[AArch64][llvm] Add support for optional register in SYS alias instructions

Add support for future AArch64 instructions in the `SYS` alias encoding
space which may support an optional register as an operand. For example:

```
   SYS #4, c8, c5, #2, {<Xt>}
```

Currently, AArch64 `SYS` alias instructions fall into two categories:
  * a register value must be present (indicated by any value except `XZR`)
  * no register value must be present (this value must be `XZR`)

This is defined by the optional "needsreg" parameter in TableGen (e.g. see
the multiclass definition for TLBI), which defaults to requiring a register
operand for the instruction to be valid.

Expand the code so that if an optional register operand is specified
in TableGen, then if it's not `XZR`, encode this value in the `SYS` alias
and for disassembly, print this value if it's not `XZR`. Don't produce an
error message if the register operand is missing or unexpected, if it is
specified as an optional register.

If a mandatory or optional register is not specified in the TableGen
(i.e. no register operand should be present), then disassemble to a SYS
alias instead if the register value is not xzr/x31 (encoded as 0b11111).

For instructions taking no register operands, this is specified in the
Arm ARM, with language similar to:
```
  Rt should be encoded as 0b11111. If the Rt field is not set to 0b11111,
  it is CONSTRAINED UNPREDICTABLE whether:
    * The instruction is UNDEFINED.
    * The instruction behaves as if the Rt field is set to 0b11111.
```
and since we want to follow "should" directives, and not encourage
undefined behaviour, these are not considered valid instructions, so
should not be assembled or disassembled as such.

Author: jthackray

llvm/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp

@@ -3908,6 +3908,8 @@ bool AArch64AsmParser::parseSysAlias(StringRef Name, SMLoc NameLoc,
   StringRef Op = Tok.getString();
   SMLoc S = Tok.getLoc();
   bool ExpectRegister = true;
+  bool OptionalRegister = false;
+  bool hasAll = getSTI().hasFeature(AArch64::FeatureAll);
 
   if (Mnemonic == "ic") {
     const AArch64IC::IC *IC = AArch64IC::lookupICByName(Op);
@@ -3956,7 +3958,6 @@ bool AArch64AsmParser::parseSysAlias(StringRef Name, SMLoc NameLoc,
     if (Op.lower() != "rctx")
       return TokError("invalid operand for prediction restriction instruction");
 
-    bool hasAll = getSTI().hasFeature(AArch64::FeatureAll);
     bool hasPredres = hasAll || getSTI().hasFeature(AArch64::FeaturePredRes);
     bool hasSpecres2 = hasAll || getSTI().hasFeature(AArch64::FeatureSPECRES2);
 
@@ -3989,10 +3990,12 @@ bool AArch64AsmParser::parseSysAlias(StringRef Name, SMLoc NameLoc,
     HasRegister = true;
   }
 
-  if (ExpectRegister && !HasRegister)
-    return TokError("specified " + Mnemonic + " op requires a register");
-  else if (!ExpectRegister && HasRegister)
-    return TokError("specified " + Mnemonic + " op does not use a register");
+  if (!OptionalRegister) {
+    if (ExpectRegister && !HasRegister)
+      return TokError("specified " + Mnemonic + " op requires a register");
+    else if (!ExpectRegister && HasRegister)
+      return TokError("specified " + Mnemonic + " op does not use a register");
+  }
 
   if (parseToken(AsmToken::EndOfStatement, "unexpected token in argument list"))
     return true;

llvm/lib/Target/AArch64/MCTargetDesc/AArch64InstPrinter.cpp

@@ -917,6 +917,7 @@ bool AArch64InstPrinter::printSysAlias(const MCInst *MI,
   Encoding |= Op1Val << 11;
 
   bool NeedsReg;
+  bool OptionalReg = false;
   std::string Ins;
   std::string Name;
 
@@ -1017,13 +1018,24 @@ bool AArch64InstPrinter::printSysAlias(const MCInst *MI,
   else
     return false;
 
+  StringRef Reg = getRegisterName(MI->getOperand(4).getReg());
+  bool NotXZR = Reg != "xzr";
+
+  // If a mandatory or optional register is not specified in the TableGen
+  // (i.e. no register operand should be present), and the register value
+  // is not xzr/x31, then disassemble to a SYS alias instead.
+  if (NotXZR && !NeedsReg && !OptionalReg)
+    return false;
+
   std::string Str = Ins + Name;
   llvm::transform(Str, Str.begin(), ::tolower);
 
   O << '\t' << Str;
-  if (NeedsReg) {
-    O << ", ";
-    printRegName(O, MI->getOperand(4).getReg());
+
+  // For optional registers, don't print the value if it's xzr/x31
+  // since this defaults to xzr/x31 if register is not specified.
+  if (NeedsReg || (OptionalReg && NotXZR)) {
+    O << ", " << Reg;
   }
 
   return true;

llvm/lib/Target/AArch64/Utils/AArch64BaseInfo.h

@@ -409,6 +409,16 @@ struct SysAliasReg : SysAlias {
       : SysAlias(N, E, F), NeedsReg(R) {}
 };
 
+struct SysAliasOptionalReg : SysAlias {
+  bool NeedsReg;
+  bool OptionalReg;
+  constexpr SysAliasOptionalReg(const char *N, uint16_t E, bool R, bool O)
+      : SysAlias(N, E), NeedsReg(R), OptionalReg(O) {}
+  constexpr SysAliasOptionalReg(const char *N, uint16_t E, bool R, bool O,
+                                FeatureBitset F)
+      : SysAlias(N, E, F), NeedsReg(R), OptionalReg(O) {}
+};
+
 struct SysAliasImm : SysAlias {
   uint16_t ImmValue;
   constexpr SysAliasImm(const char *N, uint16_t E, uint16_t I)