[X86][SSE] pslldq/psrldq byte shifts/rotation for SSE2

[RKSimon]

Summary

This patch builds on http://reviews.llvm.org/D5598 to improve byte shift/rotation shuffles on pre-SSSE3 (palignr) targets.

I've also made use of the ability of the SLLDQ/SRLDQ instructions to implicitly shift in zero bytes to avoid the need to create a zero register if we had used palignr with a zero register.

[chandlerc]

lib/Target/X86/X86ISelLowering.cpp

7545

Please use the zeroable computation routine that the rest of the vector shuffle lowering uses. There are lots of other patterns that show up which this doesn't necessarily catch.

[RKSimon]

Updated patch for SSE2 byte shifts/rotations.

As requested, I've re-implemented the byte shift shuffles using the zeroable computation routine. I've split the byte shift shuffle detection into a new function lowerVectorShuffleAsByteShift() as it makes it easier to do the zeroable element matching - lowerVectorShuffleAsByteRotate now only adds the basic PSRLDQ/PSLLDQ/POR pattern for pre-SSSE3 rotates.

I also made a fix to ISD::isBuildVectorAllZeros (and ISD::isBuildVectorAllOnes whilst I was there) so that the zero detection can properly see through multiple bitcast operations.

[chandlerc]

lib/Target/X86/X86ISelLowering.cpp

7451–7456

The rest of this comment is now misleading.

I think it should now say something like "It does not check for the profitability of lowering either as PALIGNR or PSRLDQ/PSLLDQ/POR, only whether the mask is valid to lower in that form."

7550–7552

Rather than this, I would suggest testing for SSSE3 prior to calling this in the routines where it is only profitable with SSSE3. That localizes the profitability breakdown to those routines which are already broken out by element size.

7610–7616

The formatting here isn't right. Can you use clang-format? The curly at the least should be on the prior line.

7634–7641

The bitwise ands here are weird, and I think you'll need some comments explaining the nature of your search strategy here.

7655

It would be really nice to write this logic once in a lambda, and run it over a reversed mask.

[RKSimon]

Updated patch for SSE2 byte shifts/rotations.

The patch now (re)uses a lambda to detect the sequential 'shifted' shuffle mask - I've split the shift and zeroable matching sections of the byte-shift routine to make it easier to understand/maintain.

Also, tidied up comments and pushed the byte-rotate profitability check up into the calling functions.

[RKSimon]

ping

[RKSimon]

ping2

[RKSimon]

ping

Chandler - are you happy with the changes I made to the patch? Thanks, Simon.

[RKSimon]

rebased against trunk

[andreadb]

Hi Simon,
For what is worth, the patch LGTM. Chandler - do you have any final comments?

[RKSimon]

Closed by commit rL222340 (authored by @RKSimon).

lib/Target/X86/X86ISelLowering.cpp

 
 /// \brief Try to lower a vector shuffle as a byte rotation.
 ///
-/// We have a generic PALIGNR instruction in x86 that will do an arbitrary
-/// byte-rotation of a the concatentation of two vectors. This routine will
+/// SSSE3 has a generic PALIGNR instruction in x86 that will do an arbitrary
+/// byte-rotation of a concatenation of two vectors; pre-SSSE3 can use
+/// a PSRLDQ/PSLLDQ/POR pattern to get a similar effect. This routine will
 /// try to generically lower a vector shuffle through such an instruction. It
 /// does not check for the availability of PALIGNR-based lowerings, only the
 /// applicability of this strategy to the given mask. This matches shuffle

 static SDValue lowerVectorShuffleAsByteRotate(SDLoc DL, MVT VT, SDValue V1,
                                               SDValue V2,
                                               ArrayRef<int> Mask,
+                                              const X86Subtarget *Subtarget,
                                               SelectionDAG &DAG) {
   assert(!isNoopShuffleMask(Mask) && "We shouldn't lower no-op shuffles!");
 

   else if (!Hi)
     Hi = Lo;
 
-  // Cast the inputs to v16i8 to match PALIGNR.
-  Lo = DAG.getNode(ISD::BITCAST, DL, MVT::v16i8, Lo);
-  Hi = DAG.getNode(ISD::BITCAST, DL, MVT::v16i8, Hi);
-
   assert(VT.getSizeInBits() == 128 &&
          "Rotate-based lowering only supports 128-bit lowering!");
   assert(Mask.size() <= 16 &&
          "Can shuffle at most 16 bytes in a 128-bit vector!");
+
   // The actual rotate instruction rotates bytes, so we need to scale the
   // rotation based on how many bytes are in the vector.
   int Scale = 16 / Mask.size();
+  int LoByteShift = 16 - Rotation * Scale;
+  int HiByteShift = Rotation * Scale;
 
-  return DAG.getNode(ISD::BITCAST, DL, VT,
-                     DAG.getNode(X86ISD::PALIGNR, DL, MVT::v16i8, Hi, Lo,
-                                 DAG.getConstant(Rotation * Scale, MVT::i8)));
+  // If one of the inputs is zero, then we can use PSRLDQ/PSLLDQ to
+  // implicitly shift in zero bytes and reduce register usage.
+  bool LoIsZero = ISD::isBuildVectorAllZeros(Lo.getNode());
+  bool HiIsZero = ISD::isBuildVectorAllZeros(Hi.getNode());
+
+  if (LoIsZero) {
+    // Cast the inputs to v2i64 to match PSRLDQ.
+    Hi = DAG.getNode(ISD::BITCAST, DL, MVT::v2i64, Hi);
+
+    SDValue HiShift = DAG.getNode(X86ISD::VSRLDQ, DL, MVT::v2i64, Hi,
+                                  DAG.getConstant(8 * HiByteShift, MVT::i8));
+    return DAG.getNode(ISD::BITCAST, DL, VT, HiShift);
+  }
+
+  if (HiIsZero) {
+    // Cast the inputs to v2i64 to match PSLLDQ.
+    Lo = DAG.getNode(ISD::BITCAST, DL, MVT::v2i64, Lo);
+
+    SDValue LoShift = DAG.getNode(X86ISD::VSHLDQ, DL, MVT::v2i64, Lo,
+                                  DAG.getConstant(8 * LoByteShift, MVT::i8));
+    return DAG.getNode(ISD::BITCAST, DL, VT, LoShift);
+  }
+
+  if (Subtarget->hasSSSE3()) {
+    // Cast the inputs to v16i8 to match PALIGNR.
+    Lo = DAG.getNode(ISD::BITCAST, DL, MVT::v16i8, Lo);
+    Hi = DAG.getNode(ISD::BITCAST, DL, MVT::v16i8, Hi);
+
+    return DAG.getNode(ISD::BITCAST, DL, VT,
+                       DAG.getNode(X86ISD::PALIGNR, DL, MVT::v16i8, Hi, Lo,
+                                   DAG.getConstant(Rotation * Scale, MVT::i8)));
+  }
+
+  // On pre-SSSE3 targets we will be able to shuffle/unpack more efficiently for
+  // 32-bit or 64-bit integers than using the PSRLDQ/PSLLDQ/POR pattern.
+  if(VT.getScalarSizeInBits() < 32) {
+    // Cast the inputs to v2i64 to match PSLLDQ/PSRLDQ.
+    Lo = DAG.getNode(ISD::BITCAST, DL, MVT::v2i64, Lo);
+    Hi = DAG.getNode(ISD::BITCAST, DL, MVT::v2i64, Hi);
+
+    SDValue LoShift = DAG.getNode(X86ISD::VSHLDQ, DL, MVT::v2i64, Lo,
+                                  DAG.getConstant(8 * LoByteShift, MVT::i8));
+    SDValue HiShift = DAG.getNode(X86ISD::VSRLDQ, DL, MVT::v2i64, Hi,
+                                  DAG.getConstant(8 * HiByteShift, MVT::i8));
+    return DAG.getNode(ISD::BITCAST, DL, VT,
+                       DAG.getNode(ISD::OR, DL, MVT::v2i64, LoShift, HiShift));
+  }
+
+  return SDValue();
 }
 
 /// \brief Compute whether each element of a shuffle is zeroable.

                                                   Subtarget, DAG))
       return Blend;
 
-  // Try to use rotation instructions if available.
-  if (Subtarget->hasSSSE3())
-    if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
-            DL, MVT::v2i64, V1, V2, Mask, DAG))
-      return Rotate;
+  // Try to use byte rotation instructions.
+  if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
+          DL, MVT::v2i64, V1, V2, Mask, Subtarget, DAG))
+    return Rotate;
 
   // We implement this with SHUFPD which is pretty lame because it will likely
   // incur 2 cycles of stall for integer vectors on Nehalem and older chips.

                                                   Subtarget, DAG))
       return Blend;
 
-  // Try to use rotation instructions if available.
-  if (Subtarget->hasSSSE3())
-    if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
-            DL, MVT::v4i32, V1, V2, Mask, DAG))
-      return Rotate;
+  // Try to use byte rotation instructions.
+  if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
+          DL, MVT::v4i32, V1, V2, Mask, Subtarget, DAG))
+    return Rotate;
 
   // We implement this with SHUFPS because it can blend from two vectors.
   // Because we're going to eventually use SHUFPS, we use SHUFPS even to build

   if (isShuffleEquivalent(Mask, 4, 4, 5, 5, 6, 6, 7, 7))
     return DAG.getNode(X86ISD::UNPCKH, DL, MVT::v8i16, V, V);
 
-  // Try to use rotation instructions if available.
-  if (Subtarget->hasSSSE3())
-    if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
-            DL, MVT::v8i16, V, V, Mask, DAG))
-      return Rotate;
+  // Try to use byte rotation instructions.
+  if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
+          DL, MVT::v8i16, V, V, Mask, Subtarget, DAG))
+    return Rotate;
 
   // Simplify the 1-into-3 and 3-into-1 cases with a single pshufd. For all
   // such inputs we can swap two of the dwords across the half mark and end up

                                                   Subtarget, DAG))
       return Blend;
 
-  // Try to use rotation instructions if available.
-  if (Subtarget->hasSSSE3())
-    if (SDValue Rotate = lowerVectorShuffleAsByteRotate(DL, MVT::v8i16, V1, V2, Mask, DAG))
-      return Rotate;
+  // Try to use byte rotation instructions.
+  if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
+          DL, MVT::v8i16, V1, V2, Mask, Subtarget, DAG))
 
   if (NumV1Inputs + NumV2Inputs <= 4)
     return lowerV8I16BasicBlendVectorShuffle(DL, V1, V2, Mask, Subtarget, DAG);

   ArrayRef<int> OrigMask = SVOp->getMask();
   assert(OrigMask.size() == 16 && "Unexpected mask size for v16 shuffle!");
 
-  // Try to use rotation instructions if available.
-  if (Subtarget->hasSSSE3())
-    if (SDValue Rotate = lowerVectorShuffleAsByteRotate(DL, MVT::v16i8, V1, V2,
-                                                        OrigMask, DAG))
-      return Rotate;
+  // Try to use byte rotation instructions.
+  if (SDValue Rotate = lowerVectorShuffleAsByteRotate(
+          DL, MVT::v16i8, V1, V2, OrigMask, Subtarget, DAG))
+    return Rotate;
 
   // Try to use a zext lowering.
   if (SDValue ZExt = lowerVectorShuffleAsZeroOrAnyExtend(

test/CodeGen/X86/palignr.ll

 ;
 ; CHECK-YONAH-LABEL: test6:
 ; CHECK-YONAH:       # BB#0:
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; CHECK-YONAH-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; CHECK-YONAH-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,0,1,2,4,5,6,7]
-; CHECK-YONAH-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,0,2,1,4,5,6,7]
-; CHECK-YONAH-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-YONAH-NEXT:    psrldq $6, %xmm0 {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; CHECK-YONAH-NEXT:    pslldq $10, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5]
+; CHECK-YONAH-NEXT:    por %xmm1, %xmm0
 ; CHECK-YONAH-NEXT:    retl
   %C = shufflevector <8 x i16> %A, <8 x i16> %B, <8 x i32> < i32 3, i32 4, i32 undef, i32 6, i32 7, i32 8, i32 9, i32 10 >
   ret <8 x i16> %C

 ;
 ; CHECK-YONAH-LABEL: test7:
 ; CHECK-YONAH:       # BB#0:
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-YONAH-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,1,4,5,6,7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,2,0]
-; CHECK-YONAH-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,2,3,0,4,5,6,7]
-; CHECK-YONAH-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-YONAH-NEXT:    psrldq $10, %xmm0 # xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; CHECK-YONAH-NEXT:    pslldq $6, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9]
+; CHECK-YONAH-NEXT:    por %xmm1, %xmm0
 ; CHECK-YONAH-NEXT:    retl
   %C = shufflevector <8 x i16> %A, <8 x i16> %B, <8 x i32> < i32 undef, i32 6, i32 undef, i32 8, i32 9, i32 10, i32 11, i32 12 >
   ret <8 x i16> %C

 ;
 ; CHECK-YONAH-LABEL: test8:
 ; CHECK-YONAH:       # BB#0:
-; CHECK-YONAH-NEXT:    pxor %xmm3, %xmm3
-; CHECK-YONAH-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-YONAH-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,2,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,2,3,3,4,5,6,7]
-; CHECK-YONAH-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[3,1,2,0]
-; CHECK-YONAH-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,7,6,7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,1,2,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[1,2,3,0,4,5,6,7]
-; CHECK-YONAH-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm4[0]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,2,3,3,4,5,6,7]
-; CHECK-YONAH-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,2,0]
-; CHECK-YONAH-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,3,2,1]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,2,2,4,5,6,7]
-; CHECK-YONAH-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,6,7,4]
-; CHECK-YONAH-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; CHECK-YONAH-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; CHECK-YONAH-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; CHECK-YONAH-NEXT:    packuswb %xmm0, %xmm2
-; CHECK-YONAH-NEXT:    movdqa %xmm2, %xmm0
+; CHECK-YONAH-NEXT:    psrldq $5, %xmm0 # xmm0 = xmm0[5,6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero
+; CHECK-YONAH-NEXT:    pslldq $11, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4]
+; CHECK-YONAH-NEXT:    por %xmm1, %xmm0
 ; CHECK-YONAH-NEXT:    retl
   %C = shufflevector <16 x i8> %A, <16 x i8> %B, <16 x i32> < i32 5, i32 6, i32 7, i32 undef, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20 >
   ret <16 x i8> %C

 ;
 ; CHECK-YONAH-LABEL: test9:
 ; CHECK-YONAH:       # BB#0:
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,2,1,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,3,4,5,6,7]
-; CHECK-YONAH-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,0,3]
-; CHECK-YONAH-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,1,2,4,5,6,7]
-; CHECK-YONAH-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,6,7,4]
+; CHECK-YONAH-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-YONAH-NEXT:    psrldq $2, %xmm0 # xmm0 = xmm0[2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero,zero
+; CHECK-YONAH-NEXT:    pslldq $14, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1]
+; CHECK-YONAH-NEXT:    por %xmm0, %xmm1
+; CHECK-YONAH-NEXT:    movdqa %xmm1, %xmm0
 ; CHECK-YONAH-NEXT:    retl
   %C = shufflevector <8 x i16> %B, <8 x i16> %A, <8 x i32> < i32 undef, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0 >
   ret <8 x i16> %C

test/CodeGen/X86/vector-shuffle-128-v16.ll

 define <16 x i8> @shuffle_v16i8_31_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14(<16 x i8> %a, <16 x i8> %b) {
 ; SSE2-LABEL: shuffle_v16i8_31_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[3,1,2,0]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[3,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,3,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,3,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,4,5,6]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,0,2,3,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; SSE2-NEXT:    packuswb %xmm3, %xmm0
+; SSE2-NEXT:    psrldq $15, %xmm1 {{.*#+}} xmm1 = xmm1[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $1, %xmm0 {{.*#+}} xmm0 = zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v16i8_31_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14:

 define <16 x i8> @shuffle_v16i8_15_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14(<16 x i8> %a, <16 x i8> %b) {
 ; SSE2-LABEL: shuffle_v16i8_15_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,3,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[3,1,2,0]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm3[0,3,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; SSE2-NEXT:    packuswb %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq $15, %xmm1 {{.*#+}} xmm1 = xmm1[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $1, %xmm0 {{.*#+}} xmm0 = zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v16i8_15_00_01_02_03_04_05_06_07_08_09_10_11_12_13_14:

 define <16 x i8> @shuffle_v16i8_17_18_19_20_21_22_23_24_25_26_27_28_29_30_31_00(<16 x i8> %a, <16 x i8> %b) {
 ; SSE2-LABEL: shuffle_v16i8_17_18_19_20_21_22_23_24_25_26_27_28_29_30_31_00:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,1,0,1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[3,1,2,0]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm3[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,2,3,1,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm4[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,2,0]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,4,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,6,7,7]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,6,4]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT:    packuswb %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psrldq $1, %xmm1 {{.*#+}} xmm1 = xmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero
+; SSE2-NEXT:    pslldq $15, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v16i8_17_18_19_20_21_22_23_24_25_26_27_28_29_30_31_00:

 define <16 x i8> @shuffle_v16i8_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_16(<16 x i8> %a, <16 x i8> %b) {
 ; SSE2-LABEL: shuffle_v16i8_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_16:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,0,1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[3,1,2,0]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm3[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,2,3,1,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm4[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,0]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,6,7,7]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,4]
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; SSE2-NEXT:    packuswb %xmm0, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psrldq $1, %xmm0 {{.*#+}} xmm0 = xmm0[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero
+; SSE2-NEXT:    pslldq $15, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v16i8_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_16:

 define <16 x i8> @shuffle_v16i8_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_00(<16 x i8> %a, <16 x i8> %b) {
 ; SSE2-LABEL: shuffle_v16i8_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_00:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,1,0,1]
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,3,1,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[3,1,2,0]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm1[0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,1,0,1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[3,1,2,0]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[0,2,3,1,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; SSE2-NEXT:    packuswb %xmm3, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq $1, %xmm1 {{.*#+}} xmm1 = xmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15],zero
+; SSE2-NEXT:    pslldq $15, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v16i8_01_02_03_04_05_06_07_08_09_10_11_12_13_14_15_00:

 define <16 x i8> @shuffle_v16i8_15_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30(<16 x i8> %a, <16 x i8> %b) {
 ; SSE2-LABEL: shuffle_v16i8_15_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm1[2,3,0,1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[3,1,2,0]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[3,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,3,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,3,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,4,5,6]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,0,2,3,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; SSE2-NEXT:    packuswb %xmm3, %xmm0
+; SSE2-NEXT:    psrldq $15, %xmm0 {{.*#+}} xmm0 = xmm0[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $1, %xmm1 {{.*#+}} xmm1 = zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v16i8_15_16_17_18_19_20_21_22_23_24_25_26_27_28_29_30:

test/CodeGen/X86/vector-shuffle-128-v8.ll

 define <8 x i16> @shuffle_v8i16_def01234(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_def01234:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,0,1]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,3,1,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,0]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrldq $10, %xmm1 {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $6, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_def01234:

 define <8 x i16> @shuffle_v8i16_ueuu123u(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_ueuu123u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,5,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,4,5,7]
+; SSE2-NEXT:    psrldq $10, %xmm1 {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $6, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_ueuu123u:

 define <8 x i16> @shuffle_v8i16_56701234(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_56701234:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,1,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,4,5,6]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq $10, %xmm1 {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $6, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_56701234:

 define <8 x i16> @shuffle_v8i16_u6uu123u(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_u6uu123u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,0,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,6,7,7]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq $10, %xmm1 {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $6, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_u6uu123u:

 define <8 x i16> @shuffle_v8i16_uuuu123u(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_uuuu123u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,6,7,7]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq $10, %xmm1 {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $6, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_uuuu123u:

 define <8 x i16> @shuffle_v8i16_bcdef012(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_bcdef012:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm0[3,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
+; SSE2-NEXT:    psrldq $6, %xmm1 {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $10, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5]
+; SSE2-NEXT:    por
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_bcdef012:

 define <8 x i16> @shuffle_v8i16_ucdeuu1u(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_ucdeuu1u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,7,7]
+; SSE2-NEXT:    psrldq $6, %xmm1 {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $10, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5]
+; SSE2-NEXT:    por
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_ucdeuu1u:

 define <8 x i16> @shuffle_v8i16_34567012(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_34567012:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,1,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,6,7,4]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq $6, %xmm1 {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $10, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_34567012:

 define <8 x i16> @shuffle_v8i16_u456uu1u(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_u456uu1u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,5,7]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq $6, %xmm1 {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $10, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_u456uu1u:

 define <8 x i16> @shuffle_v8i16_u456uuuu(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_u456uuuu:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,1,2,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq $6, %xmm1 {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $10, %xmm0 {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_u456uuuu:

 define <8 x i16> @shuffle_v8i16_3456789a(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_3456789a:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    psrldq $6, %xmm0 {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $10, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_3456789a:

 define <8 x i16> @shuffle_v8i16_u456uu9u(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_u456uu9u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,1,2,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,7,7]
+; SSE2-NEXT:    psrldq $6, %xmm0 {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $10, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_u456uu9u:

 define <8 x i16> @shuffle_v8i16_56789abc(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_56789abc:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,1,0,1]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,3,1,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,2,0]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,2,3,0,4,5,6,7]
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    psrldq $10, %xmm0 {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $6, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_56789abc:

 define <8 x i16> @shuffle_v8i16_u6uu9abu(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_u6uu9abu:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,6,5,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,4,5,7]
+; SSE2-NEXT:    psrldq $10, %xmm0 {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pslldq $6, %xmm1 {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9]
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_u6uu9abu: