[TableGen] Give the option of tolerating duplicate register names

A number of architectures re-use the same register names (e.g. for both 32-bit 
FPRs and 64-bit FPRs). They are currently unable to use the tablegen'erated 
MatchRegisterName and MatchRegisterAltName, as tablegen (when built with 
asserts enabled) will fail.

When the AllowDuplicateRegisterNames in AsmParser is set, duplicated register 
names will be tolerated. A backend can then coerce registers to the desired 
register class by (for instance) implementing validateTargetOperandClass.

At least the in-tree Sparc backend could benefit from this, as does RISC-V 
(single and double precision floating point registers).

Differential Revision: https://reviews.llvm.org/D39845

llvm-svn: 320018

llvm/include/llvm/TableGen/StringMatcher.h

@@ -43,11 +43,12 @@ class StringMatcher {
                 const std::vector<StringPair> &matches, raw_ostream &os)
     : StrVariableName(strVariableName), Matches(matches), OS(os) {}
 
-  void Emit(unsigned Indent = 0) const;
+  void Emit(unsigned Indent = 0, bool IgnoreDuplicates = false) const;
 
 private:
-  bool EmitStringMatcherForChar(const std::vector<const StringPair*> &Matches,
-                                unsigned CharNo, unsigned IndentCount) const;
+  bool EmitStringMatcherForChar(const std::vector<const StringPair *> &Matches,
+                                unsigned CharNo, unsigned IndentCount,
+                                bool IgnoreDuplicates) const;
 };
 
 } // end namespace llvm

llvm/include/llvm/Target/Target.td

@@ -1174,6 +1174,14 @@ class AsmParser {
   // several registers share the same alias (i.e. not a 1:1 mapping).
   bit ShouldEmitMatchRegisterAltName = 0;
 
+  // Set to true if MatchRegisterName and MatchRegisterAltName functions
+  // should be generated even if there are duplicate register names. The
+  // target is responsible for coercing aliased registers as necessary
+  // (e.g. in validateTargetOperandClass), and there are no guarantees about
+  // which numeric register identifier will be returned in the case of
+  // multiple matches.
+  bit AllowDuplicateRegisterNames = 0;
+
   // HasMnemonicFirst - Set to false if target instructions don't always
   // start with a mnemonic as the first token.
   bit HasMnemonicFirst = 1;

llvm/lib/TableGen/StringMatcher.cpp

@@ -46,17 +46,18 @@ FindFirstNonCommonLetter(const std::vector<const
 /// code to verify that CharNo and later are the same.
 ///
 /// \return - True if control can leave the emitted code fragment.
-bool StringMatcher::
-EmitStringMatcherForChar(const std::vector<const StringPair*> &Matches,
-                         unsigned CharNo, unsigned IndentCount) const {
+bool StringMatcher::EmitStringMatcherForChar(
+    const std::vector<const StringPair *> &Matches, unsigned CharNo,
+    unsigned IndentCount, bool IgnoreDuplicates) const {
   assert(!Matches.empty() && "Must have at least one string to match!");
-  std::string Indent(IndentCount*2+4, ' ');
+  std::string Indent(IndentCount * 2 + 4, ' ');
 
   // If we have verified that the entire string matches, we're done: output the
   // matching code.
   if (CharNo == Matches[0]->first.size()) {
-    assert(Matches.size() == 1 && "Had duplicate keys to match on");
-
+    if (Matches.size() > 1 && !IgnoreDuplicates)
+      report_fatal_error("Had duplicate keys to match on");
+
     // If the to-execute code has \n's in it, indent each subsequent line.
     StringRef Code = Matches[0]->second;
 
@@ -100,8 +101,9 @@ EmitStringMatcherForChar(const std::vector<const StringPair*> &Matches,
          << NumChars << ") != 0)\n";
       OS << Indent << "  break;\n";
     }
-
-    return EmitStringMatcherForChar(Matches, FirstNonCommonLetter, IndentCount);
+
+    return EmitStringMatcherForChar(Matches, FirstNonCommonLetter, IndentCount,
+                                    IgnoreDuplicates);
   }
 
   // Otherwise, we have multiple possible things, emit a switch on the
@@ -116,7 +118,8 @@ EmitStringMatcherForChar(const std::vector<const StringPair*> &Matches,
        << LI->second.size() << " string";
     if (LI->second.size() != 1) OS << 's';
     OS << " to match.\n";
-    if (EmitStringMatcherForChar(LI->second, CharNo+1, IndentCount+1))
+    if (EmitStringMatcherForChar(LI->second, CharNo + 1, IndentCount + 1,
+                                 IgnoreDuplicates))
       OS << Indent << "  break;\n";
   }
 
@@ -126,7 +129,7 @@ EmitStringMatcherForChar(const std::vector<const StringPair*> &Matches,
 
 /// Emit - Top level entry point.
 ///
-void StringMatcher::Emit(unsigned Indent) const {
+void StringMatcher::Emit(unsigned Indent, bool IgnoreDuplicates) const {
   // If nothing to match, just fall through.
   if (Matches.empty()) return;
 
@@ -146,7 +149,7 @@ void StringMatcher::Emit(unsigned Indent) const {
     OS.indent(Indent*2+2) << "case " << LI->first << ":\t // "
        << LI->second.size()
        << " string" << (LI->second.size() == 1 ? "" : "s") << " to match.\n";
-    if (EmitStringMatcherForChar(LI->second, 0, Indent))
+    if (EmitStringMatcherForChar(LI->second, 0, Indent, IgnoreDuplicates))
       OS.indent(Indent*2+4) << "break;\n";
   }
 

llvm/test/TableGen/AllowDuplicateRegisterNames.td

@@ -0,0 +1,86 @@
+// RUN: llvm-tblgen -gen-asm-matcher -I %p/../../include %s | FileCheck %s
+
+// Check that MatchRegisterName and MatchRegisterAltName are generated
+// correctly when multiple registers are defined with the same name and
+// AllowDuplicateRegisterNames is set.
+
+include "llvm/Target/Target.td"
+
+def ArchInstrInfo : InstrInfo;
+
+def ArchAsmParser : AsmParser {
+  let AllowDuplicateRegisterNames = 1;
+  let ShouldEmitMatchRegisterAltName = 1;
+}
+
+def Arch : Target {
+  let InstructionSet = ArchInstrInfo;
+  let AssemblyParsers = [ArchAsmParser];
+}
+
+let Namespace = "Arch" in {
+class ArchReg<string n, list <string> alt, list <RegAltNameIndex> altidx>
+    : Register<n> {
+  let AltNames = alt;
+  let RegAltNameIndices = altidx;
+}
+
+def ABIRegAltName : RegAltNameIndex;
+
+foreach i = 0-3 in {
+  def R#i#_32 : ArchReg<"r"#i, ["x"#i], [ABIRegAltName]>;
+  def R#i#_64 : ArchReg<"r"#i, ["x"#i], [ABIRegAltName]>;
+}
+} // Namespace = "Arch"
+
+def GPR32 : RegisterClass<"Arch", [i32], 32, (add
+    (sequence "R%u_32", 0, 3)
+)>;
+
+def GPR64 : RegisterClass<"Arch", [i64], 64, (add
+    (sequence "R%u_64", 0, 3)
+)>;
+
+// CHECK: static unsigned MatchRegisterName(StringRef Name) {
+// CHECK:   switch (Name.size()) {
+// CHECK:   default: break;
+// CHECK:   case 2:  // 8 strings to match.
+// CHECK:     if (Name[0] != 'r')
+// CHECK:       break;
+// CHECK:     switch (Name[1]) {
+// CHECK:     default: break;
+// CHECK:     case '0':  // 2 strings to match.
+// CHECK:       return 1;  // "r0"
+// CHECK:     case '1':  // 2 strings to match.
+// CHECK:       return 3;  // "r1"
+// CHECK:     case '2':  // 2 strings to match.
+// CHECK:       return 5;  // "r2"
+// CHECK:     case '3':  // 2 strings to match.
+// CHECK:       return 7;  // "r3"
+// CHECK:     }
+// CHECK:     break;
+// CHECK:   }
+// CHECK:   return 0;
+// CHECK: }
+
+// CHECK: static unsigned MatchRegisterAltName(StringRef Name) {
+// CHECK:   switch (Name.size()) {
+// CHECK:   default: break;
+// CHECK:   case 2:  // 8 strings to match.
+// CHECK:     if (Name[0] != 'x')
+// CHECK:       break;
+// CHECK:     switch (Name[1]) {
+// CHECK:     default: break;
+// CHECK:     case '0':  // 2 strings to match.
+// CHECK:       return 1;  // "x0"
+// CHECK:     case '1':  // 2 strings to match.
+// CHECK:       return 3;  // "x1"
+// CHECK:     case '2':  // 2 strings to match.
+// CHECK:       return 5;  // "x2"
+// CHECK:     case '3':  // 2 strings to match.
+// CHECK:       return 7;  // "x3"
+// CHECK:     }
+// CHECK:     break;
+// CHECK:   }
+// CHECK:   return 0;
+// CHECK: }

llvm/utils/TableGen/AsmMatcherEmitter.cpp

@@ -2438,7 +2438,9 @@ static void emitMatchRegisterName(CodeGenTarget &Target, Record *AsmParser,
 
   OS << "static unsigned MatchRegisterName(StringRef Name) {\n";
 
-  StringMatcher("Name", Matches, OS).Emit();
+  bool IgnoreDuplicates =
+      AsmParser->getValueAsBit("AllowDuplicateRegisterNames");
+  StringMatcher("Name", Matches, OS).Emit(0, IgnoreDuplicates);
 
   OS << "  return 0;\n";
   OS << "}\n\n";
@@ -2469,7 +2471,9 @@ static void emitMatchRegisterAltName(CodeGenTarget &Target, Record *AsmParser,
 
   OS << "static unsigned MatchRegisterAltName(StringRef Name) {\n";
 
-  StringMatcher("Name", Matches, OS).Emit();
+  bool IgnoreDuplicates =
+      AsmParser->getValueAsBit("AllowDuplicateRegisterNames");
+  StringMatcher("Name", Matches, OS).Emit(0, IgnoreDuplicates);
 
   OS << "  return 0;\n";
   OS << "}\n\n";