| 1 | //===- mlir-pdll.cpp - MLIR PDLL frontend -----------------------*- C++ -*-===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | #include "mlir/IR/BuiltinOps.h" |
| 10 | #include "mlir/Support/FileUtilities.h" |
| 11 | #include "mlir/Support/ToolUtilities.h" |
| 12 | #include "mlir/Tools/PDLL/AST/Context.h" |
| 13 | #include "mlir/Tools/PDLL/AST/Nodes.h" |
| 14 | #include "mlir/Tools/PDLL/CodeGen/CPPGen.h" |
| 15 | #include "mlir/Tools/PDLL/CodeGen/MLIRGen.h" |
| 16 | #include "mlir/Tools/PDLL/ODS/Context.h" |
| 17 | #include "mlir/Tools/PDLL/Parser/Parser.h" |
| 18 | #include "llvm/Support/CommandLine.h" |
| 19 | #include "llvm/Support/InitLLVM.h" |
| 20 | #include "llvm/Support/SourceMgr.h" |
| 21 | #include "llvm/Support/ToolOutputFile.h" |
| 22 | #include <set> |
| 23 | |
| 24 | using namespace mlir; |
| 25 | using namespace mlir::pdll; |
| 26 | |
| 27 | //===----------------------------------------------------------------------===// |
| 28 | // main |
| 29 | //===----------------------------------------------------------------------===// |
| 30 | |
| 31 | /// The desired output type. |
| 32 | enum class OutputType { |
| 33 | AST, |
| 34 | MLIR, |
| 35 | CPP, |
| 36 | }; |
| 37 | |
| 38 | static LogicalResult |
| 39 | processBuffer(raw_ostream &os, std::unique_ptr<llvm::MemoryBuffer> chunkBuffer, |
| 40 | OutputType outputType, std::vector<std::string> &includeDirs, |
| 41 | bool dumpODS, std::set<std::string> *includedFiles) { |
| 42 | llvm::SourceMgr sourceMgr; |
| 43 | sourceMgr.setIncludeDirs(includeDirs); |
| 44 | sourceMgr.AddNewSourceBuffer(F: std::move(chunkBuffer), IncludeLoc: SMLoc()); |
| 45 | |
| 46 | // If we are dumping ODS information, also enable documentation to ensure the |
| 47 | // summary and description information is imported as well. |
| 48 | bool enableDocumentation = dumpODS; |
| 49 | |
| 50 | ods::Context odsContext; |
| 51 | ast::Context astContext(odsContext); |
| 52 | FailureOr<ast::Module *> module = |
| 53 | parsePDLLAST(ctx&: astContext, sourceMgr, enableDocumentation); |
| 54 | if (failed(Result: module)) |
| 55 | return failure(); |
| 56 | |
| 57 | // Add the files that were included to the set. |
| 58 | if (includedFiles) { |
| 59 | for (unsigned i = 1, e = sourceMgr.getNumBuffers(); i < e; ++i) { |
| 60 | includedFiles->insert( |
| 61 | x: sourceMgr.getMemoryBuffer(i: i + 1)->getBufferIdentifier().str()); |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | // Print out the ODS information if requested. |
| 66 | if (dumpODS) |
| 67 | odsContext.print(os&: llvm::errs()); |
| 68 | |
| 69 | // Generate the output. |
| 70 | if (outputType == OutputType::AST) { |
| 71 | (*module)->print(os); |
| 72 | return success(); |
| 73 | } |
| 74 | |
| 75 | MLIRContext mlirContext; |
| 76 | OwningOpRef<ModuleOp> pdlModule = |
| 77 | codegenPDLLToMLIR(&mlirContext, astContext, sourceMgr, **module); |
| 78 | if (!pdlModule) |
| 79 | return failure(); |
| 80 | |
| 81 | if (outputType == OutputType::MLIR) { |
| 82 | pdlModule->print(os, OpPrintingFlags().enableDebugInfo()); |
| 83 | return success(); |
| 84 | } |
| 85 | codegenPDLLToCPP(**module, *pdlModule, os); |
| 86 | return success(); |
| 87 | } |
| 88 | |
| 89 | /// Create a dependency file for `-d` option. |
| 90 | /// |
| 91 | /// This functionality is generally only for the benefit of the build system, |
| 92 | /// and is modeled after the same option in TableGen. |
| 93 | static LogicalResult |
| 94 | createDependencyFile(StringRef outputFilename, StringRef dependencyFile, |
| 95 | std::set<std::string> &includedFiles) { |
| 96 | if (outputFilename == "-" ) { |
| 97 | llvm::errs() << "error: the option -d must be used together with -o\n" ; |
| 98 | return failure(); |
| 99 | } |
| 100 | |
| 101 | std::string errorMessage; |
| 102 | std::unique_ptr<llvm::ToolOutputFile> outputFile = |
| 103 | openOutputFile(outputFilename: dependencyFile, errorMessage: &errorMessage); |
| 104 | if (!outputFile) { |
| 105 | llvm::errs() << errorMessage << "\n" ; |
| 106 | return failure(); |
| 107 | } |
| 108 | |
| 109 | outputFile->os() << outputFilename << ":" ; |
| 110 | for (const auto &includeFile : includedFiles) |
| 111 | outputFile->os() << ' ' << includeFile; |
| 112 | outputFile->os() << "\n" ; |
| 113 | outputFile->keep(); |
| 114 | return success(); |
| 115 | } |
| 116 | |
| 117 | int main(int argc, char **argv) { |
| 118 | // FIXME: This is necessary because we link in TableGen, which defines its |
| 119 | // options as static variables.. some of which overlap with our options. |
| 120 | llvm::cl::ResetCommandLineParser(); |
| 121 | |
| 122 | llvm::cl::opt<std::string> inputFilename( |
| 123 | llvm::cl::Positional, llvm::cl::desc("<input file>" ), llvm::cl::init(Val: "-" ), |
| 124 | llvm::cl::value_desc("filename" )); |
| 125 | |
| 126 | llvm::cl::opt<std::string> outputFilename( |
| 127 | "o" , llvm::cl::desc("Output filename" ), llvm::cl::value_desc("filename" ), |
| 128 | llvm::cl::init(Val: "-" )); |
| 129 | |
| 130 | llvm::cl::list<std::string> includeDirs( |
| 131 | "I" , llvm::cl::desc("Directory of include files" ), |
| 132 | llvm::cl::value_desc("directory" ), llvm::cl::Prefix); |
| 133 | |
| 134 | llvm::cl::opt<bool> dumpODS( |
| 135 | "dump-ods" , |
| 136 | llvm::cl::desc( |
| 137 | "Print out the parsed ODS information from the input file" ), |
| 138 | llvm::cl::init(Val: false)); |
| 139 | llvm::cl::opt<std::string> inputSplitMarker{ |
| 140 | "split-input-file" , llvm::cl::ValueOptional, |
| 141 | llvm::cl::callback(CB: [&](const std::string &str) { |
| 142 | // Implicit value: use default marker if flag was used without value. |
| 143 | if (str.empty()) |
| 144 | inputSplitMarker.setValue(V: kDefaultSplitMarker); |
| 145 | }), |
| 146 | llvm::cl::desc("Split the input file into chunks using the given or " |
| 147 | "default marker and process each chunk independently" ), |
| 148 | llvm::cl::init(Val: "" )}; |
| 149 | llvm::cl::opt<std::string> outputSplitMarker( |
| 150 | "output-split-marker" , |
| 151 | llvm::cl::desc("Split marker to use for merging the ouput" ), |
| 152 | llvm::cl::init(Val: kDefaultSplitMarker)); |
| 153 | llvm::cl::opt<enum OutputType> outputType( |
| 154 | "x" , llvm::cl::init(Val: OutputType::AST), |
| 155 | llvm::cl::desc("The type of output desired" ), |
| 156 | llvm::cl::values(clEnumValN(OutputType::AST, "ast" , |
| 157 | "generate the AST for the input file" ), |
| 158 | clEnumValN(OutputType::MLIR, "mlir" , |
| 159 | "generate the PDL MLIR for the input file" ), |
| 160 | clEnumValN(OutputType::CPP, "cpp" , |
| 161 | "generate a C++ source file containing the " |
| 162 | "patterns for the input file" ))); |
| 163 | llvm::cl::opt<std::string> dependencyFilename( |
| 164 | "d" , llvm::cl::desc("Dependency filename" ), |
| 165 | llvm::cl::value_desc("filename" ), llvm::cl::init(Val: "" )); |
| 166 | llvm::cl::opt<bool> writeIfChanged( |
| 167 | "write-if-changed" , |
| 168 | llvm::cl::desc("Only write to the output file if it changed" )); |
| 169 | |
| 170 | // `ResetCommandLineParser` at the above unregistered the "D" option |
| 171 | // of `llvm-tblgen`, which causes tblgen usage to fail due to |
| 172 | // "Unknnown command line argument '-D...`" when a macros name is |
| 173 | // present. The following is a workaround to re-register it again. |
| 174 | llvm::cl::list<std::string> macroNames( |
| 175 | "D" , |
| 176 | llvm::cl::desc("Name of the macro to be defined -- ignored by mlir-pdll" ), |
| 177 | llvm::cl::value_desc("macro name" ), llvm::cl::Prefix); |
| 178 | |
| 179 | llvm::InitLLVM y(argc, argv); |
| 180 | llvm::cl::ParseCommandLineOptions(argc, argv, Overview: "PDLL Frontend" ); |
| 181 | |
| 182 | // Set up the input file. |
| 183 | std::string errorMessage; |
| 184 | std::unique_ptr<llvm::MemoryBuffer> inputFile = |
| 185 | openInputFile(inputFilename, errorMessage: &errorMessage); |
| 186 | if (!inputFile) { |
| 187 | llvm::errs() << errorMessage << "\n" ; |
| 188 | return 1; |
| 189 | } |
| 190 | |
| 191 | // If we are creating a dependency file, we'll also need to track what files |
| 192 | // get included during processing. |
| 193 | std::set<std::string> includedFilesStorage; |
| 194 | std::set<std::string> *includedFiles = nullptr; |
| 195 | if (!dependencyFilename.empty()) |
| 196 | includedFiles = &includedFilesStorage; |
| 197 | |
| 198 | // The split-input-file mode is a very specific mode that slices the file |
| 199 | // up into small pieces and checks each independently. |
| 200 | std::string outputStr; |
| 201 | llvm::raw_string_ostream outputStrOS(outputStr); |
| 202 | auto processFn = [&](std::unique_ptr<llvm::MemoryBuffer> chunkBuffer, |
| 203 | raw_ostream &os) { |
| 204 | return processBuffer(os, chunkBuffer: std::move(chunkBuffer), outputType, includeDirs, |
| 205 | dumpODS, includedFiles); |
| 206 | }; |
| 207 | if (failed(Result: splitAndProcessBuffer(originalBuffer: std::move(inputFile), processChunkBuffer: processFn, os&: outputStrOS, |
| 208 | inputSplitMarker, outputSplitMarker))) |
| 209 | return 1; |
| 210 | |
| 211 | // Write the output. |
| 212 | bool shouldWriteOutput = true; |
| 213 | if (writeIfChanged) { |
| 214 | // Only update the real output file if there are any differences. This |
| 215 | // prevents recompilation of all the files depending on it if there aren't |
| 216 | // any. |
| 217 | if (auto existingOrErr = |
| 218 | llvm::MemoryBuffer::getFile(Filename: outputFilename, /*IsText=*/true)) |
| 219 | if (std::move(existingOrErr.get())->getBuffer() == outputStr) |
| 220 | shouldWriteOutput = false; |
| 221 | } |
| 222 | |
| 223 | // Populate the output file if necessary. |
| 224 | if (shouldWriteOutput) { |
| 225 | std::unique_ptr<llvm::ToolOutputFile> outputFile = |
| 226 | openOutputFile(outputFilename, errorMessage: &errorMessage); |
| 227 | if (!outputFile) { |
| 228 | llvm::errs() << errorMessage << "\n" ; |
| 229 | return 1; |
| 230 | } |
| 231 | outputFile->os() << outputStr; |
| 232 | outputFile->keep(); |
| 233 | } |
| 234 | |
| 235 | // Always write the depfile, even if the main output hasn't changed. If it's |
| 236 | // missing, Ninja considers the output dirty. |
| 237 | if (!dependencyFilename.empty()) { |
| 238 | if (failed(Result: createDependencyFile(outputFilename, dependencyFile: dependencyFilename, |
| 239 | includedFiles&: includedFilesStorage))) |
| 240 | return 1; |
| 241 | } |
| 242 | |
| 243 | return 0; |
| 244 | } |
| 245 | |