2 files changed, 23 insertions, 6 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index d65a65e365c..ea97dc2dccd 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -13709,10 +13709,6 @@ static SDValue lowerV8I64VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
   assert(V2.getSimpleValueType() == MVT::v8i64 && "Bad operand type!");
   assert(Mask.size() == 8 && "Unexpected mask size for v8 shuffle!");
 
-  if (SDValue Shuf128 =
-          lowerV4X128VectorShuffle(DL, MVT::v8i64, Mask, V1, V2, DAG))
-    return Shuf128;
-
   if (V2.isUndef()) {
     // When the shuffle is mirrored between the 128-bit lanes of the unit, we
     // can use lower latency instructions that will operate on all four
@@ -13734,6 +13730,10 @@ static SDValue lowerV8I64VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
                          getV4X86ShuffleImm8ForMask(Repeated256Mask, DL, DAG));
   }
 
+  if (SDValue Shuf128 =
+          lowerV4X128VectorShuffle(DL, MVT::v8i64, Mask, V1, V2, DAG))
+    return Shuf128;
+
   // Try to use shift instructions.
   if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v8i64, V1, V2, Mask,
                                                 Zeroable, Subtarget, DAG))
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-512-v8.ll b/llvm/test/CodeGen/X86/vector-shuffle-512-v8.ll
index 6c980559721..1d17ef109d2 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-512-v8.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-512-v8.ll
@@ -1165,14 +1165,31 @@ define <8 x i64> @shuffle_v8i64_70000000(<8 x i64> %a, <8 x i64> %b) {
 define <8 x i64> @shuffle_v8i64_01014545(<8 x i64> %a, <8 x i64> %b) {
 ; AVX512F-LABEL: shuffle_v8i64_01014545:
 ; AVX512F:       # BB#0:
-; AVX512F-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,0,1,4,5,4,5]
+; AVX512F-NEXT:    vpermpd {{.*#+}} zmm0 = zmm0[0,1,0,1,4,5,4,5]
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512F-32-LABEL: shuffle_v8i64_01014545:
 ; AVX512F-32:       # BB#0:
-; AVX512F-32-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,0,1,4,5,4,5]
+; AVX512F-32-NEXT:    vpermpd {{.*#+}} zmm0 = zmm0[0,1,0,1,4,5,4,5]
+; AVX512F-32-NEXT:    retl
+
+  %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
+  ret <8 x i64> %shuffle
+}
+
+define <8 x i64> @shuffle_v8i64_01014545_mem(<8 x i64>* %ptr, <8 x i64> %b) {
+; AVX512F-LABEL: shuffle_v8i64_01014545_mem:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpermpd {{.*#+}} zmm0 = mem[0,1,0,1,4,5,4,5]
+; AVX512F-NEXT:    retq
+;
+; AVX512F-32-LABEL: shuffle_v8i64_01014545_mem:
+; AVX512F-32:       # BB#0:
+; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT:    vpermpd {{.*#+}} zmm0 = mem[0,1,0,1,4,5,4,5]
 ; AVX512F-32-NEXT:    retl
 
+  %a = load <8 x i64>, <8 x i64>* %ptr
   %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
   ret <8 x i64> %shuffle
 }