[OptTable] Make ValuesCode initialisation of Options constexpr

[serge-sans-paille]

Summary

Current implementation requires a copy of the initialization array to a
vector to be able to modify their Values field.

This is inefficient: it requires a large copy to update a value, while
TableGen has all information to avoid this overwrite.

Modify TableGen to emit the Values code and use it to perform the
initialisation.

The impact on performance is not amazing compared to the actual
compilation, but still noticeable:

https://llvm-compile-time-tracker.com/compare.php?from=d9ab3e82f30d646deff054230b0c742704a1cf26&to=f2b37fb65d5149f70b43d1801beb5239285a2a20&stat=instructions:u

[serge-sans-paille]

(rebased on main)

[serge-sans-paille]

@nikic: gentle ping

[nikic]

llvm/utils/TableGen/OptParserEmitter.cpp

268

Isn't this assert the wrong way around?

[serge-sans-paille]

nits + rebased on main

[nikic]

LGTM

clang/include/clang/Driver/Options.td

def start_no_unused_arguments : Flag<["--"], "start-no-unused-arguments">, Flags<[CoreOption]>,
   HelpText<"Don't emit warnings about unused arguments for the following arguments">;
 def static_libgcc : Flag<["-"], "static-libgcc">;
 def static_libstdcxx : Flag<["-"], "static-libstdc++">;
 def static : Flag<["-", "--"], "static">, Group<Link_Group>, Flags<[NoArgumentUnused]>;
 def std_default_EQ : Joined<["-"], "std-default=">;
 def std_EQ : Joined<["-", "--"], "std=">, Flags<[CC1Option,FlangOption,FC1Option]>,
   Group<CompileOnly_Group>, HelpText<"Language standard to compile for">,
   ValuesCode<[{
-    const char *Values =
+    static constexpr const char VALUES_CODE [] =
     #define LANGSTANDARD(id, name, lang, desc, features) name ","
     #define LANGSTANDARD_ALIAS(id, alias) alias ","
     #include "clang/Basic/LangStandards.def"
     ;
   }]>;
 def stdlib_EQ : Joined<["-", "--"], "stdlib=">, Flags<[CC1Option]>,
   HelpText<"C++ standard library to use">, Values<"libc++,libstdc++,platform">;
 def stdlibxx_isystem : JoinedOrSeparate<["-"], "stdlib++-isystem">,
def analyzer_max_loop : Separate<["-"], "analyzer-max-loop">,
   MarshallingInfoInt<AnalyzerOpts<"maxBlockVisitOnPath">, "4">;
 def analyzer_stats : Flag<["-"], "analyzer-stats">,
   HelpText<"Print internal analyzer statistics.">,
   MarshallingInfoFlag<AnalyzerOpts<"PrintStats">>;
 
 def analyzer_checker : Separate<["-"], "analyzer-checker">,
   HelpText<"Choose analyzer checkers to enable">,
   ValuesCode<[{
-    const char *Values =
+    static constexpr const char VALUES_CODE [] =
     #define GET_CHECKERS
     #define CHECKER(FULLNAME, CLASS, HT, DOC_URI, IS_HIDDEN)  FULLNAME ","
     #include "clang/StaticAnalyzer/Checkers/Checkers.inc"
     #undef GET_CHECKERS
     #define GET_PACKAGES
     #define PACKAGE(FULLNAME)  FULLNAME ","
     #include "clang/StaticAnalyzer/Checkers/Checkers.inc"
     #undef GET_PACKAGES

clang/lib/Driver/DriverOptions.cpp

 #include "llvm/Option/OptTable.h"
 #include "llvm/Option/Option.h"
 #include <cassert>
 
 using namespace clang::driver;
 using namespace clang::driver::options;
 using namespace llvm::opt;
 
+#define OPTTABLE_VALUES_CODE
+#include "clang/Driver/Options.inc"
+#undef OPTTABLE_VALUES_CODE
+
 #define PREFIX(NAME, VALUE)                                                    \
   static constexpr llvm::StringLiteral NAME##_init[] = VALUE;                  \
   static constexpr llvm::ArrayRef<llvm::StringLiteral> NAME(                   \
       NAME##_init, std::size(NAME##_init) - 1);
 #include "clang/Driver/Options.inc"
 #undef PREFIX
 
 static constexpr OptTable::Info InfoTable[] = {

 public:
   DriverOptTable()
     : OptTable(InfoTable) {}
 };
 
 }
 
 const llvm::opt::OptTable &clang::driver::getDriverOptTable() {
-  static const DriverOptTable *Table = []() {
-    auto Result = std::make_unique<DriverOptTable>();
-    // Options.inc is included in DriverOptions.cpp, and calls OptTable's
-    // addValues function.
-    // Opt is a variable used in the code fragment in Options.inc.
-    OptTable &Opt = *Result;
-#define OPTTABLE_ARG_INIT
-#include "clang/Driver/Options.inc"
-#undef OPTTABLE_ARG_INIT
-    return Result.release();
-  }();
-  return *Table;
+  static DriverOptTable Table;
+  return Table;
 }

llvm/include/llvm/Option/OptTable.h

struct Info {
     unsigned short GroupID;
     unsigned short AliasID;
     const char *AliasArgs;
     const char *Values;
   };
 
 private:
   /// The option information table.
-  std::vector<Info> OptionInfos;
+  ArrayRef<Info> OptionInfos;
   bool IgnoreCase;
   bool GroupedShortOptions = false;
   const char *EnvVar = nullptr;
 
   unsigned InputOptionID = 0;
   unsigned UnknownOptionID = 0;
 
   /// The index of the first option which can be parsed (i.e., is not a
public:
   /// For example, a minimum length of 3 prevents "-x" from being considered
   /// near to "-S".
   ///
   /// \return The edit distance of the nearest string found.
   unsigned findNearest(StringRef Option, std::string &NearestString,
                        unsigned FlagsToInclude = 0, unsigned FlagsToExclude = 0,
                        unsigned MinimumLength = 4) const;
 
-  /// Add Values to Option's Values class
-  ///
-  /// \param [in] Option - Prefix + Name of the flag which Values will be
-  ///  changed. For example, "-analyzer-checker".
-  /// \param [in] Values - String of Values seperated by ",", such as
-  ///  "foo, bar..", where foo and bar is the argument which the Option flag
-  ///  takes
-  ///
-  /// \return true in success, and false in fail.
-  bool addValues(StringRef Option, const char *Values);
-
   /// Parse a single argument; returning the new argument and
   /// updating Index.
   ///
   /// \param [in,out] Index - The current parsing position in the argument
   /// string list; on return this will be the index of the next argument
   /// string to parse.
   /// \param [in] FlagsToInclude - Only parse options with any of these flags.
   /// Zero is the default which includes all flags.

llvm/lib/Option/OptTable.cpp

for (auto CandidatePrefix : CandidateInfo.Prefixes) {
         BestDistance = Distance;
         NearestString = (Candidate + RHS).str();
       }
     }
   }
   return BestDistance;
 }
 
-bool OptTable::addValues(StringRef Option, const char *Values) {
-  for (size_t I = FirstSearchableIndex, E = OptionInfos.size(); I < E; I++) {
-    Info &In = OptionInfos[I];
-    if (optionMatches(In, Option)) {
-      In.Values = Values;
-      return true;
-    }
-  }
-  return false;
-}
-
 // Parse a single argument, return the new argument, and update Index. If
 // GroupedShortOptions is true, -a matches "-abc" and the argument in Args will
 // be updated to "-bc". This overload does not support
 // FlagsToInclude/FlagsToExclude or case insensitive options.
 std::unique_ptr<Arg> OptTable::parseOneArgGrouped(InputArgList &Args,
                                                   unsigned &Index) const {
   // Anything that doesn't start with PrefixesUnion is an input, as is '-'
   // itself.

llvm/utils/TableGen/OptParserEmitter.cpp

for (const auto &PrefixKey : Prefix.first)
       OS << "llvm::StringLiteral(\"" << PrefixKey << "\") COMMA ";
     // Append an empty element to avoid ending up with an empty array.
     OS << "llvm::StringLiteral(\"\")})\n";
   }
   OS << "#undef COMMA\n";
   OS << "#endif // PREFIX\n\n";
 
   OS << "/////////\n";
+  OS << "// ValuesCode\n\n";
+  OS << "#ifdef OPTTABLE_VALUES_CODE\n";
+  for (const Record &R : llvm::make_pointee_range(Opts)) {
+    // The option values, if any;
+    if (!isa<UnsetInit>(R.getValueInit("ValuesCode"))) {
+      assert(isa<UnsetInit>(R.getValueInit("Values")) &&
+             "Cannot choose between Values and ValuesCode");
+      OS << "#define VALUES_CODE " << getOptionName(R) << "_Values\n";
+      OS << R.getValueAsString("ValuesCode") << "\n";
+      OS << "#undef VALUES_CODE\n";
+    }
+  }
+  OS << "#endif\n";
+
+  OS << "/////////\n";
   OS << "// Groups\n\n";
   OS << "#ifdef OPTION\n";
   for (const Record &R : llvm::make_pointee_range(Groups)) {
     // Start a single option entry.
     OS << "OPTION(";
 
     // The option prefix;
     OS << "llvm::ArrayRef<llvm::StringLiteral>()";
if (!isa<UnsetInit>(R.getValueInit("MetaVarName")))
       write_cstring(OS, R.getValueAsString("MetaVarName"));
     else
       OS << "nullptr";
 
     // The option Values. Used for shell autocompletion.
     OS << ", ";
     if (!isa<UnsetInit>(R.getValueInit("Values")))
       write_cstring(OS, R.getValueAsString("Values"));
+    else if (!isa<UnsetInit>(R.getValueInit("ValuesCode"))) {
+      OS << getOptionName(R) << "_Values";
+    }
     else
       OS << "nullptr";
   };
 
   auto IsMarshallingOption = [](const Record &R) {
     return !isa<UnsetInit>(R.getValueInit("KeyPath")) &&
            !R.getValueAsString("KeyPath").empty();
   };
void EmitOptParser(RecordKeeper &Records, raw_ostream &OS) {
   OS << "};\n";
   OS << "static const unsigned SimpleEnumValueTablesSize = "
         "std::size(SimpleEnumValueTables);\n";
 
   OS << "#endif // SIMPLE_ENUM_VALUE_TABLE\n";
   OS << "\n";
 
   OS << "\n";
-  OS << "#ifdef OPTTABLE_ARG_INIT\n";
-  OS << "//////////\n";
-  OS << "// Option Values\n\n";
-  for (const Record &R : llvm::make_pointee_range(Opts)) {
-    if (isa<UnsetInit>(R.getValueInit("ValuesCode")))
-      continue;
-    OS << "{\n";
-    OS << "bool ValuesWereAdded;\n";
-    OS << R.getValueAsString("ValuesCode");
-    OS << "\n";
-    for (StringRef Prefix : R.getValueAsListOfStrings("Prefixes")) {
-      OS << "ValuesWereAdded = Opt.addValues(";
-      std::string S(Prefix);
-      S += R.getValueAsString("Name");
-      write_cstring(OS, S);
-      OS << ", Values);\n";
-      OS << "(void)ValuesWereAdded;\n";
-      OS << "assert(ValuesWereAdded && \"Couldn't add values to "
-            "OptTable!\");\n";
-    }
-    OS << "}\n";
-  }
-  OS << "\n";
-  OS << "#endif // OPTTABLE_ARG_INIT\n";
 }
 } // end namespace llvm