[X86] Remove sse41 specific code from lowering v16i8 multiply

[craig.topper]

Summary

With the SRAs removed from the SSE2 code as proposed in D44267, then there doesn't appear to be any advantage to the sse41 code. The punpcklbw instruction and pmovsx seem to have the same latency and throughput on most CPUs. And the SSE41 code requires moving the upper 64-bits into the lower 64-bit before the sign extend can be done. The unpckhbw in sse2 code can do better than that.

[RKSimon]

test/CodeGen/X86/vector-mul.ll

968

Why wasn't this constant folded?

[craig.topper]

test/CodeGen/X86/vector-mul.ll

968

At one point that constant pool entry was used by two vector shuffles and I guess we refused to fold it due to multiple uses? Late one shuffle became UNPCKH and the other became zero_extend_vector_in_reg. The DAG combine for zero_extend_vector_in_reg was perfectly happy to overlook the multiple uses and constant fold it. This is the LCPI on the pmullw. This dropped the usage count on the original constant pool but it was too late to trigger the fold.

Should we stop the zero_extend_vector_in_reg from constant folding multiple uses?

[RKSimon]

test/CodeGen/X86/vector-mul.ll

968

Should we stop the zero_extend_vector_in_reg from constant folding multiple uses?

Probably but I'm wondering if it'd be better to wait until we've made progress simplifying the mixture of *_EXTEND/*_EXTEND_VECTOR_INREG/V*EXT that we currently use for SSE/AVX vector extensions?

[RKSimon]

LGTM - please can you add a fixme/bug about the the constant folding someplace

[craig.topper]

commited in 327869 but forgot to add the Differential Revision line

lib/Target/X86/X86ISelLowering.cpp

if (Subtarget.hasInt256()) {
                       DAG.getNode(ISD::SIGN_EXTEND, dl, ExVT, A),
                       DAG.getNode(ISD::SIGN_EXTEND, dl, ExVT, B)));
     }
 
     assert(VT == MVT::v16i8 &&
            "Pre-AVX2 support only supports v16i8 multiplication");
     MVT ExVT = MVT::v8i16;
 
-    // Extract the lo parts and sign extend to i16
-    SDValue ALo, BLo;
-    if (Subtarget.hasSSE41()) {
-      ALo = DAG.getSignExtendVectorInReg(A, dl, ExVT);
-      BLo = DAG.getSignExtendVectorInReg(B, dl, ExVT);
-    } else {
-      // We're going mask off the low byte of each result element of the mullw,
-      // so it doesn't matter what's in the high byte of each 16-bit element.
-      const int ShufMask[] = {0, -1, 1, -1, 2, -1, 3, -1,
+    // Extract the lo parts into the low byte of 16-bit elements.
+    const int LoShufMask[] = {0, -1, 1, -1, 2, -1, 3, -1,
                               4, -1, 5, -1, 6, -1, 7, -1};
-      ALo = DAG.getVectorShuffle(VT, dl, A, A, ShufMask);
-      BLo = DAG.getVectorShuffle(VT, dl, B, B, ShufMask);
-      ALo = DAG.getBitcast(ExVT, ALo);
-      BLo = DAG.getBitcast(ExVT, BLo);
-    }
+    SDValue ALo = DAG.getVectorShuffle(VT, dl, A, A, LoShufMask);
+    SDValue BLo = DAG.getVectorShuffle(VT, dl, B, B, LoShufMask);
 
-    // Extract the hi parts and sign extend to i16
-    SDValue AHi, BHi;
-    if (Subtarget.hasSSE41()) {
-      const int ShufMask[] = {8,  9,  10, 11, 12, 13, 14, 15,
-                              -1, -1, -1, -1, -1, -1, -1, -1};
-      AHi = DAG.getVectorShuffle(VT, dl, A, A, ShufMask);
-      BHi = DAG.getVectorShuffle(VT, dl, B, B, ShufMask);
-      AHi = DAG.getSignExtendVectorInReg(AHi, dl, ExVT);
-      BHi = DAG.getSignExtendVectorInReg(BHi, dl, ExVT);
-    } else {
-      // We're going mask off the low byte of each result element of the mullw,
-      // so it doesn't matter what's in the high byte of each 16-bit element.
-      const int ShufMask[] = {8,  -1, 9,  -1, 10, -1, 11, -1,
+    // Extract the hi parts into the low byte of 16-bit elements.
+    const int HiShufMask[] = {8,  -1, 9,  -1, 10, -1, 11, -1,
                               12, -1, 13, -1, 14, -1, 15, -1};
-      AHi = DAG.getVectorShuffle(VT, dl, A, A, ShufMask);
-      BHi = DAG.getVectorShuffle(VT, dl, B, B, ShufMask);
-      AHi = DAG.getBitcast(ExVT, AHi);
-      BHi = DAG.getBitcast(ExVT, BHi);
-    }
+    SDValue AHi = DAG.getVectorShuffle(VT, dl, A, A, HiShufMask);
+    SDValue BHi = DAG.getVectorShuffle(VT, dl, B, B, HiShufMask);
 
     // Multiply, mask the lower 8bits of the lo/hi results and pack
     SDValue RLo = DAG.getNode(ISD::MUL, dl, ExVT, ALo, BLo);
     SDValue RHi = DAG.getNode(ISD::MUL, dl, ExVT, AHi, BHi);
     RLo = DAG.getNode(ISD::AND, dl, ExVT, RLo, DAG.getConstant(255, dl, ExVT));
     RHi = DAG.getNode(ISD::AND, dl, ExVT, RHi, DAG.getConstant(255, dl, ExVT));
     return DAG.getNode(X86ISD::PACKUS, dl, VT, RLo, RHi);
   }

test/CodeGen/X86/combine-mul.ll

 }
 
 ; This would infinite loop because DAGCombiner wants to turn this into a shift,
 ; but x86 lowering wants to avoid non-uniform vector shift amounts.
 
 define <16 x i8> @PR35579(<16 x i8> %x) {
 ; SSE-LABEL: PR35579:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    pmovsxbw %xmm0, %xmm1
-; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,2,1,4,1,2,1,8,1,2,1,4,1,2,1]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE-NEXT:    pmullw %xmm2, %xmm0
 ; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm2, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE-NEXT:    pmovsxbw %xmm0, %xmm0
-; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; SSE-NEXT:    pand %xmm2, %xmm1
 ; SSE-NEXT:    packuswb %xmm0, %xmm1
 ; SSE-NEXT:    movdqa %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: PR35579:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpmovsxbw %xmm0, %ymm0
 ; AVX-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0

test/CodeGen/X86/pmul.ll

 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE2-NEXT:    pmullw %xmm2, %xmm0
 ; SSE2-NEXT:    pand %xmm3, %xmm0
 ; SSE2-NEXT:    packuswb %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v16i8c:
 ; SSE41:       # %bb.0: # %entry
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [117,117,117,117,117,117,117,117]
-; SSE41-NEXT:    pmullw %xmm2, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm3, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm0
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117]
 ; SSE41-NEXT:    pmullw %xmm2, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
 ; SSE41-NEXT:    pand %xmm3, %xmm0
+; SSE41-NEXT:    pmullw %xmm2, %xmm1
+; SSE41-NEXT:    pand %xmm3, %xmm1
 ; SSE41-NEXT:    packuswb %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v16i8c:
 ; AVX2:       # %bb.0: # %entry
 ; AVX2-NEXT:    vpmovsxbw %xmm0, %ymm0
 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0

 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE2-NEXT:    pmullw %xmm1, %xmm0
 ; SSE2-NEXT:    pand %xmm2, %xmm0
 ; SSE2-NEXT:    packuswb %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v16i8:
 ; SSE41:       # %bb.0: # %entry
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm3
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm2
-; SSE41-NEXT:    pmullw %xmm3, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm3, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm0
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE41-NEXT:    pmullw %xmm1, %xmm0
-; SSE41-NEXT:    pand %xmm3, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    pand %xmm1, %xmm0
+; SSE41-NEXT:    pmullw %xmm3, %xmm2
+; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    packuswb %xmm0, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v16i8:
 ; AVX2:       # %bb.0: # %entry
 ; AVX2-NEXT:    vpmovsxbw %xmm1, %ymm1
 ; AVX2-NEXT:    vpmovsxbw %xmm0, %ymm0

 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSE2-NEXT:    pmullw %xmm3, %xmm1
 ; SSE2-NEXT:    pand %xmm4, %xmm1
 ; SSE2-NEXT:    packuswb %xmm2, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v32i8c:
 ; SSE41:       # %bb.0: # %entry
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [117,117,117,117,117,117,117,117]
-; SSE41-NEXT:    pmullw %xmm4, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm5, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm0
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117]
 ; SSE41-NEXT:    pmullw %xmm4, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
 ; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    pmullw %xmm4, %xmm2
+; SSE41-NEXT:    pand %xmm5, %xmm2
 ; SSE41-NEXT:    packuswb %xmm0, %xmm2
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm3
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm5, %xmm1
 ; SSE41-NEXT:    pmullw %xmm4, %xmm3
 ; SSE41-NEXT:    pand %xmm5, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm0
-; SSE41-NEXT:    pmullw %xmm4, %xmm0
-; SSE41-NEXT:    pand %xmm5, %xmm0
-; SSE41-NEXT:    packuswb %xmm0, %xmm3
+; SSE41-NEXT:    packuswb %xmm1, %xmm3
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    movdqa %xmm3, %xmm1
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v32i8c:
 ; AVX2:       # %bb.0: # %entry
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmovsxbw %xmm1, %ymm1

 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSE2-NEXT:    pmullw %xmm3, %xmm1
 ; SSE2-NEXT:    pand %xmm4, %xmm1
 ; SSE2-NEXT:    packuswb %xmm5, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v32i8:
 ; SSE41:       # %bb.0: # %entry
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm5
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm4
-; SSE41-NEXT:    pmullw %xmm5, %xmm4
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm5, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm0
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm5 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE41-NEXT:    pmullw %xmm2, %xmm0
-; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    pand %xmm6, %xmm0
+; SSE41-NEXT:    pmullw %xmm5, %xmm4
+; SSE41-NEXT:    pand %xmm6, %xmm4
 ; SSE41-NEXT:    packuswb %xmm0, %xmm4
-; SSE41-NEXT:    pmovsxbw %xmm3, %xmm0
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm2
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw %xmm3, %xmm1
+; SSE41-NEXT:    pand %xmm6, %xmm1
 ; SSE41-NEXT:    pmullw %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm5, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm1
-; SSE41-NEXT:    pmullw %xmm0, %xmm1
-; SSE41-NEXT:    pand %xmm5, %xmm1
+; SSE41-NEXT:    pand %xmm6, %xmm2
 ; SSE41-NEXT:    packuswb %xmm1, %xmm2
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
 ; SSE41-NEXT:    movdqa %xmm2, %xmm1
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v32i8:
 ; AVX2:       # %bb.0: # %entry
 ; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2

 ; SSE2-NEXT:    pand %xmm5, %xmm3
 ; SSE2-NEXT:    packuswb %xmm6, %xmm3
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v64i8c:
 ; SSE41:       # %bb.0: # %entry
 ; SSE41-NEXT:    movdqa %xmm1, %xmm4
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [117,117,117,117,117,117,117,117]
-; SSE41-NEXT:    pmullw %xmm6, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm7, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117]
 ; SSE41-NEXT:    pmullw %xmm6, %xmm1
+; SSE41-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,255,255,255,255,255]
 ; SSE41-NEXT:    pand %xmm7, %xmm1
+; SSE41-NEXT:    pmullw %xmm6, %xmm0
+; SSE41-NEXT:    pand %xmm7, %xmm0
 ; SSE41-NEXT:    packuswb %xmm1, %xmm0
-; SSE41-NEXT:    pmovsxbw %xmm4, %xmm1
-; SSE41-NEXT:    pmullw %xmm6, %xmm1
-; SSE41-NEXT:    pand %xmm7, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm4, %xmm4
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE41-NEXT:    pmullw %xmm6, %xmm4
 ; SSE41-NEXT:    pand %xmm7, %xmm4
+; SSE41-NEXT:    pmullw %xmm6, %xmm1
+; SSE41-NEXT:    pand %xmm7, %xmm1
 ; SSE41-NEXT:    packuswb %xmm4, %xmm1
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm4
-; SSE41-NEXT:    pmullw %xmm6, %xmm4
-; SSE41-NEXT:    pand %xmm7, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm2
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE41-NEXT:    pmullw %xmm6, %xmm2
 ; SSE41-NEXT:    pand %xmm7, %xmm2
+; SSE41-NEXT:    pmullw %xmm6, %xmm4
+; SSE41-NEXT:    pand %xmm7, %xmm4
 ; SSE41-NEXT:    packuswb %xmm2, %xmm4
-; SSE41-NEXT:    pmovsxbw %xmm3, %xmm5
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm5 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw %xmm6, %xmm3
+; SSE41-NEXT:    pand %xmm7, %xmm3
 ; SSE41-NEXT:    pmullw %xmm6, %xmm5
 ; SSE41-NEXT:    pand %xmm7, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm2
-; SSE41-NEXT:    pmullw %xmm6, %xmm2
-; SSE41-NEXT:    pand %xmm7, %xmm2
-; SSE41-NEXT:    packuswb %xmm2, %xmm5
+; SSE41-NEXT:    packuswb %xmm3, %xmm5
 ; SSE41-NEXT:    movdqa %xmm4, %xmm2
 ; SSE41-NEXT:    movdqa %xmm5, %xmm3
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v64i8c:
 ; AVX2:       # %bb.0: # %entry
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; AVX2-NEXT:    vpmovsxbw %xmm2, %ymm2

 ; SSE2-NEXT:    pand %xmm8, %xmm3
 ; SSE2-NEXT:    packuswb %xmm5, %xmm3
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v64i8:
 ; SSE41:       # %bb.0: # %entry
 ; SSE41-NEXT:    movdqa %xmm1, %xmm8
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pmovsxbw %xmm4, %xmm9
-; SSE41-NEXT:    pmovsxbw %xmm0, %xmm0
-; SSE41-NEXT:    pmullw %xmm9, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm9, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm4, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm10 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE41-NEXT:    pmullw %xmm4, %xmm1
+; SSE41-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,255,255,255,255,255]
 ; SSE41-NEXT:    pand %xmm9, %xmm1
+; SSE41-NEXT:    pmullw %xmm10, %xmm0
+; SSE41-NEXT:    pand %xmm9, %xmm0
 ; SSE41-NEXT:    packuswb %xmm1, %xmm0
-; SSE41-NEXT:    pmovsxbw %xmm5, %xmm4
-; SSE41-NEXT:    pmovsxbw %xmm8, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero,xmm5[4],zero,xmm5[5],zero,xmm5[6],zero,xmm5[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm8[0],zero,xmm8[1],zero,xmm8[2],zero,xmm8[3],zero,xmm8[4],zero,xmm8[5],zero,xmm8[6],zero,xmm8[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw %xmm5, %xmm8
+; SSE41-NEXT:    pand %xmm9, %xmm8
 ; SSE41-NEXT:    pmullw %xmm4, %xmm1
 ; SSE41-NEXT:    pand %xmm9, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm5[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm4, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm8[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm5, %xmm5
-; SSE41-NEXT:    pmullw %xmm4, %xmm5
-; SSE41-NEXT:    pand %xmm9, %xmm5
-; SSE41-NEXT:    packuswb %xmm5, %xmm1
-; SSE41-NEXT:    pmovsxbw %xmm6, %xmm5
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm4
+; SSE41-NEXT:    packuswb %xmm8, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm5 = xmm6[0],zero,xmm6[1],zero,xmm6[2],zero,xmm6[3],zero,xmm6[4],zero,xmm6[5],zero,xmm6[6],zero,xmm6[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw %xmm6, %xmm2
+; SSE41-NEXT:    pand %xmm9, %xmm2
 ; SSE41-NEXT:    pmullw %xmm5, %xmm4
 ; SSE41-NEXT:    pand %xmm9, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm6[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm5, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm2
-; SSE41-NEXT:    pmullw %xmm5, %xmm2
-; SSE41-NEXT:    pand %xmm9, %xmm2
 ; SSE41-NEXT:    packuswb %xmm2, %xmm4
-; SSE41-NEXT:    pmovsxbw %xmm7, %xmm2
-; SSE41-NEXT:    pmovsxbw %xmm3, %xmm5
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm7[0],zero,xmm7[1],zero,xmm7[2],zero,xmm7[3],zero,xmm7[4],zero,xmm7[5],zero,xmm7[6],zero,xmm7[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm5 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw %xmm7, %xmm3
+; SSE41-NEXT:    pand %xmm9, %xmm3
 ; SSE41-NEXT:    pmullw %xmm2, %xmm5
 ; SSE41-NEXT:    pand %xmm9, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm7[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm3, %xmm3
-; SSE41-NEXT:    pmullw %xmm2, %xmm3
-; SSE41-NEXT:    pand %xmm9, %xmm3
 ; SSE41-NEXT:    packuswb %xmm3, %xmm5
 ; SSE41-NEXT:    movdqa %xmm4, %xmm2
 ; SSE41-NEXT:    movdqa %xmm5, %xmm3
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v64i8:
 ; AVX2:       # %bb.0: # %entry
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm4

test/CodeGen/X86/vector-idiv-sdiv-128.ll

 ; SSE41-NEXT:    psrlw $2, %xmm2
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
 ; SSE41-NEXT:    pxor %xmm3, %xmm2
 ; SSE41-NEXT:    psubb %xmm3, %xmm2
 ; SSE41-NEXT:    psrlw $7, %xmm1
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    paddb %xmm2, %xmm1
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7]
-; SSE41-NEXT:    pmullw %xmm3, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm4, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm1, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; SSE41-NEXT:    pmullw %xmm3, %xmm1
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; SSE41-NEXT:    pand %xmm4, %xmm1
+; SSE41-NEXT:    pmullw %xmm3, %xmm2
+; SSE41-NEXT:    pand %xmm4, %xmm2
 ; SSE41-NEXT:    packuswb %xmm1, %xmm2
 ; SSE41-NEXT:    psubb %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_rem7_16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpmovsxbw %xmm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [65427,65427,65427,65427,65427,65427,65427,65427]
 ; AVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovsxbw %xmm3, %xmm3
 ; AVX1-NEXT:    vpmullw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm2
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
 ; AVX1-NEXT:    vpxor %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw $7, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vpmovsxbw %xmm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX1-NEXT:    vpmovsxbw %xmm1, %xmm1
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX1-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2NOBW-LABEL: test_rem7_16i8:
 ; AVX2NOBW:       # %bb.0:
 ; AVX2NOBW-NEXT:    vpmovsxbw %xmm0, %ymm1
 ; AVX2NOBW-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
 ; AVX2NOBW-NEXT:    vpsrlw $8, %ymm1, %ymm1

test/CodeGen/X86/vector-idiv-sdiv-256.ll

 ; AVX1-NEXT:    vpand %xmm8, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsrlw $2, %xmm3, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm9 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
 ; AVX1-NEXT:    vpand %xmm9, %xmm3, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
 ; AVX1-NEXT:    vpxor %xmm7, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubb %xmm7, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpmovsxbw %xmm3, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX1-NEXT:    vpmullw %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [255,255,255,255,255,255,255,255]
 ; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; AVX1-NEXT:    vpmovsxbw %xmm3, %xmm3
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
 ; AVX1-NEXT:    vpmullw %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpackuswb %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpackuswb %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovsxbw %xmm0, %xmm3
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovsxbw %xmm4, %xmm4
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm4, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $7, %xmm1, %xmm3
 ; AVX1-NEXT:    vpand %xmm8, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm9, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm7, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubb %xmm7, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpmovsxbw %xmm1, %xmm3
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX1-NEXT:    vpmullw %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX1-NEXT:    vpmovsxbw %xmm1, %xmm1
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX1-NEXT:    vpmullw %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm6, %xmm1, %xmm1
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm3, %xmm1
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2NOBW-LABEL: test_rem7_32i8:
 ; AVX2NOBW:       # %bb.0:
 ; AVX2NOBW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2NOBW-NEXT:    vpmovsxbw %xmm1, %ymm1

test/CodeGen/X86/vector-idiv-udiv-128.ll

 ; SSE41-NEXT:    packuswb %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    psubb %xmm1, %xmm2
 ; SSE41-NEXT:    psrlw $1, %xmm2
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE41-NEXT:    paddb %xmm1, %xmm2
 ; SSE41-NEXT:    psrlw $2, %xmm2
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7]
-; SSE41-NEXT:    pmullw %xmm3, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pand %xmm4, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; SSE41-NEXT:    pmovsxbw %xmm2, %xmm2
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; SSE41-NEXT:    pmullw %xmm3, %xmm2
+; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; SSE41-NEXT:    pand %xmm4, %xmm2
+; SSE41-NEXT:    pmullw %xmm3, %xmm1
+; SSE41-NEXT:    pand %xmm4, %xmm1
 ; SSE41-NEXT:    packuswb %xmm2, %xmm1
 ; SSE41-NEXT:    psubb %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_rem7_16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [37,37,37,37,37,37,37,37]
 ; AVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
 ; AVX1-NEXT:    vpmullw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpsrlw $1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vpmovsxbw %xmm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX1-NEXT:    vpmovsxbw %xmm1, %xmm1
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX1-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2NOBW-LABEL: test_rem7_16i8:
 ; AVX2NOBW:       # %bb.0:
 ; AVX2NOBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX2NOBW-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
 ; AVX2NOBW-NEXT:    vpsrlw $8, %ymm1, %ymm1

test/CodeGen/X86/vector-idiv-udiv-256.ll

 ; AVX2-NEXT:    retq
   %res = urem <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
   ret <16 x i16> %res
 }
 
 define <32 x i8> @test_rem7_32i8(<32 x i8> %a) nounwind {
 ; AVX1-LABEL: test_rem7_32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [37,37,37,37,37,37,37,37]
-; AVX1-NEXT:    vpmullw %xmm1, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [37,37,37,37,37,37,37,37]
+; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm1[2,3,0,1]
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
-; AVX1-NEXT:    vpmullw %xmm1, %xmm4, %xmm4
+; AVX1-NEXT:    vpmullw %xmm3, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
-; AVX1-NEXT:    vpackuswb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsubb %xmm3, %xmm2, %xmm4
+; AVX1-NEXT:    vpackuswb %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpsubb %xmm2, %xmm1, %xmm4
 ; AVX1-NEXT:    vpsrlw $1, %xmm4, %xmm4
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; AVX1-NEXT:    vpand %xmm8, %xmm4, %xmm4
-; AVX1-NEXT:    vpaddb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $2, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpsrlw $2, %xmm2, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpmovsxbw %xmm3, %xmm6
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [7,7,7,7,7,7,7,7]
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX1-NEXT:    vpmullw %xmm7, %xmm6, %xmm6
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
 ; AVX1-NEXT:    vpand %xmm5, %xmm6, %xmm6
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; AVX1-NEXT:    vpmovsxbw %xmm3, %xmm3
-; AVX1-NEXT:    vpmullw %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpackuswb %xmm3, %xmm6, %xmm3
-; AVX1-NEXT:    vpsubb %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX1-NEXT:    vpmullw %xmm1, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX1-NEXT:    vpmullw %xmm7, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpackuswb %xmm6, %xmm2, %xmm2
+; AVX1-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm6 = xmm6[0],zero,xmm6[1],zero,xmm6[2],zero,xmm6[3],zero,xmm6[4],zero,xmm6[5],zero,xmm6[6],zero,xmm6[7],zero
-; AVX1-NEXT:    vpmullw %xmm1, %xmm6, %xmm1
-; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpsubb %xmm1, %xmm0, %xmm3
+; AVX1-NEXT:    vpmullw %xmm3, %xmm6, %xmm3
+; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsubb %xmm2, %xmm0, %xmm3
 ; AVX1-NEXT:    vpsrlw $1, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm8, %xmm3, %xmm3
-; AVX1-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpmovsxbw %xmm1, %xmm3
+; AVX1-NEXT:    vpaddb %xmm2, %xmm3, %xmm2
+; AVX1-NEXT:    vpsrlw $2, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX1-NEXT:    vpmullw %xmm7, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX1-NEXT:    vpmovsxbw %xmm1, %xmm1
-; AVX1-NEXT:    vpmullw %xmm7, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX1-NEXT:    vpmullw %xmm7, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsubb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2NOBW-LABEL: test_rem7_32i8:
 ; AVX2NOBW:       # %bb.0:
 ; AVX2NOBW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2NOBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX2NOBW-NEXT:    vmovdqa {{.*#+}} ymm2 = [37,37,37,37,37,37,37,37,37,37,37,37,37,37,37,37]
 ; AVX2NOBW-NEXT:    vpmullw %ymm2, %ymm1, %ymm1

test/CodeGen/X86/vector-mul.ll

 ; X64-AVX-NEXT:    retq
   %1 = mul <8 x i16> %a0, <i16 17, i16 17, i16 17, i16 17, i16 17, i16 17, i16 17, i16 17>
   ret <8 x i16> %1
 }
 
 define <16 x i8> @mul_v16i8_17(<16 x i8> %a0) nounwind {
 ; X86-LABEL: mul_v16i8_17:
 ; X86:       # %bb.0:
-; X86-NEXT:    pmovsxbw %xmm0, %xmm1
-; X86-NEXT:    movdqa {{.*#+}} xmm2 = [17,17,17,17,17,17,17,17]
-; X86-NEXT:    pmullw %xmm2, %xmm1
-; X86-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; X86-NEXT:    pand %xmm3, %xmm1
-; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X86-NEXT:    pmovsxbw %xmm0, %xmm0
+; X86-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X86-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X86-NEXT:    movdqa {{.*#+}} xmm2 = [17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17]
 ; X86-NEXT:    pmullw %xmm2, %xmm0
+; X86-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
 ; X86-NEXT:    pand %xmm3, %xmm0
+; X86-NEXT:    pmullw %xmm2, %xmm1
+; X86-NEXT:    pand %xmm3, %xmm1
 ; X86-NEXT:    packuswb %xmm0, %xmm1
 ; X86-NEXT:    movdqa %xmm1, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: mul_v16i8_17:
 ; X64:       # %bb.0:
-; X64-NEXT:    pmovsxbw %xmm0, %xmm1
-; X64-NEXT:    movdqa {{.*#+}} xmm2 = [17,17,17,17,17,17,17,17]
-; X64-NEXT:    pmullw %xmm2, %xmm1
-; X64-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; X64-NEXT:    pand %xmm3, %xmm1
-; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    pmovsxbw %xmm0, %xmm0
+; X64-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-NEXT:    movdqa {{.*#+}} xmm2 = [17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17]
 ; X64-NEXT:    pmullw %xmm2, %xmm0
+; X64-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
 ; X64-NEXT:    pand %xmm3, %xmm0
+; X64-NEXT:    pmullw %xmm2, %xmm1
+; X64-NEXT:    pand %xmm3, %xmm1
 ; X64-NEXT:    packuswb %xmm0, %xmm1
 ; X64-NEXT:    movdqa %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; X64-XOP-LABEL: mul_v16i8_17:
 ; X64-XOP:       # %bb.0:
-; X64-XOP-NEXT:    vpmovsxbw %xmm0, %xmm1
-; X64-XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [17,17,17,17,17,17,17,17]
+; X64-XOP-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17]
 ; X64-XOP-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
-; X64-XOP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-XOP-NEXT:    vpmovsxbw %xmm0, %xmm0
+; X64-XOP-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; X64-XOP-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; X64-XOP-NEXT:    vpperm {{.*#+}} xmm0 = xmm1[0,2,4,6,8,10,12,14],xmm0[0,2,4,6,8,10,12,14]
+; X64-XOP-NEXT:    vpperm {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],xmm1[0,2,4,6,8,10,12,14]
 ; X64-XOP-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: mul_v16i8_17:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    vpmovsxbw %xmm0, %ymm0
 ; X64-AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>

 ; X64-AVX-NEXT:    retq
   %1 = mul <8 x i16> %a0, <i16 2, i16 3, i16 9, i16 17, i16 33, i16 65, i16 129, i16 257>
   ret <8 x i16> %1
 }
 
 define <16 x i8> @mul_v16i8_2_3_9_17_33_65_129_2_3_9_17_33_65_129_2_3(<16 x i8> %a0) nounwind {
 ; X86-LABEL: mul_v16i8_2_3_9_17_33_65_129_2_3_9_17_33_65_129_2_3:
 ; X86:       # %bb.0:
-; X86-NEXT:    pmovsxbw %xmm0, %xmm1
-; X86-NEXT:    pmullw {{\.LCPI.*}}, %xmm1
+; X86-NEXT:    movdqa {{.*#+}} xmm2 = [2,3,9,17,33,65,129,2,3,9,17,33,65,129,2,3]
+; X86-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X86-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X86-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X86-NEXT:    pmullw %xmm2, %xmm0
 ; X86-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; X86-NEXT:    pand %xmm2, %xmm1
-; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X86-NEXT:    pmovsxbw %xmm0, %xmm0
-; X86-NEXT:    pmullw {{\.LCPI.*}}, %xmm0
 ; X86-NEXT:    pand %xmm2, %xmm0
+; X86-NEXT:    pmullw {{\.LCPI.*}}, %xmm1
+; X86-NEXT:    pand %xmm2, %xmm1
 ; X86-NEXT:    packuswb %xmm0, %xmm1
 ; X86-NEXT:    movdqa %xmm1, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: mul_v16i8_2_3_9_17_33_65_129_2_3_9_17_33_65_129_2_3:
 ; X64:       # %bb.0:
-; X64-NEXT:    pmovsxbw %xmm0, %xmm1
-; X64-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; X64-NEXT:    movdqa {{.*#+}} xmm2 = [2,3,9,17,33,65,129,2,3,9,17,33,65,129,2,3]
+; X64-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-NEXT:    pmullw %xmm2, %xmm0
 ; X64-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; X64-NEXT:    pand %xmm2, %xmm1
-; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    pmovsxbw %xmm0, %xmm0
-; X64-NEXT:    pmullw {{.*}}(%rip), %xmm0
 ; X64-NEXT:    pand %xmm2, %xmm0
+; X64-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; X64-NEXT:    pand %xmm2, %xmm1
 ; X64-NEXT:    packuswb %xmm0, %xmm1
 ; X64-NEXT:    movdqa %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; X64-XOP-LABEL: mul_v16i8_2_3_9_17_33_65_129_2_3_9_17_33_65_129_2_3:
 ; X64-XOP:       # %bb.0:
-; X64-XOP-NEXT:    vpmovsxbw %xmm0, %xmm1
-; X64-XOP-NEXT:    vpmullw {{.*}}(%rip), %xmm1, %xmm1
-; X64-XOP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-XOP-NEXT:    vpmovsxbw %xmm0, %xmm0
+; X64-XOP-NEXT:    vmovdqa {{.*#+}} xmm1 = [2,3,9,17,33,65,129,2,3,9,17,33,65,129,2,3]
+; X64-XOP-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-XOP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-XOP-NEXT:    vpmullw %xmm1, %xmm2, %xmm1
+; X64-XOP-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; X64-XOP-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
-; X64-XOP-NEXT:    vpperm {{.*#+}} xmm0 = xmm1[0,2,4,6,8,10,12,14],xmm0[0,2,4,6,8,10,12,14]
+; X64-XOP-NEXT:    vpperm {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],xmm1[0,2,4,6,8,10,12,14]
 ; X64-XOP-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: mul_v16i8_2_3_9_17_33_65_129_2_3_9_17_33_65_129_2_3:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    vpmovsxbw %xmm0, %ymm0
 ; X64-AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>

 ; X64-AVX-NEXT:    retq
   %1 = mul <8 x i16> %a0, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
   ret <8 x i16> %1
 }
 
 define <16 x i8> @mul_v16i8_31(<16 x i8> %a0) nounwind {
 ; X86-LABEL: mul_v16i8_31:
 ; X86:       # %bb.0:
-; X86-NEXT:    pmovsxbw %xmm0, %xmm1
-; X86-NEXT:    movdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31]
-; X86-NEXT:    pmullw %xmm2, %xmm1
-; X86-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; X86-NEXT:    pand %xmm3, %xmm1
-; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X86-NEXT:    pmovsxbw %xmm0, %xmm0
+; X86-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X86-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X86-NEXT:    movdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
 ; X86-NEXT:    pmullw %xmm2, %xmm0
+; X86-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
 ; X86-NEXT:    pand %xmm3, %xmm0
+; X86-NEXT:    pmullw %xmm2, %xmm1
+; X86-NEXT:    pand %xmm3, %xmm1
 ; X86-NEXT:    packuswb %xmm0, %xmm1
 ; X86-NEXT:    movdqa %xmm1, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: mul_v16i8_31:
 ; X64:       # %bb.0:
-; X64-NEXT:    pmovsxbw %xmm0, %xmm1
-; X64-NEXT:    movdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31]
-; X64-NEXT:    pmullw %xmm2, %xmm1
-; X64-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; X64-NEXT:    pand %xmm3, %xmm1
-; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    pmovsxbw %xmm0, %xmm0
+; X64-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-NEXT:    movdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
 ; X64-NEXT:    pmullw %xmm2, %xmm0
+; X64-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
 ; X64-NEXT:    pand %xmm3, %xmm0
+; X64-NEXT:    pmullw %xmm2, %xmm1
+; X64-NEXT:    pand %xmm3, %xmm1
 ; X64-NEXT:    packuswb %xmm0, %xmm1
 ; X64-NEXT:    movdqa %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; X64-XOP-LABEL: mul_v16i8_31:
 ; X64-XOP:       # %bb.0:
-; X64-XOP-NEXT:    vpmovsxbw %xmm0, %xmm1
-; X64-XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31]
+; X64-XOP-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
 ; X64-XOP-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
-; X64-XOP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-XOP-NEXT:    vpmovsxbw %xmm0, %xmm0
+; X64-XOP-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; X64-XOP-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; X64-XOP-NEXT:    vpperm {{.*#+}} xmm0 = xmm1[0,2,4,6,8,10,12,14],xmm0[0,2,4,6,8,10,12,14]
+; X64-XOP-NEXT:    vpperm {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],xmm1[0,2,4,6,8,10,12,14]
 ; X64-XOP-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: mul_v16i8_31:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    vpmovsxbw %xmm0, %ymm0
 ; X64-AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>

 ; X64-AVX-NEXT:    retq
   %1 = mul <8 x i16> %a0, <i16 0, i16 1, i16 7, i16 15, i16 31, i16 63, i16 127, i16 255>
   ret <8 x i16> %1
 }
 
 define <16 x i8> @mul_v16i8_0_1_3_7_15_31_63_127_0_1_3_7_15_31_63_127(<16 x i8> %a0) nounwind {
 ; X86-LABEL: mul_v16i8_0_1_3_7_15_31_63_127_0_1_3_7_15_31_63_127:
 ; X86:       # %bb.0:
-; X86-NEXT:    pmovsxbw %xmm0, %xmm1
-; X86-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,3,7,15,31,63,127]
-; X86-NEXT:    pmullw %xmm2, %xmm1
-; X86-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; X86-NEXT:    pand %xmm3, %xmm1
-; X86-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X86-NEXT:    pmovsxbw %xmm0, %xmm0
+; X86-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,3,7,15,31,63,127,0,1,3,7,15,31,63,127]
+; X86-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X86-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X86-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; X86-NEXT:    pmullw %xmm2, %xmm0
-; X86-NEXT:    pand %xmm3, %xmm0
+; X86-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; X86-NEXT:    pand %xmm2, %xmm0
+; X86-NEXT:    pmullw {{\.LCPI.*}}, %xmm1
+; X86-NEXT:    pand %xmm2, %xmm1
 ; X86-NEXT:    packuswb %xmm0, %xmm1
 ; X86-NEXT:    movdqa %xmm1, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: mul_v16i8_0_1_3_7_15_31_63_127_0_1_3_7_15_31_63_127:
 ; X64:       # %bb.0:
-; X64-NEXT:    pmovsxbw %xmm0, %xmm1
-; X64-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,3,7,15,31,63,127]
-; X64-NEXT:    pmullw %xmm2, %xmm1
-; X64-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
-; X64-NEXT:    pand %xmm3, %xmm1
-; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-NEXT:    pmovsxbw %xmm0, %xmm0
+; X64-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,3,7,15,31,63,127,0,1,3,7,15,31,63,127]
+; X64-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; X64-NEXT:    pmullw %xmm2, %xmm0
-; X64-NEXT:    pand %xmm3, %xmm0
+; X64-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; X64-NEXT:    pand %xmm2, %xmm0
+; X64-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; X64-NEXT:    pand %xmm2, %xmm1
 ; X64-NEXT:    packuswb %xmm0, %xmm1
 ; X64-NEXT:    movdqa %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; X64-XOP-LABEL: mul_v16i8_0_1_3_7_15_31_63_127_0_1_3_7_15_31_63_127:
 ; X64-XOP:       # %bb.0:
-; X64-XOP-NEXT:    vpmovsxbw %xmm0, %xmm1
-; X64-XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,3,7,15,31,63,127]
-; X64-XOP-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
-; X64-XOP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-XOP-NEXT:    vpmovsxbw %xmm0, %xmm0
-; X64-XOP-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; X64-XOP-NEXT:    vpperm {{.*#+}} xmm0 = xmm1[0,2,4,6,8,10,12,14],xmm0[0,2,4,6,8,10,12,14]
+; X64-XOP-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,3,7,15,31,63,127,0,1,3,7,15,31,63,127]
+; X64-XOP-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-XOP-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X64-XOP-NEXT:    vpmullw %xmm1, %xmm2, %xmm1
+; X64-XOP-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X64-XOP-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; X64-XOP-NEXT:    vpperm {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],xmm1[0,2,4,6,8,10,12,14]
 ; X64-XOP-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: mul_v16i8_0_1_3_7_15_31_63_127_0_1_3_7_15_31_63_127:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    vpmovsxbw %xmm0, %ymm0
 ; X64-AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>