AMDGPU/GlobalISel: Add AGPR bank and RegBankSelect mfma intrinsics

[arsenm]

You may be able to re-use the existing DAG select tests at this point?

llvm/lib/Target/AMDGPU/AMDGPUGenRegisterBankInfo.def

35

I think a 96 bit one is needed for symmetry to legalize all the vector operations

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

176

Indented too much, and no else after return

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.h

139

This can be static

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

119

I think the cost actually needs to go up if src and dst are AGPRs since there is no direct copy between them

[kerbowa]

Address comments.

[kerbowa]

In D70871#1764421, @arsenm wrote:

You may be able to re-use the existing DAG select tests at this point?

We need to work on legalizing/selecting wide stores first.

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUGenRegisterBankInfo.def

36

This is going to cause an annoying merge conflict with one of my patches. It would be nice to start finally having tablegen emit this

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

143–144

Should just directly check if they are AGPR. This incorrectly covers VGPR copies?

2125

Extra newline

[kerbowa]

Fix AGPR to AGPR copy cost.

[arsenm]

LGTM

[kerbowa]

Fix potential merge conflict.

[arsenm]

LGTM, I don't think it's avoidable since the renumbering of the hardcoded indexes is the painful part

llvm/lib/Target/AMDGPU/AMDGPUGenRegisterBankInfo.def

enum PartialMappingIdx {
   PM_VGPR16 = 17,
   PM_VGPR32 = 18,
   PM_VGPR64 = 19,
   PM_VGPR128 = 20,
   PM_VGPR256 = 21,
   PM_VGPR512 = 22,
   PM_VGPR1024 = 23,
   PM_SGPR96 = 24,
-  PM_VGPR96 = 25
+  PM_VGPR96 = 25,
+  PM_AGPR32 = 31,
+  PM_AGPR64 = 32,
+  PM_AGPR128 = 33,
+  PM_AGPR512 = 35,
+  PM_AGPR1024 = 36
 };
 
 const RegisterBankInfo::PartialMapping PartMappings[] {
   // StartIdx, Length, RegBank
   {0, 1,  SCCRegBank},
   {0, 1,  VCCRegBank},
 
   {0, 1,  SGPRRegBank}, // SGPR begin
const RegisterBankInfo::PartialMapping PartMappings[] {
   {0, 16, VGPRRegBank},
   {0, 32, VGPRRegBank},
   {0, 64, VGPRRegBank},
   {0, 128, VGPRRegBank},
   {0, 256, VGPRRegBank},
   {0, 512, VGPRRegBank},
   {0, 1024, VGPRRegBank},
   {0, 96, SGPRRegBank},
-  {0, 96, VGPRRegBank}
-};
+  {0, 96, VGPRRegBank},
 
+  {0, 32, AGPRRegBank}, // AGPR begin
+  {0, 64, AGPRRegBank},
+  {0, 128, AGPRRegBank},
+  {0, 512, AGPRRegBank},
+  {0, 1024, AGPRRegBank}
+};
 const RegisterBankInfo::ValueMapping ValMappings[] {
   // SCC
   {&PartMappings[0], 1},
 
   // VCC
   {&PartMappings[1], 1},
 
   // SGPRs
const RegisterBankInfo::ValueMapping ValMappings[] {
   {&PartMappings[11], 1}, // 16
   {&PartMappings[12], 1}, // 32
   {&PartMappings[13], 1}, // 64
   {&PartMappings[14], 1}, // 128
   {&PartMappings[15], 1}, // 256
   {&PartMappings[16], 1}, // 512
   {&PartMappings[17], 1}, // 1024
   {&PartMappings[18], 1},
-  {&PartMappings[19], 1}
+  {&PartMappings[19], 1},
+
+  // AGPRs
+  {nullptr, 0},
+  {nullptr, 0},
+  {nullptr, 0},
+  {nullptr, 0},
+  {nullptr, 0},
+  {&PartMappings[20], 1}, // 32
+  {&PartMappings[21], 1}, // 64
+  {&PartMappings[22], 1}, // 128
+  {nullptr, 0},
+  {&PartMappings[23], 1}, // 512
+  {&PartMappings[24], 1}, // 1024
 };
 
 const RegisterBankInfo::PartialMapping SGPROnly64BreakDown[] {
   {0, 32, SGPRRegBank}, // 32-bit op
   {0, 32, SGPRRegBank}, // 2x32-bit op
   {32, 32, SGPRRegBank},
   {0, 64, SGPRRegBank}, // <2x32-bit> op
 
const RegisterBankInfo::ValueMapping ValMappingsSGPR64OnlyVGPR32[] {
 
   /*32-bit vgpr*/     {&SGPROnly64BreakDown[4], 1},
   /*2 x 32-bit vgpr*/ {&SGPROnly64BreakDown[5], 2}
 };
 
 enum ValueMappingIdx {
   SCCStartIdx = 0,
   SGPRStartIdx = 2,
-  VGPRStartIdx = 13
+  VGPRStartIdx = 13,
+  AGPRStartIdx = 26
 };
 
 const RegisterBankInfo::ValueMapping *getValueMapping(unsigned BankID,
                                                       unsigned Size) {
   unsigned Idx;
   switch (Size) {
   case 1:
     if (BankID == AMDGPU::SCCRegBankID)
       return &ValMappings[0];
     if (BankID == AMDGPU::VCCRegBankID)
       return &ValMappings[1];
 
     // 1-bit values not from a compare etc.
     Idx = BankID == AMDGPU::SGPRRegBankID ? PM_SGPR1 : PM_VGPR1;
     break;
   case 96:
     assert(BankID != AMDGPU::VCCRegBankID);
     Idx = BankID == AMDGPU::SGPRRegBankID ? PM_SGPR96 : PM_VGPR96;
     break;
   default:
     assert(BankID != AMDGPU::VCCRegBankID);
-    Idx = BankID == AMDGPU::VGPRRegBankID ? VGPRStartIdx : SGPRStartIdx;
+
+    switch (BankID) {
+      case AMDGPU::VGPRRegBankID:
+        Idx = VGPRStartIdx;
+        break;
+      case AMDGPU::SGPRRegBankID:
+        Idx = SGPRStartIdx;
+        break;
+      case AMDGPU::AGPRRegBankID:
+        Idx = AGPRStartIdx;
+        break;
+      default: llvm_unreachable("Invalid register bank");
+    }
+
     Idx += Log2_32_Ceil(Size);
     break;
   }
 
   assert(Log2_32_Ceil(Size) == Log2_32_Ceil(ValMappings[Idx].BreakDown->Length));
   assert(BankID == ValMappings[Idx].BreakDown->RegBank->getID());
 
   return &ValMappings[Idx];

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.h

const ValueMapping *getSGPROpMapping(Register Reg,
                                        const MachineRegisterInfo &MRI,
                                        const TargetRegisterInfo &TRI) const;
 
   // Return a value mapping for an operand that is required to be a VGPR.
   const ValueMapping *getVGPROpMapping(Register Reg,
                                        const MachineRegisterInfo &MRI,
                                        const TargetRegisterInfo &TRI) const;
 
+  // Return a value mapping for an operand that is required to be a AGPR.
+  const ValueMapping *getAGPROpMapping(Register Reg,
+                                       const MachineRegisterInfo &MRI,
+                                       const TargetRegisterInfo &TRI) const;
+
   /// Split 64-bit value \p Reg into two 32-bit halves and populate them into \p
   /// Regs. This appropriately sets the regbank of the new registers.
   void split64BitValueForMapping(MachineIRBuilder &B,
                                  SmallVector<Register, 2> &Regs,
                                  LLT HalfTy,
                                  Register Reg) const;
 
   template <unsigned NumOps>
class AMDGPURegisterBankInfo : public AMDGPUGenRegisterBankInfo {
   getInstrAlternativeMappingsIntrinsic(
       const MachineInstr &MI, const MachineRegisterInfo &MRI) const;
 
   RegisterBankInfo::InstructionMappings
   getInstrAlternativeMappingsIntrinsicWSideEffects(
       const MachineInstr &MI, const MachineRegisterInfo &MRI) const;
 
   bool isSALUMapping(const MachineInstr &MI) const;
+  bool isVectorRegisterBank(const RegisterBank &Bank) const;
+
   const InstructionMapping &getDefaultMappingSOP(const MachineInstr &MI) const;
   const InstructionMapping &getDefaultMappingVOP(const MachineInstr &MI) const;
   const InstructionMapping &getDefaultMappingAllVGPR(
     const MachineInstr &MI) const;
 
   const InstructionMapping &getImageMapping(const MachineRegisterInfo &MRI,
                                             const MachineInstr &MI,
                                             int RsrcIdx) const;

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

AMDGPURegisterBankInfo::AMDGPURegisterBankInfo(const GCNSubtarget &ST)
   const RegisterBank &RBSGPR = getRegBank(AMDGPU::SGPRRegBankID);
   (void)RBSGPR;
   assert(&RBSGPR == &AMDGPU::SGPRRegBank);
 
   const RegisterBank &RBVGPR = getRegBank(AMDGPU::VGPRRegBankID);
   (void)RBVGPR;
   assert(&RBVGPR == &AMDGPU::VGPRRegBank);
 
+  const RegisterBank &RBAGPR = getRegBank(AMDGPU::AGPRRegBankID);
+  (void)RBAGPR;
+  assert(&RBAGPR == &AMDGPU::AGPRRegBank);
 }
 
 unsigned AMDGPURegisterBankInfo::copyCost(const RegisterBank &Dst,
                                           const RegisterBank &Src,
                                           unsigned Size) const {
   // TODO: Should there be a UniformVGPRRegBank which can use readfirstlane?
   if (Dst.getID() == AMDGPU::SGPRRegBankID &&
-      Src.getID() == AMDGPU::VGPRRegBankID) {
+      isVectorRegisterBank(Src)) {
     return std::numeric_limits<unsigned>::max();
   }
 
   // Bool values are tricky, because the meaning is based on context. The SCC
   // and VCC banks are for the natural scalar and vector conditions produced by
   // a compare.
   //
   // Legalization doesn't know about the necessary context, so an s1 use may
   // have been a truncate from an arbitrary value, in which case a copy (lowered
   // as a compare with 0) needs to be inserted.
   if (Size == 1 &&
       (Dst.getID() == AMDGPU::SCCRegBankID ||
        Dst.getID() == AMDGPU::SGPRRegBankID) &&
-      (Src.getID() == AMDGPU::SGPRRegBankID ||
-       Src.getID() == AMDGPU::VGPRRegBankID ||
+      (isVectorRegisterBank(Src) ||
+       Src.getID() == AMDGPU::SGPRRegBankID ||
        Src.getID() == AMDGPU::VCCRegBankID))
     return std::numeric_limits<unsigned>::max();
 
   if (Dst.getID() == AMDGPU::SCCRegBankID &&
       Src.getID() == AMDGPU::VCCRegBankID)
     return std::numeric_limits<unsigned>::max();
 
   return RegisterBankInfo::copyCost(Dst, Src, Size);
 }
 
 unsigned AMDGPURegisterBankInfo::getBreakDownCost(
   const ValueMapping &ValMapping,
   const RegisterBank *CurBank) const {
   // Check if this is a breakdown for G_LOAD to move the pointer from SGPR to
   // VGPR.
   // FIXME: Is there a better way to do this?
   if (ValMapping.NumBreakDowns >= 2 || ValMapping.BreakDown[0].Length >= 64)
     return 10; // This is expensive.
unsigned AMDGPURegisterBankInfo::getBreakDownCost(
   return 1;
 }
 
 const RegisterBank &AMDGPURegisterBankInfo::getRegBankFromRegClass(
     const TargetRegisterClass &RC) const {
   if (&RC == &AMDGPU::SReg_1RegClass)
     return AMDGPU::VCCRegBank;
 
-  return TRI->isSGPRClass(&RC) ? AMDGPU::SGPRRegBank : AMDGPU::VGPRRegBank;
+  if (TRI->isSGPRClass(&RC))
+      return AMDGPU::SGPRRegBank;
+  else if (TRI->isAGPRClass(&RC))
+      return AMDGPU::AGPRRegBank;
+  else
+    return AMDGPU::VGPRRegBank;
 }
 
 template <unsigned NumOps>
 RegisterBankInfo::InstructionMappings
 AMDGPURegisterBankInfo::addMappingFromTable(
     const MachineInstr &MI, const MachineRegisterInfo &MRI,
     const std::array<unsigned, NumOps> RegSrcOpIdx,
     ArrayRef<OpRegBankEntry<NumOps>> Table) const {

 bool AMDGPURegisterBankInfo::isSALUMapping(const MachineInstr &MI) const {
   const MachineFunction &MF = *MI.getParent()->getParent();
   const MachineRegisterInfo &MRI = MF.getRegInfo();
   for (unsigned i = 0, e = MI.getNumOperands();i != e; ++i) {
     if (!MI.getOperand(i).isReg())
       continue;
     Register Reg = MI.getOperand(i).getReg();
     if (const RegisterBank *Bank = getRegBank(Reg, MRI, *TRI)) {
-      if (Bank->getID() == AMDGPU::VGPRRegBankID)
+      if (isVectorRegisterBank(*Bank))
         return false;
 
       assert(Bank->getID() == AMDGPU::SGPRRegBankID ||
              Bank->getID() == AMDGPU::SCCRegBankID);
     }
   }
   return true;
 }
 
+bool AMDGPURegisterBankInfo::isVectorRegisterBank(const RegisterBank &Bank) const {
+  unsigned BankID = Bank.getID();
+  return BankID == AMDGPU::VGPRRegBankID || BankID == AMDGPU::AGPRRegBankID;
+}
+
 const RegisterBankInfo::InstructionMapping &
 AMDGPURegisterBankInfo::getDefaultMappingSOP(const MachineInstr &MI) const {
   const MachineFunction &MF = *MI.getParent()->getParent();
   const MachineRegisterInfo &MRI = MF.getRegInfo();
   SmallVector<const ValueMapping*, 8> OpdsMapping(MI.getNumOperands());
 
   for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
     unsigned Size = getSizeInBits(MI.getOperand(i).getReg(), MRI, *TRI);
AMDGPURegisterBankInfo::getInstrMappingForLoad(const MachineInstr &MI) const {
   // handle that during instruction selection?
 }
 
 unsigned
 AMDGPURegisterBankInfo::getRegBankID(Register Reg,
                                      const MachineRegisterInfo &MRI,
                                      const TargetRegisterInfo &TRI,
                                      unsigned Default) const {
-
   const RegisterBank *Bank = getRegBank(Reg, MRI, TRI);
   return Bank ? Bank->getID() : Default;
 }
 
 
 static unsigned regBankUnion(unsigned RB0, unsigned RB1) {
   return (RB0 == AMDGPU::SGPRRegBankID && RB1 == AMDGPU::SGPRRegBankID) ?
     AMDGPU::SGPRRegBankID : AMDGPU::VGPRRegBankID;

 const RegisterBankInfo::ValueMapping *
 AMDGPURegisterBankInfo::getVGPROpMapping(Register Reg,
                                          const MachineRegisterInfo &MRI,
                                          const TargetRegisterInfo &TRI) const {
   unsigned Size = getSizeInBits(Reg, MRI, TRI);
   return AMDGPU::getValueMapping(AMDGPU::VGPRRegBankID, Size);
 }
 
+const RegisterBankInfo::ValueMapping *
+AMDGPURegisterBankInfo::getAGPROpMapping(Register Reg,
+                                         const MachineRegisterInfo &MRI,
+                                         const TargetRegisterInfo &TRI) const {
+  unsigned Size = getSizeInBits(Reg, MRI, TRI);
+  return AMDGPU::getValueMapping(AMDGPU::AGPRRegBankID, Size);
+}
+
+
 ///
 /// This function must return a legal mapping, because
 /// AMDGPURegisterBankInfo::getInstrAlternativeMappings() is not called
 /// in RegBankSelect::Mode::Fast.  Any mapping that would cause a
 /// VGPR to SGPR generated is illegal.
 ///
 const RegisterBankInfo::InstructionMapping &
 AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
case AMDGPU::G_INTRINSIC: {
     }
     case Intrinsic::amdgcn_if_break: {
       unsigned Size = getSizeInBits(MI.getOperand(0).getReg(), MRI, *TRI);
       OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
       OpdsMapping[2] = AMDGPU::getValueMapping(AMDGPU::VCCRegBankID, 1);
       OpdsMapping[3] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
       break;
     }
+    case Intrinsic::amdgcn_mfma_f32_4x4x1f32:
+    case Intrinsic::amdgcn_mfma_f32_4x4x4f16:
+    case Intrinsic::amdgcn_mfma_i32_4x4x4i8:
+    case Intrinsic::amdgcn_mfma_f32_4x4x2bf16:
+    case Intrinsic::amdgcn_mfma_f32_16x16x1f32:
+    case Intrinsic::amdgcn_mfma_f32_16x16x4f32:
+    case Intrinsic::amdgcn_mfma_f32_16x16x4f16:
+    case Intrinsic::amdgcn_mfma_f32_16x16x16f16:
+    case Intrinsic::amdgcn_mfma_i32_16x16x4i8:
+    case Intrinsic::amdgcn_mfma_i32_16x16x16i8:
+    case Intrinsic::amdgcn_mfma_f32_16x16x2bf16:
+    case Intrinsic::amdgcn_mfma_f32_16x16x8bf16:
+    case Intrinsic::amdgcn_mfma_f32_32x32x1f32:
+    case Intrinsic::amdgcn_mfma_f32_32x32x2f32:
+    case Intrinsic::amdgcn_mfma_f32_32x32x4f16:
+    case Intrinsic::amdgcn_mfma_f32_32x32x8f16:
+    case Intrinsic::amdgcn_mfma_i32_32x32x4i8:
+    case Intrinsic::amdgcn_mfma_i32_32x32x8i8:
+    case Intrinsic::amdgcn_mfma_f32_32x32x2bf16:
+    case Intrinsic::amdgcn_mfma_f32_32x32x4bf16: {
+      // Default for MAI intrinsics.
+      // srcC can also be an immediate which can be folded later.
+      // FIXME: Should we eventually add an alternative mapping with AGPR src
+      // for srcA/srcB?
+      //
+      // vdst, srcA, srcB, srcC
+      OpdsMapping[0] = getAGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);
+      OpdsMapping[2] = getVGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);
+      OpdsMapping[3] = getVGPROpMapping(MI.getOperand(3).getReg(), MRI, *TRI);
+      OpdsMapping[4] = getAGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI);
+      break;
+    }
     }
     break;
   }
   case AMDGPU::G_INTRINSIC_W_SIDE_EFFECTS: {
     auto IntrID = MI.getIntrinsicID();
     switch (IntrID) {
     case Intrinsic::amdgcn_s_getreg:
     case Intrinsic::amdgcn_s_memtime:

llvm/lib/Target/AMDGPU/AMDGPURegisterBanks.td

 def VGPRRegBank : RegisterBank<"VGPR",
   [VGPR_32, VReg_64, VReg_96, VReg_128, VReg_256, VReg_512, VReg_1024]
 >;
 
 def SCCRegBank : RegisterBank <"SCC", [SReg_32, SCC_CLASS]>;
 
 // It is helpful to distinguish conditions from ordinary SGPRs.
 def VCCRegBank : RegisterBank <"VCC", [SReg_1]>;
+
+def AGPRRegBank : RegisterBank <"AGPR",
+  [AGPR_32, AReg_64, AReg_128, AReg_512, AReg_1024]
+>;

llvm/lib/Target/AMDGPU/SIRegisterInfo.h

bool isSGPRReg(const MachineRegisterInfo &MRI, unsigned Reg) const {
     const TargetRegisterClass *RC;
     if (Register::isVirtualRegister(Reg))
       RC = MRI.getRegClass(Reg);
     else
       RC = getPhysRegClass(Reg);
     return isSGPRClass(RC);
   }
 
+  /// \returns true if this class contains only AGPR registers
+  bool isAGPRClass(const TargetRegisterClass *RC) const {
+    return hasAGPRs(RC) && !hasVGPRs(RC);
+  }
+
   /// \returns true if this class contains VGPR registers.
   bool hasVGPRs(const TargetRegisterClass *RC) const;
 
   /// \returns true if this class contains AGPR registers.
   bool hasAGPRs(const TargetRegisterClass *RC) const;
 
   /// \returns true if this class contains any vector registers.
   bool hasVectorRegisters(const TargetRegisterClass *RC) const {

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.mfma.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=gfx908 -run-pass=regbankselect -regbankselect-fast -verify-machineinstrs %s -o - | FileCheck %s
+# RUN: llc -march=amdgcn -mcpu=gfx908 -run-pass=regbankselect -regbankselect-greedy -verify-machineinstrs %s -o - | FileCheck %s
+
+---
+name: mfma_f32_32x32x1f32_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x1f32_vva
+    ; CHECK: liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x1f32), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<32 x s32>), 0, 0, 0
+    ; CHECK: $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 = COPY [[INT]](<32 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    %3:_(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x1f32), %0, %1, %2, 0, 0, 0
+    $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 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x1f32_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x1f32_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<32 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<32 x s32>) = COPY [[COPY2]](<32 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x1f32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<32 x s32>), 0, 0, 0
+    ; CHECK: $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 = COPY [[INT]](<32 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<32 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    %3:_(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x1f32), %0, %1, %2, 0, 0, 0
+    $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 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x1f32_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x1f32_vva
+    ; CHECK: liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x1f32), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x1f32), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x1f32_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x1f32_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<16 x s32>) = COPY [[COPY2]](<16 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x1f32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x1f32), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_4x4x1f32_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3
+
+    ; CHECK-LABEL: name: mfma_f32_4x4x1f32_vva
+    ; CHECK: liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x1f32), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x1f32), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_4x4x1f32_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+
+    ; CHECK-LABEL: name: mfma_f32_4x4x1f32_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<4 x s32>) = COPY [[COPY2]](<4 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x1f32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x1f32), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x2f32_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x2f32_vva
+    ; CHECK: liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x2f32), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x2f32), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x2f32_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x2f32_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<16 x s32>) = COPY [[COPY2]](<16 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x2f32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x2f32), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x4f32_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x4f32_vva
+    ; CHECK: liveins: $vgpr0, $vgpr1, $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x4f32), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x4f32), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x4f32_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x4f32_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<4 x s32>) = COPY [[COPY2]](<4 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x4f32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x4f32), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x4f16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x4f16_vva
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x4f16), [[COPY]](<4 x s16>), [[COPY1]](<4 x s16>), [[COPY2]](<32 x s32>), 0, 0, 0
+    ; CHECK: $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 = COPY [[INT]](<32 x s32>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    %3:_(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x4f16), %0, %1, %2, 0, 0, 0
+    $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 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x4f16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x4f16_sss
+    ; CHECK: liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr32_sgpr33
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr34_sgpr35
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<32 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY]](<4 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY1]](<4 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<32 x s32>) = COPY [[COPY2]](<32 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x4f16), [[COPY3]](<4 x s16>), [[COPY4]](<4 x s16>), [[COPY5]](<32 x s32>), 0, 0, 0
+    ; CHECK: $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 = COPY [[INT]](<32 x s32>)
+    %0:_(<4 x s16>) = COPY $sgpr32_sgpr33
+    %1:_(<4 x s16>) = COPY $sgpr34_sgpr35
+    %2:_(<32 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    %3:_(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x4f16), %0, %1, %2, 0, 0, 0
+    $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 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x4f16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x4f16_vva
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x4f16), [[COPY]](<4 x s16>), [[COPY1]](<4 x s16>), [[COPY2]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x4f16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x4f16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x4f16_sss
+    ; CHECK: liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr32_sgpr33
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr34_sgpr35
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY]](<4 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY1]](<4 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<16 x s32>) = COPY [[COPY2]](<16 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x4f16), [[COPY3]](<4 x s16>), [[COPY4]](<4 x s16>), [[COPY5]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(<4 x s16>) = COPY $sgpr32_sgpr33
+    %1:_(<4 x s16>) = COPY $sgpr34_sgpr35
+    %2:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x4f16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_4x4x4f16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3
+
+    ; CHECK-LABEL: name: mfma_f32_4x4x4f16_vva
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x4f16), [[COPY]](<4 x s16>), [[COPY1]](<4 x s16>), [[COPY2]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x4f16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_4x4x4f16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3
+
+    ; CHECK-LABEL: name: mfma_f32_4x4x4f16_sss
+    ; CHECK: liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr32_sgpr33
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr34_sgpr35
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY]](<4 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY1]](<4 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<4 x s32>) = COPY [[COPY2]](<4 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x4f16), [[COPY3]](<4 x s16>), [[COPY4]](<4 x s16>), [[COPY5]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(<4 x s16>) = COPY $sgpr32_sgpr33
+    %1:_(<4 x s16>) = COPY $sgpr34_sgpr35
+    %2:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x4f16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x8f16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x8f16_vva
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x8f16), [[COPY]](<4 x s16>), [[COPY1]](<4 x s16>), [[COPY2]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x8f16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x8f16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x8f16_sss
+    ; CHECK: liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr32_sgpr33
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr34_sgpr35
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY]](<4 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY1]](<4 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<16 x s32>) = COPY [[COPY2]](<16 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x8f16), [[COPY3]](<4 x s16>), [[COPY4]](<4 x s16>), [[COPY5]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(<4 x s16>) = COPY $sgpr32_sgpr33
+    %1:_(<4 x s16>) = COPY $sgpr34_sgpr35
+    %2:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x8f16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x16f16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x16f16_vva
+    ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<4 x s16>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x16f16), [[COPY]](<4 x s16>), [[COPY1]](<4 x s16>), [[COPY2]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
+    %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x16f16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x16f16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x16f16_sss
+    ; CHECK: liveins: $sgpr32_sgpr33, $sgpr34_sgpr35, $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr32_sgpr33
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<4 x s16>) = COPY $sgpr34_sgpr35
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY]](<4 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<4 x s16>) = COPY [[COPY1]](<4 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<4 x s32>) = COPY [[COPY2]](<4 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x16f16), [[COPY3]](<4 x s16>), [[COPY4]](<4 x s16>), [[COPY5]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(<4 x s16>) = COPY $sgpr32_sgpr33
+    %1:_(<4 x s16>) = COPY $sgpr34_sgpr35
+    %2:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x16f16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_i32_32x32x4i8_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+
+    ; CHECK-LABEL: name: mfma_i32_32x32x4i8_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.32x32x4i8), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<32 x s32>), 0, 0, 0
+    ; CHECK: $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 = COPY [[INT]](<32 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr2
+    %2:_(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    %3:_(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.32x32x4i8), %0, %1, %2, 0, 0, 0
+    $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 = COPY %3
+...
+
+---
+name: mfma_i32_32x32x4i8_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+
+    ; CHECK-LABEL: name: mfma_i32_32x32x4i8_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<32 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<32 x s32>) = COPY [[COPY2]](<32 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.32x32x4i8), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<32 x s32>), 0, 0, 0
+    ; CHECK: $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 = COPY [[INT]](<32 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<32 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    %3:_(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.32x32x4i8), %0, %1, %2, 0, 0, 0
+    $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 = COPY %3
+...
+
+---
+name: mfma_i32_16x16x4i8_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+
+    ; CHECK-LABEL: name: mfma_i32_16x16x4i8_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.16x16x4i8), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr2
+    %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.16x16x4i8), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_i32_16x16x4i8_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+
+    ; CHECK-LABEL: name: mfma_i32_16x16x4i8_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<16 x s32>) = COPY [[COPY2]](<16 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.16x16x4i8), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.16x16x4i8), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_i32_4x4x4i8_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3
+
+    ; CHECK-LABEL: name: mfma_i32_4x4x4i8_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.4x4x4i8), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr2
+    %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.4x4x4i8), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_i32_4x4x4i8_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+
+    ; CHECK-LABEL: name: mfma_i32_4x4x4i8_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<4 x s32>) = COPY [[COPY2]](<4 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.4x4x4i8), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.4x4x4i8), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_i32_32x32x8i8_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+
+    ; CHECK-LABEL: name: mfma_i32_32x32x8i8_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.32x32x8i8), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr2
+    %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.32x32x8i8), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_i32_32x32x8i8_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+
+    ; CHECK-LABEL: name: mfma_i32_32x32x8i8_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<16 x s32>) = COPY [[COPY2]](<16 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.32x32x8i8), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.32x32x8i8), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_i32_16x16x16i8_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3
+
+    ; CHECK-LABEL: name: mfma_i32_16x16x16i8_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.16x16x16i8), [[COPY]](s32), [[COPY1]](s32), [[COPY2]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr2
+    %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.16x16x16i8), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_i32_16x16x16i8_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+
+    ; CHECK-LABEL: name: mfma_i32_16x16x16i8_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<4 x s32>) = COPY [[COPY2]](<4 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.16x16x16i8), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(s32) = COPY $sgpr32
+    %1:_(s32) = COPY $sgpr33
+    %2:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.i32.16x16x16i8), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x2bf16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x2bf16_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x2bf16), [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[COPY2]](<32 x s32>), 0, 0, 0
+    ; CHECK: $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 = COPY [[INT]](<32 x s32>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr2
+    %2:_(<32 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31
+    %3:_(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x2bf16), %0, %1, %2, 0, 0, 0
+    $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 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x2bf16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x2bf16_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<32 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY]](<2 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY1]](<2 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<32 x s32>) = COPY [[COPY2]](<32 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x2bf16), [[COPY3]](<2 x s16>), [[COPY4]](<2 x s16>), [[COPY5]](<32 x s32>), 0, 0, 0
+    ; CHECK: $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 = COPY [[INT]](<32 x s32>)
+    %0:_(<2 x s16>) = COPY $sgpr32
+    %1:_(<2 x s16>) = COPY $sgpr33
+    %2:_(<32 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23_sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31
+    %3:_(<32 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x2bf16), %0, %1, %2, 0, 0, 0
+    $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 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x2bf16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x2bf16_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x2bf16), [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[COPY2]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr2
+    %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x2bf16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x2bf16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x2bf16_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY]](<2 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY1]](<2 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<16 x s32>) = COPY [[COPY2]](<16 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x2bf16), [[COPY3]](<2 x s16>), [[COPY4]](<2 x s16>), [[COPY5]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(<2 x s16>) = COPY $sgpr32
+    %1:_(<2 x s16>) = COPY $sgpr33
+    %2:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x2bf16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_4x4x2bf16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3
+
+    ; CHECK-LABEL: name: mfma_f32_4x4x2bf16_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x2bf16), [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[COPY2]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr2
+    %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x2bf16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_4x4x2bf16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+
+    ; CHECK-LABEL: name: mfma_f32_4x4x2bf16_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY]](<2 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY1]](<2 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<4 x s32>) = COPY [[COPY2]](<4 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x2bf16), [[COPY3]](<2 x s16>), [[COPY4]](<2 x s16>), [[COPY5]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(<2 x s16>) = COPY $sgpr32
+    %1:_(<2 x s16>) = COPY $sgpr33
+    %2:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.4x4x2bf16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x4bf16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x4bf16_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x4bf16), [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[COPY2]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr2
+    %2:_(<16 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x4bf16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_32x32x4bf16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+
+    ; CHECK-LABEL: name: mfma_f32_32x32x4bf16_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY]](<2 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY1]](<2 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<16 x s32>) = COPY [[COPY2]](<16 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x4bf16), [[COPY3]](<2 x s16>), [[COPY4]](<2 x s16>), [[COPY5]](<16 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[INT]](<16 x s32>)
+    %0:_(<2 x s16>) = COPY $sgpr32
+    %1:_(<2 x s16>) = COPY $sgpr33
+    %2:_(<16 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15
+    %3:_(<16 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.32x32x4bf16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x8bf16_vva
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x8bf16_vva
+    ; CHECK: liveins: $vgpr0, $vgpr2, $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr2
+    ; CHECK: [[COPY2:%[0-9]+]]:agpr(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x8bf16), [[COPY]](<2 x s16>), [[COPY1]](<2 x s16>), [[COPY2]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = COPY $vgpr2
+    %2:_(<4 x s32>) = COPY $agpr0_agpr1_agpr2_agpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x8bf16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...
+
+---
+name: mfma_f32_16x16x8bf16_sss
+legalized: true
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+
+    ; CHECK-LABEL: name: mfma_f32_16x16x8bf16_sss
+    ; CHECK: liveins: $sgpr32, $sgpr33, $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr32
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr33
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY]](<2 x s16>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(<2 x s16>) = COPY [[COPY1]](<2 x s16>)
+    ; CHECK: [[COPY5:%[0-9]+]]:agpr(<4 x s32>) = COPY [[COPY2]](<4 x s32>)
+    ; CHECK: [[INT:%[0-9]+]]:agpr(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x8bf16), [[COPY3]](<2 x s16>), [[COPY4]](<2 x s16>), [[COPY5]](<4 x s32>), 0, 0, 0
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[INT]](<4 x s32>)
+    %0:_(<2 x s16>) = COPY $sgpr32
+    %1:_(<2 x s16>) = COPY $sgpr33
+    %2:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.amdgcn.mfma.f32.16x16x8bf16), %0, %1, %2, 0, 0, 0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...