[llvm] r366799 - [TargetLowering] Add SimplifyMultipleUseDemandedBits

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Sat Jul 27 09:44:03 PDT 2019


Should be fixed at rL367171 - please can you confirm?

On 27/07/2019 01:54, Jordan Rupprecht wrote:
> We're also seeing some tests failures (possibly broken by a different 
> commit) that use the llvm api and are now timing out, and when we kill 
> them we see this stack trace:
>
>     @     0x55e2fdc55b13        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55ad2        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc54c35         96 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc574c5        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc54c35         96 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc560d2        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55ad2        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc563fa        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc56df8        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55d65        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc5643d        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55ed1        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc56187        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55d21        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc5643d        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55a8a        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55ad2        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55ad2        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55ad2        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55ad2        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55a8a        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55a8a        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc563fa        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55ed1        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc56187        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55d21        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc5643d        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc55a8a        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @     0x55e2fdc563fa        832 
>  llvm::SelectionDAG::computeKnownBits()
>     @ ... and at least 221 more frames
>
> So... something related to the depth checking is not working correctly
>
> On Fri, Jul 26, 2019 at 3:53 PM Jordan Rupprecht <rupprecht at google.com 
> <mailto:rupprecht at google.com>> wrote:
>
>     Sorry for the late follow up, it looks like this might be causing
>     some compiler timeouts when processing some files with msan & -O2
>     (& maybe other opts) on this file:
>     https://boringssl.googlesource.com/boringssl/+/2623/crypto/ec/p224-64.c
>
>
>     I'm making some progress on a reduced test case now
>
>     On Tue, Jul 23, 2019 at 5:38 AM Simon Pilgrim via llvm-commits
>     <llvm-commits at lists.llvm.org <mailto:llvm-commits at lists.llvm.org>>
>     wrote:
>
>         Author: rksimon
>         Date: Tue Jul 23 05:39:08 2019
>         New Revision: 366799
>
>         URL: http://llvm.org/viewvc/llvm-project?rev=366799&view=rev
>         Log:
>         [TargetLowering] Add SimplifyMultipleUseDemandedBits
>
>         This patch introduces the DAG version of
>         SimplifyMultipleUseDemandedBits, which attempts to peek
>         through ops (mainly and/or/xor so far) that don't contribute
>         to the demandedbits/elts of a node - which means we can do
>         this even in cases where we have multiple uses of an op, which
>         normally requires us to demanded all bits/elts. The intention
>         is to remove a similar instruction -
>         SelectionDAG::GetDemandedBits - once
>         SimplifyMultipleUseDemandedBits has matured.
>
>         The InstCombine version of SimplifyMultipleUseDemandedBits can
>         constant fold which I haven't added here yet, and so far I've
>         only wired this up to some basic binops
>         (and/or/xor/add/sub/mul) to demonstrate its use.
>
>         We do see a couple of regressions that need to be addressed:
>
>             AMDGPU unsigned dot product codegen retains an AND mask
>         (for ZERO_EXTEND) that it previously removed (but otherwise
>         the dotproduct codegen is a lot better).
>
>             X86/AVX2 has poor handling of vector
>         ANY_EXTEND/ANY_EXTEND_VECTOR_INREG - it prematurely gets
>         converted to ZERO_EXTEND_VECTOR_INREG.
>
>         The code owners have confirmed its ok for these cases to fixed
>         up in future patches.
>
>         Differential Revision: https://reviews.llvm.org/D63281
>
>         Modified:
>             llvm/trunk/include/llvm/CodeGen/TargetLowering.h
>             llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp
>             llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp
>             llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll
>             llvm/trunk/test/CodeGen/AMDGPU/idot4s.ll
>             llvm/trunk/test/CodeGen/AMDGPU/idot4u.ll
>             llvm/trunk/test/CodeGen/AMDGPU/idot8s.ll
>             llvm/trunk/test/CodeGen/AMDGPU/idot8u.ll
>             llvm/trunk/test/CodeGen/AMDGPU/sdiv.ll
>         llvm/trunk/test/CodeGen/SystemZ/store_nonbytesized_vecs.ll
>             llvm/trunk/test/CodeGen/X86/2012-08-07-CmpISelBug.ll
>             llvm/trunk/test/CodeGen/X86/vector-fshl-128.ll
>             llvm/trunk/test/CodeGen/X86/vector-reduce-mul-widen.ll
>             llvm/trunk/test/CodeGen/X86/vector-reduce-mul.ll
>
>         Modified: llvm/trunk/include/llvm/CodeGen/TargetLowering.h
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/CodeGen/TargetLowering.h?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/include/llvm/CodeGen/TargetLowering.h (original)
>         +++ llvm/trunk/include/llvm/CodeGen/TargetLowering.h Tue Jul
>         23 05:39:08 2019
>         @@ -3065,6 +3065,14 @@ public:
>            bool SimplifyDemandedBits(SDValue Op, const APInt
>         &DemandedMask,
>                                      DAGCombinerInfo &DCI) const;
>
>         +  /// More limited version of SimplifyDemandedBits that can
>         be used to "look
>         +  /// through" ops that don't contribute to the
>         DemandedBits/DemandedElts -
>         +  /// bitwise ops etc.
>         +  SDValue SimplifyMultipleUseDemandedBits(SDValue Op, const
>         APInt &DemandedBits,
>         +                                          const APInt
>         &DemandedElts,
>         +                                          SelectionDAG &DAG,
>         +                                          unsigned Depth) const;
>         +
>            /// Look at Vector Op. At this point, we know that only the
>         DemandedElts
>            /// elements of the result of Op are ever used downstream. 
>         If we can use
>            /// this information to simplify Op, create a new
>         simplified DAG node and
>         @@ -3139,6 +3147,13 @@ public:
>         TargetLoweringOpt &TLO,
>                                                           unsigned
>         Depth = 0) const;
>
>         +  /// More limited version of SimplifyDemandedBits that can
>         be used to "look
>         +  /// through" ops that don't contribute to the
>         DemandedBits/DemandedElts -
>         +  /// bitwise ops etc.
>         +  virtual SDValue SimplifyMultipleUseDemandedBitsForTargetNode(
>         +      SDValue Op, const APInt &DemandedBits, const APInt
>         &DemandedElts,
>         +      SelectionDAG &DAG, unsigned Depth) const;
>         +
>            /// This method returns the constant pool value that will
>         be loaded by LD.
>            /// NOTE: You must check for implicit extensions of the
>         constant by LD.
>            virtual const Constant
>         *getTargetConstantFromLoad(LoadSDNode *LD) const;
>
>         Modified: llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp
>         (original)
>         +++ llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp Tue
>         Jul 23 05:39:08 2019
>         @@ -564,6 +564,61 @@ bool TargetLowering::SimplifyDemandedBit
>                                        AssumeSingleUse);
>          }
>
>         +// TODO: Can we merge SelectionDAG::GetDemandedBits into this?
>         +// TODO: Under what circumstances can we create nodes?
>         BITCAST? Constant?
>         +SDValue TargetLowering::SimplifyMultipleUseDemandedBits(
>         +    SDValue Op, const APInt &DemandedBits, const APInt
>         &DemandedElts,
>         +    SelectionDAG &DAG, unsigned Depth) const {
>         +  KnownBits LHSKnown, RHSKnown;
>         +  switch (Op.getOpcode()) {
>         +  case ISD::AND: {
>         +    LHSKnown = DAG.computeKnownBits(Op.getOperand(0),
>         DemandedElts, Depth + 1);
>         +    RHSKnown = DAG.computeKnownBits(Op.getOperand(1),
>         DemandedElts, Depth + 1);
>         +
>         +    // If all of the demanded bits are known 1 on one side,
>         return the other.
>         +    // These bits cannot contribute to the result of the
>         'and' in this
>         +    // context.
>         +    if (DemandedBits.isSubsetOf(LHSKnown.Zero | RHSKnown.One))
>         +      return Op.getOperand(0);
>         +    if (DemandedBits.isSubsetOf(RHSKnown.Zero | LHSKnown.One))
>         +      return Op.getOperand(1);
>         +    break;
>         +  }
>         +  case ISD::OR: {
>         +    LHSKnown = DAG.computeKnownBits(Op.getOperand(0),
>         DemandedElts, Depth + 1);
>         +    RHSKnown = DAG.computeKnownBits(Op.getOperand(1),
>         DemandedElts, Depth + 1);
>         +
>         +    // If all of the demanded bits are known zero on one
>         side, return the
>         +    // other.  These bits cannot contribute to the result of
>         the 'or' in this
>         +    // context.
>         +    if (DemandedBits.isSubsetOf(LHSKnown.One | RHSKnown.Zero))
>         +      return Op.getOperand(0);
>         +    if (DemandedBits.isSubsetOf(RHSKnown.One | LHSKnown.Zero))
>         +      return Op.getOperand(1);
>         +    break;
>         +  }
>         +  case ISD::XOR: {
>         +    LHSKnown = DAG.computeKnownBits(Op.getOperand(0),
>         DemandedElts, Depth + 1);
>         +    RHSKnown = DAG.computeKnownBits(Op.getOperand(1),
>         DemandedElts, Depth + 1);
>         +
>         +    // If all of the demanded bits are known zero on one
>         side, return the
>         +    // other.
>         +    if (DemandedBits.isSubsetOf(RHSKnown.Zero))
>         +      return Op.getOperand(0);
>         +    if (DemandedBits.isSubsetOf(LHSKnown.Zero))
>         +      return Op.getOperand(1);
>         +    break;
>         +  }
>         +  default:
>         +    if (Op.getOpcode() >= ISD::BUILTIN_OP_END)
>         +      if (SDValue V =
>         SimplifyMultipleUseDemandedBitsForTargetNode(
>         +              Op, DemandedBits, DemandedElts, DAG, Depth))
>         +        return V;
>         +    break;
>         +  }
>         +  return SDValue();
>         +}
>         +
>          /// Look at Op. At this point, we know that only the
>         OriginalDemandedBits of the
>          /// result of Op are ever used downstream. If we can use this
>         information to
>          /// simplify Op, create a new simplified DAG node and return
>         true, returning the
>         @@ -834,6 +889,20 @@ bool TargetLowering::SimplifyDemandedBit
>                return true;
>              assert(!Known2.hasConflict() && "Bits known to be one AND
>         zero?");
>
>         +    // Attempt to avoid multi-use ops if we don't need
>         anything from them.
>         +    if (!DemandedBits.isAllOnesValue() ||
>         !DemandedElts.isAllOnesValue()) {
>         +      SDValue DemandedOp0 = SimplifyMultipleUseDemandedBits(
>         +          Op0, DemandedBits, DemandedElts, TLO.DAG, Depth + 1);
>         +      SDValue DemandedOp1 = SimplifyMultipleUseDemandedBits(
>         +          Op1, DemandedBits, DemandedElts, TLO.DAG, Depth + 1);
>         +      if (DemandedOp0 || DemandedOp1) {
>         +        Op0 = DemandedOp0 ? DemandedOp0 : Op0;
>         +        Op1 = DemandedOp1 ? DemandedOp1 : Op1;
>         +        SDValue NewOp = TLO.DAG.getNode(Op.getOpcode(), dl,
>         VT, Op0, Op1);
>         +        return TLO.CombineTo(Op, NewOp);
>         +      }
>         +    }
>         +
>              // If all of the demanded bits are known one on one side,
>         return the other.
>              // These bits cannot contribute to the result of the 'and'.
>              if (DemandedBits.isSubsetOf(Known2.Zero | Known.One))
>         @@ -869,6 +938,20 @@ bool TargetLowering::SimplifyDemandedBit
>                return true;
>              assert(!Known2.hasConflict() && "Bits known to be one AND
>         zero?");
>
>         +    // Attempt to avoid multi-use ops if we don't need
>         anything from them.
>         +    if (!DemandedBits.isAllOnesValue() ||
>         !DemandedElts.isAllOnesValue()) {
>         +      SDValue DemandedOp0 = SimplifyMultipleUseDemandedBits(
>         +          Op0, DemandedBits, DemandedElts, TLO.DAG, Depth + 1);
>         +      SDValue DemandedOp1 = SimplifyMultipleUseDemandedBits(
>         +          Op1, DemandedBits, DemandedElts, TLO.DAG, Depth + 1);
>         +      if (DemandedOp0 || DemandedOp1) {
>         +        Op0 = DemandedOp0 ? DemandedOp0 : Op0;
>         +        Op1 = DemandedOp1 ? DemandedOp1 : Op1;
>         +        SDValue NewOp = TLO.DAG.getNode(Op.getOpcode(), dl,
>         VT, Op0, Op1);
>         +        return TLO.CombineTo(Op, NewOp);
>         +      }
>         +    }
>         +
>              // If all of the demanded bits are known zero on one
>         side, return the other.
>              // These bits cannot contribute to the result of the 'or'.
>              if (DemandedBits.isSubsetOf(Known2.One | Known.Zero))
>         @@ -901,6 +984,20 @@ bool TargetLowering::SimplifyDemandedBit
>                return true;
>              assert(!Known2.hasConflict() && "Bits known to be one AND
>         zero?");
>
>         +    // Attempt to avoid multi-use ops if we don't need
>         anything from them.
>         +    if (!DemandedBits.isAllOnesValue() ||
>         !DemandedElts.isAllOnesValue()) {
>         +      SDValue DemandedOp0 = SimplifyMultipleUseDemandedBits(
>         +          Op0, DemandedBits, DemandedElts, TLO.DAG, Depth + 1);
>         +      SDValue DemandedOp1 = SimplifyMultipleUseDemandedBits(
>         +          Op1, DemandedBits, DemandedElts, TLO.DAG, Depth + 1);
>         +      if (DemandedOp0 || DemandedOp1) {
>         +        Op0 = DemandedOp0 ? DemandedOp0 : Op0;
>         +        Op1 = DemandedOp1 ? DemandedOp1 : Op1;
>         +        SDValue NewOp = TLO.DAG.getNode(Op.getOpcode(), dl,
>         VT, Op0, Op1);
>         +        return TLO.CombineTo(Op, NewOp);
>         +      }
>         +    }
>         +
>              // If all of the demanded bits are known zero on one
>         side, return the other.
>              // These bits cannot contribute to the result of the 'xor'.
>              if (DemandedBits.isSubsetOf(Known.Zero))
>         @@ -1663,6 +1760,7 @@ bool TargetLowering::SimplifyDemandedBit
>              // Add, Sub, and Mul don't demand any bits in positions
>         beyond that
>              // of the highest bit demanded of them.
>              SDValue Op0 = Op.getOperand(0), Op1 = Op.getOperand(1);
>         +    SDNodeFlags Flags = Op.getNode()->getFlags();
>              unsigned DemandedBitsLZ = DemandedBits.countLeadingZeros();
>              APInt LoMask = APInt::getLowBitsSet(BitWidth, BitWidth -
>         DemandedBitsLZ);
>              if (SimplifyDemandedBits(Op0, LoMask, DemandedElts,
>         Known2, TLO,
>         @@ -1671,7 +1769,6 @@ bool TargetLowering::SimplifyDemandedBit
>                                       Depth + 1) ||
>                  // See if the operation should be performed at a
>         smaller bit width.
>                  ShrinkDemandedOp(Op, BitWidth, DemandedBits, TLO)) {
>         -      SDNodeFlags Flags = Op.getNode()->getFlags();
>                if (Flags.hasNoSignedWrap() || Flags.hasNoUnsignedWrap()) {
>                  // Disable the nsw and nuw flags. We can no longer
>         guarantee that we
>                  // won't wrap after simplification.
>         @@ -1684,6 +1781,23 @@ bool TargetLowering::SimplifyDemandedBit
>                return true;
>              }
>
>         +    // Attempt to avoid multi-use ops if we don't need
>         anything from them.
>         +    if (!LoMask.isAllOnesValue() ||
>         !DemandedElts.isAllOnesValue()) {
>         +      SDValue DemandedOp0 = SimplifyMultipleUseDemandedBits(
>         +          Op0, LoMask, DemandedElts, TLO.DAG, Depth + 1);
>         +      SDValue DemandedOp1 = SimplifyMultipleUseDemandedBits(
>         +          Op1, LoMask, DemandedElts, TLO.DAG, Depth + 1);
>         +      if (DemandedOp0 || DemandedOp1) {
>         +        Flags.setNoSignedWrap(false);
>         +        Flags.setNoUnsignedWrap(false);
>         +        Op0 = DemandedOp0 ? DemandedOp0 : Op0;
>         +        Op1 = DemandedOp1 ? DemandedOp1 : Op1;
>         +        SDValue NewOp =
>         +            TLO.DAG.getNode(Op.getOpcode(), dl, VT, Op0, Op1,
>         Flags);
>         +        return TLO.CombineTo(Op, NewOp);
>         +      }
>         +    }
>         +
>              // If we have a constant operand, we may be able to turn
>         it into -1 if we
>              // do not demand the high bits. This can make the
>         constant smaller to
>              // encode, allow more general folding, or match
>         specialized instruction
>         @@ -2357,6 +2471,19 @@ bool TargetLowering::SimplifyDemandedBit
>            return false;
>          }
>
>         +SDValue
>         TargetLowering::SimplifyMultipleUseDemandedBitsForTargetNode(
>         +    SDValue Op, const APInt &DemandedBits, const APInt
>         &DemandedElts,
>         +    SelectionDAG &DAG, unsigned Depth) const {
>         +  assert(
>         +      (Op.getOpcode() >= ISD::BUILTIN_OP_END ||
>         +       Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN ||
>         +       Op.getOpcode() == ISD::INTRINSIC_W_CHAIN ||
>         +       Op.getOpcode() == ISD::INTRINSIC_VOID) &&
>         +      "Should use SimplifyMultipleUseDemandedBits if you
>         don't know whether Op"
>         +      " is a target node!");
>         +  return SDValue();
>         +}
>         +
>          const Constant
>         *TargetLowering::getTargetConstantFromLoad(LoadSDNode*) const {
>            return nullptr;
>          }
>
>         Modified: llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp
>         (original)
>         +++ llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp Tue
>         Jul 23 05:39:08 2019
>         @@ -4308,6 +4308,7 @@ bool BoUpSLP::BlockScheduling::trySchedu
>              resetSchedule();
>              initialFillReadyList(ReadyInsts);
>            }
>         +  assert(Bundle && "Failed to find schedule bundle");
>
>            LLVM_DEBUG(dbgs() << "SLP: try schedule bundle " << *Bundle
>         << " in block "
>                              << BB->getName() << "\n");
>
>         Modified: llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll (original)
>         +++ llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll Tue Jul
>         23 05:39:08 2019
>         @@ -265,8 +265,7 @@ define void @test_32bit_opnd1_better(i32
>          define i32 @test_nouseful_bits(i8 %a, i32 %b) {
>          ; CHECK-LABEL: test_nouseful_bits:
>          ; CHECK:       // %bb.0:
>         -; CHECK-NEXT:    mov w8, w0
>         -; CHECK-NEXT:    bfi w8, w8, #8, #24
>         +; CHECK-NEXT:    orr w8, w0, w8, lsl #8
>          ; CHECK-NEXT:    mov w9, w0
>          ; CHECK-NEXT:    bfi w9, w8, #8, #24
>          ; CHECK-NEXT:    bfi w0, w9, #8, #24
>
>         Modified: llvm/trunk/test/CodeGen/AMDGPU/idot4s.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/idot4s.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/AMDGPU/idot4s.ll (original)
>         +++ llvm/trunk/test/CodeGen/AMDGPU/idot4s.ll Tue Jul 23
>         05:39:08 2019
>         @@ -899,41 +899,28 @@ define amdgpu_kernel void @idot4_acc16_v
>          ; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
>          ; GFX7-NEXT:    s_mov_b32 s3, 0xf000
>          ; GFX7-NEXT:    s_mov_b32 s2, -1
>         -; GFX7-NEXT:    s_mov_b32 s8, 0xffff
>          ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
>          ; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
>          ; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
>          ; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
>          ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX7-NEXT:    s_sext_i32_i8 s6, s4
>         -; GFX7-NEXT:    s_bfe_i32 s7, s4, 0x80008
>         -; GFX7-NEXT:    s_sext_i32_i8 s10, s5
>         +; GFX7-NEXT:    s_ashr_i32 s6, s4, 24
>         +; GFX7-NEXT:    s_bfe_i32 s7, s4, 0x80010
>         +; GFX7-NEXT:    s_bfe_i32 s10, s5, 0x80010
>          ; GFX7-NEXT:    s_bfe_i32 s11, s5, 0x80008
>         -; GFX7-NEXT:    s_bfe_i32 s12, s5, 0x80010
>         -; GFX7-NEXT:    s_ashr_i32 s5, s5, 24
>         -; GFX7-NEXT:    v_mov_b32_e32 v3, s11
>         -; GFX7-NEXT:    v_mov_b32_e32 v4, s10
>         -; GFX7-NEXT:    s_bfe_i32 s9, s4, 0x80010
>         -; GFX7-NEXT:    v_mov_b32_e32 v2, s12
>         -; GFX7-NEXT:    s_ashr_i32 s4, s4, 24
>         +; GFX7-NEXT:    s_ashr_i32 s9, s5, 24
>         +; GFX7-NEXT:    s_sext_i32_i8 s5, s5
>         +; GFX7-NEXT:    s_bfe_i32 s8, s4, 0x80008
>         +; GFX7-NEXT:    s_sext_i32_i8 s4, s4
>          ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v1, s4, v1
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v2, s9, v2
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v3, s7, v3
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v4, s6, v4
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
>         -; GFX7-NEXT:    v_and_b32_e32 v2, s8, v2
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
>         -; GFX7-NEXT:    v_and_b32_e32 v4, s8, v4
>         -; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
>         -; GFX7-NEXT:    v_or_b32_e32 v2, v4, v3
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
>         +; GFX7-NEXT:    v_mov_b32_e32 v2, s11
>         +; GFX7-NEXT:    v_mov_b32_e32 v3, s10
>          ; GFX7-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s4, v1, v0
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s8, v2, v0
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s7, v3, v0
>         +; GFX7-NEXT:    v_mov_b32_e32 v1, s9
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s6, v1, v0
>          ; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
>          ; GFX7-NEXT:    s_endpgm
>          ;
>
>         Modified: llvm/trunk/test/CodeGen/AMDGPU/idot4u.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/idot4u.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/AMDGPU/idot4u.ll (original)
>         +++ llvm/trunk/test/CodeGen/AMDGPU/idot4u.ll Tue Jul 23
>         05:39:08 2019
>         @@ -1802,33 +1802,23 @@ define amdgpu_kernel void @udot4_acc16_v
>          ; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
>          ; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
>          ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX7-NEXT:    s_and_b32 s11, s4, s8
>         -; GFX7-NEXT:    s_bfe_u32 s6, s4, 0x80008
>         -; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
>         -; GFX7-NEXT:    s_lshr_b32 s10, s5, 24
>         -; GFX7-NEXT:    s_and_b32 s8, s5, s8
>         -; GFX7-NEXT:    v_mov_b32_e32 v4, s9
>         -; GFX7-NEXT:    s_lshr_b32 s7, s4, 24
>         -; GFX7-NEXT:    v_mov_b32_e32 v2, s10
>         -; GFX7-NEXT:    s_bfe_u32 s5, s5, 0x80010
>         -; GFX7-NEXT:    v_mov_b32_e32 v3, s8
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s7, v2
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v4, s6, v4
>         -; GFX7-NEXT:    s_bfe_u32 s4, s4, 0x80010
>         +; GFX7-NEXT:    s_lshr_b32 s6, s4, 24
>         +; GFX7-NEXT:    s_bfe_u32 s7, s4, 0x80008
>         +; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
>         +; GFX7-NEXT:    s_bfe_u32 s12, s5, 0x80010
>         +; GFX7-NEXT:    s_lshr_b32 s9, s5, 24
>         +; GFX7-NEXT:    s_and_b32 s5, s5, s8
>         +; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
>         +; GFX7-NEXT:    s_and_b32 s4, s4, s8
>          ; GFX7-NEXT:    v_mov_b32_e32 v1, s5
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v1, s4, v1
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v3, s11, v3
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
>         -; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
>         -; GFX7-NEXT:    v_or_b32_e32 v2, v3, v4
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
>         +; GFX7-NEXT:    v_mov_b32_e32 v2, s10
>         +; GFX7-NEXT:    v_mov_b32_e32 v3, s12
>          ; GFX7-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v2, v0
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
>         +; GFX7-NEXT:    v_mov_b32_e32 v1, s9
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v1, v0
>          ; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
>          ; GFX7-NEXT:    s_endpgm
>          ;
>         @@ -2023,23 +2013,23 @@ define amdgpu_kernel void @udot4_acc8_ve
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v1, s9, v1
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s7, v2
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v3, s6, v3
>         -; GFX7-NEXT:    s_and_b32 s4, s4, s8
>         +; GFX7-NEXT:    s_and_b32 s5, s4, s8
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
>          ; GFX7-NEXT:    v_and_b32_e32 v2, s8, v2
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
>          ; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
>         -; GFX7-NEXT:    v_or_b32_e32 v2, s4, v3
>         +; GFX7-NEXT:    v_or_b32_e32 v2, s5, v3
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
>          ; GFX7-NEXT:    v_and_b32_e32 v2, 0xffff, v2
>          ; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
>          ; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v1
>          ; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v1
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 24, v1
>          ; GFX7-NEXT:    s_waitcnt vmcnt(0)
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v3
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
>          ; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
>          ; GFX7-NEXT:    s_endpgm
>          ;
>         @@ -2055,31 +2045,32 @@ define amdgpu_kernel void @udot4_acc8_ve
>          ; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
>          ; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
>          ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
>         +; GFX8-NEXT:    v_mov_b32_e32 v3, s0
>         +; GFX8-NEXT:    v_mov_b32_e32 v4, s1
>         +; GFX8-NEXT:    s_and_b32 s7, s1, s8
>          ; GFX8-NEXT:    s_lshr_b32 s2, s0, 24
>          ; GFX8-NEXT:    s_lshr_b32 s3, s1, 24
>          ; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
>         -; GFX8-NEXT:    s_and_b32 s7, s1, s8
>         -; GFX8-NEXT:    v_mov_b32_e32 v3, s0
>         -; GFX8-NEXT:    v_mov_b32_e32 v4, s1
>          ; GFX8-NEXT:    v_mul_u32_u24_sdwa v3, v3, v4 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
>         -; GFX8-NEXT:    s_bfe_u32 s4, s0, 0x80010
>         -; GFX8-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX8-NEXT:    s_and_b32 s5, s0, s8
>          ; GFX8-NEXT:    v_mov_b32_e32 v4, s7
>         +; GFX8-NEXT:    v_mul_u32_u24_e32 v4, s5, v4
>         +; GFX8-NEXT:    s_bfe_u32 s4, s0, 0x80010
>         +; GFX8-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX8-NEXT:    v_mov_b32_e32 v6, s3
>          ; GFX8-NEXT:    v_mov_b32_e32 v7, s2
>         -; GFX8-NEXT:    v_mul_u32_u24_e32 v4, s5, v4
>         +; GFX8-NEXT:    v_or_b32_sdwa v3, v4, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>          ; GFX8-NEXT:    v_mul_u32_u24_e32 v5, s4, v5
>          ; GFX8-NEXT:    v_mul_u32_u24_sdwa v6, v7, v6 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX8-NEXT:    v_and_b32_e32 v3, 0xffff, v3
>          ; GFX8-NEXT:    v_or_b32_sdwa v5, v5, v6 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX8-NEXT:    v_or_b32_sdwa v3, v4, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
>         +; GFX8-NEXT:    v_or_b32_e32 v4, v3, v5
>         +; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v4
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
>         -; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v4, v2
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
>         +; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v5
>         +; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         +; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX8-NEXT:    flat_store_byte v[0:1], v2
>          ; GFX8-NEXT:    s_endpgm
>          ;
>         @@ -2101,20 +2092,21 @@ define amdgpu_kernel void @udot4_acc8_ve
>          ; GFX9-NODL-NEXT:    s_lshr_b32 s4, s3, 24
>          ; GFX9-NODL-NEXT:    v_mul_lo_u16_e32 v3, s2, v3
>          ; GFX9-NODL-NEXT:    v_mul_lo_u16_sdwa v4, s2, v4
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1
>         src1_sel:BYTE_1
>         -; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s1
>          ; GFX9-NODL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s1
>          ; GFX9-NODL-NEXT:    s_lshr_b32 s5, s2, 24
>          ; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s4
>          ; GFX9-NODL-NEXT:    v_mul_lo_u16_sdwa v4, s5, v4
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX9-NODL-NEXT:    v_mul_lo_u16_e32 v5, s0, v5
>         +; GFX9-NODL-NEXT:    v_and_b32_e32 v3, 0xffff, v3
>          ; GFX9-NODL-NEXT:    v_or_b32_sdwa v4, v5, v4 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX9-NODL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
>         +; GFX9-NODL-NEXT:    v_or_b32_e32 v4, v3, v4
>         +; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v5, 8, v4
>          ; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v3, v2
>         -; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v2, v4
>         -; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         +; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v2, v5
>         +; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         +; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX9-NODL-NEXT:    s_endpgm
>          ;
>         @@ -2136,20 +2128,21 @@ define amdgpu_kernel void @udot4_acc8_ve
>          ; GFX9-DL-NEXT:    s_lshr_b32 s4, s3, 24
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v3, s2, v3
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, s2, v4
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1
>         src1_sel:BYTE_1
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s1
>          ; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s1
>          ; GFX9-DL-NEXT:    s_lshr_b32 s5, s2, 24
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s4
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, s5, v4
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v5, s0, v5
>         +; GFX9-DL-NEXT:    v_and_b32_e32 v3, 0xffff, v3
>          ; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v5, v4 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
>         +; GFX9-DL-NEXT:    v_or_b32_e32 v4, v3, v4
>         +; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 8, v4
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
>         -; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         +; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
>         +; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         +; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX9-DL-NEXT:    s_endpgm
>          ;
>         @@ -2167,27 +2160,28 @@ define amdgpu_kernel void @udot4_acc8_ve
>          ; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s1
>          ; GFX10-DL-NEXT:    global_load_ubyte v3, v[0:1], off
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_lshr_b32 s0, s3, 24
>         -; GFX10-DL-NEXT:    s_lshr_b32 s5, s4, 24
>         -; GFX10-DL-NEXT:    s_lshr_b32 s1, s3, 16
>         -; GFX10-DL-NEXT:    s_lshr_b32 s6, s4, 16
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v4, s3, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v5, s4, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
>         +; GFX10-DL-NEXT:    s_lshr_b32 s0, s3, 24
>         +; GFX10-DL-NEXT:    s_lshr_b32 s1, s3, 16
>          ; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v6, s3, s4
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v7, s0, s5
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v8, s1, s6
>         +; GFX10-DL-NEXT:    s_lshr_b32 s3, s4, 16
>          ; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v4, v4, v5
>         +; GFX10-DL-NEXT:    s_lshr_b32 s4, s4, 24
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v5, v6, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v6, v7, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v7, v8, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v4, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v4, v7, v6 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v2, v5, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v4, 8, v2
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v6, s1, s3
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v4, v4, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v7, s0, s4
>         +; GFX10-DL-NEXT:    v_or_b32_sdwa v4, v5, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v5, v6, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v7, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v4, 0xffff, v4
>         +; GFX10-DL-NEXT:    v_or_b32_sdwa v2, v5, v2 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    v_or_b32_e32 v2, v4, v2
>         +; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v5, 8, v2
>          ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v2, v3
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v4
>         +; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v4, v3
>         +; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v5
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v2
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v3, v2
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX10-DL-NEXT:    global_store_byte v[0:1], v2, off
>
>         Modified: llvm/trunk/test/CodeGen/AMDGPU/idot8s.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/idot8s.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/AMDGPU/idot8s.ll (original)
>         +++ llvm/trunk/test/CodeGen/AMDGPU/idot8s.ll Tue Jul 23
>         05:39:08 2019
>         @@ -331,39 +331,38 @@ define amdgpu_kernel void @idot8_acc16(<
>          ; GFX8-NEXT:    s_bfe_i32 s1, s4, 0x40000
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX8-NEXT:    s_bfe_i32 s5, s4, 0x40004
>         -; GFX8-NEXT:    s_bfe_i32 s6, s4, 0x40008
>          ; GFX8-NEXT:    s_lshr_b32 s1, s2, 12
>         -; GFX8-NEXT:    s_lshr_b32 s7, s4, 12
>         -; GFX8-NEXT:    s_bfe_i32 s8, s2, 0x40004
>         -; GFX8-NEXT:    s_bfe_i32 s9, s2, 0x40008
>         -; GFX8-NEXT:    v_mov_b32_e32 v4, s6
>         -; GFX8-NEXT:    v_mov_b32_e32 v7, s5
>         +; GFX8-NEXT:    s_lshr_b32 s6, s4, 12
>         +; GFX8-NEXT:    s_bfe_i32 s8, s4, 0x40008
>         +; GFX8-NEXT:    v_mov_b32_e32 v4, s5
>         +; GFX8-NEXT:    s_bfe_i32 s7, s2, 0x40004
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v5, 12, s1
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v6, 12, s7
>         -; GFX8-NEXT:    v_mul_i32_i24_e32 v4, s9, v4
>         -; GFX8-NEXT:    s_bfe_i32 s10, s4, 0x40010
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v6, 12, s6
>         +; GFX8-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX8-NEXT:    s_bfe_i32 s5, s2, 0x40008
>         +; GFX8-NEXT:    s_bfe_i32 s1, s4, 0x40010
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
>         -; GFX8-NEXT:    s_bfe_i32 s12, s4, 0x40014
>         -; GFX8-NEXT:    s_bfe_i32 s11, s2, 0x40010
>         -; GFX8-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX8-NEXT:    s_bfe_i32 s14, s4, 0x40018
>         -; GFX8-NEXT:    s_bfe_i32 s13, s2, 0x40014
>         -; GFX8-NEXT:    v_mov_b32_e32 v9, s12
>         -; GFX8-NEXT:    s_bfe_i32 s15, s2, 0x40018
>         +; GFX8-NEXT:    s_bfe_i32 s8, s4, 0x40014
>         +; GFX8-NEXT:    v_mov_b32_e32 v8, s1
>         +; GFX8-NEXT:    s_bfe_i32 s6, s2, 0x40010
>         +; GFX8-NEXT:    s_bfe_i32 s9, s4, 0x40018
>         +; GFX8-NEXT:    v_mov_b32_e32 v9, s8
>         +; GFX8-NEXT:    s_bfe_i32 s1, s2, 0x40014
>         +; GFX8-NEXT:    s_bfe_i32 s8, s2, 0x40018
>          ; GFX8-NEXT:    s_ashr_i32 s4, s4, 28
>         -; GFX8-NEXT:    v_mov_b32_e32 v10, s14
>         +; GFX8-NEXT:    v_mov_b32_e32 v10, s9
>          ; GFX8-NEXT:    s_ashr_i32 s2, s2, 28
>         +; GFX8-NEXT:    v_mov_b32_e32 v11, s4
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
>         -; GFX8-NEXT:    v_mad_i32_i24 v2, s8, v7, v2
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v4, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s7, v4, v2
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s5, v7, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, v5, v6, v2
>         -; GFX8-NEXT:    v_mad_i32_i24 v2, s11, v8, v2
>         -; GFX8-NEXT:    v_mad_i32_i24 v2, s13, v9, v2
>         -; GFX8-NEXT:    v_mad_i32_i24 v2, s15, v10, v2
>         -; GFX8-NEXT:    v_mov_b32_e32 v3, s4
>         -; GFX8-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s6, v8, v2
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s1, v9, v2
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s8, v10, v2
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s2, v11, v2
>          ; GFX8-NEXT:    flat_store_short v[0:1], v2
>          ; GFX8-NEXT:    s_endpgm
>          ;
>         @@ -382,39 +381,38 @@ define amdgpu_kernel void @idot8_acc16(<
>          ; GFX9-NEXT:    s_bfe_i32 s1, s4, 0x40000
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX9-NEXT:    s_bfe_i32 s5, s4, 0x40004
>         -; GFX9-NEXT:    s_bfe_i32 s6, s4, 0x40008
>          ; GFX9-NEXT:    s_lshr_b32 s1, s2, 12
>         -; GFX9-NEXT:    s_lshr_b32 s7, s4, 12
>         -; GFX9-NEXT:    s_bfe_i32 s8, s2, 0x40004
>         -; GFX9-NEXT:    s_bfe_i32 s9, s2, 0x40008
>         -; GFX9-NEXT:    v_mov_b32_e32 v4, s6
>         -; GFX9-NEXT:    v_mov_b32_e32 v7, s5
>         +; GFX9-NEXT:    s_lshr_b32 s6, s4, 12
>         +; GFX9-NEXT:    s_bfe_i32 s8, s4, 0x40008
>         +; GFX9-NEXT:    v_mov_b32_e32 v4, s5
>         +; GFX9-NEXT:    s_bfe_i32 s7, s2, 0x40004
>          ; GFX9-NEXT:    v_lshlrev_b16_e64 v5, 12, s1
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v6, 12, s7
>         -; GFX9-NEXT:    v_mul_i32_i24_e32 v4, s9, v4
>         -; GFX9-NEXT:    s_bfe_i32 s10, s4, 0x40010
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v6, 12, s6
>         +; GFX9-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX9-NEXT:    s_bfe_i32 s5, s2, 0x40008
>         +; GFX9-NEXT:    s_bfe_i32 s1, s4, 0x40010
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
>         -; GFX9-NEXT:    s_bfe_i32 s12, s4, 0x40014
>         -; GFX9-NEXT:    s_bfe_i32 s11, s2, 0x40010
>         -; GFX9-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX9-NEXT:    s_bfe_i32 s14, s4, 0x40018
>         -; GFX9-NEXT:    s_bfe_i32 s13, s2, 0x40014
>         -; GFX9-NEXT:    v_mov_b32_e32 v9, s12
>         -; GFX9-NEXT:    s_bfe_i32 s15, s2, 0x40018
>         +; GFX9-NEXT:    s_bfe_i32 s8, s4, 0x40014
>         +; GFX9-NEXT:    v_mov_b32_e32 v8, s1
>         +; GFX9-NEXT:    s_bfe_i32 s6, s2, 0x40010
>         +; GFX9-NEXT:    s_bfe_i32 s9, s4, 0x40018
>         +; GFX9-NEXT:    v_mov_b32_e32 v9, s8
>         +; GFX9-NEXT:    s_bfe_i32 s1, s2, 0x40014
>         +; GFX9-NEXT:    s_bfe_i32 s8, s2, 0x40018
>          ; GFX9-NEXT:    s_ashr_i32 s4, s4, 28
>         -; GFX9-NEXT:    v_mov_b32_e32 v10, s14
>         +; GFX9-NEXT:    v_mov_b32_e32 v10, s9
>          ; GFX9-NEXT:    s_ashr_i32 s2, s2, 28
>         +; GFX9-NEXT:    v_mov_b32_e32 v11, s4
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
>         -; GFX9-NEXT:    v_mad_i32_i24 v2, s8, v7, v2
>         -; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s7, v4, v2
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s5, v7, v2
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, v5, v6, v2
>         -; GFX9-NEXT:    v_mad_i32_i24 v2, s11, v8, v2
>         -; GFX9-NEXT:    v_mad_i32_i24 v2, s13, v9, v2
>         -; GFX9-NEXT:    v_mad_i32_i24 v2, s15, v10, v2
>         -; GFX9-NEXT:    v_mov_b32_e32 v3, s4
>         -; GFX9-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s6, v8, v2
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s1, v9, v2
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s8, v10, v2
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s2, v11, v2
>          ; GFX9-NEXT:    global_store_short v[0:1], v2, off
>          ; GFX9-NEXT:    s_endpgm
>          ;
>         @@ -433,39 +431,38 @@ define amdgpu_kernel void @idot8_acc16(<
>          ; GFX9-DL-NEXT:    s_bfe_i32 s1, s4, 0x40000
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX9-DL-NEXT:    s_bfe_i32 s5, s4, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_i32 s6, s4, 0x40008
>          ; GFX9-DL-NEXT:    s_lshr_b32 s1, s2, 12
>         -; GFX9-DL-NEXT:    s_lshr_b32 s7, s4, 12
>         -; GFX9-DL-NEXT:    s_bfe_i32 s8, s2, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_i32 s9, s2, 0x40008
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s6
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s5
>         +; GFX9-DL-NEXT:    s_lshr_b32 s6, s4, 12
>         +; GFX9-DL-NEXT:    s_bfe_i32 s8, s4, 0x40008
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
>         +; GFX9-DL-NEXT:    s_bfe_i32 s7, s2, 0x40004
>          ; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v5, 12, s1
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v6, 12, s7
>         -; GFX9-DL-NEXT:    v_mul_i32_i24_e32 v4, s9, v4
>         -; GFX9-DL-NEXT:    s_bfe_i32 s10, s4, 0x40010
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v6, 12, s6
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX9-DL-NEXT:    s_bfe_i32 s5, s2, 0x40008
>         +; GFX9-DL-NEXT:    s_bfe_i32 s1, s4, 0x40010
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
>         -; GFX9-DL-NEXT:    s_bfe_i32 s12, s4, 0x40014
>         -; GFX9-DL-NEXT:    s_bfe_i32 s11, s2, 0x40010
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX9-DL-NEXT:    s_bfe_i32 s14, s4, 0x40018
>         -; GFX9-DL-NEXT:    s_bfe_i32 s13, s2, 0x40014
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s12
>         -; GFX9-DL-NEXT:    s_bfe_i32 s15, s2, 0x40018
>         +; GFX9-DL-NEXT:    s_bfe_i32 s8, s4, 0x40014
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s1
>         +; GFX9-DL-NEXT:    s_bfe_i32 s6, s2, 0x40010
>         +; GFX9-DL-NEXT:    s_bfe_i32 s9, s4, 0x40018
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s8
>         +; GFX9-DL-NEXT:    s_bfe_i32 s1, s2, 0x40014
>         +; GFX9-DL-NEXT:    s_bfe_i32 s8, s2, 0x40018
>          ; GFX9-DL-NEXT:    s_ashr_i32 s4, s4, 28
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v10, s14
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v10, s9
>          ; GFX9-DL-NEXT:    s_ashr_i32 s2, s2, 28
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v11, s4
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s0, v3, v2
>         -; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s8, v7, v2
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s7, v4, v2
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s5, v7, v2
>          ; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v5, v6, v2
>         -; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s11, v8, v2
>         -; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s13, v9, v2
>         -; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s15, v10, v2
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
>         -; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s2, v3, v2
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s6, v8, v2
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s1, v9, v2
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s8, v10, v2
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s2, v11, v2
>          ; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
>          ; GFX9-DL-NEXT:    s_endpgm
>          ;
>         @@ -496,26 +493,25 @@ define amdgpu_kernel void @idot8_acc16(<
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v5, v5, v2
>          ; GFX10-DL-NEXT:    s_bfe_i32 s9, s2, 0x40010
>          ; GFX10-DL-NEXT:    s_bfe_i32 s10, s4, 0x40010
>         -; GFX10-DL-NEXT:    v_mul_i32_i24_e64 v6, s1, s8
>         +; GFX10-DL-NEXT:    s_bfe_i32 s11, s2, 0x40014
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v4, 12, v4
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v5, 12, v5
>         -; GFX10-DL-NEXT:    s_bfe_i32 s1, s2, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_i32 s8, s4, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_i32 s11, s2, 0x40018
>         +; GFX10-DL-NEXT:    s_bfe_i32 s12, s4, 0x40014
>         +; GFX10-DL-NEXT:    s_bfe_i32 s13, s2, 0x40018
>         +; GFX10-DL-NEXT:    s_bfe_i32 s14, s4, 0x40018
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v4, v4, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v2, v5, v2
>         -; GFX10-DL-NEXT:    s_bfe_i32 s12, s4, 0x40018
>         -; GFX10-DL-NEXT:    s_ashr_i32 s2, s2, 28
>         -; GFX10-DL-NEXT:    s_ashr_i32 s4, s4, 28
>          ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX10-DL-NEXT:    v_mad_i32_i24 v3, s5, s6, v3
>          ; GFX10-DL-NEXT:    v_mad_i32_i24 v3, s7, s0, v3
>         -; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v6
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0
>         src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    s_ashr_i32 s0, s2, 28
>         +; GFX10-DL-NEXT:    v_mad_i32_i24 v3, s1, s8, v3
>         +; GFX10-DL-NEXT:    s_ashr_i32 s1, s4, 28
>          ; GFX10-DL-NEXT:    v_mad_u32_u24 v2, v4, v2, v3
>          ; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s9, s10, v2
>         -; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s1, s8, v2
>          ; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s11, s12, v2
>         -; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s2, s4, v2
>         +; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s13, s14, v2
>         +; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s0, s1, v2
>          ; GFX10-DL-NEXT:    global_store_short v[0:1], v2, off
>          ; GFX10-DL-NEXT:    s_endpgm
>                                                 <8 x i4> addrspace(1)*
>         %src2,
>         @@ -668,21 +664,20 @@ define amdgpu_kernel void @idot8_acc8(<8
>          ; GFX8-NEXT:    s_bfe_i32 s7, s1, 0x40000
>          ; GFX8-NEXT:    s_lshr_b32 s5, s1, 12
>          ; GFX8-NEXT:    s_bfe_i32 s9, s1, 0x40004
>         -; GFX8-NEXT:    s_bfe_i32 s11, s1, 0x40008
>          ; GFX8-NEXT:    s_bfe_i32 s6, s0, 0x40000
>         -; GFX8-NEXT:    v_mov_b32_e32 v6, s7
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v4, 12, s4
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v5, 12, s5
>         +; GFX8-NEXT:    v_mov_b32_e32 v5, s7
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v3, 12, s4
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v4, 12, s5
>         +; GFX8-NEXT:    s_bfe_i32 s11, s1, 0x40008
>          ; GFX8-NEXT:    s_bfe_i32 s8, s0, 0x40004
>         -; GFX8-NEXT:    s_bfe_i32 s10, s0, 0x40008
>         -; GFX8-NEXT:    v_mov_b32_e32 v3, s11
>         -; GFX8-NEXT:    v_mov_b32_e32 v7, s9
>         +; GFX8-NEXT:    v_mov_b32_e32 v6, s9
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
>         -; GFX8-NEXT:    v_mul_i32_i24_e32 v3, s10, v3
>         +; GFX8-NEXT:    s_bfe_i32 s10, s0, 0x40008
>         +; GFX8-NEXT:    v_mov_b32_e32 v7, s11
>          ; GFX8-NEXT:    s_bfe_i32 s13, s1, 0x40010
>         +; GFX8-NEXT:    v_and_b32_e32 v3, s2, v3
>          ; GFX8-NEXT:    v_and_b32_e32 v4, s2, v4
>         -; GFX8-NEXT:    v_and_b32_e32 v5, s2, v5
>          ; GFX8-NEXT:    s_bfe_i32 s15, s1, 0x40014
>          ; GFX8-NEXT:    s_bfe_i32 s12, s0, 0x40010
>          ; GFX8-NEXT:    v_mov_b32_e32 v8, s13
>         @@ -694,10 +689,10 @@ define amdgpu_kernel void @idot8_acc8(<8
>          ; GFX8-NEXT:    v_mov_b32_e32 v10, s17
>          ; GFX8-NEXT:    s_ashr_i32 s0, s0, 28
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX8-NEXT:    v_mad_i32_i24 v2, s6, v6, v2
>         -; GFX8-NEXT:    v_mad_i32_i24 v2, s8, v7, v2
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, v4, v5, v2
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s6, v5, v2
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s8, v6, v2
>         +; GFX8-NEXT:    v_mad_i32_i24 v2, s10, v7, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
>          ; GFX8-NEXT:    v_mad_i32_i24 v2, s12, v8, v2
>          ; GFX8-NEXT:    v_mad_i32_i24 v2, s14, v9, v2
>          ; GFX8-NEXT:    v_mad_i32_i24 v2, s16, v10, v2
>         @@ -722,21 +717,20 @@ define amdgpu_kernel void @idot8_acc8(<8
>          ; GFX9-NEXT:    s_bfe_i32 s7, s1, 0x40000
>          ; GFX9-NEXT:    s_lshr_b32 s5, s1, 12
>          ; GFX9-NEXT:    s_bfe_i32 s9, s1, 0x40004
>         -; GFX9-NEXT:    s_bfe_i32 s11, s1, 0x40008
>          ; GFX9-NEXT:    s_bfe_i32 s6, s0, 0x40000
>         -; GFX9-NEXT:    v_mov_b32_e32 v6, s7
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v4, 12, s4
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v5, 12, s5
>         +; GFX9-NEXT:    v_mov_b32_e32 v5, s7
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v3, 12, s4
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v4, 12, s5
>         +; GFX9-NEXT:    s_bfe_i32 s11, s1, 0x40008
>          ; GFX9-NEXT:    s_bfe_i32 s8, s0, 0x40004
>         -; GFX9-NEXT:    s_bfe_i32 s10, s0, 0x40008
>         -; GFX9-NEXT:    v_mov_b32_e32 v3, s11
>         -; GFX9-NEXT:    v_mov_b32_e32 v7, s9
>         +; GFX9-NEXT:    v_mov_b32_e32 v6, s9
>         +; GFX9-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
>         -; GFX9-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
>         -; GFX9-NEXT:    v_mul_i32_i24_e32 v3, s10, v3
>         +; GFX9-NEXT:    s_bfe_i32 s10, s0, 0x40008
>         +; GFX9-NEXT:    v_mov_b32_e32 v7, s11
>          ; GFX9-NEXT:    s_bfe_i32 s13, s1, 0x40010
>         +; GFX9-NEXT:    v_and_b32_e32 v3, s2, v3
>          ; GFX9-NEXT:    v_and_b32_e32 v4, s2, v4
>         -; GFX9-NEXT:    v_and_b32_e32 v5, s2, v5
>          ; GFX9-NEXT:    s_bfe_i32 s15, s1, 0x40014
>          ; GFX9-NEXT:    s_bfe_i32 s12, s0, 0x40010
>          ; GFX9-NEXT:    v_mov_b32_e32 v8, s13
>         @@ -748,10 +742,10 @@ define amdgpu_kernel void @idot8_acc8(<8
>          ; GFX9-NEXT:    v_mov_b32_e32 v10, s17
>          ; GFX9-NEXT:    s_ashr_i32 s0, s0, 28
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX9-NEXT:    v_mad_i32_i24 v2, s6, v6, v2
>         -; GFX9-NEXT:    v_mad_i32_i24 v2, s8, v7, v2
>         -; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, v4, v5, v2
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s6, v5, v2
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s8, v6, v2
>         +; GFX9-NEXT:    v_mad_i32_i24 v2, s10, v7, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
>          ; GFX9-NEXT:    v_mad_i32_i24 v2, s12, v8, v2
>          ; GFX9-NEXT:    v_mad_i32_i24 v2, s14, v9, v2
>          ; GFX9-NEXT:    v_mad_i32_i24 v2, s16, v10, v2
>         @@ -776,21 +770,20 @@ define amdgpu_kernel void @idot8_acc8(<8
>          ; GFX9-DL-NEXT:    s_bfe_i32 s7, s1, 0x40000
>          ; GFX9-DL-NEXT:    s_lshr_b32 s5, s1, 12
>          ; GFX9-DL-NEXT:    s_bfe_i32 s9, s1, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_i32 s11, s1, 0x40008
>          ; GFX9-DL-NEXT:    s_bfe_i32 s6, s0, 0x40000
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s7
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v4, 12, s4
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v5, 12, s5
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s7
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v3, 12, s4
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v4, 12, s5
>         +; GFX9-DL-NEXT:    s_bfe_i32 s11, s1, 0x40008
>          ; GFX9-DL-NEXT:    s_bfe_i32 s8, s0, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_i32 s10, s0, 0x40008
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s11
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s9
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s9
>         +; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
>         -; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
>         -; GFX9-DL-NEXT:    v_mul_i32_i24_e32 v3, s10, v3
>         +; GFX9-DL-NEXT:    s_bfe_i32 s10, s0, 0x40008
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s11
>          ; GFX9-DL-NEXT:    s_bfe_i32 s13, s1, 0x40010
>         +; GFX9-DL-NEXT:    v_and_b32_e32 v3, s2, v3
>          ; GFX9-DL-NEXT:    v_and_b32_e32 v4, s2, v4
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v5, s2, v5
>          ; GFX9-DL-NEXT:    s_bfe_i32 s15, s1, 0x40014
>          ; GFX9-DL-NEXT:    s_bfe_i32 s12, s0, 0x40010
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s13
>         @@ -802,10 +795,10 @@ define amdgpu_kernel void @idot8_acc8(<8
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v10, s17
>          ; GFX9-DL-NEXT:    s_ashr_i32 s0, s0, 28
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s6, v6, v2
>         -; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s8, v7, v2
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v4, v5, v2
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s6, v5, v2
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s8, v6, v2
>         +; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s10, v7, v2
>         +; GFX9-DL-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
>          ; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s12, v8, v2
>          ; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s14, v9, v2
>          ; GFX9-DL-NEXT:    v_mad_i32_i24 v2, s16, v10, v2
>         @@ -842,26 +835,25 @@ define amdgpu_kernel void @idot8_acc8(<8
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v2, v5, v2
>          ; GFX10-DL-NEXT:    s_bfe_i32 s10, s4, 0x40010
>          ; GFX10-DL-NEXT:    s_bfe_i32 s11, s5, 0x40010
>         -; GFX10-DL-NEXT:    v_mul_i32_i24_e64 v5, s1, s9
>         +; GFX10-DL-NEXT:    s_bfe_i32 s12, s4, 0x40014
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v4, 12, v4
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v2, 12, v2
>         -; GFX10-DL-NEXT:    s_bfe_i32 s1, s4, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_i32 s9, s5, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_i32 s12, s4, 0x40018
>         +; GFX10-DL-NEXT:    s_bfe_i32 s13, s5, 0x40014
>         +; GFX10-DL-NEXT:    s_bfe_i32 s14, s4, 0x40018
>         +; GFX10-DL-NEXT:    s_bfe_i32 s15, s5, 0x40018
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v4, v4, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v2, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX10-DL-NEXT:    s_bfe_i32 s2, s5, 0x40018
>         -; GFX10-DL-NEXT:    s_ashr_i32 s4, s4, 28
>         -; GFX10-DL-NEXT:    s_ashr_i32 s5, s5, 28
>          ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX10-DL-NEXT:    v_mad_i32_i24 v3, s6, s7, v3
>          ; GFX10-DL-NEXT:    v_mad_i32_i24 v3, s8, s0, v3
>         -; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v5
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0
>         src1_sel:BYTE_0
>         +; GFX10-DL-NEXT:    s_ashr_i32 s0, s4, 28
>         +; GFX10-DL-NEXT:    v_mad_i32_i24 v3, s1, s9, v3
>         +; GFX10-DL-NEXT:    s_ashr_i32 s1, s5, 28
>          ; GFX10-DL-NEXT:    v_mad_u32_u24 v2, v4, v2, v3
>          ; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s10, s11, v2
>         -; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s1, s9, v2
>         -; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s12, s2, v2
>         -; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s4, s5, v2
>         +; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s12, s13, v2
>         +; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s14, s15, v2
>         +; GFX10-DL-NEXT:    v_mad_i32_i24 v2, s0, s1, v2
>          ; GFX10-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX10-DL-NEXT:    s_endpgm
>                                                 <8 x i4> addrspace(1)*
>         %src2,
>         @@ -1582,69 +1574,57 @@ define amdgpu_kernel void @idot8_acc16_v
>          ; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
>          ; GFX7-NEXT:    s_mov_b32 s7, 0xf000
>          ; GFX7-NEXT:    s_mov_b32 s6, -1
>         -; GFX7-NEXT:    s_mov_b32 s0, 0xffff
>         +; GFX7-NEXT:    s_mov_b32 s2, 0xffff
>          ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX7-NEXT:    s_load_dword s1, s[8:9], 0x0
>         +; GFX7-NEXT:    s_load_dword s0, s[8:9], 0x0
>          ; GFX7-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
>         -; GFX7-NEXT:    s_load_dword s2, s[10:11], 0x0
>         +; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
>          ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX7-NEXT:    s_bfe_i32 s8, s1, 0x40010
>         -; GFX7-NEXT:    s_bfe_i32 s9, s1, 0x40014
>         -; GFX7-NEXT:    s_bfe_i32 s15, s2, 0x40010
>         -; GFX7-NEXT:    s_bfe_i32 s16, s2, 0x40014
>         -; GFX7-NEXT:    s_bfe_i32 s17, s2, 0x40018
>         -; GFX7-NEXT:    s_ashr_i32 s18, s2, 28
>         -; GFX7-NEXT:    s_bfe_i32 s19, s2, 0x40000
>         -; GFX7-NEXT:    s_bfe_i32 s20, s2, 0x40004
>         -; GFX7-NEXT:    s_bfe_i32 s21, s2, 0x40008
>         -; GFX7-NEXT:    s_bfe_i32 s2, s2, 0x4000c
>         -; GFX7-NEXT:    s_bfe_i32 s10, s1, 0x40018
>         -; GFX7-NEXT:    s_ashr_i32 s11, s1, 28
>         -; GFX7-NEXT:    s_bfe_i32 s12, s1, 0x40000
>         +; GFX7-NEXT:    s_ashr_i32 s8, s0, 28
>         +; GFX7-NEXT:    s_bfe_i32 s9, s0, 0x40018
>         +; GFX7-NEXT:    s_bfe_i32 s16, s1, 0x40018
>         +; GFX7-NEXT:    s_bfe_i32 s17, s1, 0x40014
>         +; GFX7-NEXT:    s_bfe_i32 s18, s1, 0x40010
>         +; GFX7-NEXT:    s_bfe_i32 s19, s1, 0x40000
>         +; GFX7-NEXT:    s_bfe_i32 s20, s1, 0x40004
>         +; GFX7-NEXT:    s_bfe_i32 s21, s1, 0x40008
>         +; GFX7-NEXT:    s_ashr_i32 s15, s1, 28
>         +; GFX7-NEXT:    s_bfe_i32 s1, s1, 0x4000c
>         +; GFX7-NEXT:    s_bfe_i32 s10, s0, 0x40014
>         +; GFX7-NEXT:    s_bfe_i32 s11, s0, 0x40010
>         +; GFX7-NEXT:    s_bfe_i32 s12, s0, 0x40000
>          ; GFX7-NEXT:    v_mov_b32_e32 v4, s19
>         -; GFX7-NEXT:    s_bfe_i32 s13, s1, 0x40004
>         +; GFX7-NEXT:    s_bfe_i32 s13, s0, 0x40004
>          ; GFX7-NEXT:    v_mov_b32_e32 v3, s20
>         -; GFX7-NEXT:    s_bfe_i32 s14, s1, 0x40008
>         +; GFX7-NEXT:    s_bfe_i32 s14, s0, 0x40008
>          ; GFX7-NEXT:    v_mov_b32_e32 v2, s21
>         -; GFX7-NEXT:    s_bfe_i32 s1, s1, 0x4000c
>         -; GFX7-NEXT:    v_mov_b32_e32 v1, s2
>         -; GFX7-NEXT:    v_mov_b32_e32 v5, s18
>         -; GFX7-NEXT:    v_mov_b32_e32 v6, s17
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v1, s1, v1
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v2, s14, v2
>         +; GFX7-NEXT:    s_bfe_i32 s0, s0, 0x4000c
>         +; GFX7-NEXT:    v_mov_b32_e32 v1, s1
>         +; GFX7-NEXT:    v_mul_i32_i24_e32 v1, s0, v1
>         +; GFX7-NEXT:    v_mul_i32_i24_e32 v8, s14, v2
>          ; GFX7-NEXT:    v_mul_i32_i24_e32 v3, s13, v3
>          ; GFX7-NEXT:    v_mul_i32_i24_e32 v4, s12, v4
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v5, s11, v5
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v6, s10, v6
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
>         -; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
>         +; GFX7-NEXT:    v_and_b32_e32 v8, s2, v8
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
>         -; GFX7-NEXT:    v_and_b32_e32 v4, s0, v4
>         -; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
>         -; GFX7-NEXT:    v_or_b32_e32 v2, v4, v3
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
>         -; GFX7-NEXT:    v_and_b32_e32 v6, s0, v6
>         +; GFX7-NEXT:    v_and_b32_e32 v4, s2, v4
>         +; GFX7-NEXT:    v_or_b32_e32 v3, v4, v3
>         +; GFX7-NEXT:    v_or_b32_e32 v1, v8, v1
>         +; GFX7-NEXT:    v_alignbit_b32 v4, v1, v3, 16
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
>         +; GFX7-NEXT:    v_mov_b32_e32 v5, s18
>         +; GFX7-NEXT:    v_mov_b32_e32 v6, s17
>          ; GFX7-NEXT:    v_mov_b32_e32 v7, s16
>         -; GFX7-NEXT:    v_mov_b32_e32 v8, s15
>         -; GFX7-NEXT:    v_or_b32_e32 v3, v6, v5
>         -; GFX7-NEXT:    v_alignbit_b32 v5, v1, v2, 16
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v7, s9, v7
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v8, s8, v8
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
>         -; GFX7-NEXT:    v_and_b32_e32 v8, s0, v8
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
>         -; GFX7-NEXT:    v_or_b32_e32 v4, v8, v7
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v3
>          ; GFX7-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v5, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v6, v0
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v3
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v7, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v8, v0
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s14, v2, v0
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s11, v5, v0
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s10, v6, v0
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s9, v7, v0
>         +; GFX7-NEXT:    v_mov_b32_e32 v1, s15
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s8, v1, v0
>          ; GFX7-NEXT:    buffer_store_short v0, off, s[4:7], 0
>          ; GFX7-NEXT:    s_endpgm
>          ;
>         @@ -1662,26 +1642,25 @@ define amdgpu_kernel void @idot8_acc16_v
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v3, 12, s2
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v4, 12, s4
>          ; GFX8-NEXT:    s_lshr_b32 s0, s2, 4
>         -; GFX8-NEXT:    s_lshr_b32 s1, s2, 8
>         -; GFX8-NEXT:    s_lshr_b32 s5, s4, 4
>         +; GFX8-NEXT:    s_lshr_b32 s1, s4, 4
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v5, 12, s0
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v6, 12, s1
>         +; GFX8-NEXT:    s_lshr_b32 s5, s2, 8
>          ; GFX8-NEXT:    s_lshr_b32 s6, s4, 8
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v5, 12, s1
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v6, 12, s0
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v7, 12, s6
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v3, 12, v3
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v4, 12, v4
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v7, 12, s5
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s6
>          ; GFX8-NEXT:    s_lshr_b32 s0, s2, 12
>          ; GFX8-NEXT:    s_lshr_b32 s1, s4, 12
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v5, 12, v5
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v9, 12, s0
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v10, 12, s1
>          ; GFX8-NEXT:    s_lshr_b32 s5, s2, 16
>          ; GFX8-NEXT:    s_lshr_b32 s6, s4, 16
>         -; GFX8-NEXT:    v_mul_u32_u24_e32 v5, v5, v7
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v11, 12, s5
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v12, 12, s6
>          ; GFX8-NEXT:    s_lshr_b32 s0, s2, 20
>         @@ -1695,26 +1674,26 @@ define amdgpu_kernel void @idot8_acc16_v
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v15, 12, s5
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v17, 12, s6
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v16, 12, s6
>          ; GFX8-NEXT:    s_lshr_b32 s0, s2, 28
>          ; GFX8-NEXT:    s_lshr_b32 s1, s4, 28
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v16, 12, s0
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v17, 12, s0
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v18, 12, s1
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v15, 12, v15
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v17, 12, v17
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v16, 12, v16
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v17, 12, v17
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v18, 12, v18
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, v3, v4, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, v6, v8, v2
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v5, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, v5, v6, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, v7, v8, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, v9, v10, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, v11, v12, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, v13, v14, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, v15, v17, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, v16, v18, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, v15, v16, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, v17, v18, v2
>          ; GFX8-NEXT:    flat_store_short v[0:1], v2
>          ; GFX8-NEXT:    s_endpgm
>          ;
>         @@ -1776,7 +1755,7 @@ define amdgpu_kernel void @idot8_acc16_v
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         +; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         @@ -1843,7 +1822,7 @@ define amdgpu_kernel void @idot8_acc16_v
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         +; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         @@ -1911,7 +1890,7 @@ define amdgpu_kernel void @idot8_acc16_v
>          ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX10-DL-NEXT:    v_add_nc_u32_e32 v2, v3, v2
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v3
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0
>         src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX10-DL-NEXT:    v_add_nc_u32_e32 v2, v2, v5
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v5
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         @@ -1994,7 +1973,7 @@ define amdgpu_kernel void @idot8_acc8_ve
>          ; GFX7-NEXT:    v_mul_i32_i24_e32 v1, s2, v1
>          ; GFX7-NEXT:    v_mul_i32_i24_e32 v2, s15, v2
>          ; GFX7-NEXT:    v_mul_i32_i24_e32 v3, s14, v3
>         -; GFX7-NEXT:    v_mul_i32_i24_e32 v4, s13, v4
>         +; GFX7-NEXT:    v_mul_i32_i24_e32 v9, s13, v4
>          ; GFX7-NEXT:    v_mul_i32_i24_e32 v5, s12, v5
>          ; GFX7-NEXT:    v_mul_i32_i24_e32 v6, s11, v6
>          ; GFX7-NEXT:    v_mul_i32_i24_e32 v7, s10, v7
>         @@ -2002,36 +1981,36 @@ define amdgpu_kernel void @idot8_acc8_ve
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
>          ; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
>         -; GFX7-NEXT:    v_and_b32_e32 v4, s0, v4
>         +; GFX7-NEXT:    v_and_b32_e32 v9, s0, v9
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
>          ; GFX7-NEXT:    v_and_b32_e32 v6, s0, v6
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 8, v7
>          ; GFX7-NEXT:    v_and_b32_e32 v8, s0, v8
>          ; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
>         -; GFX7-NEXT:    v_or_b32_e32 v2, v4, v3
>         +; GFX7-NEXT:    v_or_b32_e32 v2, v9, v3
>          ; GFX7-NEXT:    v_or_b32_e32 v3, v6, v5
>         -; GFX7-NEXT:    v_or_b32_e32 v4, v8, v7
>         +; GFX7-NEXT:    v_or_b32_e32 v5, v8, v7
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
>          ; GFX7-NEXT:    v_and_b32_e32 v2, s1, v2
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
>         -; GFX7-NEXT:    v_and_b32_e32 v4, s1, v4
>         +; GFX7-NEXT:    v_and_b32_e32 v5, s1, v5
>          ; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
>         -; GFX7-NEXT:    v_or_b32_e32 v2, v4, v3
>         +; GFX7-NEXT:    v_or_b32_e32 v2, v5, v3
>          ; GFX7-NEXT:    v_alignbit_b32 v3, v1, v2, 8
>         -; GFX7-NEXT:    v_alignbit_b32 v4, v1, v2, 16
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 24, v2
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 24, v1
>         +; GFX7-NEXT:    v_alignbit_b32 v5, v1, v2, 16
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v2
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 24, v1
>          ; GFX7-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v5, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v6, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v7, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v8, v0
>         +; GFX7-NEXT:    v_mad_i32_i24 v0, s13, v4, v0
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v7
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v8
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>          ; GFX7-NEXT:    buffer_store_byte v0, off, s[4:7], 0
>          ; GFX7-NEXT:    s_endpgm
>          ;
>         @@ -2068,55 +2047,56 @@ define amdgpu_kernel void @idot8_acc8_ve
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
>         -; GFX8-NEXT:    v_mul_u32_u24_sdwa v3, v3, v7 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         -; GFX8-NEXT:    v_mul_u32_u24_sdwa v4, v4, v8 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         -; GFX8-NEXT:    v_mul_u32_u24_sdwa v5, v5, v9 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         -; GFX8-NEXT:    v_mul_u32_u24_sdwa v6, v6, v10 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         +; GFX8-NEXT:    s_lshr_b32 s5, s4, 20
>         +; GFX8-NEXT:    s_lshr_b32 s6, s4, 16
>          ; GFX8-NEXT:    s_lshr_b32 s0, s2, 20
>         +; GFX8-NEXT:    v_mul_u32_u24_sdwa v6, v6, v10 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         +; GFX8-NEXT:    v_mul_u32_u24_sdwa v5, v5, v9 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>          ; GFX8-NEXT:    s_lshr_b32 s1, s2, 16
>         -; GFX8-NEXT:    s_lshr_b32 s5, s2, 28
>         -; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
>         -; GFX8-NEXT:    s_lshr_b32 s6, s4, 20
>         -; GFX8-NEXT:    s_lshr_b32 s7, s4, 16
>         +; GFX8-NEXT:    v_mul_u32_u24_sdwa v3, v3, v7 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         +; GFX8-NEXT:    v_mul_u32_u24_sdwa v4, v4, v8 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         +; GFX8-NEXT:    s_lshr_b32 s7, s2, 28
>          ; GFX8-NEXT:    s_lshr_b32 s8, s4, 28
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v9, 12, s1
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v10, 12, s0
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v13, 12, s6
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v14, 12, s5
>         +; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
>          ; GFX8-NEXT:    s_lshr_b32 s4, s4, 24
>          ; GFX8-NEXT:    v_or_b32_sdwa v5, v6, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>          ; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX8-NEXT:    v_or_b32_sdwa v3, v5, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v7, 12, s2
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s5
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v9, 12, s1
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v10, 12, s0
>         +; GFX8-NEXT:    v_lshlrev_b16_e64 v8, 12, s7
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v11, 12, s4
>          ; GFX8-NEXT:    v_lshlrev_b16_e64 v12, 12, s8
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v13, 12, s7
>         -; GFX8-NEXT:    v_lshlrev_b16_e64 v14, 12, s6
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
>         -; GFX8-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
>         +; GFX8-NEXT:    v_or_b32_sdwa v3, v5, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
>          ; GFX8-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
>         -; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
>         -; GFX8-NEXT:    v_mul_u32_u24_sdwa v7, v7, v11 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         -; GFX8-NEXT:    v_mul_u32_u24_sdwa v8, v8, v12 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v7, 12, v7
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
>         +; GFX8-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
>          ; GFX8-NEXT:    v_mul_u32_u24_sdwa v9, v9, v13 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>          ; GFX8-NEXT:    v_mul_u32_u24_sdwa v10, v10, v14
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0
>         src1_sel:BYTE_0
>         +; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
>          ; GFX8-NEXT:    v_or_b32_sdwa v9, v9, v10 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX8-NEXT:    v_mul_u32_u24_sdwa v7, v7, v11 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>         +; GFX8-NEXT:    v_mul_u32_u24_sdwa v8, v8, v12 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0
>          ; GFX8-NEXT:    v_or_b32_sdwa v7, v7, v8 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX8-NEXT:    v_or_b32_sdwa v4, v9, v7 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v4
>         +; GFX8-NEXT:    v_and_b32_e32 v4, 0xffff, v9
>         +; GFX8-NEXT:    v_or_b32_e32 v5, v4, v7
>         +; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v5
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
>         -; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v5, v2
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
>         +; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v6, v2
>         +; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
>          ; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
>         -; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v6
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         +; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v7, v2
>         +; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v5, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
>         +; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v5, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
>          ; GFX8-NEXT:    flat_store_byte v[0:1], v2
>          ; GFX8-NEXT:    s_endpgm
>          ;
>         @@ -2153,55 +2133,56 @@ define amdgpu_kernel void @idot8_acc8_ve
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
>         +; GFX9-NEXT:    s_lshr_b32 s0, s2, 20
>         +; GFX9-NEXT:    s_lshr_b32 s5, s4, 20
>         +; GFX9-NEXT:    s_lshr_b32 s6, s4, 16
>         +; GFX9-NEXT:    s_lshr_b32 s1, s2, 16
>          ; GFX9-NEXT:    v_mul_lo_u16_e32 v6, v6, v10
>          ; GFX9-NEXT:    v_mul_lo_u16_sdwa v5, v5, v9 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX9-NEXT:    v_mul_lo_u16_sdwa v4, v4, v8 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX9-NEXT:    v_mul_lo_u16_e32 v3, v3, v7
>         -; GFX9-NEXT:    s_lshr_b32 s0, s2, 20
>         -; GFX9-NEXT:    s_lshr_b32 s1, s2, 16
>         -; GFX9-NEXT:    s_lshr_b32 s5, s2, 28
>         -; GFX9-NEXT:    s_lshr_b32 s2, s2, 24
>         -; GFX9-NEXT:    s_lshr_b32 s6, s4, 20
>         -; GFX9-NEXT:    s_lshr_b32 s7, s4, 16
>         +; GFX9-NEXT:    s_lshr_b32 s7, s2, 28
>          ; GFX9-NEXT:    s_lshr_b32 s8, s4, 28
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v10, 12, s1
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v11, 12, s0
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v14, 12, s6
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v15, 12, s5
>         +; GFX9-NEXT:    s_lshr_b32 s2, s2, 24
>          ; GFX9-NEXT:    s_lshr_b32 s4, s4, 24
>          ; GFX9-NEXT:    v_or_b32_sdwa v5, v6, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>          ; GFX9-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v8, 12, s2
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v9, 12, s7
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v12, 12, s4
>         +; GFX9-NEXT:    v_lshlrev_b16_e64 v13, 12, s8
>          ; GFX9-NEXT:    v_or_b32_sdwa v3, v5, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v9, 12, s2
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v10, 12, s5
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v11, 12, s1
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v12, 12, s0
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v13, 12, s4
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v14, 12, s8
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v15, 12, s7
>         -; GFX9-NEXT:    v_lshlrev_b16_e64 v16, 12, s6
>         -; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
>         -; GFX9-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v15, 12, v15
>         +; GFX9-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
>          ; GFX9-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
>         -; GFX9-NEXT:    v_ashrrev_i16_e32 v16, 12, v16
>         -; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
>         -; GFX9-NEXT:    v_mul_lo_u16_sdwa v12, v12, v16
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-NEXT:    v_mul_lo_u16_e32 v11, v11, v15
>         -; GFX9-NEXT:    v_mul_lo_u16_sdwa v10, v10, v14
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-NEXT:    v_mul_lo_u16_e32 v9, v9, v13
>         -; GFX9-NEXT:    v_or_b32_sdwa v7, v11, v12 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX9-NEXT:    v_or_b32_sdwa v8, v9, v10 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX9-NEXT:    v_or_b32_sdwa v4, v7, v8 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         +; GFX9-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
>         +; GFX9-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
>         +; GFX9-NEXT:    v_mul_lo_u16_sdwa v11, v11, v15
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-NEXT:    v_mul_lo_u16_e32 v10, v10, v14
>         +; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
>         +; GFX9-NEXT:    v_or_b32_sdwa v7, v10, v11 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX9-NEXT:    v_mul_lo_u16_sdwa v9, v9, v13 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-NEXT:    v_mul_lo_u16_e32 v8, v8, v12
>         +; GFX9-NEXT:    v_or_b32_sdwa v8, v8, v9 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX9-NEXT:    v_and_b32_e32 v4, 0xffff, v7
>         +; GFX9-NEXT:    v_or_b32_e32 v5, v4, v8
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
>         -; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
>         -; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
>         +; GFX9-NEXT:    v_add_u32_e32 v2, v2, v6
>         +; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
>         -; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
>         +; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v5
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
>         -; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         +; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         +; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX9-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX9-NEXT:    s_endpgm
>          ;
>         @@ -2238,55 +2219,56 @@ define amdgpu_kernel void @idot8_acc8_ve
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v6, 12, v6
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
>         +; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 20
>         +; GFX9-DL-NEXT:    s_lshr_b32 s5, s4, 20
>         +; GFX9-DL-NEXT:    s_lshr_b32 s6, s4, 16
>         +; GFX9-DL-NEXT:    s_lshr_b32 s1, s2, 16
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v6, v6, v10
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v5, v5, v9
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, v4, v8
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v3, v3, v7
>         -; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 20
>         -; GFX9-DL-NEXT:    s_lshr_b32 s1, s2, 16
>         -; GFX9-DL-NEXT:    s_lshr_b32 s5, s2, 28
>         -; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 24
>         -; GFX9-DL-NEXT:    s_lshr_b32 s6, s4, 20
>         -; GFX9-DL-NEXT:    s_lshr_b32 s7, s4, 16
>         +; GFX9-DL-NEXT:    s_lshr_b32 s7, s2, 28
>          ; GFX9-DL-NEXT:    s_lshr_b32 s8, s4, 28
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v10, 12, s1
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v11, 12, s0
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v14, 12, s6
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v15, 12, s5
>         +; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 24
>          ; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 24
>          ; GFX9-DL-NEXT:    v_or_b32_sdwa v5, v6, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>          ; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v8, 12, s2
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v9, 12, s7
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v12, 12, s4
>         +; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v13, 12, s8
>          ; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v5, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v9, 12, s2
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v10, 12, s5
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v11, 12, s1
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v12, 12, s0
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v13, 12, s4
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v14, 12, s8
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v15, 12, s7
>         -; GFX9-DL-NEXT:    v_lshlrev_b16_e64 v16, 12, s6
>         -; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
>         -; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v10, 12, v10
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v14, 12, v14
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v11, 12, v11
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v15, 12, v15
>         +; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v8, 12, v8
>          ; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v12, 12, v12
>         -; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v16, 12, v16
>         -; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
>         -; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v12, v12, v16
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v11, v11, v15
>         -; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v10, v10, v14
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v9, v9, v13
>         -; GFX9-DL-NEXT:    v_or_b32_sdwa v7, v11, v12 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_or_b32_sdwa v8, v9, v10 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v7, v8 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v9, 12, v9
>         +; GFX9-DL-NEXT:    v_ashrrev_i16_e32 v13, 12, v13
>         +; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v11, v11, v15
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v10, v10, v14
>         +; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
>         +; GFX9-DL-NEXT:    v_or_b32_sdwa v7, v10, v11 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v9, v9, v13
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v8, v8, v12
>         +; GFX9-DL-NEXT:    v_or_b32_sdwa v8, v8, v9 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_and_b32_e32 v4, 0xffff, v7
>         +; GFX9-DL-NEXT:    v_or_b32_e32 v5, v4, v8
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
>         -; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
>         +; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v6
>         +; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
>         -; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
>         +; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v3, 8, v5
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v3
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         +; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         +; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX9-DL-NEXT:    s_endpgm
>          ;
>         @@ -2332,86 +2314,87 @@ define amdgpu_kernel void @idot8_acc8_ve
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v9, 12, v9
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v19, 12, v10
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v11, 12, v11
>         +; GFX10-DL-NEXT:    s_lshr_b32 s0, s4, 16
>         +; GFX10-DL-NEXT:    s_lshr_b32 s1, s4, 20
>         +; GFX10-DL-NEXT:    s_lshr_b32 s7, s5, 20
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v15, 12, v6
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v7, 12, v7
>         -; GFX10-DL-NEXT:    s_lshr_b32 s1, s4, 20
>         -; GFX10-DL-NEXT:    s_lshr_b32 s6, s4, 24
>         -; GFX10-DL-NEXT:    s_lshr_b32 s0, s4, 16
>         +; GFX10-DL-NEXT:    s_lshr_b32 s6, s5, 16
>         +; GFX10-DL-NEXT:    s_lshr_b32 s8, s4, 24
>         +; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v12, 12, s1
>         +; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v13, 12, s0
>          ; GFX10-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX10-DL-NEXT:    s_lshr_b32 s8, s5, 20
>          ; GFX10-DL-NEXT:    s_lshr_b32 s9, s5, 24
>         -; GFX10-DL-NEXT:    s_lshr_b32 s7, s5, 16
>          ; GFX10-DL-NEXT:    s_lshr_b32 s5, s5, 28
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v23, v15, v2
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v27, v15, v2
>         +; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v15, 12, s6
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v10, v19, v2
>         +; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v23, 12, s7
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v5, v5, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v8, v8, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v4, v4, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v9, v9, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v22, v7, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v11, v11, v2
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v15, v15, v2
>          ; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v5, v5, v8
>         -; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v12, 12, s4
>          ; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v4, v4, v9
>         -; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v13, 12, s6
>         +; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v16, 12, s4
>          ; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v7, v22, v11
>         -; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v15, 12, s0
>         -; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v31, 12, s8
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v23, v23, v10
>         -; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v27, 12, s1
>         -; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v16, 12, s5
>         -; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v17, 12, s9
>         -; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v19, 12, s7
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v8, v12, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v9, v13, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v11, v15, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v12, v16, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v13, v17, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v15, v19, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v10, v27, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v14, v31, v2
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v6, v23, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v17, 12, s8
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v13, v13, v2
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v14, v23, v2
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v23, v27, v10
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v12, v12, v2
>         +; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v31, 12, s5
>         +; GFX10-DL-NEXT:    v_lshlrev_b16_e64 v19, 12, s9
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v8, v16, v2
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v9, v17, v2
>         +; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v27, 12, v12
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v12, v31, v2
>         +; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v15, 12, v15
>         +; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v11, 12, v13
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v13, v19, v2
>         +; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v19, 12, v14
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v7, v7, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v6, v23, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v4, v4, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v5, v5, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v9, 12, v9
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v12, 12, v12
>         -; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v13, 12, v13
>          ; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v8, 12, v8
>         -; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v9, 12, v9
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v6, v7, v6 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v10, v27, v2
>          ; GFX10-DL-NEXT:    v_or_b32_sdwa v4, v4, v5 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         -; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v19, 12, v10
>         -; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v15, 12, v15
>         -; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v11, 12, v11
>         -; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v23, 12, v14
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v5, v8, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v7, v9, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v13, v13, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v9, v11, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v12, v12, v2
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v5, v19, v2
>         +; GFX10-DL-NEXT:    v_or_b32_sdwa v6, v7, v6 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    v_ashrrev_i16_e64 v13, 12, v13
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v14, v11, v2
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v15, v15, v2
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v5, v10, v5
>          ; GFX10-DL-NEXT:    v_or_b32_sdwa v4, v6, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v10, v15, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v8, v19, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v11, v23, v2
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v7, v7, v13
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v5, v5, v12
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v6, v9, v10
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v23, v8, v2
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v9, v9, v2
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v7, v13, v2
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v11, v14, v15
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v12, v12, v2
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v5, v5, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v6, v9, v7
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v8, v11, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v7, v23, v12
>          ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v9, 8, v4
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v8, v8, v11
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v7, v7, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v6, v6, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v8, v8, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v5, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v5, v6, v8 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v2, v7, v2 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v2, v5, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v11, v6, s2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_or_b32_sdwa v5, v8, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v7, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
>         +; GFX10-DL-NEXT:    v_or_b32_sdwa v2, v11, v2 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    v_or_b32_e32 v2, v5, v2
>          ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v4, v3
>          ; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v9
>         -; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0
>         src1_sel:BYTE_2
>         +; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v4, 8, v2
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v2
>         +; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v5
>          ; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v4
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v2
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v3, v2
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>
>         Modified: llvm/trunk/test/CodeGen/AMDGPU/idot8u.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/idot8u.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/AMDGPU/idot8u.ll (original)
>         +++ llvm/trunk/test/CodeGen/AMDGPU/idot8u.ll Tue Jul 23
>         05:39:08 2019
>         @@ -314,35 +314,34 @@ define amdgpu_kernel void @udot8_acc16(<
>          ; GFX8-NEXT:    s_and_b32 s1, s4, 15
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         +; GFX8-NEXT:    s_bfe_u32 s6, s4, 0x40008
>          ; GFX8-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX8-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40008
>         +; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX8-NEXT:    v_mov_b32_e32 v5, s6
>         +; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x40008
>          ; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         -; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40014
>         -; GFX8-NEXT:    s_bfe_u32 s12, s4, 0x40018
>         -; GFX8-NEXT:    s_lshr_b32 s14, s4, 28
>         -; GFX8-NEXT:    s_bfe_u32 s4, s4, 0x4000c
>         -; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x40008
>         -; GFX8-NEXT:    v_mov_b32_e32 v5, s5
>         -; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x4000c
>         -; GFX8-NEXT:    v_mov_b32_e32 v6, s4
>         -; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40010
>         +; GFX8-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40014
>          ; GFX8-NEXT:    v_mov_b32_e32 v7, s8
>         -; GFX8-NEXT:    s_bfe_u32 s11, s2, 0x40014
>         -; GFX8-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX8-NEXT:    s_bfe_u32 s13, s2, 0x40018
>         -; GFX8-NEXT:    v_mov_b32_e32 v9, s12
>         +; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX8-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40018
>         +; GFX8-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX8-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX8-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX8-NEXT:    v_and_b32_e32 v2, 0xffff, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
>         -; GFX8-NEXT:    v_mov_b32_e32 v3, s14
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>         +; GFX8-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX8-NEXT:    flat_store_short v[0:1], v2
>          ; GFX8-NEXT:    s_endpgm
>         @@ -362,35 +361,34 @@ define amdgpu_kernel void @udot8_acc16(<
>          ; GFX9-NEXT:    s_and_b32 s1, s4, 15
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX9-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         +; GFX9-NEXT:    s_bfe_u32 s6, s4, 0x40008
>          ; GFX9-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX9-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX9-NEXT:    s_bfe_u32 s5, s4, 0x40008
>         +; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX9-NEXT:    v_mov_b32_e32 v5, s6
>         +; GFX9-NEXT:    s_bfe_u32 s5, s2, 0x40008
>          ; GFX9-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         -; GFX9-NEXT:    s_bfe_u32 s10, s4, 0x40014
>         -; GFX9-NEXT:    s_bfe_u32 s12, s4, 0x40018
>         -; GFX9-NEXT:    s_lshr_b32 s14, s4, 28
>         -; GFX9-NEXT:    s_bfe_u32 s4, s4, 0x4000c
>         -; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x40008
>         -; GFX9-NEXT:    v_mov_b32_e32 v5, s5
>         -; GFX9-NEXT:    s_bfe_u32 s7, s2, 0x4000c
>         -; GFX9-NEXT:    v_mov_b32_e32 v6, s4
>         -; GFX9-NEXT:    s_bfe_u32 s9, s2, 0x40010
>         +; GFX9-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX9-NEXT:    s_bfe_u32 s9, s4, 0x40014
>          ; GFX9-NEXT:    v_mov_b32_e32 v7, s8
>         -; GFX9-NEXT:    s_bfe_u32 s11, s2, 0x40014
>         -; GFX9-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX9-NEXT:    s_bfe_u32 s13, s2, 0x40018
>         -; GFX9-NEXT:    v_mov_b32_e32 v9, s12
>         +; GFX9-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX9-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX9-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX9-NEXT:    s_bfe_u32 s9, s2, 0x40018
>         +; GFX9-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX9-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX9-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
>         -; GFX9-NEXT:    v_mov_b32_e32 v3, s14
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>         +; GFX9-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX9-NEXT:    global_store_short v[0:1], v2, off
>          ; GFX9-NEXT:    s_endpgm
>         @@ -406,81 +404,26 @@ define amdgpu_kernel void @udot8_acc16(<
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
>          ; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
>          ; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
>         -; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40008
>         -; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         -; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x40014
>         -; GFX9-DL-NEXT:    s_bfe_u32 s12, s4, 0x40018
>         -; GFX9-DL-NEXT:    s_lshr_b32 s14, s4, 28
>         -; GFX9-DL-NEXT:    s_bfe_u32 s4, s4, 0x4000c
>         -; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40008
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x4000c
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s4
>         -; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40010
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s8
>         -; GFX9-DL-NEXT:    s_bfe_u32 s11, s2, 0x40014
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX9-DL-NEXT:    s_bfe_u32 s13, s2, 0x40018
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s12
>         -; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v2, 0xffff, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s14
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>         +; GFX9-DL-NEXT:    v_dot8_u32_u4 v2, s2, v3, v2
>          ; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
>          ; GFX9-DL-NEXT:    s_endpgm
>          ;
>          ; GFX10-DL-LABEL: udot8_acc16:
>          ; GFX10-DL:       ; %bb.0: ; %entry
>         -; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
>         -; GFX10-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
>         +; GFX10-DL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x34
>          ; GFX10-DL-NEXT:    ; implicit-def: $vcc_hi
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
>         -; GFX10-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
>         -; GFX10-DL-NEXT:    v_mov_b32_e32 v0, s0
>         -; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s1
>         +; GFX10-DL-NEXT:    v_mov_b32_e32 v0, s4
>         +; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s5
>         +; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
>          ; GFX10-DL-NEXT:    global_load_ushort v2, v[0:1], off
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_and_b32 s0, s2, 15
>         -; GFX10-DL-NEXT:    s_and_b32 s1, s4, 15
>         -; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40004
>         -; GFX10-DL-NEXT:    s_bfe_u32 s6, s4, 0x40004
>         -; GFX10-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
>         -; GFX10-DL-NEXT:    s_bfe_u32 s8, s4, 0x40008
>         -; GFX10-DL-NEXT:    s_bfe_u32 s9, s2, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s10, s4, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s11, s2, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s12, s4, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s13, s2, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_u32 s14, s4, 0x40014
>         -; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
>         -; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40018
>         -; GFX10-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         -; GFX10-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s5, s6, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v2, 0xffff, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s7, s8, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s9, s10, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s11, s12, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s13, s14, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s2, s4, v2
>         +; GFX10-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
>         +; GFX10-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
>         +; GFX10-DL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
>         +; GFX10-DL-NEXT:    v_dot8_u32_u4 v2, s0, s1, v2
>          ; GFX10-DL-NEXT:    global_store_short v[0:1], v2, off
>          ; GFX10-DL-NEXT:    s_endpgm
>                                                 <8 x i4> addrspace(1)*
>         %src2,
>         @@ -616,35 +559,34 @@ define amdgpu_kernel void @udot8_acc8(<8
>          ; GFX8-NEXT:    s_and_b32 s1, s4, 15
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         +; GFX8-NEXT:    s_bfe_u32 s6, s4, 0x40008
>          ; GFX8-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX8-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40008
>         +; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX8-NEXT:    v_mov_b32_e32 v5, s6
>         +; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x40008
>          ; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         -; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40014
>         -; GFX8-NEXT:    s_bfe_u32 s12, s4, 0x40018
>         -; GFX8-NEXT:    s_lshr_b32 s14, s4, 28
>         -; GFX8-NEXT:    s_bfe_u32 s4, s4, 0x4000c
>         -; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x40008
>         -; GFX8-NEXT:    v_mov_b32_e32 v5, s5
>         -; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x4000c
>         -; GFX8-NEXT:    v_mov_b32_e32 v6, s4
>         -; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40010
>         +; GFX8-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40014
>          ; GFX8-NEXT:    v_mov_b32_e32 v7, s8
>         -; GFX8-NEXT:    s_bfe_u32 s11, s2, 0x40014
>         -; GFX8-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX8-NEXT:    s_bfe_u32 s13, s2, 0x40018
>         -; GFX8-NEXT:    v_mov_b32_e32 v9, s12
>         +; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX8-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40018
>         +; GFX8-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX8-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX8-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX8-NEXT:    v_and_b32_e32 v2, 0xff, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
>         -; GFX8-NEXT:    v_mov_b32_e32 v3, s14
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>         +; GFX8-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX8-NEXT:    flat_store_byte v[0:1], v2
>          ; GFX8-NEXT:    s_endpgm
>         @@ -664,35 +606,34 @@ define amdgpu_kernel void @udot8_acc8(<8
>          ; GFX9-NEXT:    s_and_b32 s1, s4, 15
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX9-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         +; GFX9-NEXT:    s_bfe_u32 s6, s4, 0x40008
>          ; GFX9-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX9-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX9-NEXT:    s_bfe_u32 s5, s4, 0x40008
>         +; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX9-NEXT:    v_mov_b32_e32 v5, s6
>         +; GFX9-NEXT:    s_bfe_u32 s5, s2, 0x40008
>          ; GFX9-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         -; GFX9-NEXT:    s_bfe_u32 s10, s4, 0x40014
>         -; GFX9-NEXT:    s_bfe_u32 s12, s4, 0x40018
>         -; GFX9-NEXT:    s_lshr_b32 s14, s4, 28
>         -; GFX9-NEXT:    s_bfe_u32 s4, s4, 0x4000c
>         -; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x40008
>         -; GFX9-NEXT:    v_mov_b32_e32 v5, s5
>         -; GFX9-NEXT:    s_bfe_u32 s7, s2, 0x4000c
>         -; GFX9-NEXT:    v_mov_b32_e32 v6, s4
>         -; GFX9-NEXT:    s_bfe_u32 s9, s2, 0x40010
>         +; GFX9-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX9-NEXT:    s_bfe_u32 s9, s4, 0x40014
>          ; GFX9-NEXT:    v_mov_b32_e32 v7, s8
>         -; GFX9-NEXT:    s_bfe_u32 s11, s2, 0x40014
>         -; GFX9-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX9-NEXT:    s_bfe_u32 s13, s2, 0x40018
>         -; GFX9-NEXT:    v_mov_b32_e32 v9, s12
>         +; GFX9-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX9-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX9-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX9-NEXT:    s_bfe_u32 s9, s2, 0x40018
>         +; GFX9-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX9-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX9-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-NEXT:    v_and_b32_e32 v2, 0xff, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
>         -; GFX9-NEXT:    v_mov_b32_e32 v3, s14
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>         +; GFX9-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX9-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX9-NEXT:    s_endpgm
>         @@ -708,81 +649,26 @@ define amdgpu_kernel void @udot8_acc8(<8
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
>          ; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
>          ; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
>         -; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40008
>         -; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         -; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x40014
>         -; GFX9-DL-NEXT:    s_bfe_u32 s12, s4, 0x40018
>         -; GFX9-DL-NEXT:    s_lshr_b32 s14, s4, 28
>         -; GFX9-DL-NEXT:    s_bfe_u32 s4, s4, 0x4000c
>         -; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40008
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x4000c
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s4
>         -; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40010
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s8
>         -; GFX9-DL-NEXT:    s_bfe_u32 s11, s2, 0x40014
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX9-DL-NEXT:    s_bfe_u32 s13, s2, 0x40018
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s12
>         -; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v2, 0xff, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s14
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>         +; GFX9-DL-NEXT:    v_dot8_u32_u4 v2, s2, v3, v2
>          ; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX9-DL-NEXT:    s_endpgm
>          ;
>          ; GFX10-DL-LABEL: udot8_acc8:
>          ; GFX10-DL:       ; %bb.0: ; %entry
>         -; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
>         -; GFX10-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
>         +; GFX10-DL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x34
>          ; GFX10-DL-NEXT:    ; implicit-def: $vcc_hi
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
>         -; GFX10-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
>         -; GFX10-DL-NEXT:    v_mov_b32_e32 v0, s0
>         -; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s1
>         +; GFX10-DL-NEXT:    v_mov_b32_e32 v0, s4
>         +; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s5
>         +; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
>          ; GFX10-DL-NEXT:    global_load_ubyte v2, v[0:1], off
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_and_b32 s0, s2, 15
>         -; GFX10-DL-NEXT:    s_and_b32 s1, s4, 15
>         -; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40004
>         -; GFX10-DL-NEXT:    s_bfe_u32 s6, s4, 0x40004
>         -; GFX10-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
>         -; GFX10-DL-NEXT:    s_bfe_u32 s8, s4, 0x40008
>         -; GFX10-DL-NEXT:    s_bfe_u32 s9, s2, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s10, s4, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s11, s2, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s12, s4, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s13, s2, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_u32 s14, s4, 0x40014
>         -; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
>         -; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40018
>         -; GFX10-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         -; GFX10-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s5, s6, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v2, 0xff, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s7, s8, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s9, s10, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s11, s12, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s13, s14, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s2, s4, v2
>         +; GFX10-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
>         +; GFX10-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
>         +; GFX10-DL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
>         +; GFX10-DL-NEXT:    v_dot8_u32_u4 v2, s0, s1, v2
>          ; GFX10-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX10-DL-NEXT:    s_endpgm
>                                                <8 x i4> addrspace(1)*
>         %src2,
>         @@ -920,35 +806,32 @@ define amdgpu_kernel void @udot8_acc4(<8
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40004
>          ; GFX8-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>          ; GFX8-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX8-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         +; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX8-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX8-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX8-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX8-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX8-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX8-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX8-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX8-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX8-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX8-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX8-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX8-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         +; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         +; GFX8-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40014
>         +; GFX8-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX8-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40018
>          ; GFX8-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX8-NEXT:    v_mov_b32_e32 v9, s13
>         +; GFX8-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX8-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX8-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v5, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX8-NEXT:    v_and_b32_e32 v2, 15, v2
>         @@ -971,35 +854,32 @@ define amdgpu_kernel void @udot8_acc4(<8
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX9-NEXT:    s_bfe_u32 s5, s4, 0x40004
>          ; GFX9-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>          ; GFX9-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX9-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         +; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX9-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX9-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX9-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX9-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX9-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX9-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX9-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX9-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX9-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX9-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX9-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX9-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX9-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX9-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         +; GFX9-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         +; GFX9-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX9-NEXT:    s_bfe_u32 s9, s4, 0x40014
>         +; GFX9-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX9-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX9-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX9-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX9-NEXT:    s_bfe_u32 s9, s2, 0x40018
>          ; GFX9-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX9-NEXT:    v_mov_b32_e32 v9, s13
>         +; GFX9-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX9-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
>         @@ -1017,86 +897,27 @@ define amdgpu_kernel void @udot8_acc4(<8
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
>          ; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
>          ; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
>         -; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX9-DL-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX9-DL-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX9-DL-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX9-DL-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX9-DL-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         -; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s13
>         -; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         -; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>         +; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>         +; GFX9-DL-NEXT:    v_dot8_u32_u4 v2, s2, v3, v2
>          ; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>          ; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX9-DL-NEXT:    s_endpgm
>          ;
>          ; GFX10-DL-LABEL: udot8_acc4:
>          ; GFX10-DL:       ; %bb.0: ; %entry
>         -; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
>         -; GFX10-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
>         +; GFX10-DL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x34
>          ; GFX10-DL-NEXT:    ; implicit-def: $vcc_hi
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
>         -; GFX10-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
>         -; GFX10-DL-NEXT:    v_mov_b32_e32 v0, s0
>         -; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s1
>         +; GFX10-DL-NEXT:    v_mov_b32_e32 v0, s4
>         +; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s5
>         +; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
>          ; GFX10-DL-NEXT:    global_load_ubyte v2, v[0:1], off
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_and_b32 s0, s2, 15
>         -; GFX10-DL-NEXT:    s_and_b32 s1, s4, 15
>         -; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40004
>         -; GFX10-DL-NEXT:    s_bfe_u32 s6, s4, 0x40004
>         -; GFX10-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
>         -; GFX10-DL-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s9, s4, 0x40008
>         -; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s4, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40010
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s5, s6, v2
>         -; GFX10-DL-NEXT:    v_mul_u32_u24_e64 v3, s8, s0
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_u32 s6, s4, 0x40014
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s7, s9, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v3, 15, v3
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v2, v2, v3
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
>         -; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40018
>         -; GFX10-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         -; GFX10-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s5, s6, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s2, s4, v2
>         +; GFX10-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
>         +; GFX10-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
>         +; GFX10-DL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
>         +; GFX10-DL-NEXT:    v_dot8_u32_u4 v2, s0, s1, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>          ; GFX10-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX10-DL-NEXT:    s_endpgm
>         @@ -1219,35 +1040,32 @@ define amdgpu_kernel void @udot8_Commuta
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40004
>          ; GFX8-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>          ; GFX8-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX8-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         +; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX8-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX8-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX8-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX8-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX8-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX8-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX8-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX8-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX8-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX8-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX8-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX8-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         +; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         +; GFX8-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40014
>         +; GFX8-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX8-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40018
>          ; GFX8-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX8-NEXT:    v_mov_b32_e32 v9, s13
>         +; GFX8-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX8-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX8-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v5
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX8-NEXT:    v_and_b32_e32 v2, 15, v2
>         @@ -1270,35 +1088,32 @@ define amdgpu_kernel void @udot8_Commuta
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX9-NEXT:    s_bfe_u32 s5, s4, 0x40004
>          ; GFX9-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>          ; GFX9-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX9-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         +; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX9-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX9-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX9-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX9-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX9-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX9-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX9-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX9-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX9-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX9-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX9-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX9-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX9-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX9-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         +; GFX9-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         +; GFX9-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX9-NEXT:    s_bfe_u32 s9, s4, 0x40014
>         +; GFX9-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX9-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX9-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX9-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX9-NEXT:    s_bfe_u32 s9, s2, 0x40018
>          ; GFX9-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX9-NEXT:    v_mov_b32_e32 v9, s13
>         +; GFX9-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX9-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX9-NEXT:    v_add_u32_e32 v2, v5, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
>         @@ -1321,35 +1136,32 @@ define amdgpu_kernel void @udot8_Commuta
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
>          ; GFX9-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         +; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX9-DL-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX9-DL-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX9-DL-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX9-DL-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX9-DL-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         +; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x40014
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX9-DL-NEXT:    s_bfe_u32 s9, s2, 0x40018
>          ; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s13
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX9-DL-NEXT:    v_add_u32_e32 v2, v5, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>         +; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>         @@ -1373,27 +1185,24 @@ define amdgpu_kernel void @udot8_Commuta
>          ; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40004
>          ; GFX10-DL-NEXT:    s_bfe_u32 s6, s4, 0x40004
>          ; GFX10-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
>         -; GFX10-DL-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         +; GFX10-DL-NEXT:    s_bfe_u32 s8, s4, 0x40008
>         +; GFX10-DL-NEXT:    s_bfe_u32 s9, s2, 0x4000c
>         +; GFX10-DL-NEXT:    s_bfe_u32 s10, s4, 0x4000c
>         +; GFX10-DL-NEXT:    s_bfe_u32 s11, s2, 0x40010
>         +; GFX10-DL-NEXT:    s_bfe_u32 s12, s4, 0x40010
>         +; GFX10-DL-NEXT:    s_bfe_u32 s13, s2, 0x40014
>         +; GFX10-DL-NEXT:    s_bfe_u32 s14, s4, 0x40014
>          ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s4, 0x40008
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s5, s6, v2
>         -; GFX10-DL-NEXT:    v_mul_u32_u24_e64 v3, s8, s1
>         -; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_u32 s6, s4, 0x40014
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s7, s0, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v3, 15, v3
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40010
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v2, v3, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>          ; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
>          ; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40018
>          ; GFX10-DL-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX10-DL-NEXT:    s_lshr_b32 s4, s4, 28
>          ; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s5, s6, v2
>         +; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s7, s8, v2
>         +; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s9, s10, v2
>         +; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s11, s12, v2
>         +; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s13, s14, v2
>          ; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>          ; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s2, s4, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>         @@ -1987,53 +1796,43 @@ define amdgpu_kernel void @udot8_acc16_v
>          ; GFX7-NEXT:    s_bfe_u32 s20, s1, 0x4000c
>          ; GFX7-NEXT:    v_mov_b32_e32 v2, s20
>          ; GFX7-NEXT:    v_mov_b32_e32 v4, s18
>         -; GFX7-NEXT:    s_bfe_u32 s14, s1, 0x40014
>         -; GFX7-NEXT:    s_bfe_u32 s15, s1, 0x40010
>         -; GFX7-NEXT:    s_lshr_b32 s16, s1, 28
>         -; GFX7-NEXT:    s_bfe_u32 s17, s1, 0x40018
>         +; GFX7-NEXT:    s_bfe_u32 s15, s1, 0x40018
>         +; GFX7-NEXT:    s_bfe_u32 s16, s1, 0x40014
>         +; GFX7-NEXT:    s_bfe_u32 s17, s1, 0x40010
>          ; GFX7-NEXT:    s_and_b32 s19, s1, 15
>         +; GFX7-NEXT:    s_lshr_b32 s14, s1, 28
>          ; GFX7-NEXT:    s_bfe_u32 s1, s1, 0x40008
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s13, v2
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v4, s11, v4
>         -; GFX7-NEXT:    s_bfe_u32 s2, s0, 0x40014
>         -; GFX7-NEXT:    s_bfe_u32 s8, s0, 0x40010
>         -; GFX7-NEXT:    s_lshr_b32 s9, s0, 28
>         -; GFX7-NEXT:    v_mov_b32_e32 v6, s16
>         -; GFX7-NEXT:    s_bfe_u32 s10, s0, 0x40018
>         +; GFX7-NEXT:    s_lshr_b32 s2, s0, 28
>         +; GFX7-NEXT:    s_bfe_u32 s8, s0, 0x40018
>         +; GFX7-NEXT:    s_bfe_u32 s9, s0, 0x40014
>         +; GFX7-NEXT:    s_bfe_u32 s10, s0, 0x40010
>          ; GFX7-NEXT:    s_and_b32 s12, s0, 15
>          ; GFX7-NEXT:    v_mov_b32_e32 v3, s19
>          ; GFX7-NEXT:    s_bfe_u32 s0, s0, 0x40008
>          ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
>         -; GFX7-NEXT:    v_mov_b32_e32 v5, s17
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v6, s9, v6
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v1, s0, v1
>         +; GFX7-NEXT:    v_mul_u32_u24_e32 v8, s0, v1
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v3, s12, v3
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
>         -; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
>         -; GFX7-NEXT:    v_or_b32_e32 v2, v3, v4
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v5, s10, v5
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
>         -; GFX7-NEXT:    v_mov_b32_e32 v8, s14
>         -; GFX7-NEXT:    v_or_b32_e32 v3, v5, v6
>         -; GFX7-NEXT:    v_alignbit_b32 v5, v1, v2, 16
>         +; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
>         +; GFX7-NEXT:    v_or_b32_e32 v2, v8, v2
>         +; GFX7-NEXT:    v_alignbit_b32 v4, v2, v3, 16
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
>         +; GFX7-NEXT:    v_mov_b32_e32 v5, s17
>         +; GFX7-NEXT:    v_mov_b32_e32 v6, s16
>          ; GFX7-NEXT:    v_mov_b32_e32 v7, s15
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v8, s2, v8
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v7, s8, v7
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 16, v8
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
>         -; GFX7-NEXT:    v_or_b32_e32 v4, v7, v8
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v4
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v3
>          ; GFX7-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v5, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v6, v0
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v3
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v7, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v8, v0
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v5, v0
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v6, v0
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v7, v0
>         +; GFX7-NEXT:    v_mov_b32_e32 v1, s14
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s2, v1, v0
>          ; GFX7-NEXT:    buffer_store_short v0, off, s[4:7], 0
>          ; GFX7-NEXT:    s_endpgm
>          ;
>         @@ -2052,35 +1851,34 @@ define amdgpu_kernel void @udot8_acc16_v
>          ; GFX8-NEXT:    s_and_b32 s1, s4, 15
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         +; GFX8-NEXT:    s_bfe_u32 s6, s4, 0x40008
>          ; GFX8-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX8-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40008
>         +; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX8-NEXT:    v_mov_b32_e32 v5, s6
>         +; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x40008
>          ; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         -; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40014
>         -; GFX8-NEXT:    s_bfe_u32 s12, s4, 0x40018
>         -; GFX8-NEXT:    s_lshr_b32 s14, s4, 28
>         -; GFX8-NEXT:    s_bfe_u32 s4, s4, 0x4000c
>         -; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x40008
>         -; GFX8-NEXT:    v_mov_b32_e32 v5, s5
>         -; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x4000c
>         -; GFX8-NEXT:    v_mov_b32_e32 v6, s4
>         -; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40010
>         +; GFX8-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40014
>          ; GFX8-NEXT:    v_mov_b32_e32 v7, s8
>         -; GFX8-NEXT:    s_bfe_u32 s11, s2, 0x40014
>         -; GFX8-NEXT:    v_mov_b32_e32 v8, s10
>         -; GFX8-NEXT:    s_bfe_u32 s13, s2, 0x40018
>         -; GFX8-NEXT:    v_mov_b32_e32 v9, s12
>         +; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX8-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40018
>         +; GFX8-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX8-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX8-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX8-NEXT:    v_and_b32_e32 v2, 0xffff, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v6, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v7, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s11, v8, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s13, v9, v2
>         -; GFX8-NEXT:    v_mov_b32_e32 v3, s14
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>         +; GFX8-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX8-NEXT:    flat_store_short v[0:1], v2
>          ; GFX8-NEXT:    s_endpgm
>         @@ -2131,7 +1929,7 @@ define amdgpu_kernel void @udot8_acc16_v
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:BYTE_0
>         +; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         @@ -2186,7 +1984,7 @@ define amdgpu_kernel void @udot8_acc16_v
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:BYTE_0
>         +; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         @@ -2237,7 +2035,7 @@ define amdgpu_kernel void @udot8_acc16_v
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v3
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v3, s1, s5
>          ; GFX10-DL-NEXT:    s_pack_ll_b32_b16 s1, s6, s4
>         -; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0
>         src1_sel:BYTE_0
>         +; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v2, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX10-DL-NEXT:    v_pk_mul_lo_u16 v4, s0, s1
>          ; GFX10-DL-NEXT:    v_add_nc_u32_e32 v2, v2, v3
>         @@ -2293,64 +2091,64 @@ define amdgpu_kernel void @udot8_acc8_ve
>          ; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
>          ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
>          ; GFX7-NEXT:    s_bfe_u32 s2, s0, 0x4000c
>         -; GFX7-NEXT:    s_lshr_b32 s11, s0, 28
>         +; GFX7-NEXT:    s_bfe_u32 s9, s0, 0x40004
>          ; GFX7-NEXT:    s_bfe_u32 s14, s1, 0x4000c
>         +; GFX7-NEXT:    s_bfe_u32 s16, s1, 0x40004
>          ; GFX7-NEXT:    s_lshr_b32 s18, s1, 28
>         -; GFX7-NEXT:    s_bfe_u32 s20, s1, 0x40014
>         -; GFX7-NEXT:    v_mov_b32_e32 v4, s18
>         +; GFX7-NEXT:    v_mov_b32_e32 v6, s16
>          ; GFX7-NEXT:    v_mov_b32_e32 v8, s14
>          ; GFX7-NEXT:    s_bfe_u32 s15, s1, 0x40008
>         -; GFX7-NEXT:    s_bfe_u32 s16, s1, 0x40004
>          ; GFX7-NEXT:    s_and_b32 s17, s1, 15
>          ; GFX7-NEXT:    s_bfe_u32 s19, s1, 0x40018
>         -; GFX7-NEXT:    s_bfe_u32 s1, s1, 0x40010
>         -; GFX7-NEXT:    s_bfe_u32 s13, s0, 0x40014
>         -; GFX7-NEXT:    v_mov_b32_e32 v2, s20
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s13, v2
>         +; GFX7-NEXT:    s_bfe_u32 s20, s1, 0x40014
>         +; GFX7-NEXT:    s_lshr_b32 s11, s0, 28
>         +; GFX7-NEXT:    v_mov_b32_e32 v4, s18
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v4, s11, v4
>         +; GFX7-NEXT:    v_mul_u32_u24_e32 v6, s9, v6
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v8, s2, v8
>         +; GFX7-NEXT:    s_bfe_u32 s1, s1, 0x40010
>          ; GFX7-NEXT:    s_bfe_u32 s8, s0, 0x40008
>          ; GFX7-NEXT:    v_mov_b32_e32 v7, s15
>         -; GFX7-NEXT:    s_bfe_u32 s9, s0, 0x40004
>         -; GFX7-NEXT:    v_mov_b32_e32 v6, s16
>          ; GFX7-NEXT:    s_and_b32 s10, s0, 15
>         +; GFX7-NEXT:    v_mov_b32_e32 v5, s17
>          ; GFX7-NEXT:    s_bfe_u32 s12, s0, 0x40018
>          ; GFX7-NEXT:    v_mov_b32_e32 v3, s19
>         +; GFX7-NEXT:    s_bfe_u32 s13, s0, 0x40014
>         +; GFX7-NEXT:    v_mov_b32_e32 v2, s20
>         +; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s13, v2
>          ; GFX7-NEXT:    s_bfe_u32 s0, s0, 0x40010
>          ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
>         -; GFX7-NEXT:    v_mov_b32_e32 v5, s17
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v6, s9, v6
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v1, s0, v1
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v3, s12, v3
>         -; GFX7-NEXT:    v_mul_u32_u24_e32 v7, s8, v7
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 8, v8
>         -; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
>         -; GFX7-NEXT:    v_or_b32_e32 v2, v3, v4
>         -; GFX7-NEXT:    v_or_b32_e32 v4, v7, v8
>          ; GFX7-NEXT:    v_mul_u32_u24_e32 v5, s10, v5
>         +; GFX7-NEXT:    v_mul_u32_u24_e32 v7, s8, v7
>          ; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 8, v6
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
>         -; GFX7-NEXT:    v_or_b32_e32 v3, v5, v6
>         -; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
>         -; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
>         -; GFX7-NEXT:    v_or_b32_e32 v2, v3, v4
>         -; GFX7-NEXT:    v_alignbit_b32 v3, v1, v2, 8
>         -; GFX7-NEXT:    v_alignbit_b32 v4, v1, v2, 16
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 24, v2
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
>         -; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 24, v1
>         +; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 8, v8
>         +; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
>         +; GFX7-NEXT:    v_or_b32_e32 v4, v5, v6
>         +; GFX7-NEXT:    v_or_b32_e32 v5, v7, v8
>         +; GFX7-NEXT:    v_mul_u32_u24_e32 v9, s0, v1
>         +; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
>         +; GFX7-NEXT:    v_or_b32_e32 v2, v9, v2
>         +; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
>         +; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
>         +; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
>         +; GFX7-NEXT:    v_or_b32_e32 v3, v4, v5
>         +; GFX7-NEXT:    v_alignbit_b32 v4, v2, v3, 8
>         +; GFX7-NEXT:    v_alignbit_b32 v5, v2, v3, 16
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v3
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v2
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
>         +; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 24, v2
>          ; GFX7-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v3
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v5, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
>          ; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v6, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v7, v0
>         -; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v8, v0
>         +; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v7
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v8
>         +; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
>          ; GFX7-NEXT:    buffer_store_byte v0, off, s[4:7], 0
>          ; GFX7-NEXT:    s_endpgm
>          ;
>         @@ -2383,41 +2181,42 @@ define amdgpu_kernel void @udot8_acc8_ve
>          ; GFX8-NEXT:    v_mul_u32_u24_e32 v3, s10, v3
>          ; GFX8-NEXT:    v_mul_u32_u24_e32 v5, s9, v6
>          ; GFX8-NEXT:    v_mul_u32_u24_sdwa v6, v8, v7 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX8-NEXT:    s_bfe_u32 s0, s2, 0x40014
>         +; GFX8-NEXT:    s_bfe_u32 s1, s4, 0x40014
>         +; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40010
>         +; GFX8-NEXT:    s_lshr_b32 s7, s2, 28
>         +; GFX8-NEXT:    s_lshr_b32 s8, s4, 28
>          ; GFX8-NEXT:    v_or_b32_e32 v5, v5, v6
>          ; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX8-NEXT:    v_or_b32_sdwa v3, v5, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX8-NEXT:    s_bfe_u32 s0, s2, 0x40014
>         -; GFX8-NEXT:    s_lshr_b32 s1, s2, 28
>         -; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40014
>         -; GFX8-NEXT:    s_bfe_u32 s6, s4, 0x40010
>         -; GFX8-NEXT:    s_lshr_b32 s7, s4, 28
>         +; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x40010
>          ; GFX8-NEXT:    s_bfe_u32 s4, s4, 0x40018
>         -; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x40010
>         -; GFX8-NEXT:    s_bfe_u32 s2, s2, 0x40018
>         -; GFX8-NEXT:    v_mov_b32_e32 v6, s4
>         +; GFX8-NEXT:    v_mov_b32_e32 v6, s8
>          ; GFX8-NEXT:    v_mov_b32_e32 v7, s7
>         -; GFX8-NEXT:    v_mov_b32_e32 v8, s1
>         -; GFX8-NEXT:    v_mov_b32_e32 v9, s6
>         -; GFX8-NEXT:    v_mov_b32_e32 v10, s5
>         -; GFX8-NEXT:    v_mov_b32_e32 v11, s0
>         +; GFX8-NEXT:    v_mov_b32_e32 v8, s5
>         +; GFX8-NEXT:    v_mov_b32_e32 v9, s1
>         +; GFX8-NEXT:    v_mov_b32_e32 v10, s0
>         +; GFX8-NEXT:    v_mul_u32_u24_sdwa v6, v7, v6 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX8-NEXT:    v_mul_u32_u24_e32 v7, s6, v8
>         +; GFX8-NEXT:    v_mul_u32_u24_sdwa v8, v10, v9 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX8-NEXT:    s_bfe_u32 s2, s2, 0x40018
>         +; GFX8-NEXT:    v_mov_b32_e32 v4, s4
>          ; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
>         -; GFX8-NEXT:    v_mul_u32_u24_sdwa v7, v8, v7 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX8-NEXT:    v_mul_u32_u24_e32 v6, s2, v6
>         -; GFX8-NEXT:    v_mul_u32_u24_e32 v8, s8, v9
>         -; GFX8-NEXT:    v_mul_u32_u24_sdwa v9, v11, v10
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX8-NEXT:    v_or_b32_e32 v8, v8, v9
>         -; GFX8-NEXT:    v_or_b32_sdwa v6, v6, v7 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX8-NEXT:    v_or_b32_sdwa v4, v8, v6 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v4
>         +; GFX8-NEXT:    v_mul_u32_u24_e32 v4, s2, v4
>         +; GFX8-NEXT:    v_or_b32_e32 v7, v7, v8
>         +; GFX8-NEXT:    v_or_b32_sdwa v4, v4, v6 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX8-NEXT:    v_and_b32_e32 v6, 0xffff, v7
>         +; GFX8-NEXT:    v_or_b32_e32 v4, v6, v4
>         +; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v4
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
>          ; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v5, v2
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
>         +; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
>          ; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         -; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
>          ; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v6
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         -; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         +; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v7, v2
>         +; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v4, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
>         +; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v4, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
>          ; GFX8-NEXT:    flat_store_byte v[0:1], v2
>          ; GFX8-NEXT:    s_endpgm
>          ;
>         @@ -2448,35 +2247,36 @@ define amdgpu_kernel void @udot8_acc8_ve
>          ; GFX9-NEXT:    v_mul_lo_u16_sdwa v4, s8, v4 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX9-NEXT:    v_mul_lo_u16_e32 v5, s9, v5
>          ; GFX9-NEXT:    v_mul_lo_u16_sdwa v6, s10, v6 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-NEXT:    s_bfe_u32 s0, s4, 0x40010
>         +; GFX9-NEXT:    s_bfe_u32 s1, s4, 0x40014
>          ; GFX9-NEXT:    v_or_b32_e32 v3, v3, v4
>          ; GFX9-NEXT:    v_or_b32_sdwa v4, v5, v6 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-NEXT:    s_bfe_u32 s1, s4, 0x40014
>         -; GFX9-NEXT:    s_bfe_u32 s5, s4, 0x40018
>         -; GFX9-NEXT:    s_bfe_u32 s0, s4, 0x40010
>         -; GFX9-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x40018
>          ; GFX9-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x40010
>         -; GFX9-NEXT:    v_mov_b32_e32 v4, s0
>         -; GFX9-NEXT:    s_bfe_u32 s7, s2, 0x40014
>         -; GFX9-NEXT:    v_mov_b32_e32 v5, s1
>         -; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x40018
>         -; GFX9-NEXT:    v_mov_b32_e32 v6, s5
>         -; GFX9-NEXT:    s_lshr_b32 s2, s2, 28
>         -; GFX9-NEXT:    v_mov_b32_e32 v7, s4
>         -; GFX9-NEXT:    v_mul_lo_u16_e32 v4, s6, v4
>         -; GFX9-NEXT:    v_mul_lo_u16_sdwa v5, s7, v5 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-NEXT:    v_mul_lo_u16_e32 v6, s8, v6
>         -; GFX9-NEXT:    v_mul_lo_u16_sdwa v7, s2, v7 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-NEXT:    v_or_b32_e32 v4, v4, v5
>         -; GFX9-NEXT:    v_or_b32_sdwa v5, v6, v7 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-NEXT:    v_or_b32_sdwa v4, v4, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
>         +; GFX9-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX9-NEXT:    v_mov_b32_e32 v7, s0
>         +; GFX9-NEXT:    s_bfe_u32 s5, s2, 0x40010
>         +; GFX9-NEXT:    v_mov_b32_e32 v8, s1
>         +; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x40014
>         +; GFX9-NEXT:    s_bfe_u32 s0, s2, 0x40018
>         +; GFX9-NEXT:    v_mov_b32_e32 v9, s7
>         +; GFX9-NEXT:    s_lshr_b32 s1, s2, 28
>         +; GFX9-NEXT:    v_mov_b32_e32 v10, s4
>         +; GFX9-NEXT:    v_mul_lo_u16_e32 v7, s5, v7
>         +; GFX9-NEXT:    v_mul_lo_u16_sdwa v8, s6, v8 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
>         +; GFX9-NEXT:    v_or_b32_e32 v7, v7, v8
>         +; GFX9-NEXT:    v_mul_lo_u16_e32 v9, s0, v9
>         +; GFX9-NEXT:    v_mul_lo_u16_sdwa v10, s1, v10 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-NEXT:    v_or_b32_sdwa v8, v9, v10 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-NEXT:    v_and_b32_e32 v5, 0xffff, v7
>         +; GFX9-NEXT:    v_or_b32_e32 v4, v5, v8
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
>         -; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
>         -; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
>         +; GFX9-NEXT:    v_add_u32_e32 v2, v2, v6
>         +; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         -; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
>         +; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
>          ; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
>          ; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
>          ; GFX9-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         @@ -2511,35 +2311,36 @@ define amdgpu_kernel void @udot8_acc8_ve
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, s8, v4
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v5, s9, v5
>          ; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v6, s10, v6
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-DL-NEXT:    s_bfe_u32 s0, s4, 0x40010
>         +; GFX9-DL-NEXT:    s_bfe_u32 s1, s4, 0x40014
>          ; GFX9-DL-NEXT:    v_or_b32_e32 v3, v3, v4
>          ; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v5, v6 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-DL-NEXT:    s_bfe_u32 s1, s4, 0x40014
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40018
>         -; GFX9-DL-NEXT:    s_bfe_u32 s0, s4, 0x40010
>         -; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x40018
>          ; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40010
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s0
>         -; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x40014
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s1
>         -; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x40018
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s5
>         -; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s4
>         -; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v4, s6, v4
>         -; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v5, s7, v5
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v6, s8, v6
>         -; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v7, s2, v7
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_or_b32_e32 v4, v4, v5
>         -; GFX9-DL-NEXT:    v_or_b32_sdwa v5, v6, v7 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v4, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v5, 8, v3
>         +; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s0
>         +; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40010
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s1
>         +; GFX9-DL-NEXT:    s_bfe_u32 s6, s2, 0x40014
>         +; GFX9-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s7
>         +; GFX9-DL-NEXT:    s_lshr_b32 s1, s2, 28
>         +; GFX9-DL-NEXT:    v_mov_b32_e32 v10, s4
>         +; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v7, s5, v7
>         +; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v8, s6, v8
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
>         +; GFX9-DL-NEXT:    v_or_b32_e32 v7, v7, v8
>         +; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v9, s0, v9
>         +; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v10, s1, v10
>         dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_or_b32_sdwa v8, v9, v10 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX9-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v7
>         +; GFX9-DL-NEXT:    v_or_b32_e32 v4, v5, v8
>          ; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
>         -; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
>         -; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_2
>         +; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v6
>         +; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         -; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
>         +; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
>          ; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v3, 8, v4
>          ; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v3
>          ; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>         @@ -2575,34 +2376,35 @@ define amdgpu_kernel void @udot8_acc8_ve
>          ; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40014
>          ; GFX10-DL-NEXT:    v_and_b32_sdwa v5, v5, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>          ; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v7, s8, s10
>         -; GFX10-DL-NEXT:    s_lshr_b32 s1, s2, 28
>         +; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40014
>          ; GFX10-DL-NEXT:    v_or_b32_sdwa v4, v6, v4 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         -; GFX10-DL-NEXT:    s_lshr_b32 s6, s4, 28
>         -; GFX10-DL-NEXT:    s_bfe_u32 s5, s4, 0x40014
>         +; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40010
>         +; GFX10-DL-NEXT:    s_lshr_b32 s6, s2, 28
>          ; GFX10-DL-NEXT:    v_or_b32_sdwa v5, v7, v5 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         -; GFX10-DL-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s2, s2, 0x40018
>         -; GFX10-DL-NEXT:    s_bfe_u32 s4, s4, 0x40018
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v7, s0, s1
>         +; GFX10-DL-NEXT:    s_bfe_u32 s7, s4, 0x40010
>         +; GFX10-DL-NEXT:    s_lshr_b32 s8, s4, 28
>         +; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
>          ; GFX10-DL-NEXT:    v_or_b32_sdwa v4, v4, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v5, s0, s5
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v11, s1, s6
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v8, s7, s8
>         -; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v9, s2, s4
>         -; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v7, 8, v4
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v5, v5, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v11, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v5, v8, v5 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40018
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v6, v7, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v5, s5, s7
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v7, s6, s8
>         +; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v8, 8, v4
>         +; GFX10-DL-NEXT:    v_mul_lo_u16_e64 v9, s0, s1
>         +; GFX10-DL-NEXT:    v_or_b32_sdwa v5, v5, v6 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         +; GFX10-DL-NEXT:    v_and_b32_sdwa v2, v7, v2 dst_sel:BYTE_1
>         dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_and_b32_e32 v5, 0xffff, v5
>          ; GFX10-DL-NEXT:    v_or_b32_sdwa v2, v9, v2 dst_sel:WORD_1
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
>         -; GFX10-DL-NEXT:    v_or_b32_sdwa v2, v5, v2 dst_sel:DWORD
>         dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
>         +; GFX10-DL-NEXT:    v_or_b32_e32 v2, v5, v2
>         +; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
>          ; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v6, v4, v3
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v6, v7
>         -; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0
>         src1_sel:BYTE_2
>         +; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v4, v3
>         +; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v8
>         +; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v4
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>         -; GFX10-DL-NEXT:    v_lshrrev_b32_e32 v4, 8, v2
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v2
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v4
>         +; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v5
>         +; GFX10-DL-NEXT:    v_add_nc_u32_e32 v3, v3, v6
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v3, v3, v2
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
>          ; GFX10-DL-NEXT:    v_add_nc_u32_sdwa v2, v3, v2
>         dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
>          ; GFX10-DL-NEXT:    global_store_byte v[0:1], v2, off
>         @@ -2706,35 +2508,32 @@ define amdgpu_kernel void @udot8_acc4_ve
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX8-NEXT:    s_bfe_u32 s5, s4, 0x40004
>          ; GFX8-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>          ; GFX8-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX8-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         +; GFX8-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX8-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX8-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX8-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX8-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX8-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX8-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX8-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX8-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX8-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX8-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX8-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX8-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         +; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         +; GFX8-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX8-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x40014
>         +; GFX8-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX8-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX8-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX8-NEXT:    s_bfe_u32 s9, s2, 0x40018
>          ; GFX8-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX8-NEXT:    v_mov_b32_e32 v9, s13
>         +; GFX8-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX8-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX8-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX8-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v5, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX8-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX8-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>          ; GFX8-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX8-NEXT:    v_and_b32_e32 v2, 15, v2
>         @@ -2757,35 +2556,32 @@ define amdgpu_kernel void @udot8_acc4_ve
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s1
>          ; GFX9-NEXT:    s_bfe_u32 s5, s4, 0x40004
>          ; GFX9-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>          ; GFX9-NEXT:    v_mov_b32_e32 v4, s5
>          ; GFX9-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         +; GFX9-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         +; GFX9-NEXT:    v_mov_b32_e32 v5, s6
>          ; GFX9-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX9-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX9-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX9-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX9-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX9-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX9-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX9-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX9-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX9-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX9-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX9-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX9-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         +; GFX9-NEXT:    s_bfe_u32 s8, s4, 0x40010
>         +; GFX9-NEXT:    v_mov_b32_e32 v6, s7
>         +; GFX9-NEXT:    s_bfe_u32 s6, s2, 0x4000c
>         +; GFX9-NEXT:    s_bfe_u32 s9, s4, 0x40014
>         +; GFX9-NEXT:    v_mov_b32_e32 v7, s8
>         +; GFX9-NEXT:    s_bfe_u32 s7, s2, 0x40010
>         +; GFX9-NEXT:    s_bfe_u32 s10, s4, 0x40018
>         +; GFX9-NEXT:    v_mov_b32_e32 v8, s9
>         +; GFX9-NEXT:    s_bfe_u32 s8, s2, 0x40014
>         +; GFX9-NEXT:    s_bfe_u32 s9, s2, 0x40018
>          ; GFX9-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX9-NEXT:    v_mov_b32_e32 v9, s13
>         +; GFX9-NEXT:    v_mov_b32_e32 v9, s10
>          ; GFX9-NEXT:    s_lshr_b32 s2, s2, 28
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX9-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s5, v5, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s6, v6, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s7, v7, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s8, v8, v2
>         +; GFX9-NEXT:    v_mad_u32_u24 v2, s9, v9, v2
>          ; GFX9-NEXT:    v_mov_b32_e32 v3, s4
>          ; GFX9-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>          ; GFX9-NEXT:    v_and_b32_e32 v2, 15, v2
>         @@ -2803,86 +2599,27 @@ define amdgpu_kernel void @udot8_acc4_ve
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
>          ; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
>          ; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX9-DL-NEXT:    s_and_b32 s0, s2, 15
>         -; GFX9-DL-NEXT:    s_and_b32 s1, s4, 15
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s4, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_u32 s6, s4, 0x40008
>         -; GFX9-DL-NEXT:    s_bfe_u32 s7, s4, 0x4000c
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s1, s2, 0x40004
>         -; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x40008
>         -; GFX9-DL-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s7
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v6, s6
>         -; GFX9-DL-NEXT:    v_mul_u32_u24_e32 v5, s8, v5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x40010
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v5, 15, v5
>         -; GFX9-DL-NEXT:    s_bfe_u32 s11, s4, 0x40014
>         -; GFX9-DL-NEXT:    s_bfe_u32 s10, s2, 0x40010
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v7, s9
>         -; GFX9-DL-NEXT:    s_bfe_u32 s13, s4, 0x40018
>         -; GFX9-DL-NEXT:    s_bfe_u32 s12, s2, 0x40014
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v8, s11
>         -; GFX9-DL-NEXT:    s_bfe_u32 s14, s2, 0x40018
>         -; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX9-DL-NEXT:    v_mov_b32_e32 v9, s13
>         -; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         -; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s1, v4, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
>         -; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v5
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s10, v7, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s12, v8, v2
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s14, v9, v2
>          ; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
>         -; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
>         +; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
>         +; GFX9-DL-NEXT:    v_dot8_u32_u4 v2, s2, v3, v2
>          ; GFX9-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>          ; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX9-DL-NEXT:    s_endpgm
>          ;
>          ; GFX10-DL-LABEL: udot8_acc4_vecMul:
>          ; GFX10-DL:       ; %bb.0: ; %entry
>         -; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
>         -; GFX10-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
>         +; GFX10-DL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x34
>          ; GFX10-DL-NEXT:    ; implicit-def: $vcc_hi
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
>         -; GFX10-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
>         -; GFX10-DL-NEXT:    v_mov_b32_e32 v0, s0
>         -; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s1
>         +; GFX10-DL-NEXT:    v_mov_b32_e32 v0, s4
>         +; GFX10-DL-NEXT:    v_mov_b32_e32 v1, s5
>         +; GFX10-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
>          ; GFX10-DL-NEXT:    global_load_ubyte v2, v[0:1], off
>          ; GFX10-DL-NEXT:    s_waitcnt lgkmcnt(0)
>         -; GFX10-DL-NEXT:    s_and_b32 s0, s2, 15
>         -; GFX10-DL-NEXT:    s_and_b32 s1, s4, 15
>         -; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40004
>         -; GFX10-DL-NEXT:    s_bfe_u32 s6, s4, 0x40004
>         -; GFX10-DL-NEXT:    s_bfe_u32 s7, s2, 0x40008
>         -; GFX10-DL-NEXT:    s_bfe_u32 s8, s2, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s9, s4, 0x40008
>         -; GFX10-DL-NEXT:    s_waitcnt vmcnt(0)
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s4, 0x4000c
>         -; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40010
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s5, s6, v2
>         -; GFX10-DL-NEXT:    v_mul_u32_u24_e64 v3, s8, s0
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40010
>         -; GFX10-DL-NEXT:    s_bfe_u32 s5, s2, 0x40014
>         -; GFX10-DL-NEXT:    s_bfe_u32 s6, s4, 0x40014
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s7, s9, v2
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v3, 15, v3
>         -; GFX10-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>         -; GFX10-DL-NEXT:    v_add_nc_u32_e32 v2, v2, v3
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    s_bfe_u32 s0, s2, 0x40018
>         -; GFX10-DL-NEXT:    s_bfe_u32 s1, s4, 0x40018
>         -; GFX10-DL-NEXT:    s_lshr_b32 s2, s2, 28
>         -; GFX10-DL-NEXT:    s_lshr_b32 s4, s4, 28
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s5, s6, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s0, s1, v2
>         -; GFX10-DL-NEXT:    v_mad_u32_u24 v2, s2, s4, v2
>         +; GFX10-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
>         +; GFX10-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
>         +; GFX10-DL-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
>         +; GFX10-DL-NEXT:    v_dot8_u32_u4 v2, s0, s1, v2
>          ; GFX10-DL-NEXT:    v_and_b32_e32 v2, 15, v2
>          ; GFX10-DL-NEXT:    global_store_byte v[0:1], v2, off
>          ; GFX10-DL-NEXT:    s_endpgm
>
>         Modified: llvm/trunk/test/CodeGen/AMDGPU/sdiv.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/sdiv.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/AMDGPU/sdiv.ll (original)
>         +++ llvm/trunk/test/CodeGen/AMDGPU/sdiv.ll Tue Jul 23 05:39:08
>         2019
>         @@ -1931,19 +1931,19 @@ define amdgpu_kernel void @v_sdiv_i24(i3
>          ; GCN-NEXT:    s_waitcnt vmcnt(1)
>          ; GCN-NEXT:    v_or_b32_e32 v0, v0, v1
>          ; GCN-NEXT:    s_waitcnt vmcnt(0)
>         -; GCN-NEXT:    v_or_b32_e32 v1, v2, v3
>         -; GCN-NEXT:    v_xor_b32_e32 v2, v0, v1
>         +; GCN-NEXT:    v_or_b32_e32 v2, v2, v3
>         +; GCN-NEXT:    v_xor_b32_e32 v1, v1, v3
>         +; GCN-NEXT:    v_ashrrev_i32_e32 v1, 30, v1
>          ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, v0
>         -; GCN-NEXT:    v_cvt_f32_i32_e32 v1, v1
>         -; GCN-NEXT:    v_ashrrev_i32_e32 v2, 30, v2
>         -; GCN-NEXT:    v_rcp_iflag_f32_e32 v3, v1
>         -; GCN-NEXT:    v_or_b32_e32 v2, 1, v2
>         +; GCN-NEXT:    v_cvt_f32_i32_e32 v2, v2
>         +; GCN-NEXT:    v_or_b32_e32 v1, 1, v1
>         +; GCN-NEXT:    v_rcp_iflag_f32_e32 v3, v2
>          ; GCN-NEXT:    v_mul_f32_e32 v3, v0, v3
>          ; GCN-NEXT:    v_trunc_f32_e32 v3, v3
>         -; GCN-NEXT:    v_mad_f32 v0, -v3, v1, v0
>         +; GCN-NEXT:    v_mad_f32 v0, -v3, v2, v0
>          ; GCN-NEXT:    v_cvt_i32_f32_e32 v3, v3
>         -; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, |v1|
>         -; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v2, vcc
>         +; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, |v2|
>         +; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
>          ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v3
>          ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
>          ; GCN-NEXT:    buffer_store_dword v0, off, s[4:7], 0
>         @@ -1970,21 +1970,21 @@ define amdgpu_kernel void @v_sdiv_i24(i3
>          ; TONGA-NEXT:    s_waitcnt vmcnt(1)
>          ; TONGA-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
>          ; TONGA-NEXT:    v_or_b32_e32 v1, v1, v2
>         -; TONGA-NEXT:    v_cvt_f32_i32_e32 v2, v1
>         +; TONGA-NEXT:    v_cvt_f32_i32_e32 v1, v1
>          ; TONGA-NEXT:    s_waitcnt vmcnt(0)
>         -; TONGA-NEXT:    v_or_b32_e32 v0, v3, v0
>         -; TONGA-NEXT:    v_cvt_f32_i32_e32 v3, v0
>         -; TONGA-NEXT:    v_xor_b32_e32 v0, v0, v1
>         -; TONGA-NEXT:    v_rcp_iflag_f32_e32 v4, v2
>         +; TONGA-NEXT:    v_or_b32_e32 v3, v3, v0
>         +; TONGA-NEXT:    v_cvt_f32_i32_e32 v3, v3
>         +; TONGA-NEXT:    v_xor_b32_e32 v0, v0, v2
>         +; TONGA-NEXT:    v_rcp_iflag_f32_e32 v4, v1
>          ; TONGA-NEXT:    v_ashrrev_i32_e32 v0, 30, v0
>          ; TONGA-NEXT:    v_or_b32_e32 v0, 1, v0
>         -; TONGA-NEXT:    v_mul_f32_e32 v1, v3, v4
>         -; TONGA-NEXT:    v_trunc_f32_e32 v1, v1
>         -; TONGA-NEXT:    v_mad_f32 v3, -v1, v2, v3
>         -; TONGA-NEXT:    v_cvt_i32_f32_e32 v1, v1
>         -; TONGA-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
>         +; TONGA-NEXT:    v_mul_f32_e32 v2, v3, v4
>         +; TONGA-NEXT:    v_trunc_f32_e32 v2, v2
>         +; TONGA-NEXT:    v_mad_f32 v3, -v2, v1, v3
>         +; TONGA-NEXT:    v_cvt_i32_f32_e32 v2, v2
>         +; TONGA-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v1|
>          ; TONGA-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
>         -; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
>         +; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
>          ; TONGA-NEXT:    v_bfe_i32 v0, v0, 0, 24
>          ; TONGA-NEXT:    buffer_store_dword v0, off, s[0:3], 0
>          ; TONGA-NEXT:    s_endpgm
>         @@ -2011,18 +2011,18 @@ define amdgpu_kernel void @v_sdiv_i24(i3
>          ; GFX9-NEXT:    s_waitcnt vmcnt(0)
>          ; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
>          ; GFX9-NEXT:    v_or_b32_e32 v2, v2, v3
>         -; GFX9-NEXT:    v_cvt_f32_i32_e32 v3, v2
>         -; GFX9-NEXT:    v_cvt_f32_i32_e32 v1, v0
>         -; GFX9-NEXT:    v_xor_b32_e32 v0, v0, v2
>         -; GFX9-NEXT:    v_ashrrev_i32_e32 v0, 30, v0
>         -; GFX9-NEXT:    v_rcp_iflag_f32_e32 v4, v3
>         -; GFX9-NEXT:    v_or_b32_e32 v0, 1, v0
>         -; GFX9-NEXT:    v_mul_f32_e32 v2, v1, v4
>         -; GFX9-NEXT:    v_trunc_f32_e32 v2, v2
>         -; GFX9-NEXT:    v_cvt_i32_f32_e32 v4, v2
>         -; GFX9-NEXT:    v_mad_f32 v1, -v2, v3, v1
>         -; GFX9-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v3|
>         -; GFX9-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
>         +; GFX9-NEXT:    v_cvt_f32_i32_e32 v2, v2
>         +; GFX9-NEXT:    v_cvt_f32_i32_e32 v0, v0
>         +; GFX9-NEXT:    v_xor_b32_e32 v1, v1, v3
>         +; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 30, v1
>         +; GFX9-NEXT:    v_rcp_iflag_f32_e32 v4, v2
>         +; GFX9-NEXT:    v_or_b32_e32 v1, 1, v1
>         +; GFX9-NEXT:    v_mul_f32_e32 v3, v0, v4
>         +; GFX9-NEXT:    v_trunc_f32_e32 v3, v3
>         +; GFX9-NEXT:    v_cvt_i32_f32_e32 v4, v3
>         +; GFX9-NEXT:    v_mad_f32 v0, -v3, v2, v0
>         +; GFX9-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, |v2|
>         +; GFX9-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
>          ; GFX9-NEXT:    v_add_u32_e32 v0, v4, v0
>          ; GFX9-NEXT:    v_bfe_i32 v0, v0, 0, 24
>          ; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], 0
>
>         Modified:
>         llvm/trunk/test/CodeGen/SystemZ/store_nonbytesized_vecs.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/SystemZ/store_nonbytesized_vecs.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/SystemZ/store_nonbytesized_vecs.ll
>         (original)
>         +++ llvm/trunk/test/CodeGen/SystemZ/store_nonbytesized_vecs.ll
>         Tue Jul 23 05:39:08 2019
>         @@ -120,15 +120,15 @@ define void @fun2(<8 x i32> %src, <8 x i
>          define void @fun3(<3 x i31>* %src, <3 x i31>* %p)
>          ; CHECK-LABEL: fun3:
>          ; CHECK:       # %bb.0:
>         -; CHECK-NEXT:    llgf %r1, 0(%r2)
>          ; CHECK-NEXT:    llgf %r0, 3(%r2)
>         -; CHECK-NEXT:    sllg %r4, %r1, 62
>         +; CHECK-NEXT:    llgf %r1, 6(%r2)
>         +; CHECK-NEXT:    llgf %r2, 0(%r2)
>         +; CHECK-NEXT:    rosbg %r1, %r0, 0, 32, 31
>         +; CHECK-NEXT:    sllg %r4, %r2, 62
>          ; CHECK-NEXT:    rosbg %r4, %r0, 0, 32, 31
>         -; CHECK-NEXT:    llgf %r0, 6(%r2)
>         -; CHECK-NEXT:    ogr %r0, %r4
>         -; CHECK-NEXT:    st %r0, 8(%r3)
>          ; CHECK-NEXT:    srlg %r0, %r4, 32
>         -; CHECK-NEXT:    sllg %r1, %r1, 30
>         +; CHECK-NEXT:    st %r1, 8(%r3)
>         +; CHECK-NEXT:    sllg %r1, %r2, 30
>          ; CHECK-NEXT:    lr %r1, %r0
>          ; CHECK-NEXT:    nihh %r1, 8191
>          ; CHECK-NEXT:    stg %r1, 0(%r3)
>
>         Modified: llvm/trunk/test/CodeGen/X86/2012-08-07-CmpISelBug.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2012-08-07-CmpISelBug.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/X86/2012-08-07-CmpISelBug.ll
>         (original)
>         +++ llvm/trunk/test/CodeGen/X86/2012-08-07-CmpISelBug.ll Tue
>         Jul 23 05:39:08 2019
>         @@ -8,14 +8,13 @@
>          define void @foo(i8 %arg4, i32 %arg5, i32* %arg14) nounwind {
>          ; CHECK-LABEL: foo:
>          ; CHECK:       ## %bb.0: ## %bb
>         +; CHECK-NEXT:    ## kill: def $edi killed $edi def $rdi
>          ; CHECK-NEXT:    andl $32, %edi
>         -; CHECK-NEXT:    orl $1601159181, %edi ## imm = 0x5F6FC00D
>         -; CHECK-NEXT:    andl %edi, %esi
>         -; CHECK-NEXT:    xorb $-14, %dil
>         -; CHECK-NEXT:    addb $82, %dil
>         -; CHECK-NEXT:    shrl $5, %esi
>         -; CHECK-NEXT:    movzbl %dil, %eax
>         -; CHECK-NEXT:    testb %sil, %sil
>         +; CHECK-NEXT:    leal 13(%rdi), %eax
>         +; CHECK-NEXT:    xorb $-14, %al
>         +; CHECK-NEXT:    addb $82, %al
>         +; CHECK-NEXT:    movzbl %al, %eax
>         +; CHECK-NEXT:    testl %esi, %edi
>          ; CHECK-NEXT:    movl $1, %ecx
>          ; CHECK-NEXT:    cmovnel %eax, %ecx
>          ; CHECK-NEXT:    xorb $81, %cl
>
>         Modified: llvm/trunk/test/CodeGen/X86/vector-fshl-128.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-fshl-128.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/X86/vector-fshl-128.ll (original)
>         +++ llvm/trunk/test/CodeGen/X86/vector-fshl-128.ll Tue Jul 23
>         05:39:08 2019
>         @@ -498,7 +498,6 @@ define <4 x i32> @var_funnnel_v4i32(<4 x
>          define <8 x i16> @var_funnnel_v8i16(<8 x i16> %x, <8 x i16>
>         %y, <8 x i16> %amt) nounwind {
>          ; SSE2-LABEL: var_funnnel_v8i16:
>          ; SSE2:       # %bb.0:
>         -; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
>          ; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
>          ; SSE2-NEXT:    psubw %xmm2, %xmm3
>          ; SSE2-NEXT:    psllw $12, %xmm3
>         @@ -531,6 +530,7 @@ define <8 x i16> @var_funnnel_v8i16(<8 x
>          ; SSE2-NEXT:    pandn %xmm1, %xmm4
>          ; SSE2-NEXT:    psrlw $1, %xmm1
>          ; SSE2-NEXT:    pand %xmm3, %xmm1
>         +; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
>          ; SSE2-NEXT:    pxor %xmm3, %xmm3
>          ; SSE2-NEXT:    movdqa %xmm2, %xmm5
>          ; SSE2-NEXT:    punpckhwd {{.*#+}} xmm5 =
>         xmm5[4],xmm3[4],xmm5[5],xmm3[5],xmm5[6],xmm3[6],xmm5[7],xmm3[7]
>         @@ -768,7 +768,6 @@ define <8 x i16> @var_funnnel_v8i16(<8 x
>          ;
>          ; X32-SSE-LABEL: var_funnnel_v8i16:
>          ; X32-SSE:       # %bb.0:
>         -; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
>          ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 =
>         [16,16,16,16,16,16,16,16]
>          ; X32-SSE-NEXT:    psubw %xmm2, %xmm3
>          ; X32-SSE-NEXT:    psllw $12, %xmm3
>         @@ -801,6 +800,7 @@ define <8 x i16> @var_funnnel_v8i16(<8 x
>          ; X32-SSE-NEXT:    pandn %xmm1, %xmm4
>          ; X32-SSE-NEXT:    psrlw $1, %xmm1
>          ; X32-SSE-NEXT:    pand %xmm3, %xmm1
>         +; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
>          ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
>          ; X32-SSE-NEXT:    movdqa %xmm2, %xmm5
>          ; X32-SSE-NEXT:    punpckhwd {{.*#+}} xmm5 =
>         xmm5[4],xmm3[4],xmm5[5],xmm3[5],xmm5[6],xmm3[6],xmm5[7],xmm3[7]
>
>         Modified: llvm/trunk/test/CodeGen/X86/vector-reduce-mul-widen.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-reduce-mul-widen.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/X86/vector-reduce-mul-widen.ll
>         (original)
>         +++ llvm/trunk/test/CodeGen/X86/vector-reduce-mul-widen.ll Tue
>         Jul 23 05:39:08 2019
>         @@ -1820,17 +1820,17 @@ define i8 @test_v16i8(<16 x i8> %a0) {
>          ; AVX2-LABEL: test_v16i8:
>          ; AVX2:       # %bb.0:
>          ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
>         -; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm0 =
>         xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
>         -; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm1 =
>         xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
>         -; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
>         +; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm0 =
>         xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
>         +; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 =
>         xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
>         +; AVX2-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
>         +; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
>         +; AVX2-NEXT:    vpackuswb %xmm0, %xmm0, %xmm1
>         +; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
>         +; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 =
>         xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
>         +; AVX2-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
>          ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 =
>         [255,255,255,255,255,255,255,255]
>         -; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
>         -; AVX2-NEXT:    vpackuswb %xmm0, %xmm0, %xmm2
>         -; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
>         -; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm2 =
>         xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
>         -; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
>         -; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
>         -; AVX2-NEXT:    vpackuswb %xmm0, %xmm0, %xmm2
>         +; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
>         +; AVX2-NEXT:    vpackuswb %xmm0, %xmm2, %xmm2
>          ; AVX2-NEXT:    vpsrld $16, %xmm2, %xmm2
>          ; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm2 =
>         xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
>          ; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
>         @@ -1840,6 +1840,7 @@ define i8 @test_v16i8(<16 x i8> %a0) {
>          ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
>          ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
>          ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
>         +; AVX2-NEXT:    vzeroupper
>          ; AVX2-NEXT:    retq
>          ;
>          ; AVX512BW-LABEL: test_v16i8:
>
>         Modified: llvm/trunk/test/CodeGen/X86/vector-reduce-mul.ll
>         URL:
>         http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-reduce-mul.ll?rev=366799&r1=366798&r2=366799&view=diff
>         ==============================================================================
>         --- llvm/trunk/test/CodeGen/X86/vector-reduce-mul.ll (original)
>         +++ llvm/trunk/test/CodeGen/X86/vector-reduce-mul.ll Tue Jul
>         23 05:39:08 2019
>         @@ -1792,14 +1792,11 @@ define i8 @test_v16i8(<16 x i8> %a0) {
>          ; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm1 =
>         xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
>          ; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm0 =
>         xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
>          ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
>         -; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 =
>         [255,255,255,255,255,255,255,255]
>         -; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
>         -; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,2,3]
>         -; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
>         -; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
>         -; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
>         -; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
>         -; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
>         +; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
>         +; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
>         +; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
>         +; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 =
>         xmm0[4],zero,xmm0[6],zero,xmm0[4],zero,xmm0[6],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
>         +; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
>          ; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
>          ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
>          ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
>
>
>         _______________________________________________
>         llvm-commits mailing list
>         llvm-commits at lists.llvm.org <mailto:llvm-commits at lists.llvm.org>
>         https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190727/643f6d03/attachment-0001.html>


More information about the llvm-commits mailing list