[AArch64][SVE] Asm: Support for saturating INC/DEC (32bit scalar) instructions.

The variants added by this patch are:
- SQINC     signed increment, e.g. sqinc x0, w0, all, mul #4
- SQDEC     signed decrement, e.g. sqdec x0, w0, all, mul #4
- UQINC   unsigned increment, e.g. uqinc w0, all, mul #4
- UQDEC   unsigned decrement, e.g. uqdec w0, all, mul #4
 
This patch includes asmparser changes to parse a GPR64 as a GPR32 in
order to satisfy the constraint check:
  x0 == GPR64(w0)
in:
  sqinc x0, w0, all, mul #4
         ^___^ (must match)

Reviewers: rengolin, fhahn, SjoerdMeijer, samparker, javed.absar

Reviewed By: fhahn

Differential Revision: https://reviews.llvm.org/D47716



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334980 91177308-0d34-0410-b5e6-96231b3b80d8

Author: sdesmalen-arm

lib/Target/AArch64/AArch64InstrFormats.td

@@ -179,11 +179,23 @@ def CondCode : AsmOperandClass {
 // A 32-bit register pasrsed as 64-bit
 def GPR32as64Operand : AsmOperandClass {
   let Name = "GPR32as64";
+  let ParserMethod =
+      "tryParseGPROperand<false, RegConstraintEqualityTy::EqualsSubReg>";
 }
 def GPR32as64 : RegisterOperand<GPR32> {
   let ParserMatchClass = GPR32as64Operand;
 }
 
+// A 64-bit register pasrsed as 32-bit
+def GPR64as32Operand : AsmOperandClass {
+  let Name = "GPR64as32";
+  let ParserMethod =
+      "tryParseGPROperand<false, RegConstraintEqualityTy::EqualsSuperReg>";
+}
+def GPR64as32 : RegisterOperand<GPR64, "printGPR64as32"> {
+  let ParserMatchClass = GPR64as32Operand;
+}
+
 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
 // are encoded as the eight bit value 'abcdefgh'.

lib/Target/AArch64/AArch64SVEInstrInfo.td

@@ -527,21 +527,37 @@ let Predicates = [HasSVE] in {
   defm CMPLO_WIDE_PPzZZ : sve_int_cmp_1_wide<0b110, "cmplo">;
   defm CMPLS_WIDE_PPzZZ : sve_int_cmp_1_wide<0b111, "cmpls">;
 
+  defm SQINCB_XPiWdI : sve_int_pred_pattern_b_s32<0b00000, "sqincb">;
+  defm UQINCB_WPiI   : sve_int_pred_pattern_b_u32<0b00001, "uqincb">;
+  defm SQDECB_XPiWdI : sve_int_pred_pattern_b_s32<0b00010, "sqdecb">;
+  defm UQDECB_WPiI   : sve_int_pred_pattern_b_u32<0b00011, "uqdecb">;
   defm SQINCB_XPiI   : sve_int_pred_pattern_b_x64<0b00100, "sqincb">;
   defm UQINCB_XPiI   : sve_int_pred_pattern_b_x64<0b00101, "uqincb">;
   defm SQDECB_XPiI   : sve_int_pred_pattern_b_x64<0b00110, "sqdecb">;
   defm UQDECB_XPiI   : sve_int_pred_pattern_b_x64<0b00111, "uqdecb">;
 
+  defm SQINCH_XPiWdI : sve_int_pred_pattern_b_s32<0b01000, "sqinch">;
+  defm UQINCH_WPiI   : sve_int_pred_pattern_b_u32<0b01001, "uqinch">;
+  defm SQDECH_XPiWdI : sve_int_pred_pattern_b_s32<0b01010, "sqdech">;
+  defm UQDECH_WPiI   : sve_int_pred_pattern_b_u32<0b01011, "uqdech">;
   defm SQINCH_XPiI   : sve_int_pred_pattern_b_x64<0b01100, "sqinch">;
   defm UQINCH_XPiI   : sve_int_pred_pattern_b_x64<0b01101, "uqinch">;
   defm SQDECH_XPiI   : sve_int_pred_pattern_b_x64<0b01110, "sqdech">;
   defm UQDECH_XPiI   : sve_int_pred_pattern_b_x64<0b01111, "uqdech">;
 
+  defm SQINCW_XPiWdI : sve_int_pred_pattern_b_s32<0b10000, "sqincw">;
+  defm UQINCW_WPiI   : sve_int_pred_pattern_b_u32<0b10001, "uqincw">;
+  defm SQDECW_XPiWdI : sve_int_pred_pattern_b_s32<0b10010, "sqdecw">;
+  defm UQDECW_WPiI   : sve_int_pred_pattern_b_u32<0b10011, "uqdecw">;
   defm SQINCW_XPiI   : sve_int_pred_pattern_b_x64<0b10100, "sqincw">;
   defm UQINCW_XPiI   : sve_int_pred_pattern_b_x64<0b10101, "uqincw">;
   defm SQDECW_XPiI   : sve_int_pred_pattern_b_x64<0b10110, "sqdecw">;
   defm UQDECW_XPiI   : sve_int_pred_pattern_b_x64<0b10111, "uqdecw">;
 
+  defm SQINCD_XPiWdI : sve_int_pred_pattern_b_s32<0b11000, "sqincd">;
+  defm UQINCD_WPiI   : sve_int_pred_pattern_b_u32<0b11001, "uqincd">;
+  defm SQDECD_XPiWdI : sve_int_pred_pattern_b_s32<0b11010, "sqdecd">;
+  defm UQDECD_WPiI   : sve_int_pred_pattern_b_u32<0b11011, "uqdecd">;
   defm SQINCD_XPiI   : sve_int_pred_pattern_b_x64<0b11100, "sqincd">;
   defm UQINCD_XPiI   : sve_int_pred_pattern_b_x64<0b11101, "uqincd">;
   defm SQDECD_XPiI   : sve_int_pred_pattern_b_x64<0b11110, "sqdecd">;

lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp

@@ -66,6 +66,12 @@ enum class RegKind {
   SVEPredicateVector
 };
 
+enum RegConstraintEqualityTy {
+  EqualsReg,
+  EqualsSuperReg,
+  EqualsSubReg
+};
+
 class AArch64AsmParser : public MCTargetAsmParser {
 private:
   StringRef Mnemonic; ///< Instruction mnemonic.
@@ -92,7 +98,8 @@ class AArch64AsmParser : public MCTargetAsmParser {
   bool parseOperand(OperandVector &Operands, bool isCondCode,
                     bool invertCondCode);
 
-  bool showMatchError(SMLoc Loc, unsigned ErrCode, OperandVector &Operands);
+  bool showMatchError(SMLoc Loc, unsigned ErrCode, uint64_t ErrorInfo,
+                      OperandVector &Operands);
 
   bool parseDirectiveArch(SMLoc L);
   bool parseDirectiveCPU(SMLoc L);
@@ -139,7 +146,8 @@ class AArch64AsmParser : public MCTargetAsmParser {
   bool tryParseNeonVectorRegister(OperandVector &Operands);
   OperandMatchResultTy tryParseVectorIndex(OperandVector &Operands);
   OperandMatchResultTy tryParseGPRSeqPair(OperandVector &Operands);
-  template <bool ParseShiftExtend>
+  template <bool ParseShiftExtend,
+            RegConstraintEqualityTy EqTy = RegConstraintEqualityTy::EqualsReg>
   OperandMatchResultTy tryParseGPROperand(OperandVector &Operands);
   template <bool ParseShiftExtend, bool ParseSuffix>
   OperandMatchResultTy tryParseSVEDataVector(OperandVector &Operands);
@@ -177,6 +185,8 @@ class AArch64AsmParser : public MCTargetAsmParser {
     setAvailableFeatures(ComputeAvailableFeatures(getSTI().getFeatureBits()));
   }
 
+  bool regsEqual(const MCParsedAsmOperand &Op1,
+                 const MCParsedAsmOperand &Op2) const override;
   bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
                         SMLoc NameLoc, OperandVector &Operands) override;
   bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc) override;
@@ -231,6 +241,10 @@ class AArch64Operand : public MCParsedAsmOperand {
     RegKind Kind;
     int ElementWidth;
 
+    // The register may be allowed as a different register class,
+    // e.g. for GPR64as32 or GPR32as64.
+    RegConstraintEqualityTy EqualityTy;
+
     // In some cases the shift/extend needs to be explicitly parsed together
     // with the register, rather than as a separate operand. This is needed
     // for addressing modes where the instruction as a whole dictates the
@@ -446,6 +460,11 @@ class AArch64Operand : public MCParsedAsmOperand {
     return Reg.RegNum;
   }
 
+  RegConstraintEqualityTy getRegEqualityTy() const {
+    assert(Kind == k_Register && "Invalid access!");
+    return Reg.EqualityTy;
+  }
+
   unsigned getVectorListStart() const {
     assert(Kind == k_VectorList && "Invalid access!");
     return VectorList.RegNum;
@@ -1002,6 +1021,11 @@ class AArch64Operand : public MCParsedAsmOperand {
       AArch64MCRegisterClasses[AArch64::GPR64RegClassID].contains(Reg.RegNum);
   }
 
+  bool isGPR64as32() const {
+    return Kind == k_Register && Reg.Kind == RegKind::Scalar &&
+      AArch64MCRegisterClasses[AArch64::GPR32RegClassID].contains(Reg.RegNum);
+  }
+
   bool isWSeqPair() const {
     return Kind == k_Register && Reg.Kind == RegKind::Scalar &&
            AArch64MCRegisterClasses[AArch64::WSeqPairsClassRegClassID].contains(
@@ -1318,6 +1342,18 @@ class AArch64Operand : public MCParsedAsmOperand {
     Inst.addOperand(MCOperand::createReg(Reg));
   }
 
+  void addGPR64as32Operands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    assert(
+        AArch64MCRegisterClasses[AArch64::GPR32RegClassID].contains(getReg()));
+
+    const MCRegisterInfo *RI = Ctx.getRegisterInfo();
+    uint32_t Reg = RI->getRegClass(AArch64::GPR64RegClassID).getRegister(
+        RI->getEncodingValue(getReg()));
+
+    Inst.addOperand(MCOperand::createReg(Reg));
+  }
+
   template <int Width>
   void addFPRasZPRRegOperands(MCInst &Inst, unsigned N) const {
     unsigned Base;
@@ -1668,13 +1704,15 @@ class AArch64Operand : public MCParsedAsmOperand {
 
   static std::unique_ptr<AArch64Operand>
   CreateReg(unsigned RegNum, RegKind Kind, SMLoc S, SMLoc E, MCContext &Ctx,
+            RegConstraintEqualityTy EqTy = RegConstraintEqualityTy::EqualsReg,
             AArch64_AM::ShiftExtendType ExtTy = AArch64_AM::LSL,
             unsigned ShiftAmount = 0,
             unsigned HasExplicitAmount = false) {
     auto Op = make_unique<AArch64Operand>(k_Register, Ctx);
     Op->Reg.RegNum = RegNum;
     Op->Reg.Kind = Kind;
     Op->Reg.ElementWidth = 0;
+    Op->Reg.EqualityTy = EqTy;
     Op->Reg.ShiftExtend.Type = ExtTy;
     Op->Reg.ShiftExtend.Amount = ShiftAmount;
     Op->Reg.ShiftExtend.HasExplicitAmount = HasExplicitAmount;
@@ -1692,7 +1730,7 @@ class AArch64Operand : public MCParsedAsmOperand {
     assert((Kind == RegKind::NeonVector || Kind == RegKind::SVEDataVector ||
             Kind == RegKind::SVEPredicateVector) &&
            "Invalid vector kind");
-    auto Op = CreateReg(RegNum, Kind, S, E, Ctx, ExtTy, ShiftAmount,
+    auto Op = CreateReg(RegNum, Kind, S, E, Ctx, EqualsReg, ExtTy, ShiftAmount,
                         HasExplicitAmount);
     Op->Reg.ElementWidth = ElementWidth;
     return Op;
@@ -3164,7 +3202,7 @@ AArch64AsmParser::tryParseGPR64sp0Operand(OperandVector &Operands) {
   return MatchOperand_Success;
 }
 
-template <bool ParseShiftExtend>
+template <bool ParseShiftExtend, RegConstraintEqualityTy EqTy>
 OperandMatchResultTy
 AArch64AsmParser::tryParseGPROperand(OperandVector &Operands) {
   SMLoc StartLoc = getLoc();
@@ -3177,7 +3215,7 @@ AArch64AsmParser::tryParseGPROperand(OperandVector &Operands) {
   // No shift/extend is the default.
   if (!ParseShiftExtend || getParser().getTok().isNot(AsmToken::Comma)) {
     Operands.push_back(AArch64Operand::CreateReg(
-        RegNum, RegKind::Scalar, StartLoc, getLoc(), getContext()));
+        RegNum, RegKind::Scalar, StartLoc, getLoc(), getContext(), EqTy));
     return MatchOperand_Success;
   }
 
@@ -3191,10 +3229,10 @@ AArch64AsmParser::tryParseGPROperand(OperandVector &Operands) {
     return Res;
 
   auto Ext = static_cast<AArch64Operand*>(ExtOpnd.back().get());
-  Operands.push_back(AArch64Operand::CreateReg(RegNum, RegKind::Scalar,
-                     StartLoc, Ext->getEndLoc(), getContext(),
-                     Ext->getShiftExtendType(), Ext->getShiftExtendAmount(),
-                     Ext->hasShiftExtendAmount()));
+  Operands.push_back(AArch64Operand::CreateReg(
+      RegNum, RegKind::Scalar, StartLoc, Ext->getEndLoc(), getContext(), EqTy,
+      Ext->getShiftExtendType(), Ext->getShiftExtendAmount(),
+      Ext->hasShiftExtendAmount()));
 
   return MatchOperand_Success;
 }
@@ -3412,6 +3450,30 @@ bool AArch64AsmParser::parseOperand(OperandVector &Operands, bool isCondCode,
   }
 }
 
+bool AArch64AsmParser::regsEqual(const MCParsedAsmOperand &Op1,
+                                 const MCParsedAsmOperand &Op2) const {
+  auto &AOp1 = static_cast<const AArch64Operand&>(Op1);
+  auto &AOp2 = static_cast<const AArch64Operand&>(Op2);
+  if (AOp1.getRegEqualityTy() == RegConstraintEqualityTy::EqualsReg &&
+      AOp2.getRegEqualityTy() == RegConstraintEqualityTy::EqualsReg)
+    return MCTargetAsmParser::regsEqual(Op1, Op2);
+
+  assert(AOp1.isScalarReg() && AOp2.isScalarReg() &&
+         "Testing equality of non-scalar registers not supported");
+
+  // Check if a registers match their sub/super register classes.
+  if (AOp1.getRegEqualityTy() == EqualsSuperReg)
+    return getXRegFromWReg(Op1.getReg()) == Op2.getReg();
+  if (AOp1.getRegEqualityTy() == EqualsSubReg)
+    return getWRegFromXReg(Op1.getReg()) == Op2.getReg();
+  if (AOp2.getRegEqualityTy() == EqualsSuperReg)
+    return getXRegFromWReg(Op2.getReg()) == Op1.getReg();
+  if (AOp2.getRegEqualityTy() == EqualsSubReg)
+    return getWRegFromXReg(Op2.getReg()) == Op1.getReg();
+
+  return false;
+}
+
 /// ParseInstruction - Parse an AArch64 instruction mnemonic followed by its
 /// operands.
 bool AArch64AsmParser::ParseInstruction(ParseInstructionInfo &Info,
@@ -3765,10 +3827,22 @@ static std::string AArch64MnemonicSpellCheck(StringRef S, uint64_t FBS,
                                              unsigned VariantID = 0);
 
 bool AArch64AsmParser::showMatchError(SMLoc Loc, unsigned ErrCode,
+                                      uint64_t ErrorInfo,
                                       OperandVector &Operands) {
   switch (ErrCode) {
-  case Match_InvalidTiedOperand:
-    return Error(Loc, "operand must match destination register");
+  case Match_InvalidTiedOperand: {
+    RegConstraintEqualityTy EqTy =
+        static_cast<const AArch64Operand &>(*Operands[ErrorInfo])
+            .getRegEqualityTy();
+    switch (EqTy) {
+    case RegConstraintEqualityTy::EqualsSubReg:
+      return Error(Loc, "operand must be 64-bit form of destination register");
+    case RegConstraintEqualityTy::EqualsSuperReg:
+      return Error(Loc, "operand must be 32-bit form of destination register");
+    case RegConstraintEqualityTy::EqualsReg:
+      return Error(Loc, "operand must match destination register");
+    }
+  }
   case Match_MissingFeature:
     return Error(Loc,
                  "instruction requires a CPU feature not currently enabled");
@@ -4389,7 +4463,7 @@ bool AArch64AsmParser::MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
     return Error(IDLoc, Msg);
   }
   case Match_MnemonicFail:
-    return showMatchError(IDLoc, MatchResult, Operands);
+    return showMatchError(IDLoc, MatchResult, ErrorInfo, Operands);
   case Match_InvalidOperand: {
     SMLoc ErrorLoc = IDLoc;
 
@@ -4408,7 +4482,7 @@ bool AArch64AsmParser::MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
         ((AArch64Operand &)*Operands[ErrorInfo]).isTokenSuffix())
       MatchResult = Match_InvalidSuffix;
 
-    return showMatchError(ErrorLoc, MatchResult, Operands);
+    return showMatchError(ErrorLoc, MatchResult, ErrorInfo, Operands);
   }
   case Match_InvalidTiedOperand:
   case Match_InvalidMemoryIndexed1:
@@ -4546,7 +4620,7 @@ bool AArch64AsmParser::MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
     SMLoc ErrorLoc = ((AArch64Operand &)*Operands[ErrorInfo]).getStartLoc();
     if (ErrorLoc == SMLoc())
       ErrorLoc = IDLoc;
-    return showMatchError(ErrorLoc, MatchResult, Operands);
+    return showMatchError(ErrorLoc, MatchResult, ErrorInfo, Operands);
   }
   }
 

lib/Target/AArch64/InstPrinter/AArch64InstPrinter.cpp

@@ -1527,3 +1527,10 @@ void AArch64InstPrinter::printExactFPImm(const MCInst *MI, unsigned OpNum,
   unsigned Val = MI->getOperand(OpNum).getImm();
   O << "#" << (Val ? Imm1Desc->Repr : Imm0Desc->Repr);
 }
+
+void AArch64InstPrinter::printGPR64as32(const MCInst *MI, unsigned OpNum,
+                                        const MCSubtargetInfo &STI,
+                                        raw_ostream &O) {
+  unsigned Reg = MI->getOperand(OpNum).getReg();
+  O << getRegisterName(getWRegFromXReg(Reg));
+}

lib/Target/AArch64/InstPrinter/AArch64InstPrinter.h

@@ -180,6 +180,8 @@ class AArch64InstPrinter : public MCInstPrinter {
   template <char = 0>
   void printSVERegOp(const MCInst *MI, unsigned OpNum,
                     const MCSubtargetInfo &STI, raw_ostream &O);
+  void printGPR64as32(const MCInst *MI, unsigned OpNum,
+                      const MCSubtargetInfo &STI, raw_ostream &O);
   template <int Width>
   void printZPRasFPR(const MCInst *MI, unsigned OpNum,
                      const MCSubtargetInfo &STI, raw_ostream &O);

lib/Target/AArch64/SVEInstrFormats.td

@@ -333,9 +333,32 @@ class sve_int_pred_pattern_b<bits<5> opc, string asm, RegisterOperand dt,
   let Inst{9-5}   = pattern;
   let Inst{4-0}   = Rdn;
 
+  // Signed 32bit forms require their GPR operand printed.
+  let AsmString = !if(!eq(opc{2,0}, 0b00),
+                      !strconcat(asm, "\t$Rdn, $_Rdn, $pattern, mul $imm4"),
+                      !strconcat(asm, "\t$Rdn, $pattern, mul $imm4"));
+
   let Constraints = "$Rdn = $_Rdn";
 }
 
+multiclass sve_int_pred_pattern_b_s32<bits<5> opc, string asm> {
+  def NAME : sve_int_pred_pattern_b<opc, asm, GPR64z, GPR64as32>;
+
+  def : InstAlias<asm # "\t$Rd, $Rn, $pattern",
+                  (!cast<Instruction>(NAME) GPR64z:$Rd, GPR64as32:$Rn, sve_pred_enum:$pattern, 1), 1>;
+  def : InstAlias<asm # "\t$Rd, $Rn",
+                  (!cast<Instruction>(NAME) GPR64z:$Rd, GPR64as32:$Rn, 0b11111, 1), 2>;
+}
+
+multiclass sve_int_pred_pattern_b_u32<bits<5> opc, string asm> {
+  def NAME : sve_int_pred_pattern_b<opc, asm, GPR32z, GPR32z>;
+
+  def : InstAlias<asm # "\t$Rdn, $pattern",
+                  (!cast<Instruction>(NAME) GPR32z:$Rdn, sve_pred_enum:$pattern, 1), 1>;
+  def : InstAlias<asm # "\t$Rdn",
+                  (!cast<Instruction>(NAME) GPR32z:$Rdn, 0b11111, 1), 2>;
+}
+
 multiclass sve_int_pred_pattern_b_x64<bits<5> opc, string asm> {
   def NAME : sve_int_pred_pattern_b<opc, asm, GPR64z, GPR64z>;
 

test/MC/AArch64/SVE/sqdecb-diagnostics.s

@@ -19,6 +19,20 @@ sqdecb sp
 // CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
 
 
+// ------------------------------------------------------------------------- //
+// Operands not matching up
+
+sqdecb x0, w1
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: operand must be 32-bit form of destination register
+// CHECK-NEXT: sqdecb x0, w1
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+sqdecb x0, x1
+// CHECK: [[@LINE-1]]:{{[0-9]+}}: error: invalid predicate pattern
+// CHECK-NEXT: sqdecb x0, x1
+// CHECK-NOT: [[@LINE-1]]:{{[0-9]+}}:
+
+
 // ------------------------------------------------------------------------- //
 // Immediate not compatible with encode/decode function.