[X86][SSE] Avoid unnecessary PMOVZX in v4i8 sum reduction

This should go away once D66004 has landed and we can simplify shuffle chains using demanded elts.

llvm-svn: 374658

1 files changed, 18 insertions, 7 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 6838dbbd08d..1f66355fe7c 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -36241,13 +36241,24 @@ static SDValue combineReductionToHorizontal(SDNode *ExtElt, SelectionDAG &DAG,
 
   // vXi8 reduction - sub 128-bit vector.
   if (VecVT == MVT::v4i8 || VecVT == MVT::v8i8) {
-    // Pad with zero.
-    if (VecVT == MVT::v4i8)
-      Rdx = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v8i8, Rdx,
-                        DAG.getConstant(0, DL, VecVT));
-    // Pad with undef.
-    Rdx = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v16i8, Rdx,
-                      DAG.getUNDEF(MVT::v8i8));
+    if (VecVT == MVT::v4i8) {
+      // Pad with zero.
+      if (Subtarget.hasSSE41()) {
+        Rdx = DAG.getBitcast(MVT::i32, Rdx);
+        Rdx = DAG.getNode(ISD::INSERT_VECTOR_ELT, DL, MVT::v4i32,
+                          DAG.getConstant(0, DL, MVT::v4i32), Rdx,
+                          DAG.getIntPtrConstant(0, DL));
+        Rdx = DAG.getBitcast(MVT::v16i8, Rdx);
+      } else {
+        Rdx = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v8i8, Rdx,
+                          DAG.getConstant(0, DL, VecVT));
+      }
+    }
+    if (Rdx.getValueType() == MVT::v8i8) {
+      // Pad with undef.
+      Rdx = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v16i8, Rdx,
+                        DAG.getUNDEF(MVT::v8i8));
+    }
     Rdx = DAG.getNode(X86ISD::PSADBW, DL, MVT::v2i64, Rdx,
                       DAG.getConstant(0, DL, MVT::v16i8));
     Rdx = DAG.getBitcast(MVT::v16i8, Rdx);