DAG: Fix vector widening fcanonicalize

llvm-svn: 338715

2 files changed, 21 insertions, 0 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index a7ccb3c05ec..ab868c34805 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -2394,6 +2394,7 @@ void DAGTypeLegalizer::WidenVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::FSIN:
   case ISD::FSQRT:
   case ISD::FTRUNC:
+  case ISD::FCANONICALIZE:
     Res = WidenVecRes_Unary(N);
     break;
   case ISD::FMA:
diff --git a/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll b/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
index 440904f9cd2..a3022d1e0b9 100644
--- a/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
@@ -6,6 +6,7 @@ declare half @llvm.fabs.f16(half) #0
 declare half @llvm.canonicalize.f16(half) #0
 declare <2 x half> @llvm.fabs.v2f16(<2 x half>) #0
 declare <2 x half> @llvm.canonicalize.v2f16(<2 x half>) #0
+declare <3 x half> @llvm.canonicalize.v3f16(<3 x half>) #0
 declare <4 x half> @llvm.canonicalize.v4f16(<4 x half>) #0
 declare i32 @llvm.amdgcn.workitem.id.x() #0
 
@@ -477,6 +478,25 @@ define amdgpu_kernel void @test_fold_canonicalize_snan3_value_v2f16(<2 x half> a
   ret void
 }
 
+; FIXME: Extra 4th component handled
+; GCN-LABEL: {{^}}v_test_canonicalize_var_v3f16:
+; GFX9: s_waitcnt
+; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX9-NEXT: s_setpc_b64
+
+; VI-DAG: v_max_f16_sdwa [[CANON_ELT3:v[0-9]+]], v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-DAG: v_max_f16_e32 [[CANON_ELT2:v[0-9]+]], v1, v1
+; VI-DAG: v_max_f16_sdwa [[CANON_ELT1:v[0-9]+]], v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-DAG: v_max_f16_e32 [[CANON_ELT0:v[0-9]+]], v0, v0
+; VI-DAG: v_or_b32_e32 v0, [[CANON_ELT0]], [[CANON_ELT1]]
+; VI-DAG: v_or_b32_e32 v1, [[CANON_ELT2]], [[CANON_ELT3]]
+; VI: s_setpc_b64
+define <3 x half> @v_test_canonicalize_var_v3f16(<3 x half> %val) #1 {
+  %canonicalized = call <3 x half> @llvm.canonicalize.v3f16(<3 x half> %val)
+  ret <3 x half> %canonicalized
+}
+
 ; GCN-LABEL: {{^}}v_test_canonicalize_var_v4f16:
 ; GFX9: s_waitcnt
 ; GFX9-NEXT: v_pk_max_f16 v0, v0, v0