[X86] Don't use kunpck for vXi1 concat_vectors if the upper bits are undef.

This can be efficiently selected by a COPY_TO_REGCLASS without the need for an extra instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@319726 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index ee8351c81ce5ef7164f47cb196dd959ec61313af..234af1601786eaed4b5f90db7f95cd8694300602 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8318,6 +8318,11 @@ static SDValue LowerCONCAT_VECTORSvXi1(SDValue Op,
          V1.getValueType().getVectorNumElements() == NumElems/2 &&
          "Unexpected operands in CONCAT_VECTORS");
 
+  // If this can be done with a subreg insert do that first.
+  SDValue ZeroIdx = DAG.getIntPtrConstant(0, dl);
+  if (V2.isUndef())
+    return DAG.getNode(ISD::INSERT_SUBVECTOR, dl, ResVT, Undef, V1, ZeroIdx);
+
   if (ResVT.getSizeInBits() >= 16)
     return Op; // The operation is legal with KUNPCK
 
@@ -8327,9 +8332,6 @@ static SDValue LowerCONCAT_VECTORSvXi1(SDValue Op,
   if (IsZeroV1 && IsZeroV2)
     return ZeroVec;
 
-  SDValue ZeroIdx = DAG.getIntPtrConstant(0, dl);
-  if (V2.isUndef())
-    return DAG.getNode(ISD::INSERT_SUBVECTOR, dl, ResVT, Undef, V1, ZeroIdx);
   if (IsZeroV2)
     return DAG.getNode(ISD::INSERT_SUBVECTOR, dl, ResVT, ZeroVec, V1, ZeroIdx);
 

diff --git a/test/CodeGen/X86/avx512-skx-insert-subvec.ll b/test/CodeGen/X86/avx512-skx-insert-subvec.ll
index 91c3b73a95904e97c1d9cc3193efb4148a92cfae..4c650903e4716a5c3c6dc3667c653d7d5fe5cc7b 100644 (file)
--- a/test/CodeGen/X86/avx512-skx-insert-subvec.ll
+++ b/test/CodeGen/X86/avx512-skx-insert-subvec.ll
@@ -100,7 +100,6 @@ define <16 x i1> @test6(<2 x i1> %a, <2 x i1>%b) {
 ; CHECK-NEXT:    kshiftlb $2, %k0, %k0
 ; CHECK-NEXT:    kshiftrb $2, %k0, %k0
 ; CHECK-NEXT:    korb %k1, %k0, %k0
-; CHECK-NEXT:    kunpckbw %k0, %k0, %k0
 ; CHECK-NEXT:    vpmovm2b %k0, %xmm0
 ; CHECK-NEXT:    retq
 
@@ -119,8 +118,6 @@ define <32 x i1> @test7(<4 x i1> %a, <4 x i1>%b) {
 ; CHECK-NEXT:    kshiftlb $4, %k0, %k0
 ; CHECK-NEXT:    kshiftrb $4, %k0, %k0
 ; CHECK-NEXT:    korb %k1, %k0, %k0
-; CHECK-NEXT:    kunpckbw %k0, %k0, %k0
-; CHECK-NEXT:    kunpckwd %k0, %k0, %k0
 ; CHECK-NEXT:    vpmovm2b %k0, %ymm0
 ; CHECK-NEXT:    retq