Fix vectorization of canonicalize

llvm-svn: 342390

3 files changed, 21 insertions, 0 deletions

diff --git a/llvm/include/llvm/CodeGen/BasicTTIImpl.h b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
index f76a2426377..0cb1ab62785 100644
--- a/llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -1036,6 +1036,9 @@ public:
     case Intrinsic::fabs:
       ISDs.push_back(ISD::FABS);
       break;
+    case Intrinsic::canonicalize:
+      ISDs.push_back(ISD::FCANONICALIZE);
+      break;
     case Intrinsic::minnum:
       ISDs.push_back(ISD::FMINNUM);
       if (FMF.noNaNs())
diff --git a/llvm/lib/Analysis/VectorUtils.cpp b/llvm/lib/Analysis/VectorUtils.cpp
index bbdea4de4e2..2f4f72aaa6c 100644
--- a/llvm/lib/Analysis/VectorUtils.cpp
+++ b/llvm/lib/Analysis/VectorUtils.cpp
@@ -70,6 +70,7 @@ bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) {
   case Intrinsic::ctlz:
   case Intrinsic::cttz:
   case Intrinsic::powi:
+  case Intrinsic::canonicalize:
     return true;
   default:
     return false;
diff --git a/llvm/test/Transforms/SLPVectorizer/AMDGPU/packed-math.ll b/llvm/test/Transforms/SLPVectorizer/AMDGPU/packed-math.ll
index 14e2cde0e29..55905a4c444 100644
--- a/llvm/test/Transforms/SLPVectorizer/AMDGPU/packed-math.ll
+++ b/llvm/test/Transforms/SLPVectorizer/AMDGPU/packed-math.ll
@@ -179,8 +179,25 @@ define amdgpu_kernel void @test1_fabs_scalar_fma_v2f16(half addrspace(3)* %a, ha
   ret void
 }
 
+; GCN-LABEL: @canonicalize_v2f16
+; GFX9: load <2 x half>
+; GFX9: call <2 x half> @llvm.canonicalize.v2f16(
+; GFX9: store <2 x half>
+define amdgpu_kernel void @canonicalize_v2f16(half addrspace(3)* %a, half addrspace(3)* %c) {
+  %i0 = load half, half addrspace(3)* %a, align 2
+  %canonicalize0 = call half @llvm.canonicalize.f16(half %i0)
+  %arrayidx3 = getelementptr inbounds half, half addrspace(3)* %a, i64 1
+  %i3 = load half, half addrspace(3)* %arrayidx3, align 2
+  %canonicalize1 = call half @llvm.canonicalize.f16(half %i3)
+  store half %canonicalize0, half addrspace(3)* %c, align 2
+  %arrayidx5 = getelementptr inbounds half, half addrspace(3)* %c, i64 1
+  store half %canonicalize1, half addrspace(3)* %arrayidx5, align 2
+  ret void
+}
+
 declare half @llvm.fabs.f16(half) #1
 declare half @llvm.fma.f16(half, half, half) #1
+declare half @llvm.canonicalize.f16(half) #1
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }