[AVX-512] Add support for lowering shuffles to VALIGND/VALIGNQ

[craig.topper]

Summary

VALIGND and VALIGNQ are similar to PALIGNR but instead of working on a 128-bit lane they work on the entire vector register. This change leverages the shuffle rotate detection code used for PALIGNR to detect these cases.

[RKSimon]

Please can you add the new tests to trunk with the existing lowering?

For 128-bit vectors, we favor VALIGND/Q over PALIGNR so that we naturely match the vector element size in case the shuffle is part of a masked operation

Would we be better off putting this logic into combineSelect() and keep to PALIGNR? Detect mask/maskz selects taking a bitcasted target shuffles/logicals/whatever and attempt to narrow/widen them back? Not certain if this would actually work in general, its just a thought.

[craig.topper]

That sounds like a good idea. We can leave it as PALIGNR and not even try to do PALIGND/Q for 128-bit until combineSelect. Less code in shuffle lowering is probably best.

[craig.topper]

Removing 128-bit support at Simon's suggestion.

[RKSimon]

Add current codegen to trunk to show the diffs in the tests?

lib/Target/X86/X86ISelLowering.cpp

12210

Will this ever fire?

[craig.topper]

Updating tests to only show difference in codegen with this patch.

lib/Target/X86/X86ISelLowering.cpp

12210

VALIGND/Q work on the whole vector. PALIGNR works on 128-bit lanes so they should cover different cases.

[RKSimon]

LGTM - please can you create a PR about the AVX512 combineSelect shuffle/logical widening/narrowing idea (unless you intend to do it very soon).

I'll see if we can add matchVectorShuffleAsRotate to matchBinaryPermuteVectorShuffle.

lib/Target/X86/X86ISelLowering.cpp

if (SDValue BlendPerm =
             lowerVectorShuffleAsBlendAndPermute(DL, VT, V1, V2, Mask, DAG))
       return BlendPerm;
 
   V1 = DAG.getVectorShuffle(VT, DL, V1, DAG.getUNDEF(VT), V1Mask);
   V2 = DAG.getVectorShuffle(VT, DL, V2, DAG.getUNDEF(VT), V2Mask);
   return DAG.getVectorShuffle(VT, DL, V1, V2, BlendMask);
 }
 
-/// \brief Try to lower a vector shuffle as a byte rotation.
-///
-/// SSSE3 has a generic PALIGNR instruction in x86 that will do an arbitrary
-/// byte-rotation of the concatenation of two vectors; pre-SSSE3 can use
-/// a PSRLDQ/PSLLDQ/POR pattern to get a similar effect. This routine will
-/// try to generically lower a vector shuffle through such an pattern. It
-/// does not check for the profitability of lowering either as PALIGNR or
-/// PSRLDQ/PSLLDQ/POR, only whether the mask is valid to lower in that form.
-/// This matches shuffle vectors that look like:
+/// \brief Try to lower a vector shuffle as a rotation.
 ///
-///   v8i16 [11, 12, 13, 14, 15, 0, 1, 2]
-///
-/// Essentially it concatenates V1 and V2, shifts right by some number of
-/// elements, and takes the low elements as the result. Note that while this is
-/// specified as a *right shift* because x86 is little-endian, it is a *left
-/// rotate* of the vector lanes.
-static int matchVectorShuffleAsByteRotate(MVT VT, SDValue &V1, SDValue &V2,
-                                          ArrayRef<int> Mask) {
-  // Don't accept any shuffles with zero elements.
-  if (any_of(Mask, [](int M) { return M == SM_SentinelZero; }))
-    return -1;
-
-  // PALIGNR works on 128-bit lanes.
-  SmallVector<int, 16> RepeatedMask;
-  if (!is128BitLaneRepeatedShuffleMask(VT, Mask, RepeatedMask))
-    return -1;
-
-  int NumElts = RepeatedMask.size();
+/// This is used for support PALIGNR for SSSE3 or VALIGND/Q for AVX512.
+static int matchVectorShuffleAsRotate(SDValue &V1, SDValue &V2,
+  int NumElts = Mask.size();
 
   // We need to detect various ways of spelling a rotation:
   //   [11, 12, 13, 14, 15,  0,  1,  2]
   //   [-1, 12, 13, 14, -1, -1,  1, -1]
   //   [-1, -1, -1, -1, -1, -1,  1,  2]
   //   [ 3,  4,  5,  6,  7,  8,  9, 10]
   //   [-1,  4,  5,  6, -1, -1,  9, -1]
   //   [-1,  4,  5,  6, -1, -1, -1, -1]
   int Rotation = 0;
   SDValue Lo, Hi;
   for (int i = 0; i < NumElts; ++i) {
-    int M = RepeatedMask[i];
+    int M = Mask[i];
     assert((M == SM_SentinelUndef || (0 <= M && M < (2*NumElts))) &&
            "Unexpected mask index.");
     if (M < 0)
       continue;
 
     // Determine where a rotated vector would have started.
     int StartIdx = i - (M % NumElts);
     if (StartIdx == 0)
static int matchVectorShuffleAsRotate(SDValue &V1, SDValue &V2,
   if (!Lo)
     Lo = Hi;
   else if (!Hi)
     Hi = Lo;
 
   V1 = Lo;
   V2 = Hi;
 
+  return Rotation;
+}
+
+/// \brief Try to lower a vector shuffle as a byte rotation.
+///
+/// SSSE3 has a generic PALIGNR instruction in x86 that will do an arbitrary
+/// byte-rotation of the concatenation of two vectors; pre-SSSE3 can use
+/// a PSRLDQ/PSLLDQ/POR pattern to get a similar effect. This routine will
+/// try to generically lower a vector shuffle through such an pattern. It
+/// does not check for the profitability of lowering either as PALIGNR or
+/// PSRLDQ/PSLLDQ/POR, only whether the mask is valid to lower in that form.
+/// This matches shuffle vectors that look like:
+///
+///   v8i16 [11, 12, 13, 14, 15, 0, 1, 2]
+///
+/// Essentially it concatenates V1 and V2, shifts right by some number of
+/// elements, and takes the low elements as the result. Note that while this is
+/// specified as a *right shift* because x86 is little-endian, it is a *left
+/// rotate* of the vector lanes.
+static int matchVectorShuffleAsByteRotate(MVT VT, SDValue &V1, SDValue &V2,
+                                          ArrayRef<int> Mask) {
+  // Don't accept any shuffles with zero elements.
+  if (any_of(Mask, [](int M) { return M == SM_SentinelZero; }))
+    return -1;
+
+  // PALIGNR works on 128-bit lanes.
+  SmallVector<int, 16> RepeatedMask;
+  if (!is128BitLaneRepeatedShuffleMask(VT, Mask, RepeatedMask))
+    return -1;
+
+  int Rotation = matchVectorShuffleAsRotate(V1, V2, RepeatedMask);
+  if (Rotation <= 0)
+    return -1;
+
   // PALIGNR rotates bytes, so we need to scale the
   // rotation based on how many bytes are in the vector lane.
+  int NumElts = RepeatedMask.size();
   int Scale = 16 / NumElts;
   return Rotation * Scale;
 }
 
 static SDValue lowerVectorShuffleAsByteRotate(const SDLoc &DL, MVT VT,
                                               SDValue V1, SDValue V2,
                                               ArrayRef<int> Mask,
                                               const X86Subtarget &Subtarget,
static SDValue lowerVectorShuffleAsByteRotate(const SDLoc &DL, MVT VT,
   SDValue LoShift = DAG.getNode(X86ISD::VSHLDQ, DL, MVT::v16i8, Lo,
                                 DAG.getConstant(LoByteShift, DL, MVT::i8));
   SDValue HiShift = DAG.getNode(X86ISD::VSRLDQ, DL, MVT::v16i8, Hi,
                                 DAG.getConstant(HiByteShift, DL, MVT::i8));
   return DAG.getBitcast(VT,
                         DAG.getNode(ISD::OR, DL, MVT::v16i8, LoShift, HiShift));
 }
 
+/// \brief Try to lower a vector shuffle as a dword/qword rotation.
+///
+/// AVX512 has a VALIGND/VALIGNQ instructions that will do an arbitrary
+/// rotation of the concatenation of two vectors; This routine will
+/// try to generically lower a vector shuffle through such an pattern.
+///
+/// Essentially it concatenates V1 and V2, shifts right by some number of
+/// elements, and takes the low elements as the result. Note that while this is
+/// specified as a *right shift* because x86 is little-endian, it is a *left
+/// rotate* of the vector lanes.
+static SDValue lowerVectorShuffleAsRotate(const SDLoc &DL, MVT VT,
+                                          SDValue V1, SDValue V2,
+                                          ArrayRef<int> Mask,
+                                          const X86Subtarget &Subtarget,
+                                          SelectionDAG &DAG) {
+  assert((VT.getScalarType() == MVT::i32 || VT.getScalarType() == MVT::i64) &&
+         "Only 32-bit and 64-bit elements are supported!");
+
+  // 128/256-bit vectors are only supported with VLX.
+  assert((Subtarget.hasVLX() || (!VT.is128BitVector() && !VT.is256BitVector()))
+         && "VLX required for 128/256-bit vectors");
+
+  SDValue Lo = V1, Hi = V2;
+  int Rotation = matchVectorShuffleAsRotate(Lo, Hi, Mask);
+  if (Rotation <= 0)
+    return SDValue();
+
+  return DAG.getNode(X86ISD::VALIGN, DL, VT, Lo, Hi,
+                     DAG.getConstant(Rotation, DL, MVT::i8));
+}
+
 /// \brief Try to lower a vector shuffle as a bit shift (shifts in zeros).
 ///
 /// Attempts to match a shuffle mask against the PSLL(W/D/Q/DQ) and
 /// PSRL(W/D/Q/DQ) SSE2 and AVX2 logical bit-shift instructions. The function
 /// matches elements from one of the input vectors shuffled to the left or
 /// right with zeroable elements 'shifted in'. It handles both the strictly
 /// bit-wise element shifts and the byte shift across an entire 128-bit double
 /// quad word lane.
if (SDValue Blend = lowerVectorShuffleAsBlend(DL, MVT::v2i64, V1, V2, Mask,
                                                   Subtarget, DAG))
       return Blend;
 
   // Use dedicated unpack instructions for masks that match their pattern.
   if (SDValue V =
           lowerVectorShuffleWithUNPCK(DL, MVT::v2i64, Mask, V1, V2, DAG))
     return V;
 
+  // If we have VLX support, we can use VALIGN.
+  if (Subtarget.hasVLX())
+    if (SDValue Rotate = lowerVectorShuffleAsRotate(DL, MVT::v2i64, V1, V2,
+                                                    Mask, Subtarget, DAG))
+      return Rotate;
+
   // Try to use byte rotation instructions.
   // Its more profitable for pre-SSSE3 to use shuffles/unpacks.
   if (Subtarget.hasSSSE3())
     if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
             DL, MVT::v2i64, V1, V2, Mask, Subtarget, DAG))
       return Rotate;
 
   // If we have direct support for blends, we should lower by decomposing into
if (SDValue Masked =
           lowerVectorShuffleAsBitMask(DL, MVT::v4i32, V1, V2, Mask, DAG))
     return Masked;
 
   // Use dedicated unpack instructions for masks that match their pattern.
   if (SDValue V =
           lowerVectorShuffleWithUNPCK(DL, MVT::v4i32, Mask, V1, V2, DAG))
     return V;
 
+  // If we have VLX support, we can use VALIGN.
+  if (Subtarget.hasVLX())
+    if (SDValue Rotate = lowerVectorShuffleAsRotate(DL, MVT::v4i32, V1, V2,
+                                                    Mask, Subtarget, DAG))
+      return Rotate;
+
   // Try to use byte rotation instructions.
   // Its more profitable for pre-SSSE3 to use shuffles/unpacks.
   if (Subtarget.hasSSSE3())
     if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
             DL, MVT::v4i32, V1, V2, Mask, Subtarget, DAG))
       return Rotate;
 
   // If we have direct support for blends, we should lower by decomposing into
return DAG.getNode(X86ISD::VPERMI, DL, MVT::v4i64, V1,
                        getV4X86ShuffleImm8ForMask(Mask, DL, DAG));
   }
 
   // Try to use shift instructions.
   if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v4i64, V1, V2, Mask,
                                                 Subtarget, DAG))
     return Shift;
 
+  // If we have VLX support, we can use VALIGN.
+  if (Subtarget.hasVLX())
+    if (SDValue Rotate = lowerVectorShuffleAsRotate(DL, MVT::v8i32, V1, V2,
+                                                    Mask, Subtarget, DAG))
+      return Rotate;
+
+  // Try to use PALIGNR.
   if (SDValue Rotate = lowerVectorShuffleAsByteRotate(DL, MVT::v4i64, V1, V2,
                                                       Mask, Subtarget, DAG))
     return Rotate;
 
   // Use dedicated unpack instructions for masks that match their pattern.
   if (SDValue V =
           lowerVectorShuffleWithUNPCK(DL, MVT::v4i64, Mask, V1, V2, DAG))
     return V;
if (SDValue V =
       return V;
   }
 
   // Try to use shift instructions.
   if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v8i32, V1, V2, Mask,
                                                 Subtarget, DAG))
     return Shift;
 
+  // If we have VLX support, we can use VALIGN.
+  if (Subtarget.hasVLX())
+    if (SDValue Rotate = lowerVectorShuffleAsRotate(DL, MVT::v8i32, V1, V2,
+                                                    Mask, Subtarget, DAG))
+      return Rotate;
+
   // Try to use byte rotation instructions.
   if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
           DL, MVT::v8i32, V1, V2, Mask, Subtarget, DAG))
     return Rotate;
 
   // Try to create an in-lane repeating shuffle mask and then shuffle the
   // results into the target lanes.
   if (SDValue V = lowerShuffleAsRepeatedMaskAndLanePermute(
if (is256BitLaneRepeatedShuffleMask(MVT::v8i64, Mask, Repeated256Mask))
                          getV4X86ShuffleImm8ForMask(Repeated256Mask, DL, DAG));
   }
 
   // Try to use shift instructions.
   if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v8i64, V1, V2, Mask,
                                                 Subtarget, DAG))
     return Shift;
 
+  // Try to use VALIGN.
+  if (SDValue Rotate = lowerVectorShuffleAsRotate(DL, MVT::v8i64, V1, V2,
+                                                  Mask, Subtarget, DAG))
+    return Rotate;
+
+  // Try to use PALIGNR.
   if (SDValue Rotate = lowerVectorShuffleAsByteRotate(DL, MVT::v8i64, V1, V2,
                                                       Mask, Subtarget, DAG))
     return Rotate;
 
   if (SDValue Unpck =
           lowerVectorShuffleWithUNPCK(DL, MVT::v8i64, Mask, V1, V2, DAG))
     return Unpck;
 
if (SDValue V =
       return V;
   }
 
   // Try to use shift instructions.
   if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v16i32, V1, V2, Mask,
                                                 Subtarget, DAG))
     return Shift;
 
+  // Try to use VALIGN.
+  if (SDValue Rotate = lowerVectorShuffleAsRotate(DL, MVT::v16i32, V1, V2,
+                                                  Mask, Subtarget, DAG))
+    return Rotate;
+
   // Try to use byte rotation instructions.
   if (Subtarget.hasBWI())
     if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
             DL, MVT::v16i32, V1, V2, Mask, Subtarget, DAG))
       return Rotate;
 
   return lowerVectorShuffleWithPERMV(DL, MVT::v16i32, Mask, V1, V2, DAG);
 }
 
 /// \brief Handle lowering of 32-lane 16-bit integer shuffles.
 static SDValue lowerV32I16VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,

test/CodeGen/X86/vector-shuffle-128-v2.ll

 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v2i64_12:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v2i64_12:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v2i64_12:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v2i64_12:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v2i64_12:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignq {{.*#+}} xmm0 = xmm0[1],xmm1[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 2>
   ret <2 x i64> %shuffle
 }
 define <2 x i64> @shuffle_v2i64_12_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
 ; SSE2-LABEL: shuffle_v2i64_12_copy:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    shufpd {{.*#+}} xmm1 = xmm1[1],xmm2[0]
 ; SSE2-NEXT:    movapd %xmm1, %xmm0

 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v2i64_12_copy:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm2 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v2i64_12_copy:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v2i64_12_copy:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v2i64_12_copy:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v2i64_12_copy:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignq {{.*#+}} xmm0 = xmm1[1],xmm2[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 1, i32 2>
   ret <2 x i64> %shuffle
 }
 define <2 x i64> @shuffle_v2i64_13(<2 x i64> %a, <2 x i64> %b) {
 ; SSE-LABEL: shuffle_v2i64_13:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSE-NEXT:    retq

 ; SSSE3-NEXT:    palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v2i64_30:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v2i64_30:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v2i64_30:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v2i64_30:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v2i64_30:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignq {{.*#+}} xmm0 = xmm1[1],xmm0[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 0>
   ret <2 x i64> %shuffle
 }
 define <2 x i64> @shuffle_v2i64_30_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64> %b) {
 ; SSE2-LABEL: shuffle_v2i64_30_copy:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    shufpd {{.*#+}} xmm2 = xmm2[1],xmm1[0]
 ; SSE2-NEXT:    movapd %xmm2, %xmm0

 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v2i64_30_copy:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v2i64_30_copy:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
-; AVX-NEXT:    retq
+; AVX1-LABEL: shuffle_v2i64_30_copy:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v2i64_30_copy:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm2[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v2i64_30_copy:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignq {{.*#+}} xmm0 = xmm2[1],xmm1[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 3, i32 0>
   ret <2 x i64> %shuffle
 }
 define <2 x i64> @shuffle_v2i64_31(<2 x i64> %a, <2 x i64> %b) {
 ; SSE-LABEL: shuffle_v2i64_31:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm0[1]
 ; SSE-NEXT:    movdqa %xmm1, %xmm0

test/CodeGen/X86/vector-shuffle-128-v4.ll

 ; SSSE3-NEXT:    palignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v4i32_7012:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v4i32_7012:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: shuffle_v4i32_7012:
+; AVX1OR2:       # BB#0:
+; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
+; AVX1OR2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i32_7012:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignd {{.*#+}} xmm0 = xmm1[3],xmm0[0,1,2]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 7, i32 0, i32 1, i32 2>
   ret <4 x i32> %shuffle
 }
 
 define <4 x i32> @shuffle_v4i32_6701(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: shuffle_v4i32_6701:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]

 ; SSSE3-NEXT:    palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v4i32_6701:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v4i32_6701:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: shuffle_v4i32_6701:
+; AVX1OR2:       # BB#0:
+; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
+; AVX1OR2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i32_6701:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignq {{.*#+}} xmm0 = xmm1[1],xmm0[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
   ret <4 x i32> %shuffle
 }
 
 define <4 x i32> @shuffle_v4i32_5670(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: shuffle_v4i32_5670:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]

 ; SSSE3-NEXT:    palignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v4i32_5670:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v4i32_5670:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: shuffle_v4i32_5670:
+; AVX1OR2:       # BB#0:
+; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
+; AVX1OR2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i32_5670:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignd {{.*#+}} xmm0 = xmm1[1,2,3],xmm0[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 5, i32 6, i32 7, i32 0>
   ret <4 x i32> %shuffle
 }
 
 define <4 x i32> @shuffle_v4i32_1234(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: shuffle_v4i32_1234:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[3,0]

 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v4i32_1234:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v4i32_1234:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: shuffle_v4i32_1234:
+; AVX1OR2:       # BB#0:
+; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
+; AVX1OR2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i32_1234:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignd {{.*#+}} xmm0 = xmm0[1,2,3],xmm1[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
   ret <4 x i32> %shuffle
 }
 
 define <4 x i32> @shuffle_v4i32_2345(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: shuffle_v4i32_2345:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]

 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v4i32_2345:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v4i32_2345:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: shuffle_v4i32_2345:
+; AVX1OR2:       # BB#0:
+; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
+; AVX1OR2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i32_2345:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignq {{.*#+}} xmm0 = xmm0[1],xmm1[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
   ret <4 x i32> %shuffle
 }
 
 define <4 x i32> @shuffle_v4i32_40u1(<4 x i32> %a, <4 x i32> %b) {
 ; SSE-LABEL: shuffle_v4i32_40u1:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]

 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v4i32_3456:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v4i32_3456:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: shuffle_v4i32_3456:
+; AVX1OR2:       # BB#0:
+; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
+; AVX1OR2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i32_3456:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignd {{.*#+}} xmm0 = xmm0[3],xmm1[0,1,2]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
   ret <4 x i32> %shuffle
 }
 
 define <4 x i32> @shuffle_v4i32_0u1u(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: shuffle_v4i32_0u1u:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]

test/CodeGen/X86/vector-shuffle-128-v8.ll

 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v8i16_456789AB:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v8i16_456789AB:
-; AVX:       # BB#0:
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
-; AVX-NEXT:    retq
+; AVX1OR2-LABEL: shuffle_v8i16_456789AB:
+; AVX1OR2:       # BB#0:
+; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
+; AVX1OR2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v8i16_456789AB:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignq {{.*#+}} xmm0 = xmm0[1],xmm1[0]
+; AVX512VL-NEXT:    retq
   %shuffle = shufflevector <8 x i16> %a, <8 x i16> %b, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>
   ret <8 x i16> %shuffle
 }
 
 define <8 x i16> @shuffle_v8i16_00000000(<8 x i16> %a, <8 x i16> %b) {
 ; SSE-LABEL: shuffle_v8i16_00000000:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]

test/CodeGen/X86/vector-shuffle-256-v4.ll

 ; ALL:       # BB#0:
 ; ALL-NEXT:    vbroadcastsd (%rdi), %ymm0
 ; ALL-NEXT:    retq
   %tmp = load i64, i64* %ptr, align 1
   %tmp1 = insertelement <2 x i64> undef, i64 %tmp, i32 0
   %tmp2 = shufflevector <2 x i64> %tmp1, <2 x i64> undef, <4 x i32> zeroinitializer
   ret <4 x i64> %tmp2
 }
+
+define <4 x i64> @shuffle_v4i64_1234(<4 x i64> %a, <4 x i64> %b) {
+; AVX1-LABEL: shuffle_v4i64_1234:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendpd {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX1-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[1],ymm1[0],ymm0[3],ymm1[2]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v4i64_1234:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[1,2,3,0]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i64_1234:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignd {{.*#+}} ymm0 = ymm0[1,2,3,4,5,6,7],ymm1[0]
+; AVX512VL-NEXT:    retq
+  %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
+  ret <4 x i64> %shuffle
+}
+
+define <4 x i64> @shuffle_v4i64_1230(<4 x i64> %a) {
+; AVX1-LABEL: shuffle_v4i64_1230:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX1-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[1],ymm1[0],ymm0[3],ymm1[2]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v4i64_1230:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[1,2,3,0]
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i64_1230:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[1,2,3,0]
+; AVX512VL-NEXT:    retq
+  %shuffle = shufflevector <4 x i64> %a, <4 x i64> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 0>
+  ret <4 x i64> %shuffle
+}

test/CodeGen/X86/vector-shuffle-256-v8.ll

 ; ALL:       # BB#0:
 ; ALL-NEXT:    vbroadcastss (%rdi), %ymm0
 ; ALL-NEXT:    retq
   %tmp = load i32, i32* %ptr, align 4
   %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
   %tmp2 = shufflevector <4 x i32> %tmp1, <4 x i32> undef, <8 x i32> zeroinitializer
   ret <8 x i32> %tmp2
 }
+
+define <8 x i32> @shuffle_v8i32_12345678(<8 x i32> %a, <8 x i32> %b) {
+; AVX1-LABEL: shuffle_v8i32_12345678:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX1-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0],ymm0[3,0],ymm1[4,4],ymm0[7,4]
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,2],ymm1[2,0],ymm0[5,6],ymm1[6,4]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v8i32_12345678:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7]
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,2,3,4,5,6,7,0]
+; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v8i32_12345678:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignd {{.*#+}} ymm0 = ymm0[1,2,3,4,5,6,7],ymm1[0]
+; AVX512VL-NEXT:    retq
+  %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
+  ret <8 x i32> %shuffle
+}
+
+define <8 x i32> @shuffle_v8i32_12345670(<8 x i32> %a) {
+; AVX1-LABEL: shuffle_v8i32_12345670:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX1-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0],ymm0[3,0],ymm1[4,4],ymm0[7,4]
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,2],ymm1[2,0],ymm0[5,6],ymm1[6,4]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v8i32_12345670:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,2,3,4,5,6,7,0]
+; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v8i32_12345670:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    valignd {{.*#+}} ymm0 = ymm0[1,2,3,4,5,6,7,0]
+; AVX512VL-NEXT:    retq
+  %shuffle = shufflevector <8 x i32> %a, <8 x i32> undef, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0>
+  ret <8 x i32> %shuffle
+}

test/CodeGen/X86/vector-shuffle-512-v16.ll

 define <16 x i32> @shuffle_v16i32_16_zz_17_zz_18_zz_19_zz_20_zz_21_zz_22_zz_23_zz(<16 x i32> %a) {
 ; ALL-LABEL: shuffle_v16i32_16_zz_17_zz_18_zz_19_zz_20_zz_21_zz_22_zz_23_zz:
 ; ALL:       # BB#0:
 ; ALL-NEXT:    vpmovzxdq {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero
 ; ALL-NEXT:    retq
   %shuffle = shufflevector <16 x i32> zeroinitializer, <16 x i32> %a, <16 x i32> <i32 16, i32 0, i32 17, i32 0, i32 18, i32 0, i32 19, i32 0, i32 20, i32 0, i32 21, i32 0, i32 22, i32 0, i32 23, i32 0>
   ret <16 x i32> %shuffle
 }
+
+define <16 x i32> @shuffle_v16i32_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_16(<16 x i32> %a, <16 x i32> %b) {
+; ALL-LABEL: shuffle_v16i32_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_16:
+; ALL:       # BB#0:
+; ALL-NEXT:    valignd {{.*#+}} zmm0 = zmm0[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],zmm1[0]
+; ALL-NEXT:    retq
+  %shuffle = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32><i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16>
+  ret <16 x i32> %shuffle
+}
+
+define <16 x i32> @shuffle_v16i32_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_00(<16 x i32> %a) {
+; ALL-LABEL: shuffle_v16i32_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_00:
+; ALL:       # BB#0:
+; ALL-NEXT:    valignd {{.*#+}} zmm0 = zmm0[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,0]
+; ALL-NEXT:    retq
+  %shuffle = shufflevector <16 x i32> %a, <16 x i32> undef, <16 x i32><i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 0>
+  ret <16 x i32> %shuffle
+}

test/CodeGen/X86/vector-shuffle-512-v8.ll

 ; AVX512F-32-LABEL: shuffle_v8f64_0zzzzzzz:
 ; AVX512F-32:       # BB#0:
 ; AVX512F-32-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
 ; AVX512F-32-NEXT:    vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
 ; AVX512F-32-NEXT:    retl
   %shuffle = shufflevector <8 x double> %a, <8 x double> zeroinitializer, <8 x i32> <i32 0, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
   ret <8 x double> %shuffle
 }
+
+define <8 x i64> @shuffle_v8i64_12345678(<8 x i64> %a, <8 x i64> %b) {
+;
+; AVX512F-LABEL: shuffle_v8i64_12345678:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    valignq {{.*#+}} zmm0 = zmm0[1,2,3,4,5,6,7],zmm1[0]
+; AVX512F-NEXT:    retq
+;
+; AVX512F-32-LABEL: shuffle_v8i64_12345678:
+; AVX512F-32:       # BB#0:
+; AVX512F-32-NEXT:    valignq {{.*#+}} zmm0 = zmm0[1,2,3,4,5,6,7],zmm1[0]
+; AVX512F-32-NEXT:    retl
+  %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
+  ret <8 x i64> %shuffle
+}
+
+define <8 x i64> @shuffle_v8i64_12345670(<8 x i64> %a) {
+;
+; AVX512F-LABEL: shuffle_v8i64_12345670:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    valignq {{.*#+}} zmm0 = zmm0[1,2,3,4,5,6,7,0]
+; AVX512F-NEXT:    retq
+;
+; AVX512F-32-LABEL: shuffle_v8i64_12345670:
+; AVX512F-32:       # BB#0:
+; AVX512F-32-NEXT:    valignq {{.*#+}} zmm0 = zmm0[1,2,3,4,5,6,7,0]
+; AVX512F-32-NEXT:    retl
+  %shuffle = shufflevector <8 x i64> %a, <8 x i64> undef, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0>
+  ret <8 x i64> %shuffle
+}

test/CodeGen/X86/vector-shuffle-v1.ll

 ; VL_BW_DQ-LABEL: shuf2i1_1_2:
 ; VL_BW_DQ:       # BB#0:
 ; VL_BW_DQ-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; VL_BW_DQ-NEXT:    vptestmq %xmm0, %xmm0, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2q %k0, %xmm0
 ; VL_BW_DQ-NEXT:    movb $1, %al
 ; VL_BW_DQ-NEXT:    kmovb %eax, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2q %k0, %xmm1
-; VL_BW_DQ-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
+; VL_BW_DQ-NEXT:    valignq {{.*#+}} xmm0 = xmm0[1],xmm1[0]
 ; VL_BW_DQ-NEXT:    vpmovq2m %xmm0, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2q %k0, %xmm0
 ; VL_BW_DQ-NEXT:    retq
   %b = shufflevector <2 x i1> %a, <2 x i1> <i1 1, i1 0>, <2 x i32> <i32 1, i32 2>
   ret <2 x i1> %b
 }