caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[DirectX] make firstbitlow intrinsic use first argument instead of return for overload type (#145350)

Browse Source 
fixes #144966
Easy fix just add `dx_firstbitlow` to
`DirectXTTIImpl::isTargetIntrinsicWithOverloadTypeAtArg`
This commit is contained in:
Farzon Lotfi
2025-06-23 14:04:06 -04:00
committed by GitHub
parent ff865b639a
commit 0f173a0f9a
2 changed files with 18 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp
View File

@@ -30,6 +30,7 @@ bool DirectXTTIImpl::isTargetIntrinsicWithOverloadTypeAtArg(Intrinsic::ID ID,
switch (ID) {
case Intrinsic::dx_asdouble:
case Intrinsic::dx_isinf:
case Intrinsic::dx_firstbitlow:
return OpdIdx == 0;
default:
return OpdIdx == -1;

21
llvm/test/CodeGen/DirectX/firstbitlow.ll
View File

@@ -41,7 +41,20 @@ entry:
ret <4 x i32> %2
}
declare i32 @llvm.dx.firstbitlow.i16(i16)
declare i32 @llvm.dx.firstbitlow.i32(i32)
declare i32 @llvm.dx.firstbitlow.i64(i64)
declare <4 x i32> @llvm.dx.firstbitlow.v4i32(<4 x i32>)
define noundef <4 x i32> @test_firstbitlow_vec4_i16(<4 x i16> noundef %a) {
entry:
; CHECK: [[ee0:%.*]] = extractelement <4 x i16> %a, i64 0
; CHECK: [[ie0:%.*]] = call i32 @dx.op.unaryBits.i16(i32 32, i16 [[ee0]])
; CHECK: [[ee1:%.*]] = extractelement <4 x i16> %a, i64 1
; CHECK: [[ie1:%.*]] = call i32 @dx.op.unaryBits.i16(i32 32, i16 [[ee1]])
; CHECK: [[ee2:%.*]] = extractelement <4 x i16> %a, i64 2
; CHECK: [[ie2:%.*]] = call i32 @dx.op.unaryBits.i16(i32 32, i16 [[ee2]])
; CHECK: [[ee3:%.*]] = extractelement <4 x i16> %a, i64 3
; CHECK: [[ie3:%.*]] = call i32 @dx.op.unaryBits.i16(i32 32, i16 [[ee3]])
; CHECK: insertelement <4 x i32> poison, i32 [[ie0]], i64 0
; CHECK: insertelement <4 x i32> %{{.*}}, i32 [[ie1]], i64 1
; CHECK: insertelement <4 x i32> %{{.*}}, i32 [[ie2]], i64 2
; CHECK: insertelement <4 x i32> %{{.*}}, i32 [[ie3]], i64 3
%2 = call <4 x i32> @llvm.dx.firstbitlow.v4i16(<4 x i16> %a)
ret <4 x i32> %2
}