diff --git a/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp b/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp
index 84b2603b4c575..ffd900c68893f 100644
--- a/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp
+++ b/llvm/lib/Target/DirectX/DirectXTargetTransformInfo.cpp
@@ -31,6 +31,8 @@ bool DirectXTTIImpl::isTargetIntrinsicWithOverloadTypeAtArg(Intrinsic::ID ID,
   case Intrinsic::dx_asdouble:
   case Intrinsic::dx_isinf:
   case Intrinsic::dx_firstbitlow:
+  case Intrinsic::dx_firstbituhigh:
+  case Intrinsic::dx_firstbitshigh:
     return OpdIdx == 0;
   default:
     return OpdIdx == -1;
diff --git a/llvm/test/CodeGen/DirectX/firstbithigh.ll b/llvm/test/CodeGen/DirectX/firstbithigh.ll
index 794b0f20a0269..bf9efd30d9b2f 100644
--- a/llvm/test/CodeGen/DirectX/firstbithigh.ll
+++ b/llvm/test/CodeGen/DirectX/firstbithigh.ll
@@ -4,42 +4,42 @@
 
 define noundef i32 @test_firstbithigh_ushort(i16 noundef %a) {
 entry:
-; CHECK: call i32 @dx.op.unaryBits.i16(i32 33, i16 %{{.*}}) #[[#ATTR:]]
+; CHECK: call i32 @dx.op.unaryBits.i16(i32 33, i16 %{{.*}})
   %elt.firstbithigh = call i32 @llvm.dx.firstbituhigh.i16(i16 %a)
   ret i32 %elt.firstbithigh
 }
 
 define noundef i32 @test_firstbithigh_short(i16 noundef %a) {
 entry:
-; CHECK: call i32 @dx.op.unaryBits.i16(i32 34, i16 %{{.*}}) #[[#ATTR]]
+; CHECK: call i32 @dx.op.unaryBits.i16(i32 34, i16 %{{.*}})
   %elt.firstbithigh = call i32 @llvm.dx.firstbitshigh.i16(i16 %a)
   ret i32 %elt.firstbithigh
 }
 
 define noundef i32 @test_firstbithigh_uint(i32 noundef %a) {
 entry:
-; CHECK: call i32 @dx.op.unaryBits.i32(i32 33, i32 %{{.*}}) #[[#ATTR]]
+; CHECK: call i32 @dx.op.unaryBits.i32(i32 33, i32 %{{.*}})
   %elt.firstbithigh = call i32 @llvm.dx.firstbituhigh.i32(i32 %a)
   ret i32 %elt.firstbithigh
 }
 
 define noundef i32 @test_firstbithigh_int(i32 noundef %a) {
 entry:
-; CHECK: call i32 @dx.op.unaryBits.i32(i32 34, i32 %{{.*}}) #[[#ATTR]]
+; CHECK: call i32 @dx.op.unaryBits.i32(i32 34, i32 %{{.*}})
   %elt.firstbithigh = call i32 @llvm.dx.firstbitshigh.i32(i32 %a)
   ret i32 %elt.firstbithigh
 }
 
 define noundef i32 @test_firstbithigh_ulong(i64 noundef %a) {
 entry:
-; CHECK: call i32 @dx.op.unaryBits.i64(i32 33, i64 %{{.*}}) #[[#ATTR]]
+; CHECK: call i32 @dx.op.unaryBits.i64(i32 33, i64 %{{.*}})
   %elt.firstbithigh = call i32 @llvm.dx.firstbituhigh.i64(i64 %a)
   ret i32 %elt.firstbithigh
 }
 
 define noundef i32 @test_firstbithigh_long(i64 noundef %a) {
 entry:
-; CHECK: call i32 @dx.op.unaryBits.i64(i32 34, i64 %{{.*}}) #[[#ATTR]]
+; CHECK: call i32 @dx.op.unaryBits.i64(i32 34, i64 %{{.*}})
   %elt.firstbithigh = call i32 @llvm.dx.firstbitshigh.i64(i64 %a)
   ret i32 %elt.firstbithigh
 }
@@ -47,17 +47,18 @@ entry:
 define noundef <4 x i32> @test_firstbituhigh_vec4_i32(<4 x i32> noundef %a)  {
 entry:
   ; CHECK: [[ee0:%.*]] = extractelement <4 x i32> %a, i64 0
-  ; CHECK: [[ie0:%.*]] = call i32 @dx.op.unaryBits.i32(i32 33, i32 [[ee0]]) #[[#ATTR]]
+  ; CHECK: [[ie0:%.*]] = call i32 @dx.op.unaryBits.i32(i32 33, i32 [[ee0]])
   ; CHECK: [[ee1:%.*]] = extractelement <4 x i32> %a, i64 1
-  ; CHECK: [[ie1:%.*]] = call i32 @dx.op.unaryBits.i32(i32 33, i32 [[ee1]]) #[[#ATTR]]
+  ; CHECK: [[ie1:%.*]] = call i32 @dx.op.unaryBits.i32(i32 33, i32 [[ee1]])
   ; CHECK: [[ee2:%.*]] = extractelement <4 x i32> %a, i64 2
-  ; CHECK: [[ie2:%.*]] = call i32 @dx.op.unaryBits.i32(i32 33, i32 [[ee2]]) #[[#ATTR]]
+  ; CHECK: [[ie2:%.*]] = call i32 @dx.op.unaryBits.i32(i32 33, i32 [[ee2]])
   ; CHECK: [[ee3:%.*]] = extractelement <4 x i32> %a, i64 3
-  ; CHECK: [[ie3:%.*]] = call i32 @dx.op.unaryBits.i32(i32 33, i32 [[ee3]]) #[[#ATTR]]
-  ; 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
+  ; CHECK: [[ie3:%.*]] = call i32 @dx.op.unaryBits.i32(i32 33, i32 [[ee3]])
+  ; CHECK: [[rt0:%.*]] = insertelement <4 x i32> poison, i32 [[ie0]], i64 0
+  ; CHECK: [[rt1:%.*]] = insertelement <4 x i32> [[rt0]], i32 [[ie1]], i64 1
+  ; CHECK: [[rt2:%.*]] = insertelement <4 x i32> [[rt1]], i32 [[ie2]], i64 2
+  ; CHECK: [[rt3:%.*]] = insertelement <4 x i32> [[rt2]], i32 [[ie3]], i64 3
+  ; CHECK: ret <4 x i32> [[rt3]]
   %2 = call <4 x i32> @llvm.dx.firstbituhigh.v4i32(<4 x i32> %a)
   ret <4 x i32> %2
 }
@@ -65,29 +66,56 @@ entry:
 define noundef <4 x i32> @test_firstbitshigh_vec4_i32(<4 x i32> noundef %a)  {
 entry:
   ; CHECK: [[ee0:%.*]] = extractelement <4 x i32> %a, i64 0
-  ; CHECK: [[ie0:%.*]] = call i32 @dx.op.unaryBits.i32(i32 34, i32 [[ee0]]) #[[#ATTR]]
+  ; CHECK: [[ie0:%.*]] = call i32 @dx.op.unaryBits.i32(i32 34, i32 [[ee0]])
   ; CHECK: [[ee1:%.*]] = extractelement <4 x i32> %a, i64 1
-  ; CHECK: [[ie1:%.*]] = call i32 @dx.op.unaryBits.i32(i32 34, i32 [[ee1]]) #[[#ATTR]]
+  ; CHECK: [[ie1:%.*]] = call i32 @dx.op.unaryBits.i32(i32 34, i32 [[ee1]])
   ; CHECK: [[ee2:%.*]] = extractelement <4 x i32> %a, i64 2
-  ; CHECK: [[ie2:%.*]] = call i32 @dx.op.unaryBits.i32(i32 34, i32 [[ee2]]) #[[#ATTR]]
+  ; CHECK: [[ie2:%.*]] = call i32 @dx.op.unaryBits.i32(i32 34, i32 [[ee2]])
   ; CHECK: [[ee3:%.*]] = extractelement <4 x i32> %a, i64 3
-  ; CHECK: [[ie3:%.*]] = call i32 @dx.op.unaryBits.i32(i32 34, i32 [[ee3]]) #[[#ATTR]]
-  ; 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
+  ; CHECK: [[ie3:%.*]] = call i32 @dx.op.unaryBits.i32(i32 34, i32 [[ee3]])
+  ; CHECK: [[rt0:%.*]] = insertelement <4 x i32> poison, i32 [[ie0]], i64 0
+  ; CHECK: [[rt1:%.*]] = insertelement <4 x i32> [[rt0]], i32 [[ie1]], i64 1
+  ; CHECK: [[rt2:%.*]] = insertelement <4 x i32> [[rt1]], i32 [[ie2]], i64 2
+  ; CHECK: [[rt3:%.*]] = insertelement <4 x i32> [[rt2]], i32 [[ie3]], i64 3
+  ; CHECK: ret <4 x i32> [[rt3]]
   %2 = call <4 x i32> @llvm.dx.firstbitshigh.v4i32(<4 x i32> %a)
   ret <4 x i32> %2
 }
 
-; CHECK: attributes #[[#ATTR]] = {{{.*}} memory(none) {{.*}}}
-
-declare i32 @llvm.dx.firstbituhigh.i16(i16)
-declare i32 @llvm.dx.firstbituhigh.i32(i32)
-declare i32 @llvm.dx.firstbituhigh.i64(i64)
-declare <4 x i32> @llvm.dx.firstbituhigh.v4i32(<4 x i32>)
+define noundef <4 x i32> @test_firstbituhigh_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 33, i16 [[ee0]])
+  ; CHECK: [[ee1:%.*]] = extractelement <4 x i16> %a, i64 1
+  ; CHECK: [[ie1:%.*]] = call i32 @dx.op.unaryBits.i16(i32 33, i16 [[ee1]])
+  ; CHECK: [[ee2:%.*]] = extractelement <4 x i16> %a, i64 2
+  ; CHECK: [[ie2:%.*]] = call i32 @dx.op.unaryBits.i16(i32 33, i16 [[ee2]])
+  ; CHECK: [[ee3:%.*]] = extractelement <4 x i16> %a, i64 3
+  ; CHECK: [[ie3:%.*]] = call i32 @dx.op.unaryBits.i16(i32 33, i16 [[ee3]])
+  ; CHECK: [[rt0:%.*]] = insertelement <4 x i32> poison, i32 [[ie0]], i64 0
+  ; CHECK: [[rt1:%.*]] = insertelement <4 x i32> [[rt0]], i32 [[ie1]], i64 1
+  ; CHECK: [[rt2:%.*]] = insertelement <4 x i32> [[rt1]], i32 [[ie2]], i64 2
+  ; CHECK: [[rt3:%.*]] = insertelement <4 x i32> [[rt2]], i32 [[ie3]], i64 3
+  ; CHECK: ret <4 x i32> [[rt3]]
+  %2 = call <4 x i32> @llvm.dx.firstbituhigh.v4i16(<4 x i16> %a)
+  ret <4 x i32> %2
+}
 
-declare i32 @llvm.dx.firstbitshigh.i16(i16)
-declare i32 @llvm.dx.firstbitshigh.i32(i32)
-declare i32 @llvm.dx.firstbitshigh.i64(i64)
-declare <4 x i32> @llvm.dx.firstbitshigh.v4i32(<4 x i32>)
+define noundef <4 x i32> @test_firstbitshigh_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 34, i16 [[ee0]])
+  ; CHECK: [[ee1:%.*]] = extractelement <4 x i16> %a, i64 1
+  ; CHECK: [[ie1:%.*]] = call i32 @dx.op.unaryBits.i16(i32 34, i16 [[ee1]])
+  ; CHECK: [[ee2:%.*]] = extractelement <4 x i16> %a, i64 2
+  ; CHECK: [[ie2:%.*]] = call i32 @dx.op.unaryBits.i16(i32 34, i16 [[ee2]])
+  ; CHECK: [[ee3:%.*]] = extractelement <4 x i16> %a, i64 3
+  ; CHECK: [[ie3:%.*]] = call i32 @dx.op.unaryBits.i16(i32 34, i16 [[ee3]])
+  ; CHECK: [[rt0:%.*]] = insertelement <4 x i32> poison, i32 [[ie0]], i64 0
+  ; CHECK: [[rt1:%.*]] = insertelement <4 x i32> [[rt0]], i32 [[ie1]], i64 1
+  ; CHECK: [[rt2:%.*]] = insertelement <4 x i32> [[rt1]], i32 [[ie2]], i64 2
+  ; CHECK: [[rt3:%.*]] = insertelement <4 x i32> [[rt2]], i32 [[ie3]], i64 3
+  ; CHECK: ret <4 x i32> [[rt3]]
+  %2 = call <4 x i32> @llvm.dx.firstbitshigh.v4i16(<4 x i16> %a)
+  ret <4 x i32> %2
+}
diff --git a/llvm/test/CodeGen/DirectX/firstbitlow.ll b/llvm/test/CodeGen/DirectX/firstbitlow.ll
index fe3786d3fbebf..7f5f0b2dd7ed7 100644
--- a/llvm/test/CodeGen/DirectX/firstbitlow.ll
+++ b/llvm/test/CodeGen/DirectX/firstbitlow.ll
@@ -33,10 +33,11 @@ entry:
   ; CHECK: [[ie2:%.*]] = call i32 @dx.op.unaryBits.i32(i32 32, i32 [[ee2]])
   ; CHECK: [[ee3:%.*]] = extractelement <4 x i32> %a, i64 3
   ; CHECK: [[ie3:%.*]] = call i32 @dx.op.unaryBits.i32(i32 32, i32 [[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
+  ; CHECK: [[rt0:%.*]] = insertelement <4 x i32> poison, i32 [[ie0]], i64 0
+  ; CHECK: [[rt1:%.*]] = insertelement <4 x i32> [[rt0]], i32 [[ie1]], i64 1
+  ; CHECK: [[rt2:%.*]] = insertelement <4 x i32> [[rt1]], i32 [[ie2]], i64 2
+  ; CHECK: [[rt3:%.*]] = insertelement <4 x i32> [[rt2]], i32 [[ie3]], i64 3
+  ; CHECK: ret <4 x i32> [[rt3]]
   %2 = call <4 x i32> @llvm.dx.firstbitlow.v4i32(<4 x i32> %a)
   ret <4 x i32> %2
 }
@@ -51,10 +52,11 @@ entry:
   ; 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
+  ; CHECK: [[rt0:%.*]] = insertelement <4 x i32> poison, i32 [[ie0]], i64 0
+  ; CHECK: [[rt1:%.*]] = insertelement <4 x i32> [[rt0]], i32 [[ie1]], i64 1
+  ; CHECK: [[rt2:%.*]] = insertelement <4 x i32> [[rt1]], i32 [[ie2]], i64 2
+  ; CHECK: [[rt3:%.*]] = insertelement <4 x i32> [[rt2]], i32 [[ie3]], i64 3
+  ; CHECK: ret <4 x i32> [[rt3]]
   %2 = call <4 x i32> @llvm.dx.firstbitlow.v4i16(<4 x i16> %a)
   ret <4 x i32> %2
 }