-
Notifications
You must be signed in to change notification settings - Fork 12.4k
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.
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
[DirectX] Handle dx.RawBuffer in DXILResourceAccess #121725
Open
bogner
wants to merge
1
commit into
llvm:users/bogner/121282
Choose a base branch
from
bogner:2025-01-05-rawbuffer-resource-access
base: users/bogner/121282
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
[DirectX] Handle dx.RawBuffer in DXILResourceAccess #121725
bogner
wants to merge
1
commit into
llvm:users/bogner/121282
from
bogner:2025-01-05-rawbuffer-resource-access
+469
−77
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This adds handling for raw and structured buffers when lowering resource access via `llvm.dx.resource.getpointer`. Fixes llvm#121714
@llvm/pr-subscribers-backend-directx Author: Justin Bogner (bogner) ChangesThis adds handling for raw and structured buffers when lowering resource access via Fixes #121714 Patch is 25.43 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/121725.diff 3 Files Affected:
diff --git a/llvm/lib/Target/DirectX/DXILResourceAccess.cpp b/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
index 837624935c5fae..b0074b58e403c4 100644
--- a/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
+++ b/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
@@ -21,99 +21,207 @@
using namespace llvm;
-static void replaceTypedBufferAccess(IntrinsicInst *II,
- dxil::ResourceTypeInfo &RTI) {
- const DataLayout &DL = II->getDataLayout();
+static Value *calculateGEPOffset(GetElementPtrInst *GEP, Value *PrevOffset,
+ dxil::ResourceTypeInfo &RTI) {
+ assert(!PrevOffset && "Non-constant GEP chains not handled yet");
+
+ const DataLayout &DL = GEP->getDataLayout();
+
+ uint64_t ScalarSize = 1;
+ if (RTI.isTyped()) {
+ Type *ContainedType = RTI.getHandleTy()->getTypeParameter(0);
+ // We need the size of an element in bytes so that we can calculate the
+ // offset in elements given a total offset in bytes.
+ Type *ScalarType = ContainedType->getScalarType();
+ ScalarSize = DL.getTypeSizeInBits(ScalarType) / 8;
+ }
+
+ APInt ConstantOffset(DL.getIndexTypeSizeInBits(GEP->getType()), 0);
+ if (GEP->accumulateConstantOffset(DL, ConstantOffset)) {
+ APInt Scaled = ConstantOffset.udiv(ScalarSize);
+ return ConstantInt::get(Type::getInt32Ty(GEP->getContext()), Scaled);
+ }
+
+ auto IndexIt = GEP->idx_begin();
+ assert(cast<ConstantInt>(IndexIt)->getZExtValue() == 0 &&
+ "GEP is not indexing through pointer");
+ ++IndexIt;
+ Value *Offset = *IndexIt;
+ assert(++IndexIt == GEP->idx_end() && "Too many indices in GEP");
+ return Offset;
+}
+
+static void createTypedBufferStore(IntrinsicInst *II, StoreInst *SI,
+ Value *Offset, dxil::ResourceTypeInfo &RTI) {
+ IRBuilder<> Builder(SI);
+ Type *ContainedType = RTI.getHandleTy()->getTypeParameter(0);
+ Type *LoadType = StructType::get(ContainedType, Builder.getInt1Ty());
+
+ Value *V = SI->getValueOperand();
+ if (V->getType() == ContainedType) {
+ // V is already the right type.
+ assert(!Offset && "store of whole element has offset?");
+ } else if (V->getType() == ContainedType->getScalarType()) {
+ // We're storing a scalar, so we need to load the current value and only
+ // replace the relevant part.
+ auto *Load = Builder.CreateIntrinsic(
+ LoadType, Intrinsic::dx_resource_load_typedbuffer,
+ {II->getOperand(0), II->getOperand(1)});
+ auto *Struct = Builder.CreateExtractValue(Load, {0});
+
+ // If we have an offset from seeing a GEP earlier, use that. Otherwise, 0.
+ if (!Offset)
+ Offset = ConstantInt::get(Builder.getInt32Ty(), 0);
+ V = Builder.CreateInsertElement(Struct, V, Offset);
+ } else {
+ llvm_unreachable("Store to typed resource has invalid type");
+ }
+
+ auto *Inst = Builder.CreateIntrinsic(
+ Builder.getVoidTy(), Intrinsic::dx_resource_store_typedbuffer,
+ {II->getOperand(0), II->getOperand(1), V});
+ SI->replaceAllUsesWith(Inst);
+}
+
+static void createRawStore(IntrinsicInst *II, StoreInst *SI, Value *Offset) {
+ IRBuilder<> Builder(SI);
+
+ if (!Offset)
+ Offset = ConstantInt::get(Builder.getInt32Ty(), 0);
+ Value *V = SI->getValueOperand();
+ // TODO: break up larger types
+ auto *Inst = Builder.CreateIntrinsic(
+ Builder.getVoidTy(), Intrinsic::dx_resource_store_rawbuffer,
+ {II->getOperand(0), II->getOperand(1), Offset, V});
+ SI->replaceAllUsesWith(Inst);
+}
+
+static void createStoreIntrinsic(IntrinsicInst *II, StoreInst *SI,
+ Value *Offset, dxil::ResourceTypeInfo &RTI) {
+ switch (RTI.getResourceKind()) {
+ case dxil::ResourceKind::TypedBuffer:
+ return createTypedBufferStore(II, SI, Offset, RTI);
+ case dxil::ResourceKind::RawBuffer:
+ case dxil::ResourceKind::StructuredBuffer:
+ return createRawStore(II, SI, Offset);
+ case dxil::ResourceKind::Texture1D:
+ case dxil::ResourceKind::Texture2D:
+ case dxil::ResourceKind::Texture2DMS:
+ case dxil::ResourceKind::Texture3D:
+ case dxil::ResourceKind::TextureCube:
+ case dxil::ResourceKind::Texture1DArray:
+ case dxil::ResourceKind::Texture2DArray:
+ case dxil::ResourceKind::Texture2DMSArray:
+ case dxil::ResourceKind::TextureCubeArray:
+ case dxil::ResourceKind::FeedbackTexture2D:
+ case dxil::ResourceKind::FeedbackTexture2DArray:
+ // TODO: handle these
+ return;
+ case dxil::ResourceKind::CBuffer:
+ case dxil::ResourceKind::Sampler:
+ case dxil::ResourceKind::TBuffer:
+ case dxil::ResourceKind::RTAccelerationStructure:
+ case dxil::ResourceKind::Invalid:
+ case dxil::ResourceKind::NumEntries:
+ llvm_unreachable("Invalid resource kind for store");
+ }
+ llvm_unreachable("Unhandled case in switch");
+}
+
+static void createTypedBufferLoad(IntrinsicInst *II, LoadInst *LI,
+ Value *Offset, dxil::ResourceTypeInfo &RTI) {
+ IRBuilder<> Builder(LI);
+ Type *ContainedType = RTI.getHandleTy()->getTypeParameter(0);
+ Type *LoadType = StructType::get(ContainedType, Builder.getInt1Ty());
- auto *HandleType = cast<TargetExtType>(II->getOperand(0)->getType());
- assert(HandleType->getName() == "dx.TypedBuffer" &&
- "Unexpected typed buffer type");
- Type *ContainedType = HandleType->getTypeParameter(0);
+ Value *V =
+ Builder.CreateIntrinsic(LoadType, Intrinsic::dx_resource_load_typedbuffer,
+ {II->getOperand(0), II->getOperand(1)});
+ V = Builder.CreateExtractValue(V, {0});
- Type *LoadType =
- StructType::get(ContainedType, Type::getInt1Ty(II->getContext()));
+ if (Offset)
+ V = Builder.CreateExtractElement(V, Offset);
- // We need the size of an element in bytes so that we can calculate the offset
- // in elements given a total offset in bytes later.
- Type *ScalarType = ContainedType->getScalarType();
- uint64_t ScalarSize = DL.getTypeSizeInBits(ScalarType) / 8;
+ LI->replaceAllUsesWith(V);
+}
+static void createRawLoad(IntrinsicInst *II, LoadInst *LI, Value *Offset) {
+ IRBuilder<> Builder(LI);
+ // TODO: break up larger types
+ Type *LoadType = StructType::get(LI->getType(), Builder.getInt1Ty());
+ if (!Offset)
+ Offset = ConstantInt::get(Builder.getInt32Ty(), 0);
+ Value *V =
+ Builder.CreateIntrinsic(LoadType, Intrinsic::dx_resource_load_rawbuffer,
+ {II->getOperand(0), II->getOperand(1), Offset});
+ V = Builder.CreateExtractValue(V, {0});
+
+ LI->replaceAllUsesWith(V);
+}
+
+static void createLoadIntrinsic(IntrinsicInst *II, LoadInst *LI, Value *Offset,
+ dxil::ResourceTypeInfo &RTI) {
+ switch (RTI.getResourceKind()) {
+ case dxil::ResourceKind::TypedBuffer:
+ return createTypedBufferLoad(II, LI, Offset, RTI);
+ case dxil::ResourceKind::RawBuffer:
+ case dxil::ResourceKind::StructuredBuffer:
+ return createRawLoad(II, LI, Offset);
+ case dxil::ResourceKind::Texture1D:
+ case dxil::ResourceKind::Texture2D:
+ case dxil::ResourceKind::Texture2DMS:
+ case dxil::ResourceKind::Texture3D:
+ case dxil::ResourceKind::TextureCube:
+ case dxil::ResourceKind::Texture1DArray:
+ case dxil::ResourceKind::Texture2DArray:
+ case dxil::ResourceKind::Texture2DMSArray:
+ case dxil::ResourceKind::TextureCubeArray:
+ case dxil::ResourceKind::FeedbackTexture2D:
+ case dxil::ResourceKind::FeedbackTexture2DArray:
+ case dxil::ResourceKind::CBuffer:
+ case dxil::ResourceKind::TBuffer:
+ // TODO: handle these
+ return;
+ case dxil::ResourceKind::Sampler:
+ case dxil::ResourceKind::RTAccelerationStructure:
+ case dxil::ResourceKind::Invalid:
+ case dxil::ResourceKind::NumEntries:
+ llvm_unreachable("Invalid resource kind for load");
+ }
+ llvm_unreachable("Unhandled case in switch");
+}
+
+static void
+replaceAccess(IntrinsicInst *II, dxil::ResourceTypeInfo &RTI) {
// Process users keeping track of indexing accumulated from GEPs.
- struct AccessAndIndex {
+ struct AccessAndOffset {
User *Access;
- Value *Index;
+ Value *Offset;
};
- SmallVector<AccessAndIndex> Worklist;
+ SmallVector<AccessAndOffset> Worklist;
for (User *U : II->users())
Worklist.push_back({U, nullptr});
SmallVector<Instruction *> DeadInsts;
while (!Worklist.empty()) {
- AccessAndIndex Current = Worklist.back();
+ AccessAndOffset Current = Worklist.back();
Worklist.pop_back();
if (auto *GEP = dyn_cast<GetElementPtrInst>(Current.Access)) {
IRBuilder<> Builder(GEP);
- Value *Index;
- APInt ConstantOffset(DL.getIndexTypeSizeInBits(GEP->getType()), 0);
- if (GEP->accumulateConstantOffset(DL, ConstantOffset)) {
- APInt Scaled = ConstantOffset.udiv(ScalarSize);
- Index = ConstantInt::get(Builder.getInt32Ty(), Scaled);
- } else {
- auto IndexIt = GEP->idx_begin();
- assert(cast<ConstantInt>(IndexIt)->getZExtValue() == 0 &&
- "GEP is not indexing through pointer");
- ++IndexIt;
- Index = *IndexIt;
- assert(++IndexIt == GEP->idx_end() && "Too many indices in GEP");
- }
-
+ Value *Offset = calculateGEPOffset(GEP, Current.Offset, RTI);
for (User *U : GEP->users())
- Worklist.push_back({U, Index});
+ Worklist.push_back({U, Offset});
DeadInsts.push_back(GEP);
} else if (auto *SI = dyn_cast<StoreInst>(Current.Access)) {
assert(SI->getValueOperand() != II && "Pointer escaped!");
- IRBuilder<> Builder(SI);
-
- Value *V = SI->getValueOperand();
- if (V->getType() == ContainedType) {
- // V is already the right type.
- } else if (V->getType() == ScalarType) {
- // We're storing a scalar, so we need to load the current value and only
- // replace the relevant part.
- auto *Load = Builder.CreateIntrinsic(
- LoadType, Intrinsic::dx_resource_load_typedbuffer,
- {II->getOperand(0), II->getOperand(1)});
- auto *Struct = Builder.CreateExtractValue(Load, {0});
-
- // If we have an offset from seeing a GEP earlier, use it.
- Value *IndexOp = Current.Index
- ? Current.Index
- : ConstantInt::get(Builder.getInt32Ty(), 0);
- V = Builder.CreateInsertElement(Struct, V, IndexOp);
- } else {
- llvm_unreachable("Store to typed resource has invalid type");
- }
-
- auto *Inst = Builder.CreateIntrinsic(
- Builder.getVoidTy(), Intrinsic::dx_resource_store_typedbuffer,
- {II->getOperand(0), II->getOperand(1), V});
- SI->replaceAllUsesWith(Inst);
+ createStoreIntrinsic(II, SI, Current.Offset, RTI);
DeadInsts.push_back(SI);
} else if (auto *LI = dyn_cast<LoadInst>(Current.Access)) {
- IRBuilder<> Builder(LI);
- Value *V = Builder.CreateIntrinsic(
- LoadType, Intrinsic::dx_resource_load_typedbuffer,
- {II->getOperand(0), II->getOperand(1)});
- V = Builder.CreateExtractValue(V, {0});
-
- if (Current.Index)
- V = Builder.CreateExtractElement(V, Current.Index);
-
- LI->replaceAllUsesWith(V);
+ createLoadIntrinsic(II, LI, Current.Offset, RTI);
DeadInsts.push_back(LI);
} else
@@ -137,15 +245,8 @@ static bool transformResourcePointers(Function &F, DXILResourceTypeMap &DRTM) {
Resources.emplace_back(II, DRTM[HandleTy]);
}
- for (auto &[II, RI] : Resources) {
- if (RI.isTyped()) {
- Changed = true;
- replaceTypedBufferAccess(II, RI);
- }
-
- // TODO: handle other resource types. We should probably have an
- // `unreachable` here once we've added support for all of them.
- }
+ for (auto &[II, RI] : Resources)
+ replaceAccess(II, RI);
return Changed;
}
diff --git a/llvm/test/CodeGen/DirectX/ResourceAccess/load_rawbuffer.ll b/llvm/test/CodeGen/DirectX/ResourceAccess/load_rawbuffer.ll
new file mode 100644
index 00000000000000..5d2e4041c3ea02
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/ResourceAccess/load_rawbuffer.ll
@@ -0,0 +1,167 @@
+; RUN: opt -S -dxil-resource-access %s | FileCheck %s
+
+target triple = "dxil-pc-shadermodel6.6-compute"
+
+declare void @f32_user(float)
+declare void @v4f32_user(<4 x float>)
+declare void @i32_user(i32)
+declare void @v4i32_user(<4 x i32>)
+declare void @v3f16_user(<3 x half>)
+declare void @v4f64_user(<4 x double>)
+
+; CHECK-LABEL: define void @loadf32_struct
+define void @loadf32_struct(i32 %index) {
+ %buffer = call target("dx.RawBuffer", float, 0, 0, 0)
+ @llvm.dx.resource.handlefrombinding.tdx.RawBuffer_f32_0_0_0(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; CHECK-NOT: @llvm.dx.resource.getpointer
+ %ptr = call ptr @llvm.dx.resource.getpointer(
+ target("dx.RawBuffer", float, 0, 0, 0) %buffer, i32 %index)
+
+ ; CHECK: %[[LOAD:.*]] = call { float, i1 } @llvm.dx.resource.load.rawbuffer.f32.tdx.RawBuffer_f32_0_0_0t(target("dx.RawBuffer", float, 0, 0, 0) %buffer, i32 %index, i32 0)
+ ; CHECK: %[[VAL:.*]] = extractvalue { float, i1 } %[[LOAD]], 0
+ ; CHECK: call void @f32_user(float %[[VAL]])
+ %data = load float, ptr %ptr
+ call void @f32_user(float %data)
+
+ ret void
+}
+
+; CHECK-LABEL: define void @loadf32_byte
+define void @loadf32_byte(i32 %offset) {
+ %buffer = call target("dx.RawBuffer", i8, 0, 0, 0)
+ @llvm.dx.resource.handlefrombinding.tdx.RawBuffer_i8_0_0_0(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; CHECK-NOT: @llvm.dx.resource.getpointer
+ %ptr = call ptr @llvm.dx.resource.getpointer(
+ target("dx.RawBuffer", i8, 0, 0, 0) %buffer, i32 %offset)
+
+ ; CHECK: %[[LOAD:.*]] = call { float, i1 } @llvm.dx.resource.load.rawbuffer.f32.tdx.RawBuffer_i8_0_0_0t(target("dx.RawBuffer", i8, 0, 0, 0) %buffer, i32 %offset, i32 0)
+ ; CHECK: %[[VAL:.*]] = extractvalue { float, i1 } %[[LOAD]], 0
+ ; CHECK: call void @f32_user(float %[[VAL]])
+ %data = load float, ptr %ptr
+ call void @f32_user(float %data)
+
+ ret void
+}
+
+; CHECK-LABEL: define void @loadv4f32_struct
+define void @loadv4f32_struct(i32 %index) {
+ %buffer = call target("dx.RawBuffer", <4 x float>, 0, 0, 0)
+ @llvm.dx.resource.handlefrombinding.tdx.RawBuffer_v4f32_0_0_0(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; CHECK-NOT: @llvm.dx.resource.getpointer
+ %ptr = call ptr @llvm.dx.resource.getpointer(
+ target("dx.RawBuffer", <4 x float>, 0, 0, 0) %buffer, i32 %index)
+
+ ; CHECK: %[[LOAD:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.rawbuffer.v4f32.tdx.RawBuffer_v4f32_0_0_0t(target("dx.RawBuffer", <4 x float>, 0, 0, 0) %buffer, i32 %index, i32 0)
+ ; CHECK: %[[VAL:.*]] = extractvalue { <4 x float>, i1 } %[[LOAD]], 0
+ ; CHECK: call void @v4f32_user(<4 x float> %[[VAL]])
+ %data = load <4 x float>, ptr %ptr
+ call void @v4f32_user(<4 x float> %data)
+
+ ret void
+}
+
+; CHECK-LABEL: define void @loadv4f32_byte
+define void @loadv4f32_byte(i32 %offset) {
+ %buffer = call target("dx.RawBuffer", i8, 0, 0, 0)
+ @llvm.dx.resource.handlefrombinding.tdx.RawBuffer_i8_0_0_0(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; CHECK-NOT: @llvm.dx.resource.getpointer
+ %ptr = call ptr @llvm.dx.resource.getpointer(
+ target("dx.RawBuffer", i8, 0, 0, 0) %buffer, i32 %offset)
+
+ ; CHECK: %[[LOAD:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.rawbuffer.v4f32.tdx.RawBuffer_i8_0_0_0t(target("dx.RawBuffer", i8, 0, 0, 0) %buffer, i32 %offset, i32 0)
+ ; CHECK: %[[VAL:.*]] = extractvalue { <4 x float>, i1 } %[[LOAD]], 0
+ ; CHECK: call void @v4f32_user(<4 x float> %[[VAL]]
+ %data = load <4 x float>, ptr %ptr
+ call void @v4f32_user(<4 x float> %data)
+
+ ret void
+}
+
+; CHECK-LABEL: define void @loadelements
+define void @loadelements(i32 %index) {
+ %buffer = call target("dx.RawBuffer", {<4 x float>, <4 x i32>}, 0, 0, 0)
+ @llvm.dx.resource.handlefrombinding.tdx.RawBuffer_sl_v4f32v4i32s_0_0_0(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; CHECK-NOT: @llvm.dx.resource.getpointer
+ %ptr = call ptr @llvm.dx.resource.getpointer(
+ target("dx.RawBuffer", {<4 x float>, <4 x i32>}, 0, 0, 0) %buffer,
+ i32 %index)
+
+ ; CHECK: %[[LOADF32:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.rawbuffer.v4f32.tdx.RawBuffer_sl_v4f32v4i32s_0_0_0t(target("dx.RawBuffer", { <4 x float>, <4 x i32> }, 0, 0, 0) %buffer, i32 %index, i32 0)
+ ; CHECK: %[[VALF32:.*]] = extractvalue { <4 x float>, i1 } %[[LOADF32]], 0
+ ; CHECK: call void @v4f32_user(<4 x float> %[[VALF32]]
+ %dataf32 = load <4 x float>, ptr %ptr
+ call void @v4f32_user(<4 x float> %dataf32)
+
+ ; CHECK: %[[LOADI32:.*]] = call { <4 x i32>, i1 } @llvm.dx.resource.load.rawbuffer.v4i32.tdx.RawBuffer_sl_v4f32v4i32s_0_0_0t(target("dx.RawBuffer", { <4 x float>, <4 x i32> }, 0, 0, 0) %buffer, i32 %index, i32 16)
+ ; CHECK: %[[VALI32:.*]] = extractvalue { <4 x i32>, i1 } %[[LOADI32]], 0
+ ; CHECK: call void @v4i32_user(<4 x i32> %[[VALI32]]
+ %addri32 = getelementptr inbounds nuw i8, ptr %ptr, i32 16
+ %datai32 = load <4 x i32>, ptr %addri32
+ call void @v4i32_user(<4 x i32> %datai32)
+
+ ret void
+}
+
+; CHECK-LABEL: define void @loadnested
+define void @loadnested(i32 %index) {
+ %buffer = call
+ target("dx.RawBuffer", {i32, {<4 x float>, <3 x half>}}, 0, 0, 0)
+ @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; CHECK-NOT: @llvm.dx.resource.getpointer
+ %ptr = call ptr @llvm.dx.resource.getpointer(
+ target("dx.RawBuffer", {i32, {<4 x float>, <3 x half>}}, 0, 0, 0) %buffer,
+ i32 %index)
+
+ ; CHECK: %[[LOADI32:.*]] = call { i32, i1 } @llvm.dx.resource.load.rawbuffer.i32.tdx.RawBuffer_sl_i32sl_v4f32v3f16ss_0_0_0t(target("dx.RawBuffer", { i32, { <4 x float>, <3 x half> } }, 0, 0, 0) %buffer, i32 %index, i32 0)
+ ; CHECK: %[[VALI32:.*]] = extractvalue { i32, i1 } %[[LOADI32]], 0
+ ; CHECK: call void @i32_user(i32 %[[VALI32]])
+ %datai32 = load i32, ptr %ptr
+ call void @i32_user(i32 %datai32)
+
+ ; CHECK: %[[LOADF32:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.rawbuffer.v4f32.tdx.RawBuffer_sl_i32sl_v4f32v3f16ss_0_0_0t(target("dx.RawBuffer", { i32, { <4 x float>, <3 x half> } }, 0, 0, 0) %buffer, i32 %index, i32 4)
+ ; CHECK: %[[VALF32:.*]] = extractvalue { <4 x float>, i1 } %[[LOADF32]], 0
+ ; CHECK: call void @v4f32_user(<4 x float> %[[VALF32]])
+ %addrf32 = getelementptr inbounds nuw i8, ptr %ptr, i32 4
+ %dataf32 = load <4 x float>, ptr %addrf32
+ call void @v4f32_user(<4 x float> %dataf32)
+
+ ; CHECK: %[[LOADF16:.*]] = call { <3 x half>, i1 } @llvm.dx.resource.load.rawbuffer.v3f16.tdx.RawBuffer_sl_i32sl_v4f32v3f16ss_0_0_0t(target("dx.RawBuffer", { i32, { <4 x float>, <3 x half> } }, 0, 0, 0) %buffer, i32 %index, i32 20)
+ ; CHECK: %[[VALF16:.*]] = extractvalue { <3 x half>, i1 } %[[LOADF16]], 0
+ ; CHECK: call void @v3f16_user(<3 x half> %[[VALF16]])
+ %addrf16 = getelementptr inbounds nuw i8, ptr %ptr, i32 20
+ %dataf16 = load <3 x half>, ptr %addrf16
+ call void @v3f16_user(<3 x half> %dataf16)
+
+ ret void
+}
+
+; byteaddressbuf.Load<int64_t4>
+; CHECK-LABEL: define void @loadv4f64_byte
+define void @loadv4f64_byte(i32 %offset) {
+ %buffer = call target("dx.RawBuffer", i8, 0, 0, 0)
+ @llvm.dx.resource.handlefrombinding.tdx.RawBuffer_i8_0_0_0(
+ i32 0, i32 0, i32 1, i32 0, i1 false)
+
+ ; CHECK-NOT: @llvm.dx.resource.getpointer
+ %ptr = call ptr @llvm.dx.resource.getpointer(
+ target("dx.RawBuffer", i8, 0, 0, 0) %buffer, i32 %offset)
+
+ ; CHECK: %[[LOAD:.*]] = call { <4 x double>, i1 } @llvm.dx.resource.load.rawbuffer.v4f64.tdx.RawBuffer_i8_0_0_0t(target("dx.RawBuffer", i8, 0, 0, 0) %buffer, i32 %offset, i32 0)
+ ; CHECK: %[[VAL:.*]] = extractvalue { <4 x double>, i1 } %[[LOAD]], 0
+ ; CHECK: call void @v4f64_user(<4 x double> %[[VAL]])
+ %data = load <4 x double>, ptr %ptr
+ call void @v4f64_user(<4 x double> %data)
+
+ ret void
+}
diff --git a/llvm/test/CodeGen/DirectX/ResourceAccess/store_rawbuffer.ll b/llvm/test/CodeGen/DirectX/ResourceAccess/store_rawbuffer.ll
new file mode 100644
index 00000000000000..b19f9d04a2dffa
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/ResourceAccess/store_rawbuffer.ll
@@ -0,0 +1,124 @@
+; RUN: opt -S -dxil-resource-access %s | FileCheck %s
+
+target triple = "dxil-pc-shadermodel6.6-compute"
+
+; CHECK-LABEL: define void @storef32_struct
+define void @storef32_struct(i32 %index, float %data) {
+ %buffer = call target("dx....
[truncated]
|
You can test this locally with the following command:git-clang-format --diff ae36a7c2aa0626febb00b7bcf581a00ea0042d23 192d20b16e418b89a549e93406f782ac0a0fa093 --extensions cpp -- llvm/lib/Target/DirectX/DXILResourceAccess.cpp View the diff from clang-format here.diff --git a/llvm/lib/Target/DirectX/DXILResourceAccess.cpp b/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
index b0074b58e4..5164fd3813 100644
--- a/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
+++ b/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
@@ -22,7 +22,7 @@
using namespace llvm;
static Value *calculateGEPOffset(GetElementPtrInst *GEP, Value *PrevOffset,
- dxil::ResourceTypeInfo &RTI) {
+ dxil::ResourceTypeInfo &RTI) {
assert(!PrevOffset && "Non-constant GEP chains not handled yet");
const DataLayout &DL = GEP->getDataLayout();
@@ -191,8 +191,7 @@ static void createLoadIntrinsic(IntrinsicInst *II, LoadInst *LI, Value *Offset,
llvm_unreachable("Unhandled case in switch");
}
-static void
-replaceAccess(IntrinsicInst *II, dxil::ResourceTypeInfo &RTI) {
+static void replaceAccess(IntrinsicInst *II, dxil::ResourceTypeInfo &RTI) {
// Process users keeping track of indexing accumulated from GEPs.
struct AccessAndOffset {
User *Access;
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This adds handling for raw and structured buffers when lowering resource access via
llvm.dx.resource.getpointer
.Fixes #121714