[llvm] 940d7cd - [X86] SimplifyDemandedVectorElts - adjust X86ISD::ANDNP demanded elts based off constant masks
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Fri Mar 4 05:53:57 PST 2022
Author: Simon Pilgrim
Date: 2022-03-04T13:40:56Z
New Revision: 940d7cd59f88d78af443091e75a33a39ce098fbe
URL: https://github.com/llvm/llvm-project/commit/940d7cd59f88d78af443091e75a33a39ce098fbe
DIFF: https://github.com/llvm/llvm-project/commit/940d7cd59f88d78af443091e75a33a39ce098fbe.diff
LOG: [X86] SimplifyDemandedVectorElts - adjust X86ISD::ANDNP demanded elts based off constant masks
Similar to what we already do in combineAndnp, if either operand is a constant then we can improve the demanded elts/bits.
Added:
Modified:
llvm/lib/Target/X86/X86ISelLowering.cpp
llvm/test/CodeGen/X86/vector-fshl-512.ll
llvm/test/CodeGen/X86/vector-fshr-512.ll
llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll
llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll
llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 26c64bd71777c..7cdb4f754ba82 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -40540,21 +40540,43 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
SDValue LHS = Op.getOperand(0);
SDValue RHS = Op.getOperand(1);
- APInt RHSUndef, RHSZero;
- if (SimplifyDemandedVectorElts(RHS, DemandedElts, RHSUndef, RHSZero, TLO,
- Depth + 1))
- return true;
+ auto GetDemandedMasks = [&](SDValue Op, bool Invert = false) {
+ APInt UndefElts;
+ SmallVector<APInt> EltBits;
+ int NumElts = VT.getVectorNumElements();
+ int EltSizeInBits = VT.getScalarSizeInBits();
+ APInt OpBits = APInt::getAllOnes(EltSizeInBits);
+ APInt OpElts = DemandedElts;
+ if (getTargetConstantBitsFromNode(Op, EltSizeInBits, UndefElts,
+ EltBits)) {
+ OpBits.clearAllBits();
+ OpElts.clearAllBits();
+ for (int I = 0; I != NumElts; ++I)
+ if (DemandedElts[I] && ((Invert && !EltBits[I].isAllOnes()) ||
+ (!Invert && !EltBits[I].isZero()))) {
+ OpBits |= Invert ? ~EltBits[I] : EltBits[I];
+ OpElts.setBit(I);
+ }
+ }
+ return std::make_pair(OpBits, OpElts);
+ };
+ std::pair<APInt, APInt> DemandLHS = GetDemandedMasks(RHS);
+ std::pair<APInt, APInt> DemandRHS = GetDemandedMasks(LHS, true);
APInt LHSUndef, LHSZero;
- if (SimplifyDemandedVectorElts(LHS, DemandedElts, LHSUndef, LHSZero, TLO,
- Depth + 1))
+ APInt RHSUndef, RHSZero;
+ if (SimplifyDemandedVectorElts(LHS, DemandLHS.second, LHSUndef, LHSZero,
+ TLO, Depth + 1))
+ return true;
+ if (SimplifyDemandedVectorElts(RHS, DemandRHS.second, RHSUndef, RHSZero,
+ TLO, Depth + 1))
return true;
if (!DemandedElts.isAllOnes()) {
- SDValue NewLHS = SimplifyMultipleUseDemandedVectorElts(
- LHS, DemandedElts, TLO.DAG, Depth + 1);
- SDValue NewRHS = SimplifyMultipleUseDemandedVectorElts(
- RHS, DemandedElts, TLO.DAG, Depth + 1);
+ SDValue NewLHS = SimplifyMultipleUseDemandedBits(
+ LHS, DemandLHS.first, DemandLHS.second, TLO.DAG, Depth + 1);
+ SDValue NewRHS = SimplifyMultipleUseDemandedBits(
+ RHS, DemandRHS.first, DemandRHS.second, TLO.DAG, Depth + 1);
if (NewLHS || NewRHS) {
NewLHS = NewLHS ? NewLHS : LHS;
NewRHS = NewRHS ? NewRHS : RHS;
diff --git a/llvm/test/CodeGen/X86/vector-fshl-512.ll b/llvm/test/CodeGen/X86/vector-fshl-512.ll
index e902e403d33f1..6818d6fe379d7 100644
--- a/llvm/test/CodeGen/X86/vector-fshl-512.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-512.ll
@@ -544,17 +544,15 @@ define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i
define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
; AVX512F-LABEL: splatvar_funnnel_v32i16:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
-; AVX512F-NEXT: vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT: vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm5
-; AVX512F-NEXT: vpsllw %xmm4, %ymm5, %ymm5
+; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm4
+; AVX512F-NEXT: vpsllw %xmm4, %ymm3, %ymm3
; AVX512F-NEXT: vpsllw %xmm4, %ymm0, %ymm0
-; AVX512F-NEXT: vinserti64x4 $1, %ymm5, %zmm0, %zmm0
-; AVX512F-NEXT: vpandn %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT: vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512F-NEXT: vpsrlw $1, %ymm3, %ymm3
+; AVX512F-NEXT: vpandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT: vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
; AVX512F-NEXT: vpsrlw %xmm2, %ymm3, %ymm3
; AVX512F-NEXT: vpsrlw $1, %ymm1, %ymm1
; AVX512F-NEXT: vpsrlw %xmm2, %ymm1, %ymm1
@@ -564,17 +562,15 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i
;
; AVX512VL-LABEL: splatvar_funnnel_v32i16:
; AVX512VL: # %bb.0:
-; AVX512VL-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT: vpand %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm5
-; AVX512VL-NEXT: vpsllw %xmm4, %ymm5, %ymm5
+; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
+; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm4
+; AVX512VL-NEXT: vpsllw %xmm4, %ymm3, %ymm3
; AVX512VL-NEXT: vpsllw %xmm4, %ymm0, %ymm0
-; AVX512VL-NEXT: vinserti64x4 $1, %ymm5, %zmm0, %zmm0
-; AVX512VL-NEXT: vpandn %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512VL-NEXT: vpsrlw $1, %ymm3, %ymm3
+; AVX512VL-NEXT: vpandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
; AVX512VL-NEXT: vpsrlw %xmm2, %ymm3, %ymm3
; AVX512VL-NEXT: vpsrlw $1, %ymm1, %ymm1
; AVX512VL-NEXT: vpsrlw %xmm2, %ymm1, %ymm1
diff --git a/llvm/test/CodeGen/X86/vector-fshr-512.ll b/llvm/test/CodeGen/X86/vector-fshr-512.ll
index cb5cfa3fb140b..176af60d68706 100644
--- a/llvm/test/CodeGen/X86/vector-fshr-512.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-512.ll
@@ -547,8 +547,8 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i
; AVX512F-LABEL: splatvar_funnnel_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
-; AVX512F-NEXT: vpand %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT: vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT: vpand %xmm3, %xmm2, %xmm4
+; AVX512F-NEXT: vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm5
; AVX512F-NEXT: vpsrlw %xmm4, %ymm5, %ymm5
; AVX512F-NEXT: vpsrlw %xmm4, %ymm1, %ymm1
@@ -567,8 +567,8 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i
; AVX512VL-LABEL: splatvar_funnnel_v32i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT: vpand %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT: vpand %xmm3, %xmm2, %xmm4
+; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm5
; AVX512VL-NEXT: vpsrlw %xmm4, %ymm5, %ymm5
; AVX512VL-NEXT: vpsrlw %xmm4, %ymm1, %ymm1
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll
index 137d846d0adf9..4d7be08e28190 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll
@@ -320,104 +320,103 @@ define void @vf8(<40 x i16>* %in.vec, <8 x i16>* %out.vec0, <8 x i16>* %out.vec1
; SSE-LABEL: vf8:
; SSE: # %bb.0:
; SSE-NEXT: movdqa 64(%rdi), %xmm11
-; SSE-NEXT: movdqa (%rdi), %xmm12
+; SSE-NEXT: movdqa (%rdi), %xmm14
; SSE-NEXT: movdqa 16(%rdi), %xmm15
-; SSE-NEXT: movdqa 32(%rdi), %xmm2
-; SSE-NEXT: movdqa 48(%rdi), %xmm14
-; SSE-NEXT: movdqa {{.*#+}} xmm4 = [65535,65535,65535,65535,0,65535,65535,65535]
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm14[0,1,0,3]
-; SSE-NEXT: pand %xmm4, %xmm5
-; SSE-NEXT: pandn %xmm2, %xmm4
-; SSE-NEXT: por %xmm5, %xmm4
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm15[3,1,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[2,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm12[0,2,2,3]
+; SSE-NEXT: movdqa 32(%rdi), %xmm8
+; SSE-NEXT: movdqa 48(%rdi), %xmm12
+; SSE-NEXT: movdqa {{.*#+}} xmm3 = [65535,65535,65535,65535,0,65535,65535,65535]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm12[0,1,0,3]
+; SSE-NEXT: pand %xmm3, %xmm4
+; SSE-NEXT: pandn %xmm8, %xmm3
+; SSE-NEXT: por %xmm4, %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm15[3,1,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[2,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm14[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm7[0,3,2,3,4,5,6,7]
-; SSE-NEXT: punpckldq {{.*#+}} xmm7 = xmm7[0],xmm5[0],xmm7[1],xmm5[1]
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm4[2,3]
-; SSE-NEXT: movaps {{.*#+}} xmm4 = [65535,65535,65535,65535,65535,65535,65535,0]
-; SSE-NEXT: andps %xmm4, %xmm7
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm11[0,1,0,1]
-; SSE-NEXT: movaps %xmm4, %xmm10
-; SSE-NEXT: pandn %xmm5, %xmm10
-; SSE-NEXT: por %xmm7, %xmm10
-; SSE-NEXT: movdqa %xmm15, %xmm5
-; SSE-NEXT: psrlq $48, %xmm5
-; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm12[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm7[1,2,2,3,4,5,6,7]
-; SSE-NEXT: punpckldq {{.*#+}} xmm6 = xmm6[0],xmm5[0],xmm6[1],xmm5[1]
+; SSE-NEXT: punpckldq {{.*#+}} xmm7 = xmm7[0],xmm4[0],xmm7[1],xmm4[1]
+; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm3[2,3]
+; SSE-NEXT: movaps {{.*#+}} xmm10 = [65535,65535,65535,65535,65535,65535,65535,0]
+; SSE-NEXT: andps %xmm10, %xmm7
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm11[0,1,0,1]
+; SSE-NEXT: movaps %xmm10, %xmm9
+; SSE-NEXT: pandn %xmm4, %xmm9
+; SSE-NEXT: por %xmm7, %xmm9
+; SSE-NEXT: movdqa %xmm15, %xmm4
+; SSE-NEXT: psrlq $48, %xmm4
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm14[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm7[1,2,2,3,4,5,6,7]
+; SSE-NEXT: punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1]
; SSE-NEXT: movdqa {{.*#+}} xmm7 = [0,0,0,65535,65535,65535,65535,65535]
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm14[1,3,2,3]
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm2[0,2,2,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,6,4,7]
-; SSE-NEXT: pand %xmm7, %xmm0
-; SSE-NEXT: pandn %xmm6, %xmm7
-; SSE-NEXT: por %xmm0, %xmm7
-; SSE-NEXT: pand %xmm4, %xmm7
-; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm11[0,1,2,0]
-; SSE-NEXT: pshufd {{.*#+}} xmm9 = xmm11[0,1,0,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm12[1,3,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm8[0,2,2,3]
+; SSE-NEXT: punpckldq {{.*#+}} xmm6 = xmm6[0],xmm4[0],xmm6[1],xmm4[1]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm6[0,1,2,3,7,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,2,1]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,0,0,0,4,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,5,6,4,7]
+; SSE-NEXT: pand %xmm7, %xmm4
+; SSE-NEXT: pandn %xmm5, %xmm7
+; SSE-NEXT: por %xmm4, %xmm7
+; SSE-NEXT: pand %xmm10, %xmm7
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm11[0,1,2,0]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm11[0,1,0,3]
+; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshufd {{.*#+}} xmm13 = xmm11[0,1,1,3]
; SSE-NEXT: psllq $48, %xmm11
-; SSE-NEXT: pandn %xmm11, %xmm4
-; SSE-NEXT: por %xmm7, %xmm4
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm2[0,0,2,3]
-; SSE-NEXT: movdqa {{.*#+}} xmm6 = [65535,65535,65535,0,0,0,65535,65535]
-; SSE-NEXT: movdqa %xmm6, %xmm7
-; SSE-NEXT: pandn %xmm5, %xmm7
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm12[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm5 = xmm5[2],xmm15[2],xmm5[3],xmm15[3]
-; SSE-NEXT: pand %xmm6, %xmm5
-; SSE-NEXT: por %xmm7, %xmm5
-; SSE-NEXT: movdqa %xmm14, %xmm7
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[1,0],xmm2[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[2,0],xmm2[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,6,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm8[0,1,2,3,4,5,6,5]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm7[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,1],xmm0[2,0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm2[2,2,2,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm6, %xmm7
-; SSE-NEXT: pandn %xmm0, %xmm7
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm12[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm15[0,2,2,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm3[0,3,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[1,0,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm0
-; SSE-NEXT: por %xmm7, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm14[0,2,2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm14 = xmm14[2,0],xmm2[3,0]
-; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm2[1,1,1,1]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm14[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,4,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm9[0,1,2,3,4,5,5,6]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm2[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm3[0,1,0,3,4,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm13[0,1,2,3,4,5,4,7]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm7[0],xmm1[1],xmm7[1]
-; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[2,0],xmm15[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm15 = xmm15[0,1],xmm12[0,2]
+; SSE-NEXT: pandn %xmm11, %xmm10
+; SSE-NEXT: por %xmm7, %xmm10
+; SSE-NEXT: movdqa %xmm12, %xmm5
+; SSE-NEXT: pshufd {{.*#+}} xmm11 = xmm12[0,2,2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[1,0],xmm8[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[2,0],xmm8[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[0,0,1,3]
+; SSE-NEXT: movdqa {{.*#+}} xmm2 = [65535,65535,65535,0,0,0,65535,65535]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm14[0,1,1,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm15[2],xmm4[3],xmm15[3]
+; SSE-NEXT: pand %xmm2, %xmm4
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm14[1,1,1,1]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm15[0,2,2,3]
+; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1]
+; SSE-NEXT: movdqa %xmm2, %xmm6
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,3,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[1,0,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm2, %xmm1
+; SSE-NEXT: shufps {{.*#+}} xmm14 = xmm14[2,0],xmm15[3,0]
+; SSE-NEXT: shufps {{.*#+}} xmm15 = xmm15[0,1],xmm14[0,2]
+; SSE-NEXT: movdqa %xmm2, %xmm7
; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm15[0,1,2,3,4,6,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[2,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm3
-; SSE-NEXT: pandn %xmm1, %xmm6
-; SSE-NEXT: por %xmm3, %xmm6
-; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,1],xmm2[2,0]
-; SSE-NEXT: movdqa %xmm10, (%rsi)
-; SSE-NEXT: movdqa %xmm4, (%rdx)
-; SSE-NEXT: movaps %xmm5, (%rcx)
-; SSE-NEXT: movaps %xmm0, (%r8)
-; SSE-NEXT: movaps %xmm6, (%r9)
+; SSE-NEXT: pand %xmm2, %xmm3
+; SSE-NEXT: pandn %xmm12, %xmm2
+; SSE-NEXT: por %xmm2, %xmm4
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm12[0,1,2,3,6,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,6,5]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm2[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm0[2,0]
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm8[2,2,2,2,4,5,6,7]
+; SSE-NEXT: pandn %xmm0, %xmm6
+; SSE-NEXT: por %xmm6, %xmm1
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[2,0],xmm8[3,0]
+; SSE-NEXT: pandn %xmm8, %xmm7
+; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,1],xmm5[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm8[0,1,2,3,7,4,6,7]
+; SSE-NEXT: pshufhw $148, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = mem[0,1,2,3,4,5,5,6]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm0[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,0]
+; SSE-NEXT: por %xmm7, %xmm3
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm11[0,1,0,3,4,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm13[0,1,2,3,4,5,4,7]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm0[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm2[2,0]
+; SSE-NEXT: movdqa %xmm9, (%rsi)
+; SSE-NEXT: movdqa %xmm10, (%rdx)
+; SSE-NEXT: movaps %xmm4, (%rcx)
+; SSE-NEXT: movaps %xmm1, (%r8)
+; SSE-NEXT: movaps %xmm3, (%r9)
; SSE-NEXT: retq
;
; AVX1-LABEL: vf8:
@@ -660,56 +659,55 @@ define void @vf16(<80 x i16>* %in.vec, <16 x i16>* %out.vec0, <16 x i16>* %out.v
; SSE-LABEL: vf16:
; SSE: # %bb.0:
; SSE-NEXT: subq $24, %rsp
-; SSE-NEXT: movdqa 144(%rdi), %xmm13
-; SSE-NEXT: movdqa 80(%rdi), %xmm7
-; SSE-NEXT: movdqa 96(%rdi), %xmm14
-; SSE-NEXT: movdqa 128(%rdi), %xmm5
-; SSE-NEXT: movdqa 112(%rdi), %xmm11
-; SSE-NEXT: movdqa 64(%rdi), %xmm4
+; SSE-NEXT: movdqa 144(%rdi), %xmm12
+; SSE-NEXT: movdqa 80(%rdi), %xmm5
+; SSE-NEXT: movdqa 96(%rdi), %xmm10
+; SSE-NEXT: movdqa 128(%rdi), %xmm8
+; SSE-NEXT: movdqa 112(%rdi), %xmm14
+; SSE-NEXT: movdqa 64(%rdi), %xmm7
; SSE-NEXT: movdqa (%rdi), %xmm6
-; SSE-NEXT: movdqa 16(%rdi), %xmm12
-; SSE-NEXT: movdqa 32(%rdi), %xmm9
+; SSE-NEXT: movdqa 16(%rdi), %xmm9
+; SSE-NEXT: movdqa 32(%rdi), %xmm4
; SSE-NEXT: movdqa 48(%rdi), %xmm3
; SSE-NEXT: movdqa {{.*#+}} xmm0 = [65535,65535,65535,65535,0,65535,65535,65535]
; SSE-NEXT: movdqa %xmm0, %xmm1
-; SSE-NEXT: pandn %xmm9, %xmm1
+; SSE-NEXT: pandn %xmm4, %xmm1
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,1,0,3]
-; SSE-NEXT: movdqa %xmm3, %xmm8
+; SSE-NEXT: movdqa %xmm3, %xmm11
; SSE-NEXT: pand %xmm0, %xmm2
; SSE-NEXT: por %xmm1, %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm12[3,1,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[3,1,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm6[0,2,2,3]
-; SSE-NEXT: movdqa %xmm6, %xmm15
+; SSE-NEXT: movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,3,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm2[2,3]
-; SSE-NEXT: movaps {{.*#+}} xmm10 = [65535,65535,65535,65535,65535,65535,65535,0]
-; SSE-NEXT: andps %xmm10, %xmm3
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm4[0,1,0,1]
-; SSE-NEXT: movaps %xmm10, %xmm2
+; SSE-NEXT: movaps {{.*#+}} xmm13 = [65535,65535,65535,65535,65535,65535,65535,0]
+; SSE-NEXT: andps %xmm13, %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,1,0,1]
+; SSE-NEXT: movaps %xmm13, %xmm2
; SSE-NEXT: pandn %xmm1, %xmm2
; SSE-NEXT: por %xmm3, %xmm2
; SSE-NEXT: movdqa %xmm2, (%rsp) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm5[0,1,0,3]
-; SSE-NEXT: movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm8[0,1,0,3]
; SSE-NEXT: pand %xmm0, %xmm1
-; SSE-NEXT: pandn %xmm11, %xmm0
+; SSE-NEXT: pandn %xmm14, %xmm0
; SSE-NEXT: por %xmm1, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm14[3,1,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm10[3,1,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm7[0,2,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm5[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,3,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm0[2,3]
-; SSE-NEXT: andps %xmm10, %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm13[0,1,0,1]
-; SSE-NEXT: movdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movaps %xmm10, %xmm1
+; SSE-NEXT: andps %xmm13, %xmm2
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm12[0,1,0,1]
+; SSE-NEXT: movdqa %xmm12, %xmm15
+; SSE-NEXT: movaps %xmm13, %xmm1
; SSE-NEXT: andnps %xmm0, %xmm1
; SSE-NEXT: orps %xmm2, %xmm1
; SSE-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm12, %xmm0
+; SSE-NEXT: movdqa %xmm9, %xmm0
; SSE-NEXT: psrlq $48, %xmm0
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[1,2,2,3,4,5,6,7]
@@ -717,10 +715,9 @@ define void @vf16(<80 x i16>* %in.vec, <16 x i16>* %out.vec0, <16 x i16>* %out.v
; SSE-NEXT: movdqa {{.*#+}} xmm0 = [0,0,0,65535,65535,65535,65535,65535]
; SSE-NEXT: movdqa %xmm0, %xmm2
; SSE-NEXT: pandn %xmm1, %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm8[1,3,2,3]
-; SSE-NEXT: movdqa %xmm8, %xmm6
-; SSE-NEXT: movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm9[0,2,2,3]
+; SSE-NEXT: movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm11[1,3,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm3[0,1,2,3,7,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
@@ -728,20 +725,26 @@ define void @vf16(<80 x i16>* %in.vec, <16 x i16>* %out.vec0, <16 x i16>* %out.v
; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,6,4,7]
; SSE-NEXT: pand %xmm0, %xmm1
; SSE-NEXT: por %xmm2, %xmm1
-; SSE-NEXT: movdqa %xmm4, %xmm2
+; SSE-NEXT: movdqa %xmm7, %xmm2
+; SSE-NEXT: movdqa %xmm7, %xmm12
+; SSE-NEXT: movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: psllq $48, %xmm2
-; SSE-NEXT: movaps %xmm10, %xmm3
+; SSE-NEXT: movaps %xmm13, %xmm3
; SSE-NEXT: andnps %xmm2, %xmm3
-; SSE-NEXT: pand %xmm10, %xmm1
+; SSE-NEXT: pand %xmm13, %xmm1
; SSE-NEXT: orps %xmm1, %xmm3
; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm14, %xmm1
+; SSE-NEXT: movdqa %xmm10, %xmm1
; SSE-NEXT: psrlq $48, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm7[0,3,2,3]
+; SSE-NEXT: movdqa %xmm5, %xmm6
+; SSE-NEXT: movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm5[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[1,2,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm5[1,3,2,3]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm11[0,2,2,3]
+; SSE-NEXT: movdqa %xmm8, %xmm5
+; SSE-NEXT: movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm8[1,3,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm14[0,2,2,3]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm3[0,1,2,3,7,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
@@ -750,144 +753,134 @@ define void @vf16(<80 x i16>* %in.vec, <16 x i16>* %out.vec0, <16 x i16>* %out.v
; SSE-NEXT: pand %xmm0, %xmm1
; SSE-NEXT: pandn %xmm2, %xmm0
; SSE-NEXT: por %xmm1, %xmm0
-; SSE-NEXT: pand %xmm10, %xmm0
-; SSE-NEXT: movdqa %xmm13, %xmm1
-; SSE-NEXT: psllq $48, %xmm1
-; SSE-NEXT: pandn %xmm1, %xmm10
-; SSE-NEXT: por %xmm0, %xmm10
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[0,0,2,3]
-; SSE-NEXT: movdqa {{.*#+}} xmm5 = [65535,65535,65535,0,0,0,65535,65535]
-; SSE-NEXT: movdqa %xmm5, %xmm2
-; SSE-NEXT: pandn %xmm1, %xmm2
+; SSE-NEXT: pand %xmm13, %xmm0
; SSE-NEXT: movdqa %xmm15, %xmm8
; SSE-NEXT: movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm15[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm13 = xmm1[0,1,2,3,4,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm13 = xmm13[2],xmm12[2],xmm13[3],xmm12[3]
-; SSE-NEXT: pand %xmm5, %xmm13
-; SSE-NEXT: por %xmm2, %xmm13
-; SSE-NEXT: movdqa %xmm6, %xmm1
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm9[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm9[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm4[0,1,2,0]
-; SSE-NEXT: movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm15, %xmm1
+; SSE-NEXT: psllq $48, %xmm1
+; SSE-NEXT: pandn %xmm1, %xmm13
+; SSE-NEXT: por %xmm0, %xmm13
+; SSE-NEXT: movdqa %xmm11, %xmm0
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0],xmm4[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm4[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,1,3]
+; SSE-NEXT: movaps {{.*#+}} xmm1 = [65535,65535,65535,0,0,0,65535,65535]
+; SSE-NEXT: movaps %xmm1, %xmm2
+; SSE-NEXT: andnps %xmm0, %xmm2
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm7[0,1,1,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm15 = xmm3[0,1,2,3,4,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm15 = xmm15[2],xmm9[2],xmm15[3],xmm9[3]
+; SSE-NEXT: pand %xmm1, %xmm15
+; SSE-NEXT: por %xmm2, %xmm15
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm12[0,1,2,0]
; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,6,5]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm13 = xmm13[0,1],xmm2[2,0]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm11[0,0,2,3]
-; SSE-NEXT: movdqa %xmm5, %xmm3
-; SSE-NEXT: pandn %xmm1, %xmm3
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,1,1,3]
-; SSE-NEXT: movdqa %xmm7, %xmm15
-; SSE-NEXT: movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm14[2],xmm2[3],xmm14[3]
-; SSE-NEXT: pand %xmm5, %xmm2
-; SSE-NEXT: por %xmm3, %xmm2
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT: movaps %xmm7, %xmm1
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm11[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm11[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,5,6,7]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm6[0,1,2,0]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm0[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm15 = xmm15[0,1],xmm2[2,0]
+; SSE-NEXT: movdqa %xmm5, %xmm2
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,0],xmm14[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm14[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,0,1,3]
+; SSE-NEXT: movaps %xmm1, %xmm3
+; SSE-NEXT: andnps %xmm2, %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm6[0,1,1,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm10[2],xmm0[3],xmm10[3]
+; SSE-NEXT: pand %xmm1, %xmm0
+; SSE-NEXT: por %xmm3, %xmm0
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm8[0,1,2,0]
; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,6,5]
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1],xmm1[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm9[2,2,2,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm0
-; SSE-NEXT: pandn %xmm1, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm8[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm12[0,2,2,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm3[0,3,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm1[1,0,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm5, %xmm3
-; SSE-NEXT: por %xmm0, %xmm3
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; SSE-NEXT: movaps %xmm8, %xmm0
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm9[3,0]
-; SSE-NEXT: movaps %xmm9, %xmm1
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,7,4,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm4[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,6]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm0[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm1[2,0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm11[2,2,2,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm4
-; SSE-NEXT: pandn %xmm0, %xmm4
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm15[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm14[0,2,2,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm1[0,3,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[1,0,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm5, %xmm1
-; SSE-NEXT: por %xmm4, %xmm1
-; SSE-NEXT: movaps %xmm7, %xmm0
-; SSE-NEXT: movaps %xmm7, %xmm15
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm11[3,0]
-; SSE-NEXT: movaps %xmm11, %xmm4
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm0[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm4[0,1,2,3,7,4,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm6[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,5,6]
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[3,1],xmm0[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,0]
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm9[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm8[0,2,2,3]
-; SSE-NEXT: pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
-; SSE-NEXT: # xmm7 = mem[0,1,1,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,1,0,3,4,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,5,4,7]
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[3,1],xmm4[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1]
-; SSE-NEXT: movdqa %xmm5, %xmm0
-; SSE-NEXT: pandn %xmm4, %xmm0
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm12[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[0,1],xmm4[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm12[0,1,2,3,4,6,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1],xmm2[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,0]
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm4[2,2,2,2,4,5,6,7]
+; SSE-NEXT: movdqa %xmm1, %xmm3
+; SSE-NEXT: pandn %xmm2, %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm7[1,1,1,1]
+; SSE-NEXT: movdqa %xmm7, %xmm6
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm9[0,2,2,3]
+; SSE-NEXT: punpckldq {{.*#+}} xmm5 = xmm5[0],xmm2[0],xmm5[1],xmm2[1]
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm5[0,3,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[1,0,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm1, %xmm2
+; SSE-NEXT: por %xmm3, %xmm2
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm11 = xmm11[2,0],xmm4[3,0]
+; SSE-NEXT: movdqa %xmm1, %xmm8
+; SSE-NEXT: pandn %xmm4, %xmm8
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm11[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,7,4,6,7]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm12[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,5,6]
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[3,1],xmm4[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm5[2,0]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm14[2,2,2,2,4,5,6,7]
+; SSE-NEXT: movdqa %xmm1, %xmm5
+; SSE-NEXT: pandn %xmm4, %xmm5
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm3[1,1,1,1]
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm10[0,2,2,3]
+; SSE-NEXT: punpckldq {{.*#+}} xmm7 = xmm7[0],xmm4[0],xmm7[1],xmm4[1]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm7[0,3,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[2,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm5, %xmm4
-; SSE-NEXT: por %xmm0, %xmm4
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm7[2,0]
-; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm11[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm15[0,2,2,3]
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm6[0,1,1,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm7[0,1,0,3,4,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,7]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm7[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm7 = xmm7[0],xmm8[0],xmm7[1],xmm8[1]
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[2,0],xmm14[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm14 = xmm14[0,1],xmm6[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm14[0,1,2,3,4,6,6,7]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[1,0,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm1, %xmm4
+; SSE-NEXT: por %xmm5, %xmm4
+; SSE-NEXT: movdqa %xmm6, %xmm5
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[2,0],xmm9[3,0]
+; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,1],xmm5[0,2]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm10[3,0]
+; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[0,1],xmm3[0,2]
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
+; SSE-NEXT: movaps %xmm11, %xmm5
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[2,0],xmm14[3,0]
+; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm9[0,1,2,3,4,6,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm7[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm7[2,1,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm1, %xmm7
+; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm10[0,1,2,3,4,6,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm6[2,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm5, %xmm6
-; SSE-NEXT: pandn %xmm7, %xmm5
-; SSE-NEXT: por %xmm6, %xmm5
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,1],xmm0[2,0]
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 16(%rsi)
-; SSE-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, (%rsi)
-; SSE-NEXT: movdqa %xmm10, 16(%rdx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, (%rdx)
-; SSE-NEXT: movaps %xmm2, 16(%rcx)
-; SSE-NEXT: movaps %xmm13, (%rcx)
-; SSE-NEXT: movaps %xmm1, 16(%r8)
-; SSE-NEXT: movaps %xmm3, (%r8)
-; SSE-NEXT: movaps %xmm5, 16(%r9)
-; SSE-NEXT: movaps %xmm4, (%r9)
+; SSE-NEXT: pand %xmm1, %xmm6
+; SSE-NEXT: pandn %xmm14, %xmm1
+; SSE-NEXT: shufps {{.*#+}} xmm14 = xmm14[0,1],xmm5[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm14[0,1,2,3,7,4,6,7]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm9[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,5,6]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1],xmm5[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm3[2,0]
+; SSE-NEXT: por %xmm8, %xmm7
+; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
+; SSE-NEXT: # xmm3 = mem[0,2,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm12[0,1,1,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,3,4,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,4,7]
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[3,1],xmm3[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm5[2,0]
+; SSE-NEXT: por %xmm6, %xmm1
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm11[0,2,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm9[0,1,1,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,3,4,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,4,7]
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[3,1],xmm3[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,0]
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT: movaps %xmm3, 16(%rsi)
+; SSE-NEXT: movaps (%rsp), %xmm3 # 16-byte Reload
+; SSE-NEXT: movaps %xmm3, (%rsi)
+; SSE-NEXT: movdqa %xmm13, 16(%rdx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT: movaps %xmm3, (%rdx)
+; SSE-NEXT: movaps %xmm0, 16(%rcx)
+; SSE-NEXT: movaps %xmm15, (%rcx)
+; SSE-NEXT: movaps %xmm4, 16(%r8)
+; SSE-NEXT: movaps %xmm2, (%r8)
+; SSE-NEXT: movaps %xmm1, 16(%r9)
+; SSE-NEXT: movaps %xmm7, (%r9)
; SSE-NEXT: addq $24, %rsp
; SSE-NEXT: retq
;
@@ -1332,61 +1325,62 @@ define void @vf16(<80 x i16>* %in.vec, <16 x i16>* %out.vec0, <16 x i16>* %out.v
define void @vf32(<160 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.vec1, <32 x i16>* %out.vec2, <32 x i16>* %out.vec3, <32 x i16>* %out.vec4) nounwind {
; SSE-LABEL: vf32:
; SSE: # %bb.0:
-; SSE-NEXT: subq $344, %rsp # imm = 0x158
-; SSE-NEXT: movdqa 304(%rdi), %xmm15
-; SSE-NEXT: movdqa 240(%rdi), %xmm9
+; SSE-NEXT: subq $408, %rsp # imm = 0x198
+; SSE-NEXT: movdqa 304(%rdi), %xmm9
+; SSE-NEXT: movdqa 240(%rdi), %xmm8
; SSE-NEXT: movdqa 256(%rdi), %xmm12
-; SSE-NEXT: movdqa 288(%rdi), %xmm10
-; SSE-NEXT: movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 288(%rdi), %xmm14
; SSE-NEXT: movdqa 272(%rdi), %xmm5
-; SSE-NEXT: movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa 144(%rdi), %xmm7
-; SSE-NEXT: movdqa 80(%rdi), %xmm11
-; SSE-NEXT: movdqa 96(%rdi), %xmm3
-; SSE-NEXT: movdqa 128(%rdi), %xmm4
-; SSE-NEXT: movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm5, (%rsp) # 16-byte Spill
+; SSE-NEXT: movdqa 144(%rdi), %xmm6
+; SSE-NEXT: movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 80(%rdi), %xmm13
+; SSE-NEXT: movdqa 96(%rdi), %xmm10
+; SSE-NEXT: movdqa 128(%rdi), %xmm3
+; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: movdqa 112(%rdi), %xmm2
; SSE-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: movdqa {{.*#+}} xmm0 = [65535,65535,65535,65535,0,65535,65535,65535]
; SSE-NEXT: movdqa %xmm0, %xmm1
; SSE-NEXT: pandn %xmm2, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm4[0,1,0,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,1,0,3]
; SSE-NEXT: pand %xmm0, %xmm2
; SSE-NEXT: por %xmm1, %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm3[3,1,2,3]
-; SSE-NEXT: movdqa %xmm3, %xmm14
-; SSE-NEXT: movdqa %xmm3, (%rsp) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm10[3,1,2,3]
+; SSE-NEXT: movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm11[0,2,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm13[0,2,2,3]
+; SSE-NEXT: movdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,3,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm2[2,3]
-; SSE-NEXT: movaps {{.*#+}} xmm13 = [65535,65535,65535,65535,65535,65535,65535,0]
-; SSE-NEXT: andps %xmm13, %xmm3
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,1,0,1]
-; SSE-NEXT: movaps %xmm13, %xmm2
+; SSE-NEXT: movaps {{.*#+}} xmm7 = [65535,65535,65535,65535,65535,65535,65535,0]
+; SSE-NEXT: andps %xmm7, %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,1,0,1]
+; SSE-NEXT: movaps %xmm7, %xmm2
; SSE-NEXT: pandn %xmm1, %xmm2
; SSE-NEXT: por %xmm3, %xmm2
; SSE-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: movdqa %xmm0, %xmm1
; SSE-NEXT: pandn %xmm5, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm10[0,1,0,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm14[0,1,0,3]
; SSE-NEXT: pand %xmm0, %xmm2
; SSE-NEXT: por %xmm1, %xmm2
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm12[3,1,2,3]
+; SSE-NEXT: movdqa %xmm12, %xmm5
; SSE-NEXT: movdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm9[0,2,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm8[0,2,2,3]
+; SSE-NEXT: movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,3,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm2[2,3]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm15[0,1,0,1]
-; SSE-NEXT: movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movaps %xmm13, %xmm2
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[0,1,0,1]
+; SSE-NEXT: movaps %xmm7, %xmm2
; SSE-NEXT: andnps %xmm1, %xmm2
; SSE-NEXT: movdqa 32(%rdi), %xmm4
; SSE-NEXT: movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: andps %xmm13, %xmm3
+; SSE-NEXT: andps %xmm7, %xmm3
; SSE-NEXT: orps %xmm3, %xmm2
; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: movdqa %xmm0, %xmm1
@@ -1396,22 +1390,22 @@ define void @vf32(<160 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
; SSE-NEXT: pand %xmm0, %xmm2
; SSE-NEXT: por %xmm1, %xmm2
-; SSE-NEXT: movdqa 16(%rdi), %xmm6
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm6[3,1,2,3]
-; SSE-NEXT: movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 16(%rdi), %xmm11
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm11[3,1,2,3]
+; SSE-NEXT: movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa (%rdi), %xmm4
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
-; SSE-NEXT: movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa (%rdi), %xmm6
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm6[0,2,2,3]
+; SSE-NEXT: movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,3,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm2[2,3]
; SSE-NEXT: movdqa 64(%rdi), %xmm1
; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
-; SSE-NEXT: movaps %xmm13, %xmm2
+; SSE-NEXT: movaps %xmm7, %xmm2
; SSE-NEXT: andnps %xmm1, %xmm2
-; SSE-NEXT: andps %xmm13, %xmm3
+; SSE-NEXT: andps %xmm7, %xmm3
; SSE-NEXT: orps %xmm3, %xmm2
; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: movdqa 208(%rdi), %xmm1
@@ -1422,38 +1416,35 @@ define void @vf32(<160 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pandn %xmm2, %xmm0
; SSE-NEXT: por %xmm1, %xmm0
-; SSE-NEXT: movdqa 176(%rdi), %xmm10
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm10[3,1,2,3]
-; SSE-NEXT: movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 176(%rdi), %xmm1
+; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
-; SSE-NEXT: movdqa 160(%rdi), %xmm3
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
-; SSE-NEXT: movdqa %xmm3, %xmm8
-; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 160(%rdi), %xmm15
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm15[0,2,2,3]
+; SSE-NEXT: movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,3,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm0[2,3]
; SSE-NEXT: movdqa 224(%rdi), %xmm0
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
-; SSE-NEXT: movaps %xmm13, %xmm1
+; SSE-NEXT: movaps %xmm7, %xmm1
; SSE-NEXT: andnps %xmm0, %xmm1
-; SSE-NEXT: andps %xmm13, %xmm2
+; SSE-NEXT: andps %xmm7, %xmm2
; SSE-NEXT: orps %xmm2, %xmm1
; SSE-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm14, %xmm0
-; SSE-NEXT: psrlq $48, %xmm0
-; SSE-NEXT: movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm11[0,3,2,3]
+; SSE-NEXT: psrlq $48, %xmm10
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm13[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[1,2,2,3,4,5,6,7]
-; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm10[0],xmm1[1],xmm10[1]
; SSE-NEXT: movdqa {{.*#+}} xmm0 = [0,0,0,65535,65535,65535,65535,65535]
; SSE-NEXT: movdqa %xmm0, %xmm2
; SSE-NEXT: pandn %xmm1, %xmm2
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm14[1,3,2,3]
-; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[0,2,2,3]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm4[1,3,2,3]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm10[0,2,2,3]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm3[0,1,2,3,7,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
@@ -1461,25 +1452,25 @@ define void @vf32(<160 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,6,4,7]
; SSE-NEXT: pand %xmm0, %xmm1
; SSE-NEXT: por %xmm2, %xmm1
-; SSE-NEXT: movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm7, %xmm2
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
+; SSE-NEXT: movdqa %xmm12, %xmm2
; SSE-NEXT: psllq $48, %xmm2
-; SSE-NEXT: movaps %xmm13, %xmm3
+; SSE-NEXT: movaps %xmm7, %xmm3
; SSE-NEXT: andnps %xmm2, %xmm3
-; SSE-NEXT: pand %xmm13, %xmm1
+; SSE-NEXT: pand %xmm7, %xmm1
; SSE-NEXT: orps %xmm1, %xmm3
; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $48, %xmm12
-; SSE-NEXT: movdqa %xmm9, %xmm5
-; SSE-NEXT: movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm9[0,3,2,3]
+; SSE-NEXT: movdqa %xmm5, %xmm1
+; SSE-NEXT: psrlq $48, %xmm1
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm8[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[1,2,2,3,4,5,6,7]
-; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm12[0],xmm2[1],xmm12[1]
+; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; SSE-NEXT: movdqa %xmm0, %xmm1
; SSE-NEXT: pandn %xmm2, %xmm1
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm9[1,3,2,3]
-; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm14[1,3,2,3]
+; SSE-NEXT: movdqa %xmm14, %xmm13
+; SSE-NEXT: movdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd $232, (%rsp), %xmm3 # 16-byte Folded Reload
; SSE-NEXT: # xmm3 = mem[0,2,2,3]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm3[0,1,2,3,7,5,6,7]
@@ -1488,20 +1479,22 @@ define void @vf32(<160 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,6,4,7]
; SSE-NEXT: pand %xmm0, %xmm2
; SSE-NEXT: por %xmm1, %xmm2
-; SSE-NEXT: psllq $48, %xmm15
-; SSE-NEXT: movdqa %xmm13, %xmm3
-; SSE-NEXT: pandn %xmm15, %xmm3
-; SSE-NEXT: pand %xmm13, %xmm2
+; SSE-NEXT: movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm9, %xmm1
+; SSE-NEXT: psllq $48, %xmm1
+; SSE-NEXT: movdqa %xmm7, %xmm3
+; SSE-NEXT: pandn %xmm1, %xmm3
+; SSE-NEXT: pand %xmm7, %xmm2
; SSE-NEXT: por %xmm2, %xmm3
; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $48, %xmm6
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm4[0,3,2,3]
+; SSE-NEXT: psrlq $48, %xmm11
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm6[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[1,2,2,3,4,5,6,7]
-; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm6[0],xmm2[1],xmm6[1]
+; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm11[0],xmm2[1],xmm11[1]
; SSE-NEXT: movdqa %xmm0, %xmm1
; SSE-NEXT: pandn %xmm2, %xmm1
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm15[1,3,2,3]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm11[1,3,2,3]
; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
; SSE-NEXT: # xmm3 = mem[0,2,2,3]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
@@ -1511,19 +1504,22 @@ define void @vf32(<160 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,6,4,7]
; SSE-NEXT: pand %xmm0, %xmm2
; SSE-NEXT: por %xmm1, %xmm2
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; SSE-NEXT: movdqa %xmm8, %xmm1
; SSE-NEXT: psllq $48, %xmm1
-; SSE-NEXT: movdqa %xmm13, %xmm3
+; SSE-NEXT: movdqa %xmm7, %xmm3
; SSE-NEXT: pandn %xmm1, %xmm3
-; SSE-NEXT: pand %xmm13, %xmm2
+; SSE-NEXT: pand %xmm7, %xmm2
; SSE-NEXT: por %xmm2, %xmm3
; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $48, %xmm10
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm8[0,3,2,3]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
+; SSE-NEXT: movdqa %xmm14, %xmm1
+; SSE-NEXT: psrlq $48, %xmm1
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm15[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[1,2,2,3,4,5,6,7]
-; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm8[1,3,2,3]
+; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm5[1,3,2,3]
; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
; SSE-NEXT: # xmm3 = mem[0,2,2,3]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
@@ -1534,299 +1530,289 @@ define void @vf32(<160 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pand %xmm0, %xmm1
; SSE-NEXT: pandn %xmm2, %xmm0
; SSE-NEXT: por %xmm1, %xmm0
-; SSE-NEXT: pand %xmm13, %xmm0
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT: movdqa %xmm4, %xmm1
+; SSE-NEXT: pand %xmm7, %xmm0
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT: movdqa %xmm6, %xmm1
; SSE-NEXT: psllq $48, %xmm1
-; SSE-NEXT: pandn %xmm1, %xmm13
-; SSE-NEXT: por %xmm0, %xmm13
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm10[0,0,2,3]
-; SSE-NEXT: movdqa {{.*#+}} xmm6 = [65535,65535,65535,0,0,0,65535,65535]
-; SSE-NEXT: movdqa %xmm6, %xmm2
-; SSE-NEXT: pandn %xmm1, %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm11[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,7,6,7]
-; SSE-NEXT: punpckhdq (%rsp), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT: # xmm0 = xmm0[2],mem[2],xmm0[3],mem[3]
-; SSE-NEXT: pand %xmm6, %xmm0
-; SSE-NEXT: por %xmm2, %xmm0
-; SSE-NEXT: movdqa %xmm14, %xmm1
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm10[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm10[2,3]
-; SSE-NEXT: movdqa %xmm10, %xmm14
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm7[0,1,2,0]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,6,5]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,0]
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm10[0,0,2,3]
-; SSE-NEXT: movdqa %xmm6, %xmm2
-; SSE-NEXT: pandn %xmm1, %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm5[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT: punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm5[2],xmm1[3],xmm5[3]
-; SSE-NEXT: pand %xmm6, %xmm1
-; SSE-NEXT: por %xmm2, %xmm1
-; SSE-NEXT: movdqa %xmm1, %xmm3
-; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[1,0],xmm10[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[2,0],xmm10[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm9[0,1,2,3,6,5,6,7]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm10[0,1,2,0]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,6,5]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm2[2,0]
-; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,0,2,3]
-; SSE-NEXT: movdqa %xmm6, %xmm2
-; SSE-NEXT: pandn %xmm1, %xmm2
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm11 = xmm1[0,1,2,3,4,7,6,7]
+; SSE-NEXT: pandn %xmm1, %xmm7
+; SSE-NEXT: por %xmm0, %xmm7
+; SSE-NEXT: movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm4, %xmm0
+; SSE-NEXT: movdqa %xmm10, %xmm7
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0],xmm10[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm10[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,1,3]
+; SSE-NEXT: movaps {{.*#+}} xmm10 = [65535,65535,65535,0,0,0,65535,65535]
+; SSE-NEXT: movaps %xmm10, %xmm1
+; SSE-NEXT: andnps %xmm0, %xmm1
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT: punpckhdq {{.*#+}} xmm11 = xmm11[2],xmm3[2],xmm11[3],xmm3[3]
-; SSE-NEXT: pand %xmm6, %xmm11
-; SSE-NEXT: por %xmm2, %xmm11
-; SSE-NEXT: movdqa %xmm15, %xmm1
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm7[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm7[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,1,1,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,7,6,7]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
+; SSE-NEXT: pand %xmm10, %xmm2
+; SSE-NEXT: por %xmm1, %xmm2
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm12[0,1,2,0]
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,5]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm0[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,0]
+; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movaps (%rsp), %xmm15 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm13 = xmm13[1,0],xmm15[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm13 = xmm13[2,0],xmm15[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm13 = xmm13[0,0,1,3]
+; SSE-NEXT: movaps %xmm10, %xmm1
+; SSE-NEXT: andnps %xmm13, %xmm1
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm12[0,1,2,0]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,6,5]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm11 = xmm11[0,1],xmm2[2,0]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm15[0,0,2,3]
-; SSE-NEXT: movdqa %xmm6, %xmm2
-; SSE-NEXT: pandn %xmm1, %xmm2
-; SSE-NEXT: pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = mem[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm1[0,1,2,3,4,7,6,7]
-; SSE-NEXT: punpckhdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
-; SSE-NEXT: # xmm7 = xmm7[2],mem[2],xmm7[3],mem[3]
-; SSE-NEXT: pand %xmm6, %xmm7
-; SSE-NEXT: por %xmm2, %xmm7
-; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[1,0],xmm15[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[2,0],xmm15[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,1,1,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm8[0,1,2,3,6,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm4[0,1,2,0]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,6,5]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm2[2,0]
-; SSE-NEXT: movdqa %xmm14, %xmm0
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm14[2,2,2,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm6, %xmm2
-; SSE-NEXT: pandn %xmm1, %xmm2
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = mem[1,1,1,1]
-; SSE-NEXT: pshufd $232, (%rsp), %xmm4 # 16-byte Folded Reload
-; SSE-NEXT: # xmm4 = mem[0,2,2,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm4[0,3,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm14 = xmm1[1,0,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm14
-; SSE-NEXT: por %xmm2, %xmm14
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm12[0,1,1,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,7,6,7]
+; SSE-NEXT: punpckhdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = xmm2[2],mem[2],xmm2[3],mem[3]
+; SSE-NEXT: pand %xmm10, %xmm2
+; SSE-NEXT: por %xmm1, %xmm2
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm13[0,1,2,3,6,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[0,1,2,0]
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,5]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm0[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,0]
+; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm11, %xmm0
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0],xmm13[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm13[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,1,3]
+; SSE-NEXT: movaps %xmm10, %xmm1
+; SSE-NEXT: andnps %xmm0, %xmm1
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm9[0,1,1,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,7,6,7]
+; SSE-NEXT: punpckhdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = xmm2[2],mem[2],xmm2[3],mem[3]
+; SSE-NEXT: pand %xmm10, %xmm2
+; SSE-NEXT: por %xmm1, %xmm2
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm8[0,1,2,0]
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,5]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm0[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,0]
+; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[1,0],xmm1[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[2,0],xmm1[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,0,1,3]
+; SSE-NEXT: movaps %xmm10, %xmm1
+; SSE-NEXT: andnps %xmm5, %xmm1
+; SSE-NEXT: pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = mem[0,1,1,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm14[2],xmm2[3],xmm14[3]
+; SSE-NEXT: pand %xmm10, %xmm2
+; SSE-NEXT: por %xmm1, %xmm2
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm5[0,1,2,3,6,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm6[0,1,2,0]
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,5]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm0[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,0]
+; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm7[2,2,2,2,4,5,6,7]
+; SSE-NEXT: movdqa %xmm10, %xmm1
+; SSE-NEXT: pandn %xmm0, %xmm1
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm3[1,1,1,1]
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
+; SSE-NEXT: movdqa %xmm4, %xmm6
+; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm3[0,3,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[1,0,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm10, %xmm0
+; SSE-NEXT: por %xmm1, %xmm0
; SSE-NEXT: movdqa %xmm0, %xmm2
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm2[0,1,2,3,7,4,6,7]
-; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT: # xmm2 = mem[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,6]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm14 = xmm14[0,1],xmm2[2,0]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[2,2,2,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm6, %xmm2
-; SSE-NEXT: pandn %xmm1, %xmm2
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = mem[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm5[0,2,2,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm4[0,3,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm8 = xmm1[1,0,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm8
-; SSE-NEXT: por %xmm2, %xmm8
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,7,4,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm10[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,6]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,1],xmm2[2,0]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm4[2,2,2,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm6, %xmm0
-; SSE-NEXT: pandn %xmm1, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,3,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm1[1,0,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm2
-; SSE-NEXT: por %xmm0, %xmm2
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm4[3,0]
-; SSE-NEXT: movaps %xmm4, %xmm1
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,7,4,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm12[0,1,0,3]
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm7[3,0]
+; SSE-NEXT: movaps %xmm10, %xmm0
+; SSE-NEXT: andnps %xmm7, %xmm0
+; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm3[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm7[0,1,2,3,7,4,6,7]
+; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT: # xmm1 = mem[0,1,0,3]
; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,6]
; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm0[2,3]
; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,0]
+; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm15[2,2,2,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm6, %xmm3
-; SSE-NEXT: pandn %xmm0, %xmm3
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT: # xmm0 = mem[1,1,1,1]
-; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = mem[0,2,2,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm1[0,3,2,3,4,5,6,7]
+; SSE-NEXT: movdqa %xmm10, %xmm1
+; SSE-NEXT: pandn %xmm0, %xmm1
+; SSE-NEXT: movdqa %xmm12, %xmm4
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm12[1,1,1,1]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
+; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm3[0,3,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[1,0,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm1
-; SSE-NEXT: por %xmm3, %xmm1
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT: movaps %xmm4, %xmm0
+; SSE-NEXT: pshuflw {{.*#+}} xmm8 = xmm0[1,0,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm10, %xmm8
+; SSE-NEXT: por %xmm1, %xmm8
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm15[3,0]
-; SSE-NEXT: movaps %xmm15, %xmm3
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm0[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm3[0,1,2,3,7,4,6,7]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm9[0,1,0,3]
+; SSE-NEXT: movdqa %xmm10, %xmm11
+; SSE-NEXT: pandn %xmm15, %xmm11
+; SSE-NEXT: shufps {{.*#+}} xmm15 = xmm15[0,1],xmm0[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm15[0,1,2,3,7,4,6,7]
+; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
+; SSE-NEXT: # xmm3 = mem[0,1,0,3]
; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,5,6]
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1],xmm0[2,3]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,0]
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Folded Reload
-; SSE-NEXT: # xmm10 = mem[1,1,1,1]
-; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[0,2,2,3]
-; SSE-NEXT: pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT: # xmm0 = mem[0,1,1,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,3,4,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,7]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm3[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm10[0],xmm3[1],xmm10[1]
-; SSE-NEXT: movdqa %xmm6, %xmm5
-; SSE-NEXT: pandn %xmm3, %xmm5
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT: movaps (%rsp), %xmm10 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm10[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[0,1],xmm3[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm10[0,1,2,3,4,6,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm10 = xmm3[2,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm10
-; SSE-NEXT: por %xmm5, %xmm10
-; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[0,1],xmm0[2,0]
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT: # xmm0 = mem[1,1,1,1]
-; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[0,2,2,3]
-; SSE-NEXT: pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT: # xmm5 = mem[0,1,1,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,3,4,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,4,7]
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[3,1],xmm3[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; SSE-NEXT: movdqa %xmm6, %xmm0
-; SSE-NEXT: pandn %xmm3, %xmm0
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm12[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[0,1],xmm3[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm12[0,1,2,3,4,6,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm12 = xmm3[2,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm12
-; SSE-NEXT: por %xmm0, %xmm12
-; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[0,1],xmm5[2,0]
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm15[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm9[0,1,1,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,3,4,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,4,7]
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[3,1],xmm3[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; SSE-NEXT: movdqa %xmm6, %xmm0
-; SSE-NEXT: pandn %xmm3, %xmm0
+; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,1],xmm3[2,0]
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm13[2,2,2,2,4,5,6,7]
+; SSE-NEXT: movdqa %xmm10, %xmm3
+; SSE-NEXT: pandn %xmm0, %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm9[1,1,1,1]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm12[0,2,2,3]
+; SSE-NEXT: punpckldq {{.*#+}} xmm7 = xmm7[0],xmm0[0],xmm7[1],xmm0[1]
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm7[0,3,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm15 = xmm0[1,0,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm10, %xmm15
+; SSE-NEXT: por %xmm3, %xmm15
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,1,1]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm7[0,2,2,3]
+; SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+; SSE-NEXT: movdqa %xmm6, %xmm1
; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm3[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm4[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm9 = xmm3[2,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm9
-; SSE-NEXT: por %xmm0, %xmm9
-; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,1],xmm5[2,0]
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT: # xmm0 = mem[1,1,1,1]
-; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[0,2,2,3]
-; SSE-NEXT: pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT: # xmm5 = mem[0,1,1,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,3,4,5,6,7]
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,4,7]
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[3,1],xmm3[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm15 = xmm15[2,0],xmm0[3,0]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm15[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm6[3,0]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm3[0,2]
+; SSE-NEXT: movdqa %xmm4, %xmm3
+; SSE-NEXT: movdqa %xmm5, %xmm4
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm5[3,0]
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm3[0,2]
+; SSE-NEXT: movdqa %xmm7, %xmm3
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm7[3,0]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm2[0,2]
+; SSE-NEXT: movaps %xmm3, %xmm5
+; SSE-NEXT: movdqa %xmm12, %xmm3
+; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[2,0],xmm12[3,0]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm9[0,2]
+; SSE-NEXT: movaps %xmm3, %xmm6
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT: movaps %xmm9, %xmm7
+; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[2,0],xmm13[3,0]
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm0[1,0,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm10, %xmm3
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,6,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm14 = xmm0[2,1,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm10, %xmm14
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm4[0,1,2,3,4,6,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[2,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm6, %xmm0
-; SSE-NEXT: pandn %xmm3, %xmm6
-; SSE-NEXT: por %xmm0, %xmm6
-; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,1],xmm5[2,0]
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 32(%rsi)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, (%rsi)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 48(%rsi)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 16(%rsi)
-; SSE-NEXT: movdqa %xmm13, 32(%rdx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, (%rdx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 48(%rdx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 16(%rdx)
-; SSE-NEXT: movaps %xmm7, 32(%rcx)
-; SSE-NEXT: movaps %xmm11, (%rcx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 48(%rcx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 16(%rcx)
-; SSE-NEXT: movaps %xmm1, 32(%r8)
-; SSE-NEXT: movaps %xmm2, (%r8)
+; SSE-NEXT: pand %xmm10, %xmm0
+; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,6,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm12 = xmm5[2,1,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm10, %xmm12
+; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm6[0,1,2,3,4,6,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[2,1,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm10, %xmm5
+; SSE-NEXT: movdqa %xmm10, %xmm6
+; SSE-NEXT: movdqa %xmm10, %xmm1
+; SSE-NEXT: pandn %xmm13, %xmm10
+; SSE-NEXT: shufps {{.*#+}} xmm13 = xmm13[0,1],xmm7[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm13[0,1,2,3,7,4,6,7]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm13[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,5,6]
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[3,1],xmm7[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm15 = xmm15[0,1],xmm4[2,0]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm2[2,2,2,2,4,5,6,7]
+; SSE-NEXT: pandn %xmm4, %xmm6
+; SSE-NEXT: por %xmm6, %xmm3
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT: movaps %xmm7, %xmm4
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm2[3,0]
+; SSE-NEXT: pandn %xmm2, %xmm1
+; SSE-NEXT: movaps %xmm2, %xmm6
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,1],xmm4[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm6[0,1,2,3,7,4,6,7]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm2[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5,5,6]
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[3,1],xmm4[2,3]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm6[2,0]
+; SSE-NEXT: por {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Folded Reload
+; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
+; SSE-NEXT: # xmm4 = mem[0,2,2,3]
+; SSE-NEXT: pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
+; SSE-NEXT: # xmm6 = mem[0,1,1,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,1,0,3,4,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5,4,7]
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[3,1],xmm4[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm14 = xmm14[0,1],xmm6[2,0]
+; SSE-NEXT: por %xmm11, %xmm0
+; SSE-NEXT: pshufd $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
+; SSE-NEXT: # xmm6 = mem[0,2,2,3]
+; SSE-NEXT: pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
+; SSE-NEXT: # xmm4 = mem[0,1,1,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm6[0,1,0,3,4,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,4,7]
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[3,1],xmm6[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm4[2,0]
+; SSE-NEXT: por %xmm1, %xmm12
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm7[0,2,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,1,1,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,3,4,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,7]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[0,1],xmm2[2,0]
+; SSE-NEXT: por %xmm5, %xmm10
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[0,2,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm13[0,1,1,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,3,4,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,7]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[0,1],xmm2[2,0]
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 32(%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, (%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 48(%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 16(%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 32(%rdx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, (%rdx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 48(%rdx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 16(%rdx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 32(%rcx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, (%rcx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 48(%rcx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 16(%rcx)
+; SSE-NEXT: movaps %xmm3, 32(%r8)
+; SSE-NEXT: movaps %xmm15, (%r8)
; SSE-NEXT: movaps %xmm8, 48(%r8)
-; SSE-NEXT: movaps %xmm14, 16(%r8)
-; SSE-NEXT: movaps %xmm6, (%r9)
-; SSE-NEXT: movaps %xmm9, 32(%r9)
-; SSE-NEXT: movaps %xmm12, 48(%r9)
-; SSE-NEXT: movaps %xmm10, 16(%r9)
-; SSE-NEXT: addq $344, %rsp # imm = 0x158
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 16(%r8)
+; SSE-NEXT: movaps %xmm10, (%r9)
+; SSE-NEXT: movaps %xmm12, 32(%r9)
+; SSE-NEXT: movaps %xmm0, 48(%r9)
+; SSE-NEXT: movaps %xmm14, 16(%r9)
+; SSE-NEXT: addq $408, %rsp # imm = 0x198
; SSE-NEXT: retq
;
; AVX1-LABEL: vf32:
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll
index a104919603f18..52303572d11cb 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll
@@ -372,126 +372,119 @@ define void @vf4(<24 x i16>* %in.vec, <4 x i16>* %out.vec0, <4 x i16>* %out.vec1
define void @vf8(<48 x i16>* %in.vec, <8 x i16>* %out.vec0, <8 x i16>* %out.vec1, <8 x i16>* %out.vec2, <8 x i16>* %out.vec3, <8 x i16>* %out.vec4, <8 x i16>* %out.vec5) nounwind {
; SSE-LABEL: vf8:
; SSE: # %bb.0:
-; SSE-NEXT: movdqa 64(%rdi), %xmm4
-; SSE-NEXT: movdqa 80(%rdi), %xmm11
-; SSE-NEXT: movdqa (%rdi), %xmm15
-; SSE-NEXT: movdqa 16(%rdi), %xmm13
+; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rax
+; SSE-NEXT: movdqa 64(%rdi), %xmm6
+; SSE-NEXT: movdqa 80(%rdi), %xmm10
+; SSE-NEXT: movdqa (%rdi), %xmm13
+; SSE-NEXT: movdqa 16(%rdi), %xmm14
; SSE-NEXT: movdqa 32(%rdi), %xmm9
-; SSE-NEXT: movdqa 48(%rdi), %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm9[1,1,1,1]
-; SSE-NEXT: movdqa %xmm9, %xmm7
-; SSE-NEXT: movdqa %xmm9, %xmm12
-; SSE-NEXT: psrlq $16, %xmm12
-; SSE-NEXT: pshufd {{.*#+}} xmm10 = xmm1[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm14 = xmm10[0,1,0,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm4, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm4[2,2,3,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm6 = xmm6[0],xmm11[0],xmm6[1],xmm11[1],xmm6[2],xmm11[2],xmm6[3],xmm11[3]
-; SSE-NEXT: movdqa %xmm11, %xmm3
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm4[3,0]
-; SSE-NEXT: movaps %xmm4, %xmm5
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,0],xmm11[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm11[2,3]
-; SSE-NEXT: pslld $16, %xmm11
+; SSE-NEXT: movdqa 48(%rdi), %xmm11
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm9[0,1,1,2,4,5,6,7]
+; SSE-NEXT: movdqa {{.*#+}} xmm1 = [65535,65535,65535,0,0,0,65535,65535]
+; SSE-NEXT: movdqa %xmm1, %xmm4
+; SSE-NEXT: pandn %xmm0, %xmm4
+; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm13[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm12 = xmm8[0,1,2,3,4,6,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm12 = xmm12[2],xmm14[2],xmm12[3],xmm14[3]
+; SSE-NEXT: pand %xmm1, %xmm12
+; SSE-NEXT: por %xmm4, %xmm12
+; SSE-NEXT: movdqa %xmm6, %xmm0
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm6[2,2,3,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm7 = xmm7[0],xmm10[0],xmm7[1],xmm10[1],xmm7[2],xmm10[2],xmm7[3],xmm10[3]
+; SSE-NEXT: movdqa %xmm10, %xmm3
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm6[3,0]
+; SSE-NEXT: movaps %xmm6, %xmm15
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[1,0],xmm10[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[2,0],xmm10[2,3]
+; SSE-NEXT: pslld $16, %xmm10
; SSE-NEXT: psrldq {{.*#+}} xmm0 = xmm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm11[0],xmm0[1],xmm11[1],xmm0[2],xmm11[2],xmm0[3],xmm11[3]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm14[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm14 = xmm14[0],xmm12[0],xmm14[1],xmm12[1]
-; SSE-NEXT: movdqa {{.*#+}} xmm2 = [65535,65535,65535,0,0,0,65535,65535]
-; SSE-NEXT: movdqa %xmm2, %xmm12
-; SSE-NEXT: pandn %xmm14, %xmm12
-; SSE-NEXT: pshufd {{.*#+}} xmm14 = xmm15[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm11 = xmm14[0,1,2,3,4,6,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm11 = xmm11[2],xmm13[2],xmm11[3],xmm13[3]
-; SSE-NEXT: pand %xmm2, %xmm11
-; SSE-NEXT: por %xmm12, %xmm11
-; SSE-NEXT: pshufd {{.*#+}} xmm12 = xmm9[2,2,3,3]
-; SSE-NEXT: pshufd {{.*#+}} xmm9 = xmm9[0,1,0,3]
-; SSE-NEXT: shufps {{.*#+}} xmm11 = xmm11[0,1],xmm0[2,0]
-; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm10[0,1,1,3,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[3,1],xmm0[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm8[0],xmm0[1],xmm8[1]
-; SSE-NEXT: movdqa %xmm13, %xmm8
-; SSE-NEXT: psrld $16, %xmm8
-; SSE-NEXT: pshufhw {{.*#+}} xmm10 = xmm14[0,1,2,3,5,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm10 = xmm10[2],xmm8[2],xmm10[3],xmm8[3]
-; SSE-NEXT: pand %xmm2, %xmm10
-; SSE-NEXT: pandn %xmm0, %xmm2
-; SSE-NEXT: por %xmm10, %xmm2
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm6[2,0]
-; SSE-NEXT: movdqa %xmm1, %xmm10
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
-; SSE-NEXT: psrldq {{.*#+}} xmm7 = xmm7[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm7 = xmm7[0],xmm0[0]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm10[0],xmm0[1],xmm10[1],xmm0[2],xmm10[2],xmm0[3],xmm10[3]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm11[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm2[0,1,0,2,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm5[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[0,1],xmm0[2,0]
+; SSE-NEXT: movdqa %xmm14, %xmm0
+; SSE-NEXT: psrld $16, %xmm0
+; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm8[0,1,2,3,5,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm5 = xmm5[2],xmm0[2],xmm5[3],xmm0[3]
+; SSE-NEXT: pand %xmm1, %xmm5
+; SSE-NEXT: pandn %xmm9, %xmm1
+; SSE-NEXT: por %xmm5, %xmm1
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm2[0,1,1,3,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[3,1],xmm0[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm7[2,0]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm11[1,1,1,1]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm9[2,2,3,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm9[0,1,0,3]
+; SSE-NEXT: psrldq {{.*#+}} xmm9 = xmm9[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm9 = xmm9[0],xmm0[0]
; SSE-NEXT: movdqa {{.*#+}} xmm0 = [65535,65535,65535,0,0,65535,65535,65535]
-; SSE-NEXT: movdqa %xmm0, %xmm6
-; SSE-NEXT: pandn %xmm7, %xmm6
-; SSE-NEXT: movdqa %xmm15, %xmm7
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[1,0],xmm13[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[2,0],xmm13[2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm7[0,2,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[1,0,2,3,4,5,6,7]
-; SSE-NEXT: pand %xmm0, %xmm1
-; SSE-NEXT: por %xmm6, %xmm1
-; SSE-NEXT: movdqa {{.*#+}} xmm6 = [65535,65535,65535,65535,65535,0,0,0]
-; SSE-NEXT: pand %xmm6, %xmm1
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,1],xmm3[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm5[0,1,2,3,4,6,6,7]
+; SSE-NEXT: movdqa %xmm0, %xmm7
+; SSE-NEXT: pandn %xmm9, %xmm7
+; SSE-NEXT: movdqa %xmm13, %xmm4
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,0],xmm14[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm14[2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm4[0,2,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[1,0,2,3,4,5,6,7]
+; SSE-NEXT: pand %xmm0, %xmm5
+; SSE-NEXT: por %xmm7, %xmm5
+; SSE-NEXT: movdqa {{.*#+}} xmm7 = [65535,65535,65535,65535,65535,0,0,0]
+; SSE-NEXT: pand %xmm7, %xmm5
+; SSE-NEXT: shufps {{.*#+}} xmm15 = xmm15[0,1],xmm3[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm15[0,1,2,3,4,6,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,2,0]
-; SSE-NEXT: pshufhw {{.*#+}} xmm8 = xmm3[0,1,2,3,4,6,5,4]
-; SSE-NEXT: movdqa %xmm6, %xmm3
-; SSE-NEXT: pandn %xmm8, %xmm3
-; SSE-NEXT: por %xmm1, %xmm3
-; SSE-NEXT: movdqa %xmm10, %xmm1
-; SSE-NEXT: psrlq $48, %xmm1
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm12 = xmm12[0],xmm1[0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm7[3,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm0, %xmm1
-; SSE-NEXT: pandn %xmm12, %xmm0
-; SSE-NEXT: por %xmm1, %xmm0
-; SSE-NEXT: pand %xmm6, %xmm0
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm5[0,1,2,3,7,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,2]
-; SSE-NEXT: movdqa %xmm6, %xmm5
-; SSE-NEXT: pandn %xmm1, %xmm5
+; SSE-NEXT: pshufhw {{.*#+}} xmm9 = xmm3[0,1,2,3,4,6,5,4]
+; SSE-NEXT: movdqa %xmm7, %xmm3
+; SSE-NEXT: pandn %xmm9, %xmm3
+; SSE-NEXT: por %xmm5, %xmm3
+; SSE-NEXT: movdqa %xmm11, %xmm5
+; SSE-NEXT: psrlq $48, %xmm5
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm5[0]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,1,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm0, %xmm4
+; SSE-NEXT: pandn %xmm2, %xmm0
+; SSE-NEXT: por %xmm4, %xmm0
+; SSE-NEXT: pand %xmm7, %xmm0
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm15[0,1,2,3,7,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,1,0,2]
+; SSE-NEXT: movdqa %xmm7, %xmm4
+; SSE-NEXT: pandn %xmm2, %xmm4
+; SSE-NEXT: por %xmm0, %xmm4
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm14[1,1,1,1]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm13[2,3,2,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm8[0,1,2,3,4,5,4,6]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm11[1]
+; SSE-NEXT: movss {{.*#+}} xmm0 = xmm2[0],xmm0[1,2,3]
+; SSE-NEXT: andps %xmm7, %xmm0
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm6[0,2,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,6]
+; SSE-NEXT: movdqa %xmm7, %xmm5
+; SSE-NEXT: pandn %xmm2, %xmm5
; SSE-NEXT: por %xmm0, %xmm5
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm13[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm15[2,3,2,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm9[0,1,2,3,4,5,4,6]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm10[1]
-; SSE-NEXT: movss {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
-; SSE-NEXT: andps %xmm6, %xmm0
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm4[0,2,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,4,6]
-; SSE-NEXT: movdqa %xmm6, %xmm7
-; SSE-NEXT: pandn %xmm1, %xmm7
+; SSE-NEXT: psrlq $48, %xmm14
+; SSE-NEXT: psrldq {{.*#+}} xmm13 = xmm13[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm13 = xmm13[0],xmm14[0],xmm13[1],xmm14[1],xmm13[2],xmm14[2],xmm13[3],xmm14[3]
+; SSE-NEXT: psrld $16, %xmm11
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm8[0,1,2,3,4,5,5,7]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm11[1]
+; SSE-NEXT: movss {{.*#+}} xmm0 = xmm13[0],xmm0[1,2,3]
+; SSE-NEXT: andps %xmm7, %xmm0
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm6[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,5,7]
+; SSE-NEXT: pandn %xmm2, %xmm7
; SSE-NEXT: por %xmm0, %xmm7
-; SSE-NEXT: psrlq $48, %xmm13
-; SSE-NEXT: psrldq {{.*#+}} xmm15 = xmm15[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm15 = xmm15[0],xmm13[0],xmm15[1],xmm13[1],xmm15[2],xmm13[2],xmm15[3],xmm13[3]
-; SSE-NEXT: movdqa %xmm10, %xmm1
-; SSE-NEXT: psrld $16, %xmm1
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm9[0,1,2,3,4,5,5,7]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; SSE-NEXT: movss {{.*#+}} xmm0 = xmm15[0],xmm0[1,2,3]
-; SSE-NEXT: andps %xmm6, %xmm0
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm4[3,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,5,7]
-; SSE-NEXT: pandn %xmm1, %xmm6
-; SSE-NEXT: por %xmm0, %xmm6
-; SSE-NEXT: movaps %xmm11, (%rsi)
-; SSE-NEXT: movaps %xmm2, (%rdx)
+; SSE-NEXT: movaps %xmm12, (%rsi)
+; SSE-NEXT: movaps %xmm1, (%rdx)
; SSE-NEXT: movdqa %xmm3, (%rcx)
-; SSE-NEXT: movdqa %xmm5, (%r8)
-; SSE-NEXT: movdqa %xmm7, (%r9)
-; SSE-NEXT: movdqa %xmm6, (%rax)
+; SSE-NEXT: movdqa %xmm4, (%r8)
+; SSE-NEXT: movdqa %xmm5, (%r9)
+; SSE-NEXT: movdqa %xmm7, (%rax)
; SSE-NEXT: retq
;
; AVX1-LABEL: vf8:
@@ -807,288 +800,275 @@ define void @vf8(<48 x i16>* %in.vec, <8 x i16>* %out.vec0, <8 x i16>* %out.vec1
define void @vf16(<96 x i16>* %in.vec, <16 x i16>* %out.vec0, <16 x i16>* %out.vec1, <16 x i16>* %out.vec2, <16 x i16>* %out.vec3, <16 x i16>* %out.vec4, <16 x i16>* %out.vec5) nounwind {
; SSE-LABEL: vf16:
; SSE: # %bb.0:
-; SSE-NEXT: subq $88, %rsp
-; SSE-NEXT: movdqa 160(%rdi), %xmm4
-; SSE-NEXT: movdqa 176(%rdi), %xmm12
-; SSE-NEXT: movdqa 64(%rdi), %xmm0
-; SSE-NEXT: movdqa 80(%rdi), %xmm7
-; SSE-NEXT: movaps (%rdi), %xmm1
+; SSE-NEXT: subq $72, %rsp
+; SSE-NEXT: movdqa 112(%rdi), %xmm4
+; SSE-NEXT: movdqa 128(%rdi), %xmm11
+; SSE-NEXT: movdqa 64(%rdi), %xmm2
+; SSE-NEXT: movdqa 80(%rdi), %xmm12
+; SSE-NEXT: movdqa (%rdi), %xmm3
+; SSE-NEXT: movdqa %xmm3, (%rsp) # 16-byte Spill
+; SSE-NEXT: movdqa 16(%rdi), %xmm15
+; SSE-NEXT: movdqa 32(%rdi), %xmm0
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 48(%rdi), %xmm5
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,2,4,5,6,7]
+; SSE-NEXT: movdqa {{.*#+}} xmm7 = [65535,65535,65535,0,0,0,65535,65535]
+; SSE-NEXT: movdqa %xmm7, %xmm1
+; SSE-NEXT: pandn %xmm0, %xmm1
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,1,0,3]
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm15[2],xmm0[3],xmm15[3]
+; SSE-NEXT: pand %xmm7, %xmm0
+; SSE-NEXT: por %xmm1, %xmm0
+; SSE-NEXT: movdqa %xmm2, %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm2[2,2,3,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm6 = xmm6[0],xmm12[0],xmm6[1],xmm12[1],xmm6[2],xmm12[2],xmm6[3],xmm12[3]
+; SSE-NEXT: movdqa %xmm12, %xmm1
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
; SSE-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa 16(%rdi), %xmm5
-; SSE-NEXT: movdqa 32(%rdi), %xmm9
-; SSE-NEXT: movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa 48(%rdi), %xmm1
-; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $16, %xmm9
-; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm1[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm10 = xmm6[0,1,0,2,4,5,6,7]
-; SSE-NEXT: movdqa %xmm0, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm15 = xmm0[2,2,3,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm15 = xmm15[0],xmm7[0],xmm15[1],xmm7[1],xmm15[2],xmm7[2],xmm15[3],xmm7[3]
-; SSE-NEXT: movdqa %xmm7, %xmm2
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm0[3,0]
; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,0],xmm12[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm12[2,3]
+; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pslld $16, %xmm12
+; SSE-NEXT: psrldq {{.*#+}} xmm3 = xmm3[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm12[0],xmm3[1],xmm12[1],xmm3[2],xmm12[2],xmm3[3],xmm12[3]
+; SSE-NEXT: movdqa %xmm5, %xmm12
+; SSE-NEXT: pshufd {{.*#+}} xmm10 = xmm5[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm10[0,1,0,2,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1],xmm1[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,0]
; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0],xmm7[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm7[2,3]
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pslld $16, %xmm7
-; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm7[0],xmm1[1],xmm7[1],xmm1[2],xmm7[2],xmm1[3],xmm7[3]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm10[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm10 = xmm10[0],xmm9[0],xmm10[1],xmm9[1]
-; SSE-NEXT: movdqa {{.*#+}} xmm3 = [65535,65535,65535,0,0,0,65535,65535]
-; SSE-NEXT: movdqa %xmm3, %xmm0
-; SSE-NEXT: pandn %xmm10, %xmm0
-; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
-; SSE-NEXT: # xmm11 = mem[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm11[0,1,2,3,4,6,6,7]
-; SSE-NEXT: movdqa %xmm5, %xmm2
-; SSE-NEXT: punpckhdq {{.*#+}} xmm7 = xmm7[2],xmm5[2],xmm7[3],xmm5[3]
-; SSE-NEXT: pand %xmm3, %xmm7
-; SSE-NEXT: por %xmm0, %xmm7
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm1[2,0]
-; SSE-NEXT: movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm4, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm4[2,2,3,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm12[0],xmm0[1],xmm12[1],xmm0[2],xmm12[2],xmm0[3],xmm12[3]
-; SSE-NEXT: movdqa %xmm12, %xmm4
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm1[3,0]
-; SSE-NEXT: movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm1, %xmm4
-; SSE-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm12[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm12[2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm11[0,1,1,2,4,5,6,7]
+; SSE-NEXT: movdqa %xmm7, %xmm5
+; SSE-NEXT: movdqa %xmm7, %xmm2
+; SSE-NEXT: pandn %xmm1, %xmm2
+; SSE-NEXT: movdqa 96(%rdi), %xmm1
+; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm1[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm8 = xmm7[0,1,2,3,4,6,6,7]
+; SSE-NEXT: movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: punpckhdq {{.*#+}} xmm8 = xmm8[2],xmm4[2],xmm8[3],xmm4[3]
+; SSE-NEXT: pand %xmm5, %xmm8
+; SSE-NEXT: por %xmm2, %xmm8
+; SSE-NEXT: movdqa 160(%rdi), %xmm3
+; SSE-NEXT: movdqa 176(%rdi), %xmm2
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm3[2,2,3,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE-NEXT: movdqa %xmm2, %xmm1
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm3[3,0]
; SSE-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pslld $16, %xmm12
-; SSE-NEXT: psrldq {{.*#+}} xmm4 = xmm4[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm12[0],xmm4[1],xmm12[1],xmm4[2],xmm12[2],xmm4[3],xmm12[3]
-; SSE-NEXT: movdqa 128(%rdi), %xmm5
+; SSE-NEXT: movdqa %xmm3, %xmm1
+; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,0],xmm2[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm2[2,3]
+; SSE-NEXT: pslld $16, %xmm2
+; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; SSE-NEXT: movdqa 144(%rdi), %xmm13
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm13[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm9 = xmm2[0,1,0,2,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm9[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,1],xmm1[2,0]
+; SSE-NEXT: movdqa %xmm15, %xmm1
+; SSE-NEXT: psrld $16, %xmm1
+; SSE-NEXT: pshufhw $237, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
+; SSE-NEXT: # xmm9 = mem[0,1,2,3,5,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm9 = xmm9[2],xmm1[2],xmm9[3],xmm1[3]
; SSE-NEXT: movdqa %xmm5, %xmm1
-; SSE-NEXT: movdqa %xmm5, %xmm14
-; SSE-NEXT: psrlq $16, %xmm1
-; SSE-NEXT: movdqa 144(%rdi), %xmm5
-; SSE-NEXT: movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm12 = xmm5[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm8 = xmm12[0,1,0,2,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[3,1],xmm8[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm8 = xmm8[0],xmm1[0],xmm8[1],xmm1[1]
-; SSE-NEXT: movdqa %xmm3, %xmm7
-; SSE-NEXT: pandn %xmm8, %xmm7
-; SSE-NEXT: movdqa 112(%rdi), %xmm10
-; SSE-NEXT: movdqa 96(%rdi), %xmm13
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm13[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm9 = xmm5[0,1,2,3,4,6,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm9 = xmm9[2],xmm10[2],xmm9[3],xmm10[3]
-; SSE-NEXT: pand %xmm3, %xmm9
-; SSE-NEXT: por %xmm7, %xmm9
-; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,1],xmm4[2,0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm6[0,1,1,3,4,5,6,7]
-; SSE-NEXT: movdqa %xmm15, %xmm6
-; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[3,1],xmm4[1,3]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm1[1,1,1,1]
-; SSE-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm7[0],xmm4[1],xmm7[1]
-; SSE-NEXT: movdqa %xmm3, %xmm7
-; SSE-NEXT: pandn %xmm4, %xmm7
-; SSE-NEXT: movdqa %xmm2, %xmm4
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
+; SSE-NEXT: pandn %xmm14, %xmm1
+; SSE-NEXT: pand %xmm5, %xmm9
+; SSE-NEXT: por %xmm1, %xmm9
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm10[0,1,1,3,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[3,1],xmm1[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,1],xmm6[2,0]
; SSE-NEXT: psrld $16, %xmm4
-; SSE-NEXT: pshufhw {{.*#+}} xmm8 = xmm11[0,1,2,3,5,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm8 = xmm8[2],xmm4[2],xmm8[3],xmm4[3]
-; SSE-NEXT: pand %xmm3, %xmm8
-; SSE-NEXT: por %xmm7, %xmm8
-; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,1],xmm6[2,0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm12[0,1,1,3,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm4[1,3]
-; SSE-NEXT: movdqa %xmm14, %xmm11
-; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm14[1,1,1,1]
-; SSE-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1]
-; SSE-NEXT: movdqa %xmm10, %xmm6
-; SSE-NEXT: psrld $16, %xmm6
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm5 = xmm5[2],xmm6[2],xmm5[3],xmm6[3]
-; SSE-NEXT: pand %xmm3, %xmm5
-; SSE-NEXT: pandn %xmm4, %xmm3
-; SSE-NEXT: por %xmm5, %xmm3
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm0[2,0]
-; SSE-NEXT: movdqa %xmm1, %xmm4
-; SSE-NEXT: movdqa %xmm1, %xmm14
-; SSE-NEXT: psrldq {{.*#+}} xmm4 = xmm4[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
+; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,5,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm7 = xmm7[2],xmm4[2],xmm7[3],xmm4[3]
+; SSE-NEXT: pand %xmm5, %xmm7
+; SSE-NEXT: pandn %xmm11, %xmm5
+; SSE-NEXT: por %xmm7, %xmm5
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,3,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm1[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,1],xmm0[2,0]
+; SSE-NEXT: movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm14, %xmm1
+; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm12[1,1,1,1]
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm0[0]
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; SSE-NEXT: movdqa {{.*#+}} xmm0 = [65535,65535,65535,0,0,65535,65535,65535]
-; SSE-NEXT: movdqa %xmm0, %xmm6
-; SSE-NEXT: pandn %xmm4, %xmm6
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[1,0],xmm2[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[2,0],xmm2[2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm7[0,2,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm4[1,0,2,3,4,5,6,7]
-; SSE-NEXT: pand %xmm0, %xmm5
-; SSE-NEXT: por %xmm6, %xmm5
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: shufps $132, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = xmm1[0,1],mem[0,2]
-; SSE-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm1[0,1,2,3,4,6,6,7]
+; SSE-NEXT: movdqa %xmm0, %xmm10
+; SSE-NEXT: pandn %xmm1, %xmm10
+; SSE-NEXT: movaps (%rsp), %xmm1 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm15[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm15[2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm1[0,2,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm7[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm7[1,0,2,3,4,5,6,7]
+; SSE-NEXT: pand %xmm0, %xmm2
+; SSE-NEXT: por %xmm10, %xmm2
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT: shufps $132, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
+; SSE-NEXT: # xmm4 = xmm4[0,1],mem[0,2]
+; SSE-NEXT: movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm4[0,1,2,3,4,6,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm7[0,1,2,0]
+; SSE-NEXT: pshufhw {{.*#+}} xmm10 = xmm7[0,1,2,3,4,6,5,4]
+; SSE-NEXT: movdqa {{.*#+}} xmm7 = [65535,65535,65535,65535,65535,0,0,0]
+; SSE-NEXT: movdqa %xmm7, %xmm4
+; SSE-NEXT: pandn %xmm10, %xmm4
+; SSE-NEXT: pand %xmm7, %xmm2
+; SSE-NEXT: por %xmm2, %xmm4
+; SSE-NEXT: movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm11, %xmm2
+; SSE-NEXT: movdqa %xmm11, %xmm14
+; SSE-NEXT: movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: psrldq {{.*#+}} xmm2 = xmm2[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm13[1,1,1,1]
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm4[0]
+; SSE-NEXT: movdqa %xmm0, %xmm4
+; SSE-NEXT: pandn %xmm2, %xmm4
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[1,0],xmm2[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[2,0],xmm2[2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm10[0,2,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm6[1,0,2,3,4,5,6,7]
+; SSE-NEXT: pand %xmm0, %xmm6
+; SSE-NEXT: por %xmm4, %xmm6
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: shufps $132, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = xmm2[0,1],mem[0,2]
+; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm2[0,1,2,3,4,6,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,2,0]
; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,6,5,4]
-; SSE-NEXT: movdqa {{.*#+}} xmm6 = [65535,65535,65535,65535,65535,0,0,0]
-; SSE-NEXT: movdqa %xmm6, %xmm1
-; SSE-NEXT: pandn %xmm4, %xmm1
-; SSE-NEXT: pand %xmm6, %xmm5
-; SSE-NEXT: por %xmm5, %xmm1
-; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm11, %xmm4
-; SSE-NEXT: movdqa %xmm11, (%rsp) # 16-byte Spill
-; SSE-NEXT: psrldq {{.*#+}} xmm4 = xmm4[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm5[0]
-; SSE-NEXT: movdqa %xmm0, %xmm5
+; SSE-NEXT: movdqa %xmm7, %xmm5
; SSE-NEXT: pandn %xmm4, %xmm5
-; SSE-NEXT: movdqa %xmm13, %xmm4
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,0],xmm10[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm10[2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm15 = xmm4[0,2,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm15[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[1,0,2,3,4,5,6,7]
+; SSE-NEXT: pand %xmm7, %xmm6
+; SSE-NEXT: por %xmm6, %xmm5
+; SSE-NEXT: movdqa %xmm12, %xmm4
+; SSE-NEXT: psrlq $48, %xmm4
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm2[2,2,3,3]
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm4[0]
+; SSE-NEXT: movdqa %xmm0, %xmm4
+; SSE-NEXT: pandn %xmm6, %xmm4
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,3,3,4,5,6,7]
; SSE-NEXT: pand %xmm0, %xmm1
-; SSE-NEXT: por %xmm5, %xmm1
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT: shufps $132, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT: # xmm5 = xmm5[0,1],mem[0,2]
-; SSE-NEXT: movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,6,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,1,2,0]
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,6,5,4]
-; SSE-NEXT: movdqa %xmm6, %xmm15
-; SSE-NEXT: pandn %xmm5, %xmm15
-; SSE-NEXT: pand %xmm6, %xmm1
-; SSE-NEXT: por %xmm1, %xmm15
-; SSE-NEXT: movdqa %xmm12, %xmm1
+; SSE-NEXT: por %xmm4, %xmm1
+; SSE-NEXT: pshufhw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
+; SSE-NEXT: # xmm4 = mem[0,1,2,3,7,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,0,2]
+; SSE-NEXT: movdqa %xmm7, %xmm11
+; SSE-NEXT: pandn %xmm4, %xmm11
+; SSE-NEXT: pand %xmm7, %xmm1
+; SSE-NEXT: por %xmm1, %xmm11
+; SSE-NEXT: movdqa %xmm13, %xmm1
; SSE-NEXT: psrlq $48, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm14[2,2,3,3]
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm1[0]
-; SSE-NEXT: movdqa %xmm0, %xmm1
-; SSE-NEXT: pandn %xmm5, %xmm1
-; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm7[3,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm0, %xmm5
-; SSE-NEXT: por %xmm1, %xmm5
-; SSE-NEXT: pshufhw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = mem[0,1,2,3,7,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,2]
-; SSE-NEXT: movdqa %xmm6, %xmm12
-; SSE-NEXT: pandn %xmm1, %xmm12
-; SSE-NEXT: pand %xmm6, %xmm5
-; SSE-NEXT: por %xmm5, %xmm12
-; SSE-NEXT: psrlq $48, %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm11[2,2,3,3]
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm2[0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm4[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm14[2,2,3,3]
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm1[0]
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm10[3,1,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,3,3,4,5,6,7]
; SSE-NEXT: pand %xmm0, %xmm1
-; SSE-NEXT: pandn %xmm5, %xmm0
+; SSE-NEXT: pandn %xmm4, %xmm0
; SSE-NEXT: por %xmm1, %xmm0
; SSE-NEXT: pshufhw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT: # xmm1 = mem[0,1,2,3,7,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,2]
-; SSE-NEXT: movdqa %xmm6, %xmm11
-; SSE-NEXT: pandn %xmm1, %xmm11
-; SSE-NEXT: pand %xmm6, %xmm0
-; SSE-NEXT: por %xmm0, %xmm11
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm4[1,1,1,1]
-; SSE-NEXT: pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = mem[2,3,2,3]
+; SSE-NEXT: movdqa %xmm7, %xmm4
+; SSE-NEXT: pandn %xmm1, %xmm4
+; SSE-NEXT: pand %xmm7, %xmm0
+; SSE-NEXT: por %xmm0, %xmm4
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm15[1,1,1,1]
+; SSE-NEXT: movdqa (%rsp), %xmm10 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm10[2,3,2,3]
; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE-NEXT: pshufd {{.*#+}} xmm14 = xmm14[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm14[0,1,2,3,4,5,4,6]
-; SSE-NEXT: punpckhqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT: # xmm5 = xmm5[1],mem[1]
-; SSE-NEXT: movss {{.*#+}} xmm5 = xmm1[0],xmm5[1,2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm14 = xmm2[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm14[0,1,2,3,4,5,4,6]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm6 = xmm6[1],xmm12[1]
+; SSE-NEXT: movss {{.*#+}} xmm6 = xmm1[0],xmm6[1,2,3]
; SSE-NEXT: pshuflw $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT: # xmm0 = mem[0,2,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,6]
-; SSE-NEXT: movdqa %xmm6, %xmm7
-; SSE-NEXT: pandn %xmm0, %xmm7
-; SSE-NEXT: andps %xmm6, %xmm5
-; SSE-NEXT: por %xmm5, %xmm7
-; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm10[1,1,1,1]
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm13[2,3,2,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
-; SSE-NEXT: pshufd $196, (%rsp), %xmm0 # 16-byte Folded Reload
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5,4,6]
+; SSE-NEXT: movdqa %xmm7, %xmm2
+; SSE-NEXT: pandn %xmm1, %xmm2
+; SSE-NEXT: andps %xmm7, %xmm6
+; SSE-NEXT: por %xmm6, %xmm2
+; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT: # xmm1 = mem[1,1,1,1]
+; SSE-NEXT: pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
+; SSE-NEXT: # xmm6 = mem[2,3,2,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm6 = xmm6[0],xmm1[0],xmm6[1],xmm1[1],xmm6[2],xmm1[2],xmm6[3],xmm1[3]
+; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT: # xmm0 = mem[0,1,0,3]
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,6]
-; SSE-NEXT: punpckhqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT: # xmm0 = xmm0[1],mem[1]
-; SSE-NEXT: movss {{.*#+}} xmm0 = xmm5[0],xmm0[1,2,3]
-; SSE-NEXT: pshuflw $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT: # xmm5 = mem[0,2,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,4,6]
-; SSE-NEXT: movdqa %xmm6, %xmm2
-; SSE-NEXT: pandn %xmm5, %xmm2
-; SSE-NEXT: andps %xmm6, %xmm0
-; SSE-NEXT: por %xmm0, %xmm2
-; SSE-NEXT: psrlq $48, %xmm4
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3]
-; SSE-NEXT: movdqa %xmm0, %xmm5
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: psrld $16, %xmm1
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm14[0,1,2,3,4,5,5,7]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; SSE-NEXT: movss {{.*#+}} xmm0 = xmm5[0],xmm0[1,2,3]
-; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
-; SSE-NEXT: # xmm4 = mem[3,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,5,7]
-; SSE-NEXT: movdqa %xmm6, %xmm5
-; SSE-NEXT: pandn %xmm4, %xmm5
-; SSE-NEXT: andps %xmm6, %xmm0
-; SSE-NEXT: por %xmm0, %xmm5
-; SSE-NEXT: psrlq $48, %xmm10
-; SSE-NEXT: psrldq {{.*#+}} xmm13 = xmm13[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm13 = xmm13[0],xmm10[0],xmm13[1],xmm10[1],xmm13[2],xmm10[2],xmm13[3],xmm10[3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5,4,6]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm13[1]
+; SSE-NEXT: movss {{.*#+}} xmm1 = xmm6[0],xmm1[1,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm3[0,2,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5,4,6]
+; SSE-NEXT: movdqa %xmm7, %xmm0
+; SSE-NEXT: pandn %xmm6, %xmm0
+; SSE-NEXT: andps %xmm7, %xmm1
+; SSE-NEXT: por %xmm1, %xmm0
+; SSE-NEXT: psrlq $48, %xmm15
+; SSE-NEXT: psrldq {{.*#+}} xmm10 = xmm10[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm10 = xmm10[0],xmm15[0],xmm10[1],xmm15[1],xmm10[2],xmm15[2],xmm10[3],xmm15[3]
+; SSE-NEXT: psrld $16, %xmm12
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm14[0,1,2,3,4,5,5,7]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm12[1]
+; SSE-NEXT: movss {{.*#+}} xmm1 = xmm10[0],xmm1[1,2,3]
+; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
+; SSE-NEXT: # xmm6 = mem[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm12 = xmm6[0,1,2,3,4,4,5,7]
+; SSE-NEXT: movdqa %xmm7, %xmm14
+; SSE-NEXT: pandn %xmm12, %xmm14
+; SSE-NEXT: andps %xmm7, %xmm1
+; SSE-NEXT: por %xmm1, %xmm14
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT: psrlq $48, %xmm6
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: psrld $16, %xmm1
-; SSE-NEXT: pshufhw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT: # xmm0 = mem[0,1,2,3,4,5,5,7]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; SSE-NEXT: movss {{.*#+}} xmm0 = xmm13[0],xmm0[1,2,3]
-; SSE-NEXT: andps %xmm6, %xmm0
-; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
-; SSE-NEXT: # xmm4 = mem[3,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,5,7]
-; SSE-NEXT: pandn %xmm4, %xmm6
-; SSE-NEXT: por %xmm0, %xmm6
-; SSE-NEXT: movaps %xmm9, 16(%rsi)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, (%rsi)
-; SSE-NEXT: movaps %xmm3, 16(%rdx)
-; SSE-NEXT: movaps %xmm8, (%rdx)
-; SSE-NEXT: movdqa %xmm15, 16(%rcx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, (%rcx)
-; SSE-NEXT: movdqa %xmm11, 16(%r8)
-; SSE-NEXT: movdqa %xmm12, (%r8)
-; SSE-NEXT: movdqa %xmm2, 16(%r9)
-; SSE-NEXT: movdqa %xmm7, (%r9)
+; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1],xmm1[2],xmm6[2],xmm1[3],xmm6[3]
+; SSE-NEXT: movdqa %xmm1, %xmm10
+; SSE-NEXT: psrld $16, %xmm13
+; SSE-NEXT: pshufhw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT: # xmm1 = mem[0,1,2,3,4,5,5,7]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm13[1]
+; SSE-NEXT: movss {{.*#+}} xmm1 = xmm10[0],xmm1[1,2,3]
+; SSE-NEXT: andps %xmm7, %xmm1
+; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,5,7]
+; SSE-NEXT: pandn %xmm3, %xmm7
+; SSE-NEXT: por %xmm1, %xmm7
+; SSE-NEXT: movaps %xmm8, 16(%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, (%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, 16(%rdx)
+; SSE-NEXT: movaps %xmm9, (%rdx)
+; SSE-NEXT: movdqa %xmm5, 16(%rcx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT: movaps %xmm1, (%rcx)
+; SSE-NEXT: movdqa %xmm4, 16(%r8)
+; SSE-NEXT: movdqa %xmm11, (%r8)
+; SSE-NEXT: movdqa %xmm0, 16(%r9)
+; SSE-NEXT: movdqa %xmm2, (%r9)
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT: movdqa %xmm6, 16(%rax)
-; SSE-NEXT: movdqa %xmm5, (%rax)
-; SSE-NEXT: addq $88, %rsp
+; SSE-NEXT: movdqa %xmm7, 16(%rax)
+; SSE-NEXT: movdqa %xmm14, (%rax)
+; SSE-NEXT: addq $72, %rsp
; SSE-NEXT: retq
;
; AVX1-LABEL: vf16:
@@ -1656,234 +1636,227 @@ define void @vf32(<192 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-LABEL: vf32:
; SSE: # %bb.0:
; SSE-NEXT: subq $456, %rsp # imm = 0x1C8
-; SSE-NEXT: movdqa 256(%rdi), %xmm2
-; SSE-NEXT: movdqa 272(%rdi), %xmm13
-; SSE-NEXT: movdqa 64(%rdi), %xmm0
-; SSE-NEXT: movdqa 80(%rdi), %xmm6
-; SSE-NEXT: movdqa (%rdi), %xmm4
+; SSE-NEXT: movdqa 208(%rdi), %xmm8
+; SSE-NEXT: movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 224(%rdi), %xmm6
+; SSE-NEXT: movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 64(%rdi), %xmm2
+; SSE-NEXT: movdqa 80(%rdi), %xmm7
+; SSE-NEXT: movdqa (%rdi), %xmm3
+; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 16(%rdi), %xmm4
; SSE-NEXT: movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa 16(%rdi), %xmm7
-; SSE-NEXT: movdqa 32(%rdi), %xmm9
-; SSE-NEXT: movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa 48(%rdi), %xmm1
-; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $16, %xmm9
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,3,2,3]
-; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshuflw {{.*#+}} xmm10 = xmm1[0,1,0,2,4,5,6,7]
+; SSE-NEXT: movdqa 32(%rdi), %xmm0
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 48(%rdi), %xmm5
+; SSE-NEXT: movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,2,4,5,6,7]
+; SSE-NEXT: movdqa {{.*#+}} xmm10 = [65535,65535,65535,0,0,0,65535,65535]
+; SSE-NEXT: movdqa %xmm10, %xmm1
+; SSE-NEXT: pandn %xmm0, %xmm1
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm3[0,1,0,3]
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
+; SSE-NEXT: pand %xmm10, %xmm0
+; SSE-NEXT: por %xmm1, %xmm0
; SSE-NEXT: movdqa %xmm0, %xmm1
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm0[2,2,3,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm6[0],xmm3[1],xmm6[1],xmm3[2],xmm6[2],xmm3[3],xmm6[3]
-; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm6, %xmm3
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm0[3,0]
-; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0],xmm6[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm6[2,3]
+; SSE-NEXT: movdqa %xmm2, %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,2,3,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1],xmm0[2],xmm7[2],xmm0[3],xmm7[3]
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm7, %xmm0
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm2[3,0]
; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pslld $16, %xmm6
-; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1],xmm1[2],xmm6[2],xmm1[3],xmm6[3]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm10[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm10 = xmm10[0],xmm9[0],xmm10[1],xmm9[1]
-; SSE-NEXT: movdqa {{.*#+}} xmm14 = [65535,65535,65535,0,0,0,65535,65535]
-; SSE-NEXT: movdqa %xmm14, %xmm0
-; SSE-NEXT: pandn %xmm10, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm4[0,1,0,3]
-; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm7[2],xmm3[3],xmm7[3]
-; SSE-NEXT: pand %xmm14, %xmm3
-; SSE-NEXT: por %xmm0, %xmm3
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm1[2,0]
-; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm2, %xmm0
-; SSE-NEXT: pshufd {{.*#+}} xmm10 = xmm2[2,2,3,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm10 = xmm10[0],xmm13[0],xmm10[1],xmm13[1],xmm10[2],xmm13[2],xmm10[3],xmm13[3]
-; SSE-NEXT: movdqa %xmm13, %xmm1
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
-; SSE-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm2, %xmm1
; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0],xmm13[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm13[2,3]
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pslld $16, %xmm13
-; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm13[0],xmm1[1],xmm13[1],xmm1[2],xmm13[2],xmm1[3],xmm13[3]
-; SSE-NEXT: movdqa 224(%rdi), %xmm2
-; SSE-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $16, %xmm2
-; SSE-NEXT: movdqa 240(%rdi), %xmm0
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,0],xmm7[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm7[2,3]
+; SSE-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pslld $16, %xmm7
+; SSE-NEXT: psrldq {{.*#+}} xmm3 = xmm3[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm7[0],xmm3[1],xmm7[1],xmm3[2],xmm7[2],xmm3[3],xmm7[3]
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm5[0,3,2,3]
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm11 = xmm0[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm11[0,1,0,2,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm3[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
-; SSE-NEXT: movdqa %xmm14, %xmm2
-; SSE-NEXT: pandn %xmm3, %xmm2
-; SSE-NEXT: movdqa 208(%rdi), %xmm3
-; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1],xmm0[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,0]
+; SSE-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm6[0,1,1,2,4,5,6,7]
+; SSE-NEXT: movdqa %xmm10, %xmm1
+; SSE-NEXT: pandn %xmm0, %xmm1
; SSE-NEXT: movdqa 192(%rdi), %xmm0
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,0,3]
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE-NEXT: pand %xmm14, %xmm0
-; SSE-NEXT: por %xmm2, %xmm0
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa 160(%rdi), %xmm0
-; SSE-NEXT: movdqa 176(%rdi), %xmm2
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSE-NEXT: movdqa %xmm2, %xmm3
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm0[3,0]
-; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm8[2],xmm0[3],xmm8[3]
+; SSE-NEXT: pand %xmm10, %xmm0
+; SSE-NEXT: por %xmm1, %xmm0
; SSE-NEXT: movdqa %xmm0, %xmm3
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0],xmm2[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm2[2,3]
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pslld $16, %xmm2
-; SSE-NEXT: psrldq {{.*#+}} xmm3 = xmm3[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; SSE-NEXT: movdqa 128(%rdi), %xmm2
-; SSE-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $16, %xmm2
-; SSE-NEXT: movdqa 144(%rdi), %xmm0
+; SSE-NEXT: movdqa 256(%rdi), %xmm4
+; SSE-NEXT: movdqa 272(%rdi), %xmm0
+; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm4[2,2,3,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE-NEXT: movdqa %xmm0, %xmm1
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm4[3,0]
+; SSE-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm4, %xmm1
+; SSE-NEXT: movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,0],xmm0[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm0[2,3]
+; SSE-NEXT: movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pslld $16, %xmm0
+; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; SSE-NEXT: movdqa 240(%rdi), %xmm0
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm0[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm8[0,1,0,2,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1],xmm4[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; SSE-NEXT: movdqa %xmm14, %xmm2
-; SSE-NEXT: pandn %xmm4, %xmm2
-; SSE-NEXT: movdqa 112(%rdi), %xmm4
-; SSE-NEXT: movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,3,2,3]
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,0,2,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm0[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm1[2,0]
+; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 128(%rdi), %xmm0
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,2,4,5,6,7]
+; SSE-NEXT: movdqa %xmm10, %xmm3
+; SSE-NEXT: pandn %xmm0, %xmm3
+; SSE-NEXT: movdqa 112(%rdi), %xmm1
+; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: movdqa 96(%rdi), %xmm0
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm15 = xmm0[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm15[0,1,2,3,4,6,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
-; SSE-NEXT: pand %xmm14, %xmm0
-; SSE-NEXT: por %xmm2, %xmm0
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,0]
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa 352(%rdi), %xmm0
-; SSE-NEXT: movdqa 368(%rdi), %xmm3
-; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm0[2,2,3,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; SSE-NEXT: movdqa %xmm3, %xmm4
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm0[3,0]
-; SSE-NEXT: movaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm0, %xmm5
-; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,0],xmm3[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm3[2,3]
+; SSE-NEXT: pshufd {{.*#+}} xmm13 = xmm0[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm13[0,1,2,3,4,6,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE-NEXT: pand %xmm10, %xmm0
+; SSE-NEXT: por %xmm3, %xmm0
+; SSE-NEXT: movdqa %xmm0, %xmm1
+; SSE-NEXT: movdqa 160(%rdi), %xmm5
+; SSE-NEXT: movdqa 176(%rdi), %xmm3
+; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm5[2,2,3,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm6 = xmm6[0],xmm3[0],xmm6[1],xmm3[1],xmm6[2],xmm3[2],xmm6[3],xmm3[3]
+; SSE-NEXT: movdqa %xmm3, %xmm0
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm5[3,0]
; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm5, %xmm4
+; SSE-NEXT: movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[1,0],xmm3[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[2,0],xmm3[2,3]
+; SSE-NEXT: movaps %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pslld $16, %xmm3
-; SSE-NEXT: psrldq {{.*#+}} xmm5 = xmm5[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm3[0],xmm5[1],xmm3[1],xmm5[2],xmm3[2],xmm5[3],xmm3[3]
-; SSE-NEXT: movdqa 320(%rdi), %xmm3
-; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $16, %xmm3
-; SSE-NEXT: movdqa 336(%rdi), %xmm0
+; SSE-NEXT: psrldq {{.*#+}} xmm4 = xmm4[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; SSE-NEXT: movdqa 144(%rdi), %xmm0
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm9 = xmm0[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm12 = xmm9[0,1,0,2,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[3,1],xmm12[1,3]
-; SSE-NEXT: punpckldq {{.*#+}} xmm12 = xmm12[0],xmm3[0],xmm12[1],xmm3[1]
-; SSE-NEXT: movdqa %xmm14, %xmm3
-; SSE-NEXT: pandn %xmm12, %xmm3
-; SSE-NEXT: movdqa 304(%rdi), %xmm13
-; SSE-NEXT: movdqa 288(%rdi), %xmm0
+; SSE-NEXT: pshufd {{.*#+}} xmm14 = xmm0[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm14[0,1,0,2,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[3,1],xmm3[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,0]
+; SSE-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa 320(%rdi), %xmm0
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshufd {{.*#+}} xmm12 = xmm0[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm12[0,1,2,3,4,6,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm6 = xmm6[2],xmm13[2],xmm6[3],xmm13[3]
-; SSE-NEXT: pand %xmm14, %xmm6
-; SSE-NEXT: por %xmm3, %xmm6
-; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,1],xmm5[2,0]
-; SSE-NEXT: movaps %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshuflw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[0,1,1,3,4,5,6,7]
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1],xmm3[1,3]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm4[1,1,1,1]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
-; SSE-NEXT: movdqa %xmm14, %xmm5
-; SSE-NEXT: pandn %xmm3, %xmm5
-; SSE-NEXT: movdqa %xmm7, %xmm6
-; SSE-NEXT: movdqa %xmm7, %xmm3
-; SSE-NEXT: psrld $16, %xmm3
-; SSE-NEXT: pshufhw $237, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
-; SSE-NEXT: # xmm7 = mem[0,1,2,3,5,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm7 = xmm7[2],xmm3[2],xmm7[3],xmm3[3]
-; SSE-NEXT: pand %xmm14, %xmm7
-; SSE-NEXT: por %xmm5, %xmm7
-; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm0[2,0]
-; SSE-NEXT: movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm11[0,1,1,3,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[3,1],xmm3[1,3]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,1,1]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
-; SSE-NEXT: movdqa %xmm14, %xmm5
-; SSE-NEXT: pandn %xmm3, %xmm5
+; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm0[0,1,1,2,4,5,6,7]
+; SSE-NEXT: movdqa %xmm10, %xmm4
+; SSE-NEXT: pandn %xmm3, %xmm4
+; SSE-NEXT: movdqa 304(%rdi), %xmm15
+; SSE-NEXT: movdqa 288(%rdi), %xmm1
+; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm1[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm5[0,1,2,3,4,6,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm15[2],xmm1[3],xmm15[3]
+; SSE-NEXT: pand %xmm10, %xmm1
+; SSE-NEXT: por %xmm4, %xmm1
+; SSE-NEXT: movdqa %xmm1, %xmm3
+; SSE-NEXT: movdqa 352(%rdi), %xmm12
+; SSE-NEXT: movdqa 368(%rdi), %xmm11
+; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm12[2,2,3,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm8 = xmm8[0],xmm11[0],xmm8[1],xmm11[1],xmm8[2],xmm11[2],xmm8[3],xmm11[3]
+; SSE-NEXT: movdqa %xmm11, %xmm0
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm12[3,0]
+; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm12, %xmm9
+; SSE-NEXT: movaps %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[1,0],xmm11[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm12 = xmm12[2,0],xmm11[2,3]
+; SSE-NEXT: pslld $16, %xmm11
+; SSE-NEXT: psrldq {{.*#+}} xmm9 = xmm9[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm9 = xmm9[0],xmm11[0],xmm9[1],xmm11[1],xmm9[2],xmm11[2],xmm9[3],xmm11[3]
+; SSE-NEXT: movdqa 336(%rdi), %xmm0
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm0[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm1[0,1,0,2,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[3,1],xmm7[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm9[2,0]
+; SSE-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT: movdqa %xmm3, %xmm7
+; SSE-NEXT: psrld $16, %xmm7
+; SSE-NEXT: pshufhw $237, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
+; SSE-NEXT: # xmm11 = mem[0,1,2,3,5,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm11 = xmm11[2],xmm7[2],xmm11[3],xmm7[3]
+; SSE-NEXT: movdqa %xmm10, %xmm7
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT: pandn %xmm9, %xmm7
+; SSE-NEXT: pand %xmm10, %xmm11
+; SSE-NEXT: por %xmm7, %xmm11
+; SSE-NEXT: pshuflw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; SSE-NEXT: # xmm7 = mem[0,1,1,3,4,5,6,7]
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[3,1],xmm7[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm11 = xmm11[0,1],xmm4[2,0]
+; SSE-NEXT: movaps %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT: movdqa %xmm11, %xmm3
-; SSE-NEXT: psrld $16, %xmm3
+; SSE-NEXT: movdqa %xmm11, %xmm7
+; SSE-NEXT: psrld $16, %xmm7
; SSE-NEXT: pshufhw $237, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT: # xmm0 = mem[0,1,2,3,5,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE-NEXT: pand %xmm14, %xmm0
-; SSE-NEXT: por %xmm5, %xmm0
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm10[2,0]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm7[2],xmm0[3],xmm7[3]
+; SSE-NEXT: movdqa %xmm10, %xmm7
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT: pandn %xmm4, %xmm7
+; SSE-NEXT: pand %xmm10, %xmm0
+; SSE-NEXT: por %xmm7, %xmm0
+; SSE-NEXT: pshuflw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; SSE-NEXT: # xmm7 = mem[0,1,1,3,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm7[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,0]
; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm8[0,1,1,3,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1],xmm3[1,3]
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT: # xmm5 = mem[1,1,1,1]
-; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
-; SSE-NEXT: movdqa %xmm14, %xmm5
-; SSE-NEXT: pandn %xmm3, %xmm5
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT: psrld $16, %xmm3
-; SSE-NEXT: pshufhw {{.*#+}} xmm15 = xmm15[0,1,2,3,5,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm15 = xmm15[2],xmm3[2],xmm15[3],xmm3[3]
-; SSE-NEXT: pand %xmm14, %xmm15
-; SSE-NEXT: por %xmm5, %xmm15
-; SSE-NEXT: shufps {{.*#+}} xmm15 = xmm15[0,1],xmm1[2,0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm9[0,1,1,3,4,5,6,7]
-; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1],xmm1[1,3]
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,1,1]
-; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
-; SSE-NEXT: movdqa %xmm13, %xmm3
-; SSE-NEXT: psrld $16, %xmm3
-; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm12[0,1,2,3,5,7,6,7]
-; SSE-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE-NEXT: pand %xmm14, %xmm0
-; SSE-NEXT: pandn %xmm1, %xmm14
-; SSE-NEXT: por %xmm0, %xmm14
-; SSE-NEXT: shufps {{.*#+}} xmm14 = xmm14[0,1],xmm2[2,0]
-; SSE-NEXT: movdqa %xmm4, %xmm0
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: psrld $16, %xmm2
+; SSE-NEXT: pshufhw {{.*#+}} xmm13 = xmm13[0,1,2,3,5,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm13 = xmm13[2],xmm2[2],xmm13[3],xmm2[3]
+; SSE-NEXT: movdqa %xmm10, %xmm2
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT: pandn %xmm7, %xmm2
+; SSE-NEXT: pand %xmm10, %xmm13
+; SSE-NEXT: por %xmm2, %xmm13
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm14[0,1,1,3,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[3,1],xmm2[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm13 = xmm13[0,1],xmm6[2,0]
+; SSE-NEXT: movdqa %xmm15, %xmm0
+; SSE-NEXT: psrld $16, %xmm0
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm5[0,1,2,3,5,7,6,7]
+; SSE-NEXT: punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE-NEXT: pand %xmm10, %xmm2
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
+; SSE-NEXT: pandn %xmm14, %xmm10
+; SSE-NEXT: por %xmm2, %xmm10
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm1[0,1,1,3,4,5,6,7]
+; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[3,1],xmm0[1,3]
+; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[0,1],xmm8[2,0]
+; SSE-NEXT: movaps %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm9, %xmm0
; SSE-NEXT: psrldq {{.*#+}} xmm0 = xmm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm10[1,1,1,1]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[1,1,1,1]
; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; SSE-NEXT: movdqa {{.*#+}} xmm1 = [65535,65535,65535,0,0,65535,65535,65535]
; SSE-NEXT: movdqa %xmm1, %xmm2
; SSE-NEXT: pandn %xmm0, %xmm2
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT: movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,0],xmm6[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm6[2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm4[0,2,2,3,4,5,6,7]
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[1,0],xmm3[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[2,0],xmm3[2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm5[0,2,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm0[1,0,2,3,4,5,6,7]
; SSE-NEXT: pand %xmm1, %xmm3
@@ -1895,23 +1868,23 @@ define void @vf32(<192 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,2,0]
; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm0[0,1,2,3,4,6,5,4]
-; SSE-NEXT: movdqa {{.*#+}} xmm12 = [65535,65535,65535,65535,65535,0,0,0]
-; SSE-NEXT: movdqa %xmm12, %xmm5
-; SSE-NEXT: pandn %xmm2, %xmm5
-; SSE-NEXT: pand %xmm12, %xmm3
-; SSE-NEXT: por %xmm3, %xmm5
-; SSE-NEXT: movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm7, %xmm2
+; SSE-NEXT: movdqa {{.*#+}} xmm10 = [65535,65535,65535,65535,65535,0,0,0]
+; SSE-NEXT: movdqa %xmm10, %xmm6
+; SSE-NEXT: pandn %xmm2, %xmm6
+; SSE-NEXT: pand %xmm10, %xmm3
+; SSE-NEXT: por %xmm3, %xmm6
+; SSE-NEXT: movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm4, %xmm2
; SSE-NEXT: psrldq {{.*#+}} xmm2 = xmm2[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[1,1,1,1]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm8[1,1,1,1]
; SSE-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
; SSE-NEXT: movdqa %xmm1, %xmm3
; SSE-NEXT: pandn %xmm2, %xmm3
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[1,0],xmm11[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[2,0],xmm11[2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm5[0,2,2,3,4,5,6,7]
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[1,0],xmm11[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[2,0],xmm11[2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm6[0,2,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[1,0,2,3,4,5,6,7]
; SSE-NEXT: pand %xmm1, %xmm2
@@ -1923,94 +1896,73 @@ define void @vf32(<192 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm0[0,1,2,3,4,6,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,2,0]
; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,5,4]
-; SSE-NEXT: movdqa %xmm12, %xmm6
-; SSE-NEXT: pandn %xmm3, %xmm6
-; SSE-NEXT: pand %xmm12, %xmm2
-; SSE-NEXT: por %xmm2, %xmm6
-; SSE-NEXT: movdqa %xmm6, (%rsp) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm8, %xmm2
+; SSE-NEXT: movdqa %xmm10, %xmm4
+; SSE-NEXT: pandn %xmm3, %xmm4
+; SSE-NEXT: pand %xmm10, %xmm2
+; SSE-NEXT: por %xmm2, %xmm4
+; SSE-NEXT: movdqa %xmm4, (%rsp) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm14, %xmm2
; SSE-NEXT: psrldq {{.*#+}} xmm2 = xmm2[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm9[1,1,1,1]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm14[1,1,1,1]
; SSE-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
; SSE-NEXT: movdqa %xmm1, %xmm3
; SSE-NEXT: pandn %xmm2, %xmm3
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; SSE-NEXT: movdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[1,0],xmm13[0,0]
-; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[2,0],xmm13[2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm8[0,2,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm7[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm7[1,0,2,3,4,5,6,7]
-; SSE-NEXT: pand %xmm1, %xmm7
-; SSE-NEXT: por %xmm3, %xmm7
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,0],xmm15[0,0]
+; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm15[2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm2[0,2,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[1,0,2,3,4,5,6,7]
+; SSE-NEXT: pand %xmm1, %xmm4
+; SSE-NEXT: por %xmm3, %xmm4
; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT: shufps $132, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; SSE-NEXT: # xmm0 = xmm0[0,1],mem[0,2]
; SSE-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm0[0,1,2,3,4,6,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,2,0]
-; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,6,5,4]
-; SSE-NEXT: movdqa %xmm12, %xmm2
-; SSE-NEXT: pandn %xmm3, %xmm2
-; SSE-NEXT: pand %xmm12, %xmm7
-; SSE-NEXT: por %xmm7, %xmm2
-; SSE-NEXT: movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT: movdqa %xmm11, %xmm3
+; SSE-NEXT: pshufhw {{.*#+}} xmm11 = xmm3[0,1,2,3,4,6,5,4]
+; SSE-NEXT: movdqa %xmm10, %xmm3
+; SSE-NEXT: pandn %xmm11, %xmm3
+; SSE-NEXT: pand %xmm10, %xmm4
+; SSE-NEXT: por %xmm4, %xmm3
+; SSE-NEXT: movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm7, %xmm3
; SSE-NEXT: psrldq {{.*#+}} xmm3 = xmm3[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm13[1,1,1,1]
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm7[0]
-; SSE-NEXT: movdqa %xmm1, %xmm7
-; SSE-NEXT: pandn %xmm3, %xmm7
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm11[1,1,1,1]
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
+; SSE-NEXT: movdqa %xmm1, %xmm4
+; SSE-NEXT: pandn %xmm3, %xmm4
; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,0],xmm0[0,0]
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm0[2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm3[0,2,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm6[1,0,2,3,4,5,6,7]
-; SSE-NEXT: pand %xmm1, %xmm6
-; SSE-NEXT: por %xmm7, %xmm6
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT: shufps $132, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT: # xmm2 = xmm2[0,1],mem[0,2]
-; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm2[0,1,2,3,4,6,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm7[0,1,2,0]
-; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,6,5,4]
-; SSE-NEXT: movdqa %xmm12, %xmm0
-; SSE-NEXT: pandn %xmm7, %xmm0
-; SSE-NEXT: pand %xmm12, %xmm6
-; SSE-NEXT: por %xmm6, %xmm0
-; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $48, %xmm10
-; SSE-NEXT: pshufd $250, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
-; SSE-NEXT: # xmm7 = mem[2,2,3,3]
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm7 = xmm7[0],xmm10[0]
-; SSE-NEXT: movdqa %xmm1, %xmm6
-; SSE-NEXT: pandn %xmm7, %xmm6
-; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,3,2,3]
-; SSE-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,1,3,3,4,5,6,7]
-; SSE-NEXT: pand %xmm1, %xmm4
-; SSE-NEXT: por %xmm6, %xmm4
-; SSE-NEXT: pshufhw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT: # xmm6 = mem[0,1,2,3,7,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,1,0,2]
-; SSE-NEXT: movdqa %xmm12, %xmm0
-; SSE-NEXT: pandn %xmm6, %xmm0
-; SSE-NEXT: pand %xmm12, %xmm4
-; SSE-NEXT: por %xmm4, %xmm0
+; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm3[0,2,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm7[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm7 = xmm7[1,0,2,3,4,5,6,7]
+; SSE-NEXT: pand %xmm1, %xmm7
+; SSE-NEXT: por %xmm4, %xmm7
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
+; SSE-NEXT: shufps $132, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
+; SSE-NEXT: # xmm15 = xmm15[0,1],mem[0,2]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm15[0,1,2,3,4,6,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,2,0]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,6,5,4]
+; SSE-NEXT: movdqa %xmm10, %xmm0
+; SSE-NEXT: pandn %xmm4, %xmm0
+; SSE-NEXT: pand %xmm10, %xmm7
+; SSE-NEXT: por %xmm7, %xmm0
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT: movdqa %xmm7, %xmm4
+; SSE-NEXT: movdqa %xmm9, %xmm4
; SSE-NEXT: psrlq $48, %xmm4
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm10[2,2,3,3]
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm6 = xmm6[0],xmm4[0]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm9[2,2,3,3]
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm7 = xmm7[0],xmm4[0]
; SSE-NEXT: movdqa %xmm1, %xmm4
-; SSE-NEXT: pandn %xmm6, %xmm4
+; SSE-NEXT: pandn %xmm7, %xmm4
; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[3,1,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,1,3,3,4,5,6,7]
@@ -2019,14 +1971,34 @@ define void @vf32(<192 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshufhw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
; SSE-NEXT: # xmm4 = mem[0,1,2,3,7,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,0,2]
-; SSE-NEXT: movdqa %xmm12, %xmm0
+; SSE-NEXT: movdqa %xmm10, %xmm0
; SSE-NEXT: pandn %xmm4, %xmm0
-; SSE-NEXT: pand %xmm12, %xmm5
+; SSE-NEXT: pand %xmm10, %xmm5
; SSE-NEXT: por %xmm5, %xmm0
; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: movdqa %xmm13, %xmm4
+; SSE-NEXT: psrlq $48, %xmm8
+; SSE-NEXT: pshufd $250, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
+; SSE-NEXT: # xmm5 = mem[2,2,3,3]
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm8[0]
+; SSE-NEXT: movdqa %xmm1, %xmm4
+; SSE-NEXT: pandn %xmm5, %xmm4
+; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm6[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm5[0,3,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,1,3,3,4,5,6,7]
+; SSE-NEXT: pand %xmm1, %xmm5
+; SSE-NEXT: por %xmm4, %xmm5
+; SSE-NEXT: pshufhw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
+; SSE-NEXT: # xmm4 = mem[0,1,2,3,7,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,0,2]
+; SSE-NEXT: movdqa %xmm10, %xmm0
+; SSE-NEXT: pandn %xmm4, %xmm0
+; SSE-NEXT: pand %xmm10, %xmm5
+; SSE-NEXT: por %xmm5, %xmm0
+; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: movdqa %xmm11, %xmm4
; SSE-NEXT: psrlq $48, %xmm4
-; SSE-NEXT: pshufd {{.*#+}} xmm5 = xmm11[2,2,3,3]
+; SSE-NEXT: pshufd $250, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
+; SSE-NEXT: # xmm5 = mem[2,2,3,3]
; SSE-NEXT: punpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm4[0]
; SSE-NEXT: movdqa %xmm1, %xmm4
; SSE-NEXT: pandn %xmm5, %xmm4
@@ -2035,18 +2007,18 @@ define void @vf32(<192 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,1,3,3,4,5,6,7]
; SSE-NEXT: pand %xmm1, %xmm3
; SSE-NEXT: por %xmm4, %xmm3
-; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm2[0,1,2,3,7,5,6,7]
+; SSE-NEXT: pshufhw {{.*#+}} xmm4 = xmm15[0,1,2,3,7,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,1,0,2]
-; SSE-NEXT: movdqa %xmm12, %xmm0
-; SSE-NEXT: pandn %xmm4, %xmm0
-; SSE-NEXT: pand %xmm12, %xmm3
-; SSE-NEXT: por %xmm3, %xmm0
-; SSE-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT: psrlq $48, %xmm9
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT: pshufd {{.*#+}} xmm4 = xmm5[2,2,3,3]
-; SSE-NEXT: punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm9[0]
-; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm8[3,1,2,3,4,5,6,7]
+; SSE-NEXT: movdqa %xmm10, %xmm11
+; SSE-NEXT: pandn %xmm4, %xmm11
+; SSE-NEXT: pand %xmm10, %xmm3
+; SSE-NEXT: por %xmm3, %xmm11
+; SSE-NEXT: movdqa %xmm14, %xmm3
+; SSE-NEXT: psrlq $48, %xmm3
+; SSE-NEXT: pshufd $250, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
+; SSE-NEXT: # xmm4 = mem[2,2,3,3]
+; SSE-NEXT: punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm3[0]
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[3,1,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,3,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,1,3,3,4,5,6,7]
; SSE-NEXT: pand %xmm1, %xmm2
@@ -2055,50 +2027,32 @@ define void @vf32(<192 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: pshufhw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
; SSE-NEXT: # xmm2 = mem[0,1,2,3,7,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,1,0,2]
-; SSE-NEXT: movdqa %xmm12, %xmm13
-; SSE-NEXT: pandn %xmm2, %xmm13
-; SSE-NEXT: pand %xmm12, %xmm1
-; SSE-NEXT: por %xmm1, %xmm13
+; SSE-NEXT: movdqa %xmm10, %xmm14
+; SSE-NEXT: pandn %xmm2, %xmm14
+; SSE-NEXT: pand %xmm10, %xmm1
+; SSE-NEXT: por %xmm1, %xmm14
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
; SSE-NEXT: pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
; SSE-NEXT: # xmm2 = mem[2,3,2,3]
; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
-; SSE-NEXT: # xmm9 = mem[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm9[0,1,2,3,4,5,4,6]
+; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm9[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm8[0,1,2,3,4,5,4,6]
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
; SSE-NEXT: punpckhqdq {{.*#+}} xmm3 = xmm3[1],xmm6[1]
; SSE-NEXT: movss {{.*#+}} xmm3 = xmm2[0],xmm3[1,2,3]
-; SSE-NEXT: pshuflw $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT: # xmm2 = mem[0,2,2,3,4,5,6,7]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm7[0,2,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,6]
-; SSE-NEXT: movdqa %xmm12, %xmm11
-; SSE-NEXT: pandn %xmm2, %xmm11
-; SSE-NEXT: andps %xmm12, %xmm3
-; SSE-NEXT: por %xmm3, %xmm11
+; SSE-NEXT: movdqa %xmm10, %xmm9
+; SSE-NEXT: pandn %xmm2, %xmm9
+; SSE-NEXT: andps %xmm10, %xmm3
+; SSE-NEXT: por %xmm3, %xmm9
; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
; SSE-NEXT: # xmm2 = mem[1,1,1,1]
-; SSE-NEXT: pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[2,3,2,3]
-; SSE-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; SSE-NEXT: pshufd {{.*#+}} xmm8 = xmm10[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm8[0,1,2,3,4,5,4,6]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm7[1]
-; SSE-NEXT: movss {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3]
-; SSE-NEXT: pshuflw $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[0,2,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,4,6]
-; SSE-NEXT: movdqa %xmm12, %xmm10
-; SSE-NEXT: pandn %xmm3, %xmm10
-; SSE-NEXT: andps %xmm12, %xmm2
-; SSE-NEXT: por %xmm2, %xmm10
-; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT: # xmm2 = mem[1,1,1,1]
-; SSE-NEXT: pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT: # xmm3 = mem[2,3,2,3]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
+; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm15[2,3,2,3]
; SSE-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
; SSE-NEXT: # xmm1 = mem[0,1,0,3]
@@ -2111,16 +2065,17 @@ define void @vf32(<192 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: # xmm3 = mem[0,2,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,0,3]
; SSE-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,4,6]
-; SSE-NEXT: movdqa %xmm12, %xmm7
-; SSE-NEXT: pandn %xmm3, %xmm7
-; SSE-NEXT: andps %xmm12, %xmm2
-; SSE-NEXT: por %xmm2, %xmm7
+; SSE-NEXT: movdqa %xmm10, %xmm5
+; SSE-NEXT: pandn %xmm3, %xmm5
+; SSE-NEXT: andps %xmm10, %xmm2
+; SSE-NEXT: por %xmm2, %xmm5
; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
; SSE-NEXT: # xmm2 = mem[1,1,1,1]
; SSE-NEXT: pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
; SSE-NEXT: # xmm3 = mem[2,3,2,3]
; SSE-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm5[0,1,0,3]
+; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT: # xmm1 = mem[0,1,0,3]
; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,5,4,6]
; SSE-NEXT: punpckhqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
@@ -2130,122 +2085,138 @@ define void @vf32(<192 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.
; SSE-NEXT: # xmm3 = mem[0,2,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,0,3]
; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm3[0,1,2,3,4,5,4,6]
-; SSE-NEXT: movdqa %xmm12, %xmm4
+; SSE-NEXT: movdqa %xmm10, %xmm4
; SSE-NEXT: pandn %xmm1, %xmm4
-; SSE-NEXT: andps %xmm12, %xmm2
+; SSE-NEXT: andps %xmm10, %xmm2
; SSE-NEXT: por %xmm2, %xmm4
-; SSE-NEXT: psrlq $48, %xmm0
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE-NEXT: movdqa %xmm1, %xmm2
-; SSE-NEXT: psrld $16, %xmm6
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm9[0,1,2,3,4,5,5,7]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm6[1]
+; SSE-NEXT: pshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT: # xmm1 = mem[1,1,1,1]
+; SSE-NEXT: pshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = mem[2,3,2,3]
+; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE-NEXT: pshufd $196, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT: # xmm1 = mem[0,1,0,3]
+; SSE-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,4,6]
+; SSE-NEXT: punpckhqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT: # xmm1 = xmm1[1],mem[1]
; SSE-NEXT: movss {{.*#+}} xmm1 = xmm2[0],xmm1[1,2,3]
-; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT: # xmm2 = mem[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm12[0,2,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm9 = xmm2[0,1,2,3,4,4,5,7]
-; SSE-NEXT: movdqa %xmm12, %xmm3
-; SSE-NEXT: pandn %xmm9, %xmm3
-; SSE-NEXT: andps %xmm12, %xmm1
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,6]
+; SSE-NEXT: movdqa %xmm10, %xmm3
+; SSE-NEXT: pandn %xmm2, %xmm3
+; SSE-NEXT: andps %xmm10, %xmm1
; SSE-NEXT: por %xmm1, %xmm3
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT: psrlq $48, %xmm2
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSE-NEXT: movdqa %xmm1, %xmm9
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT: psrld $16, %xmm2
-; SSE-NEXT: pshufhw {{.*#+}} xmm1 = xmm8[0,1,2,3,4,5,5,7]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm2[1]
-; SSE-NEXT: movss {{.*#+}} xmm1 = xmm9[0],xmm1[1,2,3]
-; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT: # xmm6 = mem[3,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,5,7]
-; SSE-NEXT: movdqa %xmm12, %xmm9
-; SSE-NEXT: pandn %xmm6, %xmm9
-; SSE-NEXT: andps %xmm12, %xmm1
-; SSE-NEXT: por %xmm1, %xmm9
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT: psrlq $48, %xmm0
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
; SSE-NEXT: psrldq {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
; SSE-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE-NEXT: movdqa %xmm1, %xmm6
+; SSE-NEXT: psrld $16, %xmm6
+; SSE-NEXT: pshufhw {{.*#+}} xmm2 = xmm8[0,1,2,3,4,5,5,7]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm6[1]
+; SSE-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
+; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm7[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm8 = xmm1[0,1,2,3,4,4,5,7]
+; SSE-NEXT: movdqa %xmm10, %xmm1
+; SSE-NEXT: pandn %xmm8, %xmm1
+; SSE-NEXT: andps %xmm10, %xmm2
+; SSE-NEXT: por %xmm2, %xmm1
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT: psrlq $48, %xmm6
+; SSE-NEXT: psrldq {{.*#+}} xmm15 = xmm15[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm15 = xmm15[0],xmm6[0],xmm15[1],xmm6[1],xmm15[2],xmm6[2],xmm15[3],xmm6[3]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT: psrld $16, %xmm7
+; SSE-NEXT: pshufhw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = mem[0,1,2,3,4,5,5,7]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm7[1]
+; SSE-NEXT: movss {{.*#+}} xmm2 = xmm15[0],xmm2[1,2,3]
+; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; SSE-NEXT: # xmm7 = mem[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm7[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,4,5,7]
+; SSE-NEXT: movdqa %xmm10, %xmm15
+; SSE-NEXT: pandn %xmm7, %xmm15
+; SSE-NEXT: andps %xmm10, %xmm2
+; SSE-NEXT: por %xmm2, %xmm15
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE-NEXT: psrlq $48, %xmm0
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT: psrld $16, %xmm2
-; SSE-NEXT: pshufhw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = mem[0,1,2,3,4,5,5,7]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm2[1]
-; SSE-NEXT: movss {{.*#+}} xmm1 = xmm6[0],xmm1[1,2,3]
-; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT: # xmm6 = mem[3,1,2,3,4,5,6,7]
-; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,1,0,3]
-; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,5,7]
-; SSE-NEXT: movdqa %xmm12, %xmm2
-; SSE-NEXT: pandn %xmm6, %xmm2
-; SSE-NEXT: andps %xmm12, %xmm1
-; SSE-NEXT: por %xmm1, %xmm2
+; SSE-NEXT: psrldq {{.*#+}} xmm2 = xmm2[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; SSE-NEXT: movdqa %xmm2, %xmm6
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT: psrld $16, %xmm7
+; SSE-NEXT: pshufhw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = mem[0,1,2,3,4,5,5,7]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm7[1]
+; SSE-NEXT: movss {{.*#+}} xmm2 = xmm6[0],xmm2[1,2,3]
+; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; SSE-NEXT: # xmm7 = mem[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufd {{.*#+}} xmm7 = xmm7[0,1,0,3]
+; SSE-NEXT: pshufhw {{.*#+}} xmm8 = xmm7[0,1,2,3,4,4,5,7]
+; SSE-NEXT: movdqa %xmm10, %xmm7
+; SSE-NEXT: pandn %xmm8, %xmm7
+; SSE-NEXT: andps %xmm10, %xmm2
+; SSE-NEXT: por %xmm2, %xmm7
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
; SSE-NEXT: psrlq $48, %xmm6
-; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT: psrldq {{.*#+}} xmm5 = xmm5[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT: punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
+; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: psrldq {{.*#+}} xmm2 = xmm2[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm6[0],xmm2[1],xmm6[1],xmm2[2],xmm6[2],xmm2[3],xmm6[3]
+; SSE-NEXT: movdqa %xmm2, %xmm8
; SSE-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
; SSE-NEXT: psrld $16, %xmm6
-; SSE-NEXT: pshufhw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT: # xmm1 = mem[0,1,2,3,4,5,5,7]
-; SSE-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],xmm6[1]
-; SSE-NEXT: movss {{.*#+}} xmm1 = xmm5[0],xmm1[1,2,3]
-; SSE-NEXT: andps %xmm12, %xmm1
-; SSE-NEXT: pshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT: # xmm6 = mem[3,1,2,3,4,5,6,7]
+; SSE-NEXT: pshufhw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT: # xmm2 = mem[0,1,2,3,4,5,5,7]
+; SSE-NEXT: punpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm6[1]
+; SSE-NEXT: movss {{.*#+}} xmm2 = xmm8[0],xmm2[1,2,3]
+; SSE-NEXT: andps %xmm10, %xmm2
+; SSE-NEXT: pshuflw {{.*#+}} xmm6 = xmm12[3,1,2,3,4,5,6,7]
; SSE-NEXT: pshufd {{.*#+}} xmm6 = xmm6[0,1,0,3]
; SSE-NEXT: pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,5,7]
-; SSE-NEXT: pandn %xmm6, %xmm12
-; SSE-NEXT: por %xmm1, %xmm12
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, 48(%rsi)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, 16(%rsi)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, 32(%rsi)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, (%rsi)
-; SSE-NEXT: movaps %xmm14, 48(%rdx)
-; SSE-NEXT: movaps %xmm15, 16(%rdx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, 32(%rdx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, (%rdx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, 16(%rcx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, 48(%rcx)
-; SSE-NEXT: movaps (%rsp), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, 32(%rcx)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT: movaps %xmm1, (%rcx)
-; SSE-NEXT: movdqa %xmm13, 48(%r8)
-; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT: movaps %xmm0, 16(%r8)
+; SSE-NEXT: pandn %xmm6, %xmm10
+; SSE-NEXT: por %xmm2, %xmm10
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, 48(%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, 16(%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, 32(%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, (%rsi)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, 48(%rdx)
+; SSE-NEXT: movaps %xmm13, 16(%rdx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, 32(%rdx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, (%rdx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, 16(%rcx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, 48(%rcx)
+; SSE-NEXT: movaps (%rsp), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, 32(%rcx)
+; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT: movaps %xmm2, (%rcx)
+; SSE-NEXT: movdqa %xmm14, 48(%r8)
+; SSE-NEXT: movdqa %xmm11, 16(%r8)
; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT: movaps %xmm0, 32(%r8)
; SSE-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; SSE-NEXT: movaps %xmm0, (%r8)
-; SSE-NEXT: movdqa %xmm4, 48(%r9)
-; SSE-NEXT: movdqa %xmm7, 16(%r9)
-; SSE-NEXT: movdqa %xmm10, 32(%r9)
-; SSE-NEXT: movdqa %xmm11, (%r9)
+; SSE-NEXT: movdqa %xmm3, 48(%r9)
+; SSE-NEXT: movdqa %xmm4, 16(%r9)
+; SSE-NEXT: movdqa %xmm5, 32(%r9)
+; SSE-NEXT: movdqa %xmm9, (%r9)
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT: movdqa %xmm12, 48(%rax)
-; SSE-NEXT: movdqa %xmm2, 16(%rax)
-; SSE-NEXT: movdqa %xmm9, 32(%rax)
-; SSE-NEXT: movdqa %xmm3, (%rax)
+; SSE-NEXT: movdqa %xmm10, 48(%rax)
+; SSE-NEXT: movdqa %xmm7, 16(%rax)
+; SSE-NEXT: movdqa %xmm15, 32(%rax)
+; SSE-NEXT: movdqa %xmm1, (%rax)
; SSE-NEXT: addq $456, %rsp # imm = 0x1C8
; SSE-NEXT: retq
;
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll
index 9d7b11119d5ec..f671c4d1d2175 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll
@@ -895,7 +895,7 @@ define void @vf16(<16 x i16>* %in.vecptr0, <16 x i16>* %in.vecptr1, <16 x i16>*
; AVX1-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm7[0],xmm4[0],xmm7[1],xmm4[1],xmm7[2],xmm4[2],xmm7[3],xmm4[3]
; AVX1-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[0,1,2,2,4,5,6,7]
; AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; AVX1-NEXT: vpshuflw {{.*#+}} xmm3 = xmm3[2,2,2,2,4,5,6,7]
+; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
; AVX1-NEXT: vpshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,4,6,7]
; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm2
; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm13[0,1,2,3],xmm0[4],xmm13[5,6,7]
@@ -1767,7 +1767,7 @@ define void @vf32(<32 x i16>* %in.vecptr0, <32 x i16>* %in.vecptr1, <32 x i16>*
; AVX1-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
; AVX1-NEXT: vpshuflw {{.*#+}} xmm3 = xmm4[0,1,2,2,4,5,6,7]
; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[0,0,2,1]
-; AVX1-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[2,2,2,2,4,5,6,7]
+; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; AVX1-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,4,6,7]
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm3, %ymm0
; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm5[0,1,2,3],xmm14[4],xmm5[5,6,7]
@@ -1870,7 +1870,7 @@ define void @vf32(<32 x i16>* %in.vecptr0, <32 x i16>* %in.vecptr1, <32 x i16>*
; AVX1-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
; AVX1-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,1,2,2,4,5,6,7]
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; AVX1-NEXT: vpshuflw {{.*#+}} xmm3 = xmm3[2,2,2,2,4,5,6,7]
+; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
; AVX1-NEXT: vpshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,4,6,7]
; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm3
; AVX1-NEXT: vpblendw {{.*#+}} xmm14 = xmm14[0,1,2,3],xmm2[4],xmm14[5,6,7]
More information about the llvm-commits
mailing list