[RISCV] Add Custom Parser for Atomic Memory Operands

Summary:
GCC Accepts both (reg) and 0(reg) for atomic instruction memory
operands. These instructions do not allow for an offset in their
encoding, so in the latter case, the 0 is silently dropped.

Due to how we have structured the RISCVAsmParser, the easiest way to add
support for parsing this offset is to add a custom AsmOperand and
parser. This parser drops all the parens, and just keeps the register.

This commit also adds a custom printer for these operands, which matches
the GCC canonical printer, printing both `(a0)` and `0(a0)` as `(a0)`.

Reviewers: asb, lewis-revill

Reviewed By: asb

Subscribers: s.egerton, hiraditya, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, kito-cheng, shiva0217, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, rkruppe, jfb, PkmX, jocewei, psnobl, benna, Jim, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65205

llvm-svn: 367553
diff --git a/llvm/lib/Target/RISCV/AsmParser/RISCVAsmParser.cpp b/llvm/lib/Target/RISCV/AsmParser/RISCVAsmParser.cpp
index 1d166c8..93a2dd3 100644
--- a/llvm/lib/Target/RISCV/AsmParser/RISCVAsmParser.cpp
+++ b/llvm/lib/Target/RISCV/AsmParser/RISCVAsmParser.cpp
@@ -127,6 +127,7 @@
   OperandMatchResultTy parseRegister(OperandVector &Operands,
                                      bool AllowParens = false);
   OperandMatchResultTy parseMemOpBaseReg(OperandVector &Operands);
+  OperandMatchResultTy parseAtomicMemOp(OperandVector &Operands);
   OperandMatchResultTy parseOperandWithModifier(OperandVector &Operands);
   OperandMatchResultTy parseBareSymbol(OperandVector &Operands);
   OperandMatchResultTy parseCallSymbol(OperandVector &Operands);
@@ -575,6 +576,15 @@
 
   bool isSImm21Lsb0JAL() const { return isBareSimmNLsb0<21>(); }
 
+  bool isImmZero() const {
+    if (!isImm())
+      return false;
+    int64_t Imm;
+    RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
+    bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
+    return IsConstantImm && (Imm == 0) && VK == RISCVMCExpr::VK_RISCV_None;
+  }
+
   /// getStartLoc - Gets location of the first token of this operand
   SMLoc getStartLoc() const override { return StartLoc; }
   /// getEndLoc - Gets location of the last token of this operand
@@ -1282,6 +1292,73 @@
   return MatchOperand_Success;
 }
 
+OperandMatchResultTy RISCVAsmParser::parseAtomicMemOp(OperandVector &Operands) {
+  // Atomic operations such as lr.w, sc.w, and amo*.w accept a "memory operand"
+  // as one of their register operands, such as `(a0)`. This just denotes that
+  // the register (in this case `a0`) contains a memory address.
+  //
+  // Normally, we would be able to parse these by putting the parens into the
+  // instruction string. However, GNU as also accepts a zero-offset memory
+  // operand (such as `0(a0)`), and ignores the 0. Normally this would be parsed
+  // with parseImmediate followed by parseMemOpBaseReg, but these instructions
+  // do not accept an immediate operand, and we do not want to add a "dummy"
+  // operand that is silently dropped.
+  //
+  // Instead, we use this custom parser. This will: allow (and discard) an
+  // offset if it is zero; require (and discard) parentheses; and add only the
+  // parsed register operand to `Operands`.
+  //
+  // These operands are printed with RISCVInstPrinter::printAtomicMemOp, which
+  // will only print the register surrounded by parentheses (which GNU as also
+  // uses as its canonical representation for these operands).
+  std::unique_ptr<RISCVOperand> OptionalImmOp;
+
+  if (getLexer().isNot(AsmToken::LParen)) {
+    // Parse an Integer token. We do not accept arbritrary constant expressions
+    // in the offset field (because they may include parens, which complicates
+    // parsing a lot).
+    int64_t ImmVal;
+    SMLoc ImmStart = getLoc();
+    if (getParser().parseIntToken(ImmVal,
+                                  "expected '(' or optional integer offset"))
+      return MatchOperand_ParseFail;
+
+    // Create a RISCVOperand for checking later (so the error messages are
+    // nicer), but we don't add it to Operands.
+    SMLoc ImmEnd = getLoc();
+    OptionalImmOp =
+        RISCVOperand::createImm(MCConstantExpr::create(ImmVal, getContext()),
+                                ImmStart, ImmEnd, isRV64());
+  }
+
+  if (getLexer().isNot(AsmToken::LParen)) {
+    Error(getLoc(), OptionalImmOp ? "expected '(' after optional integer offset"
+                                  : "expected '(' or optional integer offset");
+    return MatchOperand_ParseFail;
+  }
+  getParser().Lex(); // Eat '('
+
+  if (parseRegister(Operands) != MatchOperand_Success) {
+    Error(getLoc(), "expected register");
+    return MatchOperand_ParseFail;
+  }
+
+  if (getLexer().isNot(AsmToken::RParen)) {
+    Error(getLoc(), "expected ')'");
+    return MatchOperand_ParseFail;
+  }
+  getParser().Lex(); // Eat ')'
+
+  // Deferred Handling of non-zero offsets. This makes the error messages nicer.
+  if (OptionalImmOp && !OptionalImmOp->isImmZero()) {
+    Error(OptionalImmOp->getStartLoc(), "optional integer offset must be 0",
+          SMRange(OptionalImmOp->getStartLoc(), OptionalImmOp->getEndLoc()));
+    return MatchOperand_ParseFail;
+  }
+
+  return MatchOperand_Success;
+}
+
 /// Looks at a token type and creates the relevant operand from this
 /// information, adding to Operands. If operand was parsed, returns false, else
 /// true.
diff --git a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVInstPrinter.cpp b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVInstPrinter.cpp
index fe37b70..d745220 100644
--- a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVInstPrinter.cpp
+++ b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVInstPrinter.cpp
@@ -112,3 +112,15 @@
       static_cast<RISCVFPRndMode::RoundingMode>(MI->getOperand(OpNo).getImm());
   O << RISCVFPRndMode::roundingModeToString(FRMArg);
 }
+
+void RISCVInstPrinter::printAtomicMemOp(const MCInst *MI, unsigned OpNo,
+                                        const MCSubtargetInfo &STI,
+                                        raw_ostream &O) {
+  const MCOperand &MO = MI->getOperand(OpNo);
+
+  assert(MO.isReg() && "printAtomicMemOp can only print register operands");
+  O << "(";
+  printRegName(O, MO.getReg());
+  O << ")";
+  return;
+}
diff --git a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVInstPrinter.h b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVInstPrinter.h
index 5ca1d3f..908b2de 100644
--- a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVInstPrinter.h
+++ b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVInstPrinter.h
@@ -37,6 +37,8 @@
                      const MCSubtargetInfo &STI, raw_ostream &O);
   void printFRMArg(const MCInst *MI, unsigned OpNo, const MCSubtargetInfo &STI,
                    raw_ostream &O);
+  void printAtomicMemOp(const MCInst *MI, unsigned OpNo,
+                        const MCSubtargetInfo &STI, raw_ostream &O);
 
   // Autogenerated by tblgen.
   void printInstruction(const MCInst *MI, const MCSubtargetInfo &STI,
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoA.td b/llvm/lib/Target/RISCV/RISCVInstrInfoA.td
index b768c934..1484ba9 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoA.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoA.td
@@ -12,14 +12,32 @@
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
+// Operand and SDNode transformation definitions.
+//===----------------------------------------------------------------------===//
+
+// A parse method for (${gpr}) or 0(${gpr}), where the 0 is be silently ignored.
+// Used for GNU as Compatibility.
+def AtomicMemOpOperand : AsmOperandClass {
+  let Name = "AtomicMemOpOperand";
+  let RenderMethod = "addRegOperands";
+  let PredicateMethod = "isReg";
+  let ParserMethod = "parseAtomicMemOp";
+}
+
+def GPRMemAtomic : RegisterOperand<GPR> {
+  let ParserMatchClass = AtomicMemOpOperand;
+  let PrintMethod = "printAtomicMemOp";
+}
+
+//===----------------------------------------------------------------------===//
 // Instruction class templates
 //===----------------------------------------------------------------------===//
 
 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in
 class LR_r<bit aq, bit rl, bits<3> funct3, string opcodestr>
     : RVInstRAtomic<0b00010, aq, rl, funct3, OPC_AMO,
-                    (outs GPR:$rd), (ins GPR:$rs1),
-                    opcodestr, "$rd, (${rs1})"> {
+                    (outs GPR:$rd), (ins GPRMemAtomic:$rs1),
+                    opcodestr, "$rd, $rs1"> {
   let rs2 = 0;
 }
 
@@ -33,8 +51,8 @@
 let hasSideEffects = 0, mayLoad = 1, mayStore = 1 in
 class AMO_rr<bits<5> funct5, bit aq, bit rl, bits<3> funct3, string opcodestr>
     : RVInstRAtomic<funct5, aq, rl, funct3, OPC_AMO,
-                    (outs GPR:$rd), (ins GPR:$rs1, GPR:$rs2),
-                    opcodestr, "$rd, $rs2, (${rs1})">;
+                    (outs GPR:$rd), (ins GPRMemAtomic:$rs1, GPR:$rs2),
+                    opcodestr, "$rd, $rs2, $rs1">;
 
 multiclass AMO_rr_aq_rl<bits<5> funct5, bits<3> funct3, string opcodestr> {
   def ""     : AMO_rr<funct5, 0, 0, funct3, opcodestr>;