[AMDGPU] gfx908 register file changes

lib/Target/AMDGPU/AMDGPURegisterInfo.td

 //===-- AMDGPURegisterInfo.td - AMDGPU register info -------*- tablegen -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // Tablegen register definitions common to all hw codegen targets.
 //
 //===----------------------------------------------------------------------===//
 
 let Namespace = "AMDGPU" in {
 
-foreach Index = 0-15 in {
+foreach Index = 0-31 in {
   def sub#Index : SubRegIndex<32, !shl(Index, 5)>;
 }
 
 }
 
 include "SIRegisterInfo.td"

lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

 using namespace llvm;
 using namespace llvm::AMDGPU;
 using namespace llvm::amdhsa;
 
 namespace {
 
 class AMDGPUAsmParser;
 
-enum RegisterKind { IS_UNKNOWN, IS_VGPR, IS_SGPR, IS_TTMP, IS_SPECIAL };
+enum RegisterKind { IS_UNKNOWN, IS_VGPR, IS_SGPR, IS_AGPR, IS_TTMP, IS_SPECIAL };
 
 //===----------------------------------------------------------------------===//
 // Operand
 //===----------------------------------------------------------------------===//
 
 class AMDGPUOperand : public MCParsedAsmOperand {
   enum KindTy {
     Token,
enum ImmTy {
     ImmTyInterpAttr,
     ImmTyAttrChan,
     ImmTyOpSel,
     ImmTyOpSelHi,
     ImmTyNegLo,
     ImmTyNegHi,
     ImmTySwizzle,
     ImmTyGprIdxMode,
+    ImmTyHigh,
+    ImmTyBLGP,
+    ImmTyCBSZ,
+    ImmTyABID,
     ImmTyEndpgm,
-    ImmTyHigh
   };
 
 private:
   struct TokOp {
     const char *Data;
     unsigned Length;
   };
 
public:
   bool isVSrcF16() const {
     return isVCSrcF16() || isLiteralImm(MVT::f16);
   }
 
   bool isVSrcV2F16() const {
     return isVSrcF16() || isLiteralImm(MVT::v2f16);
   }
 
+  bool isVISrcB32() const {
+    return isRegOrInlineNoMods(AMDGPU::VGPR_32RegClassID, MVT::i32);
+  }
+
+  bool isVISrcB16() const {
+    return isRegOrInlineNoMods(AMDGPU::VGPR_32RegClassID, MVT::i16);
+  }
+
+  bool isVISrcV2B16() const {
+    return isVISrcB16();
+  }
+
+  bool isVISrcF32() const {
+    return isRegOrInlineNoMods(AMDGPU::VGPR_32RegClassID, MVT::f32);
+  }
+
+  bool isVISrcF16() const {
+    return isRegOrInlineNoMods(AMDGPU::VGPR_32RegClassID, MVT::f16);
+  }
+
+  bool isVISrcV2F16() const {
+    return isVISrcF16() || isVISrcB32();
+  }
+
+  bool isAISrcB32() const {
+    return isRegOrInlineNoMods(AMDGPU::AGPR_32RegClassID, MVT::i32);
+  }
+
+  bool isAISrcB16() const {
+    return isRegOrInlineNoMods(AMDGPU::AGPR_32RegClassID, MVT::i16);
+  }
+
+  bool isAISrcV2B16() const {
+    return isAISrcB16();
+  }
+
+  bool isAISrcF32() const {
+    return isRegOrInlineNoMods(AMDGPU::AGPR_32RegClassID, MVT::f32);
+  }
+
+  bool isAISrcF16() const {
+    return isRegOrInlineNoMods(AMDGPU::AGPR_32RegClassID, MVT::f16);
+  }
+
+  bool isAISrcV2F16() const {
+    return isAISrcF16() || isAISrcB32();
+  }
+
+  bool isAISrc_128B32() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_128RegClassID, MVT::i32);
+  }
+
+  bool isAISrc_128B16() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_128RegClassID, MVT::i16);
+  }
+
+  bool isAISrc_128V2B16() const {
+    return isAISrc_128B16();
+  }
+
+  bool isAISrc_128F32() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_128RegClassID, MVT::f32);
+  }
+
+  bool isAISrc_128F16() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_128RegClassID, MVT::f16);
+  }
+
+  bool isAISrc_128V2F16() const {
+    return isAISrc_128F16() || isAISrc_128B32();
+  }
+
+  bool isAISrc_512B32() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_512RegClassID, MVT::i32);
+  }
+
+  bool isAISrc_512B16() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_512RegClassID, MVT::i16);
+  }
+
+  bool isAISrc_512V2B16() const {
+    return isAISrc_512B16();
+  }
+
+  bool isAISrc_512F32() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_512RegClassID, MVT::f32);
+  }
+
+  bool isAISrc_512F16() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_512RegClassID, MVT::f16);
+  }
+
+  bool isAISrc_512V2F16() const {
+    return isAISrc_512F16() || isAISrc_512B32();
+  }
+
+  bool isAISrc_1024B32() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_1024RegClassID, MVT::i32);
+  }
+
+  bool isAISrc_1024B16() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_1024RegClassID, MVT::i16);
+  }
+
+  bool isAISrc_1024V2B16() const {
+    return isAISrc_1024B16();
+  }
+
+  bool isAISrc_1024F32() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_1024RegClassID, MVT::f32);
+  }
+
+  bool isAISrc_1024F16() const {
+    return isRegOrInlineNoMods(AMDGPU::AReg_1024RegClassID, MVT::f16);
+  }
+
+  bool isAISrc_1024V2F16() const {
+    return isAISrc_1024F16() || isAISrc_1024B32();
+  }
+
   bool isKImmFP32() const {
     return isLiteralImm(MVT::f32);
   }
 
   bool isKImmFP16() const {
     return isLiteralImm(MVT::f16);
   }
 
public:
   bool isHwreg() const;
   bool isSendMsg() const;
   bool isSwizzle() const;
   bool isSMRDOffset8() const;
   bool isSMRDOffset20() const;
   bool isSMRDLiteralOffset() const;
   bool isDPP8() const;
   bool isDPPCtrl() const;
+  bool isBLGP() const;
+  bool isCBSZ() const;
+  bool isABID() const;
   bool isGPRIdxMode() const;
   bool isS16Imm() const;
   bool isU16Imm() const;
   bool isEndpgm() const;
 
   StringRef getExpressionAsToken() const {
     assert(isExpr());
     const MCSymbolRefExpr *S = cast<MCSymbolRefExpr>(Expr);
static void printImmTy(raw_ostream& OS, ImmTy Type) {
     case ImmTyAttrChan: OS << "AttrChan"; break;
     case ImmTyOpSel: OS << "OpSel"; break;
     case ImmTyOpSelHi: OS << "OpSelHi"; break;
     case ImmTyNegLo: OS << "NegLo"; break;
     case ImmTyNegHi: OS << "NegHi"; break;
     case ImmTySwizzle: OS << "Swizzle"; break;
     case ImmTyGprIdxMode: OS << "GprIdxMode"; break;
     case ImmTyHigh: OS << "High"; break;
-    case ImmTyEndpgm:
-      OS << "Endpgm";
-      break;
+    case ImmTyBLGP: OS << "BLGP"; break;
+    case ImmTyCBSZ: OS << "CBSZ"; break;
+    case ImmTyABID: OS << "ABID"; break;
+    case ImmTyEndpgm: OS << "Endpgm"; break;
     }
   }
 
   void print(raw_ostream &OS) const override {
     switch (Kind) {
     case Register:
       OS << "<register " << getReg() << " mods: " << Reg.Mods << '>';
       break;
void initialize(MCContext &Context) {
     Ctx = &Context;
     usesSgprAt(SgprIndexUnusedMin = -1);
     usesVgprAt(VgprIndexUnusedMin = -1);
   }
 
   void usesRegister(RegisterKind RegKind, unsigned DwordRegIndex, unsigned RegWidth) {
     switch (RegKind) {
       case IS_SGPR: usesSgprAt(DwordRegIndex + RegWidth - 1); break;
+      case IS_AGPR: // fall through
       case IS_VGPR: usesVgprAt(DwordRegIndex + RegWidth - 1); break;
       default: break;
     }
   }
 };
 
 class AMDGPUAsmParser : public MCTargetAsmParser {
   MCAsmParser &Parser;
public:
   OperandMatchResultTy parseSDWADstUnused(OperandVector &Operands);
   void cvtSdwaVOP1(MCInst &Inst, const OperandVector &Operands);
   void cvtSdwaVOP2(MCInst &Inst, const OperandVector &Operands);
   void cvtSdwaVOP2b(MCInst &Inst, const OperandVector &Operands);
   void cvtSdwaVOPC(MCInst &Inst, const OperandVector &Operands);
   void cvtSDWA(MCInst &Inst, const OperandVector &Operands,
                 uint64_t BasicInstType, bool skipVcc = false);
 
+  AMDGPUOperand::Ptr defaultBLGP() const;
+  AMDGPUOperand::Ptr defaultCBSZ() const;
+  AMDGPUOperand::Ptr defaultABID() const;
+
   OperandMatchResultTy parseEndpgmOp(OperandVector &Operands);
   AMDGPUOperand::Ptr defaultEndpgmImmOperands() const;
 };
 
 struct OptionalOperand {
   const char *Name;
   AMDGPUOperand::ImmTy Type;
   bool IsBit;

 }
 
 static const fltSemantics *getOpFltSemantics(uint8_t OperandType) {
   switch (OperandType) {
   case AMDGPU::OPERAND_REG_IMM_INT32:
   case AMDGPU::OPERAND_REG_IMM_FP32:
   case AMDGPU::OPERAND_REG_INLINE_C_INT32:
   case AMDGPU::OPERAND_REG_INLINE_C_FP32:
+  case AMDGPU::OPERAND_REG_INLINE_AC_INT32:
+  case AMDGPU::OPERAND_REG_INLINE_AC_FP32:
     return &APFloat::IEEEsingle();
   case AMDGPU::OPERAND_REG_IMM_INT64:
   case AMDGPU::OPERAND_REG_IMM_FP64:
   case AMDGPU::OPERAND_REG_INLINE_C_INT64:
   case AMDGPU::OPERAND_REG_INLINE_C_FP64:
     return &APFloat::IEEEdouble();
   case AMDGPU::OPERAND_REG_IMM_INT16:
   case AMDGPU::OPERAND_REG_IMM_FP16:
   case AMDGPU::OPERAND_REG_INLINE_C_INT16:
   case AMDGPU::OPERAND_REG_INLINE_C_FP16:
   case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
   case AMDGPU::OPERAND_REG_INLINE_C_V2FP16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_INT16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_FP16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_V2INT16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_V2FP16:
   case AMDGPU::OPERAND_REG_IMM_V2INT16:
   case AMDGPU::OPERAND_REG_IMM_V2FP16:
     return &APFloat::IEEEhalf();
   default:
     llvm_unreachable("unsupported fp type");
   }
 }
 
case AMDGPU::OPERAND_REG_INLINE_C_FP64:
       // unclear how we should encode them. This case should be checked earlier
       // in predicate methods (isLiteralImm())
       llvm_unreachable("fp literal in 64-bit integer instruction.");
 
     case AMDGPU::OPERAND_REG_IMM_INT32:
     case AMDGPU::OPERAND_REG_IMM_FP32:
     case AMDGPU::OPERAND_REG_INLINE_C_INT32:
     case AMDGPU::OPERAND_REG_INLINE_C_FP32:
+    case AMDGPU::OPERAND_REG_INLINE_AC_INT32:
+    case AMDGPU::OPERAND_REG_INLINE_AC_FP32:
     case AMDGPU::OPERAND_REG_IMM_INT16:
     case AMDGPU::OPERAND_REG_IMM_FP16:
     case AMDGPU::OPERAND_REG_INLINE_C_INT16:
     case AMDGPU::OPERAND_REG_INLINE_C_FP16:
     case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
     case AMDGPU::OPERAND_REG_INLINE_C_V2FP16:
+    case AMDGPU::OPERAND_REG_INLINE_AC_INT16:
+    case AMDGPU::OPERAND_REG_INLINE_AC_FP16:
+    case AMDGPU::OPERAND_REG_INLINE_AC_V2INT16:
+    case AMDGPU::OPERAND_REG_INLINE_AC_V2FP16:
     case AMDGPU::OPERAND_REG_IMM_V2INT16:
     case AMDGPU::OPERAND_REG_IMM_V2FP16: {
       bool lost;
       APFloat FPLiteral(APFloat::IEEEdouble(), Literal);
       // Convert literal to single precision
       FPLiteral.convert(*getOpFltSemantics(OpTy),
                         APFloat::rmNearestTiesToEven, &lost);
       // We allow precision lost but not overflow or underflow. This should be
void AMDGPUOperand::addLiteralImmOperand(MCInst &Inst, int64_t Val, bool ApplyModifiers) const {
 
   // We got int literal token.
   // Only sign extend inline immediates.
   switch (OpTy) {
   case AMDGPU::OPERAND_REG_IMM_INT32:
   case AMDGPU::OPERAND_REG_IMM_FP32:
   case AMDGPU::OPERAND_REG_INLINE_C_INT32:
   case AMDGPU::OPERAND_REG_INLINE_C_FP32:
+  case AMDGPU::OPERAND_REG_INLINE_AC_INT32:
+  case AMDGPU::OPERAND_REG_INLINE_AC_FP32:
   case AMDGPU::OPERAND_REG_IMM_V2INT16:
   case AMDGPU::OPERAND_REG_IMM_V2FP16:
     if (isSafeTruncation(Val, 32) &&
         AMDGPU::isInlinableLiteral32(static_cast<int32_t>(Val),
                                      AsmParser->hasInv2PiInlineImm())) {
       Inst.addOperand(MCOperand::createImm(Val));
       return;
     }
case AMDGPU::OPERAND_REG_INLINE_C_FP64:
 
     Inst.addOperand(MCOperand::createImm(Lo_32(Val)));
     return;
 
   case AMDGPU::OPERAND_REG_IMM_INT16:
   case AMDGPU::OPERAND_REG_IMM_FP16:
   case AMDGPU::OPERAND_REG_INLINE_C_INT16:
   case AMDGPU::OPERAND_REG_INLINE_C_FP16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_INT16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_FP16:
     if (isSafeTruncation(Val, 16) &&
         AMDGPU::isInlinableLiteral16(static_cast<int16_t>(Val),
                                      AsmParser->hasInv2PiInlineImm())) {
       Inst.addOperand(MCOperand::createImm(Val));
       return;
     }
 
     Inst.addOperand(MCOperand::createImm(Val & 0xffff));
     return;
 
   case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
-  case AMDGPU::OPERAND_REG_INLINE_C_V2FP16: {
+  case AMDGPU::OPERAND_REG_INLINE_C_V2FP16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_V2INT16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_V2FP16: {
     assert(isSafeTruncation(Val, 16));
     assert(AMDGPU::isInlinableLiteral16(static_cast<int16_t>(Val),
                                         AsmParser->hasInv2PiInlineImm()));
 
     Inst.addOperand(MCOperand::createImm(Val));
     return;
   }
   default:
if (Is == IS_VGPR) {
     switch (RegWidth) {
       default: return -1;
       case 1: return AMDGPU::SGPR_32RegClassID;
       case 2: return AMDGPU::SGPR_64RegClassID;
       case 4: return AMDGPU::SGPR_128RegClassID;
       case 8: return AMDGPU::SGPR_256RegClassID;
       case 16: return AMDGPU::SGPR_512RegClassID;
     }
+  } else if (Is == IS_AGPR) {
+    switch (RegWidth) {
+      default: return -1;
+      case 1: return AMDGPU::AGPR_32RegClassID;
+      case 2: return AMDGPU::AReg_64RegClassID;
+      case 4: return AMDGPU::AReg_128RegClassID;
+      case 16: return AMDGPU::AReg_512RegClassID;
+      case 32: return AMDGPU::AReg_1024RegClassID;
+    }
   }
   return -1;
 }
 
 static unsigned getSpecialRegForName(StringRef RegName) {
   return StringSwitch<unsigned>(RegName)
     .Case("exec", AMDGPU::EXEC)
     .Case("vcc", AMDGPU::VCC)
case IS_SPECIAL:
     if (Reg == AMDGPU::TMA_LO && Reg1 == AMDGPU::TMA_HI) {
       Reg = AMDGPU::TMA;
       RegWidth = 2;
       return true;
     }
     return false;
   case IS_VGPR:
   case IS_SGPR:
+  case IS_AGPR:
   case IS_TTMP:
     if (Reg1 != Reg + RegWidth) {
       return false;
     }
     RegWidth++;
     return true;
   default:
     llvm_unreachable("unexpected register kind");
   }
 }
 
 static const StringRef Registers[] = {
   { "v" },
   { "s" },
   { "ttmp" },
+  { "acc" },
+  { "a" },
 };
 
 bool
 AMDGPUAsmParser::isRegister(const AsmToken &Token,
                             const AsmToken &NextToken) const {
 
   // A list of consecutive registers: [s0,s1,s2,s3]
   if (Token.is(AsmToken::LBrac))
if (getLexer().is(AsmToken::Identifier)) {
     } else {
       unsigned RegNumIndex = 0;
       if (RegName[0] == 'v') {
         RegNumIndex = 1;
         RegKind = IS_VGPR;
       } else if (RegName[0] == 's') {
         RegNumIndex = 1;
         RegKind = IS_SGPR;
+      } else if (RegName[0] == 'a') {
+        RegNumIndex = RegName.startswith("acc") ? 3 : 1;
+        RegKind = IS_AGPR;
       } else if (RegName.startswith("ttmp")) {
         RegNumIndex = strlen("ttmp");
         RegKind = IS_TTMP;
       } else {
         return false;
       }
       if (RegName.size() > RegNumIndex) {
         // Single 32-bit register: vXX.
bool AMDGPUAsmParser::ParseAMDGPURegister(RegisterKind &RegKind, unsigned &Reg,
   }
   switch (RegKind) {
   case IS_SPECIAL:
     RegNum = 0;
     RegWidth = 1;
     break;
   case IS_VGPR:
   case IS_SGPR:
+  case IS_AGPR:
   case IS_TTMP:
   {
     unsigned Size = 1;
     if (RegKind == IS_SGPR || RegKind == IS_TTMP) {
       // SGPR and TTMP registers must be aligned. Max required alignment is 4 dwords.
       Size = std::min(RegWidth, 4u);
     }
     if (RegNum % Size != 0)
bool AMDGPUAsmParser::isInlineConstant(const MCInst &Inst,
   case 8:
     return AMDGPU::isInlinableLiteral64(Val, hasInv2PiInlineImm());
   case 4:
     return AMDGPU::isInlinableLiteral32(Val, hasInv2PiInlineImm());
   case 2: {
     const unsigned OperandType = Desc.OpInfo[OpIdx].OperandType;
     if (OperandType == AMDGPU::OPERAND_REG_INLINE_C_V2INT16 ||
         OperandType == AMDGPU::OPERAND_REG_INLINE_C_V2FP16 ||
+        OperandType == AMDGPU::OPERAND_REG_INLINE_AC_V2INT16 ||
+        OperandType == AMDGPU::OPERAND_REG_INLINE_AC_V2FP16 ||
         OperandType == AMDGPU::OPERAND_REG_IMM_V2INT16 ||
         OperandType == AMDGPU::OPERAND_REG_IMM_V2FP16) {
       return AMDGPU::isInlinableLiteralV216(Val, hasInv2PiInlineImm());
     } else {
       return AMDGPU::isInlinableLiteral16(Val, hasInv2PiInlineImm());
     }
   }
   default:
AMDGPUAsmParser::parseNamedBit(const char *Name, OperandVector &Operands,
   // use a default value.
   if (getLexer().isNot(AsmToken::EndOfStatement)) {
     switch(getLexer().getKind()) {
       case AsmToken::Identifier: {
         StringRef Tok = Parser.getTok().getString();
         if (Tok == Name) {
           if (Tok == "r128" && isGFX9())
             Error(S, "r128 modifier is not supported on this GPU");
-          if (Tok == "a16" && !isGFX9())
+          if (Tok == "a16" && !isGFX9() && !isGFX10())
             Error(S, "a16 modifier is not supported on this GPU");
           Bit = 1;
           Parser.Lex();
         } else if (Tok.startswith("no") && Tok.endswith(Name)) {
           Bit = 0;
           Parser.Lex();
         } else {
           return MatchOperand_NoMatch;
static const OptionalOperand AMDGPUOptionalOperandTable[] = {
   {"src0_sel",   AMDGPUOperand::ImmTySdwaSrc0Sel, false, nullptr},
   {"src1_sel",   AMDGPUOperand::ImmTySdwaSrc1Sel, false, nullptr},
   {"dst_unused", AMDGPUOperand::ImmTySdwaDstUnused, false, nullptr},
   {"compr", AMDGPUOperand::ImmTyExpCompr, true, nullptr },
   {"vm", AMDGPUOperand::ImmTyExpVM, true, nullptr},
   {"op_sel", AMDGPUOperand::ImmTyOpSel, false, nullptr},
   {"op_sel_hi", AMDGPUOperand::ImmTyOpSelHi, false, nullptr},
   {"neg_lo", AMDGPUOperand::ImmTyNegLo, false, nullptr},
-  {"neg_hi", AMDGPUOperand::ImmTyNegHi, false, nullptr}
+  {"neg_hi", AMDGPUOperand::ImmTyNegHi, false, nullptr},
+  {"blgp", AMDGPUOperand::ImmTyBLGP, false, nullptr},
+  {"cbsz", AMDGPUOperand::ImmTyCBSZ, false, nullptr},
+  {"abid", AMDGPUOperand::ImmTyABID, false, nullptr}
 };
 
 OperandMatchResultTy AMDGPUAsmParser::parseOptionalOperand(OperandVector &Operands) {
   unsigned size = Operands.size();
   assert(size > 0);
 
   OperandMatchResultTy res = parseOptionalOpr(Operands);
 
return (Imm >= DppCtrl::QUAD_PERM_FIRST && Imm <= DppCtrl::QUAD_PERM_LAST) ||
            (Imm == DppCtrl::BCAST15) ||
            (Imm == DppCtrl::BCAST31) ||
            (Imm >= DppCtrl::ROW_SHARE_FIRST && Imm <= DppCtrl::ROW_SHARE_LAST) ||
            (Imm >= DppCtrl::ROW_XMASK_FIRST && Imm <= DppCtrl::ROW_XMASK_LAST);
   }
   return false;
 }
 
+//===----------------------------------------------------------------------===//
+// mAI
+//===----------------------------------------------------------------------===//
+
+bool AMDGPUOperand::isBLGP() const {
+  return isImm() && getImmTy() == ImmTyBLGP && isUInt<3>(getImm());
+}
+
+bool AMDGPUOperand::isCBSZ() const {
+  return isImm() && getImmTy() == ImmTyCBSZ && isUInt<3>(getImm());
+}
+
+bool AMDGPUOperand::isABID() const {
+  return isImm() && getImmTy() == ImmTyABID && isUInt<4>(getImm());
+}
+
 bool AMDGPUOperand::isS16Imm() const {
   return isImm() && (isInt<16>(getImm()) || isUInt<16>(getImm()));
 }
 
 bool AMDGPUOperand::isU16Imm() const {
   return isImm() && isUInt<16>(getImm());
 }
 
if (Inst.getOpcode() == AMDGPU::V_MAC_F32_sdwa_vi ||
       Inst.getOpcode() == AMDGPU::V_MAC_F16_sdwa_vi)  {
     auto it = Inst.begin();
     std::advance(
       it, AMDGPU::getNamedOperandIdx(Inst.getOpcode(), AMDGPU::OpName::src2));
     Inst.insert(it, Inst.getOperand(0)); // src2 = dst
   }
 }
 
+//===----------------------------------------------------------------------===//
+// mAI
+//===----------------------------------------------------------------------===//
+
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultBLGP() const {
+  return AMDGPUOperand::CreateImm(this, 0, SMLoc(), AMDGPUOperand::ImmTyBLGP);
+}
+
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultCBSZ() const {
+  return AMDGPUOperand::CreateImm(this, 0, SMLoc(), AMDGPUOperand::ImmTyCBSZ);
+}
+
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultABID() const {
+  return AMDGPUOperand::CreateImm(this, 0, SMLoc(), AMDGPUOperand::ImmTyABID);
+}
+
 /// Force static initialization.
 extern "C" void LLVMInitializeAMDGPUAsmParser() {
   RegisterMCAsmParser<AMDGPUAsmParser> A(getTheAMDGPUTarget());
   RegisterMCAsmParser<AMDGPUAsmParser> B(getTheGCNTarget());
 }
 
 #define GET_REGISTER_MATCHER
 #define GET_MATCHER_IMPLEMENTATION

lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h

public:
   MCOperand decodeOperand_VS_64(unsigned Val) const;
   MCOperand decodeOperand_VS_128(unsigned Val) const;
   MCOperand decodeOperand_VSrc16(unsigned Val) const;
   MCOperand decodeOperand_VSrcV216(unsigned Val) const;
 
   MCOperand decodeOperand_VReg_64(unsigned Val) const;
   MCOperand decodeOperand_VReg_96(unsigned Val) const;
   MCOperand decodeOperand_VReg_128(unsigned Val) const;
+  MCOperand decodeOperand_VReg_256(unsigned Val) const;
+  MCOperand decodeOperand_VReg_512(unsigned Val) const;
 
   MCOperand decodeOperand_SReg_32(unsigned Val) const;
   MCOperand decodeOperand_SReg_32_XM0_XEXEC(unsigned Val) const;
   MCOperand decodeOperand_SReg_32_XEXEC_HI(unsigned Val) const;
   MCOperand decodeOperand_SRegOrLds_32(unsigned Val) const;
   MCOperand decodeOperand_SReg_64(unsigned Val) const;
   MCOperand decodeOperand_SReg_64_XEXEC(unsigned Val) const;
   MCOperand decodeOperand_SReg_128(unsigned Val) const;
   MCOperand decodeOperand_SReg_256(unsigned Val) const;
   MCOperand decodeOperand_SReg_512(unsigned Val) const;
 
+  MCOperand decodeOperand_AGPR_32(unsigned Val) const;
+  MCOperand decodeOperand_AReg_128(unsigned Val) const;
+  MCOperand decodeOperand_AReg_512(unsigned Val) const;
+  MCOperand decodeOperand_AReg_1024(unsigned Val) const;
+  MCOperand decodeOperand_AV_32(unsigned Val) const;
+  MCOperand decodeOperand_AV_64(unsigned Val) const;
+
   enum OpWidthTy {
     OPW32,
     OPW64,
     OPW128,
     OPW256,
     OPW512,
+    OPW1024,
     OPW16,
     OPWV216,
     OPW_LAST_,
     OPW_FIRST_ = OPW32
   };
 
   unsigned getVgprClassId(const OpWidthTy Width) const;
+  unsigned getAgprClassId(const OpWidthTy Width) const;
   unsigned getSgprClassId(const OpWidthTy Width) const;
   unsigned getTtmpClassId(const OpWidthTy Width) const;
 
   static MCOperand decodeIntImmed(unsigned Imm);
   static MCOperand decodeFPImmed(OpWidthTy Width, unsigned Imm);
   MCOperand decodeLiteralConstant() const;
 
   MCOperand decodeSrcOp(const OpWidthTy Width, unsigned Val) const;

lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp

 static DecodeStatus decodeOperand_VSrcV216(MCInst &Inst,
                                          unsigned Imm,
                                          uint64_t Addr,
                                          const void *Decoder) {
   auto DAsm = static_cast<const AMDGPUDisassembler*>(Decoder);
   return addOperand(Inst, DAsm->decodeOperand_VSrcV216(Imm));
 }
 
+static DecodeStatus decodeOperand_VS_16(MCInst &Inst,
+                                        unsigned Imm,
+                                        uint64_t Addr,
+                                        const void *Decoder) {
+  auto DAsm = static_cast<const AMDGPUDisassembler*>(Decoder);
+  return addOperand(Inst, DAsm->decodeOperand_VSrc16(Imm));
+}
+
+static DecodeStatus decodeOperand_VS_32(MCInst &Inst,
+                                        unsigned Imm,
+                                        uint64_t Addr,
+                                        const void *Decoder) {
+  auto DAsm = static_cast<const AMDGPUDisassembler*>(Decoder);
+  return addOperand(Inst, DAsm->decodeOperand_VS_32(Imm));
+}
+
+static DecodeStatus decodeOperand_SReg_32(MCInst &Inst,
+                                          unsigned Imm,
+                                          uint64_t Addr,
+                                          const void *Decoder) {
+  auto DAsm = static_cast<const AMDGPUDisassembler*>(Decoder);
+  return addOperand(Inst, DAsm->decodeOperand_SReg_32(Imm));
+}
+
 #define DECODE_SDWA(DecName) \
 DECODE_OPERAND(decodeSDWA##DecName, decodeSDWA##DecName)
 
 DECODE_SDWA(Src32)
 DECODE_SDWA(Src16)
 DECODE_SDWA(VopcDst)
 
 #include "AMDGPUGenDisassemblerTables.inc"
MCOperand AMDGPUDisassembler::decodeOperand_VGPR_32(unsigned Val) const {
 
   return createRegOperand(AMDGPU::VGPR_32RegClassID, Val);
 }
 
 MCOperand AMDGPUDisassembler::decodeOperand_VRegOrLds_32(unsigned Val) const {
   return decodeSrcOp(OPW32, Val);
 }
 
+MCOperand AMDGPUDisassembler::decodeOperand_AGPR_32(unsigned Val) const {
+  return createRegOperand(AMDGPU::AGPR_32RegClassID, Val & 255);
+}
+
+MCOperand AMDGPUDisassembler::decodeOperand_AReg_128(unsigned Val) const {
+  return createRegOperand(AMDGPU::AReg_128RegClassID, Val & 255);
+}
+
+MCOperand AMDGPUDisassembler::decodeOperand_AReg_512(unsigned Val) const {
+  return createRegOperand(AMDGPU::AReg_512RegClassID, Val & 255);
+}
+
+MCOperand AMDGPUDisassembler::decodeOperand_AReg_1024(unsigned Val) const {
+  return createRegOperand(AMDGPU::AReg_1024RegClassID, Val & 255);
+}
+
+MCOperand AMDGPUDisassembler::decodeOperand_AV_32(unsigned Val) const {
+  return decodeSrcOp(OPW32, Val);
+}
+
+MCOperand AMDGPUDisassembler::decodeOperand_AV_64(unsigned Val) const {
+  return decodeSrcOp(OPW64, Val);
+}
+
 MCOperand AMDGPUDisassembler::decodeOperand_VReg_64(unsigned Val) const {
   return createRegOperand(AMDGPU::VReg_64RegClassID, Val);
 }
 
 MCOperand AMDGPUDisassembler::decodeOperand_VReg_96(unsigned Val) const {
   return createRegOperand(AMDGPU::VReg_96RegClassID, Val);
 }
 
 MCOperand AMDGPUDisassembler::decodeOperand_VReg_128(unsigned Val) const {
   return createRegOperand(AMDGPU::VReg_128RegClassID, Val);
 }
 
+MCOperand AMDGPUDisassembler::decodeOperand_VReg_256(unsigned Val) const {
+  return createRegOperand(AMDGPU::VReg_256RegClassID, Val);
+}
+
+MCOperand AMDGPUDisassembler::decodeOperand_VReg_512(unsigned Val) const {
+  return createRegOperand(AMDGPU::VReg_512RegClassID, Val);
+}
+
 MCOperand AMDGPUDisassembler::decodeOperand_SReg_32(unsigned Val) const {
   // table-gen generated disassembler doesn't care about operand types
   // leaving only registry class so SSrc_32 operand turns into SReg_32
   // and therefore we accept immediates and literals here as well
   return decodeSrcOp(OPW32, Val);
 }
 
 MCOperand AMDGPUDisassembler::decodeOperand_SReg_32_XM0_XEXEC(

 
 MCOperand AMDGPUDisassembler::decodeFPImmed(OpWidthTy Width, unsigned Imm) {
   assert(Imm >= AMDGPU::EncValues::INLINE_FLOATING_C_MIN
       && Imm <= AMDGPU::EncValues::INLINE_FLOATING_C_MAX);
 
   // ToDo: case 248: 1/(2*PI) - is allowed only on VI
   switch (Width) {
   case OPW32:
+  case OPW128: // splat constants
+  case OPW512:
+  case OPW1024:
     return MCOperand::createImm(getInlineImmVal32(Imm));
   case OPW64:
     return MCOperand::createImm(getInlineImmVal64(Imm));
   case OPW16:
   case OPWV216:
     return MCOperand::createImm(getInlineImmVal16(Imm));
   default:
     llvm_unreachable("implement me");
unsigned AMDGPUDisassembler::getVgprClassId(const OpWidthTy Width) const {
   case OPW16:
   case OPWV216:
     return VGPR_32RegClassID;
   case OPW64: return VReg_64RegClassID;
   case OPW128: return VReg_128RegClassID;
   }
 }
 
+unsigned AMDGPUDisassembler::getAgprClassId(const OpWidthTy Width) const {
+  using namespace AMDGPU;
+
+  assert(OPW_FIRST_ <= Width && Width < OPW_LAST_);
+  switch (Width) {
+  default: // fall
+  case OPW32:
+  case OPW16:
+  case OPWV216:
+    return AGPR_32RegClassID;
+  case OPW64: return AReg_64RegClassID;
+  case OPW128: return AReg_128RegClassID;
+  case OPW512: return AReg_512RegClassID;
+  case OPW1024: return AReg_1024RegClassID;
+  }
+}
+
+
 unsigned AMDGPUDisassembler::getSgprClassId(const OpWidthTy Width) const {
   using namespace AMDGPU;
 
   assert(OPW_FIRST_ <= Width && Width < OPW_LAST_);
   switch (Width) {
   default: // fall
   case OPW32:
   case OPW16:
unsigned TTmpMax =
       (isGFX9() || isGFX10()) ? TTMP_GFX9_GFX10_MAX : TTMP_VI_MAX;
 
   return (TTmpMin <= Val && Val <= TTmpMax)? Val - TTmpMin : -1;
 }
 
 MCOperand AMDGPUDisassembler::decodeSrcOp(const OpWidthTy Width, unsigned Val) const {
   using namespace AMDGPU::EncValues;
 
-  assert(Val < 512); // enum9
+  assert(Val < 1024); // enum10
+
+  bool IsAGPR = Val & 512;
+  Val &= 511;
 
   if (VGPR_MIN <= Val && Val <= VGPR_MAX) {
-    return createRegOperand(getVgprClassId(Width), Val - VGPR_MIN);
+    return createRegOperand(IsAGPR ? getAgprClassId(Width)
+                                   : getVgprClassId(Width), Val - VGPR_MIN);
   }
   if (Val <= SGPR_MAX) {
     assert(SGPR_MIN == 0); // "SGPR_MIN <= Val" is always true and causes compilation warning.
     return createSRegOperand(getSgprClassId(Width), Val - SGPR_MIN);
   }
 
   int TTmpIdx = getTTmpIdx(Val);
   if (TTmpIdx >= 0) {

lib/Target/AMDGPU/SIDefines.h

enum OperandType : unsigned {
     OPERAND_REG_INLINE_C_INT32,
     OPERAND_REG_INLINE_C_INT64,
     OPERAND_REG_INLINE_C_FP16,
     OPERAND_REG_INLINE_C_FP32,
     OPERAND_REG_INLINE_C_FP64,
     OPERAND_REG_INLINE_C_V2FP16,
     OPERAND_REG_INLINE_C_V2INT16,
 
+    /// Operands with an AccVGPR register or inline constant
+    OPERAND_REG_INLINE_AC_INT16,
+    OPERAND_REG_INLINE_AC_INT32,
+    OPERAND_REG_INLINE_AC_FP16,
+    OPERAND_REG_INLINE_AC_FP32,
+    OPERAND_REG_INLINE_AC_V2FP16,
+    OPERAND_REG_INLINE_AC_V2INT16,
+
     OPERAND_REG_IMM_FIRST = OPERAND_REG_IMM_INT32,
     OPERAND_REG_IMM_LAST = OPERAND_REG_IMM_V2INT16,
 
     OPERAND_REG_INLINE_C_FIRST = OPERAND_REG_INLINE_C_INT16,
-    OPERAND_REG_INLINE_C_LAST = OPERAND_REG_INLINE_C_V2INT16,
+    OPERAND_REG_INLINE_C_LAST = OPERAND_REG_INLINE_AC_V2INT16,
+
+    OPERAND_REG_INLINE_AC_FIRST = OPERAND_REG_INLINE_AC_INT16,
+    OPERAND_REG_INLINE_AC_LAST = OPERAND_REG_INLINE_AC_V2INT16,
 
     OPERAND_SRC_FIRST = OPERAND_REG_IMM_INT32,
     OPERAND_SRC_LAST = OPERAND_REG_INLINE_C_LAST,
 
     // Operand for source modifiers for VOP instructions
     OPERAND_INPUT_MODS,
 
     // Operand for SDWA instructions

lib/Target/AMDGPU/SIRegisterInfo.td

class getSubRegs<int size> {
   list<SubRegIndex> ret3 = [sub0, sub1, sub2];
   list<SubRegIndex> ret4 = [sub0, sub1, sub2, sub3];
   list<SubRegIndex> ret5 = [sub0, sub1, sub2, sub3, sub4];
   list<SubRegIndex> ret8 = [sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7];
   list<SubRegIndex> ret16 = [sub0, sub1, sub2, sub3,
                              sub4, sub5, sub6, sub7,
                              sub8, sub9, sub10, sub11,
                              sub12, sub13, sub14, sub15];
+  list<SubRegIndex> ret32 = [sub0, sub1, sub2, sub3,
+                             sub4, sub5, sub6, sub7,
+                             sub8, sub9, sub10, sub11,
+                             sub12, sub13, sub14, sub15,
+                             sub16, sub17, sub18, sub19,
+                             sub20, sub21, sub22, sub23,
+                             sub24, sub25, sub26, sub27,
+                             sub28, sub29, sub30, sub31];
 
   list<SubRegIndex> ret = !if(!eq(size, 2), ret2,
                               !if(!eq(size, 3), ret3,
                                   !if(!eq(size, 4), ret4,
                                       !if(!eq(size, 5), ret5,
-                                          !if(!eq(size, 8), ret8, ret16)))));
+                                          !if(!eq(size, 8), ret8,
+                                              !if(!eq(size, 16), ret16, ret32))))));
 }
 
 //===----------------------------------------------------------------------===//
 //  Declarations that describe the SI registers
 //===----------------------------------------------------------------------===//
 class SIReg <string n, bits<16> regIdx = 0> : Register<n>,
   DwarfRegNum<[!cast<int>(HWEncoding)]> {
   let Namespace = "AMDGPU";

 
 // VGPR registers
 foreach Index = 0-255 in {
   def VGPR#Index : SIReg <"VGPR"#Index, Index> {
     let HWEncoding{8} = 1;
   }
 }
 
+// AccVGPR registers
+foreach Index = 0-255 in {
+  def AGPR#Index : SIReg <"AGPR"#Index, Index> {
+    let HWEncoding{8} = 1;
+  }
+}
+
 //===----------------------------------------------------------------------===//
 //  Groupings using register classes and tuples
 //===----------------------------------------------------------------------===//
 
 def SCC_CLASS : RegisterClass<"AMDGPU", [i1], 1, (add SCC)> {
   let CopyCost = -1;
   let isAllocatable = 0;
 }
 
 def M0_CLASS : RegisterClass<"AMDGPU", [i32], 32, (add M0)> {
   let CopyCost = 1;
   let isAllocatable = 0;
 }
 
 // TODO: Do we need to set DwarfRegAlias on register tuples?
 
 // SGPR 32-bit registers
 def SGPR_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16], 32,
                             (add (sequence "SGPR%u", 0, 105))> {
   // Give all SGPR classes higher priority than VGPR classes, because
   // we want to spill SGPRs to VGPRs.
-  let AllocationPriority = 7;
+  let AllocationPriority = 9;
 }
 
 // SGPR 64-bit registers
 def SGPR_64Regs : RegisterTuples<getSubRegs<2>.ret,
                              [(add (decimate SGPR_32, 2)),
                               (add (decimate (shl SGPR_32, 1), 2))]>;
 
 // SGPR 96-bit registers. No operations use these, but for symmetry with 96-bit VGPRs.
def SGPR_512Regs : RegisterTuples<getSubRegs<16>.ret,
                                (add (decimate (shl SGPR_32, 9), 4)),
                                (add (decimate (shl SGPR_32, 10), 4)),
                                (add (decimate (shl SGPR_32, 11), 4)),
                                (add (decimate (shl SGPR_32, 12), 4)),
                                (add (decimate (shl SGPR_32, 13), 4)),
                                (add (decimate (shl SGPR_32, 14), 4)),
                                (add (decimate (shl SGPR_32, 15), 4))]>;
 
+// SGPR 1024-bit registers
+def SGPR_1024Regs : RegisterTuples<getSubRegs<32>.ret,
+                              [(add (decimate SGPR_32, 4)),
+                               (add (decimate (shl SGPR_32, 1), 4)),
+                               (add (decimate (shl SGPR_32, 2), 4)),
+                               (add (decimate (shl SGPR_32, 3), 4)),
+                               (add (decimate (shl SGPR_32, 4), 4)),
+                               (add (decimate (shl SGPR_32, 5), 4)),
+                               (add (decimate (shl SGPR_32, 6), 4)),
+                               (add (decimate (shl SGPR_32, 7), 4)),
+                               (add (decimate (shl SGPR_32, 8), 4)),
+                               (add (decimate (shl SGPR_32, 9), 4)),
+                               (add (decimate (shl SGPR_32, 10), 4)),
+                               (add (decimate (shl SGPR_32, 11), 4)),
+                               (add (decimate (shl SGPR_32, 12), 4)),
+                               (add (decimate (shl SGPR_32, 13), 4)),
+                               (add (decimate (shl SGPR_32, 14), 4)),
+                               (add (decimate (shl SGPR_32, 15), 4)),
+                               (add (decimate (shl SGPR_32, 16), 4)),
+                               (add (decimate (shl SGPR_32, 17), 4)),
+                               (add (decimate (shl SGPR_32, 18), 4)),
+                               (add (decimate (shl SGPR_32, 19), 4)),
+                               (add (decimate (shl SGPR_32, 20), 4)),
+                               (add (decimate (shl SGPR_32, 21), 4)),
+                               (add (decimate (shl SGPR_32, 22), 4)),
+                               (add (decimate (shl SGPR_32, 23), 4)),
+                               (add (decimate (shl SGPR_32, 24), 4)),
+                               (add (decimate (shl SGPR_32, 25), 4)),
+                               (add (decimate (shl SGPR_32, 26), 4)),
+                               (add (decimate (shl SGPR_32, 27), 4)),
+                               (add (decimate (shl SGPR_32, 28), 4)),
+                               (add (decimate (shl SGPR_32, 29), 4)),
+                               (add (decimate (shl SGPR_32, 30), 4)),
+                               (add (decimate (shl SGPR_32, 31), 4))]>;
+
 // Trap handler TMP 32-bit registers
 def TTMP_32 : RegisterClass<"AMDGPU", [i32, f32, v2i16, v2f16], 32,
                             (add (sequence "TTMP%u", 0, 15))> {
   let isAllocatable = 0;
 }
 
 // Trap handler TMP 64-bit registers
 def TTMP_64Regs : RegisterTuples<getSubRegs<2>.ret,

 
 def TTMP0_TTMP1_TTMP2_TTMP3_TTMP4_TTMP5_TTMP6_TTMP7_TTMP8_TTMP9_TTMP10_TTMP11_TTMP12_TTMP13_TTMP14_TTMP15_gfx9_gfx10 :
   TmpRegTuplesBase<0, 16,
                    [TTMP0_gfx9_gfx10, TTMP1_gfx9_gfx10, TTMP2_gfx9_gfx10, TTMP3_gfx9_gfx10,
                     TTMP4_gfx9_gfx10, TTMP5_gfx9_gfx10, TTMP6_gfx9_gfx10, TTMP7_gfx9_gfx10,
                     TTMP8_gfx9_gfx10, TTMP9_gfx9_gfx10, TTMP10_gfx9_gfx10, TTMP11_gfx9_gfx10,
                     TTMP12_gfx9_gfx10, TTMP13_gfx9_gfx10, TTMP14_gfx9_gfx10, TTMP15_gfx9_gfx10]>;
 
-
 // VGPR 32-bit registers
 // i16/f16 only on VI+
 def VGPR_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16], 32,
                             (add (sequence "VGPR%u", 0, 255))> {
   let AllocationPriority = 1;
   let Size = 32;
 }
 
def VGPR_512 : RegisterTuples<getSubRegs<16>.ret,
                                (add (shl VGPR_32, 9)),
                                (add (shl VGPR_32, 10)),
                                (add (shl VGPR_32, 11)),
                                (add (shl VGPR_32, 12)),
                                (add (shl VGPR_32, 13)),
                                (add (shl VGPR_32, 14)),
                                (add (shl VGPR_32, 15))]>;
 
+// VGPR 1024-bit registers
+def VGPR_1024 : RegisterTuples<getSubRegs<32>.ret,
+                              [(add (trunc VGPR_32, 225)),
+                               (add (shl VGPR_32, 1)),
+                               (add (shl VGPR_32, 2)),
+                               (add (shl VGPR_32, 3)),
+                               (add (shl VGPR_32, 4)),
+                               (add (shl VGPR_32, 5)),
+                               (add (shl VGPR_32, 6)),
+                               (add (shl VGPR_32, 7)),
+                               (add (shl VGPR_32, 8)),
+                               (add (shl VGPR_32, 9)),
+                               (add (shl VGPR_32, 10)),
+                               (add (shl VGPR_32, 11)),
+                               (add (shl VGPR_32, 12)),
+                               (add (shl VGPR_32, 13)),
+                               (add (shl VGPR_32, 14)),
+                               (add (shl VGPR_32, 15)),
+                               (add (shl VGPR_32, 16)),
+                               (add (shl VGPR_32, 17)),
+                               (add (shl VGPR_32, 18)),
+                               (add (shl VGPR_32, 19)),
+                               (add (shl VGPR_32, 20)),
+                               (add (shl VGPR_32, 21)),
+                               (add (shl VGPR_32, 22)),
+                               (add (shl VGPR_32, 23)),
+                               (add (shl VGPR_32, 24)),
+                               (add (shl VGPR_32, 25)),
+                               (add (shl VGPR_32, 26)),
+                               (add (shl VGPR_32, 27)),
+                               (add (shl VGPR_32, 28)),
+                               (add (shl VGPR_32, 29)),
+                               (add (shl VGPR_32, 30)),
+                               (add (shl VGPR_32, 31))]>;
+
+// AccVGPR 32-bit registers
+def AGPR_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16], 32,
+                            (add (sequence "AGPR%u", 0, 255))> {
+  let AllocationPriority = 1;
+  let Size = 32;
+}
+
+// AGPR 64-bit registers
+def AGPR_64 : RegisterTuples<getSubRegs<2>.ret,
+                             [(add (trunc AGPR_32, 255)),
+                              (add (shl AGPR_32, 1))]>;
+
+// AGPR 128-bit registers
+def AGPR_128 : RegisterTuples<getSubRegs<4>.ret,
+                              [(add (trunc AGPR_32, 253)),
+                               (add (shl AGPR_32, 1)),
+                               (add (shl AGPR_32, 2)),
+                               (add (shl AGPR_32, 3))]>;
+
+// AGPR 512-bit registers
+def AGPR_512 : RegisterTuples<getSubRegs<16>.ret,
+                              [(add (trunc AGPR_32, 241)),
+                               (add (shl AGPR_32, 1)),
+                               (add (shl AGPR_32, 2)),
+                               (add (shl AGPR_32, 3)),
+                               (add (shl AGPR_32, 4)),
+                               (add (shl AGPR_32, 5)),
+                               (add (shl AGPR_32, 6)),
+                               (add (shl AGPR_32, 7)),
+                               (add (shl AGPR_32, 8)),
+                               (add (shl AGPR_32, 9)),
+                               (add (shl AGPR_32, 10)),
+                               (add (shl AGPR_32, 11)),
+                               (add (shl AGPR_32, 12)),
+                               (add (shl AGPR_32, 13)),
+                               (add (shl AGPR_32, 14)),
+                               (add (shl AGPR_32, 15))]>;
+
+// AGPR 1024-bit registers
+def AGPR_1024 : RegisterTuples<getSubRegs<32>.ret,
+                              [(add (trunc AGPR_32, 225)),
+                               (add (shl AGPR_32, 1)),
+                               (add (shl AGPR_32, 2)),
+                               (add (shl AGPR_32, 3)),
+                               (add (shl AGPR_32, 4)),
+                               (add (shl AGPR_32, 5)),
+                               (add (shl AGPR_32, 6)),
+                               (add (shl AGPR_32, 7)),
+                               (add (shl AGPR_32, 8)),
+                               (add (shl AGPR_32, 9)),
+                               (add (shl AGPR_32, 10)),
+                               (add (shl AGPR_32, 11)),
+                               (add (shl AGPR_32, 12)),
+                               (add (shl AGPR_32, 13)),
+                               (add (shl AGPR_32, 14)),
+                               (add (shl AGPR_32, 15)),
+                               (add (shl AGPR_32, 16)),
+                               (add (shl AGPR_32, 17)),
+                               (add (shl AGPR_32, 18)),
+                               (add (shl AGPR_32, 19)),
+                               (add (shl AGPR_32, 20)),
+                               (add (shl AGPR_32, 21)),
+                               (add (shl AGPR_32, 22)),
+                               (add (shl AGPR_32, 23)),
+                               (add (shl AGPR_32, 24)),
+                               (add (shl AGPR_32, 25)),
+                               (add (shl AGPR_32, 26)),
+                               (add (shl AGPR_32, 27)),
+                               (add (shl AGPR_32, 28)),
+                               (add (shl AGPR_32, 29)),
+                               (add (shl AGPR_32, 30)),
+                               (add (shl AGPR_32, 31))]>;
+
 //===----------------------------------------------------------------------===//
 //  Register classes used as source and destination
 //===----------------------------------------------------------------------===//
 
 def Pseudo_SReg_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16], 32,
   (add FP_REG, SP_REG, SCRATCH_WAVE_OFFSET_REG)> {
   let isAllocatable = 0;
   let CopyCost = -1;

 
 // Subset of SReg_32 without M0 for SMRD instructions and alike.
 // See comments in SIInstructions.td for more info.
 def SReg_32_XM0_XEXEC : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16, i1], 32,
   (add SGPR_32, VCC_LO, VCC_HI, FLAT_SCR_LO, FLAT_SCR_HI, XNACK_MASK_LO, XNACK_MASK_HI,
    SGPR_NULL, TTMP_32, TMA_LO, TMA_HI, TBA_LO, TBA_HI, SRC_SHARED_BASE, SRC_SHARED_LIMIT,
    SRC_PRIVATE_BASE, SRC_PRIVATE_LIMIT, SRC_POPS_EXITING_WAVE_ID,
    SRC_VCCZ, SRC_EXECZ, SRC_SCC)> {
-  let AllocationPriority = 8;
+  let AllocationPriority = 10;
 }
 
 def SReg_32_XEXEC_HI : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16, i1], 32,
   (add SReg_32_XM0_XEXEC, EXEC_LO, M0_CLASS)> {
-  let AllocationPriority = 8;
+  let AllocationPriority = 10;
 }
 
 def SReg_32_XM0 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16, i1], 32,
   (add SReg_32_XM0_XEXEC, EXEC_LO, EXEC_HI)> {
-  let AllocationPriority = 8;
+  let AllocationPriority = 10;
 }
 
 // Register class for all scalar registers (SGPRs + Special Registers)
 def SReg_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16, i1], 32,
   (add SReg_32_XM0, M0_CLASS, EXEC_LO, EXEC_HI, SReg_32_XEXEC_HI)> {
-  let AllocationPriority = 8;
+  let AllocationPriority = 10;
 }
 
 def SRegOrLds_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16, i1], 32,
   (add SReg_32_XM0, M0_CLASS, EXEC_LO, EXEC_HI, SReg_32_XEXEC_HI, LDS_DIRECT_CLASS)> {
   let isAllocatable = 0;
 }
 
 def SGPR_64 : RegisterClass<"AMDGPU", [v2i32, i64, v2f32, f64, v4i16, v4f16], 32, (add SGPR_64Regs)> {
   let CopyCost = 1;
-  let AllocationPriority = 9;
+  let AllocationPriority = 11;
 }
 
 // CCR (call clobbered registers) SGPR 64-bit registers
 def CCR_SGPR_64 : RegisterClass<"AMDGPU", SGPR_64.RegTypes, 32, (add (trunc SGPR_64, 16))> {
   let CopyCost = SGPR_64.CopyCost;
   let AllocationPriority = SGPR_64.AllocationPriority;
 }
 
 def TTMP_64 : RegisterClass<"AMDGPU", [v2i32, i64, f64, v4i16, v4f16], 32, (add TTMP_64Regs)> {
   let isAllocatable = 0;
 }
 
 def SReg_64_XEXEC : RegisterClass<"AMDGPU", [v2i32, i64, v2f32, f64, i1, v4i16, v4f16], 32,
   (add SGPR_64, VCC, FLAT_SCR, XNACK_MASK, TTMP_64, TBA, TMA)> {
   let CopyCost = 1;
-  let AllocationPriority = 9;
+  let AllocationPriority = 13;
 }
 
 def SReg_64 : RegisterClass<"AMDGPU", [v2i32, i64, v2f32, f64, i1, v4i16, v4f16], 32,
   (add SReg_64_XEXEC, EXEC)> {
   let CopyCost = 1;
-  let AllocationPriority = 9;
+  let AllocationPriority = 13;
 }
 
 def SReg_1_XEXEC : RegisterClass<"AMDGPU", [i1], 32,
   (add SReg_64_XEXEC, SReg_32_XM0_XEXEC)> {
   let CopyCost = 1;
   let isAllocatable = 0;
 }
 
 def SReg_1 : RegisterClass<"AMDGPU", [i1], 32,
   (add SReg_1_XEXEC, EXEC, EXEC_LO)> {
   let CopyCost = 1;
   let isAllocatable = 0;
 }
 
 // Requires 2 s_mov_b64 to copy
 let CopyCost = 2 in {
 
 // There are no 3-component scalar instructions, but this is needed
 // for symmetry with VGPRs.
 def SGPR_96 : RegisterClass<"AMDGPU", [v3i32, v3f32], 32,
   (add SGPR_96Regs)> {
-  let AllocationPriority = 10;
+  let AllocationPriority = 14;
 }
 
 def SReg_96 : RegisterClass<"AMDGPU", [v3i32, v3f32], 32,
   (add SGPR_96)> {
-  let AllocationPriority = 10;
+  let AllocationPriority = 14;
 }
 
 def SGPR_128 : RegisterClass<"AMDGPU", [v4i32, v4f32, v2i64], 32, (add SGPR_128Regs)> {
-  let AllocationPriority = 11;
+  let AllocationPriority = 15;
 }
 
 def TTMP_128 : RegisterClass<"AMDGPU", [v4i32, v4f32, v2i64], 32, (add TTMP_128Regs)> {
   let isAllocatable = 0;
 }
 
 def SReg_128 : RegisterClass<"AMDGPU", [v4i32, v4f32, v2i64, v2f64], 32,
   (add SGPR_128, TTMP_128)> {
-  let AllocationPriority = 11;
+  let AllocationPriority = 15;
 }
 
 } // End CopyCost = 2
 
 // There are no 5-component scalar instructions, but this is needed
 // for symmetry with VGPRs.
 def SGPR_160 : RegisterClass<"AMDGPU", [v5i32, v5f32], 32,
   (add SGPR_160Regs)> {
-  let AllocationPriority = 12;
+  let AllocationPriority = 16;
 }
 
 def SReg_160 : RegisterClass<"AMDGPU", [v5i32, v5f32], 32,
   (add SGPR_160)> {
-  let AllocationPriority = 12;
+  let AllocationPriority = 16;
 }
 
 def SGPR_256 : RegisterClass<"AMDGPU", [v8i32, v8f32], 32, (add SGPR_256Regs)> {
-  let AllocationPriority = 13;
+  let AllocationPriority = 17;
 }
 
 def TTMP_256 : RegisterClass<"AMDGPU", [v8i32, v8f32], 32, (add TTMP_256Regs)> {
   let isAllocatable = 0;
 }
 
 def SReg_256 : RegisterClass<"AMDGPU", [v8i32, v8f32], 32,
   (add SGPR_256, TTMP_256)> {
   // Requires 4 s_mov_b64 to copy
   let CopyCost = 4;
-  let AllocationPriority = 13;
+  let AllocationPriority = 17;
 }
 
 def SGPR_512 : RegisterClass<"AMDGPU", [v16i32, v16f32], 32, (add SGPR_512Regs)> {
-  let AllocationPriority = 14;
+  let AllocationPriority = 18;
 }
 
 def TTMP_512 : RegisterClass<"AMDGPU", [v16i32, v16f32], 32, (add TTMP_512Regs)> {
   let isAllocatable = 0;
 }
 
 def SReg_512 : RegisterClass<"AMDGPU", [v16i32, v16f32], 32,
   (add SGPR_512, TTMP_512)> {
   // Requires 8 s_mov_b64 to copy
   let CopyCost = 8;
-  let AllocationPriority = 14;
+  let AllocationPriority = 18;
 }
 
 def VRegOrLds_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16], 32,
                                  (add VGPR_32, LDS_DIRECT_CLASS)> {
   let isAllocatable = 0;
 }
 
+def SGPR_1024 : RegisterClass<"AMDGPU", [v32i32], 32, (add SGPR_1024Regs)> {
+  let AllocationPriority = 19;
+}
+
+def SReg_1024 : RegisterClass<"AMDGPU", [v32i32], 32,
+  (add SGPR_1024)> {
+  let CopyCost = 16;
+  let AllocationPriority = 19;
+}
+
 // Register class for all vector registers (VGPRs + Interploation Registers)
 def VReg_64 : RegisterClass<"AMDGPU", [i64, f64, v2i32, v2f32, v4f16, v4i16], 32, (add VGPR_64)> {
   let Size = 64;
 
   // Requires 2 v_mov_b32 to copy
   let CopyCost = 2;
   let AllocationPriority = 2;
 }

 }
 
 def VReg_512 : RegisterClass<"AMDGPU", [v16i32, v16f32], 32, (add VGPR_512)> {
   let Size = 512;
   let CopyCost = 16;
   let AllocationPriority = 7;
 }
 
+def VReg_1024 : RegisterClass<"AMDGPU", [v32i32], 32, (add VGPR_1024)> {
+  let Size = 1024;
+  let CopyCost = 32;
+  let AllocationPriority = 8;
+}
+
+def AReg_64 : RegisterClass<"AMDGPU", [i64, f64, v2i32, v2f32, v4f16, v4i16], 32, (add AGPR_64)> {
+  let Size = 64;
+
+  let CopyCost = 5;
+  let AllocationPriority = 2;
+}
+
+def AReg_128 : RegisterClass<"AMDGPU", [v4i32, v4f32, v2i64, v2f64], 32, (add AGPR_128)> {
+  let Size = 128;
+
+  // Requires 4 v_accvgpr_write and 4 v_accvgpr_read to copy + burn 1 vgpr
+  let CopyCost = 9;
+  let AllocationPriority = 4;
+}
+
+def AReg_512 : RegisterClass<"AMDGPU", [v16i32, v16f32], 32, (add AGPR_512)> {
+  let Size = 512;
+  let CopyCost = 33;
+  let AllocationPriority = 7;
+}
+
+// TODO: add v32f32 value type
+def AReg_1024 : RegisterClass<"AMDGPU", [v32i32], 32, (add AGPR_1024)> {
+  let Size = 1024;
+  let CopyCost = 65;
+  let AllocationPriority = 8;
+}
+
 def VReg_1 : RegisterClass<"AMDGPU", [i1], 32, (add VGPR_32)> {
   let Size = 32;
 }
 
 def VS_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16], 32,
                           (add VGPR_32, SReg_32, LDS_DIRECT_CLASS)> {
   let isAllocatable = 0;
 }
 
 def VS_64 : RegisterClass<"AMDGPU", [i64, f64], 32, (add VReg_64, SReg_64)> {
   let isAllocatable = 0;
 }
 
+def AV_32 : RegisterClass<"AMDGPU", [i32, f32, i16, f16, v2i16, v2f16], 32,
+                          (add AGPR_32, VGPR_32)> {
+  let isAllocatable = 0;
+}
+
+def AV_64 : RegisterClass<"AMDGPU", [i64, f64, v4f16], 32, (add AReg_64, VReg_64)> {
+  let isAllocatable = 0;
+}
+
 //===----------------------------------------------------------------------===//
 //  Register operands
 //===----------------------------------------------------------------------===//
 
 class RegImmMatcher<string name> : AsmOperandClass {
   let Name = name;
   let RenderMethod = "addRegOrImmOperands";
 }
 
-multiclass SIRegOperand <string rc, string MatchName, string opType> {
+multiclass SIRegOperand32 <string rc, string MatchName, string opType,
+                           string rc_suffix = "_32"> {
   let OperandNamespace = "AMDGPU" in {
-    def _b16 : RegisterOperand<!cast<RegisterClass>(rc#"_32")> {
+    def _b16 : RegisterOperand<!cast<RegisterClass>(rc#rc_suffix)> {
       let OperandType = opType#"_INT16";
       let ParserMatchClass = RegImmMatcher<MatchName#"B16">;
       let DecoderMethod = "decodeOperand_VSrc16";
     }
 
-    def _f16 : RegisterOperand<!cast<RegisterClass>(rc#"_32")> {
+    def _f16 : RegisterOperand<!cast<RegisterClass>(rc#rc_suffix)> {
       let OperandType = opType#"_FP16";
       let ParserMatchClass = RegImmMatcher<MatchName#"F16">;
-      let DecoderMethod = "decodeOperand_VSrc16";
+      let DecoderMethod = "decodeOperand_" # rc # "_16";
     }
 
-    def _b32 : RegisterOperand<!cast<RegisterClass>(rc#"_32")> {
+    def _b32 : RegisterOperand<!cast<RegisterClass>(rc#rc_suffix)> {
       let OperandType = opType#"_INT32";
       let ParserMatchClass = RegImmMatcher<MatchName#"B32">;
+      let DecoderMethod = "decodeOperand_" # rc # rc_suffix;
     }
 
-    def _f32 : RegisterOperand<!cast<RegisterClass>(rc#"_32")> {
+    def _f32 : RegisterOperand<!cast<RegisterClass>(rc#rc_suffix)> {
       let OperandType = opType#"_FP32";
       let ParserMatchClass = RegImmMatcher<MatchName#"F32">;
+      let DecoderMethod = "decodeOperand_" # rc # rc_suffix;
     }
 
-    def _b64 : RegisterOperand<!cast<RegisterClass>(rc#"_64")> {
-      let OperandType = opType#"_INT64";
-      let ParserMatchClass = RegImmMatcher<MatchName#"B64">;
-    }
-
-    def _f64 : RegisterOperand<!cast<RegisterClass>(rc#"_64")> {
-      let OperandType = opType#"_FP64";
-      let ParserMatchClass = RegImmMatcher<MatchName#"F64">;
-    }
-
-    def _v2b16 : RegisterOperand<!cast<RegisterClass>(rc#"_32")> {
+    def _v2b16 : RegisterOperand<!cast<RegisterClass>(rc#rc_suffix)> {
       let OperandType = opType#"_V2INT16";
       let ParserMatchClass = RegImmMatcher<MatchName#"V2B16">;
       let DecoderMethod = "decodeOperand_VSrcV216";
     }
 
-    def _v2f16 : RegisterOperand<!cast<RegisterClass>(rc#"_32")> {
+    def _v2f16 : RegisterOperand<!cast<RegisterClass>(rc#rc_suffix)> {
       let OperandType = opType#"_V2FP16";
       let ParserMatchClass = RegImmMatcher<MatchName#"V2F16">;
       let DecoderMethod = "decodeOperand_VSrcV216";
     }
   }
 }
 
+multiclass SIRegOperand <string rc, string MatchName, string opType> :
+  SIRegOperand32<rc, MatchName, opType> {
+  let OperandNamespace = "AMDGPU" in {
+    def _b64 : RegisterOperand<!cast<RegisterClass>(rc#"_64")> {
+      let OperandType = opType#"_INT64";
+      let ParserMatchClass = RegImmMatcher<MatchName#"B64">;
+    }
+
+    def _f64 : RegisterOperand<!cast<RegisterClass>(rc#"_64")> {
+      let OperandType = opType#"_FP64";
+      let ParserMatchClass = RegImmMatcher<MatchName#"F64">;
+    }
+  }
+}
+
 // FIXME: 64-bit sources can sometimes use 32-bit constants.
 multiclass RegImmOperand <string rc, string MatchName>
   : SIRegOperand<rc, MatchName, "OPERAND_REG_IMM">;
 
 multiclass RegInlineOperand <string rc, string MatchName>
   : SIRegOperand<rc, MatchName, "OPERAND_REG_INLINE_C">;
 
+multiclass RegInlineOperand32 <string rc, string MatchName,
+                               string rc_suffix = "_32">
+  : SIRegOperand32<rc, MatchName, "OPERAND_REG_INLINE_C", rc_suffix>;
+
+multiclass RegInlineOperandAC <string rc, string MatchName,
+                               string rc_suffix = "_32">
+  : SIRegOperand32<rc, MatchName, "OPERAND_REG_INLINE_AC", rc_suffix>;
+
 //===----------------------------------------------------------------------===//
 //  SSrc_* Operands with an SGPR or a 32-bit immediate
 //===----------------------------------------------------------------------===//
 
 defm SSrc : RegImmOperand<"SReg", "SSrc">;
 
 def SSrcOrLds_b32 : RegisterOperand<SRegOrLds_32> {
   let OperandNamespace = "AMDGPU";

 // 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";
 }
 
 //===----------------------------------------------------------------------===//
+//  ASrc_* Operands with an AccVGPR
+//===----------------------------------------------------------------------===//
+
+def ARegSrc_32 : RegisterOperand<AGPR_32> {
+  let DecoderMethod = "DecodeAGPR_32RegisterClass";
+  let EncoderMethod = "getAVOperandEncoding";
+}
+
+//===----------------------------------------------------------------------===//
 //  VCSrc_* Operands with an SGPR, VGPR or an inline constant
 //===----------------------------------------------------------------------===//
 
 defm VCSrc : RegInlineOperand<"VS", "VCSrc">;
+
+//===----------------------------------------------------------------------===//
+//  VISrc_* Operands with a VGPR or an inline constant
+//===----------------------------------------------------------------------===//
+
+defm VISrc : RegInlineOperand32<"VGPR", "VISrc">;
+
+//===----------------------------------------------------------------------===//
+//  AVSrc_* Operands with an AGPR or VGPR
+//===----------------------------------------------------------------------===//
+
+def AVSrc_32 : RegisterOperand<AV_32> {
+  let DecoderMethod = "DecodeAV_32RegisterClass";
+  let EncoderMethod = "getAVOperandEncoding";
+}
+
+def AVSrc_64 : RegisterOperand<AV_64> {
+  let DecoderMethod = "DecodeAV_64RegisterClass";
+  let EncoderMethod = "getAVOperandEncoding";
+}
+
+//===----------------------------------------------------------------------===//
+//  ACSrc_* Operands with an AGPR or an inline constant
+//===----------------------------------------------------------------------===//
+
+defm AISrc      : RegInlineOperandAC<"AGPR", "AISrc">;
+defm AISrc_128  : RegInlineOperandAC<"AReg", "AISrc_128",  "_128">;
+defm AISrc_512  : RegInlineOperandAC<"AReg", "AISrc_512",  "_512">;
+defm AISrc_1024 : RegInlineOperandAC<"AReg", "AISrc_1024", "_1024">;

test/CodeGen/AMDGPU/ipra-regmask.ll

 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -enable-ipra -print-regusage -o /dev/null 2>&1 < %s | FileCheck %s
 ; Make sure the expected regmask is generated for sub/superregisters.
 
-; CHECK-DAG: csr Clobbered Registers: $vgpr0 $vgpr0_vgpr1_vgpr2_vgpr3 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 $vgpr0_vgpr1 $vgpr0_vgpr1_vgpr2 {{$}}
+; CHECK-DAG: csr Clobbered Registers: $vgpr0 $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 $vgpr0_vgpr1_vgpr2_vgpr3 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 $vgpr0_vgpr1 $vgpr0_vgpr1_vgpr2 {{$}}
 define void @csr() #0 {
   call void asm sideeffect "", "~{v0},~{v36},~{v37}"() #0
   ret void
 }
 
-; CHECK-DAG: subregs_for_super Clobbered Registers: $vgpr0 $vgpr1 $vgpr0_vgpr1_vgpr2_vgpr3 $vgpr1_vgpr2_vgpr3_vgpr4 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16 $vgpr0_vgpr1 $vgpr1_vgpr2 $vgpr0_vgpr1_vgpr2 $vgpr1_vgpr2_vgpr3 {{$}}
+; CHECK-DAG: subregs_for_super Clobbered Registers: $vgpr0 $vgpr1 $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 $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 $vgpr0_vgpr1_vgpr2_vgpr3 $vgpr1_vgpr2_vgpr3_vgpr4 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16 $vgpr0_vgpr1 $vgpr1_vgpr2 $vgpr0_vgpr1_vgpr2 $vgpr1_vgpr2_vgpr3 {{$}}
 define void @subregs_for_super() #0 {
   call void asm sideeffect "", "~{v0},~{v1}"() #0
   ret void
 }
 
-; CHECK-DAG: clobbered_reg_with_sub Clobbered Registers: $vgpr0 $vgpr1 $vgpr0_vgpr1_vgpr2_vgpr3 $vgpr1_vgpr2_vgpr3_vgpr4 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16 $vgpr0_vgpr1 $vgpr1_vgpr2 $vgpr0_vgpr1_vgpr2 $vgpr1_vgpr2_vgpr3 {{$}}
+; CHECK-DAG: Clobbered Registers: $vgpr0 $vgpr1 $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 $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 $vgpr0_vgpr1_vgpr2_vgpr3 $vgpr1_vgpr2_vgpr3_vgpr4 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 $vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16 $vgpr0_vgpr1 $vgpr1_vgpr2 $vgpr0_vgpr1_vgpr2 $vgpr1_vgpr2_vgpr3 {{$}}
 define void @clobbered_reg_with_sub() #0 {
   call void asm sideeffect "", "~{v[0:1]}"() #0
   ret void
 }
 
 ; CHECK-DAG: nothing Clobbered Registers: {{$}}
 define void @nothing() #0 {
   ret void