[DirectX] Handle <1 x ...> loads in DXILResourceAccess #137076
Merged
Conversation
This file contains hidden or 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
We can end up with loads of single element vectors when we have scalar values, because the vectorizer may introduce these to use ops like shufflevector in some cases. Make sure we're maintaining the correct type when translating these into resource load operations. Fixes llvm#136409.
|
@llvm/pr-subscribers-backend-directx Author: Justin Bogner (bogner) ChangesWe can end up with loads of single element vectors when we have scalar values, because the vectorizer may introduce these to use ops like shufflevector in some cases. Make sure we're maintaining the correct type when translating these into resource load operations. Fixes #136409. Full diff: https://github.com/llvm/llvm-project/pull/137076.diff 2 Files Affected:
diff --git a/llvm/lib/Target/DirectX/DXILResourceAccess.cpp b/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
index 3b8f7140d3122..16337f1237e00 100644
--- a/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
+++ b/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
@@ -143,6 +143,13 @@ static void createTypedBufferLoad(IntrinsicInst *II, LoadInst *LI,
if (Offset)
V = Builder.CreateExtractElement(V, Offset);
+ // If we loaded a <1 x ...> instead of a scalar (presumably to feed a
+ // shufflevector), then make sure we're maintaining the resulting type.
+ if (auto *VT = dyn_cast<FixedVectorType>(LI->getType()))
+ if (VT->getNumElements() == 1 && !isa<FixedVectorType>(V->getType()))
+ V = Builder.CreateInsertElement(PoisonValue::get(VT), V,
+ Builder.getInt32(0));
+
LI->replaceAllUsesWith(V);
}
diff --git a/llvm/test/CodeGen/DirectX/ResourceAccess/load_typedbuffer.ll b/llvm/test/CodeGen/DirectX/ResourceAccess/load_typedbuffer.ll
index 8769e6ec66d8e..7d8099ef0d871 100644
--- a/llvm/test/CodeGen/DirectX/ResourceAccess/load_typedbuffer.ll
+++ b/llvm/test/CodeGen/DirectX/ResourceAccess/load_typedbuffer.ll
@@ -4,8 +4,9 @@ target triple = "dxil-pc-shadermodel6.6-compute"
declare void @use_float4(<4 x float>)
declare void @use_float(float)
+declare void @use_float1(<1 x float>)
-; CHECK-LABEL: define void @load_float4
+; CHECK-LABEL: define void @load_float4(
define void @load_float4(i32 %index, i32 %elemindex) {
%buffer = call target("dx.TypedBuffer", <4 x float>, 1, 0, 0)
@llvm.dx.resource.handlefrombinding.tdx.TypedBuffer_v4f32_1_0_0(
@@ -35,3 +36,48 @@ define void @load_float4(i32 %index, i32 %elemindex) {
ret void
}
+
+; CHECK-LABEL: define void @load_float(
+define void @load_float(i32 %index, i32 %elemindex) {
+ %buffer = call target("dx.TypedBuffer", float, 1, 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.TypedBuffer", float, 1, 0, 0) %buffer, i32 %index)
+
+ ; CHECK: %[[LOAD:.*]] = call { float, i1 } @llvm.dx.resource.load.typedbuffer.f32.tdx.TypedBuffer_f32_1_0_0t(target("dx.TypedBuffer", float, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VALUE:.*]] = extractvalue { float, i1 } %[[LOAD]], 0
+ ; CHECK: call void @use_float(float %[[VALUE]])
+ %data = load float, ptr %ptr
+ call void @use_float(float %data)
+
+ ; CHECK: %[[LOAD:.*]] = call { float, i1 } @llvm.dx.resource.load.typedbuffer.f32.tdx.TypedBuffer_f32_1_0_0t(target("dx.TypedBuffer", float, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VALUE:.*]] = extractvalue { float, i1 } %[[LOAD]], 0
+ ; CHECK: %[[VEC:.*]] = insertelement <1 x float> poison, float %[[VALUE]], i32 0
+ ; CHECK: call void @use_float1(<1 x float> %[[VEC]])
+ %vec_data = load <1 x float>, ptr %ptr
+ call void @use_float1(<1 x float> %vec_data)
+
+ ret void
+}
+
+; CHECK-LABEL: define void @load_float1(
+define void @load_float1(i32 %index) {
+ %buffer = call target("dx.TypedBuffer", <1 x float>, 1, 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.TypedBuffer", <1 x float>, 1, 0, 0) %buffer, i32 %index)
+
+ ; CHECK: %[[LOAD:.*]] = call { <1 x float>, i1 } @llvm.dx.resource.load.typedbuffer.v1f32.tdx.TypedBuffer_v1f32_1_0_0t(target("dx.TypedBuffer", <1 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VALUE:.*]] = extractvalue { <1 x float>, i1 } %[[LOAD]], 0
+ ; CHECK: %[[SHUF:.*]] = shufflevector <1 x float> %[[VALUE]], <1 x float> poison, <4 x i32> zeroinitializer
+ ; CHECK: call void @use_float4(<4 x float> %[[SHUF]])
+ %load = load <1 x float>, ptr %ptr
+ %shuf = shufflevector <1 x float> %load, <1 x float> poison, <4 x i32> zeroinitializer
+ call void @use_float4(<4 x float> %shuf)
+
+ ret void
+}
|
Icohedron
approved these changes
Apr 24, 2025
IanWood1
pushed a commit
to IanWood1/llvm-project
that referenced
this pull request
May 6, 2025
We can end up with loads of single element vectors when we have scalar values, because the vectorizer may introduce these to use ops like shufflevector in some cases. Make sure we're maintaining the correct type when translating these into resource load operations. Fixes llvm#136409.
IanWood1
pushed a commit
to IanWood1/llvm-project
that referenced
this pull request
May 6, 2025
We can end up with loads of single element vectors when we have scalar values, because the vectorizer may introduce these to use ops like shufflevector in some cases. Make sure we're maintaining the correct type when translating these into resource load operations. Fixes llvm#136409.
IanWood1
pushed a commit
to IanWood1/llvm-project
that referenced
this pull request
May 6, 2025
We can end up with loads of single element vectors when we have scalar values, because the vectorizer may introduce these to use ops like shufflevector in some cases. Make sure we're maintaining the correct type when translating these into resource load operations. Fixes llvm#136409.
Ankur-0429
pushed a commit
to Ankur-0429/llvm-project
that referenced
this pull request
May 9, 2025
We can end up with loads of single element vectors when we have scalar values, because the vectorizer may introduce these to use ops like shufflevector in some cases. Make sure we're maintaining the correct type when translating these into resource load operations. Fixes llvm#136409.
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.
We can end up with loads of single element vectors when we have scalar values, because the vectorizer may introduce these to use ops like shufflevector in some cases. Make sure we're maintaining the correct type when translating these into resource load operations.
Fixes #136409.