<html>
  <head>
    <meta content="text/html; charset=utf-8" http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    Yes revert if there is a problem but please raise a bugzilla with a
    repro, even if it isn't greatly reduced.<br>
    Simon.<br>
    <br>
    <div class="moz-cite-prefix">On 14/03/2016 15:44, Alina Sbirlea
      wrote:<br>
    </div>
    <blockquote
cite="mid:CALV8tXM+Nwf0Brt_fmGPv-KYgbUTtLvPiWVT=_h+rb+84mS+9g@mail.gmail.com"
      type="cite">
      <div dir="ltr">As far as I can tell the problematic change is the
        check in <span style="font-size:12.8px">llvm/trunk/lib/Target/X86/</span><span
          style="font-size:12.8px">X86ISelLowering.cpp</span>
        <div><span style="font-size:12.8px">Working on a testcase in the
            meantime.</span></div>
        <div><span style="font-size:12.8px"><br>
          </span></div>
        <div><span style="font-size:12.8px"><br>
          </span></div>
        <div><span style="font-size:12.8px">Thank you,</span></div>
        <div><span style="font-size:12.8px">Alina</span></div>
      </div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On Mon, Mar 14, 2016 at 3:31 PM, Eric
          Christopher <span dir="ltr"><<a moz-do-not-send="true"
              href="mailto:echristo@gmail.com" target="_blank">echristo@gmail.com</a>></span>
          wrote:<br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div dir="ltr">Hi Simon,
              <div><br>
              </div>
              <div>I'm seeing asserts with this and halide on x86. I'm
                working with Alina to get a testcase for you, but would
                you mind terribly if we reverted this in the meantime?</div>
              <div><br>
              </div>
              <div>-eric</div>
              <br>
              <div class="gmail_quote">
                <div dir="ltr">On Fri, Mar 11, 2016 at 2:22 PM Simon
                  Pilgrim via llvm-commits <<a moz-do-not-send="true"
                    href="mailto:llvm-commits@lists.llvm.org"
                    target="_blank">llvm-commits@lists.llvm.org</a>>
                  wrote:<br>
                </div>
                <blockquote class="gmail_quote" style="margin:0 0 0
                  .8ex;border-left:1px #ccc solid;padding-left:1ex">Author:
                  rksimon<br>
                  Date: Fri Mar 11 16:18:05 2016<br>
                  New Revision: 263303<br>
                  <br>
                  URL: <a moz-do-not-send="true"
                    href="http://llvm.org/viewvc/llvm-project?rev=263303&view=rev"
                    rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project?rev=263303&view=rev</a><br>
                  Log:<br>
                  [X86][SSE] Simplify vector LOAD + EXTEND on pre-SSE41
                  hardware<br>
                  <br>
                  Improve vector extension of vectors on hardware
                  without dedicated VSEXT/VZEXT instructions.<br>
                  <br>
                  We already convert these to
                  SIGN_EXTEND_VECTOR_INREG/ZERO_EXTEND_VECTOR_INREG but
                  can further improve this by using the legalizer
                  instead of prematurely splitting into legal vectors in
                  the combine as this only properly helps for lowering
                  to VSEXT/VZEXT.<br>
                  <br>
                  Removes a lot of unnecessary any_extend + mask pattern
                  - (Fix for PR25718).<br>
                  <br>
                  Differential Revision: <a moz-do-not-send="true"
                    href="http://reviews.llvm.org/D17932"
                    rel="noreferrer" target="_blank">http://reviews.llvm.org/D17932</a><br>
                  <br>
                  Modified:<br>
                     
                  llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeTypes.h<br>
                     
                  llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp<br>
                      llvm/trunk/lib/Target/X86/X86ISelLowering.cpp<br>
                      llvm/trunk/test/CodeGen/X86/vector-zext.ll<br>
                  <br>
                  Modified:
                  llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeTypes.h<br>
                  URL: <a moz-do-not-send="true"
href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeTypes.h?rev=263303&r1=263302&r2=263303&view=diff"
                    rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeTypes.h?rev=263303&r1=263302&r2=263303&view=diff</a><br>
==============================================================================<br>
                  ---
                  llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeTypes.h
                  (original)<br>
                  +++
                  llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeTypes.h
                  Fri Mar 11 16:18:05 2016<br>
                  @@ -653,6 +653,7 @@ private:<br>
                     void SplitVecRes_UnaryOp(SDNode *N, SDValue
                  &Lo, SDValue &Hi);<br>
                     void SplitVecRes_ExtendOp(SDNode *N, SDValue
                  &Lo, SDValue &Hi);<br>
                     void SplitVecRes_InregOp(SDNode *N, SDValue
                  &Lo, SDValue &Hi);<br>
                  +  void SplitVecRes_ExtVecInRegOp(SDNode *N, SDValue
                  &Lo, SDValue &Hi);<br>
                  <br>
                     void SplitVecRes_BITCAST(SDNode *N, SDValue
                  &Lo, SDValue &Hi);<br>
                     void SplitVecRes_BUILD_VECTOR(SDNode *N, SDValue
                  &Lo, SDValue &Hi);<br>
                  <br>
                  Modified:
                  llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp<br>
                  URL: <a moz-do-not-send="true"
href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp?rev=263303&r1=263302&r2=263303&view=diff"
                    rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp?rev=263303&r1=263302&r2=263303&view=diff</a><br>
==============================================================================<br>
                  ---
                  llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
                  (original)<br>
                  +++
                  llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
                  Fri Mar 11 16:18:05 2016<br>
                  @@ -621,6 +621,12 @@ void
                  DAGTypeLegalizer::SplitVectorResult<br>
                     
                   SplitVecRes_VECTOR_SHUFFLE(cast<ShuffleVectorSDNode>(N),
                  Lo, Hi);<br>
                       break;<br>
                  <br>
                  +  case ISD::ANY_EXTEND_VECTOR_INREG:<br>
                  +  case ISD::SIGN_EXTEND_VECTOR_INREG:<br>
                  +  case ISD::ZERO_EXTEND_VECTOR_INREG:<br>
                  +    SplitVecRes_ExtVecInRegOp(N, Lo, Hi);<br>
                  +    break;<br>
                  +<br>
                     case ISD::BITREVERSE:<br>
                     case ISD::BSWAP:<br>
                     case ISD::CONVERT_RNDSAT:<br>
                  @@ -917,6 +923,39 @@ void
                  DAGTypeLegalizer::SplitVecRes_Inreg<br>
                                      DAG.getValueType(HiVT));<br>
                   }<br>
                  <br>
                  +void
                  DAGTypeLegalizer::SplitVecRes_ExtVecInRegOp(SDNode *N,
                  SDValue &Lo,<br>
                  +                                               
                   SDValue &Hi) {<br>
                  +  unsigned Opcode = N->getOpcode();<br>
                  +  SDValue N0 = N->getOperand(0);<br>
                  +<br>
                  +  SDLoc dl(N);<br>
                  +  SDValue InLo, InHi;<br>
                  +  GetSplitVector(N0, InLo, InHi);<br>
                  +  EVT InLoVT = InLo.getValueType();<br>
                  +  unsigned InNumElements =
                  InLoVT.getVectorNumElements();<br>
                  +<br>
                  +  EVT OutLoVT, OutHiVT;<br>
                  +  std::tie(OutLoVT, OutHiVT) =
                  DAG.GetSplitDestVTs(N->getValueType(0));<br>
                  +  unsigned OutNumElements =
                  OutLoVT.getVectorNumElements();<br>
                  +  assert((2 * OutNumElements) <= InNumElements
                  &&<br>
                  +         "Illegal extend vector in reg split");<br>
                  +<br>
                  +  // *_EXTEND_VECTOR_INREG instructions extend the
                  lowest elements of the<br>
                  +  // input vector (i.e. we only use InLo):<br>
                  +  // OutLo will extend the first OutNumElements from
                  InLo.<br>
                  +  // OutHi will extend the next OutNumElements from
                  InLo.<br>
                  +<br>
                  +  // Shuffle the elements from InLo for OutHi into
                  the bottom elements to<br>
                  +  // create a 'fake' InHi.<br>
                  +  SmallVector<int, 8> SplitHi(InNumElements,
                  -1);<br>
                  +  for (unsigned i = 0; i != OutNumElements; ++i)<br>
                  +    SplitHi[i] = i + OutNumElements;<br>
                  +  InHi = DAG.getVectorShuffle(InLoVT, dl, InLo,
                  DAG.getUNDEF(InLoVT), SplitHi);<br>
                  +<br>
                  +  Lo = DAG.getNode(Opcode, dl, OutLoVT, InLo);<br>
                  +  Hi = DAG.getNode(Opcode, dl, OutHiVT, InHi);<br>
                  +}<br>
                  +<br>
                   void
                  DAGTypeLegalizer::SplitVecRes_INSERT_VECTOR_ELT(SDNode
                  *N, SDValue &Lo,<br>
                                                                       
                  SDValue &Hi) {<br>
                     SDValue Vec = N->getOperand(0);<br>
                  <br>
                  Modified:
                  llvm/trunk/lib/Target/X86/X86ISelLowering.cpp<br>
                  URL: <a moz-do-not-send="true"
href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=263303&r1=263302&r2=263303&view=diff"
                    rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=263303&r1=263302&r2=263303&view=diff</a><br>
==============================================================================<br>
                  --- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
                  (original)<br>
                  +++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Fri
                  Mar 11 16:18:05 2016<br>
                  @@ -28534,7 +28534,9 @@ static SDValue
                  combineToExtendVectorInRe<br>
                  <br>
                     // If target-size is 128-bits (or 256-bits on AVX2
                  target), then convert to<br>
                     // ISD::*_EXTEND_VECTOR_INREG which ensures
                  lowering to X86ISD::V*EXT.<br>
                  -  if (VT.is128BitVector() || (VT.is256BitVector()
                  && Subtarget.hasInt256())) {<br>
                  +  // Also use this if we don't have SSE41 to allow
                  the legalizer do its job.<br>
                  +  if (!Subtarget.hasSSE41() || VT.is128BitVector() ||<br>
                  +      (VT.is256BitVector() &&
                  Subtarget.hasInt256())) {<br>
                       SDValue ExOp = ExtendVecSize(DL, N0,
                  VT.getSizeInBits());<br>
                       return Opcode == ISD::SIGN_EXTEND<br>
                                  ? DAG.getSignExtendVectorInReg(ExOp,
                  DL, VT)<br>
                  <br>
                  Modified: llvm/trunk/test/CodeGen/X86/vector-zext.ll<br>
                  URL: <a moz-do-not-send="true"
href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-zext.ll?rev=263303&r1=263302&r2=263303&view=diff"
                    rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-zext.ll?rev=263303&r1=263302&r2=263303&view=diff</a><br>
==============================================================================<br>
                  --- llvm/trunk/test/CodeGen/X86/vector-zext.ll
                  (original)<br>
                  +++ llvm/trunk/test/CodeGen/X86/vector-zext.ll Fri Mar
                  11 16:18:05 2016<br>
                  @@ -544,23 +544,20 @@ define <4 x i64>
                  @load_zext_4i8_to_4i64(<br>
                   ; SSE2-LABEL: load_zext_4i8_to_4i64:<br>
                   ; SSE2:       # BB#0: # %entry<br>
                   ; SSE2-NEXT:    movd {{.*#+}} xmm1 =
                  mem[0],zero,zero,zero<br>
                  -; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 =
xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]<br>
                  -; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 =
                  xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]<br>
                  -; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,1,3]<br>
                  -; SSE2-NEXT:    movdqa {{.*#+}} xmm2 =
                  [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]<br>
                  -; SSE2-NEXT:    pand %xmm2, %xmm0<br>
                  -; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]<br>
                  -; SSE2-NEXT:    pand %xmm2, %xmm1<br>
                  +; SSE2-NEXT:    pxor %xmm2, %xmm2<br>
                  +; 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]<br>
                  +; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 =
                  xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]<br>
                  +; SSE2-NEXT:    movdqa %xmm1, %xmm0<br>
                  +; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 =
                  xmm0[0],xmm2[0],xmm0[1],xmm2[1]<br>
                  +; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 =
                  xmm1[2],xmm2[2],xmm1[3],xmm2[3]<br>
                   ; SSE2-NEXT:    retq<br>
                   ;<br>
                   ; SSSE3-LABEL: load_zext_4i8_to_4i64:<br>
                   ; SSSE3:       # BB#0: # %entry<br>
                   ; SSSE3-NEXT:    movd {{.*#+}} xmm1 =
                  mem[0],zero,zero,zero<br>
                  -; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 =
xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]<br>
                  -; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 =
                  xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]<br>
                   ; SSSE3-NEXT:    movdqa %xmm1, %xmm0<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 =
xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[4],zero,zero,zero,zero,zero,zero,zero<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 =
xmm1[8],zero,zero,zero,zero,zero,zero,zero,xmm1[12],zero,zero,zero,zero,zero,zero,zero<br>
                  +; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 =
xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero<br>
                  +; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 =
xmm1[2],zero,zero,zero,zero,zero,zero,zero,xmm1[3],zero,zero,zero,zero,zero,zero,zero<br>
                   ; SSSE3-NEXT:    retq<br>
                   ;<br>
                   ; SSE41-LABEL: load_zext_4i8_to_4i64:<br>
                  @@ -625,22 +622,21 @@ define <8 x i32>
                  @load_zext_8i8_to_8i32(<br>
                   ; SSE2-LABEL: load_zext_8i8_to_8i32:<br>
                   ; SSE2:       # BB#0: # %entry<br>
                   ; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero<br>
                  -; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 =
xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]<br>
                  +; SSE2-NEXT:    pxor %xmm2, %xmm2<br>
                  +; 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]<br>
                   ; SSE2-NEXT:    movdqa %xmm1, %xmm0<br>
                  -; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 =
                  xmm0[0,0,1,1,2,2,3,3]<br>
                  -; SSE2-NEXT:    movdqa {{.*#+}} xmm2 =
                  [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]<br>
                  -; SSE2-NEXT:    pand %xmm2, %xmm0<br>
                  -; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 =
                  xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]<br>
                  -; SSE2-NEXT:    pand %xmm2, %xmm1<br>
                  +; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 =
                  xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]<br>
                  +; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 =
                  xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]<br>
                   ; SSE2-NEXT:    retq<br>
                   ;<br>
                   ; SSSE3-LABEL: load_zext_8i8_to_8i32:<br>
                   ; SSSE3:       # BB#0: # %entry<br>
                   ; SSSE3-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero<br>
                  -; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 =
xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]<br>
                  +; SSSE3-NEXT:    pxor %xmm2, %xmm2<br>
                  +; SSSE3-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]<br>
                   ; SSSE3-NEXT:    movdqa %xmm1, %xmm0<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 =
xmm0[0],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[6],zero,zero,zero<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 =
xmm1[8],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[14],zero,zero,zero<br>
                  +; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 =
                  xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]<br>
                  +; SSSE3-NEXT:    punpckhwd {{.*#+}} xmm1 =
                  xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]<br>
                   ; SSSE3-NEXT:    retq<br>
                   ;<br>
                   ; SSE41-LABEL: load_zext_8i8_to_8i32:<br>
                  @@ -674,34 +670,33 @@ entry:<br>
                   define <8 x i64> @load_zext_8i8_to_8i64(<8 x
                  i8> *%ptr) {<br>
                   ; SSE2-LABEL: load_zext_8i8_to_8i64:<br>
                   ; SSE2:       # BB#0: # %entry<br>
                  -; SSE2-NEXT:    movq {{.*#+}} xmm3 = mem[0],zero<br>
                  -; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 =
xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]<br>
                  -; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,1,0,3]<br>
                  -; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 =
                  xmm0[0,1,2,3,5,5,6,7]<br>
                  -; SSE2-NEXT:    movdqa {{.*#+}} xmm4 =
                  [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]<br>
                  -; SSE2-NEXT:    pand %xmm4, %xmm0<br>
                  -; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,1,1,3]<br>
                  -; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 =
                  xmm1[0,1,2,3,5,5,6,7]<br>
                  -; SSE2-NEXT:    pand %xmm4, %xmm1<br>
                  -; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,1,2,3]<br>
                  -; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 =
                  xmm2[0,1,2,3,5,5,6,7]<br>
                  -; SSE2-NEXT:    pand %xmm4, %xmm2<br>
                  -; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[3,1,3,3]<br>
                  -; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 =
                  xmm3[0,1,2,3,5,5,6,7]<br>
                  -; SSE2-NEXT:    pand %xmm4, %xmm3<br>
                  +; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero<br>
                  +; SSE2-NEXT:    pxor %xmm4, %xmm4<br>
                  +; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]<br>
                  +; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 =
xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]<br>
                  +; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 =
                  xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]<br>
                  +; SSE2-NEXT:    movdqa %xmm1, %xmm0<br>
                  +; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 =
                  xmm0[0],xmm4[0],xmm0[1],xmm4[1]<br>
                  +; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 =
                  xmm1[2],xmm4[2],xmm1[3],xmm4[3]<br>
                  +; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 =
xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]<br>
                  +; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 =
                  xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]<br>
                  +; SSE2-NEXT:    movdqa %xmm3, %xmm2<br>
                  +; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 =
                  xmm2[0],xmm4[0],xmm2[1],xmm4[1]<br>
                  +; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 =
                  xmm3[2],xmm4[2],xmm3[3],xmm4[3]<br>
                   ; SSE2-NEXT:    retq<br>
                   ;<br>
                   ; SSSE3-LABEL: load_zext_8i8_to_8i64:<br>
                   ; SSSE3:       # BB#0: # %entry<br>
                  -; SSSE3-NEXT:    movq {{.*#+}} xmm3 = mem[0],zero<br>
                  -; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 =
xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]<br>
                  -; SSSE3-NEXT:    movdqa %xmm3, %xmm0<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 =
xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero<br>
                  -; SSSE3-NEXT:    movdqa %xmm3, %xmm1<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 =
xmm1[4],zero,zero,zero,zero,zero,zero,zero,xmm1[6],zero,zero,zero,zero,zero,zero,zero<br>
                  +; SSSE3-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero<br>
                  +; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 =
                  [0,128,128,128,128,128,128,128,1,128,128,128,128,128,128,128]<br>
                  +; SSSE3-NEXT:    movdqa %xmm1, %xmm0<br>
                  +; SSSE3-NEXT:    pshufb %xmm4, %xmm0<br>
                  +; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 =
                  [2,128,128,128,128,128,128,128,3,128,128,128,128,128,128,128]<br>
                  +; SSSE3-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]<br>
                  +; SSSE3-NEXT:    pshufb %xmm5, %xmm1<br>
                   ; SSSE3-NEXT:    movdqa %xmm3, %xmm2<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm2 =
xmm2[8],zero,zero,zero,zero,zero,zero,zero,xmm2[10],zero,zero,zero,zero,zero,zero,zero<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm3 =
xmm3[12],zero,zero,zero,zero,zero,zero,zero,xmm3[14],zero,zero,zero,zero,zero,zero,zero<br>
                  +; SSSE3-NEXT:    pshufb %xmm4, %xmm2<br>
                  +; SSSE3-NEXT:    pshufb %xmm5, %xmm3<br>
                   ; SSSE3-NEXT:    retq<br>
                   ;<br>
                   ; SSE41-LABEL: load_zext_8i8_to_8i64:<br>
                  @@ -851,21 +846,21 @@ define <4 x i64>
                  @load_zext_4i16_to_4i64<br>
                   ; SSE2-LABEL: load_zext_4i16_to_4i64:<br>
                   ; SSE2:       # BB#0: # %entry<br>
                   ; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero<br>
                  -; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 =
                  xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]<br>
                  -; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,1,3]<br>
                  -; SSE2-NEXT:    movdqa {{.*#+}} xmm2 =
                  [65535,0,0,0,65535,0,0,0]<br>
                  -; SSE2-NEXT:    pand %xmm2, %xmm0<br>
                  -; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]<br>
                  -; SSE2-NEXT:    pand %xmm2, %xmm1<br>
                  +; SSE2-NEXT:    pxor %xmm2, %xmm2<br>
                  +; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 =
                  xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]<br>
                  +; SSE2-NEXT:    movdqa %xmm1, %xmm0<br>
                  +; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 =
                  xmm0[0],xmm2[0],xmm0[1],xmm2[1]<br>
                  +; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 =
                  xmm1[2],xmm2[2],xmm1[3],xmm2[3]<br>
                   ; SSE2-NEXT:    retq<br>
                   ;<br>
                   ; SSSE3-LABEL: load_zext_4i16_to_4i64:<br>
                   ; SSSE3:       # BB#0: # %entry<br>
                   ; SSSE3-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero<br>
                  -; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 =
                  xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]<br>
                  +; SSSE3-NEXT:    pxor %xmm2, %xmm2<br>
                  +; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 =
                  xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]<br>
                   ; SSSE3-NEXT:    movdqa %xmm1, %xmm0<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 =
xmm0[0,1],zero,zero,zero,zero,zero,zero,xmm0[4,5],zero,zero,zero,zero,zero,zero<br>
                  -; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 =
xmm1[8,9],zero,zero,zero,zero,zero,zero,xmm1[12,13],zero,zero,zero,zero,zero,zero<br>
                  +; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 =
                  xmm0[0],xmm2[0],xmm0[1],xmm2[1]<br>
                  +; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm1 =
                  xmm1[2],xmm2[2],xmm1[3],xmm2[3]<br>
                   ; SSSE3-NEXT:    retq<br>
                   ;<br>
                   ; SSE41-LABEL: load_zext_4i16_to_4i64:<br>
                  <br>
                  <br>
                  _______________________________________________<br>
                  llvm-commits mailing list<br>
                  <a moz-do-not-send="true"
                    href="mailto:llvm-commits@lists.llvm.org"
                    target="_blank">llvm-commits@lists.llvm.org</a><br>
                  <a moz-do-not-send="true"
                    href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits"
                    rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>
                </blockquote>
              </div>
            </div>
          </blockquote>
        </div>
        <br>
      </div>
    </blockquote>
    <br>
  </body>
</html>