Skip to content

Consolidate buffer store translation #7251

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pow2clk merged 4 commits into from
Mar 24, 2025
Merged

Consolidate buffer store translation #7251

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
34e8685
Consolidate buffer store translation
Mar 8, 2025
6cbfc40
Add new store tests
Mar 21, 2025
52e5abe
use maybe_unused instead of NDEBUG
Mar 24, 2025
a989fab
Remove unneeded metatdata, Capitalize variable.
Mar 24, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
175 changes: 63 additions & 112 deletions lib/HLSL/HLOperationLower.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4335,18 +4335,15 @@ void Split64bitValForStore(Type *EltTy, ArrayRef<Value *> vals, unsigned size,
}

void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
Value *offset, IRBuilder<> &Builder, hlsl::OP *OP,
Value *sampIdx = nullptr) {
Value *Idx, Value *offset, IRBuilder<> &Builder,
hlsl::OP *OP, Value *sampIdx = nullptr) {
Type *Ty = val->getType();

// This function is no longer used for lowering stores to a
// structured buffer.
DXASSERT_NOMSG(RK != DxilResource::Kind::StructuredBuffer);

OP::OpCode opcode = OP::OpCode::NumOpCodes;
bool IsTyped = true;
switch (RK) {
case DxilResource::Kind::RawBuffer:
case DxilResource::Kind::StructuredBuffer:
IsTyped = false;
opcode = OP::OpCode::RawBufferStore;
break;
case DxilResource::Kind::TypedBuffer:
Expand All @@ -4364,10 +4361,6 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
break;
}

bool isTyped = opcode == OP::OpCode::TextureStore ||
opcode == OP::OpCode::TextureStoreSample ||
RK == DxilResource::Kind::TypedBuffer;

Type *i32Ty = Builder.getInt32Ty();
Type *i64Ty = Builder.getInt64Ty();
Type *doubleTy = Builder.getDoubleTy();
Expand All @@ -4390,7 +4383,7 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
alignValue = 4;
Constant *Alignment = OP->GetI32Const(alignValue);
bool is64 = EltTy == i64Ty || EltTy == doubleTy;
if (is64 && isTyped) {
if (is64 && IsTyped) {
EltTy = i32Ty;
}

Expand All @@ -4406,38 +4399,42 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
storeArgs.emplace_back(opArg); // opcode
storeArgs.emplace_back(handle); // resource handle

unsigned offset0Idx = 0;
if (RK == DxilResource::Kind::RawBuffer ||
RK == DxilResource::Kind::TypedBuffer) {
// Offset 0
if (offset->getType()->isVectorTy()) {
Value *scalarOffset = Builder.CreateExtractElement(offset, (uint64_t)0);
storeArgs.emplace_back(scalarOffset); // offset
unsigned OffsetIdx = 0;
if (opcode == OP::OpCode::RawBufferStore ||
opcode == OP::OpCode::BufferStore) {
// Append Coord0 (Index) value.
if (Idx->getType()->isVectorTy()) {
Value *ScalarIdx = Builder.CreateExtractElement(Idx, (uint64_t)0);
storeArgs.emplace_back(ScalarIdx); // Coord0 (Index).
} else {
storeArgs.emplace_back(offset); // offset
storeArgs.emplace_back(Idx); // Coord0 (Index).
}

// Store offset0 for later use
offset0Idx = storeArgs.size() - 1;
// Store OffsetIdx representing the argument that may need to be incremented
// later to load additional chunks of data.
// Only structured buffers can use the offset parameter.
// Others must increment the index.
if (RK == DxilResource::Kind::StructuredBuffer)
OffsetIdx = storeArgs.size();
else
OffsetIdx = storeArgs.size() - 1;

// Offset 1
storeArgs.emplace_back(undefI);
// Coord1 (Offset).
// Only relevant when storing more than 4 elements to structured buffers.
storeArgs.emplace_back(offset);
} else {
// texture store
unsigned coordSize = DxilResource::GetNumCoords(RK);

// Set x first.
if (offset->getType()->isVectorTy())
storeArgs.emplace_back(Builder.CreateExtractElement(offset, (uint64_t)0));
if (Idx->getType()->isVectorTy())
storeArgs.emplace_back(Builder.CreateExtractElement(Idx, (uint64_t)0));
else
storeArgs.emplace_back(offset);

// Store offset0 for later use
offset0Idx = storeArgs.size() - 1;
storeArgs.emplace_back(Idx);

for (unsigned i = 1; i < 3; i++) {
if (i < coordSize)
storeArgs.emplace_back(Builder.CreateExtractElement(offset, i));
storeArgs.emplace_back(Builder.CreateExtractElement(Idx, i));
else
storeArgs.emplace_back(undefI);
}
Expand All @@ -4464,30 +4461,24 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
}

for (unsigned j = 0; j < storeArgsList.size(); j++) {

// For second and subsequent store calls, increment the offset0 (i.e. store
// index)
// For second and subsequent store calls, increment the resource-appropriate
// index or offset parameter.
if (j > 0) {
// Greater than four-components store is not allowed for
// TypedBuffer and Textures. So greater than four elements
// scenario should only get hit here for RawBuffer.
DXASSERT_NOMSG(RK == DxilResource::Kind::RawBuffer);
unsigned EltSize = OP->GetAllocSizeForType(EltTy);
unsigned newOffset = EltSize * MaxStoreElemCount * j;
Value *newOffsetVal = ConstantInt::get(Builder.getInt32Ty(), newOffset);
newOffsetVal =
Builder.CreateAdd(storeArgsList[0][offset0Idx], newOffsetVal);
storeArgsList[j][offset0Idx] = newOffsetVal;
unsigned NewCoord = EltSize * MaxStoreElemCount * j;
Value *NewCoordVal = ConstantInt::get(Builder.getInt32Ty(), NewCoord);
NewCoordVal = Builder.CreateAdd(storeArgsList[0][OffsetIdx], NewCoordVal);
storeArgsList[j][OffsetIdx] = NewCoordVal;
}

// values
// Set value parameters.
uint8_t mask = 0;
if (Ty->isVectorTy()) {
unsigned vecSize =
std::min((j + 1) * MaxStoreElemCount, Ty->getVectorNumElements()) -
(j * MaxStoreElemCount);
Value *emptyVal = undefVal;
if (isTyped) {
if (IsTyped) {
mask = DXIL::kCompMask_All;
emptyVal = Builder.CreateExtractElement(val, (uint64_t)0);
}
Expand All @@ -4503,7 +4494,7 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
}

} else {
if (isTyped) {
if (IsTyped) {
mask = DXIL::kCompMask_All;
storeArgsList[j].emplace_back(val);
storeArgsList[j].emplace_back(val);
Expand All @@ -4518,7 +4509,7 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
}
}

if (is64 && isTyped) {
if (is64 && IsTyped) {
unsigned size = 1;
if (Ty->isVectorTy()) {
size =
Expand Down Expand Up @@ -4576,7 +4567,8 @@ Value *TranslateResourceStore(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,

Value *val = CI->getArgOperand(HLOperandIndex::kStoreValOpIdx);
Value *offset = CI->getArgOperand(HLOperandIndex::kStoreOffsetOpIdx);
TranslateStore(RK, handle, val, offset, Builder, hlslOP);
Value *UndefI = UndefValue::get(Builder.getInt32Ty());
TranslateStore(RK, handle, val, offset, UndefI, Builder, hlslOP);

return nullptr;
}
Expand Down Expand Up @@ -7907,40 +7899,11 @@ Value *TranslateStructBufMatLd(CallInst *CI, IRBuilder<> &Builder,
void TranslateStructBufMatSt(Type *matType, IRBuilder<> &Builder, Value *handle,
hlsl::OP *OP, Value *bufIdx, Value *baseOffset,
Value *val, const DataLayout &DL) {
HLMatrixType MatTy = HLMatrixType::cast(matType);
Type *EltTy = MatTy.getElementTypeForMem();

val = MatTy.emitLoweredRegToMem(val, Builder);

unsigned EltSize = DL.getTypeAllocSize(EltTy);
Constant *Alignment = OP->GetI32Const(EltSize);
Value *offset = baseOffset;
if (baseOffset == nullptr)
offset = OP->GetU32Const(0);

unsigned matSize = MatTy.getNumElements();
Value *undefElt = UndefValue::get(EltTy);

unsigned storeSize = matSize;
if (matSize % 4) {
storeSize = matSize + 4 - (matSize & 3);
}
std::vector<Value *> elts(storeSize, undefElt);
for (unsigned i = 0; i < matSize; i++)
elts[i] = Builder.CreateExtractElement(val, i);

for (unsigned i = 0; i < matSize; i += 4) {
uint8_t mask = 0;
for (unsigned j = 0; j < 4 && (i + j) < matSize; j++) {
if (elts[i + j] != undefElt)
mask |= (1 << j);
}
GenerateStructBufSt(handle, bufIdx, offset, EltTy, OP, Builder,
{elts[i], elts[i + 1], elts[i + 2], elts[i + 3]}, mask,
Alignment);
// Update offset by 4*4bytes.
offset = Builder.CreateAdd(offset, OP->GetU32Const(4 * EltSize));
}
[[maybe_unused]] HLMatrixType MatTy = HLMatrixType::cast(matType);
DXASSERT(MatTy.getLoweredVectorType(false /*MemRepr*/) == val->getType(),
"helper type should match vectorized matrix");
TranslateStore(DxilResource::Kind::StructuredBuffer, handle, val, bufIdx,
baseOffset, Builder, OP);
}

void TranslateStructBufMatLdSt(CallInst *CI, Value *handle, HLResource::Kind RK,
Expand Down Expand Up @@ -8085,6 +8048,9 @@ void TranslateStructBufMatSubscript(CallInst *CI, Value *handle,

GEP->eraseFromParent();
} else if (StoreInst *stUser = dyn_cast<StoreInst>(subsUser)) {
// Store elements of matrix in a struct. Needs to be done one scalar at a
// time even for vectors in the case that matrix orientation spreads the
// indexed scalars throughout the matrix vector.
IRBuilder<> stBuilder(stUser);
Value *Val = stUser->getValueOperand();
if (Val->getType()->isVectorTy()) {
Expand All @@ -8108,6 +8074,9 @@ void TranslateStructBufMatSubscript(CallInst *CI, Value *handle,
LoadInst *ldUser = cast<LoadInst>(subsUser);
IRBuilder<> ldBuilder(ldUser);
Value *ldData = UndefValue::get(resultType);
// Load elements of matrix in a struct. Needs to be done one scalar at a
// time even for vectors in the case that matrix orientation spreads the
// indexed scalars throughout the matrix vector.
if (resultType->isVectorTy()) {
for (unsigned i = 0; i < resultSize; i++) {
Value *ResultElt;
Expand Down Expand Up @@ -8248,30 +8217,9 @@ void TranslateStructBufSubscriptUser(Instruction *user, Value *handle,
LdInst->eraseFromParent();
} else if (StoreInst *StInst = dyn_cast<StoreInst>(user)) {
// Store of scalar/vector within a struct or structured raw store.
Type *Ty = StInst->getValueOperand()->getType();
Type *pOverloadTy = Ty->getScalarType();
Value *offset = baseOffset;

Value *val = StInst->getValueOperand();
Value *undefVal = llvm::UndefValue::get(pOverloadTy);
Value *vals[] = {undefVal, undefVal, undefVal, undefVal};
uint8_t mask = 0;
if (Ty->isVectorTy()) {
unsigned vectorNumElements = Ty->getVectorNumElements();
DXASSERT(vectorNumElements <= 4, "up to 4 elements in vector");
assert(vectorNumElements <= 4);
for (unsigned i = 0; i < vectorNumElements; i++) {
vals[i] = Builder.CreateExtractElement(val, i);
mask |= (1 << i);
}
} else {
vals[0] = val;
mask = DXIL::kCompMask_X;
}
Constant *alignment =
OP->GetI32Const(DL.getTypeAllocSize(Ty->getScalarType()));
GenerateStructBufSt(handle, bufIdx, offset, pOverloadTy, OP, Builder, vals,
mask, alignment);
TranslateStore(DxilResource::Kind::StructuredBuffer, handle, val, bufIdx,
baseOffset, Builder, OP);
StInst->eraseFromParent();
} else if (BitCastInst *BCI = dyn_cast<BitCastInst>(user)) {
// Recurse users
Expand Down Expand Up @@ -8418,14 +8366,15 @@ void TranslateTypedBufferSubscript(CallInst *CI, HLOperationLowerHelper &helper,
User *user = *(It++);
Instruction *I = cast<Instruction>(user);
IRBuilder<> Builder(I);
Value *UndefI = UndefValue::get(Builder.getInt32Ty());
if (LoadInst *ldInst = dyn_cast<LoadInst>(user)) {
TranslateTypedBufSubscript(CI, RK, RC, handle, ldInst, Builder, hlslOP,
helper.dataLayout);
} else if (StoreInst *stInst = dyn_cast<StoreInst>(user)) {
Value *val = stInst->getValueOperand();
TranslateStore(RK, handle, val,
CI->getArgOperand(HLOperandIndex::kStoreOffsetOpIdx),
Builder, hlslOP);
CI->getArgOperand(HLOperandIndex::kSubscriptIndexOpIdx),
UndefI, Builder, hlslOP);
// delete the st
stInst->eraseFromParent();
} else if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(user)) {
Expand All @@ -8450,9 +8399,10 @@ void TranslateTypedBufferSubscript(CallInst *CI, HLOperationLowerHelper &helper,
// Generate St.
// Reset insert point, UpdateVectorElt may move SI to different block.
StBuilder.SetInsertPoint(SI);
TranslateStore(RK, handle, ldVal,
CI->getArgOperand(HLOperandIndex::kStoreOffsetOpIdx),
StBuilder, hlslOP);
TranslateStore(
RK, handle, ldVal,
CI->getArgOperand(HLOperandIndex::kSubscriptIndexOpIdx), UndefI,
StBuilder, hlslOP);
SI->eraseFromParent();
continue;
}
Expand Down Expand Up @@ -8642,9 +8592,10 @@ void TranslateHLSubscript(CallInst *CI, HLSubscriptOpcode opcode,
} else {
StoreInst *stInst = cast<StoreInst>(*U);
Value *val = stInst->getValueOperand();
Value *UndefI = UndefValue::get(Builder.getInt32Ty());
TranslateStore(RK, handle, val,
CI->getArgOperand(HLOperandIndex::kStoreOffsetOpIdx),
Builder, hlslOP, mipLevel);
CI->getArgOperand(HLOperandIndex::kSubscriptIndexOpIdx),
UndefI, Builder, hlslOP, mipLevel);
stInst->eraseFromParent();
}
Translated = true;
Expand Down
Loading