[X86] Disable combineToExtendVectorInReg under -x86-experimental-vector-widening-legalization. Add custom type legalization for extends.

If we widen illegal types instead of promoting, we should be able to rely on the type legalizer to create the vector_inreg operations for us with some caveats.

This patch disables combineToExtendVectorInReg when we are using widening.

I've enabled custom legalization for v8i8->v8i64 extends under avx512f since the type legalizer would want to create a vector_inreg with a v64i8 input type which isn't legal without avx512bw. So we go to v16i8 with custom code using the relaxation of rules we get from D54346.

I've also enable custom legalization of v8i64 and v16i32 operations with with AVX. When the input type is 128 bits, the default splitting legalization would extend first 128->256, then do the a split to two 128 pieces. Extend each half to 256 and then concat the result. The custom legalization I've added instead uses a 128->256 bit vector_inreg extend that only reads the lower 64-bits for the low half of the split. Then shuffles the high 64-bits to the low 64-bits and does another vector_inreg extend.

llvm-svn: 347172

4 files changed, 154 insertions, 94 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 1c8f6aad21c..43be941e1a4 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1104,6 +1104,14 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
       setOperationAction(ISD::SRA, VT, Custom);
     }
 
+    if (ExperimentalVectorWideningLegalization) {
+      // These types need custom splitting if their input is a 128-bit vector.
+      setOperationAction(ISD::SIGN_EXTEND,       MVT::v8i64,  Custom);
+      setOperationAction(ISD::SIGN_EXTEND,       MVT::v16i32, Custom);
+      setOperationAction(ISD::ZERO_EXTEND,       MVT::v8i64,  Custom);
+      setOperationAction(ISD::ZERO_EXTEND,       MVT::v16i32, Custom);
+    }
+
     setOperationAction(ISD::ROTL,              MVT::v8i32,  Custom);
     setOperationAction(ISD::ROTL,              MVT::v16i16, Custom);
     setOperationAction(ISD::ROTL,              MVT::v32i8,  Custom);
@@ -1368,6 +1376,13 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
     setOperationAction(ISD::SIGN_EXTEND,        MVT::v16i32, Custom);
     setOperationAction(ISD::SIGN_EXTEND,        MVT::v8i64, Custom);
 
+    if (ExperimentalVectorWideningLegalization) {
+      // Need to custom widen this if we don't have AVX512BW.
+      setOperationAction(ISD::ANY_EXTEND,         MVT::v8i8, Custom);
+      setOperationAction(ISD::ZERO_EXTEND,        MVT::v8i8, Custom);
+      setOperationAction(ISD::SIGN_EXTEND,        MVT::v8i8, Custom);
+    }
+
     for (auto VT : { MVT::v16f32, MVT::v8f64 }) {
       setOperationAction(ISD::FFLOOR,           VT, Legal);
       setOperationAction(ISD::FCEIL,            VT, Legal);
@@ -17636,6 +17651,17 @@ static SDValue LowerAVXExtend(SDValue Op, SelectionDAG &DAG,
           InVT.getVectorElementType() == MVT::i32) &&
          "Unexpected element type");
 
+  // Custom legalize v8i8->v8i64 on CPUs without avx512bw.
+  if (InVT == MVT::v8i8) {
+    if (!ExperimentalVectorWideningLegalization || VT != MVT::v8i64)
+      return SDValue();
+
+    In = DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(Op),
+                     MVT::v16i8, In, DAG.getUNDEF(MVT::v8i8));
+    // FIXME: This should be ANY_EXTEND_VECTOR_INREG for ANY_EXTEND input.
+    return DAG.getNode(ISD::ZERO_EXTEND_VECTOR_INREG, dl, VT, In);
+  }
+
   if (Subtarget.hasInt256())
     return Op;
 
@@ -20089,6 +20115,16 @@ static SDValue LowerSIGN_EXTEND(SDValue Op, const X86Subtarget &Subtarget,
           InVT.getVectorElementType() == MVT::i32) &&
          "Unexpected element type");
 
+  // Custom legalize v8i8->v8i64 on CPUs without avx512bw.
+  if (InVT == MVT::v8i8) {
+    if (!ExperimentalVectorWideningLegalization || VT != MVT::v8i64)
+      return SDValue();
+
+    In = DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(Op),
+                     MVT::v16i8, In, DAG.getUNDEF(MVT::v8i8));
+    return DAG.getNode(ISD::SIGN_EXTEND_VECTOR_INREG, dl, VT, In);
+  }
+
   if (Subtarget.hasInt256())
     return Op;
 
@@ -26307,6 +26343,41 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N,
     }
     return;
   }
+  case ISD::SIGN_EXTEND:
+  case ISD::ZERO_EXTEND: {
+    if (!ExperimentalVectorWideningLegalization)
+      return;
+
+    EVT VT = N->getValueType(0);
+    assert((VT == MVT::v16i32 || VT == MVT::v8i64) && "Unexpected VT!");
+    SDValue In = N->getOperand(0);
+    EVT InVT = In.getValueType();
+    if (InVT.is128BitVector()) {
+      // Perform custom splitting instead of the two stage extend we would get
+      // by default.
+      EVT LoVT, HiVT;
+      std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
+      assert(isTypeLegal(LoVT) && "Split VT not legal?");
+
+      bool IsSigned = N->getOpcode() == ISD::SIGN_EXTEND;
+
+      SDValue Lo = getExtendInVec(IsSigned, dl, LoVT, In, DAG);
+
+      // We need to shift the input over by half the number of elements.
+      unsigned NumElts = InVT.getVectorNumElements();
+      unsigned HalfNumElts = NumElts / 2;
+      SmallVector<int, 16> ShufMask(NumElts, SM_SentinelUndef);
+      for (unsigned i = 0; i != HalfNumElts; ++i)
+        ShufMask[i] = i + HalfNumElts;
+
+      SDValue Hi = DAG.getVectorShuffle(InVT, dl, In, In, ShufMask);
+      Hi = getExtendInVec(IsSigned, dl, HiVT, Hi, DAG);
+
+      SDValue Res = DAG.getNode(ISD::CONCAT_VECTORS, dl, VT, Lo, Hi);
+      Results.push_back(Res);
+    }
+    return;
+  }
   case ISD::FP_TO_SINT:
   case ISD::FP_TO_UINT: {
     bool IsSigned = N->getOpcode() == ISD::FP_TO_SINT;
@@ -38715,6 +38786,9 @@ combineToExtendBoolVectorInReg(SDNode *N, SelectionDAG &DAG,
 static SDValue combineToExtendVectorInReg(SDNode *N, SelectionDAG &DAG,
                                           TargetLowering::DAGCombinerInfo &DCI,
                                           const X86Subtarget &Subtarget) {
+  if (ExperimentalVectorWideningLegalization)
+    return SDValue();
+
   unsigned Opcode = N->getOpcode();
   if (Opcode != ISD::SIGN_EXTEND && Opcode != ISD::ZERO_EXTEND)
     return SDValue();
diff --git a/llvm/test/CodeGen/X86/vec_int_to_fp-widen.ll b/llvm/test/CodeGen/X86/vec_int_to_fp-widen.ll
index b87a15442db..13fe0f6220e 100644
--- a/llvm/test/CodeGen/X86/vec_int_to_fp-widen.ll
+++ b/llvm/test/CodeGen/X86/vec_int_to_fp-widen.ll
@@ -5435,13 +5435,14 @@ define <8 x float> @uitofp_load_8i8_to_8f32(<8 x i8> *%a) {
 define void @aggregate_sitofp_8i16_to_8f32(%Arguments* nocapture readonly %a0) {
 ; SSE2-LABEL: aggregate_sitofp_8i16_to_8f32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movq 24(%rdi), %rax
-; SSE2-NEXT:    movdqu 8(%rdi), %xmm0
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
+; SSE2-NEXT:    psrad $16, %xmm0
+; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3]
 ; SSE2-NEXT:    psrad $16, %xmm1
+; SSE2-NEXT:    movq 24(%rdi), %rax
 ; SSE2-NEXT:    cvtdq2ps %xmm1, %xmm1
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    psrad $16, %xmm0
 ; SSE2-NEXT:    cvtdq2ps %xmm0, %xmm0
 ; SSE2-NEXT:    movaps %xmm0, 16(%rax)
 ; SSE2-NEXT:    movaps %xmm1, (%rax)
@@ -5450,14 +5451,12 @@ define void @aggregate_sitofp_8i16_to_8f32(%Arguments* nocapture readonly %a0) {
 ; SSE41-LABEL: aggregate_sitofp_8i16_to_8f32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movq 24(%rdi), %rax
-; SSE41-NEXT:    movdqu 8(%rdi), %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovsxwd %xmm1, %xmm1
+; SSE41-NEXT:    pmovsxwd 16(%rdi), %xmm0
+; SSE41-NEXT:    pmovsxwd 8(%rdi), %xmm1
 ; SSE41-NEXT:    cvtdq2ps %xmm1, %xmm1
-; SSE41-NEXT:    pmovsxwd %xmm0, %xmm0
 ; SSE41-NEXT:    cvtdq2ps %xmm0, %xmm0
-; SSE41-NEXT:    movaps %xmm0, (%rax)
-; SSE41-NEXT:    movaps %xmm1, 16(%rax)
+; SSE41-NEXT:    movaps %xmm0, 16(%rax)
+; SSE41-NEXT:    movaps %xmm1, (%rax)
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: aggregate_sitofp_8i16_to_8f32:
diff --git a/llvm/test/CodeGen/X86/vector-sext-widen.ll b/llvm/test/CodeGen/X86/vector-sext-widen.ll
index d0969c6607e..44d9ee5b2f9 100644
--- a/llvm/test/CodeGen/X86/vector-sext-widen.ll
+++ b/llvm/test/CodeGen/X86/vector-sext-widen.ll
@@ -495,26 +495,24 @@ entry:
 define <8 x i64> @sext_16i8_to_8i64(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: sext_16i8_to_8i64:
 ; SSE2:       # %bb.0: # %entry
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
 ; SSE2-NEXT:    movdqa %xmm4, %xmm1
 ; SSE2-NEXT:    psrad $31, %xmm1
 ; SSE2-NEXT:    psrad $24, %xmm4
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrad $31, %xmm0
-; SSE2-NEXT:    psrad $24, %xmm1
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
-; SSE2-NEXT:    psrad $31, %xmm0
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    psrad $31, %xmm1
 ; SSE2-NEXT:    psrad $24, %xmm2
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3]
+; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    psrad $31, %xmm3
+; SSE2-NEXT:    psrad $24, %xmm1
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
 ; SSE2-NEXT:    movdqa %xmm3, %xmm0
 ; SSE2-NEXT:    psrad $31, %xmm0
@@ -525,26 +523,24 @@ define <8 x i64> @sext_16i8_to_8i64(<16 x i8> %A) nounwind uwtable readnone ssp
 ;
 ; SSSE3-LABEL: sext_16i8_to_8i64:
 ; SSSE3:       # %bb.0: # %entry
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,2,3]
 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
 ; SSSE3-NEXT:    movdqa %xmm4, %xmm1
 ; SSSE3-NEXT:    psrad $31, %xmm1
 ; SSSE3-NEXT:    psrad $24, %xmm4
 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSSE3-NEXT:    psrad $31, %xmm0
-; SSSE3-NEXT:    psrad $24, %xmm1
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; SSSE3-NEXT:    movdqa %xmm2, %xmm0
-; SSSE3-NEXT:    psrad $31, %xmm0
+; SSSE3-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; SSSE3-NEXT:    movdqa %xmm2, %xmm1
+; SSSE3-NEXT:    psrad $31, %xmm1
 ; SSSE3-NEXT:    psrad $24, %xmm2
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3]
+; SSSE3-NEXT:    movdqa %xmm1, %xmm3
+; SSSE3-NEXT:    psrad $31, %xmm3
+; SSSE3-NEXT:    psrad $24, %xmm1
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
 ; SSSE3-NEXT:    movdqa %xmm3, %xmm0
 ; SSSE3-NEXT:    psrad $31, %xmm0
@@ -912,8 +908,8 @@ define <8 x i64> @sext_8i16_to_8i64(<8 x i16> %A) nounwind uwtable readnone ssp
 ; SSE2-NEXT:    psrad $31, %xmm3
 ; SSE2-NEXT:    psrad $16, %xmm1
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]
 ; SSE2-NEXT:    movdqa %xmm3, %xmm0
 ; SSE2-NEXT:    psrad $31, %xmm0
 ; SSE2-NEXT:    psrad $16, %xmm3
@@ -938,8 +934,8 @@ define <8 x i64> @sext_8i16_to_8i64(<8 x i16> %A) nounwind uwtable readnone ssp
 ; SSSE3-NEXT:    psrad $31, %xmm3
 ; SSSE3-NEXT:    psrad $16, %xmm1
 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSSE3-NEXT:    pshuflw {{.*#+}} xmm3 = xmm0[0,2,2,3,4,5,6,7]
+; SSSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]
 ; SSSE3-NEXT:    movdqa %xmm3, %xmm0
 ; SSSE3-NEXT:    psrad $31, %xmm0
 ; SSSE3-NEXT:    psrad $16, %xmm3
diff --git a/llvm/test/CodeGen/X86/vector-zext-widen.ll b/llvm/test/CodeGen/X86/vector-zext-widen.ll
index bc8896f0116..2eafc7cdaf0 100644
--- a/llvm/test/CodeGen/X86/vector-zext-widen.ll
+++ b/llvm/test/CodeGen/X86/vector-zext-widen.ll
@@ -123,15 +123,14 @@ define <32 x i16> @zext_32i8_to_32i16(<32 x i8> %A) {
 ;
 ; AVX1-LABEL: zext_32i8_to_32i16:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm2
+; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
 ; AVX1-NEXT:    vmovaps %ymm2, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -398,16 +397,15 @@ entry:
 define <8 x i64> @zext_16i8_to_8i64(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: zext_16i8_to_8i64:
 ; SSE2:       # %bb.0: # %entry
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,2,3]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
 ; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
 ; SSE2-NEXT:    movdqa %xmm3, %xmm2
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
 ; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
@@ -415,15 +413,13 @@ define <8 x i64> @zext_16i8_to_8i64(<16 x i8> %A) nounwind uwtable readnone ssp
 ;
 ; SSSE3-LABEL: zext_16i8_to_8i64:
 ; SSSE3:       # %bb.0: # %entry
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [0,128,128,128,128,128,128,128,1,128,128,128,128,128,128,128]
-; SSSE3-NEXT:    pshufb %xmm4, %xmm0
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [2,128,128,128,128,128,128,128,3,128,128,128,128,128,128,128]
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]
-; SSSE3-NEXT:    pshufb %xmm5, %xmm1
+; SSSE3-NEXT:    movdqa %xmm0, %xmm3
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    movdqa %xmm3, %xmm1
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[2],zero,zero,zero,zero,zero,zero,zero,xmm1[3],zero,zero,zero,zero,zero,zero,zero
 ; SSSE3-NEXT:    movdqa %xmm3, %xmm2
-; SSSE3-NEXT:    pshufb %xmm4, %xmm2
-; SSSE3-NEXT:    pshufb %xmm5, %xmm3
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm2 = xmm2[4],zero,zero,zero,zero,zero,zero,zero,xmm2[5],zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm3 = xmm3[6],zero,zero,zero,zero,zero,zero,zero,xmm3[7],zero,zero,zero,zero,zero,zero,zero
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: zext_16i8_to_8i64:
@@ -585,15 +581,14 @@ define <16 x i32> @zext_16i16_to_16i32(<16 x i16> %A) nounwind uwtable readnone
 ;
 ; AVX1-LABEL: zext_16i16_to_16i32:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm2
+; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
 ; AVX1-NEXT:    vmovaps %ymm2, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -884,15 +879,14 @@ define <8 x i64> @zext_8i32_to_8i64(<8 x i32> %A) nounwind uwtable readnone ssp
 ;
 ; AVX1-LABEL: zext_8i32_to_8i64:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm2
+; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
 ; AVX1-NEXT:    vmovaps %ymm2, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -1162,16 +1156,15 @@ entry:
 define <8 x i64> @load_zext_8i8_to_8i64(<8 x i8> *%ptr) {
 ; SSE2-LABEL: load_zext_8i8_to_8i64:
 ; SSE2:       # %bb.0: # %entry
-; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
+; SSE2-NEXT:    movq {{.*#+}} xmm3 = mem[0],zero
 ; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
 ; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
 ; SSE2-NEXT:    movdqa %xmm3, %xmm2
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
 ; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
@@ -1179,16 +1172,14 @@ define <8 x i64> @load_zext_8i8_to_8i64(<8 x i8> *%ptr) {
 ;
 ; SSSE3-LABEL: load_zext_8i8_to_8i64:
 ; SSSE3:       # %bb.0: # %entry
-; SSSE3-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [0,128,128,128,128,128,128,128,1,128,128,128,128,128,128,128]
-; SSSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb %xmm4, %xmm0
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [2,128,128,128,128,128,128,128,3,128,128,128,128,128,128,128]
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]
-; SSSE3-NEXT:    pshufb %xmm5, %xmm1
+; SSSE3-NEXT:    movq {{.*#+}} xmm3 = mem[0],zero
+; SSSE3-NEXT:    movdqa %xmm3, %xmm0
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    movdqa %xmm3, %xmm1
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[2],zero,zero,zero,zero,zero,zero,zero,xmm1[3],zero,zero,zero,zero,zero,zero,zero
 ; SSSE3-NEXT:    movdqa %xmm3, %xmm2
-; SSSE3-NEXT:    pshufb %xmm4, %xmm2
-; SSSE3-NEXT:    pshufb %xmm5, %xmm3
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm2 = xmm2[4],zero,zero,zero,zero,zero,zero,zero,xmm2[5],zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm3 = xmm3[6],zero,zero,zero,zero,zero,zero,zero,xmm3[7],zero,zero,zero,zero,zero,zero,zero
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: load_zext_8i8_to_8i64:
@@ -2229,11 +2220,11 @@ define <32 x i32> @zext_32i8_to_32i32(<32 x i8> %x) {
 ; AVX2-LABEL: zext_32i8_to_32i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm4 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[3,1,2,3]
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
+; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm2 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm3[2,3,0,1]
 ; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
 ; AVX2-NEXT:    vmovdqa %ymm4, %ymm0
 ; AVX2-NEXT:    retq