[DirectX] make firstbitlow intrinsic use first argument instead of return for overload type

[farzonl]

fixes #144966
Easy fix just add dx_firstbitlow to DirectXTTIImpl::isTargetIntrinsicWithOverloadTypeAtArg

[llvmbot]

@llvm/pr-subscribers-backend-directx

Author: Farzon Lotfi (farzonl)

Changes

…turn for overload type

fixes #144966
Easy fix just add dx_firstbitlow to DirectXTTIImpl::isTargetIntrinsicWithOverloadTypeAtArg

---

Full diff: https://github.com/llvm/llvm-project/pull/145350.diff

2 Files Affected:

• (modified) llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp (+1)
• (modified) llvm/test/CodeGen/DirectX/firstbitlow.ll (+17-4)

diff --git a/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp b/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp
index e426c5249f930..84b2603b4c575 100644
--- a/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp
+++ b/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp
@@ -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;
diff --git a/llvm/test/CodeGen/DirectX/firstbitlow.ll b/llvm/test/CodeGen/DirectX/firstbitlow.ll
index 884ec1164fc99..fe3786d3fbebf 100644
--- a/llvm/test/CodeGen/DirectX/firstbitlow.ll
+++ b/llvm/test/CodeGen/DirectX/firstbitlow.ll
@@ -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
+}

[farzonl]

Build failure is
LLVM :: CodeGen/DirectX/min_vec_size.ll
https://github.com/llvm/llvm-project/actions/runs/15829219011/job/44617332565

thats an existing test failure.

[V-FEXrt]

Note: This change needs to me made for first bit high as well. I'll do that in a follow up