tblgen: add preprocessor as a separate mode

This patch adds a preprocessor that can expand nested for-loops for
saving some copy-n-paste in *.td files.

The preprocessor is not yet integrated with TGParser, and so it has
no direct effect on *.td inputs.  However, you may preprocess an td
input (and only preprocess it).

To test the proprecessor, type:
  tblgen -E -o $@ $<



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141079 91177308-0d34-0410-b5e6-96231b3b80d8

lib/TableGen/CMakeLists.txt

@@ -9,6 +9,7 @@ add_llvm_library(LLVMTableGen
   TableGenBackend.cpp
   TGLexer.cpp
   TGParser.cpp
+  TGPreprocessor.cpp
   )
 
 add_llvm_library_dependencies(LLVMTableGen

lib/TableGen/Main.cpp

@@ -16,6 +16,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "TGParser.h"
+#include "TGPreprocessor.h"
 #include "llvm/ADT/OwningPtr.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/MemoryBuffer.h"
@@ -43,6 +44,12 @@ namespace {
   cl::list<std::string>
   IncludeDirs("I", cl::desc("Directory of include files"),
               cl::value_desc("directory"), cl::Prefix);
+
+  cl::opt<bool>
+  PreprocessOnly("E",
+                 cl::desc("Stop after the preprocessing stage; "
+                          "This is work in progress and has no effect yet"),
+                 cl::init(false));
 }
 
 namespace llvm {
@@ -67,6 +74,22 @@ int TableGenMain(char *argv0, TableGenAction &Action) {
     // it later.
     SrcMgr.setIncludeDirs(IncludeDirs);
 
+    // TODO(clchiou): Integrate preprocessor into TGParser
+    if (PreprocessOnly) {
+      std::string Error;
+      tool_output_file Out(OutputFilename.c_str(), Error);
+      if (!Error.empty()) {
+        errs() << argv0 << ": error opening " << OutputFilename
+          << ":" << Error << "\n";
+        return 1;
+      }
+      TGPreprocessor Preprocessor(SrcMgr, Out);
+      if (Preprocessor.PreprocessFile())
+        return 1;
+      Out.keep();
+      return 0;
+    }
+
     TGParser Parser(SrcMgr, Records);
 
     if (Parser.ParseFile())

lib/TableGen/TGPreprocessor.cpp

@@ -0,0 +1,603 @@
+//===- TGPreprocessor.cpp - Preprocessor for TableGen ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Implement the Preprocessor for TableGen.
+//
+//===----------------------------------------------------------------------===//
+
+#include "TGPreprocessor.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/SourceMgr.h"
+#include "llvm/Support/ToolOutputFile.h"
+#include "llvm/TableGen/Error.h"
+#include <map>
+#include <string>
+#include <cctype>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+
+namespace llvm {
+typedef std::map<std::string, std::string> TGPPEnvironment;
+
+enum TGPPTokenKind {
+  tgpptoken_symbol,
+  tgpptoken_literal,
+  tgpptoken_newline,
+  tgpptoken_error,
+  tgpptoken_end
+};
+
+enum TGPPRecordKind {
+  tgpprecord_for,
+  tgpprecord_variable,
+  tgpprecord_literal
+};
+
+enum TGPPRangeKind {
+  tgpprange_list,
+  tgpprange_sequence
+};
+
+bool MatchSymbol(TGPPTokenKind Kind,
+                 const char *BeginOfToken, const char *EndOfToken,
+                 char Symbol);
+
+bool MatchSymbol(TGPPTokenKind Kind,
+                 const char *BeginOfToken, const char *EndOfToken,
+                 const char *Symbol);
+
+bool MatchIdNum(TGPPTokenKind Kind,
+                const char *BeginOfToken, const char *EndOfToken);
+
+bool MatchIdentifier(TGPPTokenKind Kind,
+                     const char *BeginOfToken, const char *EndOfToken);
+
+bool MatchNumber(TGPPTokenKind Kind,
+                 const char *BeginOfToken, const char *EndOfToken,
+                 long int *Val);
+
+class TGPPLexer {
+  const MemoryBuffer *CurBuf;
+  const char *CurPtr;
+  bool IsInsideMacroStatement, WasEndOfLine;
+
+  bool IsEndOfBuffer(const char *Ptr) const {
+    return (!*Ptr && Ptr == CurBuf->getBufferEnd());
+  }
+
+  bool IsNewLine() {
+    if (*CurPtr == '\r' || *CurPtr == '\n') {
+      if ((CurPtr[1] == '\r' || CurPtr[1] == '\n') && CurPtr[0] != CurPtr[1])
+        ++CurPtr;
+      return true;
+    }
+    return false;
+  }
+
+  bool MatchPrefix(const char *Prefix, const char *Ptr) const {
+    while (*Ptr == ' ' || *Ptr == '\t')
+      ++Ptr;
+    return !strncmp(Prefix, Ptr, strlen(Prefix));
+  }
+public:
+  TGPPLexer(const SourceMgr &SM)
+    : CurBuf(SM.getMemoryBuffer(0)),
+      CurPtr(CurBuf->getBufferStart()),
+      IsInsideMacroStatement(false),
+      WasEndOfLine(true) {
+  }
+
+  TGPPTokenKind NextToken(const char **BeginOfToken, const char **EndOfToken);
+};
+
+// preprocessor records
+class TGPPRecord {
+  TGPPRecordKind Kind;
+
+  // tgpprecord_for
+  std::vector<std::string> IndexVars;
+  std::vector<TGPPRange> IndexRanges;
+  TGPPRecords LoopBody;
+
+  // tgpprecord_variable, tgpprecord_literal
+  std::string Str;
+
+  bool EvaluateFor(const TGPPEnvironment &Env, raw_fd_ostream &OS) const;
+
+  bool EvaluateVariable(const TGPPEnvironment &Env, raw_fd_ostream &OS) const {
+    TGPPEnvironment::const_iterator it_val = Env.find(Str);
+    if (it_val == Env.end()) {
+      PrintError("Var is not bound to any value: " + Str);
+      return true;
+    }
+    OS << it_val->second;
+    return false;
+  }
+
+  bool EvaluateLiteral(const TGPPEnvironment &Env, raw_fd_ostream &OS) const {
+    OS << Str;
+    return false;
+  }
+
+public:
+  TGPPRecord(TGPPRecordKind K) : Kind(K) {}
+  TGPPRecord(TGPPRecordKind K, const std::string &S) : Kind(K), Str(S) {}
+
+  TGPPRecords *GetLoopBody() { return &LoopBody; }
+
+  void AppendIndex(const std::string &V, const TGPPRange &R) {
+    IndexVars.push_back(V);
+    IndexRanges.push_back(R);
+  }
+
+  bool Evaluate(const TGPPEnvironment &Env, raw_fd_ostream &OS) const;
+};
+
+class TGPPRange {
+  TGPPRangeKind Kind;
+
+  // tgpprange_list
+  std::vector<std::string> Vals;
+
+  // tgpprange_sequence
+  long int From, To;
+
+public:
+  TGPPRange() : Kind(tgpprange_list) {}
+  TGPPRange(long int F, long int T)
+    : Kind(tgpprange_sequence), From(F), To(T) {}
+
+  size_t size() const {
+    if (Kind == tgpprange_list)
+      return Vals.size();
+    else
+      return To - From + 1;
+  }
+
+  std::string at(size_t i) const {
+    if (Kind == tgpprange_list)
+      return Vals.at(i);
+    else {
+      char buf[32];
+      snprintf(buf, sizeof(buf), "%ld", From + (long int)i);
+      return std::string(buf);
+    }
+  }
+
+  void push_back(const std::string &S) {
+    if (Kind == tgpprange_list)
+      Vals.push_back(S);
+  }
+};
+} // namespace llvm
+
+using namespace llvm;
+
+bool llvm::MatchSymbol(TGPPTokenKind Kind,
+                       const char *BeginOfToken, const char *EndOfToken,
+                       char Symbol) {
+  return Kind == tgpptoken_symbol &&
+         BeginOfToken + 1 == EndOfToken &&
+         Symbol == *BeginOfToken;
+}
+
+bool llvm::MatchSymbol(TGPPTokenKind Kind,
+                       const char *BeginOfToken, const char *EndOfToken,
+                       const char *Symbol) {
+  return Kind == tgpptoken_symbol &&
+         BeginOfToken + strlen(Symbol) == EndOfToken &&
+         !strncmp(Symbol, BeginOfToken, EndOfToken - BeginOfToken);
+}
+
+bool llvm::MatchIdNum(TGPPTokenKind Kind,
+                      const char *BeginOfToken, const char *EndOfToken) {
+  if (Kind != tgpptoken_symbol)
+    return false;
+  for (const char *i = BeginOfToken; i != EndOfToken; ++i)
+    if (*i != '_' && !isalnum(*i))
+      return false;
+  return true;
+}
+
+bool llvm::MatchIdentifier(TGPPTokenKind Kind,
+                           const char *BeginOfToken, const char *EndOfToken) {
+  if (Kind != tgpptoken_symbol)
+    return false;
+
+  const char *i = BeginOfToken;
+  if (*i != '_' && !isalpha(*i))
+    return false;
+  for (++i; i != EndOfToken; ++i)
+    if (*i != '_' && !isalnum(*i))
+      return false;
+
+  return true;
+}
+
+bool llvm::MatchNumber(TGPPTokenKind Kind,
+                       const char *BeginOfToken, const char *EndOfToken,
+                       long int *Val) {
+  if (Kind != tgpptoken_symbol)
+    return false;
+  char *e;
+  *Val = strtol(BeginOfToken, &e, 10);
+  return e == EndOfToken;
+}
+
+TGPPTokenKind TGPPLexer::
+NextToken(const char **BeginOfToken, const char **EndOfToken) {
+  bool IsBeginOfLine = WasEndOfLine;
+  WasEndOfLine = false;
+
+  if (IsEndOfBuffer(CurPtr))
+    return tgpptoken_end;
+
+  else if (IsInsideMacroStatement) {
+    while (*CurPtr == ' ' || *CurPtr == '\t') // trim space, if any
+      ++CurPtr;
+
+    const char *BeginOfSymbol = CurPtr;
+
+    if (IsNewLine()) {
+      ++CurPtr;
+      IsInsideMacroStatement = false;
+      WasEndOfLine = true;
+      return tgpptoken_newline;
+    }
+
+    else if (*CurPtr == '[' || *CurPtr == ']' ||
+             *CurPtr == '(' || *CurPtr == ')' ||
+             *CurPtr == ',' || *CurPtr == '=') {
+      *BeginOfToken = BeginOfSymbol;
+      *EndOfToken = ++CurPtr;
+      return tgpptoken_symbol;
+    }
+
+    else if (*CurPtr == '_' || isalpha(*CurPtr)) {
+      ++CurPtr;
+      while (*CurPtr == '_' || isalnum(*CurPtr))
+        ++CurPtr;
+      *BeginOfToken = BeginOfSymbol;
+      *EndOfToken = CurPtr;
+      return tgpptoken_symbol;
+    }
+
+    else if (*CurPtr == '+' || *CurPtr == '-' || isdigit(*CurPtr)) {
+      ++CurPtr;
+      while (isdigit(*CurPtr))
+        ++CurPtr;
+      *BeginOfToken = BeginOfSymbol;
+      *EndOfToken = CurPtr;
+      return tgpptoken_symbol;
+    }
+
+    else {
+      PrintError(BeginOfSymbol, "Unrecognizable token");
+      return tgpptoken_error;
+    }
+  }
+
+  else if (*CurPtr == '#') {
+    if (IsBeginOfLine &&
+        (MatchPrefix("for", CurPtr + 1) ||
+         MatchPrefix("end", CurPtr + 1))) {
+      ++CurPtr;
+      IsInsideMacroStatement = true;
+      return NextToken(BeginOfToken, EndOfToken);
+    }
+
+    // special token #"# is translate to literal "
+    else if (CurPtr[1] == '"' && CurPtr[2] == '#') {
+      *BeginOfToken = ++CurPtr;
+      *EndOfToken = ++CurPtr;
+      ++CurPtr;
+      return tgpptoken_literal;
+    }
+
+    else {
+      const char *BeginOfVar = ++CurPtr; // trim '#'
+      if (*CurPtr != '_' && !isalpha(*CurPtr)) {
+        PrintError(BeginOfVar, "Variable must start with [_A-Za-z]: ");
+        return tgpptoken_error;
+      }
+      while (*CurPtr == '_' || isalnum(*CurPtr))
+        ++CurPtr;
+      if (*CurPtr != '#') {
+        PrintError(BeginOfVar, "Variable must end with #");
+        return tgpptoken_error;
+      }
+      *BeginOfToken = BeginOfVar;
+      *EndOfToken = CurPtr++; // trim '#'
+      return tgpptoken_symbol;
+    }
+  }
+
+  const char *BeginOfLiteral = CurPtr;
+  int CCommentLevel = 0;
+  bool BCPLComment = false;
+  bool StringLiteral = false;
+  for (; !IsEndOfBuffer(CurPtr); ++CurPtr) {
+    if (CCommentLevel > 0) {
+      if (CurPtr[0] == '/' && CurPtr[1] == '*') {
+        ++CurPtr;
+        ++CCommentLevel;
+      } else if (CurPtr[0] == '*' && CurPtr[1] == '/') {
+        ++CurPtr;
+        --CCommentLevel;
+      } else if (IsNewLine())
+        WasEndOfLine = true;
+    }
+
+    else if (BCPLComment) {
+      if (IsNewLine()) {
+        WasEndOfLine = true;
+        BCPLComment = false;
+      }
+    }
+
+    else if (StringLiteral) {
+      // no string escape sequence in TableGen?
+      if (*CurPtr == '"')
+        StringLiteral = false;
+    }
+
+    else if (CurPtr[0] == '/' && CurPtr[1] == '*') {
+      ++CurPtr;
+      ++CCommentLevel;
+    }
+
+    else if (CurPtr[0] == '/' && CurPtr[1] == '/') {
+      ++CurPtr;
+      BCPLComment = true;
+    }
+
+    else if (*CurPtr == '"')
+      StringLiteral = true;
+
+    else if (IsNewLine()) {
+      ++CurPtr;
+      WasEndOfLine = true;
+      break;
+    }
+
+    else if (*CurPtr == '#')
+      break;
+  }
+
+  *BeginOfToken = BeginOfLiteral;
+  *EndOfToken = CurPtr;
+  return tgpptoken_literal;
+}
+
+bool TGPPRecord::
+EvaluateFor(const TGPPEnvironment &Env, raw_fd_ostream &OS) const {
+  std::vector<TGPPRange>::const_iterator ri, re;
+
+  // calculate the min size
+  ri = IndexRanges.begin();
+  re = IndexRanges.begin();
+  size_t n = ri->size();
+  for (; ri != re; ++ri) {
+    size_t m = ri->size();
+    if (m < n)
+      n = m;
+  }
+
+  for (size_t which_val = 0; which_val < n; ++which_val) {
+    // construct nested environment
+    TGPPEnvironment NestedEnv(Env);
+    std::vector<std::string>::const_iterator vi = IndexVars.begin();
+    for (ri = IndexRanges.begin(), re = IndexRanges.end();
+        ri != re; ++vi, ++ri) {
+      NestedEnv.insert(std::make_pair(*vi, ri->at(which_val)));
+    }
+    // evalute loop body
+    for (TGPPRecords::const_iterator i = LoopBody.begin(), e = LoopBody.end();
+        i != e; ++i)
+      if (i->Evaluate(NestedEnv, OS))
+        return true;
+  }
+
+  return false;
+}
+
+bool TGPPRecord::
+Evaluate(const TGPPEnvironment &Env, raw_fd_ostream &OS) const {
+  switch (Kind) {
+  case tgpprecord_for:
+    return EvaluateFor(Env, OS);
+  case tgpprecord_variable:
+    return EvaluateVariable(Env, OS);
+  case tgpprecord_literal:
+    return EvaluateLiteral(Env, OS);
+  default:
+    PrintError("Unknown kind of record: " + Kind);
+    return true;
+  }
+  return false;
+}
+
+bool TGPreprocessor::ParseBlock(bool TopLevel) {
+  TGPPTokenKind Kind;
+  const char *BeginOfToken, *EndOfToken;
+  while ((Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken)) !=
+         tgpptoken_end) {
+    std::string Symbol(BeginOfToken, EndOfToken);
+    switch (Kind) {
+    case tgpptoken_symbol:
+      if (Symbol == "for") {
+        if (ParseForLoop())
+          return true;
+      } else if (Symbol == "end") {
+        if (TopLevel) {
+          PrintError(BeginOfToken, "No block to end here");
+          return true;
+        }
+        if ((Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken)) !=
+            tgpptoken_newline) {
+          PrintError(BeginOfToken, "Tokens after #end");
+          return true;
+        }
+        return false;
+      } else if (Symbol == "NAME") {
+        // treat '#NAME#' as a literal
+        CurRecords->push_back(
+            TGPPRecord(tgpprecord_literal,
+                       std::string("#NAME#")));
+      } else {
+        CurRecords->push_back(
+            TGPPRecord(tgpprecord_variable,
+                       std::string(BeginOfToken, EndOfToken)));
+      }
+      break;
+    case tgpptoken_literal:
+      CurRecords->push_back(
+          TGPPRecord(tgpprecord_literal,
+                     std::string(BeginOfToken, EndOfToken)));
+      break;
+    default:
+      return true;
+    }
+  }
+  return false;
+}
+
+bool TGPreprocessor::ParseForLoop() {
+  TGPPRecord ForLoopRecord(tgpprecord_for);
+
+  for (;;) {
+    TGPPTokenKind Kind;
+    const char *BeginOfToken, *EndOfToken;
+
+    Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+    if (!MatchIdentifier(Kind, BeginOfToken, EndOfToken)) {
+      PrintError(BeginOfToken, "Not an identifier");
+      return true;
+    }
+    std::string IndexVar(BeginOfToken, EndOfToken);
+
+    Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+    if (!MatchSymbol(Kind, BeginOfToken, EndOfToken, '=')) {
+      PrintError(BeginOfToken, "Need a '=' here");
+      return true;
+    }
+
+    TGPPRange Range;
+    if (ParseRange(&Range))
+      return true;
+    ForLoopRecord.AppendIndex(IndexVar, Range);
+
+    Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+    if (Kind == tgpptoken_newline)
+      break;
+    if (!MatchSymbol(Kind, BeginOfToken, EndOfToken, ',')) {
+      PrintError(BeginOfToken, "Need a ',' here");
+      return true;
+    }
+  }
+
+  // open a new level
+  TGPPRecords *LastCurRecords = CurRecords;
+  CurRecords = ForLoopRecord.GetLoopBody();
+
+  if (ParseBlock(false))
+    return true;
+
+  CurRecords = LastCurRecords;
+  CurRecords->push_back(ForLoopRecord);
+  return false;
+}
+
+bool TGPreprocessor::ParseRange(TGPPRange *Range) {
+  TGPPTokenKind Kind;
+  const char *BeginOfToken, *EndOfToken;
+
+  Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+
+  if (MatchSymbol(Kind, BeginOfToken, EndOfToken, '[')) {
+    for (;;) {
+      Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+      if (!MatchIdNum(Kind, BeginOfToken, EndOfToken)) {
+        PrintError(BeginOfToken, "Need a identifier or a number here");
+        return true;
+      }
+      Range->push_back(std::string(BeginOfToken, EndOfToken));
+
+      Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+      if (MatchSymbol(Kind, BeginOfToken, EndOfToken, ']'))
+        break;
+      if (!MatchSymbol(Kind, BeginOfToken, EndOfToken, ',')) {
+        PrintError(BeginOfToken, "Need a comma here");
+        return true;
+      }
+    }
+    return false;
+  }
+
+  else if (MatchSymbol(Kind, BeginOfToken, EndOfToken, "sequence")) {
+    long int from, to;
+
+    Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+    if (!MatchSymbol(Kind, BeginOfToken, EndOfToken, '(')) {
+      PrintError(BeginOfToken, "Need a left parentheses here");
+      return true;
+    }
+
+    Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+    if (!MatchNumber(Kind, BeginOfToken, EndOfToken, &from)) {
+      PrintError(BeginOfToken, "Not a number");
+      return true;
+    }
+
+    Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+    if (!MatchSymbol(Kind, BeginOfToken, EndOfToken, ',')) {
+      PrintError(BeginOfToken, "Need a comma here");
+      return true;
+    }
+
+    Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+    if (!MatchNumber(Kind, BeginOfToken, EndOfToken, &to)) {
+      PrintError(BeginOfToken, "Not a number");
+      return true;
+    }
+
+    Kind = Lexer->NextToken(&BeginOfToken, &EndOfToken);
+    if (!MatchSymbol(Kind, BeginOfToken, EndOfToken, ')')) {
+      PrintError(BeginOfToken, "Need a right parentheses here");
+      return true;
+    }
+
+    *Range = TGPPRange(from, to);
+    return false;
+  }
+
+  PrintError(BeginOfToken, "illegal range of loop index");
+  return true;
+}
+
+bool TGPreprocessor::PreprocessFile() {
+  TGPPLexer TheLexer(SrcMgr);
+  TGPPRecords TopLevelRecords;
+
+  Lexer = &TheLexer;
+  CurRecords = &TopLevelRecords;
+  if (ParseBlock(true))
+    return true;
+
+  TGPPEnvironment Env;
+  for (TGPPRecords::const_iterator i = TopLevelRecords.begin(),
+                                   e = TopLevelRecords.end();
+      i != e; ++i)
+    if (i->Evaluate(Env, Out.os()))
+      return true;
+
+  return false;
+}

lib/TableGen/TGPreprocessor.h

@@ -0,0 +1,52 @@
+//===- TGPreprocessor.h - Preprocessor for TableGen Files -------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class represents the Preprocessor for tablegen files.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef TGPREPROCESSOR_H
+#define TGPREPROCESSOR_H
+
+#include <vector>
+
+namespace llvm {
+class MemoryBuffer;
+class SourceMgr;
+class tool_output_file;
+
+class TGPPLexer;
+class TGPPRange;
+class TGPPRecord;
+
+typedef std::vector<TGPPRecord> TGPPRecords;
+
+class TGPreprocessor {
+  SourceMgr &SrcMgr;
+  tool_output_file &Out;
+
+  TGPPLexer *Lexer;
+  TGPPRecords *CurRecords;
+
+  bool ParseBlock(bool TopLevel);
+  bool ParseForLoop();
+  bool ParseRange(TGPPRange *Range);
+
+public:
+  TGPreprocessor(SourceMgr &SM, tool_output_file &O)
+    : SrcMgr(SM), Out(O), Lexer(NULL), CurRecords(NULL) {
+  }
+
+  /// PreprocessFile - Main entrypoint for preprocess a tblgen file.  These
+  /// preprocess routines return true on error, or false on success.
+  bool PreprocessFile();
+};
+} // namespace llvm
+
+#endif /* TGPREPROCESSOR_H */