,
F16SAMPLER1DSHADOW = 475,
F16SAMPLER2DSHADOW = 476,
F16SAMPLER1DARRAYSHADOW = 477,
F16SAMPLER2DARRAYSHADOW = 478,
F16SAMPLER2DRECTSHADOW = 479,
F16SAMPLERCUBESHADOW = 480,
F16SAMPLERCUBEARRAYSHADOW = 481,
SAMPLER = 482,
SAMPLERSHADOW = 483,
TEXTURE1D = 484,
TEXTURE2D = 485,
TEXTURE3D = 486,
TEXTURECUBE = 487,
TEXTURE1DARRAY = 488,
TEXTURE2DARRAY = 489,
ITEXTURE1D = 490,
ITEXTURE2D = 491,
ITEXTURE3D = 492,
ITEXTURECUBE = 493,
ITEXTURE1DARRAY = 494,
ITEXTURE2DARRAY = 495,
UTEXTURE1D = 496,
UTEXTURE2D = 497,
UTEXTURE3D = 498,
UTEXTURECUBE = 499,
UTEXTURE1DARRAY = 500,
UTEXTURE2DARRAY = 501,
TEXTURE2DRECT = 502,
ITEXTURE2DRECT = 503,
UTEXTURE2DRECT = 504,
TEXTUREBUFFER = 505,
ITEXTUREBUFFER = 506,
UTEXTUREBUFFER = 507,
TEXTURECUBEARRAY = 508,
ITEXTURECUBEARRAY = 509,
UTEXTURECUBEARRAY = 510,
TEXTURE2DMS = 511,
ITEXTURE2DMS = 512,
UTEXTURE2DMS = 513,
TEXTURE2DMSARRAY = 514,
ITEXTURE2DMSARRAY = 515,
UTEXTURE2DMSARRAY = 516,
F16TEXTURE1D = 517,
F16TEXTURE2D = 518,
F16TEXTURE3D = 519,
F16TEXTURE2DRECT = 520,
F16TEXTURECUBE = 521,
F16TEXTURE1DARRAY = 522,
F16TEXTURE2DARRAY = 523,
F16TEXTURECUBEARRAY = 524,
F16TEXTUREBUFFER = 525,
F16TEXTURE2DMS = 526,
F16TEXTURE2DMSARRAY = 527,
SUBPASSINPUT = 528,
SUBPASSINPUTMS = 529,
ISUBPASSINPUT = 530,
ISUBPASSINPUTMS = 531,
USUBPASSINPUT = 532,
USUBPASSINPUTMS = 533,
F16SUBPASSINPUT = 534,
F16SUBPASSINPUTMS = 535,
IMAGE1D = 536,
IIMAGE1D = 537,
UIMAGE1D = 538,
IMAGE2D = 539,
IIMAGE2D = 540,
UIMAGE2D = 541,
IMAGE3D = 542,
IIMAGE3D = 543,
UIMAGE3D = 544,
IMAGE2DRECT = 545,
IIMAGE2DRECT = 546,
UIMAGE2DRECT = 547,
IMAGECUBE = 548,
IIMAGECUBE = 549,
UIMAGECUBE = 550,
IMAGEBUFFER = 551,
IIMAGEBUFFER = 552,
UIMAGEBUFFER = 553,
IMAGE1DARRAY = 554,
IIMAGE1DARRAY = 555,
UIMAGE1DARRAY = 556,
IMAGE2DARRAY = 557,
IIMAGE2DARRAY = 558,
UIMAGE2DARRAY = 559,
IMAGECUBEARRAY = 560,
IIMAGECUBEARRAY = 561,
UIMAGECUBEARRAY = 562,
IMAGE2DMS = 563,
IIMAGE2DMS = 564,
UIMAGE2DMS = 565,
IMAGE2DMSARRAY = 566,
IIMAGE2DMSARRAY = 567,
UIMAGE2DMSARRAY = 568,
F16IMAGE1D = 569,
F16IMAGE2D = 570,
F16IMAGE3D = 571,
F16IMAGE2DRECT = 572,
F16IMAGECUBE = 573,
F16IMAGE1DARRAY = 574,
F16IMAGE2DARRAY = 575,
F16IMAGECUBEARRAY = 576,
F16IMAGEBUFFER = 577,
F16IMAGE2DMS = 578,
F16IMAGE2DMSARRAY = 579,
STRUCT = 580,
VOID = 581,
WHILE = 582,
IDENTIFIER = 583,
TYPE_NAME = 584,
FLOATCONSTANT = 585,
DOUBLECONSTANT = 586,
INT16CONSTANT = 587,
UINT16CONSTANT = 588,
INT32CONSTANT = 589,
UINT32CONSTANT = 590,
INTCONSTANT = 591,
UINTCONSTANT = 592,
INT64CONSTANT = 593,
UINT64CONSTANT = 594,
BOOLCONSTANT = 595,
FLOAT16CONSTANT = 596,
LEFT_OP = 597,
RIGHT_OP = 598,
INC_OP = 599,
DEC_OP = 600,
LE_OP = 601,
GE_OP = 602,
EQ_OP = 603,
NE_OP = 604,
AND_OP = 605,
OR_OP = 606,
XOR_OP = 607,
MUL_ASSIGN = 608,
DIV_ASSIGN = 609,
ADD_ASSIGN = 610,
MOD_ASSIGN = 611,
LEFT_ASSIGN = 612,
RIGHT_ASSIGN = 613,
AND_ASSIGN = 614,
XOR_ASSIGN = 615,
OR_ASSIGN = 616,
SUB_ASSIGN = 617,
LEFT_PAREN = 618,
RIGHT_PAREN = 619,
LEFT_BRACKET = 620,
RIGHT_BRACKET = 621,
LEFT_BRACE = 622,
RIGHT_BRACE = 623,
DOT = 624,
COMMA = 625,
COLON = 626,
EQUAL = 627,
SEMICOLON = 628,
BANG = 629,
DASH = 630,
TILDE = 631,
PLUS = 632,
STAR = 633,
SLASH = 634,
PERCENT = 635,
LEFT_ANGLE = 636,
RIGHT_ANGLE = 637,
VERTICAL_BAR = 638,
CARET = 639,
AMPERSAND = 640,
QUESTION = 641,
INVARIANT = 642,
PRECISE = 643,
HIGH_PRECISION = 644,
MEDIUM_PRECISION = 645,
LOW_PRECISION = 646,
PRECISION = 647,
PACKED = 648,
RESOURCE = 649,
SUPERP = 650
};
#endif
/* Value type. */
#if ! defined YYSTYPE && ! defined YYSTYPE_IS_DECLARED
typedef union YYSTYPE YYSTYPE;
union YYSTYPE
{
#line 70 "MachineIndependent/glslang.y" /* yacc.c:1909 */
@ -476,8 +481,10 @@ union YYSTYPE
};
} interm;
#line 480 "MachineIndependent/glslang_tab.cpp.h" /* yacc.c:1909 */
#line 485 "MachineIndependent/glslang_tab.cpp.h" /* yacc.c:1909 */
};
typedef union YYSTYPE YYSTYPE;
# define YYSTYPE_IS_TRIVIAL 1
# define YYSTYPE_IS_DECLARED 1
#endif

Просмотреть файл

@ -871,6 +871,8 @@ bool TOutputTraverser::visitAggregate(TVisit /* visit */, TIntermAggregate* node
case EOpAtomicXor: out.debug << "AtomicXor"; break;
case EOpAtomicExchange: out.debug << "AtomicExchange"; break;
case EOpAtomicCompSwap: out.debug << "AtomicCompSwap"; break;
case EOpAtomicLoad: out.debug << "AtomicLoad"; break;
case EOpAtomicStore: out.debug << "AtomicStore"; break;
case EOpAtomicCounterAdd: out.debug << "AtomicCounterAdd"; break;
case EOpAtomicCounterSubtract: out.debug << "AtomicCounterSubtract"; break;
@ -894,6 +896,8 @@ bool TOutputTraverser::visitAggregate(TVisit /* visit */, TIntermAggregate* node
case EOpImageAtomicXor: out.debug << "imageAtomicXor"; break;
case EOpImageAtomicExchange: out.debug << "imageAtomicExchange"; break;
case EOpImageAtomicCompSwap: out.debug << "imageAtomicCompSwap"; break;
case EOpImageAtomicLoad: out.debug << "imageAtomicLoad"; break;
case EOpImageAtomicStore: out.debug << "imageAtomicStore"; break;
#ifdef AMD_EXTENSIONS
case EOpImageLoadLod: out.debug << "imageLoadLod"; break;
case EOpImageStoreLod: out.debug << "imageStoreLod"; break;

Просмотреть файл

@ -524,11 +524,16 @@ void TIntermediate::mergeErrorCheck(TInfoSink& infoSink, const TIntermSymbol& sy
}
// Memory...
if (symbol.getQualifier().coherent != unitSymbol.getQualifier().coherent ||
symbol.getQualifier().volatil != unitSymbol.getQualifier().volatil ||
symbol.getQualifier().restrict != unitSymbol.getQualifier().restrict ||
symbol.getQualifier().readonly != unitSymbol.getQualifier().readonly ||
symbol.getQualifier().writeonly != unitSymbol.getQualifier().writeonly) {
if (symbol.getQualifier().coherent != unitSymbol.getQualifier().coherent ||
symbol.getQualifier().devicecoherent != unitSymbol.getQualifier().devicecoherent ||
symbol.getQualifier().queuefamilycoherent != unitSymbol.getQualifier().queuefamilycoherent ||
symbol.getQualifier().workgroupcoherent != unitSymbol.getQualifier().workgroupcoherent ||
symbol.getQualifier().subgroupcoherent != unitSymbol.getQualifier().subgroupcoherent ||
symbol.getQualifier().nonprivate != unitSymbol.getQualifier().nonprivate ||
symbol.getQualifier().volatil != unitSymbol.getQualifier().volatil ||
symbol.getQualifier().restrict != unitSymbol.getQualifier().restrict ||
symbol.getQualifier().readonly != unitSymbol.getQualifier().readonly ||
symbol.getQualifier().writeonly != unitSymbol.getQualifier().writeonly) {
error(infoSink, "Memory qualifiers must match:");
writeTypeComparison = true;
}

Просмотреть файл

@ -233,6 +233,7 @@ public:
useUnknownFormat(false),
hlslOffsets(false),
useStorageBuffer(false),
useVulkanMemoryModel(false),
hlslIoMapping(false),
textureSamplerTransformMode(EShTexSampTransKeep),
needToLegalize(false),
@ -365,6 +366,12 @@ public:
processes.addProcess("hlsl-iomap");
}
bool usingHlslIoMapping() { return hlslIoMapping; }
void setUseVulkanMemoryModel()
{
useVulkanMemoryModel = true;
processes.addProcess("use-vulkan-memory-model");
}
bool usingVulkanMemoryModel() const { return useVulkanMemoryModel; }
template<class T> T addCounterBufferName(const T& name) const { return name + implicitCounterName; }
bool hasCounterBufferName(const TString& name) const {
@ -734,6 +741,7 @@ protected:
bool useUnknownFormat;
bool hlslOffsets;
bool useStorageBuffer;
bool useVulkanMemoryModel;
bool hlslIoMapping;
std::set<TString> ioAccessed; // set of names of statically read/written I/O that might need extra checking

Просмотреть файл

@ -286,6 +286,8 @@ INSTANTIATE_TEST_CASE_P(
"spv.matrix.frag",
"spv.matrix2.frag",
"spv.memoryQualifier.frag",
"spv.memoryScopeSemantics.comp",
"spv.memoryScopeSemantics_Error.comp",
"spv.merge-unreachable.frag",
"spv.multiStruct.comp",
"spv.multiStructFuncall.frag",

Просмотреть файл

@ -5,14 +5,14 @@
"site" : "github",
"subrepo" : "KhronosGroup/SPIRV-Tools",
"subdir" : "External/spirv-tools",
"commit" : "6d27a8350fbc339909834a6ef339c805cb1ab69b"
"commit" : "7600fc0e19c3a99bd3ef2c24515cc508ca1d3cfb"
},
{
"name" : "spirv-tools/external/spirv-headers",
"site" : "github",
"subrepo" : "KhronosGroup/SPIRV-Headers",
"subdir" : "External/spirv-tools/external/spirv-headers",
"commit" : "ff684ffc6a35d2a58f0f63108877d0064ea33feb"
"commit" : "dcf23bdabacc3c54b83b1f9367e7a8adb27f8d87"
}
]
}