[SelectionDAG] Add partial BITCAST support to computeKnownBits

[RKSimon]

Summary

Adds support for bitcasting a 'small element' vector to 'large element' scalar/vector (e.g. v16i8 to v4i32 or v2i32 to i64), which is required for PR30845. We extract the knownbits for each 'small element' part and concatenate the results together.

I have the code ready for 'large element' scalar/vector to 'small element' vector bitcasting as well if you think it worth adding to this patch now before we have any useful test cases.

[RKSimon]

ping?

[efriedma]

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

2174 ↗

(On Diff #79300)

(Phabricator won't let me delete this.)

2176 ↗

(On Diff #79300)

No testcase for a big-endian target?

2179 ↗

(On Diff #79300)

You're calling computeKnownBits multiple times with different SubDemandedElts to try and get more resolution out of it? Does that actually do anything useful in practice, as opposed to calling it once? If it does, please put the explanation in a comment.

[RKSimon]

Updated based on Eli's comments.

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

2176 ↗

(On Diff #79300)

I'm struggling to create a big-endian test case for this (its only happening on x86 with use of a particular 'extending' intrinsic). So I'm proposing we don't do this for non little endian targets for now.

[efriedma]

LGTM.

llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

if (SubIdx && SubIdx->getAPIntValue().ule(NumSrcElts - NumElts)) {
       uint64_t Idx = SubIdx->getZExtValue();
       APInt DemandedSrc = DemandedElts.zext(NumSrcElts).shl(Idx);
       computeKnownBits(Src, KnownZero, KnownOne, DemandedSrc, Depth + 1);
     } else {
       computeKnownBits(Src, KnownZero, KnownOne, Depth + 1);
     }
     break;
   }
+  case ISD::BITCAST: {
+    SDValue N0 = Op.getOperand(0);
+    unsigned SubBitWidth = N0.getScalarValueSizeInBits();
+
+    // Ignore bitcasts from floating point.
+    if (!N0.getValueType().isInteger())
+      break;
+
+    // Fast handling of 'identity' bitcasts.
+    if (BitWidth == SubBitWidth) {
+      computeKnownBits(N0, KnownZero, KnownOne, DemandedElts, Depth + 1);
+      break;
+    }
+
+    // Support big-endian targets when it becomes useful.
+    bool IsLE = getDataLayout().isLittleEndian();
+    if (!IsLE)
+      break;
+
+    // Bitcast 'small element' vector to 'large element' scalar/vector.
+    if ((BitWidth % SubBitWidth) == 0) {
+      assert(N0.getValueType().isVector() && "Expected bitcast from vector");
+
+      // Collect known bits for the (larger) output by collecting the known
+      // bits from each set of sub elements and shift these into place.
+      // We need to separately call computeKnownBits for each set of
+      // sub elements as the knownbits for each is likely to be different.
+      unsigned SubScale = BitWidth / SubBitWidth;
+      APInt SubDemandedElts(NumElts * SubScale, 0);
+      for (unsigned i = 0; i != NumElts; ++i)
+        if (DemandedElts[i])
+          SubDemandedElts.setBit(i * SubScale);
+
+      for (unsigned i = 0; i != SubScale; ++i) {
+        computeKnownBits(N0, KnownZero2, KnownOne2, SubDemandedElts.shl(i),
+                         Depth + 1);
+        KnownOne |= KnownOne2.zext(BitWidth).shl(SubBitWidth * i);
+        KnownZero |= KnownZero2.zext(BitWidth).shl(SubBitWidth * i);
+      }
+    }
+
+    // TODO - support ((SubBitWidth % BitWidth) == 0) when it becomes useful.
+    break;
+  }
   case ISD::AND:
     // If either the LHS or the RHS are Zero, the result is zero.
     computeKnownBits(Op.getOperand(1), KnownZero, KnownOne, DemandedElts,
                      Depth + 1);
     computeKnownBits(Op.getOperand(0), KnownZero2, KnownOne2, DemandedElts,
                      Depth + 1);
 
     // Output known-1 bits are only known if set in both the LHS & RHS.

llvm/trunk/test/CodeGen/X86/pmul.ll

entry:
   %A = mul <64 x i8> %i, %j
   ret <64 x i8> %A
 }
 
 ; PR30845
 define <4 x i32> @mul_v4i64_zero_upper(<4 x i32> %val1, <4 x i32> %val2) {
 ; SSE2-LABEL: mul_v4i64_zero_upper:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1]
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    pmuludq %xmm1, %xmm4
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrlq $32, %xmm5
-; SSE2-NEXT:    pmuludq %xmm0, %xmm5
-; SSE2-NEXT:    psllq $32, %xmm5
-; SSE2-NEXT:    psrlq $32, %xmm0
-; SSE2-NEXT:    pmuludq %xmm1, %xmm0
-; SSE2-NEXT:    psllq $32, %xmm0
-; SSE2-NEXT:    paddq %xmm5, %xmm0
-; SSE2-NEXT:    paddq %xmm4, %xmm0
-; SSE2-NEXT:    movdqa %xmm2, %xmm1
-; SSE2-NEXT:    pmuludq %xmm3, %xmm1
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    psrlq $32, %xmm4
-; SSE2-NEXT:    pmuludq %xmm2, %xmm4
-; SSE2-NEXT:    psllq $32, %xmm4
-; SSE2-NEXT:    psrlq $32, %xmm2
-; SSE2-NEXT:    pmuludq %xmm3, %xmm2
-; SSE2-NEXT:    psllq $32, %xmm2
-; SSE2-NEXT:    paddq %xmm4, %xmm2
-; SSE2-NEXT:    paddq %xmm1, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[1,3,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm2[2],xmm3[3],xmm2[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    movdqa %xmm1, %xmm4
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm2[2],xmm4[3],xmm2[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; SSE2-NEXT:    pmuludq %xmm0, %xmm1
+; SSE2-NEXT:    pmuludq %xmm3, %xmm4
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,3,2,3]
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v4i64_zero_upper:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pxor %xmm3, %xmm3
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
+; SSE41-NEXT:    pxor %xmm2, %xmm2
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero
+; SSE41-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pmuludq %xmm1, %xmm3
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrlq $32, %xmm5
-; SSE41-NEXT:    pmuludq %xmm0, %xmm5
-; SSE41-NEXT:    psllq $32, %xmm5
-; SSE41-NEXT:    psrlq $32, %xmm0
-; SSE41-NEXT:    pmuludq %xmm1, %xmm0
-; SSE41-NEXT:    psllq $32, %xmm0
-; SSE41-NEXT:    paddq %xmm5, %xmm0
-; SSE41-NEXT:    paddq %xmm3, %xmm0
-; SSE41-NEXT:    movdqa %xmm2, %xmm1
-; SSE41-NEXT:    pmuludq %xmm4, %xmm1
-; SSE41-NEXT:    movdqa %xmm4, %xmm3
-; SSE41-NEXT:    psrlq $32, %xmm3
-; SSE41-NEXT:    pmuludq %xmm2, %xmm3
-; SSE41-NEXT:    psllq $32, %xmm3
-; SSE41-NEXT:    psrlq $32, %xmm2
-; SSE41-NEXT:    pmuludq %xmm4, %xmm2
-; SSE41-NEXT:    psllq $32, %xmm2
-; SSE41-NEXT:    paddq %xmm3, %xmm2
-; SSE41-NEXT:    paddq %xmm1, %xmm2
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,1,1,3]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,3,2,3]
+; SSE41-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; SSE41-NEXT:    pmuludq %xmm0, %xmm1
+; SSE41-NEXT:    pmuludq %xmm3, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,1,3]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm4[1,3,2,3]
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v4i64_zero_upper:
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; AVX2-NEXT:    vpmuludq %ymm1, %ymm0, %ymm0
entry:
   %rescast = bitcast <4 x i64> %res64 to <8 x i32>
   %res = shufflevector <8 x i32> %rescast, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
   ret <4 x i32> %res
 }
 
 define <4 x i32> @mul_v4i64_zero_upper_left(<4 x i32> %val1, <4 x i64> %val2) {
 ; SSE2-LABEL: mul_v4i64_zero_upper_left:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
+; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    pmuludq %xmm1, %xmm4
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrlq $32, %xmm5
-; SSE2-NEXT:    pmuludq %xmm0, %xmm5
-; SSE2-NEXT:    psllq $32, %xmm5
-; SSE2-NEXT:    psrlq $32, %xmm0
-; SSE2-NEXT:    pmuludq %xmm1, %xmm0
-; SSE2-NEXT:    psllq $32, %xmm0
-; SSE2-NEXT:    paddq %xmm5, %xmm0
-; SSE2-NEXT:    paddq %xmm4, %xmm0
-; SSE2-NEXT:    movdqa %xmm3, %xmm1
-; SSE2-NEXT:    pmuludq %xmm2, %xmm1
-; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    psrlq $32, %xmm4
-; SSE2-NEXT:    pmuludq %xmm3, %xmm4
-; SSE2-NEXT:    psllq $32, %xmm4
-; SSE2-NEXT:    psrlq $32, %xmm3
-; SSE2-NEXT:    pmuludq %xmm2, %xmm3
-; SSE2-NEXT:    psllq $32, %xmm3
-; SSE2-NEXT:    paddq %xmm4, %xmm3
-; SSE2-NEXT:    paddq %xmm1, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; SSE2-NEXT:    psrlq $32, %xmm1
+; SSE2-NEXT:    pmuludq %xmm0, %xmm1
+; SSE2-NEXT:    psllq $32, %xmm1
+; SSE2-NEXT:    paddq %xmm3, %xmm1
+; SSE2-NEXT:    movdqa %xmm4, %xmm0
+; SSE2-NEXT:    pmuludq %xmm2, %xmm0
+; SSE2-NEXT:    psrlq $32, %xmm2
+; SSE2-NEXT:    pmuludq %xmm4, %xmm2
+; SSE2-NEXT:    psllq $32, %xmm2
+; SSE2-NEXT:    paddq %xmm0, %xmm2
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,3,2,3]
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v4i64_zero_upper_left:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pxor %xmm4, %xmm4
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pmuludq %xmm1, %xmm4
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrlq $32, %xmm5
-; SSE41-NEXT:    pmuludq %xmm3, %xmm5
-; SSE41-NEXT:    psllq $32, %xmm5
-; SSE41-NEXT:    psrlq $32, %xmm3
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm4 = xmm0[0],zero,xmm0[1],zero
+; SSE41-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
+; SSE41-NEXT:    movdqa %xmm4, %xmm3
 ; SSE41-NEXT:    pmuludq %xmm1, %xmm3
-; SSE41-NEXT:    psllq $32, %xmm3
-; SSE41-NEXT:    paddq %xmm5, %xmm3
-; SSE41-NEXT:    paddq %xmm4, %xmm3
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pmuludq %xmm2, %xmm1
-; SSE41-NEXT:    movdqa %xmm2, %xmm4
-; SSE41-NEXT:    psrlq $32, %xmm4
-; SSE41-NEXT:    pmuludq %xmm0, %xmm4
-; SSE41-NEXT:    psllq $32, %xmm4
-; SSE41-NEXT:    psrlq $32, %xmm0
-; SSE41-NEXT:    pmuludq %xmm2, %xmm0
-; SSE41-NEXT:    psllq $32, %xmm0
-; SSE41-NEXT:    paddq %xmm4, %xmm0
-; SSE41-NEXT:    paddq %xmm1, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,1,1,3]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,3,2,3]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    psrlq $32, %xmm1
+; SSE41-NEXT:    pmuludq %xmm4, %xmm1
+; SSE41-NEXT:    psllq $32, %xmm1
+; SSE41-NEXT:    paddq %xmm3, %xmm1
+; SSE41-NEXT:    movdqa %xmm0, %xmm3
+; SSE41-NEXT:    pmuludq %xmm2, %xmm3
+; SSE41-NEXT:    psrlq $32, %xmm2
+; SSE41-NEXT:    pmuludq %xmm0, %xmm2
+; SSE41-NEXT:    psllq $32, %xmm2
+; SSE41-NEXT:    paddq %xmm3, %xmm2
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,1,3]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,3,2,3]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v4i64_zero_upper_left:
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; AVX2-NEXT:    vpmuludq %ymm1, %ymm0, %ymm2
 ; AVX2-NEXT:    vpsrlq $32, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmuludq %ymm1, %ymm0, %ymm0
entry:
   %rescast = bitcast <4 x i64> %res64 to <8 x i32>
   %res = shufflevector <8 x i32> %rescast, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
   ret <4 x i32> %res
 }
 
 define <4 x i32> @mul_v4i64_zero_lower(<4 x i32> %val1, <4 x i64> %val2) {
 ; SSE2-LABEL: mul_v4i64_zero_lower:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [0,4294967295,0,4294967295]
-; SSE2-NEXT:    pand %xmm4, %xmm2
-; SSE2-NEXT:    pand %xmm4, %xmm1
+; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    pmuludq %xmm1, %xmm4
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrlq $32, %xmm5
-; SSE2-NEXT:    pmuludq %xmm0, %xmm5
-; SSE2-NEXT:    psllq $32, %xmm5
-; SSE2-NEXT:    psrlq $32, %xmm0
-; SSE2-NEXT:    pmuludq %xmm1, %xmm0
-; SSE2-NEXT:    psllq $32, %xmm0
-; SSE2-NEXT:    paddq %xmm5, %xmm0
-; SSE2-NEXT:    paddq %xmm4, %xmm0
-; SSE2-NEXT:    movdqa %xmm3, %xmm1
-; SSE2-NEXT:    pmuludq %xmm2, %xmm1
-; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    psrlq $32, %xmm4
-; SSE2-NEXT:    pmuludq %xmm3, %xmm4
-; SSE2-NEXT:    psllq $32, %xmm4
-; SSE2-NEXT:    psrlq $32, %xmm3
-; SSE2-NEXT:    pmuludq %xmm2, %xmm3
-; SSE2-NEXT:    psllq $32, %xmm3
-; SSE2-NEXT:    paddq %xmm4, %xmm3
-; SSE2-NEXT:    paddq %xmm1, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [0,4294967295,0,4294967295]
+; SSE2-NEXT:    pand %xmm3, %xmm2
+; SSE2-NEXT:    pand %xmm3, %xmm1
+; SSE2-NEXT:    psrlq $32, %xmm1
+; SSE2-NEXT:    pmuludq %xmm0, %xmm1
+; SSE2-NEXT:    psllq $32, %xmm1
+; SSE2-NEXT:    psrlq $32, %xmm2
+; SSE2-NEXT:    pmuludq %xmm4, %xmm2
+; SSE2-NEXT:    psllq $32, %xmm2
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,3,2,3]
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v4i64_zero_lower:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pxor %xmm4, %xmm4
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm4[0,1],xmm1[2,3],xmm4[4,5],xmm1[6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3],xmm4[4,5],xmm2[6,7]
-; SSE41-NEXT:    movdqa %xmm0, %xmm4
-; SSE41-NEXT:    pmuludq %xmm2, %xmm4
-; SSE41-NEXT:    movdqa %xmm2, %xmm5
-; SSE41-NEXT:    psrlq $32, %xmm5
-; SSE41-NEXT:    pmuludq %xmm0, %xmm5
-; SSE41-NEXT:    psllq $32, %xmm5
-; SSE41-NEXT:    psrlq $32, %xmm0
-; SSE41-NEXT:    pmuludq %xmm2, %xmm0
-; SSE41-NEXT:    psllq $32, %xmm0
-; SSE41-NEXT:    paddq %xmm5, %xmm0
-; SSE41-NEXT:    paddq %xmm4, %xmm0
-; SSE41-NEXT:    movdqa %xmm3, %xmm2
-; SSE41-NEXT:    pmuludq %xmm1, %xmm2
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    psrlq $32, %xmm4
-; SSE41-NEXT:    pmuludq %xmm3, %xmm4
-; SSE41-NEXT:    psllq $32, %xmm4
-; SSE41-NEXT:    psrlq $32, %xmm3
-; SSE41-NEXT:    pmuludq %xmm1, %xmm3
-; SSE41-NEXT:    psllq $32, %xmm3
-; SSE41-NEXT:    paddq %xmm4, %xmm3
-; SSE41-NEXT:    paddq %xmm2, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,1,1,3]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,3,2,3]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm4 = xmm0[0],zero,xmm0[1],zero
+; SSE41-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm3[0,1],xmm1[2,3],xmm3[4,5],xmm1[6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
+; SSE41-NEXT:    psrlq $32, %xmm2
+; SSE41-NEXT:    pmuludq %xmm0, %xmm2
+; SSE41-NEXT:    psllq $32, %xmm2
+; SSE41-NEXT:    psrlq $32, %xmm1
+; SSE41-NEXT:    pmuludq %xmm4, %xmm1
+; SSE41-NEXT:    psllq $32, %xmm1
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,1,3]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,3,2,3]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v4i64_zero_lower:
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; AVX2-NEXT:    vpxor %ymm2, %ymm2, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4],ymm1[5],ymm2[6],ymm1[7]
-; AVX2-NEXT:    vpmuludq %ymm1, %ymm0, %ymm2
 ; AVX2-NEXT:    vpsrlq $32, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmuludq %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsllq $32, %ymm0, %ymm0
-; AVX2-NEXT:    vpaddq %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,1,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: mul_v4i64_zero_lower:
 ; AVX512:       # BB#0: # %entry
 ; AVX512-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; AVX512-NEXT:    vpxor %ymm2, %ymm2, %ymm2
 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4],ymm1[5],ymm2[6],ymm1[7]
-; AVX512-NEXT:    vpmuludq %ymm1, %ymm0, %ymm2
 ; AVX512-NEXT:    vpsrlq $32, %ymm1, %ymm1
 ; AVX512-NEXT:    vpmuludq %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vpsllq $32, %ymm0, %ymm0
-; AVX512-NEXT:    vpaddq %ymm0, %ymm2, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,1,3]
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; AVX512-NEXT:    retq
 entry:
   %val1a = zext <4 x i32> %val1 to <4 x i64>
   %val2a = and <4 x i64> %val2, <i64 -4294967296, i64 -4294967296, i64 -4294967296, i64 -4294967296>
   %res64 = mul <4 x i64> %val1a, %val2a
   %rescast = bitcast <4 x i64> %res64 to <8 x i32>
   %res = shufflevector <8 x i32> %rescast, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
   ret <4 x i32> %res
 }
 
 define <8 x i32> @mul_v8i64_zero_upper(<8 x i32> %val1, <8 x i32> %val2) {
 ; SSE2-LABEL: mul_v8i64_zero_upper:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    movdqa %xmm0, %xmm5
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm5 = xmm5[2],xmm6[2],xmm5[3],xmm6[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm6[0],xmm0[1],xmm6[1]
-; SSE2-NEXT:    movdqa %xmm1, %xmm4
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm6[2],xmm4[3],xmm6[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1]
-; SSE2-NEXT:    movdqa %xmm2, %xmm8
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm8 = xmm8[2],xmm6[2],xmm8[3],xmm6[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm6[0],xmm2[1],xmm6[1]
-; SSE2-NEXT:    movdqa %xmm3, %xmm7
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm7 = xmm7[2],xmm6[2],xmm7[3],xmm6[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm6[0],xmm3[1],xmm6[1]
-; SSE2-NEXT:    movdqa %xmm1, %xmm9
-; SSE2-NEXT:    pmuludq %xmm3, %xmm9
-; SSE2-NEXT:    movdqa %xmm3, %xmm6
-; SSE2-NEXT:    psrlq $32, %xmm6
-; SSE2-NEXT:    pmuludq %xmm1, %xmm6
-; SSE2-NEXT:    psllq $32, %xmm6
-; SSE2-NEXT:    psrlq $32, %xmm1
-; SSE2-NEXT:    pmuludq %xmm3, %xmm1
-; SSE2-NEXT:    psllq $32, %xmm1
-; SSE2-NEXT:    paddq %xmm6, %xmm1
-; SSE2-NEXT:    paddq %xmm9, %xmm1
-; SSE2-NEXT:    movdqa %xmm4, %xmm3
-; SSE2-NEXT:    pmuludq %xmm7, %xmm3
-; SSE2-NEXT:    movdqa %xmm7, %xmm6
-; SSE2-NEXT:    psrlq $32, %xmm6
-; SSE2-NEXT:    pmuludq %xmm4, %xmm6
-; SSE2-NEXT:    psllq $32, %xmm6
-; SSE2-NEXT:    psrlq $32, %xmm4
-; SSE2-NEXT:    pmuludq %xmm7, %xmm4
-; SSE2-NEXT:    psllq $32, %xmm4
-; SSE2-NEXT:    paddq %xmm6, %xmm4
-; SSE2-NEXT:    paddq %xmm3, %xmm4
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    pmuludq %xmm2, %xmm3
-; SSE2-NEXT:    movdqa %xmm2, %xmm6
-; SSE2-NEXT:    psrlq $32, %xmm6
-; SSE2-NEXT:    pmuludq %xmm0, %xmm6
-; SSE2-NEXT:    psllq $32, %xmm6
-; SSE2-NEXT:    psrlq $32, %xmm0
-; SSE2-NEXT:    pmuludq %xmm2, %xmm0
-; SSE2-NEXT:    psllq $32, %xmm0
-; SSE2-NEXT:    paddq %xmm6, %xmm0
-; SSE2-NEXT:    paddq %xmm3, %xmm0
-; SSE2-NEXT:    movdqa %xmm5, %xmm2
-; SSE2-NEXT:    pmuludq %xmm8, %xmm2
-; SSE2-NEXT:    movdqa %xmm8, %xmm3
-; SSE2-NEXT:    psrlq $32, %xmm3
-; SSE2-NEXT:    pmuludq %xmm5, %xmm3
-; SSE2-NEXT:    psllq $32, %xmm3
-; SSE2-NEXT:    psrlq $32, %xmm5
-; SSE2-NEXT:    pmuludq %xmm8, %xmm5
-; SSE2-NEXT:    psllq $32, %xmm5
-; SSE2-NEXT:    paddq %xmm3, %xmm5
-; SSE2-NEXT:    paddq %xmm2, %xmm5
+; SSE2-NEXT:    pxor %xmm4, %xmm4
+; SSE2-NEXT:    movdqa %xmm0, %xmm8
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm8 = xmm8[2],xmm4[2],xmm8[3],xmm4[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
+; SSE2-NEXT:    movdqa %xmm1, %xmm6
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm6 = xmm6[2],xmm4[2],xmm6[3],xmm4[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1]
+; SSE2-NEXT:    movdqa %xmm2, %xmm7
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm7 = xmm7[2],xmm4[2],xmm7[3],xmm4[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
+; SSE2-NEXT:    movdqa %xmm3, %xmm5
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm5 = xmm5[2],xmm4[2],xmm5[3],xmm4[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
+; SSE2-NEXT:    pmuludq %xmm1, %xmm3
+; SSE2-NEXT:    pmuludq %xmm6, %xmm5
+; SSE2-NEXT:    pmuludq %xmm0, %xmm2
+; SSE2-NEXT:    pmuludq %xmm8, %xmm7
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm7[1,3,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,3,2,3]
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[1,3,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
 ; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: mul_v8i64_zero_upper:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pxor %xmm6, %xmm6
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm5 = xmm0[0],zero,xmm0[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm6[2],xmm0[3],xmm6[3]
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm6[2],xmm1[3],xmm6[3]
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm8 = xmm2[0],zero,xmm2[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm6[2],xmm2[3],xmm6[3]
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm9 = xmm3[0],zero,xmm3[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm6[2],xmm3[3],xmm6[3]
-; SSE41-NEXT:    movdqa %xmm1, %xmm6
-; SSE41-NEXT:    pmuludq %xmm3, %xmm6
-; SSE41-NEXT:    movdqa %xmm3, %xmm7
-; SSE41-NEXT:    psrlq $32, %xmm7
-; SSE41-NEXT:    pmuludq %xmm1, %xmm7
-; SSE41-NEXT:    psllq $32, %xmm7
-; SSE41-NEXT:    psrlq $32, %xmm1
-; SSE41-NEXT:    pmuludq %xmm3, %xmm1
-; SSE41-NEXT:    psllq $32, %xmm1
-; SSE41-NEXT:    paddq %xmm7, %xmm1
-; SSE41-NEXT:    paddq %xmm6, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pmuludq %xmm2, %xmm3
-; SSE41-NEXT:    movdqa %xmm2, %xmm6
-; SSE41-NEXT:    psrlq $32, %xmm6
-; SSE41-NEXT:    pmuludq %xmm0, %xmm6
-; SSE41-NEXT:    psllq $32, %xmm6
-; SSE41-NEXT:    psrlq $32, %xmm0
-; SSE41-NEXT:    pmuludq %xmm2, %xmm0
-; SSE41-NEXT:    psllq $32, %xmm0
-; SSE41-NEXT:    paddq %xmm6, %xmm0
-; SSE41-NEXT:    paddq %xmm3, %xmm0
-; SSE41-NEXT:    movdqa %xmm4, %xmm2
-; SSE41-NEXT:    pmuludq %xmm9, %xmm2
-; SSE41-NEXT:    movdqa %xmm9, %xmm3
-; SSE41-NEXT:    psrlq $32, %xmm3
-; SSE41-NEXT:    pmuludq %xmm4, %xmm3
-; SSE41-NEXT:    psllq $32, %xmm3
-; SSE41-NEXT:    psrlq $32, %xmm4
-; SSE41-NEXT:    pmuludq %xmm9, %xmm4
-; SSE41-NEXT:    psllq $32, %xmm4
-; SSE41-NEXT:    paddq %xmm3, %xmm4
-; SSE41-NEXT:    paddq %xmm2, %xmm4
-; SSE41-NEXT:    movdqa %xmm5, %xmm2
-; SSE41-NEXT:    pmuludq %xmm8, %xmm2
-; SSE41-NEXT:    movdqa %xmm8, %xmm3
-; SSE41-NEXT:    psrlq $32, %xmm3
-; SSE41-NEXT:    pmuludq %xmm5, %xmm3
-; SSE41-NEXT:    psllq $32, %xmm3
-; SSE41-NEXT:    psrlq $32, %xmm5
-; SSE41-NEXT:    pmuludq %xmm8, %xmm5
-; SSE41-NEXT:    psllq $32, %xmm5
-; SSE41-NEXT:    paddq %xmm3, %xmm5
-; SSE41-NEXT:    paddq %xmm2, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,1,3]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[1,3,2,3]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,1,1,3]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[1,3,2,3]
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm8 = xmm0[0],zero,xmm0[1],zero
+; SSE41-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm6 = xmm1[0],zero,xmm1[1],zero
+; SSE41-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm7 = xmm2[0],zero,xmm2[1],zero
+; SSE41-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm5 = xmm3[0],zero,xmm3[1],zero
+; SSE41-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
+; SSE41-NEXT:    pmuludq %xmm1, %xmm3
+; SSE41-NEXT:    pmuludq %xmm0, %xmm2
+; SSE41-NEXT:    pmuludq %xmm6, %xmm5
+; SSE41-NEXT:    pmuludq %xmm8, %xmm7
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,1,1,3]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm7[1,3,2,3]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,1,1,3]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[1,3,2,3]
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX2-LABEL: mul_v8i64_zero_upper:
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; AVX2-NEXT:    vpmuludq %ymm1, %ymm0, %ymm4
-; AVX2-NEXT:    vpsrlq $32, %ymm1, %ymm5
-; AVX2-NEXT:    vpmuludq %ymm5, %ymm0, %ymm5
-; AVX2-NEXT:    vpsllq $32, %ymm5, %ymm5
-; AVX2-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmuludq %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsllq $32, %ymm0, %ymm0
-; AVX2-NEXT:    vpaddq %ymm0, %ymm5, %ymm0
-; AVX2-NEXT:    vpaddq %ymm0, %ymm4, %ymm0
 ; AVX2-NEXT:    vpmuludq %ymm3, %ymm2, %ymm1
-; AVX2-NEXT:    vpsrlq $32, %ymm3, %ymm4
-; AVX2-NEXT:    vpmuludq %ymm4, %ymm2, %ymm4
-; AVX2-NEXT:    vpsllq $32, %ymm4, %ymm4
-; AVX2-NEXT:    vpsrlq $32, %ymm2, %ymm2
-; AVX2-NEXT:    vpmuludq %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpsllq $32, %ymm2, %ymm2
-; AVX2-NEXT:    vpaddq %ymm2, %ymm4, %ymm2
-; AVX2-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,1,1,3,4,5,5,7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[1,3,2,3,5,7,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: mul_v8i64_zero_upper:
 ; AVX512:       # BB#0: # %entry