[llvm] 6917799 - [DirectX backend] change MinVectorRegisterBitWidth to 32.
Xiang Li via llvm-commits
llvm-commits at lists.llvm.org
Tue Aug 30 23:20:27 PDT 2022
Author: Xiang Li
Date: 2022-08-30T23:20:12-07:00
New Revision: 6917799e3725f50c8d76dfbbb68123904b4b00d2
URL: https://github.com/llvm/llvm-project/commit/6917799e3725f50c8d76dfbbb68123904b4b00d2
DIFF: https://github.com/llvm/llvm-project/commit/6917799e3725f50c8d76dfbbb68123904b4b00d2.diff
LOG: [DirectX backend] change MinVectorRegisterBitWidth to 32.
This is to avoid vector-combine generate vector4 on float.
Reviewed By: beanz
Differential Revision: https://reviews.llvm.org/D132826
Added:
llvm/test/CodeGen/DirectX/min_vec_size.ll
Modified:
llvm/lib/Target/DirectX/DirectXTargetTransformInfo.h
Removed:
################################################################################
diff --git a/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.h b/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.h
index 90beb386fa44d..a0c09684dc67d 100644
--- a/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.h
+++ b/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.h
@@ -33,6 +33,7 @@ class DirectXTTIImpl : public BasicTTIImplBase<DirectXTTIImpl> {
explicit DirectXTTIImpl(const DirectXTargetMachine *TM, const Function &F)
: BaseT(TM, F.getParent()->getDataLayout()), ST(TM->getSubtargetImpl(F)),
TLI(ST->getTargetLowering()) {}
+ unsigned getMinVectorRegisterBitWidth() const { return 32; }
};
} // namespace llvm
diff --git a/llvm/test/CodeGen/DirectX/min_vec_size.ll b/llvm/test/CodeGen/DirectX/min_vec_size.ll
new file mode 100644
index 0000000000000..6ae61fd518d56
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/min_vec_size.ll
@@ -0,0 +1,21 @@
+; RUN: opt -S -vector-combine < %s | FileCheck %s
+
+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"
+target triple = "dxil-unknown-shadermodel6.7-library"
+
+; Make sure vec combine min vec size is 1 instead of 4 for float.
+; CHECK:@foo()
+; CHECK-NEXT:%[[LD:[0-9]+]] = load <1 x float>, ptr @a, align 8
+; CHECK-NEXT:%insert = shufflevector <1 x float> %[[LD]], <1 x float> poison, <2 x i32> <i32 0, i32 undef>
+; CHECK-NEXT:%shuffle = shufflevector <2 x float> %insert, <2 x float> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:ret <2 x float> %shuffle
+
+ at a = external local_unnamed_addr constant float
+
+; Function Attrs: mustprogress nofree norecurse nosync nounwind readnone willreturn
+define noundef <2 x float> @foo() local_unnamed_addr {
+ %1 = load float, ptr @a, align 8
+ %insert = insertelement <2 x float> poison, float %1, i64 0
+ %shuffle = shufflevector <2 x float> %insert, <2 x float> poison, <2 x i32> zeroinitializer
+ ret <2 x float> %shuffle
+}
More information about the llvm-commits
mailing list