[x86] Lower some trunc + shuffle patterns to vpmov[q|d][b|w]

This should help in lowering the following four intrinsics:
 _mm256_cvtepi32_epi8
 _mm256_cvtepi64_epi16
 _mm256_cvtepi64_epi8
 _mm512_cvtepi64_epi8

Differential Revision: https://reviews.llvm.org/D46957

llvm-svn: 335238

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -4822,10 +4822,24 @@ static bool isUndefInRange(ArrayRef<int> Mask, unsigned Pos, unsigned Size) {
   return true;
 }
 
+/// Return true if Val falls within the specified range (L, H].
+static bool isInRange(int Val, int Low, int Hi) {
+  return (Val >= Low && Val < Hi);
+}
+
+/// Return true if the value of any element in Mask falls within the specified
+/// range (L, H].
+static bool isAnyInRange(ArrayRef<int> Mask, int Low, int Hi) {
+  for (int M : Mask)
+    if (isInRange(M, Low, Hi))
+      return true;
+  return false;
+}
+
 /// Return true if Val is undef or if its value falls within the
 /// specified range (L, H].
 static bool isUndefOrInRange(int Val, int Low, int Hi) {
-  return (Val == SM_SentinelUndef) || (Val >= Low && Val < Hi);
+  return (Val == SM_SentinelUndef) || isInRange(Val, Low, Hi);
 }
 
 /// Return true if every element in Mask is undef or if its value
@@ -4841,7 +4855,7 @@ static bool isUndefOrInRange(ArrayRef<int> Mask,
 /// Return true if Val is undef, zero or if its value falls within the
 /// specified range (L, H].
 static bool isUndefOrZeroOrInRange(int Val, int Low, int Hi) {
-  return isUndefOrZero(Val) || (Val >= Low && Val < Hi);
+  return isUndefOrZero(Val) || isInRange(Val, Low, Hi);
 }
 
 /// Return true if every element in Mask is undef, zero or if its value
@@ -4854,11 +4868,11 @@ static bool isUndefOrZeroOrInRange(ArrayRef<int> Mask, int Low, int Hi) {
 }
 
 /// Return true if every element in Mask, beginning
-/// from position Pos and ending in Pos+Size, falls within the specified
+/// from position Pos and ending in Pos + Size, falls within the specified
-/// sequential range (Low, Low+Size]. or is undef.
+/// sequence (Low, Low + Step, ..., Low + (Size - 1) * Step) or is undef.
-static bool isSequentialOrUndefInRange(ArrayRef<int> Mask,
+static bool isSequentialOrUndefInRange(ArrayRef<int> Mask, unsigned Pos,
-                                       unsigned Pos, unsigned Size, int Low) {
+                                       unsigned Size, int Low, int Step = 1) {
-  for (unsigned i = Pos, e = Pos+Size; i != e; ++i, ++Low)
+  for (unsigned i = Pos, e = Pos + Size; i != e; ++i, Low += Step)
     if (!isUndefOrEqual(Mask[i], Low))
       return false;
   return true;
@@ -9390,6 +9404,99 @@ static SDValue lowerVectorShuffleWithUNPCK(const SDLoc &DL, MVT VT,
   return SDValue();
 }
 
+static bool matchVectorShuffleAsVPMOV(ArrayRef<int> Mask, bool SwappedOps,
+                                         int Delta) {
+  int Size = (int)Mask.size();
+  int Split = Size / Delta;
+  int TruncatedVectorStart = SwappedOps ? Size : 0;
+
+  // Match for mask starting with e.g.: <8, 10, 12, 14,... or <0, 2, 4, 6,...
+  if (!isSequentialOrUndefInRange(Mask, 0, Split, TruncatedVectorStart, Delta))
+    return false;
+
+  // The rest of the mask should not refer to the truncated vector's elements.
+  if (isAnyInRange(Mask.slice(Split, Size - Split), TruncatedVectorStart,
+                   TruncatedVectorStart + Size))
+    return false;
+
+  return true;
+}
+
+// Try to lower trunc+vector_shuffle to a vpmovdb or a vpmovdw instruction.
+//
+// An example is the following:
+//
+// t0: ch = EntryToken
+//           t2: v4i64,ch = CopyFromReg t0, Register:v4i64 %0
+//         t25: v4i32 = truncate t2
+//       t41: v8i16 = bitcast t25
+//       t21: v8i16 = BUILD_VECTOR undef:i16, undef:i16, undef:i16, undef:i16,
+//       Constant:i16<0>, Constant:i16<0>, Constant:i16<0>, Constant:i16<0>
+//     t51: v8i16 = vector_shuffle<0,2,4,6,12,13,14,15> t41, t21
+//   t18: v2i64 = bitcast t51
+//
+// Without avx512vl, this is lowered to:
+//
+// vpmovqd %zmm0, %ymm0
+// vpshufb {{.*#+}} xmm0 =
+// xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+//
+// But when avx512vl is available, one can just use a single vpmovdw
+// instruction.
+static SDValue lowerVectorShuffleWithVPMOV(const SDLoc &DL, ArrayRef<int> Mask,
+                                           MVT VT, SDValue V1, SDValue V2,
+                                           SelectionDAG &DAG,
+                                           const X86Subtarget &Subtarget) {
+  if (VT != MVT::v16i8 && VT != MVT::v8i16)
+    return SDValue();
+
+  if (Mask.size() != VT.getVectorNumElements())
+    return SDValue();
+
+  bool SwappedOps = false;
+
+  if (!ISD::isBuildVectorAllZeros(V2.getNode())) {
+    if (!ISD::isBuildVectorAllZeros(V1.getNode()))
+      return SDValue();
+
+    std::swap(V1, V2);
+    SwappedOps = true;
+  }
+
+  // Look for:
+  //
+  // bitcast (truncate <8 x i32> %vec to <8 x i16>) to <16 x i8>
+  // bitcast (truncate <4 x i64> %vec to <4 x i32>) to <8 x i16>
+  //
+  // and similar ones.
+  if (V1.getOpcode() != ISD::BITCAST)
+    return SDValue();
+  if (V1.getOperand(0).getOpcode() != ISD::TRUNCATE)
+    return SDValue();
+
+  SDValue Src = V1.getOperand(0).getOperand(0);
+  MVT SrcVT = Src.getSimpleValueType();
+
+  // The vptrunc** instructions truncating 128 bit and 256 bit vectors
+  // are only available with avx512vl.
+  if (!SrcVT.is512BitVector() && !Subtarget.hasVLX())
+    return SDValue();
+
+  // Down Convert Word to Byte is only available with avx512bw. The case with
+  // 256-bit output doesn't contain a shuffle and is therefore not handled here.
+  if (SrcVT.getVectorElementType() == MVT::i16 && VT == MVT::v16i8 &&
+      !Subtarget.hasBWI())
+    return SDValue();
+
+  // The first half/quarter of the mask should refer to every second/fourth
+  // element of the vector truncated and bitcasted.
+  if (!matchVectorShuffleAsVPMOV(Mask, SwappedOps, 2) &&
+      !matchVectorShuffleAsVPMOV(Mask, SwappedOps, 4))
+    return SDValue();
+
+  return DAG.getNode(X86ISD::VTRUNC, DL, VT, Src);
+}
+
 // X86 has dedicated pack instructions that can handle specific truncation
 // operations: PACKSS and PACKUS.
 static bool matchVectorShuffleWithPACK(MVT VT, MVT &SrcVT, SDValue &V1,
@@ -14923,6 +15030,10 @@ static SDValue lowerVectorShuffle(SDValue Op, const X86Subtarget &Subtarget,
   if (canonicalizeShuffleMaskWithCommute(Mask))
     return DAG.getCommutedVectorShuffle(*SVOp);
 
+  if (SDValue V =
+          lowerVectorShuffleWithVPMOV(DL, Mask, VT, V1, V2, DAG, Subtarget))
+    return V;
+
   // For each vector width, delegate to a specialized lowering routine.
   if (VT.is128BitVector())
     return lower128BitVectorShuffle(DL, Mask, VT, V1, V2, Zeroable, Subtarget,

llvm/test/CodeGen/X86/shuffle-vs-trunc-256.ll

@@ -511,8 +511,7 @@ define <2 x i64> @trunc_v8i32_to_v8i8_return_v2i64(<8 x i32> %vec) nounwind {
 ;
 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -526,15 +525,13 @@ define <2 x i64> @trunc_v8i32_to_v8i8_return_v2i64(<8 x i32> %vec) nounwind {
 ;
 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated.vec = trunc <8 x i32> %vec to <8 x i8>
@@ -573,8 +570,7 @@ define <16 x i8> @trunc_v8i32_to_v8i8_with_zext_return_v16i8(<8 x i32> %vec) nou
 ;
 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -588,15 +584,13 @@ define <16 x i8> @trunc_v8i32_to_v8i8_with_zext_return_v16i8(<8 x i32> %vec) nou
 ;
 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated = trunc <8 x i32> %vec to <8 x i8>
@@ -636,8 +630,7 @@ define <16 x i8> @trunc_v8i32_to_v8i8_via_v8i16_return_v16i8(<8 x i32> %vec) nou
 ;
 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -651,15 +644,13 @@ define <16 x i8> @trunc_v8i32_to_v8i8_via_v8i16_return_v16i8(<8 x i32> %vec) nou
 ;
 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated = trunc <8 x i32> %vec to <8 x i16>
@@ -698,8 +689,7 @@ define <16 x i8> @trunc_v8i32_to_v8i8_return_v16i8(<8 x i32> %vec) nounwind {
 ;
 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -713,15 +703,13 @@ define <16 x i8> @trunc_v8i32_to_v8i8_return_v16i8(<8 x i32> %vec) nounwind {
 ;
 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated = trunc <8 x i32> %vec to <8 x i8>
@@ -766,8 +754,7 @@ define <2 x i64> @trunc_v4i64_to_v4i16_return_v2i64(<4 x i64> %vec) nounwind {
 ;
 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -781,15 +768,13 @@ define <2 x i64> @trunc_v4i64_to_v4i16_return_v2i64(<4 x i64> %vec) nounwind {
 ;
 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated = trunc <4 x i64> %vec to <4 x i16>
@@ -833,8 +818,7 @@ define <8 x i16> @trunc_v4i64_to_v4i16_with_zext_return_v8i16(<4 x i64> %vec) no
 ;
 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -848,15 +832,13 @@ define <8 x i16> @trunc_v4i64_to_v4i16_with_zext_return_v8i16(<4 x i64> %vec) no
 ;
 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated = trunc <4 x i64> %vec to <4 x i16>
@@ -901,8 +883,7 @@ define <8 x i16> @trunc_v4i64_to_v4i16_via_v4i32_return_v8i16(<4 x i64> %vec) no
 ;
 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -916,15 +897,13 @@ define <8 x i16> @trunc_v4i64_to_v4i16_via_v4i32_return_v8i16(<4 x i64> %vec) no
 ;
 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated = trunc <4 x i64> %vec to <4 x i32>
@@ -968,8 +947,7 @@ define <8 x i16> @trunc_v4i64_to_v4i16_return_v8i16(<4 x i64> %vec) nounwind {
 ;
 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -983,15 +961,13 @@ define <8 x i16> @trunc_v4i64_to_v4i16_return_v8i16(<4 x i64> %vec) nounwind {
 ;
 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovqw %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated = trunc <4 x i64> %vec to <4 x i16>
@@ -1034,8 +1010,7 @@ define <16 x i8> @trunc_v4i64_to_v4i8_return_v16i8(<4 x i64> %vec) nounwind {
 ;
 ; AVX512VL-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VL-NEXT:    vpmovqb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[u],zero
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
@@ -1049,15 +1024,13 @@ define <16 x i8> @trunc_v4i64_to_v4i8_return_v16i8(<4 x i64> %vec) nounwind {
 ;
 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmovqb %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[u],zero
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vpmovqd %ymm0, %xmm0
+; AVX512VBMIVL-NEXT:    vpmovqb %ymm0, %xmm0
-; AVX512VBMIVL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[u],zero
 ; AVX512VBMIVL-NEXT:    vzeroupper
 ; AVX512VBMIVL-NEXT:    retq
   %truncated = trunc <4 x i64> %vec to <4 x i8>

llvm/test/CodeGen/X86/shuffle-vs-trunc-512.ll

@@ -945,8 +945,7 @@ define <4 x double> @PR34175(<32 x i16>* %p) {
 define <16 x i8> @trunc_v8i64_to_v8i8_return_v16i8(<8 x i64> %vec) nounwind {
 ; AVX512-LABEL: trunc_v8i64_to_v8i8_return_v16i8:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512-NEXT:    vpmovqb %zmm0, %xmm0
-; AVX512-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %truncated = trunc <8 x i64> %vec to <8 x i8>