glslang: Sync with upstream 4fc7a33 for Vulkan SDK 1.2.131

Fixes #36888.
This commit is contained in:
Rémi Verschelde
2020-03-09 09:35:55 +01:00
parent 214bc9e5a1
commit 1231c2ecfc
64 changed files with 13442 additions and 11788 deletions

View File

@ -34,5 +34,6 @@ static const char* const E_SPV_EXT_shader_stencil_export = "SPV_EXT_shade
static const char* const E_SPV_EXT_shader_viewport_index_layer = "SPV_EXT_shader_viewport_index_layer";
static const char* const E_SPV_EXT_fragment_fully_covered = "SPV_EXT_fragment_fully_covered";
static const char* const E_SPV_EXT_fragment_invocation_density = "SPV_EXT_fragment_invocation_density";
static const char* const E_SPV_EXT_demote_to_helper_invocation = "SPV_EXT_demote_to_helper_invocation";
#endif // #ifndef GLSLextEXT_H

View File

@ -41,5 +41,8 @@ static const char* const E_SPV_KHR_storage_buffer_storage_class = "SPV_KHR_stora
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";
static const char* const E_SPV_EXT_physical_storage_buffer = "SPV_EXT_physical_storage_buffer";
static const char* const E_SPV_KHR_physical_storage_buffer = "SPV_KHR_physical_storage_buffer";
static const char* const E_SPV_EXT_fragment_shader_interlock = "SPV_EXT_fragment_shader_interlock";
static const char* const E_SPV_KHR_shader_clock = "SPV_KHR_shader_clock";
#endif // #ifndef GLSLextKHR_H

View File

@ -75,4 +75,7 @@ const char* const E_SPV_NV_shading_rate = "SPV_NV_shading_rate";
//SPV_NV_cooperative_matrix
const char* const E_SPV_NV_cooperative_matrix = "SPV_NV_cooperative_matrix";
//SPV_NV_shader_sm_builtins
const char* const E_SPV_NV_shader_sm_builtins = "SPV_NV_shader_sm_builtins";
#endif // #ifndef GLSLextNV_H

1689
thirdparty/glslang/SPIRV/GlslangToSpv.cpp vendored Normal file → Executable file

File diff suppressed because it is too large Load Diff

2
thirdparty/glslang/SPIRV/GlslangToSpv.h vendored Normal file → Executable file
View File

@ -40,7 +40,7 @@
#endif
#include "SpvTools.h"
#include "../glslang/Include/intermediate.h"
#include "glslang/Include/intermediate.h"
#include <string>
#include <vector>

View File

@ -61,17 +61,22 @@ namespace {
// Use by calling visit() on the root block.
class ReadableOrderTraverser {
public:
explicit ReadableOrderTraverser(std::function<void(Block*)> callback) : callback_(callback) {}
ReadableOrderTraverser(std::function<void(Block*, spv::ReachReason, Block*)> callback)
: callback_(callback) {}
// Visits the block if it hasn't been visited already and isn't currently
// being delayed. Invokes callback(block), then descends into its
// being delayed. Invokes callback(block, why, header), then descends into its
// successors. Delays merge-block and continue-block processing until all
// the branches have been completed.
void visit(Block* block)
// the branches have been completed. If |block| is an unreachable merge block or
// an unreachable continue target, then |header| is the corresponding header block.
void visit(Block* block, spv::ReachReason why, Block* header)
{
assert(block);
if (why == spv::ReachViaControlFlow) {
reachableViaControlFlow_.insert(block);
}
if (visited_.count(block) || delayed_.count(block))
return;
callback_(block);
callback_(block, why, header);
visited_.insert(block);
Block* mergeBlock = nullptr;
Block* continueBlock = nullptr;
@ -87,27 +92,40 @@ public:
delayed_.insert(continueBlock);
}
}
const auto successors = block->getSuccessors();
for (auto it = successors.cbegin(); it != successors.cend(); ++it)
visit(*it);
if (why == spv::ReachViaControlFlow) {
const auto& successors = block->getSuccessors();
for (auto it = successors.cbegin(); it != successors.cend(); ++it)
visit(*it, why, nullptr);
}
if (continueBlock) {
const spv::ReachReason continueWhy =
(reachableViaControlFlow_.count(continueBlock) > 0)
? spv::ReachViaControlFlow
: spv::ReachDeadContinue;
delayed_.erase(continueBlock);
visit(continueBlock);
visit(continueBlock, continueWhy, block);
}
if (mergeBlock) {
const spv::ReachReason mergeWhy =
(reachableViaControlFlow_.count(mergeBlock) > 0)
? spv::ReachViaControlFlow
: spv::ReachDeadMerge;
delayed_.erase(mergeBlock);
visit(mergeBlock);
visit(mergeBlock, mergeWhy, block);
}
}
private:
std::function<void(Block*)> callback_;
std::function<void(Block*, spv::ReachReason, Block*)> callback_;
// Whether a block has already been visited or is being delayed.
std::unordered_set<Block *> visited_, delayed_;
// The set of blocks that actually are reached via control flow.
std::unordered_set<Block *> reachableViaControlFlow_;
};
}
void spv::inReadableOrder(Block* root, std::function<void(Block*)> callback)
void spv::inReadableOrder(Block* root, std::function<void(Block*, spv::ReachReason, Block*)> callback)
{
ReadableOrderTraverser(callback).visit(root);
ReadableOrderTraverser(callback).visit(root, spv::ReachViaControlFlow, nullptr);
}

View File

@ -32,6 +32,8 @@
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
#ifndef GLSLANG_WEB
#include "Logger.h"
#include <algorithm>
@ -66,3 +68,5 @@ std::string SpvBuildLogger::getAllMessages() const {
}
} // end spv namespace
#endif

View File

@ -46,6 +46,14 @@ class SpvBuildLogger {
public:
SpvBuildLogger() {}
#ifdef GLSLANG_WEB
void tbdFunctionality(const std::string& f) { }
void missingFunctionality(const std::string& f) { }
void warning(const std::string& w) { }
void error(const std::string& e) { errors.push_back(e); }
std::string getAllMessages() { return ""; }
#else
// Registers a TBD functionality.
void tbdFunctionality(const std::string& f);
// Registers a missing functionality.
@ -59,6 +67,7 @@ public:
// Returns all messages accumulated in the order of:
// TBD functionalities, missing functionalities, warnings, errors.
std::string getAllMessages() const;
#endif
private:
SpvBuildLogger(const SpvBuildLogger&);

View File

@ -195,7 +195,7 @@ private:
// Header access & set methods
spirword_t magic() const { return spv[0]; } // return magic number
spirword_t bound() const { return spv[3]; } // return Id bound from header
spirword_t bound(spirword_t b) { return spv[3] = b; };
spirword_t bound(spirword_t b) { return spv[3] = b; }
spirword_t genmagic() const { return spv[2]; } // generator magic
spirword_t genmagic(spirword_t m) { return spv[2] = m; }
spirword_t schemaNum() const { return spv[4]; } // schema number from header

View File

@ -46,7 +46,9 @@
#include "SpvBuilder.h"
#ifndef GLSLANG_WEB
#include "hex_float.h"
#endif
#ifndef _WIN32
#include <cstdio>
@ -230,6 +232,11 @@ Id Builder::makePointerFromForwardPointer(StorageClass storageClass, Id forwardP
Id Builder::makeIntegerType(int width, bool hasSign)
{
#ifdef GLSLANG_WEB
assert(width == 32);
width = 32;
#endif
// try to find it
Instruction* type;
for (int t = 0; t < (int)groupedTypes[OpTypeInt].size(); ++t) {
@ -265,6 +272,11 @@ Id Builder::makeIntegerType(int width, bool hasSign)
Id Builder::makeFloatType(int width)
{
#ifdef GLSLANG_WEB
assert(width == 32);
width = 32;
#endif
// try to find it
Instruction* type;
for (int t = 0; t < (int)groupedTypes[OpTypeFloat].size(); ++t) {
@ -516,6 +528,7 @@ Id Builder::makeImageType(Id sampledType, Dim dim, bool depth, bool arrayed, boo
constantsTypesGlobals.push_back(std::unique_ptr<Instruction>(type));
module.mapInstruction(type);
#ifndef GLSLANG_WEB
// deal with capabilities
switch (dim) {
case DimBuffer:
@ -561,6 +574,7 @@ Id Builder::makeImageType(Id sampledType, Dim dim, bool depth, bool arrayed, boo
addCapability(CapabilityImageMSArray);
}
}
#endif
return type->getResultId();
}
@ -586,7 +600,7 @@ Id Builder::makeSampledImageType(Id imageType)
return type->getResultId();
}
#ifdef NV_EXTENSIONS
#ifndef GLSLANG_WEB
Id Builder::makeAccelerationStructureNVType()
{
Instruction *type;
@ -602,6 +616,7 @@ Id Builder::makeAccelerationStructureNVType()
return type->getResultId();
}
#endif
Id Builder::getDerefTypeId(Id resultId) const
{
Id typeId = getTypeId(resultId);
@ -939,6 +954,10 @@ Id Builder::makeFloatConstant(float f, bool specConstant)
Id Builder::makeDoubleConstant(double d, bool specConstant)
{
#ifdef GLSLANG_WEB
assert(0);
return NoResult;
#else
Op opcode = specConstant ? OpSpecConstant : OpConstant;
Id typeId = makeFloatType(64);
union { double db; unsigned long long ull; } u;
@ -963,10 +982,15 @@ Id Builder::makeDoubleConstant(double d, bool specConstant)
module.mapInstruction(c);
return c->getResultId();
#endif
}
Id Builder::makeFloat16Constant(float f16, bool specConstant)
{
#ifdef GLSLANG_WEB
assert(0);
return NoResult;
#else
Op opcode = specConstant ? OpSpecConstant : OpConstant;
Id typeId = makeFloatType(16);
@ -991,25 +1015,33 @@ Id Builder::makeFloat16Constant(float f16, bool specConstant)
module.mapInstruction(c);
return c->getResultId();
#endif
}
Id Builder::makeFpConstant(Id type, double d, bool specConstant)
{
assert(isFloatType(type));
#ifdef GLSLANG_WEB
const int width = 32;
assert(width == getScalarTypeWidth(type));
#else
const int width = getScalarTypeWidth(type);
#endif
switch (getScalarTypeWidth(type)) {
case 16:
return makeFloat16Constant((float)d, specConstant);
case 32:
return makeFloatConstant((float)d, specConstant);
case 64:
return makeDoubleConstant(d, specConstant);
default:
break;
}
assert(isFloatType(type));
assert(false);
return NoResult;
switch (width) {
case 16:
return makeFloat16Constant((float)d, specConstant);
case 32:
return makeFloatConstant((float)d, specConstant);
case 64:
return makeDoubleConstant(d, specConstant);
default:
break;
}
assert(false);
return NoResult;
}
Id Builder::findCompositeConstant(Op typeClass, Id typeId, const std::vector<Id>& comps)
@ -1825,7 +1857,7 @@ Id Builder::createTextureCall(Decoration precision, Id resultType, bool sparse,
if (parameters.component != NoResult)
texArgs[numArgs++] = parameters.component;
#ifdef NV_EXTENSIONS
#ifndef GLSLANG_WEB
if (parameters.granularity != NoResult)
texArgs[numArgs++] = parameters.granularity;
if (parameters.coarse != NoResult)
@ -1872,6 +1904,7 @@ Id Builder::createTextureCall(Decoration precision, Id resultType, bool sparse,
mask = (ImageOperandsMask)(mask | ImageOperandsConstOffsetsMask);
texArgs[numArgs++] = parameters.offsets;
}
#ifndef GLSLANG_WEB
if (parameters.sample) {
mask = (ImageOperandsMask)(mask | ImageOperandsSampleMask);
texArgs[numArgs++] = parameters.sample;
@ -1889,6 +1922,7 @@ Id Builder::createTextureCall(Decoration precision, Id resultType, bool sparse,
if (parameters.volatil) {
mask = mask | ImageOperandsVolatileTexelKHRMask;
}
#endif
mask = mask | signExtensionMask;
if (mask == ImageOperandsMaskNone)
--numArgs; // undo speculative reservation for the mask argument
@ -1904,10 +1938,9 @@ Id Builder::createTextureCall(Decoration precision, Id resultType, bool sparse,
opCode = OpImageSparseFetch;
else
opCode = OpImageFetch;
#ifdef NV_EXTENSIONS
#ifndef GLSLANG_WEB
} else if (parameters.granularity && parameters.coarse) {
opCode = OpImageSampleFootprintNV;
#endif
} else if (gather) {
if (parameters.Dref)
if (sparse)
@ -1919,6 +1952,7 @@ Id Builder::createTextureCall(Decoration precision, Id resultType, bool sparse,
opCode = OpImageSparseGather;
else
opCode = OpImageGather;
#endif
} else if (explicitLod) {
if (parameters.Dref) {
if (proj)
@ -2067,11 +2101,7 @@ Id Builder::createTextureQueryCall(Op opCode, const TextureParameters& parameter
break;
}
case OpImageQueryLod:
#ifdef AMD_EXTENSIONS
resultType = makeVectorType(getScalarTypeId(getTypeId(parameters.coords)), 2);
#else
resultType = makeVectorType(makeFloatType(32), 2);
#endif
break;
case OpImageQueryLevels:
case OpImageQuerySamples:
@ -2089,6 +2119,7 @@ Id Builder::createTextureQueryCall(Op opCode, const TextureParameters& parameter
if (parameters.lod)
query->addIdOperand(parameters.lod);
buildPoint->addInstruction(std::unique_ptr<Instruction>(query));
addCapability(CapabilityImageQuery);
return query->getResultId();
}
@ -2282,7 +2313,12 @@ Id Builder::createMatrixConstructor(Decoration precision, const std::vector<Id>&
int numRows = getTypeNumRows(resultTypeId);
Instruction* instr = module.getInstruction(componentTypeId);
unsigned bitCount = instr->getImmediateOperand(0);
#ifdef GLSLANG_WEB
const unsigned bitCount = 32;
assert(bitCount == instr->getImmediateOperand(0));
#else
const unsigned bitCount = instr->getImmediateOperand(0);
#endif
// Optimize matrix constructed from a bigger matrix
if (isMatrix(sources[0]) && getNumColumns(sources[0]) >= numCols && getNumRows(sources[0]) >= numRows) {

View File

@ -67,6 +67,7 @@ typedef enum {
Spv_1_2 = (1 << 16) | (2 << 8),
Spv_1_3 = (1 << 16) | (3 << 8),
Spv_1_4 = (1 << 16) | (4 << 8),
Spv_1_5 = (1 << 16) | (5 << 8),
} SpvVersion;
class Builder {
@ -105,6 +106,20 @@ public:
void addModuleProcessed(const std::string& p) { moduleProcesses.push_back(p.c_str()); }
void setEmitOpLines() { emitOpLines = true; }
void addExtension(const char* ext) { extensions.insert(ext); }
void removeExtension(const char* ext)
{
extensions.erase(ext);
}
void addIncorporatedExtension(const char* ext, SpvVersion incorporatedVersion)
{
if (getSpvVersion() < static_cast<unsigned>(incorporatedVersion))
addExtension(ext);
}
void promoteIncorporatedExtension(const char* baseExt, const char* promoExt, SpvVersion incorporatedVersion)
{
removeExtension(baseExt);
addIncorporatedExtension(promoExt, incorporatedVersion);
}
void addInclude(const std::string& name, const std::string& text)
{
spv::Id incId = getStringId(name);
@ -201,7 +216,11 @@ public:
bool isMatrixType(Id typeId) const { return getTypeClass(typeId) == OpTypeMatrix; }
bool isStructType(Id typeId) const { return getTypeClass(typeId) == OpTypeStruct; }
bool isArrayType(Id typeId) const { return getTypeClass(typeId) == OpTypeArray; }
#ifdef GLSLANG_WEB
bool isCooperativeMatrixType(Id typeId)const { return false; }
#else
bool isCooperativeMatrixType(Id typeId)const { return getTypeClass(typeId) == OpTypeCooperativeMatrixNV; }
#endif
bool isAggregateType(Id typeId) const { return isArrayType(typeId) || isStructType(typeId) || isCooperativeMatrixType(typeId); }
bool isImageType(Id typeId) const { return getTypeClass(typeId) == OpTypeImage; }
bool isSamplerType(Id typeId) const { return getTypeClass(typeId) == OpTypeSampler; }
@ -557,6 +576,14 @@ public:
// Accumulate whether anything in the chain of structures has coherent decorations.
struct CoherentFlags {
CoherentFlags() { clear(); }
#ifdef GLSLANG_WEB
void clear() { }
bool isVolatile() const { return false; }
CoherentFlags operator |=(const CoherentFlags &other) { return *this; }
#else
bool isVolatile() const { return volatil; }
unsigned coherent : 1;
unsigned devicecoherent : 1;
unsigned queuefamilycoherent : 1;
@ -577,7 +604,6 @@ public:
isImage = 0;
}
CoherentFlags() { clear(); }
CoherentFlags operator |=(const CoherentFlags &other) {
coherent |= other.coherent;
devicecoherent |= other.devicecoherent;
@ -589,6 +615,7 @@ public:
isImage |= other.isImage;
return *this;
}
#endif
};
CoherentFlags coherentFlags;
};
@ -656,16 +683,21 @@ public:
// based on the type of the base and the chain of dereferences.
Id accessChainGetInferredType();
// Add capabilities, extensions, remove unneeded decorations, etc.,
// Add capabilities, extensions, remove unneeded decorations, etc.,
// based on the resulting SPIR-V.
void postProcess();
// Prune unreachable blocks in the CFG and remove unneeded decorations.
void postProcessCFG();
#ifndef GLSLANG_WEB
// Add capabilities, extensions based on instructions in the module.
void postProcessFeatures();
// Hook to visit each instruction in a block in a function
void postProcess(Instruction&);
// Hook to visit each instruction in a reachable block in a function.
void postProcessReachable(const Instruction&);
// Hook to visit each non-32-bit sized float/int operation in a block.
void postProcessType(const Instruction&, spv::Id typeId);
#endif
void dump(std::vector<unsigned int>&) const;

View File

@ -39,6 +39,7 @@
#include <cassert>
#include <cstdlib>
#include <unordered_map>
#include <unordered_set>
#include <algorithm>
@ -51,16 +52,13 @@ namespace spv {
#include "GLSL.std.450.h"
#include "GLSL.ext.KHR.h"
#include "GLSL.ext.EXT.h"
#ifdef AMD_EXTENSIONS
#include "GLSL.ext.AMD.h"
#endif
#ifdef NV_EXTENSIONS
#include "GLSL.ext.NV.h"
#endif
}
namespace spv {
#ifndef GLSLANG_WEB
// Hook to visit each operand type and result type of an instruction.
// Will be called multiple times for one instruction, once for each typed
// operand and the result.
@ -160,7 +158,6 @@ void Builder::postProcessType(const Instruction& inst, Id typeId)
}
break;
case OpExtInst:
#if AMD_EXTENSIONS
switch (inst.getImmediateOperand(1)) {
case GLSLstd450Frexp:
case GLSLstd450FrexpStruct:
@ -176,7 +173,6 @@ void Builder::postProcessType(const Instruction& inst, Id typeId)
default:
break;
}
#endif
break;
default:
if (basicTypeOp == OpTypeFloat && width == 16)
@ -222,12 +218,10 @@ void Builder::postProcess(Instruction& inst)
addCapability(CapabilityImageQuery);
break;
#ifdef NV_EXTENSIONS
case OpGroupNonUniformPartitionNV:
addExtension(E_SPV_NV_shader_subgroup_partitioned);
addCapability(CapabilityGroupNonUniformPartitionedNV);
break;
#endif
case OpLoad:
case OpStore:
@ -326,17 +320,16 @@ void Builder::postProcess(Instruction& inst)
}
}
}
// Called for each instruction in a reachable block.
void Builder::postProcessReachable(const Instruction&)
{
// did have code here, but questionable to do so without deleting the instructions
}
#endif
// comment in header
void Builder::postProcess()
void Builder::postProcessCFG()
{
// reachableBlocks is the set of blockss reached via control flow, or which are
// unreachable continue targert or unreachable merge.
std::unordered_set<const Block*> reachableBlocks;
std::unordered_map<Block*, Block*> headerForUnreachableContinue;
std::unordered_set<Block*> unreachableMerges;
std::unordered_set<Id> unreachableDefinitions;
// Collect IDs defined in unreachable blocks. For each function, label the
// reachable blocks first. Then for each unreachable block, collect the
@ -344,16 +337,41 @@ void Builder::postProcess()
for (auto fi = module.getFunctions().cbegin(); fi != module.getFunctions().cend(); fi++) {
Function* f = *fi;
Block* entry = f->getEntryBlock();
inReadableOrder(entry, [&reachableBlocks](const Block* b) { reachableBlocks.insert(b); });
inReadableOrder(entry,
[&reachableBlocks, &unreachableMerges, &headerForUnreachableContinue]
(Block* b, ReachReason why, Block* header) {
reachableBlocks.insert(b);
if (why == ReachDeadContinue) headerForUnreachableContinue[b] = header;
if (why == ReachDeadMerge) unreachableMerges.insert(b);
});
for (auto bi = f->getBlocks().cbegin(); bi != f->getBlocks().cend(); bi++) {
Block* b = *bi;
if (reachableBlocks.count(b) == 0) {
for (auto ii = b->getInstructions().cbegin(); ii != b->getInstructions().cend(); ii++)
if (unreachableMerges.count(b) != 0 || headerForUnreachableContinue.count(b) != 0) {
auto ii = b->getInstructions().cbegin();
++ii; // Keep potential decorations on the label.
for (; ii != b->getInstructions().cend(); ++ii)
unreachableDefinitions.insert(ii->get()->getResultId());
} else if (reachableBlocks.count(b) == 0) {
// The normal case for unreachable code. All definitions are considered dead.
for (auto ii = b->getInstructions().cbegin(); ii != b->getInstructions().cend(); ++ii)
unreachableDefinitions.insert(ii->get()->getResultId());
}
}
}
// Modify unreachable merge blocks and unreachable continue targets.
// Delete their contents.
for (auto mergeIter = unreachableMerges.begin(); mergeIter != unreachableMerges.end(); ++mergeIter) {
(*mergeIter)->rewriteAsCanonicalUnreachableMerge();
}
for (auto continueIter = headerForUnreachableContinue.begin();
continueIter != headerForUnreachableContinue.end();
++continueIter) {
Block* continue_target = continueIter->first;
Block* header = continueIter->second;
continue_target->rewriteAsCanonicalUnreachableContinue(header);
}
// Remove unneeded decorations, for unreachable instructions
decorations.erase(std::remove_if(decorations.begin(), decorations.end(),
[&unreachableDefinitions](std::unique_ptr<Instruction>& I) -> bool {
@ -361,7 +379,11 @@ void Builder::postProcess()
return unreachableDefinitions.count(decoration_id) != 0;
}),
decorations.end());
}
#ifndef GLSLANG_WEB
// comment in header
void Builder::postProcessFeatures() {
// Add per-instruction capabilities, extensions, etc.,
// Look for any 8/16 bit type in physical storage buffer class, and set the
@ -371,24 +393,17 @@ void Builder::postProcess()
Instruction* type = groupedTypes[OpTypePointer][t];
if (type->getImmediateOperand(0) == (unsigned)StorageClassPhysicalStorageBufferEXT) {
if (containsType(type->getIdOperand(1), OpTypeInt, 8)) {
addExtension(spv::E_SPV_KHR_8bit_storage);
addIncorporatedExtension(spv::E_SPV_KHR_8bit_storage, spv::Spv_1_5);
addCapability(spv::CapabilityStorageBuffer8BitAccess);
}
if (containsType(type->getIdOperand(1), OpTypeInt, 16) ||
containsType(type->getIdOperand(1), OpTypeFloat, 16)) {
addExtension(spv::E_SPV_KHR_16bit_storage);
addIncorporatedExtension(spv::E_SPV_KHR_16bit_storage, spv::Spv_1_3);
addCapability(spv::CapabilityStorageBuffer16BitAccess);
}
}
}
// process all reachable instructions...
for (auto bi = reachableBlocks.cbegin(); bi != reachableBlocks.cend(); ++bi) {
const Block* block = *bi;
const auto function = [this](const std::unique_ptr<Instruction>& inst) { postProcessReachable(*inst.get()); };
std::for_each(block->getInstructions().begin(), block->getInstructions().end(), function);
}
// process all block-contained instructions
for (auto fi = module.getFunctions().cbegin(); fi != module.getFunctions().cend(); fi++) {
Function* f = *fi;
@ -422,5 +437,14 @@ void Builder::postProcess()
}
}
}
#endif
// comment in header
void Builder::postProcess() {
postProcessCFG();
#ifndef GLSLANG_WEB
postProcessFeatures();
#endif
}
}; // end spv namespace

View File

@ -67,6 +67,8 @@ spv_target_env MapToSpirvToolsEnv(const SpvVersion& spvVersion, spv::SpvBuildLog
logger->missingFunctionality("Target version for SPIRV-Tools validator");
return spv_target_env::SPV_ENV_VULKAN_1_1;
}
case glslang::EShTargetVulkan_1_2:
return spv_target_env::SPV_ENV_VULKAN_1_2;
default:
break;
}
@ -103,7 +105,7 @@ void SpirvToolsDisassemble(std::ostream& out, const std::vector<unsigned int>& s
// Apply the SPIRV-Tools validator to generated SPIR-V.
void SpirvToolsValidate(const glslang::TIntermediate& intermediate, std::vector<unsigned int>& spirv,
spv::SpvBuildLogger* logger)
spv::SpvBuildLogger* logger, bool prelegalization)
{
// validate
spv_context context = spvContextCreate(MapToSpirvToolsEnv(intermediate.getSpv(), logger));
@ -111,6 +113,7 @@ void SpirvToolsValidate(const glslang::TIntermediate& intermediate, std::vector<
spv_diagnostic diagnostic = nullptr;
spv_validator_options options = spvValidatorOptionsCreate();
spvValidatorOptionsSetRelaxBlockLayout(options, intermediate.usingHlslOffsets());
spvValidatorOptionsSetBeforeHlslLegalization(options, prelegalization);
spvValidateWithOptions(context, options, &binary, &diagnostic);
// report
@ -172,6 +175,7 @@ void SpirvToolsLegalize(const glslang::TIntermediate&, std::vector<unsigned int>
if (options->generateDebugInfo) {
optimizer.RegisterPass(spvtools::CreatePropagateLineInfoPass());
}
optimizer.RegisterPass(spvtools::CreateWrapOpKillPass());
optimizer.RegisterPass(spvtools::CreateDeadBranchElimPass());
optimizer.RegisterPass(spvtools::CreateMergeReturnPass());
optimizer.RegisterPass(spvtools::CreateInlineExhaustivePass());
@ -195,8 +199,6 @@ void SpirvToolsLegalize(const glslang::TIntermediate&, std::vector<unsigned int>
optimizer.RegisterPass(spvtools::CreateDeadInsertElimPass());
if (options->optimizeSize) {
optimizer.RegisterPass(spvtools::CreateRedundancyEliminationPass());
// TODO(greg-lunarg): Add this when AMD driver issues are resolved
// optimizer.RegisterPass(CreateCommonUniformElimPass());
}
optimizer.RegisterPass(spvtools::CreateAggressiveDCEPass());
optimizer.RegisterPass(spvtools::CreateCFGCleanupPass());

View File

@ -41,10 +41,12 @@
#ifndef GLSLANG_SPV_TOOLS_H
#define GLSLANG_SPV_TOOLS_H
#ifdef ENABLE_OPT
#include <vector>
#include <ostream>
#endif
#include "../glslang/MachineIndependent/localintermediate.h"
#include "glslang/MachineIndependent/localintermediate.h"
#include "Logger.h"
namespace glslang {
@ -59,14 +61,14 @@ struct SpvOptions {
bool validate;
};
#if ENABLE_OPT
#ifdef ENABLE_OPT
// Use the SPIRV-Tools disassembler to print SPIR-V.
void SpirvToolsDisassemble(std::ostream& out, const std::vector<unsigned int>& spirv);
// Apply the SPIRV-Tools validator to generated SPIR-V.
void SpirvToolsValidate(const glslang::TIntermediate& intermediate, std::vector<unsigned int>& spirv,
spv::SpvBuildLogger*);
spv::SpvBuildLogger*, bool prelegalization);
// Apply the SPIRV-Tools optimizer to generated SPIR-V, for the purpose of
// legalizing HLSL SPIR-V.

View File

@ -52,26 +52,16 @@ namespace spv {
extern "C" {
// Include C-based headers that don't have a namespace
#include "GLSL.std.450.h"
#ifdef AMD_EXTENSIONS
#include "GLSL.ext.AMD.h"
#endif
#ifdef NV_EXTENSIONS
#include "GLSL.ext.NV.h"
#endif
}
}
const char* GlslStd450DebugNames[spv::GLSLstd450Count];
namespace spv {
#ifdef AMD_EXTENSIONS
static const char* GLSLextAMDGetDebugNames(const char*, unsigned);
#endif
#ifdef NV_EXTENSIONS
static const char* GLSLextNVGetDebugNames(const char*, unsigned);
#endif
static void Kill(std::ostream& out, const char* message)
{
@ -82,15 +72,8 @@ static void Kill(std::ostream& out, const char* message)
// used to identify the extended instruction library imported when printing
enum ExtInstSet {
GLSL450Inst,
#ifdef AMD_EXTENSIONS
GLSLextAMDInst,
#endif
#ifdef NV_EXTENSIONS
GLSLextNVInst,
#endif
OpenCLExtInst,
};
@ -499,37 +482,29 @@ void SpirvStream::disassembleInstruction(Id resultId, Id /*typeId*/, Op opCode,
const char* name = idDescriptor[stream[word - 2]].c_str();
if (0 == memcmp("OpenCL", name, 6)) {
extInstSet = OpenCLExtInst;
#ifdef AMD_EXTENSIONS
} else if (strcmp(spv::E_SPV_AMD_shader_ballot, name) == 0 ||
strcmp(spv::E_SPV_AMD_shader_trinary_minmax, name) == 0 ||
strcmp(spv::E_SPV_AMD_shader_explicit_vertex_parameter, name) == 0 ||
strcmp(spv::E_SPV_AMD_gcn_shader, name) == 0) {
extInstSet = GLSLextAMDInst;
#endif
#ifdef NV_EXTENSIONS
}else if (strcmp(spv::E_SPV_NV_sample_mask_override_coverage, name) == 0 ||
} else if (strcmp(spv::E_SPV_NV_sample_mask_override_coverage, name) == 0 ||
strcmp(spv::E_SPV_NV_geometry_shader_passthrough, name) == 0 ||
strcmp(spv::E_SPV_NV_viewport_array2, name) == 0 ||
strcmp(spv::E_SPV_NVX_multiview_per_view_attributes, name) == 0 ||
strcmp(spv::E_SPV_NV_fragment_shader_barycentric, name) == 0 ||
strcmp(spv::E_SPV_NV_mesh_shader, name) == 0) {
extInstSet = GLSLextNVInst;
#endif
}
unsigned entrypoint = stream[word - 1];
if (extInstSet == GLSL450Inst) {
if (entrypoint < GLSLstd450Count) {
out << "(" << GlslStd450DebugNames[entrypoint] << ")";
}
#ifdef AMD_EXTENSIONS
} else if (extInstSet == GLSLextAMDInst) {
out << "(" << GLSLextAMDGetDebugNames(name, entrypoint) << ")";
#endif
#ifdef NV_EXTENSIONS
}
else if (extInstSet == GLSLextNVInst) {
out << "(" << GLSLextNVGetDebugNames(name, entrypoint) << ")";
#endif
}
}
break;
@ -648,9 +623,11 @@ static void GLSLstd450GetDebugNames(const char** names)
names[GLSLstd450InterpolateAtCentroid] = "InterpolateAtCentroid";
names[GLSLstd450InterpolateAtSample] = "InterpolateAtSample";
names[GLSLstd450InterpolateAtOffset] = "InterpolateAtOffset";
names[GLSLstd450NMin] = "NMin";
names[GLSLstd450NMax] = "NMax";
names[GLSLstd450NClamp] = "NClamp";
}
#ifdef AMD_EXTENSIONS
static const char* GLSLextAMDGetDebugNames(const char* name, unsigned entrypoint)
{
if (strcmp(name, spv::E_SPV_AMD_shader_ballot) == 0) {
@ -692,18 +669,17 @@ static const char* GLSLextAMDGetDebugNames(const char* name, unsigned entrypoint
return "Bad";
}
#endif
#ifdef NV_EXTENSIONS
static const char* GLSLextNVGetDebugNames(const char* name, unsigned entrypoint)
{
if (strcmp(name, spv::E_SPV_NV_sample_mask_override_coverage) == 0 ||
strcmp(name, spv::E_SPV_NV_geometry_shader_passthrough) == 0 ||
strcmp(name, spv::E_ARB_shader_viewport_layer_array) == 0 ||
strcmp(name, spv::E_SPV_NV_viewport_array2) == 0 ||
strcmp(spv::E_SPV_NVX_multiview_per_view_attributes, name) == 0 ||
strcmp(spv::E_SPV_NV_fragment_shader_barycentric, name) == 0 ||
strcmp(name, spv::E_SPV_NV_mesh_shader) == 0) {
strcmp(name, spv::E_SPV_NVX_multiview_per_view_attributes) == 0 ||
strcmp(name, spv::E_SPV_NV_fragment_shader_barycentric) == 0 ||
strcmp(name, spv::E_SPV_NV_mesh_shader) == 0 ||
strcmp(name, spv::E_SPV_NV_shader_image_footprint) == 0) {
switch (entrypoint) {
// NV builtins
case BuiltInViewportMaskNV: return "ViewportMaskNV";
@ -729,6 +705,8 @@ static const char* GLSLextNVGetDebugNames(const char* name, unsigned entrypoint)
case CapabilityPerViewAttributesNV: return "PerViewAttributesNV";
case CapabilityFragmentBarycentricNV: return "FragmentBarycentricNV";
case CapabilityMeshShadingNV: return "MeshShadingNV";
case CapabilityImageFootprintNV: return "ImageFootprintNV";
case CapabilitySampleMaskOverrideCoverageNV:return "SampleMaskOverrideCoverageNV";
// NV Decorations
case DecorationOverrideCoverageNV: return "OverrideCoverageNV";
@ -745,7 +723,6 @@ static const char* GLSLextNVGetDebugNames(const char* name, unsigned entrypoint)
}
return "Bad";
}
#endif
void Disassemble(std::ostream& out, const std::vector<unsigned int>& stream)
{

View File

@ -50,12 +50,8 @@ namespace spv {
// Include C-based headers that don't have a namespace
#include "GLSL.ext.KHR.h"
#include "GLSL.ext.EXT.h"
#ifdef AMD_EXTENSIONS
#include "GLSL.ext.AMD.h"
#endif
#ifdef NV_EXTENSIONS
#include "GLSL.ext.NV.h"
#endif
}
}
@ -98,22 +94,17 @@ const char* ExecutionModelString(int model)
case 4: return "Fragment";
case 5: return "GLCompute";
case 6: return "Kernel";
#ifdef NV_EXTENSIONS
case ExecutionModelTaskNV: return "TaskNV";
case ExecutionModelMeshNV: return "MeshNV";
#endif
default: return "Bad";
#ifdef NV_EXTENSIONS
case ExecutionModelRayGenerationNV: return "RayGenerationNV";
case ExecutionModelIntersectionNV: return "IntersectionNV";
case ExecutionModelAnyHitNV: return "AnyHitNV";
case ExecutionModelClosestHitNV: return "ClosestHitNV";
case ExecutionModelMissNV: return "MissNV";
case ExecutionModelCallableNV: return "CallableNV";
#endif
}
}
@ -183,13 +174,18 @@ const char* ExecutionModeString(int mode)
case 4446: return "PostDepthCoverage";
#ifdef NV_EXTENSIONS
case ExecutionModeOutputLinesNV: return "OutputLinesNV";
case ExecutionModeOutputPrimitivesNV: return "OutputPrimitivesNV";
case ExecutionModeOutputTrianglesNV: return "OutputTrianglesNV";
case ExecutionModeDerivativeGroupQuadsNV: return "DerivativeGroupQuadsNV";
case ExecutionModeDerivativeGroupLinearNV: return "DerivativeGroupLinearNV";
#endif
case ExecutionModePixelInterlockOrderedEXT: return "PixelInterlockOrderedEXT";
case ExecutionModePixelInterlockUnorderedEXT: return "PixelInterlockUnorderedEXT";
case ExecutionModeSampleInterlockOrderedEXT: return "SampleInterlockOrderedEXT";
case ExecutionModeSampleInterlockUnorderedEXT: return "SampleInterlockUnorderedEXT";
case ExecutionModeShadingRateInterlockOrderedEXT: return "ShadingRateInterlockOrderedEXT";
case ExecutionModeShadingRateInterlockUnorderedEXT: return "ShadingRateInterlockUnorderedEXT";
case ExecutionModeCeiling:
default: return "Bad";
@ -213,14 +209,12 @@ const char* StorageClassString(int StorageClass)
case 11: return "Image";
case 12: return "StorageBuffer";
#ifdef NV_EXTENSIONS
case StorageClassRayPayloadNV: return "RayPayloadNV";
case StorageClassHitAttributeNV: return "HitAttributeNV";
case StorageClassIncomingRayPayloadNV: return "IncomingRayPayloadNV";
case StorageClassShaderRecordBufferNV: return "ShaderRecordBufferNV";
case StorageClassCallableDataNV: return "CallableDataNV";
case StorageClassIncomingCallableDataNV: return "IncomingCallableDataNV";
#endif
case StorageClassPhysicalStorageBufferEXT: return "PhysicalStorageBufferEXT";
@ -282,10 +276,7 @@ const char* DecorationString(int decoration)
case DecorationCeiling:
default: return "Bad";
#ifdef AMD_EXTENSIONS
case DecorationExplicitInterpAMD: return "ExplicitInterpAMD";
#endif
#ifdef NV_EXTENSIONS
case DecorationOverrideCoverageNV: return "OverrideCoverageNV";
case DecorationPassthroughNV: return "PassthroughNV";
case DecorationViewportRelativeNV: return "ViewportRelativeNV";
@ -294,7 +285,6 @@ const char* DecorationString(int decoration)
case DecorationPerViewNV: return "PerViewNV";
case DecorationPerTaskNV: return "PerTaskNV";
case DecorationPerVertexNV: return "PerVertexNV";
#endif
case DecorationNonUniformEXT: return "DecorationNonUniformEXT";
case DecorationHlslCounterBufferGOOGLE: return "DecorationHlslCounterBufferGOOGLE";
@ -364,7 +354,6 @@ const char* BuiltInString(int builtIn)
case 4426: return "DrawIndex";
case 5014: return "FragStencilRefEXT";
#ifdef AMD_EXTENSIONS
case 4992: return "BaryCoordNoPerspAMD";
case 4993: return "BaryCoordNoPerspCentroidAMD";
case 4994: return "BaryCoordNoPerspSampleAMD";
@ -372,9 +361,6 @@ const char* BuiltInString(int builtIn)
case 4996: return "BaryCoordSmoothCentroidAMD";
case 4997: return "BaryCoordSmoothSampleAMD";
case 4998: return "BaryCoordPullModelAMD";
#endif
#ifdef NV_EXTENSIONS
case BuiltInLaunchIdNV: return "LaunchIdNV";
case BuiltInLaunchSizeNV: return "LaunchSizeNV";
case BuiltInWorldRayOriginNV: return "WorldRayOriginNV";
@ -398,15 +384,12 @@ const char* BuiltInString(int builtIn)
// case BuiltInInvocationsPerPixelNV: return "InvocationsPerPixelNV"; // superseded by BuiltInFragInvocationCountEXT
case BuiltInBaryCoordNV: return "BaryCoordNV";
case BuiltInBaryCoordNoPerspNV: return "BaryCoordNoPerspNV";
#endif
case BuiltInFragSizeEXT: return "FragSizeEXT";
case BuiltInFragInvocationCountEXT: return "FragInvocationCountEXT";
case 5264: return "FullyCoveredEXT";
#ifdef NV_EXTENSIONS
case BuiltInTaskCountNV: return "TaskCountNV";
case BuiltInPrimitiveCountNV: return "PrimitiveCountNV";
case BuiltInPrimitiveIndicesNV: return "PrimitiveIndicesNV";
@ -415,7 +398,10 @@ const char* BuiltInString(int builtIn)
case BuiltInLayerPerViewNV: return "LayerPerViewNV";
case BuiltInMeshViewCountNV: return "MeshViewCountNV";
case BuiltInMeshViewIndicesNV: return "MeshViewIndicesNV";
#endif
case BuiltInWarpsPerSMNV: return "WarpsPerSMNV";
case BuiltInSMCountNV: return "SMCountNV";
case BuiltInWarpIDNV: return "WarpIDNV";
case BuiltInSMIDNV: return "SMIDNV";
default: return "Bad";
}
@ -770,11 +756,9 @@ const char* GroupOperationString(int gop)
case GroupOperationInclusiveScan: return "InclusiveScan";
case GroupOperationExclusiveScan: return "ExclusiveScan";
case GroupOperationClusteredReduce: return "ClusteredReduce";
#ifdef NV_EXTENSIONS
case GroupOperationPartitionedReduceNV: return "PartitionedReduceNV";
case GroupOperationPartitionedInclusiveScanNV: return "PartitionedInclusiveScanNV";
case GroupOperationPartitionedExclusiveScanNV: return "PartitionedExclusiveScanNV";
#endif
default: return "Bad";
}
@ -882,26 +866,23 @@ const char* CapabilityString(int info)
case CapabilityStoragePushConstant16: return "StoragePushConstant16";
case CapabilityStorageInputOutput16: return "StorageInputOutput16";
case CapabilityStorageBuffer8BitAccess: return "CapabilityStorageBuffer8BitAccess";
case CapabilityUniformAndStorageBuffer8BitAccess: return "CapabilityUniformAndStorageBuffer8BitAccess";
case CapabilityStoragePushConstant8: return "CapabilityStoragePushConstant8";
case CapabilityStorageBuffer8BitAccess: return "StorageBuffer8BitAccess";
case CapabilityUniformAndStorageBuffer8BitAccess: return "UniformAndStorageBuffer8BitAccess";
case CapabilityStoragePushConstant8: return "StoragePushConstant8";
case CapabilityDeviceGroup: return "DeviceGroup";
case CapabilityMultiView: return "MultiView";
case CapabilityStencilExportEXT: return "StencilExportEXT";
#ifdef AMD_EXTENSIONS
case CapabilityFloat16ImageAMD: return "Float16ImageAMD";
case CapabilityImageGatherBiasLodAMD: return "ImageGatherBiasLodAMD";
case CapabilityFragmentMaskAMD: return "FragmentMaskAMD";
case CapabilityImageReadWriteLodAMD: return "ImageReadWriteLodAMD";
#endif
case CapabilityAtomicStorageOps: return "AtomicStorageOps";
case CapabilitySampleMaskPostDepthCoverage: return "SampleMaskPostDepthCoverage";
#ifdef NV_EXTENSIONS
case CapabilityGeometryShaderPassthroughNV: return "GeometryShaderPassthroughNV";
case CapabilityShaderViewportIndexLayerNV: return "ShaderViewportIndexLayerNV";
case CapabilityShaderViewportMaskNV: return "ShaderViewportMaskNV";
@ -913,33 +894,44 @@ const char* CapabilityString(int info)
case CapabilityComputeDerivativeGroupLinearNV: return "ComputeDerivativeGroupLinearNV";
case CapabilityFragmentBarycentricNV: return "FragmentBarycentricNV";
case CapabilityMeshShadingNV: return "MeshShadingNV";
// case CapabilityShadingRateNV: return "ShadingRateNV"; // superseded by CapabilityFragmentDensityEXT
#endif
case CapabilityImageFootprintNV: return "ImageFootprintNV";
// case CapabilityShadingRateNV: return "ShadingRateNV"; // superseded by FragmentDensityEXT
case CapabilitySampleMaskOverrideCoverageNV: return "SampleMaskOverrideCoverageNV";
case CapabilityFragmentDensityEXT: return "FragmentDensityEXT";
case CapabilityFragmentFullyCoveredEXT: return "FragmentFullyCoveredEXT";
case CapabilityShaderNonUniformEXT: return "CapabilityShaderNonUniformEXT";
case CapabilityRuntimeDescriptorArrayEXT: return "CapabilityRuntimeDescriptorArrayEXT";
case CapabilityInputAttachmentArrayDynamicIndexingEXT: return "CapabilityInputAttachmentArrayDynamicIndexingEXT";
case CapabilityUniformTexelBufferArrayDynamicIndexingEXT: return "CapabilityUniformTexelBufferArrayDynamicIndexingEXT";
case CapabilityStorageTexelBufferArrayDynamicIndexingEXT: return "CapabilityStorageTexelBufferArrayDynamicIndexingEXT";
case CapabilityUniformBufferArrayNonUniformIndexingEXT: return "CapabilityUniformBufferArrayNonUniformIndexingEXT";
case CapabilitySampledImageArrayNonUniformIndexingEXT: return "CapabilitySampledImageArrayNonUniformIndexingEXT";
case CapabilityStorageBufferArrayNonUniformIndexingEXT: return "CapabilityStorageBufferArrayNonUniformIndexingEXT";
case CapabilityStorageImageArrayNonUniformIndexingEXT: return "CapabilityStorageImageArrayNonUniformIndexingEXT";
case CapabilityInputAttachmentArrayNonUniformIndexingEXT: return "CapabilityInputAttachmentArrayNonUniformIndexingEXT";
case CapabilityUniformTexelBufferArrayNonUniformIndexingEXT: return "CapabilityUniformTexelBufferArrayNonUniformIndexingEXT";
case CapabilityStorageTexelBufferArrayNonUniformIndexingEXT: return "CapabilityStorageTexelBufferArrayNonUniformIndexingEXT";
case CapabilityShaderNonUniformEXT: return "ShaderNonUniformEXT";
case CapabilityRuntimeDescriptorArrayEXT: return "RuntimeDescriptorArrayEXT";
case CapabilityInputAttachmentArrayDynamicIndexingEXT: return "InputAttachmentArrayDynamicIndexingEXT";
case CapabilityUniformTexelBufferArrayDynamicIndexingEXT: return "UniformTexelBufferArrayDynamicIndexingEXT";
case CapabilityStorageTexelBufferArrayDynamicIndexingEXT: return "StorageTexelBufferArrayDynamicIndexingEXT";
case CapabilityUniformBufferArrayNonUniformIndexingEXT: return "UniformBufferArrayNonUniformIndexingEXT";
case CapabilitySampledImageArrayNonUniformIndexingEXT: return "SampledImageArrayNonUniformIndexingEXT";
case CapabilityStorageBufferArrayNonUniformIndexingEXT: return "StorageBufferArrayNonUniformIndexingEXT";
case CapabilityStorageImageArrayNonUniformIndexingEXT: return "StorageImageArrayNonUniformIndexingEXT";
case CapabilityInputAttachmentArrayNonUniformIndexingEXT: return "InputAttachmentArrayNonUniformIndexingEXT";
case CapabilityUniformTexelBufferArrayNonUniformIndexingEXT: return "UniformTexelBufferArrayNonUniformIndexingEXT";
case CapabilityStorageTexelBufferArrayNonUniformIndexingEXT: return "StorageTexelBufferArrayNonUniformIndexingEXT";
case CapabilityVulkanMemoryModelKHR: return "CapabilityVulkanMemoryModelKHR";
case CapabilityVulkanMemoryModelDeviceScopeKHR: return "CapabilityVulkanMemoryModelDeviceScopeKHR";
case CapabilityVulkanMemoryModelKHR: return "VulkanMemoryModelKHR";
case CapabilityVulkanMemoryModelDeviceScopeKHR: return "VulkanMemoryModelDeviceScopeKHR";
case CapabilityPhysicalStorageBufferAddressesEXT: return "CapabilityPhysicalStorageBufferAddressesEXT";
case CapabilityPhysicalStorageBufferAddressesEXT: return "PhysicalStorageBufferAddressesEXT";
case CapabilityVariablePointers: return "CapabilityVariablePointers";
case CapabilityVariablePointers: return "VariablePointers";
case CapabilityCooperativeMatrixNV: return "CapabilityCooperativeMatrixNV";
case CapabilityCooperativeMatrixNV: return "CooperativeMatrixNV";
case CapabilityShaderSMBuiltinsNV: return "ShaderSMBuiltinsNV";
case CapabilityFragmentShaderSampleInterlockEXT: return "CapabilityFragmentShaderSampleInterlockEXT";
case CapabilityFragmentShaderPixelInterlockEXT: return "CapabilityFragmentShaderPixelInterlockEXT";
case CapabilityFragmentShaderShadingRateInterlockEXT: return "CapabilityFragmentShaderShadingRateInterlockEXT";
case CapabilityDemoteToHelperInvocationEXT: return "DemoteToHelperInvocationEXT";
case CapabilityShaderClockKHR: return "ShaderClockKHR";
case CapabilityIntegerFunctions2INTEL: return "CapabilityIntegerFunctions2INTEL";
default: return "Bad";
}
@ -1316,7 +1308,6 @@ const char* OpcodeString(int op)
case 4430: return "OpSubgroupAllEqualKHR";
case 4432: return "OpSubgroupReadInvocationKHR";
#ifdef AMD_EXTENSIONS
case 5000: return "OpGroupIAddNonUniformAMD";
case 5001: return "OpGroupFAddNonUniformAMD";
case 5002: return "OpGroupFMinNonUniformAMD";
@ -1328,12 +1319,12 @@ const char* OpcodeString(int op)
case 5011: return "OpFragmentMaskFetchAMD";
case 5012: return "OpFragmentFetchAMD";
#endif
case OpReadClockKHR: return "OpReadClockKHR";
case OpDecorateStringGOOGLE: return "OpDecorateStringGOOGLE";
case OpMemberDecorateStringGOOGLE: return "OpMemberDecorateStringGOOGLE";
#ifdef NV_EXTENSIONS
case OpGroupNonUniformPartitionNV: return "OpGroupNonUniformPartitionNV";
case OpReportIntersectionNV: return "OpReportIntersectionNV";
case OpIgnoreIntersectionNV: return "OpIgnoreIntersectionNV";
@ -1343,13 +1334,17 @@ const char* OpcodeString(int op)
case OpExecuteCallableNV: return "OpExecuteCallableNV";
case OpImageSampleFootprintNV: return "OpImageSampleFootprintNV";
case OpWritePackedPrimitiveIndices4x8NV: return "OpWritePackedPrimitiveIndices4x8NV";
#endif
case OpTypeCooperativeMatrixNV: return "OpTypeCooperativeMatrixNV";
case OpCooperativeMatrixLoadNV: return "OpCooperativeMatrixLoadNV";
case OpCooperativeMatrixStoreNV: return "OpCooperativeMatrixStoreNV";
case OpCooperativeMatrixMulAddNV: return "OpCooperativeMatrixMulAddNV";
case OpCooperativeMatrixLengthNV: return "OpCooperativeMatrixLengthNV";
case OpDemoteToHelperInvocationEXT: return "OpDemoteToHelperInvocationEXT";
case OpIsHelperInvocationEXT: return "OpIsHelperInvocationEXT";
case OpBeginInvocationInterlockEXT: return "OpBeginInvocationInterlockEXT";
case OpEndInvocationInterlockEXT: return "OpEndInvocationInterlockEXT";
default:
return "Bad";
@ -1464,6 +1459,8 @@ void Parameterize()
InstructionDesc[OpModuleProcessed].setResultAndType(false, false);
InstructionDesc[OpTypeCooperativeMatrixNV].setResultAndType(true, false);
InstructionDesc[OpCooperativeMatrixStoreNV].setResultAndType(false, false);
InstructionDesc[OpBeginInvocationInterlockEXT].setResultAndType(false, false);
InstructionDesc[OpEndInvocationInterlockEXT].setResultAndType(false, false);
// Specific additional context-dependent operands
@ -2656,7 +2653,6 @@ void Parameterize()
InstructionDesc[OpModuleProcessed].operands.push(OperandLiteralString, "'process'");
#ifdef AMD_EXTENSIONS
InstructionDesc[OpGroupIAddNonUniformAMD].operands.push(OperandScope, "'Execution'");
InstructionDesc[OpGroupIAddNonUniformAMD].operands.push(OperandGroupOperation, "'Operation'");
InstructionDesc[OpGroupIAddNonUniformAMD].operands.push(OperandId, "'X'");
@ -2695,9 +2691,7 @@ void Parameterize()
InstructionDesc[OpFragmentFetchAMD].operands.push(OperandId, "'Image'");
InstructionDesc[OpFragmentFetchAMD].operands.push(OperandId, "'Coordinate'");
InstructionDesc[OpFragmentFetchAMD].operands.push(OperandId, "'Fragment Index'");
#endif
#ifdef NV_EXTENSIONS
InstructionDesc[OpGroupNonUniformPartitionNV].operands.push(OperandId, "X");
InstructionDesc[OpTypeAccelerationStructureNV].setResultAndType(true, false);
@ -2735,7 +2729,6 @@ void Parameterize()
InstructionDesc[OpWritePackedPrimitiveIndices4x8NV].operands.push(OperandId, "'Index Offset'");
InstructionDesc[OpWritePackedPrimitiveIndices4x8NV].operands.push(OperandId, "'Packed Indices'");
#endif
InstructionDesc[OpTypeCooperativeMatrixNV].operands.push(OperandId, "'Component Type'");
InstructionDesc[OpTypeCooperativeMatrixNV].operands.push(OperandId, "'Scope'");
@ -2762,6 +2755,10 @@ void Parameterize()
InstructionDesc[OpCooperativeMatrixMulAddNV].operands.push(OperandId, "'C'");
InstructionDesc[OpCooperativeMatrixLengthNV].operands.push(OperandId, "'Type'");
InstructionDesc[OpDemoteToHelperInvocationEXT].setResultAndType(false, false);
InstructionDesc[OpReadClockKHR].operands.push(OperandScope, "'Scope'");
}
}; // end spv namespace

View File

@ -91,6 +91,7 @@ enum AddressingModel {
AddressingModelLogical = 0,
AddressingModelPhysical32 = 1,
AddressingModelPhysical64 = 2,
AddressingModelPhysicalStorageBuffer64 = 5348,
AddressingModelPhysicalStorageBuffer64EXT = 5348,
AddressingModelMax = 0x7fffffff,
};
@ -99,6 +100,7 @@ enum MemoryModel {
MemoryModelSimple = 0,
MemoryModelGLSL450 = 1,
MemoryModelOpenCL = 2,
MemoryModelVulkan = 3,
MemoryModelVulkanKHR = 3,
MemoryModelMax = 0x7fffffff,
};
@ -154,6 +156,12 @@ enum ExecutionMode {
ExecutionModeDerivativeGroupQuadsNV = 5289,
ExecutionModeDerivativeGroupLinearNV = 5290,
ExecutionModeOutputTrianglesNV = 5298,
ExecutionModePixelInterlockOrderedEXT = 5366,
ExecutionModePixelInterlockUnorderedEXT = 5367,
ExecutionModeSampleInterlockOrderedEXT = 5368,
ExecutionModeSampleInterlockUnorderedEXT = 5369,
ExecutionModeShadingRateInterlockOrderedEXT = 5370,
ExecutionModeShadingRateInterlockUnorderedEXT = 5371,
ExecutionModeMax = 0x7fffffff,
};
@ -177,6 +185,7 @@ enum StorageClass {
StorageClassHitAttributeNV = 5339,
StorageClassIncomingRayPayloadNV = 5342,
StorageClassShaderRecordBufferNV = 5343,
StorageClassPhysicalStorageBuffer = 5349,
StorageClassPhysicalStorageBufferEXT = 5349,
StorageClassMax = 0x7fffffff,
};
@ -305,9 +314,13 @@ enum ImageOperandsShift {
ImageOperandsConstOffsetsShift = 5,
ImageOperandsSampleShift = 6,
ImageOperandsMinLodShift = 7,
ImageOperandsMakeTexelAvailableShift = 8,
ImageOperandsMakeTexelAvailableKHRShift = 8,
ImageOperandsMakeTexelVisibleShift = 9,
ImageOperandsMakeTexelVisibleKHRShift = 9,
ImageOperandsNonPrivateTexelShift = 10,
ImageOperandsNonPrivateTexelKHRShift = 10,
ImageOperandsVolatileTexelShift = 11,
ImageOperandsVolatileTexelKHRShift = 11,
ImageOperandsSignExtendShift = 12,
ImageOperandsZeroExtendShift = 13,
@ -324,9 +337,13 @@ enum ImageOperandsMask {
ImageOperandsConstOffsetsMask = 0x00000020,
ImageOperandsSampleMask = 0x00000040,
ImageOperandsMinLodMask = 0x00000080,
ImageOperandsMakeTexelAvailableMask = 0x00000100,
ImageOperandsMakeTexelAvailableKHRMask = 0x00000100,
ImageOperandsMakeTexelVisibleMask = 0x00000200,
ImageOperandsMakeTexelVisibleKHRMask = 0x00000200,
ImageOperandsNonPrivateTexelMask = 0x00000400,
ImageOperandsNonPrivateTexelKHRMask = 0x00000400,
ImageOperandsVolatileTexelMask = 0x00000800,
ImageOperandsVolatileTexelKHRMask = 0x00000800,
ImageOperandsSignExtendMask = 0x00001000,
ImageOperandsZeroExtendMask = 0x00002000,
@ -442,13 +459,17 @@ enum Decoration {
DecorationPerViewNV = 5272,
DecorationPerTaskNV = 5273,
DecorationPerVertexNV = 5285,
DecorationNonUniform = 5300,
DecorationNonUniformEXT = 5300,
DecorationRestrictPointer = 5355,
DecorationRestrictPointerEXT = 5355,
DecorationAliasedPointer = 5356,
DecorationAliasedPointerEXT = 5356,
DecorationCounterBuffer = 5634,
DecorationHlslCounterBufferGOOGLE = 5634,
DecorationHlslSemanticGOOGLE = 5635,
DecorationUserSemantic = 5635,
DecorationUserTypeGOOGLE = 5636,
DecorationMax = 0x7fffffff,
};
@ -551,6 +572,10 @@ enum BuiltIn {
BuiltInHitTNV = 5332,
BuiltInHitKindNV = 5333,
BuiltInIncomingRayFlagsNV = 5351,
BuiltInWarpsPerSMNV = 5374,
BuiltInSMCountNV = 5375,
BuiltInWarpIDNV = 5376,
BuiltInSMIDNV = 5377,
BuiltInMax = 0x7fffffff,
};
@ -619,9 +644,13 @@ enum MemorySemanticsShift {
MemorySemanticsCrossWorkgroupMemoryShift = 9,
MemorySemanticsAtomicCounterMemoryShift = 10,
MemorySemanticsImageMemoryShift = 11,
MemorySemanticsOutputMemoryShift = 12,
MemorySemanticsOutputMemoryKHRShift = 12,
MemorySemanticsMakeAvailableShift = 13,
MemorySemanticsMakeAvailableKHRShift = 13,
MemorySemanticsMakeVisibleShift = 14,
MemorySemanticsMakeVisibleKHRShift = 14,
MemorySemanticsVolatileShift = 15,
MemorySemanticsMax = 0x7fffffff,
};
@ -637,17 +666,24 @@ enum MemorySemanticsMask {
MemorySemanticsCrossWorkgroupMemoryMask = 0x00000200,
MemorySemanticsAtomicCounterMemoryMask = 0x00000400,
MemorySemanticsImageMemoryMask = 0x00000800,
MemorySemanticsOutputMemoryMask = 0x00001000,
MemorySemanticsOutputMemoryKHRMask = 0x00001000,
MemorySemanticsMakeAvailableMask = 0x00002000,
MemorySemanticsMakeAvailableKHRMask = 0x00002000,
MemorySemanticsMakeVisibleMask = 0x00004000,
MemorySemanticsMakeVisibleKHRMask = 0x00004000,
MemorySemanticsVolatileMask = 0x00008000,
};
enum MemoryAccessShift {
MemoryAccessVolatileShift = 0,
MemoryAccessAlignedShift = 1,
MemoryAccessNontemporalShift = 2,
MemoryAccessMakePointerAvailableShift = 3,
MemoryAccessMakePointerAvailableKHRShift = 3,
MemoryAccessMakePointerVisibleShift = 4,
MemoryAccessMakePointerVisibleKHRShift = 4,
MemoryAccessNonPrivatePointerShift = 5,
MemoryAccessNonPrivatePointerKHRShift = 5,
MemoryAccessMax = 0x7fffffff,
};
@ -657,8 +693,11 @@ enum MemoryAccessMask {
MemoryAccessVolatileMask = 0x00000001,
MemoryAccessAlignedMask = 0x00000002,
MemoryAccessNontemporalMask = 0x00000004,
MemoryAccessMakePointerAvailableMask = 0x00000008,
MemoryAccessMakePointerAvailableKHRMask = 0x00000008,
MemoryAccessMakePointerVisibleMask = 0x00000010,
MemoryAccessMakePointerVisibleKHRMask = 0x00000010,
MemoryAccessNonPrivatePointerMask = 0x00000020,
MemoryAccessNonPrivatePointerKHRMask = 0x00000020,
};
@ -668,6 +707,7 @@ enum Scope {
ScopeWorkgroup = 2,
ScopeSubgroup = 3,
ScopeInvocation = 4,
ScopeQueueFamily = 5,
ScopeQueueFamilyKHR = 5,
ScopeMax = 0x7fffffff,
};
@ -768,6 +808,8 @@ enum Capability {
CapabilityGroupNonUniformShuffleRelative = 66,
CapabilityGroupNonUniformClustered = 67,
CapabilityGroupNonUniformQuad = 68,
CapabilityShaderLayer = 69,
CapabilityShaderViewportIndex = 70,
CapabilitySubgroupBallotKHR = 4423,
CapabilityDrawParameters = 4427,
CapabilitySubgroupVoteKHR = 4431,
@ -796,6 +838,7 @@ enum Capability {
CapabilityFragmentMaskAMD = 5010,
CapabilityStencilExportEXT = 5013,
CapabilityImageReadWriteLodAMD = 5015,
CapabilityShaderClockKHR = 5055,
CapabilitySampleMaskOverrideCoverageNV = 5249,
CapabilityGeometryShaderPassthroughNV = 5251,
CapabilityShaderViewportIndexLayerEXT = 5254,
@ -811,28 +854,49 @@ enum Capability {
CapabilityFragmentDensityEXT = 5291,
CapabilityShadingRateNV = 5291,
CapabilityGroupNonUniformPartitionedNV = 5297,
CapabilityShaderNonUniform = 5301,
CapabilityShaderNonUniformEXT = 5301,
CapabilityRuntimeDescriptorArray = 5302,
CapabilityRuntimeDescriptorArrayEXT = 5302,
CapabilityInputAttachmentArrayDynamicIndexing = 5303,
CapabilityInputAttachmentArrayDynamicIndexingEXT = 5303,
CapabilityUniformTexelBufferArrayDynamicIndexing = 5304,
CapabilityUniformTexelBufferArrayDynamicIndexingEXT = 5304,
CapabilityStorageTexelBufferArrayDynamicIndexing = 5305,
CapabilityStorageTexelBufferArrayDynamicIndexingEXT = 5305,
CapabilityUniformBufferArrayNonUniformIndexing = 5306,
CapabilityUniformBufferArrayNonUniformIndexingEXT = 5306,
CapabilitySampledImageArrayNonUniformIndexing = 5307,
CapabilitySampledImageArrayNonUniformIndexingEXT = 5307,
CapabilityStorageBufferArrayNonUniformIndexing = 5308,
CapabilityStorageBufferArrayNonUniformIndexingEXT = 5308,
CapabilityStorageImageArrayNonUniformIndexing = 5309,
CapabilityStorageImageArrayNonUniformIndexingEXT = 5309,
CapabilityInputAttachmentArrayNonUniformIndexing = 5310,
CapabilityInputAttachmentArrayNonUniformIndexingEXT = 5310,
CapabilityUniformTexelBufferArrayNonUniformIndexing = 5311,
CapabilityUniformTexelBufferArrayNonUniformIndexingEXT = 5311,
CapabilityStorageTexelBufferArrayNonUniformIndexing = 5312,
CapabilityStorageTexelBufferArrayNonUniformIndexingEXT = 5312,
CapabilityRayTracingNV = 5340,
CapabilityVulkanMemoryModel = 5345,
CapabilityVulkanMemoryModelKHR = 5345,
CapabilityVulkanMemoryModelDeviceScope = 5346,
CapabilityVulkanMemoryModelDeviceScopeKHR = 5346,
CapabilityPhysicalStorageBufferAddresses = 5347,
CapabilityPhysicalStorageBufferAddressesEXT = 5347,
CapabilityComputeDerivativeGroupLinearNV = 5350,
CapabilityCooperativeMatrixNV = 5357,
CapabilityFragmentShaderSampleInterlockEXT = 5363,
CapabilityFragmentShaderShadingRateInterlockEXT = 5372,
CapabilityShaderSMBuiltinsNV = 5373,
CapabilityFragmentShaderPixelInterlockEXT = 5378,
CapabilityDemoteToHelperInvocationEXT = 5379,
CapabilitySubgroupShuffleINTEL = 5568,
CapabilitySubgroupBufferBlockIOINTEL = 5569,
CapabilitySubgroupImageBlockIOINTEL = 5570,
CapabilitySubgroupImageMediaBlockIOINTEL = 5579,
CapabilityIntegerFunctions2INTEL = 5584,
CapabilitySubgroupAvcMotionEstimationINTEL = 5696,
CapabilitySubgroupAvcMotionEstimationIntraINTEL = 5697,
CapabilitySubgroupAvcMotionEstimationChromaINTEL = 5698,
@ -1200,6 +1264,7 @@ enum Op {
OpGroupSMaxNonUniformAMD = 5007,
OpFragmentMaskFetchAMD = 5011,
OpFragmentFetchAMD = 5012,
OpReadClockKHR = 5056,
OpImageSampleFootprintNV = 5283,
OpGroupNonUniformPartitionNV = 5296,
OpWritePackedPrimitiveIndices4x8NV = 5299,
@ -1214,6 +1279,10 @@ enum Op {
OpCooperativeMatrixStoreNV = 5360,
OpCooperativeMatrixMulAddNV = 5361,
OpCooperativeMatrixLengthNV = 5362,
OpBeginInvocationInterlockEXT = 5364,
OpEndInvocationInterlockEXT = 5365,
OpDemoteToHelperInvocationEXT = 5380,
OpIsHelperInvocationEXT = 5381,
OpSubgroupShuffleINTEL = 5571,
OpSubgroupShuffleDownINTEL = 5572,
OpSubgroupShuffleUpINTEL = 5573,
@ -1224,6 +1293,20 @@ enum Op {
OpSubgroupImageBlockWriteINTEL = 5578,
OpSubgroupImageMediaBlockReadINTEL = 5580,
OpSubgroupImageMediaBlockWriteINTEL = 5581,
OpUCountLeadingZerosINTEL = 5585,
OpUCountTrailingZerosINTEL = 5586,
OpAbsISubINTEL = 5587,
OpAbsUSubINTEL = 5588,
OpIAddSatINTEL = 5589,
OpUAddSatINTEL = 5590,
OpIAverageINTEL = 5591,
OpUAverageINTEL = 5592,
OpIAverageRoundedINTEL = 5593,
OpUAverageRoundedINTEL = 5594,
OpISubSatINTEL = 5595,
OpUSubSatINTEL = 5596,
OpIMul32x16INTEL = 5597,
OpUMul32x16INTEL = 5598,
OpDecorateString = 5632,
OpDecorateStringGOOGLE = 5632,
OpMemberDecorateString = 5633,
@ -1714,6 +1797,7 @@ inline void HasResultAndType(Op opcode, bool *hasResult, bool *hasResultType) {
case OpGroupSMaxNonUniformAMD: *hasResult = true; *hasResultType = true; break;
case OpFragmentMaskFetchAMD: *hasResult = true; *hasResultType = true; break;
case OpFragmentFetchAMD: *hasResult = true; *hasResultType = true; break;
case OpReadClockKHR: *hasResult = true; *hasResultType = true; break;
case OpImageSampleFootprintNV: *hasResult = true; *hasResultType = true; break;
case OpGroupNonUniformPartitionNV: *hasResult = true; *hasResultType = true; break;
case OpWritePackedPrimitiveIndices4x8NV: *hasResult = false; *hasResultType = false; break;
@ -1728,6 +1812,10 @@ inline void HasResultAndType(Op opcode, bool *hasResult, bool *hasResultType) {
case OpCooperativeMatrixStoreNV: *hasResult = false; *hasResultType = false; break;
case OpCooperativeMatrixMulAddNV: *hasResult = true; *hasResultType = true; break;
case OpCooperativeMatrixLengthNV: *hasResult = true; *hasResultType = true; break;
case OpBeginInvocationInterlockEXT: *hasResult = false; *hasResultType = false; break;
case OpEndInvocationInterlockEXT: *hasResult = false; *hasResultType = false; break;
case OpDemoteToHelperInvocationEXT: *hasResult = false; *hasResultType = false; break;
case OpIsHelperInvocationEXT: *hasResult = true; *hasResultType = true; break;
case OpSubgroupShuffleINTEL: *hasResult = true; *hasResultType = true; break;
case OpSubgroupShuffleDownINTEL: *hasResult = true; *hasResultType = true; break;
case OpSubgroupShuffleUpINTEL: *hasResult = true; *hasResultType = true; break;
@ -1738,10 +1826,22 @@ inline void HasResultAndType(Op opcode, bool *hasResult, bool *hasResultType) {
case OpSubgroupImageBlockWriteINTEL: *hasResult = false; *hasResultType = false; break;
case OpSubgroupImageMediaBlockReadINTEL: *hasResult = true; *hasResultType = true; break;
case OpSubgroupImageMediaBlockWriteINTEL: *hasResult = false; *hasResultType = false; break;
case OpUCountLeadingZerosINTEL: *hasResult = true; *hasResultType = true; break;
case OpUCountTrailingZerosINTEL: *hasResult = true; *hasResultType = true; break;
case OpAbsISubINTEL: *hasResult = true; *hasResultType = true; break;
case OpAbsUSubINTEL: *hasResult = true; *hasResultType = true; break;
case OpIAddSatINTEL: *hasResult = true; *hasResultType = true; break;
case OpUAddSatINTEL: *hasResult = true; *hasResultType = true; break;
case OpIAverageINTEL: *hasResult = true; *hasResultType = true; break;
case OpUAverageINTEL: *hasResult = true; *hasResultType = true; break;
case OpIAverageRoundedINTEL: *hasResult = true; *hasResultType = true; break;
case OpUAverageRoundedINTEL: *hasResult = true; *hasResultType = true; break;
case OpISubSatINTEL: *hasResult = true; *hasResultType = true; break;
case OpUSubSatINTEL: *hasResult = true; *hasResultType = true; break;
case OpIMul32x16INTEL: *hasResult = true; *hasResultType = true; break;
case OpUMul32x16INTEL: *hasResult = true; *hasResultType = true; break;
case OpDecorateString: *hasResult = false; *hasResultType = false; break;
case OpDecorateStringGOOGLE: *hasResult = false; *hasResultType = false; break;
case OpMemberDecorateString: *hasResult = false; *hasResultType = false; break;
case OpMemberDecorateStringGOOGLE: *hasResult = false; *hasResultType = false; break;
case OpVmeImageINTEL: *hasResult = true; *hasResultType = true; break;
case OpTypeVmeImageINTEL: *hasResult = true; *hasResultType = false; break;
case OpTypeAvcImePayloadINTEL: *hasResult = true; *hasResultType = false; break;

53
thirdparty/glslang/SPIRV/spvIR.h vendored Normal file → Executable file
View File

@ -226,6 +226,36 @@ public:
return nullptr;
}
// Change this block into a canonical dead merge block. Delete instructions
// as necessary. A canonical dead merge block has only an OpLabel and an
// OpUnreachable.
void rewriteAsCanonicalUnreachableMerge() {
assert(localVariables.empty());
// Delete all instructions except for the label.
assert(instructions.size() > 0);
instructions.resize(1);
successors.clear();
Instruction* unreachable = new Instruction(OpUnreachable);
addInstruction(std::unique_ptr<Instruction>(unreachable));
}
// Change this block into a canonical dead continue target branching to the
// given header ID. Delete instructions as necessary. A canonical dead continue
// target has only an OpLabel and an unconditional branch back to the corresponding
// header.
void rewriteAsCanonicalUnreachableContinue(Block* header) {
assert(localVariables.empty());
// Delete all instructions except for the label.
assert(instructions.size() > 0);
instructions.resize(1);
successors.clear();
// Add OpBranch back to the header.
assert(header != nullptr);
Instruction* branch = new Instruction(OpBranch);
branch->addIdOperand(header->getId());
addInstruction(std::unique_ptr<Instruction>(branch));
successors.push_back(header);
}
bool isTerminated() const
{
switch (instructions.back()->getOpCode()) {
@ -235,6 +265,7 @@ public:
case OpKill:
case OpReturn:
case OpReturnValue:
case OpUnreachable:
return true;
default:
return false;
@ -268,10 +299,24 @@ protected:
bool unreachable;
};
// The different reasons for reaching a block in the inReadableOrder traversal.
enum ReachReason {
// Reachable from the entry block via transfers of control, i.e. branches.
ReachViaControlFlow = 0,
// A continue target that is not reachable via control flow.
ReachDeadContinue,
// A merge block that is not reachable via control flow.
ReachDeadMerge
};
// Traverses the control-flow graph rooted at root in an order suited for
// readable code generation. Invokes callback at every node in the traversal
// order.
void inReadableOrder(Block* root, std::function<void(Block*)> callback);
// order. The callback arguments are:
// - the block,
// - the reason we reached the block,
// - if the reason was that block is an unreachable continue or unreachable merge block
// then the last parameter is the corresponding header block.
void inReadableOrder(Block* root, std::function<void(Block*, ReachReason, Block* header)> callback);
//
// SPIR-V IR Function.
@ -321,7 +366,7 @@ public:
parameterInstructions[p]->dump(out);
// Blocks
inReadableOrder(blocks[0], [&out](const Block* b) { b->dump(out); });
inReadableOrder(blocks[0], [&out](const Block* b, ReachReason, Block*) { b->dump(out); });
Instruction end(0, 0, OpFunctionEnd);
end.dump(out);
}
@ -436,6 +481,6 @@ __inline void Block::addInstruction(std::unique_ptr<Instruction> inst)
parent.getParent().mapInstruction(raw_instruction);
}
}; // end spv namespace
} // end spv namespace
#endif // spvIR_H