[DirectX] Add 32- and 64-bit 3-element vectors to DataLayout #160955
+13
−7
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
This explicitly adds two 3-element vectors to the DataLayout so that they'll be element-aligned. We need to do this more generally for vectors, but this unblocks some very common cases. Workaround for llvm#123968
llvmbot
added
clang
|
@llvm/pr-subscribers-hlsl @llvm/pr-subscribers-backend-directx Author: Justin Bogner (bogner) ChangesThis explicitly adds two 3-element vectors to the DataLayout so that they'll be element-aligned. We need to do this more generally for vectors, but this unblocks some very common cases. Workaround for #123968 Full diff: https://github.com/llvm/llvm-project/pull/160955.diff 4 Files Affected:
diff --git a/clang/lib/Basic/Targets/DirectX.h b/clang/lib/Basic/Targets/DirectX.h
index bd13c9ee0fd05..a21a593365773 100644
--- a/clang/lib/Basic/Targets/DirectX.h
+++ b/clang/lib/Basic/Targets/DirectX.h
@@ -64,8 +64,11 @@ class LLVM_LIBRARY_VISIBILITY DirectXTargetInfo : public TargetInfo {
NoAsmVariants = true;
PlatformMinVersion = Triple.getOSVersion();
PlatformName = llvm::Triple::getOSTypeName(Triple.getOS());
+ // TODO: We need to align vectors on the element size generally, but for now
+ // we hard code this for 3-element 32- and 64-bit vectors as a workaround.
+ // See https://github.com/llvm/llvm-project/issues/123968
resetDataLayout("e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:"
- "32-f64:64-n8:16:32:64");
+ "32-f64:64-n8:16:32:64-v48:16:16-v96:32:32-v192:64:64");
TheCXXABI.set(TargetCXXABI::GenericItanium);
}
bool useFP16ConversionIntrinsics() const override { return false; }
diff --git a/clang/test/CodeGenHLSL/basic-target.c b/clang/test/CodeGenHLSL/basic-target.c
index c700e06bd5850..b9482df5a0987 100644
--- a/clang/test/CodeGenHLSL/basic-target.c
+++ b/clang/test/CodeGenHLSL/basic-target.c
@@ -6,5 +6,5 @@
// RUN: %clang -cc1 -triple dxil-pc-shadermodel6.0-domain -emit-llvm -o - %s | FileCheck %s
// RUN: %clang -cc1 -triple dxil-pc-shadermodel6.0-geometry -emit-llvm -o - %s | FileCheck %s
-// CHECK: target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
+// CHECK: target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64-v48:16:16-v96:32:32-v192:64:64"
// CHECK: target triple = "dxilv1.0-pc-shadermodel6.0-{{[a-z]+}}"
diff --git a/llvm/lib/TargetParser/TargetDataLayout.cpp b/llvm/lib/TargetParser/TargetDataLayout.cpp
index cea246e9527bd..5e6f77316780e 100644
--- a/llvm/lib/TargetParser/TargetDataLayout.cpp
+++ b/llvm/lib/TargetParser/TargetDataLayout.cpp
@@ -548,8 +548,11 @@ std::string Triple::computeDataLayout(StringRef ABIName) const {
case Triple::csky:
return computeCSKYDataLayout(*this);
case Triple::dxil:
+ // TODO: We need to align vectors on the element size generally, but for now
+ // we hard code this for 3-element 32- and 64-bit vectors as a workaround.
+ // See https://github.com/llvm/llvm-project/issues/123968
return "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-"
- "f32:32-f64:64-n8:16:32:64";
+ "f32:32-f64:64-n8:16:32:64-v48:16:16-v96:32:32-v192:64:64";
case Triple::hexagon:
return "e-m:e-p:32:32:32-a:0-n16:32-"
"i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-"
diff --git a/llvm/test/CodeGen/DirectX/CBufferAccess/memcpy.ll b/llvm/test/CodeGen/DirectX/CBufferAccess/memcpy.ll
index a78fdd5037f93..f1486f974fb36 100644
--- a/llvm/test/CodeGen/DirectX/CBufferAccess/memcpy.ll
+++ b/llvm/test/CodeGen/DirectX/CBufferAccess/memcpy.ll
@@ -74,7 +74,7 @@ entry:
; CHECK: [[UPTO1:%.*]] = insertelement <3 x double> [[UPTO0]], double [[Y]], i32 1
; CHECK: [[UPTO2:%.*]] = insertelement <3 x double> [[UPTO1]], double [[Z]], i32 2
; CHECK: [[DEST:%.*]] = getelementptr inbounds i8, ptr [[A2_COPY:%.*]], i32 0
-; CHECK: store <3 x double> [[UPTO2]], ptr [[DEST]], align 32
+; CHECK: store <3 x double> [[UPTO2]], ptr [[DEST]], align 8
; CHECK: [[LOAD:%.*]] = call { double, double } @llvm.dx.resource.load.cbufferrow.2.{{.*}}(target("dx.CBuffer", {{.*}})) [[CB]], i32 5)
; CHECK: [[X:%.*]] = extractvalue { double, double } [[LOAD]], 0
; CHECK: [[Y:%.*]] = extractvalue { double, double } [[LOAD]], 1
@@ -83,9 +83,9 @@ entry:
; CHECK: [[UPTO0:%.*]] = insertelement <3 x double> poison, double [[X]], i32 0
; CHECK: [[UPTO1:%.*]] = insertelement <3 x double> [[UPTO0]], double [[Y]], i32 1
; CHECK: [[UPTO2:%.*]] = insertelement <3 x double> [[UPTO1]], double [[Z]], i32 2
-; CHECK: [[DEST:%.*]] = getelementptr inbounds i8, ptr [[A2_COPY]], i32 32
-; CHECK: store <3 x double> [[UPTO2]], ptr [[DEST]], align 32
- call void @llvm.memcpy.p0.p2.i32(ptr align 32 %a2.copy, ptr addrspace(2) align 32 @a2, i32 64, i1 false)
+; CHECK: [[DEST:%.*]] = getelementptr inbounds i8, ptr [[A2_COPY]], i32 24
+; CHECK: store <3 x double> [[UPTO2]], ptr [[DEST]], align 8
+ call void @llvm.memcpy.p0.p2.i32(ptr align 32 %a2.copy, ptr addrspace(2) align 32 @a2, i32 48, i1 false)
; CHECK: [[CB:%.*]] = load target("dx.CBuffer", {{.*}})), ptr @CB.cb, align 4
; CHECK: [[LOAD:%.*]] = call { half, half, half, half, half, half, half, half } @llvm.dx.resource.load.cbufferrow.8.{{.*}}(target("dx.CBuffer", {{.*}})) [[CB]], i32 7)
|
llvm-beanz
reviewed
Oct 7, 2025
| // See https://github.com/llvm/llvm-project/issues/123968 | ||
| resetDataLayout("e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:" | ||
| "32-f64:64-n8:16:32:64"); | ||
| "32-f64:64-n8:16:32:64-v48:16:16-v96:32:32-v192:64:64"); |
There was a problem hiding this comment.
My one concern is that we likely need to change this back to the original value before running into the DXIL bitcode writer.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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 explicitly adds two 3-element vectors to the DataLayout so that they'll be element-aligned. We need to do this more generally for vectors, but this unblocks some very common cases.
Workaround for #123968