[llvm] 4ceade0 - [X86] Combine concat(shufps, shufps) -> shufps(concat, concat)

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Sat Mar 21 05:44:30 PDT 2020


Author: Simon Pilgrim
Date: 2020-03-21T12:44:10Z
New Revision: 4ceade04284500ca960f6c88b546ba076ec6a643

URL: https://github.com/llvm/llvm-project/commit/4ceade04284500ca960f6c88b546ba076ec6a643
DIFF: https://github.com/llvm/llvm-project/commit/4ceade04284500ca960f6c88b546ba076ec6a643.diff

LOG: [X86] Combine concat(shufps,shufps) -> shufps(concat,concat)

Now that rG18c19441d105 has improved VPERM2X128 handling, we can perform this to improve x64->x32 truncation without poor cross-lane issues.

Someday combineX86ShufflesRecursively will handle this, but we're still really bad at dealing with different vector widths.

Added: 
    

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp
    llvm/test/CodeGen/X86/masked_store_trunc.ll
    llvm/test/CodeGen/X86/masked_store_trunc_ssat.ll
    llvm/test/CodeGen/X86/masked_store_trunc_usat.ll
    llvm/test/CodeGen/X86/pr40891.ll
    llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
    llvm/test/CodeGen/X86/vector-reduce-or-bool.ll
    llvm/test/CodeGen/X86/vector-reduce-xor-bool.ll
    llvm/test/CodeGen/X86/vector-trunc-math.ll
    llvm/test/CodeGen/X86/vector-trunc-packus.ll
    llvm/test/CodeGen/X86/vector-trunc-ssat.ll
    llvm/test/CodeGen/X86/vector-trunc-usat.ll
    llvm/test/CodeGen/X86/vector-trunc.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index f590dab6c44f..c326315b38d0 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -46402,6 +46402,24 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
       })) {
     unsigned NumOps = Ops.size();
     switch (Op0.getOpcode()) {
+    case X86ISD::SHUFP: {
+      // Add SHUFPD support if/when necessary.
+      if (!IsSplat && VT.getScalarType() == MVT::f32 &&
+          llvm::all_of(Ops, [Op0](SDValue Op) {
+            return Op.getOperand(2) == Op0.getOperand(2);
+          })) {
+        SmallVector<SDValue, 2> LHS, RHS;
+        for (unsigned i = 0; i != NumOps; ++i) {
+          LHS.push_back(Ops[i].getOperand(0));
+          RHS.push_back(Ops[i].getOperand(1));
+        }
+        return DAG.getNode(Op0.getOpcode(), DL, VT,
+                           DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, LHS),
+                           DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, RHS),
+                           Op0.getOperand(2));
+      }
+      break;
+    }
     case X86ISD::PSHUFHW:
     case X86ISD::PSHUFLW:
     case X86ISD::PSHUFD:

diff  --git a/llvm/test/CodeGen/X86/masked_store_trunc.ll b/llvm/test/CodeGen/X86/masked_store_trunc.ll
index 5c8c40ed4bad..2feb0382d8c9 100644
--- a/llvm/test/CodeGen/X86/masked_store_trunc.ll
+++ b/llvm/test/CodeGen/X86/masked_store_trunc.ll
@@ -163,11 +163,9 @@ define void @truncstore_v8i64_v8i32(<8 x i64> %x, <8 x i32>* %p, <8 x i32> %mask
 ; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpxor %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm3[0,2]
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3],ymm1[2,3]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm3[0,2],ymm0[4,6],ymm3[4,6]
 ; AVX1-NEXT:    vmaskmovps %ymm0, %ymm2, (%rdi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -178,11 +176,9 @@ define void @truncstore_v8i64_v8i32(<8 x i64> %x, <8 x i32>* %p, <8 x i32> %mask
 ; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
 ; AVX2-NEXT:    vpxor %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm3[0,2]
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3],ymm1[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm3[0,2],ymm0[4,6],ymm3[4,6]
 ; AVX2-NEXT:    vpmaskmovd %ymm0, %ymm2, (%rdi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -457,11 +453,9 @@ define void @truncstore_v8i64_v8i16(<8 x i64> %x, <8 x i16>* %p, <8 x i32> %mask
 ; AVX2-LABEL: truncstore_v8i64_v8i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm4[0,2]
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm4[0,2]
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm0[2,3],ymm1[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm4[0,2],ymm0[4,6],ymm4[4,6]
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm1

diff  --git a/llvm/test/CodeGen/X86/masked_store_trunc_ssat.ll b/llvm/test/CodeGen/X86/masked_store_trunc_ssat.ll
index b79fb5c35f3d..bde5ea6e9bde 100644
--- a/llvm/test/CodeGen/X86/masked_store_trunc_ssat.ll
+++ b/llvm/test/CodeGen/X86/masked_store_trunc_ssat.ll
@@ -300,26 +300,26 @@ define void @truncstore_v8i64_v8i32(<8 x i64> %x, <8 x i32>* %p, <8 x i32> %mask
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [2147483647,2147483647]
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm9
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm4, %xmm6
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; AVX1-NEXT:    vpcmpgtq %xmm7, %xmm4, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
+; AVX1-NEXT:    vpcmpgtq %xmm6, %xmm4, %xmm7
+; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm4, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm4, %xmm5
 ; AVX1-NEXT:    vblendvpd %xmm5, %xmm0, %xmm4, %xmm0
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [18446744071562067968,18446744071562067968]
 ; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm0, %xmm10
-; AVX1-NEXT:    vblendvpd %xmm2, %xmm7, %xmm4, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm2, %xmm7
-; AVX1-NEXT:    vblendvpd %xmm6, %xmm1, %xmm4, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm1, %xmm6
+; AVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm4, %xmm1
+; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm1, %xmm2
+; AVX1-NEXT:    vblendvpd %xmm7, %xmm6, %xmm4, %xmm6
+; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm6, %xmm7
 ; AVX1-NEXT:    vblendvpd %xmm9, %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm3, %xmm4
 ; AVX1-NEXT:    vblendvpd %xmm4, %xmm3, %xmm5, %xmm3
-; AVX1-NEXT:    vblendvpd %xmm6, %xmm1, %xmm5, %xmm1
-; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm3[0,2]
-; AVX1-NEXT:    vblendvpd %xmm7, %xmm2, %xmm5, %xmm2
+; AVX1-NEXT:    vblendvpd %xmm7, %xmm6, %xmm5, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
+; AVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vblendvpd %xmm10, %xmm0, %xmm5, %xmm0
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm3[0,2],ymm0[4,6],ymm3[4,6]
 ; AVX1-NEXT:    vmaskmovps %ymm0, %ymm8, (%rdi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -331,20 +331,18 @@ define void @truncstore_v8i64_v8i32(<8 x i64> %x, <8 x i32>* %p, <8 x i32> %mask
 ; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
 ; AVX2-NEXT:    vpxor %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [2147483647,2147483647,2147483647,2147483647]
-; AVX2-NEXT:    vpcmpgtq %ymm1, %ymm3, %ymm4
-; AVX2-NEXT:    vblendvpd %ymm4, %ymm1, %ymm3, %ymm1
 ; AVX2-NEXT:    vpcmpgtq %ymm0, %ymm3, %ymm4
 ; AVX2-NEXT:    vblendvpd %ymm4, %ymm0, %ymm3, %ymm0
+; AVX2-NEXT:    vpcmpgtq %ymm1, %ymm3, %ymm4
+; AVX2-NEXT:    vblendvpd %ymm4, %ymm1, %ymm3, %ymm1
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968]
-; AVX2-NEXT:    vpcmpgtq %ymm3, %ymm0, %ymm4
-; AVX2-NEXT:    vblendvpd %ymm4, %ymm0, %ymm3, %ymm0
 ; AVX2-NEXT:    vpcmpgtq %ymm3, %ymm1, %ymm4
 ; AVX2-NEXT:    vblendvpd %ymm4, %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm3[0,2]
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
+; AVX2-NEXT:    vpcmpgtq %ymm3, %ymm0, %ymm4
+; AVX2-NEXT:    vblendvpd %ymm4, %ymm0, %ymm3, %ymm0
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3],ymm1[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm3[0,2],ymm0[4,6],ymm3[4,6]
 ; AVX2-NEXT:    vpmaskmovd %ymm0, %ymm2, (%rdi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq

diff  --git a/llvm/test/CodeGen/X86/masked_store_trunc_usat.ll b/llvm/test/CodeGen/X86/masked_store_trunc_usat.ll
index ab22ba3b8f92..84958fef6a32 100644
--- a/llvm/test/CodeGen/X86/masked_store_trunc_usat.ll
+++ b/llvm/test/CodeGen/X86/masked_store_trunc_usat.ll
@@ -240,22 +240,22 @@ define void @truncstore_v8i64_v8i32(<8 x i64> %x, <8 x i32>* %p, <8 x i32> %mask
 ; AVX1-NEXT:    vpxor %xmm3, %xmm0, %xmm4
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [9223372041149743103,9223372041149743103]
 ; AVX1-NEXT:    vpcmpgtq %xmm4, %xmm5, %xmm9
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; AVX1-NEXT:    vpxor %xmm3, %xmm6, %xmm7
-; AVX1-NEXT:    vpcmpgtq %xmm7, %xmm5, %xmm7
-; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm2
+; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm6
+; AVX1-NEXT:    vpcmpgtq %xmm6, %xmm5, %xmm6
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
+; AVX1-NEXT:    vpxor %xmm3, %xmm7, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm5, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
 ; AVX1-NEXT:    vpxor %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm5, %xmm3
 ; AVX1-NEXT:    vmovapd {{.*#+}} xmm5 = [4294967295,4294967295]
 ; AVX1-NEXT:    vblendvpd %xmm3, %xmm4, %xmm5, %xmm3
-; AVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm5, %xmm1
-; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm3[0,2]
-; AVX1-NEXT:    vblendvpd %xmm7, %xmm6, %xmm5, %xmm2
+; AVX1-NEXT:    vblendvpd %xmm2, %xmm7, %xmm5, %xmm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX1-NEXT:    vblendvpd %xmm6, %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vblendvpd %xmm9, %xmm0, %xmm5, %xmm0
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX1-NEXT:    vmaskmovps %ymm0, %ymm8, (%rdi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -268,18 +268,16 @@ define void @truncstore_v8i64_v8i32(<8 x i64> %x, <8 x i32>* %p, <8 x i32> %mask
 ; AVX2-NEXT:    vpxor %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [4294967295,4294967295,4294967295,4294967295]
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
-; AVX2-NEXT:    vpxor %ymm4, %ymm0, %ymm5
+; AVX2-NEXT:    vpxor %ymm4, %ymm1, %ymm5
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm6 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
 ; AVX2-NEXT:    vpcmpgtq %ymm5, %ymm6, %ymm5
-; AVX2-NEXT:    vblendvpd %ymm5, %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpxor %ymm4, %ymm1, %ymm4
+; AVX2-NEXT:    vblendvpd %ymm5, %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpxor %ymm4, %ymm0, %ymm4
 ; AVX2-NEXT:    vpcmpgtq %ymm4, %ymm6, %ymm4
-; AVX2-NEXT:    vblendvpd %ymm4, %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm3[0,2]
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
+; AVX2-NEXT:    vblendvpd %ymm4, %ymm0, %ymm3, %ymm0
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3],ymm1[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm3[0,2],ymm0[4,6],ymm3[4,6]
 ; AVX2-NEXT:    vpmaskmovd %ymm0, %ymm2, (%rdi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq

diff  --git a/llvm/test/CodeGen/X86/pr40891.ll b/llvm/test/CodeGen/X86/pr40891.ll
index 817e5e8fd29a..d67739767b21 100644
--- a/llvm/test/CodeGen/X86/pr40891.ll
+++ b/llvm/test/CodeGen/X86/pr40891.ll
@@ -8,11 +8,9 @@ define <8 x i32> @foo(<8 x i64> %x, <4 x i64> %y) {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; CHECK-NEXT:    vandps {{\.LCPI.*}}, %ymm1, %ymm1
-; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; CHECK-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
-; CHECK-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; CHECK-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; CHECK-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; CHECK-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; CHECK-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; CHECK-NEXT:    retl
   %a = shufflevector <4 x i64> %y, <4 x i64> <i64 12345, i64 67890, i64 13579, i64 24680>, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
   %b = and <8 x i64> %x, %a

diff  --git a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
index a18e47cd9984..3457450a3ee0 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
@@ -566,11 +566,9 @@ define i1 @trunc_v8i64_v8i1(<8 x i64>) {
 ;
 ; AVX2-LABEL: trunc_v8i64_v8i1:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpsllw $15, %xmm0, %xmm0

diff  --git a/llvm/test/CodeGen/X86/vector-reduce-or-bool.ll b/llvm/test/CodeGen/X86/vector-reduce-or-bool.ll
index a9d4f50502a7..5a0deab79ae9 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-or-bool.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-or-bool.ll
@@ -560,11 +560,9 @@ define i1 @trunc_v8i64_v8i1(<8 x i64>) {
 ;
 ; AVX2-LABEL: trunc_v8i64_v8i1:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpsllw $15, %xmm0, %xmm0

diff  --git a/llvm/test/CodeGen/X86/vector-reduce-xor-bool.ll b/llvm/test/CodeGen/X86/vector-reduce-xor-bool.ll
index fb2b99543e36..8014f1f41516 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-xor-bool.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-xor-bool.ll
@@ -611,11 +611,9 @@ define i1 @trunc_v8i64_v8i1(<8 x i64>) {
 ;
 ; AVX2-LABEL: trunc_v8i64_v8i1:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpsllw $15, %xmm0, %xmm0

diff  --git a/llvm/test/CodeGen/X86/vector-trunc-math.ll b/llvm/test/CodeGen/X86/vector-trunc-math.ll
index 38cd2a3ae968..24418ece07a5 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-math.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-math.ll
@@ -101,13 +101,11 @@ define <8 x i16> @trunc_add_v8i64_v8i16(<8 x i64> %a0, <8 x i64> %a1) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_add_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vpaddq %ymm2, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpaddq %ymm3, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vpaddq %ymm2, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
@@ -253,25 +251,21 @@ define <16 x i8> @trunc_add_v16i64_v16i8(<16 x i64> %a0, <16 x i64> %a1) nounwin
 ;
 ; AVX2-SLOW-LABEL: trunc_add_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vpaddq %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpaddq %ymm5, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vpaddq %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vpaddq %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpaddq %ymm7, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vpaddq %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vperm2i128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -574,11 +568,9 @@ define <8 x i16> @trunc_add_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_add_const_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpaddw {{.*}}(%rip), %xmm0, %xmm0
@@ -696,21 +688,17 @@ define <16 x i8> @trunc_add_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_add_const_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -959,13 +947,11 @@ define <8 x i16> @trunc_sub_v8i64_v8i16(<8 x i64> %a0, <8 x i64> %a1) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_sub_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vpsubq %ymm2, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpsubq %ymm3, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vpsubq %ymm2, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
@@ -1111,25 +1097,21 @@ define <16 x i8> @trunc_sub_v16i64_v16i8(<16 x i64> %a0, <16 x i64> %a1) nounwin
 ;
 ; AVX2-SLOW-LABEL: trunc_sub_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vpsubq %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpsubq %ymm5, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vpsubq %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vpsubq %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpsubq %ymm7, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vpsubq %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vperm2i128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -1400,11 +1382,9 @@ define <8 x i16> @trunc_sub_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_sub_const_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpsubw {{.*}}(%rip), %xmm0, %xmm0
@@ -1522,21 +1502,17 @@ define <16 x i8> @trunc_sub_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_sub_const_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -1852,19 +1828,15 @@ define <8 x i16> @trunc_mul_v8i64_v8i16(<8 x i64> %a0, <8 x i64> %a1) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_mul_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm4[0,2],ymm0[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
@@ -2426,11 +2398,9 @@ define <8 x i16> @trunc_mul_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_mul_const_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
@@ -2869,13 +2839,11 @@ define <8 x i16> @trunc_and_v8i64_v8i16(<8 x i64> %a0, <8 x i64> %a1) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_and_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vandps %ymm3, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vandps %ymm2, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
@@ -3007,25 +2975,21 @@ define <16 x i8> @trunc_and_v16i64_v16i8(<16 x i64> %a0, <16 x i64> %a1) nounwin
 ;
 ; AVX2-SLOW-LABEL: trunc_and_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vandps %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vandps %ymm5, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vandps %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vandps %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vandps %ymm7, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vandps %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -3269,11 +3233,9 @@ define <8 x i16> @trunc_and_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_and_const_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
@@ -3391,21 +3353,17 @@ define <16 x i8> @trunc_and_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_and_const_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -3646,13 +3604,11 @@ define <8 x i16> @trunc_xor_v8i64_v8i16(<8 x i64> %a0, <8 x i64> %a1) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_xor_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vxorps %ymm2, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vxorps %ymm3, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vxorps %ymm2, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
@@ -3784,25 +3740,21 @@ define <16 x i8> @trunc_xor_v16i64_v16i8(<16 x i64> %a0, <16 x i64> %a1) nounwin
 ;
 ; AVX2-SLOW-LABEL: trunc_xor_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vxorps %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vxorps %ymm5, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vxorps %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vxorps %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vxorps %ymm7, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vxorps %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -4046,11 +3998,9 @@ define <8 x i16> @trunc_xor_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_xor_const_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
@@ -4168,21 +4118,17 @@ define <16 x i8> @trunc_xor_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_xor_const_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -4423,13 +4369,11 @@ define <8 x i16> @trunc_or_v8i64_v8i16(<8 x i64> %a0, <8 x i64> %a1) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_or_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vorps %ymm2, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vorps %ymm3, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vorps %ymm2, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
@@ -4561,25 +4505,21 @@ define <16 x i8> @trunc_or_v16i64_v16i8(<16 x i64> %a0, <16 x i64> %a1) nounwind
 ;
 ; AVX2-SLOW-LABEL: trunc_or_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vorps %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vorps %ymm5, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vorps %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vorps %ymm4, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vorps %ymm7, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vorps %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0
@@ -4823,11 +4763,9 @@ define <8 x i16> @trunc_or_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_or_const_v8i64_v8i16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
@@ -4945,21 +4883,17 @@ define <16 x i8> @trunc_or_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
 ;
 ; AVX2-SLOW-LABEL: trunc_or_const_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,2],xmm4[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm4 = ymm2[2,3],ymm3[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2],ymm4[0,2],ymm2[4,6],ymm4[4,6]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm5[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm5[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm5[0,2],ymm0[4,6],ymm5[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpand %xmm4, %xmm0, %xmm0

diff  --git a/llvm/test/CodeGen/X86/vector-trunc-packus.ll b/llvm/test/CodeGen/X86/vector-trunc-packus.ll
index 80ce22ca93a7..c74b87921c44 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-packus.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-packus.ll
@@ -899,25 +899,25 @@ define <8 x i32> @trunc_packus_v8i64_v8i32(<8 x i64>* %p0) "min-legal-vector-wid
 ; AVX1-NEXT:    vmovdqa 48(%rdi), %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4294967295,4294967295]
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm8
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm4, %xmm6
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm4, %xmm7
+; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm4, %xmm6
+; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm4, %xmm7
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm4, %xmm5
 ; AVX1-NEXT:    vblendvpd %xmm5, %xmm0, %xmm4, %xmm0
 ; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
 ; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm0, %xmm9
-; AVX1-NEXT:    vblendvpd %xmm7, %xmm1, %xmm4, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm1, %xmm7
-; AVX1-NEXT:    vblendvpd %xmm6, %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm2, %xmm6
+; AVX1-NEXT:    vblendvpd %xmm7, %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm2, %xmm7
+; AVX1-NEXT:    vblendvpd %xmm6, %xmm1, %xmm4, %xmm1
+; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm1, %xmm6
 ; AVX1-NEXT:    vblendvpd %xmm8, %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm3, %xmm4
 ; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpand %xmm2, %xmm6, %xmm2
-; AVX1-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
-; AVX1-NEXT:    vpand %xmm1, %xmm7, %xmm1
+; AVX1-NEXT:    vpand %xmm1, %xmm6, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vpand %xmm2, %xmm7, %xmm2
 ; AVX1-NEXT:    vpand %xmm0, %xmm9, %xmm0
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm1[0,2],ymm0[4,6],ymm1[4,6]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_packus_v8i64_v8i32:
@@ -925,20 +925,18 @@ define <8 x i32> @trunc_packus_v8i64_v8i32(<8 x i64>* %p0) "min-legal-vector-wid
 ; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm1
 ; AVX2-SLOW-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,4294967295]
-; AVX2-SLOW-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
-; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vpcmpgtq %ymm0, %ymm2, %ymm3
 ; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
+; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm3
-; AVX2-SLOW-NEXT:    vpand %ymm0, %ymm3, %ymm0
-; AVX2-SLOW-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm2
-; AVX2-SLOW-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm3
+; AVX2-SLOW-NEXT:    vpand %ymm1, %ymm3, %ymm1
+; AVX2-SLOW-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm2
+; AVX2-SLOW-NEXT:    vpand %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: trunc_packus_v8i64_v8i32:

diff  --git a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
index 543708caaea8..35a50b59fbf3 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
@@ -905,25 +905,25 @@ define <8 x i32> @trunc_ssat_v8i64_v8i32(<8 x i64>* %p0) "min-legal-vector-width
 ; AVX1-NEXT:    vmovdqa 48(%rdi), %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [2147483647,2147483647]
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm8
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm4, %xmm6
-; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm4, %xmm7
+; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm4, %xmm6
+; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm4, %xmm7
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm4, %xmm5
 ; AVX1-NEXT:    vblendvpd %xmm5, %xmm0, %xmm4, %xmm0
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [18446744071562067968,18446744071562067968]
 ; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm0, %xmm9
-; AVX1-NEXT:    vblendvpd %xmm7, %xmm1, %xmm4, %xmm1
-; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm1, %xmm7
-; AVX1-NEXT:    vblendvpd %xmm6, %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm2, %xmm6
+; AVX1-NEXT:    vblendvpd %xmm7, %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm2, %xmm7
+; AVX1-NEXT:    vblendvpd %xmm6, %xmm1, %xmm4, %xmm1
+; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm1, %xmm6
 ; AVX1-NEXT:    vblendvpd %xmm8, %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm3, %xmm4
 ; AVX1-NEXT:    vblendvpd %xmm4, %xmm3, %xmm5, %xmm3
-; AVX1-NEXT:    vblendvpd %xmm6, %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
-; AVX1-NEXT:    vblendvpd %xmm7, %xmm1, %xmm5, %xmm1
+; AVX1-NEXT:    vblendvpd %xmm6, %xmm1, %xmm5, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vblendvpd %xmm7, %xmm2, %xmm5, %xmm2
 ; AVX1-NEXT:    vblendvpd %xmm9, %xmm0, %xmm5, %xmm0
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm1[0,2],ymm0[4,6],ymm1[4,6]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_ssat_v8i64_v8i32:
@@ -931,20 +931,18 @@ define <8 x i32> @trunc_ssat_v8i64_v8i32(<8 x i64>* %p0) "min-legal-vector-width
 ; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm1
 ; AVX2-SLOW-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2147483647,2147483647,2147483647,2147483647]
-; AVX2-SLOW-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
-; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vpcmpgtq %ymm0, %ymm2, %ymm3
 ; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
+; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968]
-; AVX2-SLOW-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm3
-; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm3
 ; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm3
+; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: trunc_ssat_v8i64_v8i32:

diff  --git a/llvm/test/CodeGen/X86/vector-trunc-usat.ll b/llvm/test/CodeGen/X86/vector-trunc-usat.ll
index 7afdb04f6b8f..8434bac380df 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-usat.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-usat.ll
@@ -611,20 +611,20 @@ define <8 x i32> @trunc_usat_v8i64_v8i32(<8 x i64>* %p0) {
 ; AVX1-NEXT:    vpxor %xmm4, %xmm0, %xmm5
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [9223372041149743103,9223372041149743103]
 ; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm6, %xmm8
-; AVX1-NEXT:    vpxor %xmm4, %xmm1, %xmm7
+; AVX1-NEXT:    vpxor %xmm4, %xmm2, %xmm7
 ; AVX1-NEXT:    vpcmpgtq %xmm7, %xmm6, %xmm7
-; AVX1-NEXT:    vpxor %xmm4, %xmm2, %xmm5
+; AVX1-NEXT:    vpxor %xmm4, %xmm1, %xmm5
 ; AVX1-NEXT:    vpcmpgtq %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpxor %xmm4, %xmm3, %xmm4
 ; AVX1-NEXT:    vpcmpgtq %xmm4, %xmm6, %xmm4
 ; AVX1-NEXT:    vmovapd {{.*#+}} xmm6 = [4294967295,4294967295]
 ; AVX1-NEXT:    vblendvpd %xmm4, %xmm3, %xmm6, %xmm3
-; AVX1-NEXT:    vblendvpd %xmm5, %xmm2, %xmm6, %xmm2
-; AVX1-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
-; AVX1-NEXT:    vblendvpd %xmm7, %xmm1, %xmm6, %xmm1
+; AVX1-NEXT:    vblendvpd %xmm5, %xmm1, %xmm6, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX1-NEXT:    vblendvpd %xmm7, %xmm2, %xmm6, %xmm2
 ; AVX1-NEXT:    vblendvpd %xmm8, %xmm0, %xmm6, %xmm0
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm1[0,2],ymm0[4,6],ymm1[4,6]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_usat_v8i64_v8i32:
@@ -633,18 +633,16 @@ define <8 x i32> @trunc_usat_v8i64_v8i32(<8 x i64>* %p0) {
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm1
 ; AVX2-SLOW-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,4294967295]
 ; AVX2-SLOW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
-; AVX2-SLOW-NEXT:    vpxor %ymm3, %ymm0, %ymm4
+; AVX2-SLOW-NEXT:    vpxor %ymm3, %ymm1, %ymm4
 ; AVX2-SLOW-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
 ; AVX2-SLOW-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
-; AVX2-SLOW-NEXT:    vblendvpd %ymm4, %ymm0, %ymm2, %ymm0
-; AVX2-SLOW-NEXT:    vpxor %ymm3, %ymm1, %ymm3
+; AVX2-SLOW-NEXT:    vblendvpd %ymm4, %ymm1, %ymm2, %ymm1
+; AVX2-SLOW-NEXT:    vpxor %ymm3, %ymm0, %ymm3
 ; AVX2-SLOW-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
-; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: trunc_usat_v8i64_v8i32:

diff  --git a/llvm/test/CodeGen/X86/vector-trunc.ll b/llvm/test/CodeGen/X86/vector-trunc.ll
index 36f7d46cd332..94b08c9abb82 100644
--- a/llvm/test/CodeGen/X86/vector-trunc.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc.ll
@@ -20,20 +20,16 @@ define <8 x i32> @trunc8i64_8i32(<8 x i64> %a) {
 ;
 ; AVX1-LABEL: trunc8i64_8i32:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc8i64_8i32:
 ; AVX2-SLOW:       # %bb.0: # %entry
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: trunc8i64_8i32:
@@ -63,20 +59,16 @@ define <8 x i32> @trunc8i64_8i32_ashr(<8 x i64> %a) {
 ;
 ; AVX1-LABEL: trunc8i64_8i32_ashr:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[1,3],xmm3[1,3]
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm2[1,3]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm2[1,3],ymm0[5,7],ymm2[5,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc8i64_8i32_ashr:
 ; AVX2-SLOW:       # %bb.0: # %entry
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[1,3],xmm2[1,3]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm2[1,3]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm2[1,3],ymm0[5,7],ymm2[5,7]
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: trunc8i64_8i32_ashr:
@@ -108,22 +100,18 @@ define <8 x i32> @trunc8i64_8i32_lshr(<8 x i64> %a) {
 ;
 ; AVX1-LABEL: trunc8i64_8i32_lshr:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[1,3],xmm3[1,3]
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm2[1,3]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm2[1,3],ymm0[5,7],ymm2[5,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc8i64_8i32_lshr:
 ; AVX2-SLOW:       # %bb.0: # %entry
-; AVX2-SLOW-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpsrlq $32, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextracti128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vpsrlq $32, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: trunc8i64_8i32_lshr:
@@ -205,11 +193,9 @@ define <8 x i16> @trunc8i64_8i16(<8 x i64> %a) {
 ;
 ; AVX2-SLOW-LABEL: trunc8i64_8i16:
 ; AVX2-SLOW:       # %bb.0: # %entry
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-SLOW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
@@ -1361,20 +1347,16 @@ define <8 x i32> @trunc2x4i64_8i32(<4 x i64> %a, <4 x i64> %b) {
 ;
 ; AVX1-LABEL: trunc2x4i64_8i32:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc2x4i64_8i32:
 ; AVX2-SLOW:       # %bb.0: # %entry
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
-; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
 ; AVX2-SLOW-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: trunc2x4i64_8i32:


        


More information about the llvm-commits mailing list