[AMDGPU] gfx11 WMMA instruction support

[Joe_Nash]

Summary

gfx11 introduces new WMMA (Wave Matrix Multiply-accumulate)
instructions.

[arsenm]

llvm/include/llvm/IR/IntrinsicsAMDGPU.td

1990

Missing the clang changes and tests for this

[Joe_Nash]

Removed clang builtin from intrinsics. The builtins can be added in a later patch.

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

3739–3740

Probably should have a verifier check this, or just rely on 0/non-0

llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp

1708

Can use Register() in place of NoRegister

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

3268

Why skip implicit operands?

[Joe_Nash]

use Register() in place of NoRegister. Add implicit operands in convertToThreeAddress

[foad]

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

3270

I think you could just raise the upper bound of the loop above to E = MI.getNumOperands() instead of adding this extra call?

[Joe_Nash]

better way to copy implicit operands

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

3739–3740

I don't know what to do here. The intrinsic has an i1 field. If you put a non-i1 value that will be reported right? Given that, we are asserting that other parts of ISel haven't transformed this value incorrectly. Also, we do the same thing in selectDotIUVOP3PMods, line 3726. Please let me know what could be done.

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

3739–3740

I'd just invert the check below to != 0. The machine verifier is certainly not enforcing this be 0/-1 for booleans. Practically speaking, this would only come up for hand written MIR

[piotr]

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

3739–3740

I don't know what to do here. The intrinsic has an i1 field. If you put a non-i1 value that will be reported right? Given that, we are asserting that other parts of ISel haven't transformed this value incorrectly. Also, we do the same thing in selectDotIUVOP3PMods, line 3726. Please let me know what could be done.

Indeed, this was strongly inspired by the existing code in selectDotIUVOP3PMods, which handles the intrinsic in a similar way.

[Joe_Nash]

invert op_sel_0 value check

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

3739

No real point to the assert anymore

llvm/include/llvm/IR/IntrinsicsAMDGPU.td

def int_amdgcn_ds_add_gs_reg_rtn :
   Intrinsic<[llvm_anyint_ty], [llvm_i32_ty, llvm_i32_ty],
             [ImmArg<ArgIndex<1>>, IntrHasSideEffects, IntrWillReturn]>;
 
 def int_amdgcn_ds_sub_gs_reg_rtn :
   ClangBuiltin<"__builtin_amdgcn_ds_sub_gs_reg_rtn">,
   Intrinsic<[llvm_anyint_ty], [llvm_i32_ty, llvm_i32_ty],
             [ImmArg<ArgIndex<1>>, IntrHasSideEffects, IntrWillReturn]>;
 
+// WMMA (Wave Matrix Multiply-Accumulate) intrinsics
+//
+// These operations perform a matrix multiplication and accumulation of
+// the form: D = A * B + C .
+
+class AMDGPUWmmaIntrinsic<LLVMType AB, LLVMType CD> :
+  Intrinsic<
+    [CD],               // %D
+    [
+      AB,               // %A
+      AB,               // %B
+      LLVMMatchType<0>, // %C
+    ],
+    [IntrNoMem, IntrConvergent, IntrWillReturn]
+>;
+
+class AMDGPUWmmaIntrinsicOPSEL<LLVMType AB, LLVMType CD> :
+  Intrinsic<
+    [CD],               // %D
+    [
+      AB,               // %A
+      AB,               // %B
+      LLVMMatchType<0>, // %C
+      llvm_i1_ty,       // %high
+    ],
+    [IntrNoMem, IntrConvergent, IntrWillReturn, ImmArg<ArgIndex<3>>]
+>;
+
+class AMDGPUWmmaIntrinsicIU<LLVMType AB, LLVMType CD> :
+  Intrinsic<
+    [CD],               // %D
+    [
+      llvm_i1_ty,       // %A_sign
+      AB,               // %A
+      llvm_i1_ty,       // %B_sign
+      AB,               // %B
+      LLVMMatchType<0>, // %C
+      llvm_i1_ty,       // %clamp
+    ],
+    [IntrNoMem, IntrConvergent, IntrWillReturn, ImmArg<ArgIndex<0>>, ImmArg<ArgIndex<2>>, ImmArg<ArgIndex<5>>]
+>;
+
+def int_amdgcn_wmma_f32_16x16x16_f16   : AMDGPUWmmaIntrinsic<llvm_v8f32_ty, llvm_anyfloat_ty>;
+def int_amdgcn_wmma_f32_16x16x16_bf16  : AMDGPUWmmaIntrinsic<llvm_v8i32_ty, llvm_anyfloat_ty>;
+def int_amdgcn_wmma_f16_16x16x16_f16   : AMDGPUWmmaIntrinsicOPSEL<llvm_v8f32_ty, llvm_anyfloat_ty>;
+def int_amdgcn_wmma_bf16_16x16x16_bf16 : AMDGPUWmmaIntrinsicOPSEL<llvm_v8i32_ty, llvm_anyint_ty>;
+def int_amdgcn_wmma_i32_16x16x16_iu8   : AMDGPUWmmaIntrinsicIU<llvm_v4i32_ty, llvm_anyint_ty>;
+def int_amdgcn_wmma_i32_16x16x16_iu4   : AMDGPUWmmaIntrinsicIU<llvm_v2i32_ty, llvm_anyint_ty>;
+
+
 //===----------------------------------------------------------------------===//
 // Deep learning intrinsics.
 //===----------------------------------------------------------------------===//
 
 // f32 %r = llvm.amdgcn.fdot2(v2f16 %a, v2f16 %b, f32 %c, i1 %clamp)
 //   %r = %a[0] * %b[0] + %a[1] * %b[1] + %c
 def int_amdgcn_fdot2 :
   ClangBuiltin<"__builtin_amdgcn_fdot2">,

llvm/lib/Target/AMDGPU/AMDGPUGISel.td

 def gi_vop3pmodsdot :
     GIComplexOperandMatcher<s32, "selectVOP3PModsDOT">,
     GIComplexPatternEquiv<VOP3PModsDOT>;
 
 def gi_dotiuvop3pmods :
     GIComplexOperandMatcher<s32, "selectDotIUVOP3PMods">,
     GIComplexPatternEquiv<DotIUVOP3PMods>;
 
+def gi_wmmaopselvop3pmods :
+    GIComplexOperandMatcher<s32, "selectWMMAOpSelVOP3PMods">,
+    GIComplexPatternEquiv<WMMAOpSelVOP3PMods>;
+
 def gi_vop3opselmods :
     GIComplexOperandMatcher<s32, "selectVOP3OpSelMods">,
     GIComplexPatternEquiv<VOP3OpSelMods>;
 
 def gi_vinterpmods :
     GIComplexOperandMatcher<s32, "selectVINTERPMods">,
     GIComplexPatternEquiv<VINTERPMods>;
 

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h

private:
   bool SelectVOP3OMods(SDValue In, SDValue &Src, SDValue &Clamp,
                        SDValue &Omod) const;
 
   bool SelectVOP3PMods(SDValue In, SDValue &Src, SDValue &SrcMods,
                        bool IsDOT = false) const;
   bool SelectVOP3PModsDOT(SDValue In, SDValue &Src, SDValue &SrcMods) const;
 
   bool SelectDotIUVOP3PMods(SDValue In, SDValue &Src) const;
+  bool SelectWMMAOpSelVOP3PMods(SDValue In, SDValue &Src) const;
 
   bool SelectVOP3OpSel(SDValue In, SDValue &Src, SDValue &SrcMods) const;
 
   bool SelectVOP3OpSelMods(SDValue In, SDValue &Src, SDValue &SrcMods) const;
   bool SelectVOP3PMadMixModsImpl(SDValue In, SDValue &Src,
                                  unsigned &Mods) const;
   bool SelectVOP3PMadMixMods(SDValue In, SDValue &Src, SDValue &SrcMods) const;
 

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

bool AMDGPUDAGToDAGISel::SelectDotIUVOP3PMods(SDValue In, SDValue &Src) const {
   unsigned SrcSign = C->getAPIntValue().getZExtValue();
   if (SrcSign == 1)
     Mods ^= SISrcMods::NEG;
 
   Src = CurDAG->getTargetConstant(Mods, SDLoc(In), MVT::i32);
   return true;
 }
 
+bool AMDGPUDAGToDAGISel::SelectWMMAOpSelVOP3PMods(SDValue In,
+                                                  SDValue &Src) const {
+  const ConstantSDNode *C = cast<ConstantSDNode>(In);
+  assert(C->getAPIntValue().getBitWidth() == 1 && "expected i1 value");
+
+  unsigned Mods = SISrcMods::OP_SEL_1;
+  unsigned SrcVal = C->getAPIntValue().getZExtValue();
+  if (SrcVal == 1)
+    Mods |= SISrcMods::OP_SEL_0;
+
+  Src = CurDAG->getTargetConstant(Mods, SDLoc(In), MVT::i32);
+  return true;
+}
+
 bool AMDGPUDAGToDAGISel::SelectVOP3OpSel(SDValue In, SDValue &Src,
                                          SDValue &SrcMods) const {
   Src = In;
   // FIXME: Handle op_sel
   SrcMods = CurDAG->getTargetConstant(0, SDLoc(In), MVT::i32);
   return true;
 }
 

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h

private:
 
   InstructionSelector::ComplexRendererFns
   selectVOP3PModsDOT(MachineOperand &Root) const;
 
   InstructionSelector::ComplexRendererFns
   selectDotIUVOP3PMods(MachineOperand &Root) const;
 
   InstructionSelector::ComplexRendererFns
+  selectWMMAOpSelVOP3PMods(MachineOperand &Root) const;
+
+  InstructionSelector::ComplexRendererFns
   selectVOP3OpSelMods(MachineOperand &Root) const;
 
   InstructionSelector::ComplexRendererFns
   selectVINTERPMods(MachineOperand &Root) const;
   InstructionSelector::ComplexRendererFns
   selectVINTERPModsHi(MachineOperand &Root) const;
 
   InstructionSelector::ComplexRendererFns

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

AMDGPUInstructionSelector::selectDotIUVOP3PMods(MachineOperand &Root) const {
   if (Root.getImm() == -1)
     Mods ^= SISrcMods::NEG;
   return {{
       [=](MachineInstrBuilder &MIB) { MIB.addImm(Mods); } // src_mods
   }};
 }
 
 InstructionSelector::ComplexRendererFns
+AMDGPUInstructionSelector::selectWMMAOpSelVOP3PMods(
+    MachineOperand &Root) const {
+  assert((Root.isImm() && (Root.getImm() == -1 || Root.getImm() == 0)) &&
+         "expected i1 value");
+  unsigned Mods = SISrcMods::OP_SEL_1;
+  if (Root.getImm() == -1)
+    Mods |= SISrcMods::OP_SEL_0;
+
+  return {{
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(Mods); } // src_mods
+  }};
+}
+
+InstructionSelector::ComplexRendererFns
 AMDGPUInstructionSelector::selectVOP3Mods_nnan(MachineOperand &Root) const {
   Register Src;
   unsigned Mods;
   std::tie(Src, Mods) = selectVOP3ModsImpl(Root);
   if (!isKnownNeverNaN(Src, *MRI))
     return None;
 
   return {{

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

case AMDGPU::G_INTRINSIC: {
     case Intrinsic::amdgcn_udot4:
     case Intrinsic::amdgcn_sdot8:
     case Intrinsic::amdgcn_udot8:
     case Intrinsic::amdgcn_fdot2_bf16_bf16:
     case Intrinsic::amdgcn_fdot2_f16_f16:
     case Intrinsic::amdgcn_fdot2_f32_bf16:
     case Intrinsic::amdgcn_sudot4:
     case Intrinsic::amdgcn_sudot8:
+    case Intrinsic::amdgcn_wmma_bf16_16x16x16_bf16:
+    case Intrinsic::amdgcn_wmma_f16_16x16x16_f16:
+    case Intrinsic::amdgcn_wmma_f32_16x16x16_bf16:
+    case Intrinsic::amdgcn_wmma_f32_16x16x16_f16:
+    case Intrinsic::amdgcn_wmma_i32_16x16x16_iu4:
+    case Intrinsic::amdgcn_wmma_i32_16x16x16_iu8:
       return getDefaultMappingVOP(MI);
     case Intrinsic::amdgcn_sbfe:
     case Intrinsic::amdgcn_ubfe:
       if (isSALUMapping(MI))
         return getDefaultMappingSOP(MI);
       return getDefaultMappingVOP(MI);
     case Intrinsic::amdgcn_ds_swizzle:
     case Intrinsic::amdgcn_ds_permute:

llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp

do {
 
     Res = tryDecodeInst(DecoderTableGFX964, MI, QW, Address);
     if (Res) break;
 
     Res = tryDecodeInst(DecoderTableGFX1064, MI, QW, Address);
     if (Res) break;
 
     Res = tryDecodeInst(DecoderTableGFX1164, MI, QW, Address);
+    if (Res)
+      break;
+
+    Res = tryDecodeInst(DecoderTableWMMAGFX1164, MI, QW, Address);
   } while (false);
 
   if (Res && (MI.getOpcode() == AMDGPU::V_MAC_F32_e64_vi ||
               MI.getOpcode() == AMDGPU::V_MAC_F32_e64_gfx6_gfx7 ||
               MI.getOpcode() == AMDGPU::V_MAC_F32_e64_gfx10 ||
               MI.getOpcode() == AMDGPU::V_MAC_LEGACY_F32_e64_gfx6_gfx7 ||
               MI.getOpcode() == AMDGPU::V_MAC_LEGACY_F32_e64_gfx10 ||
               MI.getOpcode() == AMDGPU::V_MAC_F16_e64_vi ||

llvm/lib/Target/AMDGPU/GCNHazardRecognizer.h

private:
   bool fixVMEMtoScalarWriteHazards(MachineInstr *MI);
   bool fixSMEMtoVectorWriteHazards(MachineInstr *MI);
   bool fixVcmpxExecWARHazard(MachineInstr *MI);
   bool fixLdsBranchVmemWARHazard(MachineInstr *MI);
   bool fixLdsDirectVALUHazard(MachineInstr *MI);
   bool fixLdsDirectVMEMHazard(MachineInstr *MI);
   bool fixVALUPartialForwardingHazard(MachineInstr *MI);
   bool fixVALUTransUseHazard(MachineInstr *MI);
+  bool fixWMMAHazards(MachineInstr *MI);
 
   int checkMAIHazards(MachineInstr *MI);
   int checkMAIHazards908(MachineInstr *MI);
   int checkMAIHazards90A(MachineInstr *MI);
   /// Pad the latency between neighboring MFMA instructions with s_nops. The
   /// percentage of wait states to fill with s_nops is specified by the command
   /// line option '-amdgpu-mfma-padding-ratio'.
   ///

llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp

void GCNHazardRecognizer::fixHazards(MachineInstr *MI) {
   fixVcmpxExecWARHazard(MI);
   fixLdsBranchVmemWARHazard(MI);
   if (ST.hasLdsDirect()) {
     fixLdsDirectVALUHazard(MI);
     fixLdsDirectVMEMHazard(MI);
   }
   fixVALUPartialForwardingHazard(MI);
   fixVALUTransUseHazard(MI);
+  fixWMMAHazards(MI);
 }
 
 bool GCNHazardRecognizer::fixVcmpxPermlaneHazards(MachineInstr *MI) {
   if (!ST.hasVcmpxPermlaneHazard() || !isPermlane(*MI))
     return false;
 
   const SIInstrInfo *TII = ST.getInstrInfo();
   const SIRegisterInfo *TRI = ST.getRegisterInfo();
bool GCNHazardRecognizer::fixVALUTransUseHazard(MachineInstr *MI) {
   // avoided (mask 0x0fff achieves this).
   BuildMI(*MI->getParent(), MI, MI->getDebugLoc(),
           TII.get(AMDGPU::S_WAITCNT_DEPCTR))
       .addImm(0x0fff);
 
   return true;
 }
 
+bool GCNHazardRecognizer::fixWMMAHazards(MachineInstr *MI) {
+  if (!SIInstrInfo::isWMMA(*MI))
+    return false;
+
+  const SIInstrInfo *TII = ST.getInstrInfo();
+  const SIRegisterInfo *TRI = ST.getRegisterInfo();
+
+  auto IsHazardFn = [MI, TII, TRI](const MachineInstr &I) {
+    if (!SIInstrInfo::isWMMA(I))
+      return false;
+
+    // Src0 or Src1 of the current wmma instruction overlaps with the dest of
+    // the previous wmma.
+    const Register CurSrc0Reg =
+        TII->getNamedOperand(*MI, AMDGPU::OpName::src0)->getReg();
+    const Register CurSrc1Reg =
+        TII->getNamedOperand(*MI, AMDGPU::OpName::src1)->getReg();
+
+    const Register PrevDstReg =
+        TII->getNamedOperand(I, AMDGPU::OpName::vdst)->getReg();
+
+    if (TRI->regsOverlap(PrevDstReg, CurSrc0Reg) ||
+        TRI->regsOverlap(PrevDstReg, CurSrc1Reg)) {
+      return true;
+    }
+
+    // Src2 of the current wmma instruction overlaps with the dest of the
+    // previous wmma.
+    const MachineOperand *Src2 =
+        TII->getNamedOperand(*MI, AMDGPU::OpName::src2);
+    const Register CurSrc2Reg = Src2->isReg() ? Src2->getReg() : Register();
+
+    if (CurSrc2Reg != AMDGPU::NoRegister &&
+        TRI->regsOverlap(PrevDstReg, CurSrc2Reg)) {
+
+      const MachineOperand *Src2Mods =
+          TII->getNamedOperand(*MI, AMDGPU::OpName::src2_modifiers);
+      const bool NoSrc2Mods =
+          (Src2Mods->getImm() & (SISrcMods::NEG | SISrcMods::NEG_HI)) == 0;
+      // Exception: there is no hazard if the wmma instructions are of the same
+      // type and there is no input modifier on src2 of the current instruction.
+      return !(NoSrc2Mods && (TII->pseudoToMCOpcode(I.getOpcode()) ==
+                              TII->pseudoToMCOpcode(MI->getOpcode())));
+    }
+
+    return false;
+  };
+
+  auto IsExpiredFn = [](const MachineInstr &I, int) {
+    return SIInstrInfo::isVALU(I);
+  };
+
+  if (::getWaitStatesSince(IsHazardFn, MI, IsExpiredFn) ==
+      std::numeric_limits<int>::max())
+    return false;
+
+  BuildMI(*MI->getParent(), MI, MI->getDebugLoc(), TII->get(AMDGPU::V_NOP_e32));
+
+  return true;
+}
+
 int GCNHazardRecognizer::checkNSAtoVMEMHazard(MachineInstr *MI) {
   int NSAtoVMEMWaitStates = 1;
 
   if (!ST.hasNSAtoVMEMBug())
     return 0;
 
   if (!SIInstrInfo::isMUBUF(*MI) && !SIInstrInfo::isMTBUF(*MI))
     return 0;

llvm/lib/Target/AMDGPU/SIDefines.h

enum : uint64_t {
 
   // FLAT instruction accesses FLAT_SCRATCH segment.
   FlatScratch = UINT64_C(1) << 56,
 
   // Atomic without return.
   IsAtomicNoRet = UINT64_C(1) << 57,
 
   // Atomic with return.
-  IsAtomicRet = UINT64_C(1) << 58
+  IsAtomicRet = UINT64_C(1) << 58,
+
+  // Is a WMMA instruction.
+  IsWMMA = UINT64_C(1) << 59,
 };
 
 // v_cmp_class_* etc. use a 10-bit mask for what operation is checked.
 // The result is true if any of these tests are true.
 enum ClassFlags : unsigned {
   S_NAN = 1 << 0,        // Signaling NaN
   Q_NAN = 1 << 1,        // Quiet NaN
   N_INFINITY = 1 << 2,   // Negative infinity

llvm/lib/Target/AMDGPU/SIInstrFormats.td

class InstSI <dag outs, dag ins, string asm = "",
   field bit FlatScratch = 0;
 
   // Atomic without a return.
   field bit IsAtomicNoRet = 0;
 
   // Atomic with return.
   field bit IsAtomicRet = 0;
 
+  // This bit indicates that this is one of WMMA instructions.
+  field bit IsWMMA = 0;
+
   // These need to be kept in sync with the enum in SIInstrFlags.
   let TSFlags{0} = SALU;
   let TSFlags{1} = VALU;
 
   let TSFlags{2} = SOP1;
   let TSFlags{3} = SOP2;
   let TSFlags{4} = SOPC;
   let TSFlags{5} = SOPK;
class InstSI <dag outs, dag ins, string asm = "",
   let TSFlags{55} = IsDOT;
 
   let TSFlags{56} = FlatScratch;
 
   let TSFlags{57} = IsAtomicNoRet;
 
   let TSFlags{58} = IsAtomicRet;
 
+  let TSFlags{59} = IsWMMA;
+
   let SchedRW = [Write32Bit];
 
   let AsmVariantName = AMDGPUAsmVariants.Default;
 
   // Avoid changing source registers in a way that violates constant bus read limitations.
   let hasExtraSrcRegAllocReq = !or(VOP1, VOP2, VOP3, VOPC, SDWA, VALU);
 }
 

llvm/lib/Target/AMDGPU/SIInstrInfo.h

static bool isMFMA(const MachineInstr &MI) {
     return isMAI(MI) && MI.getOpcode() != AMDGPU::V_ACCVGPR_WRITE_B32_e64 &&
            MI.getOpcode() != AMDGPU::V_ACCVGPR_READ_B32_e64;
   }
 
   static bool isDOT(const MachineInstr &MI) {
     return MI.getDesc().TSFlags & SIInstrFlags::IsDOT;
   }
 
+  static bool isWMMA(const MachineInstr &MI) {
+    return MI.getDesc().TSFlags & SIInstrFlags::IsWMMA;
+  }
+
+  bool isWMMA(uint16_t Opcode) const {
+    return get(Opcode).TSFlags & SIInstrFlags::IsWMMA;
+  }
+
   bool isDOT(uint16_t Opcode) const {
     return get(Opcode).TSFlags & SIInstrFlags::IsDOT;
   }
 
   static bool isLDSDIR(const MachineInstr &MI) {
     return MI.getDesc().TSFlags & SIInstrFlags::LDSDIR;
   }
 

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

if (NewMFMAOpc != -1) {
     for (unsigned I = 0, E = MI.getNumOperands(); I != E; ++I)
       MIB.add(MI.getOperand(I));
     updateLiveVariables(LV, MI, *MIB);
     if (LIS)
       LIS->ReplaceMachineInstrInMaps(MI, *MIB);
     return MIB;
   }
 
+  if (SIInstrInfo::isWMMA(MI)) {
+    unsigned NewOpc = AMDGPU::mapWMMA2AddrTo3AddrOpcode(MI.getOpcode());
+    MachineInstrBuilder MIB = BuildMI(MBB, MI, MI.getDebugLoc(), get(NewOpc))
+                                  .setMIFlags(MI.getFlags());
+    for (unsigned I = 0, E = MI.getNumOperands(); I != E; ++I)
+      MIB->addOperand(MI.getOperand(I));
+
+    updateLiveVariables(LV, MI, *MIB);
+    if (LIS)
+      LIS->ReplaceMachineInstrInMaps(MI, *MIB);
+
+    return MIB;
+  }
+
   // Handle MAC/FMAC.
   bool IsF16 = Opc == AMDGPU::V_MAC_F16_e32 || Opc == AMDGPU::V_MAC_F16_e64 ||
                Opc == AMDGPU::V_FMAC_F16_e32 || Opc == AMDGPU::V_FMAC_F16_e64;
   bool IsFMA = Opc == AMDGPU::V_FMAC_F32_e32 || Opc == AMDGPU::V_FMAC_F32_e64 ||
                Opc == AMDGPU::V_FMAC_LEGACY_F32_e32 ||
                Opc == AMDGPU::V_FMAC_LEGACY_F32_e64 ||
                Opc == AMDGPU::V_FMAC_F16_e32 || Opc == AMDGPU::V_FMAC_F16_e64 ||
                Opc == AMDGPU::V_FMAC_F64_e32 || Opc == AMDGPU::V_FMAC_F64_e64;

llvm/lib/Target/AMDGPU/SIInstrInfo.td

 def VOP3Mods_nnan : ComplexPattern<fAny, 2, "SelectVOP3Mods_NNaN">;
 
 def VOP3OMods : ComplexPattern<untyped, 3, "SelectVOP3OMods">;
 
 def VOP3PMods  : ComplexPattern<untyped, 2, "SelectVOP3PMods">;
 
 def VOP3PModsDOT  : ComplexPattern<untyped, 2, "SelectVOP3PModsDOT">;
 def DotIUVOP3PMods  : ComplexPattern<untyped, 1, "SelectDotIUVOP3PMods">;
+def WMMAOpSelVOP3PMods  : ComplexPattern<untyped, 1, "SelectWMMAOpSelVOP3PMods">;
 
 def VOP3OpSel  : ComplexPattern<untyped, 2, "SelectVOP3OpSel">;
 
 def VOP3OpSelMods  : ComplexPattern<untyped, 2, "SelectVOP3OpSelMods">;
 
 def VOP3PMadMixMods  : ComplexPattern<untyped, 2, "SelectVOP3PMadMixMods">;
 
 def VINTERPMods  : ComplexPattern<untyped, 2, "SelectVINTERPMods">;
class VOPProfile <list<ValueType> _ArgVT, bit _EnableF32SrcMods = 0,
   field bit HasExtSDWA = getHasSDWA<NumSrcArgs, DstVT, Src0VT, Src1VT>.ret;
   field bit HasExtSDWA9 = HasExtSDWA;
   field int NeedPatGen = PatGenMode.NoPattern;
 
   field bit IsMAI = 0;
   field bit IsVOP3P = 0;
   field bit IsDOT = 0;
   field bit IsSingle = 0;
+  field bit IsWMMA = 0;
 
   field Operand Src0PackedMod = !if(HasSrc0FloatMods, PackedF16InputMods, PackedI16InputMods);
   field Operand Src1PackedMod = !if(HasSrc1FloatMods, PackedF16InputMods, PackedI16InputMods);
   field Operand Src2PackedMod = !if(HasSrc2FloatMods, PackedF16InputMods, PackedI16InputMods);
 
   field dag Outs = !if(HasDst,(outs DstRC:$vdst),(outs));
 
   // VOP3b instructions are a special case with a second explicit

llvm/lib/Target/AMDGPU/SIRegisterInfo.td

 
 // This is for operands with the enum(9), VSrc encoding restriction,
 // but only allows VGPRs.
 def VRegSrc_32 : RegisterOperand<VGPR_32> {
   //let ParserMatchClass = RegImmMatcher<"VRegSrc32">;
   let DecoderMethod = "DecodeVS_32RegisterClass";
 }
 
+def VRegSrc_64 : RegisterOperand<VReg_64> {
+  let DecoderMethod = "decodeOperand_VReg_64";
+}
+
+def VRegSrc_128 : RegisterOperand<VReg_128> {
+  let DecoderMethod = "decodeOperand_VReg_128";
+}
+
+def VRegSrc_256 : RegisterOperand<VReg_256> {
+  let DecoderMethod = "decodeOperand_VReg_256";
+}
+
 //===----------------------------------------------------------------------===//
 // VGPRSrc_*
 //===----------------------------------------------------------------------===//
 
 // An 8-bit RegisterOperand wrapper for a VGPR
 def VGPRSrc_32 : RegisterOperand<VGPR_32> {
   let DecoderMethod = "DecodeVGPR_32RegisterClass";
 }

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h

 struct MIMGG16MappingInfo {
   MIMGBaseOpcode G;
   MIMGBaseOpcode G16;
 };
 
 LLVM_READONLY
 const MIMGLZMappingInfo *getMIMGLZMappingInfo(unsigned L);
 
+struct WMMAOpcodeMappingInfo {
+  unsigned Opcode2Addr;
+  unsigned Opcode3Addr;
+};
+
 LLVM_READONLY
 const MIMGMIPMappingInfo *getMIMGMIPMappingInfo(unsigned MIP);
 
 LLVM_READONLY
 const MIMGBiasMappingInfo *getMIMGBiasMappingInfo(unsigned Bias);
 
 LLVM_READONLY
 const MIMGOffsetMappingInfo *getMIMGOffsetMappingInfo(unsigned Offset);
const GcnBufferFormatInfo *getGcnBufferFormatInfo(uint8_t BitsPerComp,
                                                   const MCSubtargetInfo &STI);
 LLVM_READONLY
 const GcnBufferFormatInfo *getGcnBufferFormatInfo(uint8_t Format,
                                                   const MCSubtargetInfo &STI);
 
 LLVM_READONLY
 int getMCOpcode(uint16_t Opcode, unsigned Gen);
 
+LLVM_READONLY
+unsigned mapWMMA2AddrTo3AddrOpcode(unsigned Opc);
+
+LLVM_READONLY
+unsigned mapWMMA3AddrTo2AddrOpcode(unsigned Opc);
+
 void initDefaultAMDKernelCodeT(amd_kernel_code_t &Header,
                                const MCSubtargetInfo *STI);
 
 amdhsa::kernel_descriptor_t getDefaultAmdhsaKernelDescriptor(
     const MCSubtargetInfo *STI);
 
 bool isGroupSegment(const GlobalValue *GV);
 bool isGlobalSegment(const GlobalValue *GV);

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp

 #define GET_VOP2InfoTable_DECL
 #define GET_VOP2InfoTable_IMPL
 #define GET_VOP3InfoTable_DECL
 #define GET_VOP3InfoTable_IMPL
 #define GET_VOPC64DPPTable_DECL
 #define GET_VOPC64DPPTable_IMPL
 #define GET_VOPC64DPP8Table_DECL
 #define GET_VOPC64DPP8Table_IMPL
+#define GET_WMMAOpcode2AddrMappingTable_DECL
+#define GET_WMMAOpcode2AddrMappingTable_IMPL
+#define GET_WMMAOpcode3AddrMappingTable_DECL
+#define GET_WMMAOpcode3AddrMappingTable_IMPL
 #include "AMDGPUGenSearchableTables.inc"
 
 int getMTBUFBaseOpcode(unsigned Opc) {
   const MTBUFInfo *Info = getMTBUFInfoFromOpcode(Opc);
   return Info ? Info->BaseOpcode : -1;
 }
 
 int getMTBUFOpcode(unsigned BaseOpc, unsigned Elements) {
bool getMAIIsDGEMM(unsigned Opc) {
   return Info ? Info->is_dgemm : false;
 }
 
 bool getMAIIsGFX940XDL(unsigned Opc) {
   const MAIInstInfo *Info = getMAIInstInfoHelper(Opc);
   return Info ? Info->is_gfx940_xdl : false;
 }
 
+unsigned mapWMMA2AddrTo3AddrOpcode(unsigned Opc) {
+  const WMMAOpcodeMappingInfo *Info = getWMMAMappingInfoFrom2AddrOpcode(Opc);
+  return Info ? Info->Opcode3Addr : ~0u;
+}
+
+unsigned mapWMMA3AddrTo2AddrOpcode(unsigned Opc) {
+  const WMMAOpcodeMappingInfo *Info = getWMMAMappingInfoFrom3AddrOpcode(Opc);
+  return Info ? Info->Opcode2Addr : ~0u;
+}
+
 // Wrapper for Tablegen'd function.  enum Subtarget is not defined in any
 // header files, so we need to wrap it in a function that takes unsigned
 // instead.
 int getMCOpcode(uint16_t Opcode, unsigned Gen) {
   return getMCOpcodeGen(Opcode, static_cast<Subtarget>(Gen));
 }
 
 namespace IsaInfo {

llvm/lib/Target/AMDGPU/VOP3PInstructions.td

let SubtargetPredicate = HasPackedFP32Ops, isCommutable = 1 in {
   defm V_PK_MUL_F32 : VOP3PInst<"v_pk_mul_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fmul>;
   defm V_PK_ADD_F32 : VOP3PInst<"v_pk_add_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fadd>;
   defm V_PK_MOV_B32 : VOP3PInst<"v_pk_mov_b32", VOP3P_Profile<VOP_V2I32_V2I32_V2I32, VOP3_PACKED>>;
 } // End SubtargetPredicate = HasPackedFP32Ops, isCommutable = 1
 
 def : MnemonicAlias<"v_accvgpr_read",  "v_accvgpr_read_b32">;
 def : MnemonicAlias<"v_accvgpr_write", "v_accvgpr_write_b32">;
 
+class VOPProfileWMMA<VOPProfile P, string Suffix, RegisterOperand _Src01RC64, bit _HasClamp, bit _HasOpSel> : VOP3P_Profile<P> {
+  let DstRC = !if(!eq(Suffix, "_w32"), VDst_256, VDst_128);
+  let Src0RC64 = _Src01RC64;
+  let Src1RC64 = _Src01RC64;
+  let Src2RC64 = !if(!eq(Suffix, "_w32"), VISrc_256_f64, VISrc_128_f32);
+  let HasClamp = _HasClamp;
+  let HasOpSel = _HasOpSel;
+  let IsWMMA = 1;
+}
+
+def VOP_V8F32_V8F32_V8F32_V8F32 : VOPProfile <[v8f32, v8f32, v8f32, v8f32]>;
+def VOP_V8F32_V8I32_V8I32_V8F32 : VOPProfile <[v8f32, v8i32, v8i32, v8f32]>;
+def VOP_V8I32_V4I32_V4I32_V8I32 : VOPProfile <[v8i32, v4i32, v4i32, v8i32]>;
+def VOP_V8I32_V2I32_V2I32_V8I32 : VOPProfile <[v8i32, v2i32, v2i32, v8i32]>;
+def VOP_V8I32_V8I32_V8I32_V8I32 : VOPProfile <[v8i32, v8i32, v8i32, v8i32]>;
+
+def VOP_V4F32_V8F32_V8F32_V4F32 : VOPProfile <[v4f32, v8f32, v8f32, v4f32]>;
+def VOP_V4F32_V8I32_V8I32_V4F32 : VOPProfile <[v4f32, v8i32, v8i32, v4f32]>;
+def VOP_V4I32_V4I32_V4I32_V4I32 : VOPProfile <[v4i32, v4i32, v4i32, v4i32]>;
+def VOP_V4I32_V2I32_V2I32_V4I32 : VOPProfile <[v4i32, v2i32, v2i32, v4i32]>;
+def VOP_V4I32_V8I32_V8I32_V4I32 : VOPProfile <[v4i32, v8i32, v8i32, v4i32]>;
+
+class WMMAType <bits<2> val> {
+  bit hasClamp = val{0};
+  bit hasOpsel = val{1};
+}
+
+def WMMARegular      : WMMAType<0b00>;
+def WMMAUIClamp      : WMMAType<0b01>;
+def WMMAOpSel        : WMMAType<0b10>;
+
+class WMMARegularPat<Instruction Inst, SDPatternOperator node, VOPProfile P> :
+  GCNPat < (P.DstVT (node
+                                (P.Src0VT (VOP3PMods P.Src0VT:$src0, i32:$src0_modifiers)),
+                                (P.Src1VT (VOP3PMods P.Src1VT:$src1, i32:$src1_modifiers)),
+                                (P.Src2VT (VOP3PMods P.Src2VT:$src2, i32:$src2_modifiers))
+                   )),
+                   (P.DstVT (Inst i32:$src0_modifiers, P.Src0VT:$src0, i32:$src1_modifiers, P.Src1VT:$src1, $src2_modifiers, P.Src2VT:$src2))
+>;
+
+class WMMAOpSelPat<Instruction Inst, SDPatternOperator node, VOPProfile P> :
+  GCNPat < (P.DstVT (node
+                                (P.Src0VT P.Src0VT:$src0),
+                                (P.Src1VT P.Src1VT:$src1),
+                                (P.Src2VT P.Src2VT:$src2), (WMMAOpSelVOP3PMods i32:$src2_modifiers)
+                   )),
+                   (P.DstVT (Inst (i32 8), P.Src0VT:$src0, (i32 8), P.Src1VT:$src1, i32:$src2_modifiers, P.Src2VT:$src2))
+>;
+
+class WMMAUIClampPat<Instruction Inst, SDPatternOperator node, VOPProfile P> :
+  GCNPat < (P.DstVT (node
+                                (DotIUVOP3PMods i32:$src0_modifiers), (P.Src0VT P.Src0VT:$src0),
+                                (DotIUVOP3PMods i32:$src1_modifiers), (P.Src1VT P.Src1VT:$src1),
+                                (P.Src2VT P.Src2VT:$src2), (i1 timm:$clamp)
+                   )),
+                   (P.DstVT (Inst i32:$src0_modifiers, P.Src0VT:$src0, i32:$src1_modifiers, P.Src1VT:$src1, (i32 8), P.Src2VT:$src2, i1:$clamp))
+>;
+
+class WMMAOpcodeMapping<Instruction TwoAddr, Instruction ThreeAddr> {
+  Instruction Opcode2Addr = TwoAddr;
+  Instruction Opcode3Addr = ThreeAddr;
+  Predicate WaveSizePredicate;
+}
+
+def WMMAOpcode : GenericEnum {
+  let FilterClass = "VOP3P_Pseudo";
+}
+
+class WMMAMappingTable : GenericTable {
+  let FilterClass = "WMMAOpcodeMapping";
+  let CppTypeName = "WMMAOpcodeMappingInfo";
+  let Fields = ["Opcode2Addr", "Opcode3Addr"];
+  string TypeOf_Opcode2Addr = "WMMAOpcode";
+  string TypeOf_Opcode3Addr = "WMMAOpcode";
+}
+
+def WMMAOpcode2AddrMappingTable : WMMAMappingTable {
+  let PrimaryKey = ["Opcode2Addr"];
+  let PrimaryKeyName = "getWMMAMappingInfoFrom2AddrOpcode";
+}
+
+def WMMAOpcode3AddrMappingTable : WMMAMappingTable {
+  let PrimaryKey = ["Opcode3Addr"];
+  let PrimaryKeyName = "getWMMAMappingInfoFrom3AddrOpcode";
+}
+
+// The WMMA instruction has extra constraints:
+// Matrices A and B cannot overlap with D. C cannot partially overlap with D,
+// but it is OK for them to be the same (which is a typical case).
+//
+// We implement it as follows:
+// 1) Map the intrinsic to the pseudo where D is tied to C ($vdst = $src2).
+// 2) The pass twoaddressinstruction checks if src2 is live and if that is the case
+//    it converts the default pseudo to the pseudo where src2 is not the same as vdst.
+// 3) @earlyclobber on the destination satisfies the constraint during RA.
+
+multiclass WMMAInst<string Suffix, string Instr, VOPProfile P, SDPatternOperator node = null_frag, RegisterOperand _Src01RC64 = VRegSrc_256, WMMAType Type> {
+
+  defvar WMMAConstraints2Addr = "@earlyclobber $vdst,$vdst = $src2";
+  defvar WMMAConstraints3Addr = "@earlyclobber $vdst";
+
+  defvar WMMAProfile = VOPProfileWMMA<P, Suffix, _Src01RC64, Type.hasClamp, Type.hasOpsel>;
+  if !eq(Suffix, "_w32") then {
+    let Mnemonic = Instr, mayRaiseFPException = 0, ReadsModeReg = 0 in {
+      let Constraints = WMMAConstraints2Addr, isConvertibleToThreeAddress = 1 in {
+        def _twoaddr_w32 : VOP3P_Pseudo<Instr # Suffix, WMMAProfile>;
+      }
+      let Constraints = WMMAConstraints3Addr, SchedRW = [Write32Bit, Write32Bit] in {
+        def _threeaddr_w32 : VOP3P_Pseudo<Instr # Suffix, WMMAProfile>;
+      }
+    }
+    def : WMMAOpcodeMapping<!cast<Instruction>(NAME # _twoaddr_w32),
+                            !cast<Instruction>(NAME # _threeaddr_w32)>;
+  } else if !eq(Suffix, "_w64") then {
+    let Mnemonic = Instr, mayRaiseFPException = 0, ReadsModeReg = 0 in {
+      let Constraints = WMMAConstraints2Addr, isConvertibleToThreeAddress = 1 in {
+        def _twoaddr_w64 : VOP3P_Pseudo<Instr # Suffix, WMMAProfile>;
+      }
+      let Constraints = WMMAConstraints3Addr, SchedRW = [Write32Bit, Write32Bit] in {
+        def _threeaddr_w64 : VOP3P_Pseudo<Instr # Suffix, WMMAProfile>;
+      }
+    }
+    def : WMMAOpcodeMapping<!cast<Instruction>(NAME # _twoaddr_w64),
+                            !cast<Instruction>(NAME # _threeaddr_w64)>;
+  }
+
+  if !eq(Type, WMMAOpSel) then {
+    def : WMMAOpSelPat<!cast<Instruction>(NAME # _twoaddr # Suffix), node, P>;
+  } else if !eq(Type, WMMAUIClamp) then {
+    def : WMMAUIClampPat<!cast<Instruction>(NAME # _twoaddr # Suffix), node, P>;
+  } else {
+    def : WMMARegularPat<!cast<Instruction>(NAME # _twoaddr # Suffix), node, P>;
+  }
+}
+
+let WaveSizePredicate = isWave32 in {
+  defm V_WMMA_F32_16X16X16_F16   : WMMAInst<"_w32", "v_wmma_f32_16x16x16_f16",  VOP_V8F32_V8F32_V8F32_V8F32, int_amdgcn_wmma_f32_16x16x16_f16, VRegSrc_256, WMMARegular>;
+  defm V_WMMA_F32_16X16X16_BF16  : WMMAInst<"_w32", "v_wmma_f32_16x16x16_bf16", VOP_V8F32_V8I32_V8I32_V8F32, int_amdgcn_wmma_f32_16x16x16_bf16, VRegSrc_256, WMMARegular>;
+  defm V_WMMA_F16_16X16X16_F16   : WMMAInst<"_w32", "v_wmma_f16_16x16x16_f16",   VOP_V8F32_V8F32_V8F32_V8F32, int_amdgcn_wmma_f16_16x16x16_f16, VRegSrc_256, WMMAOpSel>;
+  defm V_WMMA_BF16_16X16X16_BF16 : WMMAInst<"_w32", "v_wmma_bf16_16x16x16_bf16", VOP_V8I32_V8I32_V8I32_V8I32, int_amdgcn_wmma_bf16_16x16x16_bf16, VRegSrc_256, WMMAOpSel>;
+  defm V_WMMA_I32_16X16X16_IU8   : WMMAInst<"_w32", "v_wmma_i32_16x16x16_iu8",   VOP_V8I32_V4I32_V4I32_V8I32, int_amdgcn_wmma_i32_16x16x16_iu8, VRegSrc_128, WMMAUIClamp>;
+  defm V_WMMA_I32_16X16X16_IU4   : WMMAInst<"_w32", "v_wmma_i32_16x16x16_iu4",   VOP_V8I32_V2I32_V2I32_V8I32, int_amdgcn_wmma_i32_16x16x16_iu4, VRegSrc_64,  WMMAUIClamp>;
+}
+
+let WaveSizePredicate = isWave64 in {
+  defm V_WMMA_F32_16X16X16_F16   : WMMAInst<"_w64", "v_wmma_f32_16x16x16_f16",   VOP_V4F32_V8F32_V8F32_V4F32, int_amdgcn_wmma_f32_16x16x16_f16, VRegSrc_256, WMMARegular>;
+  defm V_WMMA_F32_16X16X16_BF16  : WMMAInst<"_w64", "v_wmma_f32_16x16x16_bf16",  VOP_V4F32_V8I32_V8I32_V4F32, int_amdgcn_wmma_f32_16x16x16_bf16, VRegSrc_256, WMMARegular>;
+  defm V_WMMA_F16_16X16X16_F16   : WMMAInst<"_w64", "v_wmma_f16_16x16x16_f16",   VOP_V4F32_V8F32_V8F32_V4F32, int_amdgcn_wmma_f16_16x16x16_f16, VRegSrc_256, WMMAOpSel>;
+  defm V_WMMA_BF16_16X16X16_BF16 : WMMAInst<"_w64", "v_wmma_bf16_16x16x16_bf16", VOP_V4I32_V8I32_V8I32_V4I32, int_amdgcn_wmma_bf16_16x16x16_bf16, VRegSrc_256, WMMAOpSel>;
+  defm V_WMMA_I32_16X16X16_IU8   : WMMAInst<"_w64", "v_wmma_i32_16x16x16_iu8",   VOP_V4I32_V4I32_V4I32_V4I32, int_amdgcn_wmma_i32_16x16x16_iu8, VRegSrc_128, WMMAUIClamp>;
+  defm V_WMMA_I32_16X16X16_IU4   : WMMAInst<"_w64", "v_wmma_i32_16x16x16_iu4",   VOP_V4I32_V2I32_V2I32_V4I32, int_amdgcn_wmma_i32_16x16x16_iu4, VRegSrc_64, WMMAUIClamp>;
+}
+
 //===----------------------------------------------------------------------===//
 // Begin Real Encodings
 //===----------------------------------------------------------------------===//
 
 class VOP3P_DPP16<bits<7> op, VOP_DPP_Pseudo ps, int subtarget,
                   string opName = ps.OpName>
     : VOP3P_DPP<op, opName, ps.Pfl, 1>, SIMCInstr<ps.PseudoInstr, subtarget> {
   let hasSideEffects = ps.hasSideEffects;
multiclass VOP3P_Realtriple_gfx11<bits<7> op, string backing_ps_name = NAME,
         VOP3P_Real_dpp_gfx11<op, backing_ps_name, asmName>,
         VOP3P_Real_dpp8_gfx11<op, backing_ps_name, asmName>;
 } // End AssemblerPredicate = isGFX11Plus, DecoderNamespace = "GFX11"
 
 defm V_DOT4_I32_IU8              : VOP3P_Real_gfx11 <0x16>;
 defm V_DOT8_I32_IU4              : VOP3P_Real_gfx11 <0x18>;
 defm V_DOT2_F32_BF16             : VOP3P_Real_gfx11 <0x1a>;
 
+multiclass VOP3P_Real_WMMA <bits<7> op> {
+  let WaveSizePredicate = isWave32, DecoderNamespace = "GFX11" in {
+    defm _twoaddr_w32 : VOP3P_Real_gfx11 <op>;
+  }
+  let WaveSizePredicate = isWave64, DecoderNamespace = "WMMAGFX11" in {
+    defm _twoaddr_w64 : VOP3P_Real_gfx11 <op>;
+  }
+}
+
+defm V_WMMA_F32_16X16X16_F16   : VOP3P_Real_WMMA <0x040>;
+defm V_WMMA_F32_16X16X16_BF16  : VOP3P_Real_WMMA <0x041>;
+defm V_WMMA_F16_16X16X16_F16   : VOP3P_Real_WMMA <0x042>;
+defm V_WMMA_BF16_16X16X16_BF16 : VOP3P_Real_WMMA <0x043>;
+defm V_WMMA_I32_16X16X16_IU8   : VOP3P_Real_WMMA <0x044>;
+defm V_WMMA_I32_16X16X16_IU4   : VOP3P_Real_WMMA <0x045>;
+
 //===----------------------------------------------------------------------===//
 // GFX8 (VI)
 //===----------------------------------------------------------------------===//
 
 multiclass VOP3P_Real_vi<bits<7> op> {
   def _vi : VOP3P_Real<!cast<VOP3_Pseudo>(NAME), SIEncodingFamily.VI>,
             VOP3Pe <op, !cast<VOP3_Pseudo>(NAME).Pfl> {
     let AssemblerPredicate = HasVOP3PInsts;

llvm/lib/Target/AMDGPU/VOPInstructions.td

VOP_Pseudo <opName, "_e64", P, P.Outs64,
               !if(isVop3OpSel,
                   P.InsVOP3OpSel,
                   !if(!and(isVOP3P, P.IsPacked), P.InsVOP3P, P.Ins64)),
               "", pattern> {
 
   let VOP3_OPSEL = isVop3OpSel;
   let IsPacked = P.IsPacked;
   let IsMAI = P.IsMAI;
+  let IsWMMA = P.IsWMMA;
 
   let AsmOperands = !if(isVop3OpSel,
                         P.AsmVOP3OpSel,
                         !if(!and(isVOP3P, P.IsPacked), P.AsmVOP3P, P.Asm64));
 
   let Size = 8;
   let mayLoad = 0;
   let mayStore = 0;
class VOP3_Real <VOP_Pseudo ps, int EncodingFamily, string asm_name = ps.Mnemonic> :
   let TRANS                = ps.TRANS;
 
   VOPProfile Pfl = ps.Pfl;
 }
 
 // XXX - Is there any reason to distinguish this from regular VOP3
 // here?
 class VOP3P_Real<VOP_Pseudo ps, int EncodingFamily, string asm_name = ps.Mnemonic> :
-  VOP3_Real<ps, EncodingFamily, asm_name>;
+  VOP3_Real<ps, EncodingFamily, asm_name> {
+
+  // The v_wmma pseudos have extra constraints that we do not want to impose on the real instruction.
+  let Constraints        = !if(!eq(!substr(ps.Mnemonic,0,6), "v_wmma"), "", ps.Constraints);
+}
 
 class VOP3a<VOPProfile P> : Enc64 {
   bits<4> src0_modifiers;
   bits<9> src0;
   bits<3> src1_modifiers;
   bits<9> src1;
   bits<3> src2_modifiers;
   bits<9> src2;

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.wmma_32.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -verify-machineinstrs < %s | FileCheck %s --check-prefix=W32
+
+declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float>, <8 x float> , <8 x float>)
+declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32>, <8 x i32> , <8 x float>)
+declare <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float>, <8 x float> , <8 x float>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32>, <8 x i32> , <8 x i32>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 immarg, <4 x i32>, i1 immarg, <4 x i32> , <8 x i32>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 immarg, <2 x i32>, i1 immarg, <2 x i32> , <8 x i32>, i1 immarg)
+
+; @llvm.amdgcn.wmma.f32.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_f32_16x16x16_f16:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f32.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_bf16(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, <8 x float> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_f32_16x16x16_bf16:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x float> %C)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f16.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_lo(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_f16_16x16x16_f16_lo:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, i1 0)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_hi(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_f16_16x16x16_f16_hi:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, i1 1)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.bf16.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_lo(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_bf16_16x16x16_bf16_lo:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_hi(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_bf16_16x16x16_bf16_hi:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu8
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu4
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.wmma_64.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs < %s | FileCheck %s --check-prefix=W64
+
+declare <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float>, <8 x float>, <4 x float>)
+declare <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32>, <8 x i32>, <4 x float>)
+declare <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float>, <8 x float>, <4 x float>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32>, <8 x i32>, <4 x i32>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 immarg, <4 x i32>, i1 immarg, <4 x i32>, <4 x i32>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 immarg, <2 x i32>, i1 immarg, <2 x i32>, <4 x i32>, i1 immarg)
+
+; @llvm.amdgcn.wmma.f32.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_f32_16x16x16_f16:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f32.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_bf16(<8 x i32> %A, <8 x i32> %B, <4 x float> %C, <4 x float> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_f32_16x16x16_bf16:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x float> %C)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f16.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_lo(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_f16_16x16x16_f16_lo:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, i1 0)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_hi(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_f16_16x16x16_f16_hi:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, i1 1)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.bf16.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_lo(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_bf16_16x16x16_bf16_lo:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_hi(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_bf16_16x16x16_bf16_hi:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu8
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11]
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu4
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7]
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.wmma_32.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -verify-machineinstrs < %s | FileCheck %s --check-prefix=W32
+
+declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float>, <8 x float> , <8 x float>)
+declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32>, <8 x i32> , <8 x float>)
+declare <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float>, <8 x float> , <8 x float>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32>, <8 x i32> , <8 x i32>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 immarg, <4 x i32>, i1 immarg, <4 x i32> , <8 x i32>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 immarg, <2 x i32>, i1 immarg, <2 x i32> , <8 x i32>, i1 immarg)
+
+; @llvm.amdgcn.wmma.f32.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_f32_16x16x16_f16:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f32.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_bf16(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, <8 x float> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_f32_16x16x16_bf16:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x float> %C)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f16.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_lo(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_f16_16x16x16_f16_lo:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, i1 0)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_hi(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_f16_16x16x16_f16_hi:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, i1 1)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.bf16.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_lo(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_bf16_16x16x16_bf16_lo:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_hi(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_bf16_16x16x16_bf16_hi:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu8
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu4
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  ret void
+}
+

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.wmma_64.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs < %s | FileCheck %s --check-prefix=W64
+
+declare <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float>, <8 x float>, <4 x float>)
+declare <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32>, <8 x i32>, <4 x float>)
+declare <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float>, <8 x float>, <4 x float>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32>, <8 x i32>, <4 x i32>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 immarg, <4 x i32>, i1 immarg, <4 x i32>, <4 x i32>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 immarg, <2 x i32>, i1 immarg, <2 x i32>, <4 x i32>, i1 immarg)
+
+; @llvm.amdgcn.wmma.f32.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_f32_16x16x16_f16:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f32.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_bf16(<8 x i32> %A, <8 x i32> %B, <4 x float> %C, <4 x float> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_f32_16x16x16_bf16:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x float> %C)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f16.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_lo(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_f16_16x16x16_f16_lo:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, i1 0)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_hi(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_f16_16x16x16_f16_hi:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, i1 1)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.bf16.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_lo(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_bf16_16x16x16_bf16_lo:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_hi(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_bf16_16x16x16_bf16_hi:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+; W64-NEXT:    global_store_b128 v[20:21], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu8
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11]
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu4
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7]
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  ret void
+}
+

llvm/test/CodeGen/AMDGPU/threeaddr-wmma.mir

+# RUN: llc -march=amdgcn -mcpu=gfx1100 %s -start-after postrapseudos -verify-machineinstrs -o - | FileCheck -check-prefix=GCN %s
+
+# GCN-LABEL: test_V_WMMA_F32_16X16X16_F16_threeaddr_w32:
+# GCN:  v_wmma_f32_16x16x16_f16 v[34:41], v[0:7], v[8:15], v[16:23]
+---
+name: test_V_WMMA_F32_16X16X16_F16_threeaddr_w32
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F32_16X16X16_F16_threeaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+---
+
+# GCN-LABEL: test_V_WMMA_F32_16X16X16_BF16_threeaddr_w32:
+# GCN: v_wmma_f32_16x16x16_bf16 v[34:41], v[0:7], v[8:15], v[16:23]
+---
+name: test_V_WMMA_F32_16X16X16_BF16_threeaddr_w32
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F32_16X16X16_BF16_threeaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+---
+
+# GCN-LABEL: test_V_WMMA_F16_16X16X16_F16_threeaddr_w32:
+# GCN: v_wmma_f16_16x16x16_f16 v[34:41], v[0:7], v[8:15], v[16:23]
+---
+name: test_V_WMMA_F16_16X16X16_F16_threeaddr_w32
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X16_F16_threeaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
+---
+
+# GCN-LABEL: test_V_WMMA_BF16_16X16X16_BF16_threeaddr_w32:
+# GCN: v_wmma_bf16_16x16x16_bf16 v[34:41], v[0:7], v[8:15], v[16:23]
+---
+name: test_V_WMMA_BF16_16X16X16_BF16_threeaddr_w32
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_BF16_16X16X16_BF16_threeaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
+---
+
+# GCN-LABEL: test_V_WMMA_I32_16X16X16_IU8_threeaddr_w32:
+# GCN: v_wmma_i32_16x16x16_iu8 v[26:33], v[0:3], v[4:7], v[8:15]
+---
+name: test_V_WMMA_I32_16X16X16_IU8_threeaddr_w32
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33 = V_WMMA_I32_16X16X16_IU8_threeaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3, 8, $vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 0, 0, 0, implicit $exec
+---
+
+
+# GCN-LABEL: test_V_WMMA_I32_16X16X16_IU4_threeaddr_w32:
+# GCN: v_wmma_i32_16x16x16_iu4 v[26:33], v[0:1], v[2:3], v[8:15]
+---
+name: test_V_WMMA_I32_16X16X16_IU4_threeaddr_w32
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33 = V_WMMA_I32_16X16X16_IU4_threeaddr_w32 8, killed $vgpr0_vgpr1, 8, $vgpr2_vgpr3, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 0, 0, 0, implicit $exec
+
+
+# GCN-LABEL: test_V_WMMA_F32_16X16X16_F16_threeaddr_w64:
+# GCN: v_wmma_f32_16x16x16_f16 v[34:37], v[0:7], v[8:15], v[16:19]
+---
+name: test_V_WMMA_F32_16X16X16_F16_threeaddr_w64
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr34_vgpr35_vgpr36_vgpr37 = V_WMMA_F32_16X16X16_F16_threeaddr_w64 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19, 0, 0, implicit $exec
+---
+
+# GCN-LABEL: test_V_WMMA_F32_16X16X16_BF16_threeaddr_w64:
+# GCN: v_wmma_f32_16x16x16_bf16 v[34:37], v[0:7], v[8:15], v[16:19]
+---
+name: test_V_WMMA_F32_16X16X16_BF16_threeaddr_w64
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr34_vgpr35_vgpr36_vgpr37 = V_WMMA_F32_16X16X16_BF16_threeaddr_w64 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19, 0, 0, implicit $exec
+---
+
+# GCN-LABEL: test_V_WMMA_F16_16X16X16_F16_threeaddr_w64:
+# GCN: v_wmma_f16_16x16x16_f16 v[34:37], v[0:7], v[8:15], v[16:19]
+---
+name: test_V_WMMA_F16_16X16X16_F16_threeaddr_w64
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr34_vgpr35_vgpr36_vgpr37 = V_WMMA_F16_16X16X16_F16_threeaddr_w64 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19, 0, 0, 0, 0, implicit $exec
+---
+
+# GCN-LABEL: test_V_WMMA_BF16_16X16X16_BF16_threeaddr_w64:
+# GCN: v_wmma_bf16_16x16x16_bf16 v[34:37], v[0:7], v[8:15], v[16:19]
+---
+name: test_V_WMMA_BF16_16X16X16_BF16_threeaddr_w64
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr34_vgpr35_vgpr36_vgpr37 = V_WMMA_BF16_16X16X16_BF16_threeaddr_w64 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19, 0, 0, 0, 0, implicit $exec
+---
+
+# GCN-LABEL: test_V_WMMA_I32_16X16X16_IU8_threeaddr_w64:
+# GCN: v_wmma_i32_16x16x16_iu8 v[26:29], v[0:3], v[4:7], v[8:11]
+---
+name: test_V_WMMA_I32_16X16X16_IU8_threeaddr_w64
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr26_vgpr27_vgpr28_vgpr29 = V_WMMA_I32_16X16X16_IU8_threeaddr_w64 8, killed $vgpr0_vgpr1_vgpr2_vgpr3, 8, $vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11, 0, 0, 0, implicit $exec
+---
+
+
+# GCN-LABEL: test_V_WMMA_I32_16X16X16_IU4_threeaddr_w64:
+# GCN: v_wmma_i32_16x16x16_iu4 v[26:29], v[0:1], v[2:3], v[8:11]
+---
+name: test_V_WMMA_I32_16X16X16_IU4_threeaddr_w64
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
+    early-clobber renamable $vgpr26_vgpr27_vgpr28_vgpr29 = V_WMMA_I32_16X16X16_IU4_threeaddr_w64 8, killed $vgpr0_vgpr1, 8, $vgpr2_vgpr3, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11, 0, 0, 0, implicit $exec
+
+

llvm/test/CodeGen/AMDGPU/twoaddr-wmma.mir

+# RUN: llc -march=amdgcn -mcpu=gfx1100 %s -run-pass twoaddressinstruction -verify-machineinstrs -o - | FileCheck -check-prefix=GCN %s
+
+# GCN-LABEL: name: test_v_wmma_f32_16x16x16_f16_twoaddr_w32
+# GCN: early-clobber %2:vreg_256 = V_WMMA_F32_16X16X16_F16_threeaddr_w32 8, killed %1, 8, killed %1, 8, %0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_f32_16x16x16_f16_twoaddr_w32
+registers:
+  - { id: 0, class: vreg_256 }
+  - { id: 1, class: vreg_256 }
+  - { id: 2, class: vreg_256 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_256 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed %1:vreg_256, 8, killed %1:vreg_256, 8, %0:vreg_256, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_f32_16x16x16_bf16_twoaddr_w32
+# GCN: early-clobber %2:vreg_256 = V_WMMA_F32_16X16X16_BF16_threeaddr_w32 8, killed %1, 8, killed %1, 8, %0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_f32_16x16x16_bf16_twoaddr_w32
+registers:
+  - { id: 0, class: vreg_256 }
+  - { id: 1, class: vreg_256 }
+  - { id: 2, class: vreg_256 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_256 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed %1:vreg_256, 8, killed %1:vreg_256, 8, %0:vreg_256, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_f16_16x16x16_f16_twoaddr_w32
+# GCN: early-clobber %2:vreg_256 = V_WMMA_F16_16X16X16_F16_threeaddr_w32 8, killed %1, 8, killed %1, 8, %0, 0, 0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_f16_16x16x16_f16_twoaddr_w32
+registers:
+  - { id: 0, class: vreg_256 }
+  - { id: 1, class: vreg_256 }
+  - { id: 2, class: vreg_256 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_256 = V_WMMA_F16_16X16X16_F16_twoaddr_w32 8, killed %1:vreg_256, 8, killed %1:vreg_256, 8, %0:vreg_256, 0, 0, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_bf16_16x16x16_bf16_twoaddr_w32
+# GCN: early-clobber %2:vreg_256 = V_WMMA_BF16_16X16X16_BF16_threeaddr_w32 8, killed %1, 8, killed %1, 8, %0, 0, 0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_bf16_16x16x16_bf16_twoaddr_w32
+registers:
+  - { id: 0, class: vreg_256 }
+  - { id: 1, class: vreg_256 }
+  - { id: 2, class: vreg_256 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_256 = V_WMMA_BF16_16X16X16_BF16_twoaddr_w32 8, killed %1:vreg_256, 8, killed %1:vreg_256, 8, %0:vreg_256, 0, 0, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_i32_16x16x16_iu8_twoaddr_w32
+# GCN: early-clobber %2:vreg_256 = V_WMMA_I32_16X16X16_IU8_threeaddr_w32 8, killed %1, 8, killed %1, 8, %0, 0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_i32_16x16x16_iu8_twoaddr_w32
+registers:
+  - { id: 0, class: vreg_256 }
+  - { id: 1, class: vreg_128 }
+  - { id: 2, class: vreg_256 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_256 = V_WMMA_I32_16X16X16_IU8_twoaddr_w32 8, killed %1:vreg_128, 8, killed %1:vreg_128, 8, %0:vreg_256, 0, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_i32_16x16x16_iu4_twoaddr_w32
+# GCN: early-clobber %2:vreg_256 = V_WMMA_I32_16X16X16_IU4_threeaddr_w32 8, killed %1, 8, killed %1, 8, %0, 0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_i32_16x16x16_iu4_twoaddr_w32
+registers:
+  - { id: 0, class: vreg_256 }
+  - { id: 1, class: vreg_64 }
+  - { id: 2, class: vreg_256 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_256 = V_WMMA_I32_16X16X16_IU4_twoaddr_w32 8, killed %1:vreg_64, 8, killed %1:vreg_64, 8, %0:vreg_256, 0, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_f32_16x16x16_f16_twoaddr_w64
+# GCN: early-clobber %2:vreg_128 = V_WMMA_F32_16X16X16_F16_threeaddr_w64 8, killed %1, 8, killed %1, 8, %0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_f32_16x16x16_f16_twoaddr_w64
+registers:
+  - { id: 0, class: vreg_128 }
+  - { id: 1, class: vreg_256 }
+  - { id: 2, class: vreg_128 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_128 = V_WMMA_F32_16X16X16_F16_twoaddr_w64 8, killed %1:vreg_256, 8, killed %1:vreg_256, 8, %0:vreg_128, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_f32_16x16x16_bf16_twoaddr_w64
+# GCN: early-clobber %2:vreg_128 = V_WMMA_F32_16X16X16_BF16_threeaddr_w64 8, killed %1, 8, killed %1, 8, %0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_f32_16x16x16_bf16_twoaddr_w64
+registers:
+  - { id: 0, class: vreg_128 }
+  - { id: 1, class: vreg_256 }
+  - { id: 2, class: vreg_128 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_128 = V_WMMA_F32_16X16X16_BF16_twoaddr_w64 8, killed %1:vreg_256, 8, killed %1:vreg_256, 8, %0:vreg_128, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_f16_16x16x16_f16_twoaddr_w64
+# GCN: early-clobber %2:vreg_128 = V_WMMA_F16_16X16X16_F16_threeaddr_w64 8, killed %1, 8, killed %1, 8, %0, 0, 0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_f16_16x16x16_f16_twoaddr_w64
+registers:
+  - { id: 0, class: vreg_128 }
+  - { id: 1, class: vreg_256 }
+  - { id: 2, class: vreg_128 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_128 = V_WMMA_F16_16X16X16_F16_twoaddr_w64 8, killed %1:vreg_256, 8, killed %1:vreg_256, 8, %0:vreg_128, 0, 0, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_bf16_16x16x16_bf16_twoaddr_w64
+# GCN: early-clobber %2:vreg_128 = V_WMMA_BF16_16X16X16_BF16_threeaddr_w64 8, killed %1, 8, killed %1, 8, %0, 0, 0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_bf16_16x16x16_bf16_twoaddr_w64
+registers:
+  - { id: 0, class: vreg_128 }
+  - { id: 1, class: vreg_256 }
+  - { id: 2, class: vreg_128 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_128 = V_WMMA_BF16_16X16X16_BF16_twoaddr_w64 8, killed %1:vreg_256, 8, killed %1:vreg_256, 8, %0:vreg_128, 0, 0, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_i32_16x16x16_iu8_twoaddr_w64
+# GCN: early-clobber %2:vreg_128 = V_WMMA_I32_16X16X16_IU8_threeaddr_w64 8, killed %1, 8, killed %1, 8, %0, 0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_i32_16x16x16_iu8_twoaddr_w64
+registers:
+  - { id: 0, class: vreg_128 }
+  - { id: 1, class: vreg_128 }
+  - { id: 2, class: vreg_128 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_128 = V_WMMA_I32_16X16X16_IU8_twoaddr_w64 8, killed %1:vreg_128, 8, killed %1:vreg_128, 8, %0:vreg_128, 0, 0, 0, implicit $exec
+
+...
+
+# GCN-LABEL: name: test_v_wmma_i32_16x16x16_iu4_twoaddr_w64
+# GCN: early-clobber %2:vreg_128 = V_WMMA_I32_16X16X16_IU4_threeaddr_w64 8, killed %1, 8, killed %1, 8, %0, 0, 0, 0, implicit $exec
+
+---
+name:            test_v_wmma_i32_16x16x16_iu4_twoaddr_w64
+registers:
+  - { id: 0, class: vreg_128 }
+  - { id: 1, class: vreg_64 }
+  - { id: 2, class: vreg_128 }
+body:             |
+  bb.0:
+
+    %0 = IMPLICIT_DEF
+    %1 = IMPLICIT_DEF
+    early-clobber %2:vreg_128 = V_WMMA_I32_16X16X16_IU4_twoaddr_w64 8, killed %1:vreg_64, 8, killed %1:vreg_64, 8, %0:vreg_128, 0, 0, 0, implicit $exec
+
+...

llvm/test/CodeGen/AMDGPU/wmma-hazards.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=GCN %s
+
+---
+name:            back_to_back_WMMA1_D_overlaps_WMMA2_A
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: back_to_back_WMMA1_D_overlaps_WMMA2_A
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: V_NOP_e32 implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, implicit $exec
+...
+---
+
+---
+name:            back_to_back_WMMA1_D_overlaps_WMMA2_B
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: back_to_back_WMMA1_D_overlaps_WMMA2_B
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: V_NOP_e32 implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, implicit $exec
+...
+---
+
+---
+name:            valu_inbetween_WMMA1_D_overlaps_WMMA2_A
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: valu_inbetween_WMMA1_D_overlaps_WMMA2_A
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: $vgpr40 = V_MOV_B32_e32 0, implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    $vgpr40 = V_MOV_B32_e32 0, implicit $exec
+    early-clobber renamable $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, implicit $exec
+...
+---
+
+---
+name:            value_inbetween_WMMA1_D_overlaps_WMMA2_B
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: value_inbetween_WMMA1_D_overlaps_WMMA2_B
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: $vgpr40 = V_MOV_B32_e32 0, implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    $vgpr40 = V_MOV_B32_e32 0, implicit $exec
+    early-clobber renamable $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, implicit $exec
+...
+---
+
+---
+name:            back_to_back_WMMA1_D_overlaps_WMMA2_C
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: back_to_back_WMMA1_D_overlaps_WMMA2_C
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: V_NOP_e32 implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+...
+---
+
+---
+name:            valu_inbetween_WMMA1_D_overlaps_WMMA2_C
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: valu_inbetween_WMMA1_D_overlaps_WMMA2_C
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: $vgpr40 = V_MOV_B32_e32 0, implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    $vgpr40 = V_MOV_B32_e32 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+...
+---
+
+---
+name:            back_to_back_WMMA1_D_overlaps_WMMA2_C_same_pseudo_instruction_no_imod
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: back_to_back_WMMA1_D_overlaps_WMMA2_C_same_pseudo_instruction_no_imod
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+...
+---
+
+---
+name:            back_to_back_WMMA1_D_overlaps_WMMA2_C_same_real_instruction_no_imod
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: back_to_back_WMMA1_D_overlaps_WMMA2_C_same_real_instruction_no_imod
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47 = V_WMMA_F32_16X16X16_F16_threeaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47 = V_WMMA_F32_16X16X16_F16_threeaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+...
+---
+
+---
+name:            back_to_back_WMMA1_D_overlaps_WMMA2_C_same_pseudo_instruction_imod
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: back_to_back_WMMA1_D_overlaps_WMMA2_C_same_pseudo_instruction_imod
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: V_NOP_e32 implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 11, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 11, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+...
+---
+
+
+---
+name:            back_to_back_WMMA1_D_overlaps_WMMA2_C_diff_pseudo_instruction_no_imod
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: back_to_back_WMMA1_D_overlaps_WMMA2_C_diff_pseudo_instruction_no_imod
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: V_NOP_e32 implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+...
+---
+
+
+---
+name:            back_to_back_WMMA1_D_overlaps_WMMA2_C_diff_pseudo_instruction_imod
+body:             |
+  bb.0:
+  liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39
+    ; GCN-LABEL: name: back_to_back_WMMA1_D_overlaps_WMMA2_C_diff_pseudo_instruction_imod
+    ; GCN: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    ; GCN-NEXT: V_NOP_e32 implicit $exec
+    ; GCN-NEXT: early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 11, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_F16_twoaddr_w32 8, killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, 8, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+    early-clobber renamable $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X16_BF16_twoaddr_w32 8, killed $vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 11, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, implicit $exec
+...
+---

llvm/test/CodeGen/AMDGPU/wmma_multiple_32.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -verify-machineinstrs < %s | FileCheck %s --check-prefix=W32
+
+declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float>, <8 x float> , <8 x float>)
+declare <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32>, <8 x i32> , <8 x float>)
+declare <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float>, <8 x float> , <8 x float>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32>, <8 x i32> , <8 x i32>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 immarg, <4 x i32>, i1 immarg, <4 x i32> , <8 x i32>, i1 immarg)
+declare <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 immarg, <2 x i32>, i1 immarg, <2 x i32> , <8 x i32>, i1 immarg)
+
+; The tests demonstrate that the following WMMA register constraints are satisfied.
+;
+; v_wmma D, A, B, C
+; A and B cannot overlap with D. C cannot partially overlap with D, but it is OK for them to be the same (which is a typical case).
+;
+; In each test,
+;   - first wmma instruction: the dest register D is different than all the sources
+;   - second wmma instruction: the dest register D and src2 (C) are the same
+
+
+; @llvm.amdgcn.wmma.f32.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out, <8 x float> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_f32_16x16x16_f16:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f32_16x16x16_f16 v[28:35], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    v_wmma_f32_16x16x16_f16 v[16:23], v[8:15], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[32:35], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[28:31], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[26:27], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[26:27], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C)
+  %res2 = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float> %B, <8 x float> %B, <8 x float> %C)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  store <8 x float> %res2, <8 x float> addrspace(1)* %out2, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f32.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_bf16(<8 x i32> %A, <8 x i32> %B, <8 x float> %C, <8 x float> addrspace(1)* %out, <8 x float> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_f32_16x16x16_bf16:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f32_16x16x16_bf16 v[28:35], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    v_wmma_f32_16x16x16_bf16 v[16:23], v[8:15], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[32:35], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[28:31], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[26:27], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[26:27], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x float> %C)
+  %res2 = call <8 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32> %B, <8 x i32> %B, <8 x float> %C)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  store <8 x float> %res2, <8 x float> addrspace(1)* %out2, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f16.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_lo(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out, <8 x float> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_f16_16x16x16_f16_lo:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f16_16x16x16_f16 v[28:35], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    v_wmma_f16_16x16x16_f16 v[16:23], v[8:15], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[32:35], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[28:31], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[26:27], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[26:27], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, i1 0)
+  %res2 = call <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %B, <8 x float> %B, <8 x float> %C, i1 0)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  store <8 x float> %res2, <8 x float> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_hi(<8 x float> %A, <8 x float> %B, <8 x float> %C, <8 x float> addrspace(1)* %out, <8 x float> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_f16_16x16x16_f16_hi:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_f16_16x16x16_f16 v[28:35], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+; W32-NEXT:    v_wmma_f16_16x16x16_f16 v[16:23], v[8:15], v[8:15], v[16:23] op_sel:[0,0,1]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[32:35], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[28:31], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[26:27], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[26:27], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <8 x float> %C, i1 1)
+  %res2 = call <8 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %B, <8 x float> %B, <8 x float> %C, i1 1)
+  store <8 x float> %res, <8 x float> addrspace(1)* %out, align 32
+  store <8 x float> %res2, <8 x float> addrspace(1)* %out2, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.bf16.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_lo(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_bf16_16x16x16_bf16_lo:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_bf16_16x16x16_bf16 v[28:35], v[0:7], v[8:15], v[16:23]
+; W32-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:23], v[8:15], v[8:15], v[16:23]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[32:35], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[28:31], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[26:27], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[26:27], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %B, <8 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_hi(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_bf16_16x16x16_bf16_hi:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_bf16_16x16x16_bf16 v[28:35], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+; W32-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:23], v[8:15], v[8:15], v[16:23] op_sel:[0,0,1]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[24:25], v[32:35], off offset:16
+; W32-NEXT:    global_store_b128 v[24:25], v[28:31], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[26:27], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[26:27], v[16:19], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %B, <8 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu8
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[20:27], v[0:3], v[4:7], v[8:15]
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[4:7], v[4:7], v[8:15]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[24:27], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[20:23], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[18:19], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[18:19], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %B, i1 0, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[20:27], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0]
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[4:7], v[4:7], v[8:15] neg_lo:[0,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[24:27], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[20:23], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[18:19], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[18:19], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %B, i1 1, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[20:27], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0]
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[4:7], v[4:7], v[8:15] neg_lo:[1,0,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[24:27], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[20:23], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[18:19], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[18:19], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %B, i1 0, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[20:27], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0]
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[4:7], v[4:7], v[8:15] neg_lo:[1,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[24:27], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[20:23], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[18:19], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[18:19], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %B, i1 1, <4 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[20:27], v[0:3], v[4:7], v[8:15] clamp
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[4:7], v[4:7], v[8:15] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[24:27], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[20:23], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[18:19], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[18:19], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %B, i1 0, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[20:27], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0] clamp
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[4:7], v[4:7], v[8:15] neg_lo:[0,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[24:27], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[20:23], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[18:19], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[18:19], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %B, i1 1, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[20:27], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0] clamp
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[4:7], v[4:7], v[8:15] neg_lo:[1,0,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[24:27], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[20:23], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[18:19], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[18:19], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %B, i1 0, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed_clamp(<4 x i32> %A, <4 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[20:27], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0] clamp
+; W32-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:15], v[4:7], v[4:7], v[8:15] neg_lo:[1,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[16:17], v[24:27], off offset:16
+; W32-NEXT:    global_store_b128 v[16:17], v[20:23], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[18:19], v[12:15], off offset:16
+; W32-NEXT:    global_store_b128 v[18:19], v[8:11], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %B, i1 1, <4 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu4
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[16:23], v[0:1], v[2:3], v[4:11]
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[2:3], v[2:3], v[4:11]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[14:15], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[14:15], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %B, i1 0, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[16:23], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0]
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[2:3], v[2:3], v[4:11] neg_lo:[0,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[14:15], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[14:15], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %B, i1 1, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[16:23], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0]
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[2:3], v[2:3], v[4:11] neg_lo:[1,0,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[14:15], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[14:15], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %B, i1 0, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[16:23], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0]
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[2:3], v[2:3], v[4:11] neg_lo:[1,1,0]
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[14:15], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[14:15], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 0)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %B, i1 1, <2 x i32> %B, <8 x i32> %C, i1 0)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[16:23], v[0:1], v[2:3], v[4:11] clamp
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[2:3], v[2:3], v[4:11] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[14:15], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[14:15], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %B, i1 0, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[16:23], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0] clamp
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[2:3], v[2:3], v[4:11] neg_lo:[0,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[14:15], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[14:15], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %B, i1 1, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[16:23], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0] clamp
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[2:3], v[2:3], v[4:11] neg_lo:[1,0,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[14:15], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[14:15], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %B, i1 0, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed_clamp(<2 x i32> %A, <2 x i32> %B, <8 x i32> %C, <8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %out2) {
+; W32-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed_clamp:
+; W32:       ; %bb.0: ; %bb
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[16:23], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0] clamp
+; W32-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:11], v[2:3], v[2:3], v[4:11] neg_lo:[1,1,0] clamp
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[12:13], v[20:23], off offset:16
+; W32-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W32-NEXT:    s_clause 0x1
+; W32-NEXT:    global_store_b128 v[14:15], v[8:11], off offset:16
+; W32-NEXT:    global_store_b128 v[14:15], v[4:7], off
+; W32-NEXT:    s_endpgm
+bb:
+  %res = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <8 x i32> %C, i1 1)
+  %res2 = call <8 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %B, i1 1, <2 x i32> %B, <8 x i32> %C, i1 1)
+  store <8 x i32> %res, <8 x i32> addrspace(1)* %out, align 32
+  store <8 x i32> %res2, <8 x i32> addrspace(1)* %out2, align 32
+  ret void
+}
+

llvm/test/CodeGen/AMDGPU/wmma_multiple_64.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs < %s | FileCheck %s --check-prefix=W64
+
+declare <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float>, <8 x float>, <4 x float>)
+declare <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32>, <8 x i32>, <4 x float>)
+declare <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float>, <8 x float>, <4 x float>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32>, <8 x i32>, <4 x i32>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 immarg, <4 x i32>, i1 immarg, <4 x i32>, <4 x i32>, i1 immarg)
+declare <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 immarg, <2 x i32>, i1 immarg, <2 x i32>, <4 x i32>, i1 immarg)
+
+; The tests demonstrate that the following WMMA register constraints are satisfied.
+;
+; v_wmma D, A, B, C
+; A and B cannot overlap with D. C cannot partially overlap with D, but it is OK for them to be the same (which is a typical case).
+;
+; In each test,
+;   - first wmma instruction: the dest register D is different than all the sources
+;   - second wmma instruction: the dest register D and src2 (C) are the same
+
+
+; @llvm.amdgcn.wmma.f32.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_f32_16x16x16_f16:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f32_16x16x16_f16 v[24:27], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    v_wmma_f32_16x16x16_f16 v[16:19], v[8:15], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[24:27], off
+; W64-NEXT:    global_store_b128 v[22:23], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C)
+  %res2 = call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.f16(<8 x float> %B, <8 x float> %B, <4 x float> %C)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  store <4 x float> %res2, <4 x float> addrspace(1)* %out2, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f32.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_f32_16x16x16_bf16(<8 x i32> %A, <8 x i32> %B, <4 x float> %C, <4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_f32_16x16x16_bf16:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f32_16x16x16_bf16 v[24:27], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    v_wmma_f32_16x16x16_bf16 v[16:19], v[8:15], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[24:27], off
+; W64-NEXT:    global_store_b128 v[22:23], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x float> %C)
+  %res2 = call <4 x float> @llvm.amdgcn.wmma.f32.16x16x16.bf16(<8 x i32> %B, <8 x i32> %B, <4 x float> %C)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  store <4 x float> %res2, <4 x float> addrspace(1)* %out2, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.f16.16x16x16.f16
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_lo(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_f16_16x16x16_f16_lo:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f16_16x16x16_f16 v[24:27], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    v_wmma_f16_16x16x16_f16 v[16:19], v[8:15], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[24:27], off
+; W64-NEXT:    global_store_b128 v[22:23], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, i1 0)
+  %res2 = call <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %B, <8 x float> %B, <4 x float> %C, i1 0)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  store <4 x float> %res2, <4 x float> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_f16_16x16x16_f16_hi(<8 x float> %A, <8 x float> %B, <4 x float> %C, <4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_f16_16x16x16_f16_hi:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_f16_16x16x16_f16 v[24:27], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+; W64-NEXT:    v_wmma_f16_16x16x16_f16 v[16:19], v[8:15], v[8:15], v[16:19] op_sel:[0,0,1]
+; W64-NEXT:    global_store_b128 v[20:21], v[24:27], off
+; W64-NEXT:    global_store_b128 v[22:23], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %A, <8 x float> %B, <4 x float> %C, i1 1)
+  %res2 = call <4 x float> @llvm.amdgcn.wmma.f16.16x16x16.f16(<8 x float> %B, <8 x float> %B, <4 x float> %C, i1 1)
+  store <4 x float> %res, <4 x float> addrspace(1)* %out, align 16
+  store <4 x float> %res2, <4 x float> addrspace(1)* %out2, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.bf16.16x16x16.bf16
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_lo(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_bf16_16x16x16_bf16_lo:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_bf16_16x16x16_bf16 v[24:27], v[0:7], v[8:15], v[16:19]
+; W64-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:19], v[8:15], v[8:15], v[16:19]
+; W64-NEXT:    global_store_b128 v[20:21], v[24:27], off
+; W64-NEXT:    global_store_b128 v[22:23], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %B, <8 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_bf16_16x16x16_bf16_hi(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_bf16_16x16x16_bf16_hi:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_bf16_16x16x16_bf16 v[24:27], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+; W64-NEXT:    v_wmma_bf16_16x16x16_bf16 v[16:19], v[8:15], v[8:15], v[16:19] op_sel:[0,0,1]
+; W64-NEXT:    global_store_b128 v[20:21], v[24:27], off
+; W64-NEXT:    global_store_b128 v[22:23], v[16:19], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %A, <8 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.bf16.16x16x16.bf16(<8 x i32> %B, <8 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu8
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[16:19], v[0:3], v[4:7], v[8:11]
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[4:7], v[4:7], v[8:11]
+; W64-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W64-NEXT:    global_store_b128 v[14:15], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %B, i1 0, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[16:19], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0]
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[4:7], v[4:7], v[8:11] neg_lo:[0,1,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W64-NEXT:    global_store_b128 v[14:15], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %B, i1 1, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[16:19], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0]
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[4:7], v[4:7], v[8:11] neg_lo:[1,0,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W64-NEXT:    global_store_b128 v[14:15], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %B, i1 0, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[16:19], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0]
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[4:7], v[4:7], v[8:11] neg_lo:[1,1,0]
+; W64-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W64-NEXT:    global_store_b128 v[14:15], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %B, i1 1, <4 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[16:19], v[0:3], v[4:7], v[8:11] clamp
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[4:7], v[4:7], v[8:11] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W64-NEXT:    global_store_b128 v[14:15], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %B, i1 0, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_unsigned_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[16:19], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0] clamp
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[4:7], v[4:7], v[8:11] neg_lo:[0,1,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W64-NEXT:    global_store_b128 v[14:15], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 0, <4 x i32> %B, i1 1, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[16:19], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0] clamp
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[4:7], v[4:7], v[8:11] neg_lo:[1,0,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W64-NEXT:    global_store_b128 v[14:15], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 0, <4 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %B, i1 0, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui8_signed_signed_clamp(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui8_signed_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[16:19], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0] clamp
+; W64-NEXT:    v_wmma_i32_16x16x16_iu8 v[8:11], v[4:7], v[4:7], v[8:11] neg_lo:[1,1,0] clamp
+; W64-NEXT:    global_store_b128 v[12:13], v[16:19], off
+; W64-NEXT:    global_store_b128 v[14:15], v[8:11], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %A, i1 1, <4 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu8(i1 1, <4 x i32> %B, i1 1, <4 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+; @llvm.amdgcn.wmma.i32.16x16x16.iu4
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[12:15], v[0:1], v[2:3], v[4:7]
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[2:3], v[2:3], v[4:7]
+; W64-NEXT:    global_store_b128 v[8:9], v[12:15], off
+; W64-NEXT:    global_store_b128 v[10:11], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %B, i1 0, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[12:15], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0]
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[2:3], v[2:3], v[4:7] neg_lo:[0,1,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[12:15], off
+; W64-NEXT:    global_store_b128 v[10:11], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %B, i1 1, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[12:15], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0]
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[2:3], v[2:3], v[4:7] neg_lo:[1,0,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[12:15], off
+; W64-NEXT:    global_store_b128 v[10:11], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %B, i1 0, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[12:15], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0]
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[2:3], v[2:3], v[4:7] neg_lo:[1,1,0]
+; W64-NEXT:    global_store_b128 v[8:9], v[12:15], off
+; W64-NEXT:    global_store_b128 v[10:11], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 0)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %B, i1 1, <2 x i32> %B, <4 x i32> %C, i1 0)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[12:15], v[0:1], v[2:3], v[4:7] clamp
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[2:3], v[2:3], v[4:7] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[12:15], off
+; W64-NEXT:    global_store_b128 v[10:11], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %B, i1 0, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_unsigned_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[12:15], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0] clamp
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[2:3], v[2:3], v[4:7] neg_lo:[0,1,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[12:15], off
+; W64-NEXT:    global_store_b128 v[10:11], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 0, <2 x i32> %B, i1 1, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_unsigned_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[12:15], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0] clamp
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[2:3], v[2:3], v[4:7] neg_lo:[1,0,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[12:15], off
+; W64-NEXT:    global_store_b128 v[10:11], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 0, <2 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %B, i1 0, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+
+define amdgpu_ps void @test_wmma_i32_16x16x16_ui4_signed_signed_clamp(<2 x i32> %A, <2 x i32> %B, <4 x i32> %C, <4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %out2) {
+; W64-LABEL: test_wmma_i32_16x16x16_ui4_signed_signed_clamp:
+; W64:       ; %bb.0: ; %bb
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[12:15], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0] clamp
+; W64-NEXT:    v_wmma_i32_16x16x16_iu4 v[4:7], v[2:3], v[2:3], v[4:7] neg_lo:[1,1,0] clamp
+; W64-NEXT:    global_store_b128 v[8:9], v[12:15], off
+; W64-NEXT:    global_store_b128 v[10:11], v[4:7], off
+; W64-NEXT:    s_endpgm
+bb:
+  %res = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %A, i1 1, <2 x i32> %B, <4 x i32> %C, i1 1)
+  %res2 = call <4 x i32> @llvm.amdgcn.wmma.i32.16x16x16.iu4(i1 1, <2 x i32> %B, i1 1, <2 x i32> %B, <4 x i32> %C, i1 1)
+  store <4 x i32> %res, <4 x i32> addrspace(1)* %out, align 16
+  store <4 x i32> %res2, <4 x i32> addrspace(1)* %out2, align 16
+  ret void
+}
+

llvm/test/MC/AMDGPU/gfx11_wmma.s

+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -show-encoding %s | FileCheck --check-prefix=W32 %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -show-encoding %s | FileCheck --check-prefix=W64 %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 %s 2>&1 | FileCheck --check-prefix=W32-ERR --implicit-check-not=error: %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 %s 2>&1 | FileCheck --check-prefix=W64-ERR --implicit-check-not=error: %s
+
+//
+// Test v_wmma_f32_16x16x16_f16
+//
+
+v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+// W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0x42,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]
+// W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0x42,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:23], 1.0, v[8:15], v[16:23]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_f16 v[16:19], 1.0, v[8:15], v[16:19]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], 1.0, v[16:23]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], 1.0, v[16:19]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], 1.0
+// W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0xca,0x1b]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], 1.0
+// W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0xca,0x1b]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x40,0xcc,0x00,0x11,0x42,0x3c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x40,0xcc,0x00,0x11,0x42,0x3c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x40,0xcc,0x00,0x11,0x42,0x5c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x40,0xcc,0x00,0x11,0x42,0x5c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x40,0xcc,0x00,0x11,0x42,0x7c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x40,0xcc,0x00,0x11,0x42,0x7c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] clamp
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] clamp
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+//
+// Test v_wmma_f32_16x16x16_bf16
+//
+
+v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+// W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0x42,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19]
+// W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0x42,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:23], 1.0, v[8:15], v[16:23]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_bf16 v[16:19], 1.0, v[8:15], v[16:19]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], 1.0, v[16:23]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], 1.0, v[16:19]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], 1.0
+// W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0xca,0x1b]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], 1.0
+// W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0xca,0x1b]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x41,0xcc,0x00,0x11,0x42,0x3c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x41,0xcc,0x00,0x11,0x42,0x3c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x41,0xcc,0x00,0x11,0x42,0x5c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x41,0xcc,0x00,0x11,0x42,0x5c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x41,0xcc,0x00,0x11,0x42,0x7c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x41,0xcc,0x00,0x11,0x42,0x7c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] clamp
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] clamp
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+//
+// Test v_wmma_f16_16x16x16_f16
+//
+
+v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+// W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0x42,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]
+// W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0x42,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:23], 1.0, v[8:15], v[16:23]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f16_16x16x16_f16 v[16:19], 1.0, v[8:15], v[16:19]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], 1.0, v[16:23]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], 1.0, v[16:19]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], 1.0
+// W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0xca,0x1b]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], 1.0
+// W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0xca,0x1b]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+// W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1] ; encoding: [0x10,0x60,0x42,0xcc,0x00,0x11,0x42,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+// W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1] ; encoding: [0x10,0x60,0x42,0xcc,0x00,0x11,0x42,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x42,0xcc,0x00,0x11,0x42,0x3c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x42,0xcc,0x00,0x11,0x42,0x3c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x42,0xcc,0x00,0x11,0x42,0x5c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x42,0xcc,0x00,0x11,0x42,0x5c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x42,0xcc,0x00,0x11,0x42,0x7c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x42,0xcc,0x00,0x11,0x42,0x7c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] clamp
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] clamp
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+//
+// Test v_wmma_bf16_16x16x16_bf16
+//
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+// W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0x42,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19]
+// W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0x42,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], 1.0, v[8:15], v[16:23]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], 1.0, v[8:15], v[16:19]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], 1.0, v[16:23]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], 1.0, v[16:19]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], 1.0
+// W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0xca,0x1b]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], 1.0
+// W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0xca,0x1b]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+// W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1] ; encoding: [0x10,0x60,0x43,0xcc,0x00,0x11,0x42,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+// W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1] ; encoding: [0x10,0x60,0x43,0xcc,0x00,0x11,0x42,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x43,0xcc,0x00,0x11,0x42,0x3c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x43,0xcc,0x00,0x11,0x42,0x3c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x43,0xcc,0x00,0x11,0x42,0x5c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x43,0xcc,0x00,0x11,0x42,0x5c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x43,0xcc,0x00,0x11,0x42,0x7c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x43,0xcc,0x00,0x11,0x42,0x7c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] clamp
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] clamp
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+//
+// Test v_wmma_i32_16x16x16_iu8
+//
+
+v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15]
+// W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x22,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11]
+// W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x22,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:15], 1, v[4:7], v[8:15]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu8 v[8:11], 1, v[4:7], v[8:11]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], 1, v[8:15]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], 1, v[8:11]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], 1
+// W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], 1 ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x06,0x1a]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], 1
+// W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], 1 ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x06,0x1a]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu8 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x08,0x41,0x44,0xcc,0x00,0x09,0x22,0x3c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x08,0x41,0x44,0xcc,0x00,0x09,0x22,0x3c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x08,0x42,0x44,0xcc,0x00,0x09,0x22,0x5c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x08,0x42,0x44,0xcc,0x00,0x09,0x22,0x5c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x08,0x43,0x44,0xcc,0x00,0x09,0x22,0x7c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x08,0x43,0x44,0xcc,0x00,0x09,0x22,0x7c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] clamp
+// W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] clamp ; encoding: [0x08,0xc0,0x44,0xcc,0x00,0x09,0x22,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] clamp
+// W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] clamp ; encoding: [0x08,0xc0,0x44,0xcc,0x00,0x09,0x22,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+//
+// Test v_wmma_i32_16x16x16_iu4
+//
+
+v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11]
+// W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x12,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7]
+// W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x12,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:11], 1, v[2:3], v[4:11]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu4 v[4:7], 1, v[2:3], v[4:7]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], 1, v[4:11]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], 1, v[4:7]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], 1
+// W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], 1 ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x06,0x1a]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], 1
+// W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], 1 ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x06,0x1a]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu4 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]
+// W32-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: invalid operand for instruction
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: invalid operand for instruction
+
+v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x04,0x41,0x45,0xcc,0x00,0x05,0x12,0x3c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0] neg_hi:[1,0,0]
+// W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x04,0x41,0x45,0xcc,0x00,0x05,0x12,0x3c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x04,0x42,0x45,0xcc,0x00,0x05,0x12,0x5c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0] neg_hi:[0,1,0]
+// W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x04,0x42,0x45,0xcc,0x00,0x05,0x12,0x5c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x04,0x43,0x45,0xcc,0x00,0x05,0x12,0x7c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0] neg_hi:[1,1,0]
+// W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x04,0x43,0x45,0xcc,0x00,0x05,0x12,0x7c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] clamp
+// W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] clamp ; encoding: [0x04,0xc0,0x45,0xcc,0x00,0x05,0x12,0x1c]
+// W64-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+
+v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] clamp
+// W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] clamp ; encoding: [0x04,0xc0,0x45,0xcc,0x00,0x05,0x12,0x1c]
+// W32-ERR: :[[@LINE-2]]:{{[0-9]+}}: error: operands are not valid for this GPU or mode
+

llvm/test/MC/Disassembler/AMDGPU/gfx11_wmma.txt

+# RUN: llvm-mc -arch=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=W32 %s
+# RUN: llvm-mc -arch=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=W64 %s
+
+
+# Test v_wmma_f32_16x16x16_f16
+
+# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]                               ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0x42,0x1c]
+# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]                               ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0x42,0x1c]
+0x10,0x40,0x40,0xcc,0x00,0x11,0x42,0x1c
+
+# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], 1.0                                    ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0xca,0x1b]
+# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], 1.0                                    ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0xca,0x1b]
+0x10,0x40,0x40,0xcc,0x00,0x11,0xca,0x1b
+
+# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x40,0xcc,0x00,0x11,0x42,0x3c]
+# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x40,0xcc,0x00,0x11,0x42,0x3c]
+0x10,0x41,0x40,0xcc,0x00,0x11,0x42,0x3c
+
+# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x40,0xcc,0x00,0x11,0x42,0x5c]
+# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x40,0xcc,0x00,0x11,0x42,0x5c]
+0x10,0x42,0x40,0xcc,0x00,0x11,0x42,0x5c
+
+# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x40,0xcc,0x00,0x11,0x42,0x7c]
+# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x40,0xcc,0x00,0x11,0x42,0x7c]
+0x10,0x43,0x40,0xcc,0x00,0x11,0x42,0x7c
+
+
+# Test v_wmma_f32_16x16x16_bf16
+
+# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23]                               ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0x42,0x1c]
+# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19]                               ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0x42,0x1c]
+0x10,0x40,0x41,0xcc,0x00,0x11,0x42,0x1c
+
+# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], 1.0                                    ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0xca,0x1b]
+# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], 1.0                                    ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0xca,0x1b]
+0x10,0x40,0x41,0xcc,0x00,0x11,0xca,0x1b
+
+# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x41,0xcc,0x00,0x11,0x42,0x3c]
+# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x41,0xcc,0x00,0x11,0x42,0x3c]
+0x10,0x41,0x41,0xcc,0x00,0x11,0x42,0x3c
+
+# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x41,0xcc,0x00,0x11,0x42,0x5c]
+# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x41,0xcc,0x00,0x11,0x42,0x5c]
+0x10,0x42,0x41,0xcc,0x00,0x11,0x42,0x5c
+
+# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x41,0xcc,0x00,0x11,0x42,0x7c]
+# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x41,0xcc,0x00,0x11,0x42,0x7c]
+0x10,0x43,0x41,0xcc,0x00,0x11,0x42,0x7c
+
+
+# Test v_wmma_f16_16x16x16_f16
+
+# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]                                ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0x42,0x1c]
+# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]                                ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0x42,0x1c]
+0x10,0x40,0x42,0xcc,0x00,0x11,0x42,0x1c
+
+# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], 1.0                                     ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0xca,0x1b]
+# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], 1.0                                     ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0xca,0x1b]
+0x10,0x40,0x42,0xcc,0x00,0x11,0xca,0x1b
+
+# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]                 ; encoding: [0x10,0x60,0x42,0xcc,0x00,0x11,0x42,0x1c]
+# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]                 ; encoding: [0x10,0x60,0x42,0xcc,0x00,0x11,0x42,0x1c]
+0x10,0x60,0x42,0xcc,0x00,0x11,0x42,0x1c
+
+# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]  ; encoding: [0x10,0x41,0x42,0xcc,0x00,0x11,0x42,0x3c]
+# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]  ; encoding: [0x10,0x41,0x42,0xcc,0x00,0x11,0x42,0x3c]
+0x10,0x41,0x42,0xcc,0x00,0x11,0x42,0x3c
+
+# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]  ; encoding: [0x10,0x42,0x42,0xcc,0x00,0x11,0x42,0x5c]
+# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]  ; encoding: [0x10,0x42,0x42,0xcc,0x00,0x11,0x42,0x5c]
+0x10,0x42,0x42,0xcc,0x00,0x11,0x42,0x5c
+
+# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0]  ; encoding: [0x10,0x43,0x42,0xcc,0x00,0x11,0x42,0x7c]
+# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0]  ; encoding: [0x10,0x43,0x42,0xcc,0x00,0x11,0x42,0x7c]
+0x10,0x43,0x42,0xcc,0x00,0x11,0x42,0x7c
+
+
+# Test v_wmma_bf16_16x16x16_bf16
+
+# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0x42,0x1c]
+# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0x42,0x1c]
+0x10,0x40,0x43,0xcc,0x00,0x11,0x42,0x1c
+
+# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0xca,0x1b]
+# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0xca,0x1b]
+0x10,0x40,0x43,0xcc,0x00,0x11,0xca,0x1b
+
+# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1] ; encoding: [0x10,0x60,0x43,0xcc,0x00,0x11,0x42,0x1c]
+# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1] ; encoding: [0x10,0x60,0x43,0xcc,0x00,0x11,0x42,0x1c]
+0x10,0x60,0x43,0xcc,0x00,0x11,0x42,0x1c
+
+# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x43,0xcc,0x00,0x11,0x42,0x3c]
+# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x43,0xcc,0x00,0x11,0x42,0x3c]
+0x10,0x41,0x43,0xcc,0x00,0x11,0x42,0x3c
+
+# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x43,0xcc,0x00,0x11,0x42,0x5c]
+# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x43,0xcc,0x00,0x11,0x42,0x5c]
+0x10,0x42,0x43,0xcc,0x00,0x11,0x42,0x5c
+
+# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x43,0xcc,0x00,0x11,0x42,0x7c]
+# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x43,0xcc,0x00,0x11,0x42,0x7c]
+0x10,0x43,0x43,0xcc,0x00,0x11,0x42,0x7c
+
+
+# Test v_wmma_i32_16x16x16_iu8
+
+# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15]                               ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x22,0x1c]
+# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11]                               ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x22,0x1c]
+0x08,0x40,0x44,0xcc,0x00,0x09,0x22,0x1c
+
+# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], 1                                      ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x06,0x1a]
+# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], 1                                      ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x06,0x1a]
+0x08,0x40,0x44,0xcc,0x00,0x09,0x06,0x1a
+
+# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x08,0x41,0x44,0xcc,0x00,0x09,0x22,0x3c]
+# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x08,0x41,0x44,0xcc,0x00,0x09,0x22,0x3c]
+0x08,0x41,0x44,0xcc,0x00,0x09,0x22,0x3c
+
+# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x08,0x42,0x44,0xcc,0x00,0x09,0x22,0x5c]
+# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x08,0x42,0x44,0xcc,0x00,0x09,0x22,0x5c]
+0x08,0x42,0x44,0xcc,0x00,0x09,0x22,0x5c
+
+# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x08,0x43,0x44,0xcc,0x00,0x09,0x22,0x7c]
+# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x08,0x43,0x44,0xcc,0x00,0x09,0x22,0x7c]
+0x08,0x43,0x44,0xcc,0x00,0x09,0x22,0x7c
+
+# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] clamp                         ; encoding: [0x08,0xc0,0x44,0xcc,0x00,0x09,0x22,0x1c]
+# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] clamp                         ; encoding: [0x08,0xc0,0x44,0xcc,0x00,0x09,0x22,0x1c]
+0x08,0xc0,0x44,0xcc,0x00,0x09,0x22,0x1c
+
+
+# Test v_wmma_i32_16x16x16_iu4
+
+# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11]                             ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x12,0x1c]
+# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7]                               ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x12,0x1c]
+0x04,0x40,0x45,0xcc,0x00,0x05,0x12,0x1c
+
+# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], 1                                     ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x06,0x1a]
+# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], 1                                      ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x06,0x1a]
+0x04,0x40,0x45,0xcc,0x00,0x05,0x06,0x1a
+
+# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x04,0x41,0x45,0xcc,0x00,0x05,0x12,0x3c]
+# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x04,0x41,0x45,0xcc,0x00,0x05,0x12,0x3c]
+0x04,0x41,0x45,0xcc,0x00,0x05,0x12,0x3c
+
+# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x04,0x42,0x45,0xcc,0x00,0x05,0x12,0x5c]
+# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x04,0x42,0x45,0xcc,0x00,0x05,0x12,0x5c]
+0x04,0x42,0x45,0xcc,0x00,0x05,0x12,0x5c
+
+# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x04,0x43,0x45,0xcc,0x00,0x05,0x12,0x7c]
+# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x04,0x43,0x45,0xcc,0x00,0x05,0x12,0x7c]
+0x04,0x43,0x45,0xcc,0x00,0x05,0x12,0x7c
+
+# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] clamp                         ; encoding: [0x04,0xc0,0x45,0xcc,0x00,0x05,0x12,0x1c]
+# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] clamp                         ; encoding: [0x04,0xc0,0x45,0xcc,0x00,0x05,0x12,0x1c]
+0x04,0xc0,0x45,0xcc,0x00,0x05,0x12,0x1c
+