blob: f99d2babf8994c11de391a9e80359fb425f217db [file] [edit]
//===- Main.cpp - Top-Level TableGen implementation -----------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// TableGen is a tool which can be used to build up a description of something,
// then invoke one or more "tablegen backends" to emit information about the
// description in some predefined format. In practice, this is used by the LLVM
// code generators to automate generation of a code generator through a
// high-level description of the target.
//
//===----------------------------------------------------------------------===//
#include "llvm/TableGen/Main.h"
#include "TGLexer.h"
#include "TGParser.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/Twine.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorOr.h"
#include "llvm/Support/FileSystem.h"
#include "llvm/Support/MemoryBuffer.h"
#include "llvm/Support/Path.h"
#include "llvm/Support/SMLoc.h"
#include "llvm/Support/SourceMgr.h"
#include "llvm/Support/ToolOutputFile.h"
#include "llvm/Support/VirtualFileSystem.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/TableGen/Error.h"
#include "llvm/TableGen/Record.h"
#include "llvm/TableGen/TGTimer.h"
#include "llvm/TableGen/TableGenBackend.h"
#include <memory>
#include <string>
#include <system_error>
#include <utility>
using namespace llvm;
static cl::opt<std::string>
OutputFilename("o", cl::desc("Output filename"), cl::value_desc("filename"),
cl::init("-"));
static cl::opt<std::string>
DependFilename("d",
cl::desc("Dependency filename"),
cl::value_desc("filename"),
cl::init(""));
static cl::opt<std::string>
InputFilename(cl::Positional, cl::desc("<input file>"), cl::init("-"));
static cl::list<std::string>
IncludeDirs("I", cl::desc("Directory of include files"),
cl::value_desc("directory"), cl::Prefix);
static cl::list<std::string>
MacroNames("D", cl::desc("Name of the macro to be defined"),
cl::value_desc("macro name"), cl::Prefix);
static cl::opt<bool>
WriteIfChanged("write-if-changed", cl::desc("Only write output if it changed"));
static cl::opt<bool>
TimePhases("time-phases", cl::desc("Time phases of parser and backend"));
cl::opt<bool> llvm::EmitLongStrLiterals(
"long-string-literals",
cl::desc("when emitting large string tables, prefer string literals over "
"comma-separated char literals. This can be a readability and "
"compile-time performance win, but upsets some compilers"),
cl::Hidden, cl::init(true));
static cl::opt<bool> NoWarnOnUnusedTemplateArgs(
"no-warn-on-unused-template-args",
cl::desc("Disable unused template argument warnings."));
static cl::opt<bool> Preprocess("E", cl::desc("Write preprocessed output"));
static int reportError(const char *ProgName, Twine Msg) {
errs() << ProgName << ": " << Msg;
errs().flush();
return 1;
}
namespace {
/// Encapsulate file, line and column numbers from SourceMgr.
struct SMCoords {
unsigned Buf = 0;
unsigned Line = 0;
unsigned Col = 0;
SMCoords() = default;
SMCoords(const SourceMgr &Mgr, SMLoc Loc) {
Buf = Mgr.FindBufferContainingLoc(Loc);
// TODO: Find a faster way to do this. For example we don't need the column
// number for every token, only the first token on each output line.
std::tie(Line, Col) = Mgr.getLineAndColumn(Loc, Buf);
}
};
} // end anonymous namespace
/// Create preprocessed output for `-E` option.
static int preprocessInput(TGLexer &Lex, raw_ostream &OS) {
SMCoords Last;
while (true) {
Lex.Lex();
if (Lex.getCode() == tgtok::Eof || Lex.getCode() == tgtok::Error)
break;
SMCoords This(SrcMgr, Lex.getLoc());
if (This.Buf == Last.Buf && This.Line == Last.Line) {
// Add a single space between tokens on the same line. This is overkill in
// many cases but at least it will parse correctly.
OS << ' ';
} else {
// Always start a new line when including a new file or popping back out
// to the previous file. This is just a heuristic to make the output look
// reasonably pretty.
if (Last.Buf)
OS << '\n';
// Indent the first token on a line to its original indentation, to make
// the output look pretty.
OS.indent(This.Col - 1);
}
const char *Start = Lex.getLoc().getPointer();
const char *End = Lex.getLocRange().End.getPointer();
StringRef Text(Start, End - Start);
OS << Text;
Last = This;
// A code block is the only kind of token that can span multiple lines.
// Record the line it ends on, so that a following token on that line stays
// on the same output line. Count '\n' characters to match how SourceMgr
// counts lines.
if (Lex.getCode() == tgtok::CodeFragment)
Last.Line += Text.count('\n');
}
if (Last.Buf != 0)
OS << '\n';
return Lex.getCode() == tgtok::Error;
}
/// Create a dependency file for `-d` option.
///
/// This functionality is really only for the benefit of the build system.
/// It is similar to GCC's `-M*` family of options.
static int createDependencyFile(const TGLexer::DependenciesSetTy &Dependencies,
const char *argv0) {
if (OutputFilename == "-")
return reportError(argv0, "the option -d must be used together with -o\n");
std::error_code EC;
ToolOutputFile DepOut(DependFilename, EC, sys::fs::OF_Text);
if (EC)
return reportError(argv0, "error opening " + DependFilename + ":" +
EC.message() + "\n");
DepOut.os() << OutputFilename << ":";
// Emit the primary input file as a dependency. This matches C compilers like
// Clang and GCC. Without it, a .td file with no `include` directives would
// produce a depfile listing zero dependencies. CMake's
// `cmake_transform_depfile` then collapses that to a 0-byte file, which Ninja
// treats as a missing depfile and re-runs the rule on every incremental
// build.
if (InputFilename != "-")
DepOut.os() << ' ' << InputFilename;
for (const auto &Dep : Dependencies) {
DepOut.os() << ' ' << Dep;
}
DepOut.os() << "\n";
DepOut.keep();
return 0;
}
static int WriteOutput(const char *argv0, StringRef Filename,
StringRef Content) {
if (WriteIfChanged) {
// Only updates the real output file if there are any differences.
// This prevents recompilation of all the files depending on it if there
// aren't any.
if (auto ExistingOrErr = MemoryBuffer::getFile(Filename, /*IsText=*/true))
if (std::move(ExistingOrErr.get())->getBuffer() == Content)
return 0;
}
std::error_code EC;
// On z/OS, text files default to IBM-1047, but generated files are UTF-8.
// OF_ForceUTF8 has no effect on other platforms.
ToolOutputFile OutFile(Filename, EC,
sys::fs::OF_Text | sys::fs::OF_ForceUTF8);
if (EC)
return reportError(argv0, "error opening " + Filename + ": " +
EC.message() + "\n");
OutFile.os() << Content;
if (ErrorsPrinted == 0)
OutFile.keep();
return 0;
}
int llvm::TableGenMain(const char *argv0, MultiFileTableGenMainFn MainFn) {
ErrorOr<std::unique_ptr<MemoryBuffer>> FileOrErr =
MemoryBuffer::getFileOrSTDIN(InputFilename, /*IsText=*/true);
if (std::error_code EC = FileOrErr.getError())
return reportError(argv0, "Could not open input file '" + InputFilename +
"': " + EC.message() + "\n");
// Tell SrcMgr about this buffer, which is what TGParser will pick up.
SrcMgr.AddNewSourceBuffer(std::move(*FileOrErr), SMLoc());
// Record the location of the include directory so that the lexer can find
// it later.
SrcMgr.setIncludeDirs(IncludeDirs);
SrcMgr.setVirtualFileSystem(vfs::getRealFileSystem());
if (Preprocess) {
TGLexer Lex(SrcMgr, MacroNames);
std::string OutString;
raw_string_ostream Out(OutString);
if (int Ret = preprocessInput(Lex, Out))
return Ret;
// Return early if the lexer printed errors but still reached EOF (e.g., a
// NUL character in the source).
if (ErrorsPrinted > 0)
return reportError(argv0, Twine(ErrorsPrinted) + " errors.\n");
if (!DependFilename.empty()) {
if (int Ret = createDependencyFile(Lex.getDependencies(), argv0))
return Ret;
}
return WriteOutput(argv0, OutputFilename, OutString);
}
RecordKeeper Records;
Records.saveInputFilename(InputFilename);
TGTimer &Timer = Records.getTimer();
if (TimePhases)
Timer.startPhaseTiming();
// Parse the input file.
Timer.startTimer("Parse, build records");
TGParser Parser(SrcMgr, MacroNames, Records, NoWarnOnUnusedTemplateArgs);
if (Parser.ParseFile())
return 1;
Timer.stopTimer();
// Return early if any other errors were generated during parsing
// (e.g., assert failures).
if (ErrorsPrinted > 0)
return reportError(argv0, Twine(ErrorsPrinted) + " errors.\n");
// Write output to memory.
Timer.startBackendTimer("Backend overall");
TableGenOutputFiles OutFiles;
unsigned status = 0;
// ApplyCallback will return true if it did not apply any callback. In that
// case, attempt to apply the MainFn.
StringRef FilenamePrefix(sys::path::stem(OutputFilename));
if (TableGen::Emitter::ApplyCallback(Records, OutFiles, FilenamePrefix))
status = MainFn ? MainFn(OutFiles, Records) : 1;
Timer.stopBackendTimer();
if (status)
return 1;
// Always write the depfile, even if the main output hasn't changed.
// If it's missing, Ninja considers the output dirty. If this was below
// the early exit below and someone deleted the .inc.d file but not the .inc
// file, tablegen would never write the depfile.
if (!DependFilename.empty()) {
if (int Ret = createDependencyFile(Parser.getDependencies(), argv0))
return Ret;
}
Timer.startTimer("Write output");
if (int Ret = WriteOutput(argv0, OutputFilename, OutFiles.MainFile))
return Ret;
for (auto [Suffix, Content] : OutFiles.AdditionalFiles) {
SmallString<128> Filename(OutputFilename);
// TODO: Format using the split-file convention when writing to stdout?
if (Filename != "-") {
sys::path::replace_extension(Filename, "");
Filename.append(Suffix);
}
if (int Ret = WriteOutput(argv0, Filename, Content))
return Ret;
}
Timer.stopTimer();
Timer.stopPhaseTiming();
if (ErrorsPrinted > 0)
return reportError(argv0, Twine(ErrorsPrinted) + " errors.\n");
return 0;
}
int llvm::TableGenMain(const char *argv0, TableGenMainFn MainFn) {
return TableGenMain(argv0, [&MainFn](TableGenOutputFiles &OutFiles,
const RecordKeeper &Records) {
std::string S;
raw_string_ostream OS(S);
int Res = MainFn(OS, Records);
OutFiles = {std::move(S), {}};
return Res;
});
}