xref: /llvm-project/mlir/examples/toy/Ch7/toyc.cpp (revision 4c2f90f362e401162533ffede924ac3890c70577)
1 //===- toyc.cpp - The Toy Compiler ----------------------------------------===//
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 // This file implements the entry point for the Toy compiler.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "mlir/Dialect/Func/Extensions/AllExtensions.h"
14 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
15 #include "mlir/Dialect/LLVMIR/Transforms/InlinerInterfaceImpl.h"
16 #include "toy/AST.h"
17 #include "toy/Dialect.h"
18 #include "toy/Lexer.h"
19 #include "toy/MLIRGen.h"
20 #include "toy/Parser.h"
21 #include "toy/Passes.h"
22 
23 #include "mlir/Dialect/Affine/Passes.h"
24 #include "mlir/Dialect/LLVMIR/Transforms/Passes.h"
25 #include "mlir/ExecutionEngine/ExecutionEngine.h"
26 #include "mlir/ExecutionEngine/OptUtils.h"
27 #include "mlir/IR/AsmState.h"
28 #include "mlir/IR/BuiltinOps.h"
29 #include "mlir/IR/MLIRContext.h"
30 #include "mlir/IR/Verifier.h"
31 #include "mlir/InitAllDialects.h"
32 #include "mlir/Parser/Parser.h"
33 #include "mlir/Pass/PassManager.h"
34 #include "mlir/Target/LLVMIR/Dialect/Builtin/BuiltinToLLVMIRTranslation.h"
35 #include "mlir/Target/LLVMIR/Dialect/LLVMIR/LLVMToLLVMIRTranslation.h"
36 #include "mlir/Target/LLVMIR/Export.h"
37 #include "mlir/Transforms/Passes.h"
38 
39 #include "llvm/ADT/StringRef.h"
40 #include "llvm/ExecutionEngine/Orc/JITTargetMachineBuilder.h"
41 #include "llvm/IR/Module.h"
42 #include "llvm/Support/CommandLine.h"
43 #include "llvm/Support/ErrorOr.h"
44 #include "llvm/Support/MemoryBuffer.h"
45 #include "llvm/Support/SourceMgr.h"
46 #include "llvm/Support/TargetSelect.h"
47 #include "llvm/Support/raw_ostream.h"
48 #include <cassert>
49 #include <memory>
50 #include <string>
51 #include <system_error>
52 #include <utility>
53 
54 using namespace toy;
55 namespace cl = llvm::cl;
56 
57 static cl::opt<std::string> inputFilename(cl::Positional,
58                                           cl::desc("<input toy file>"),
59                                           cl::init("-"),
60                                           cl::value_desc("filename"));
61 
62 namespace {
63 enum InputType { Toy, MLIR };
64 } // namespace
65 static cl::opt<enum InputType> inputType(
66     "x", cl::init(Toy), cl::desc("Decided the kind of output desired"),
67     cl::values(clEnumValN(Toy, "toy", "load the input file as a Toy source.")),
68     cl::values(clEnumValN(MLIR, "mlir",
69                           "load the input file as an MLIR file")));
70 
71 namespace {
72 enum Action {
73   None,
74   DumpAST,
75   DumpMLIR,
76   DumpMLIRAffine,
77   DumpMLIRLLVM,
78   DumpLLVMIR,
79   RunJIT
80 };
81 } // namespace
82 static cl::opt<enum Action> emitAction(
83     "emit", cl::desc("Select the kind of output desired"),
84     cl::values(clEnumValN(DumpAST, "ast", "output the AST dump")),
85     cl::values(clEnumValN(DumpMLIR, "mlir", "output the MLIR dump")),
86     cl::values(clEnumValN(DumpMLIRAffine, "mlir-affine",
87                           "output the MLIR dump after affine lowering")),
88     cl::values(clEnumValN(DumpMLIRLLVM, "mlir-llvm",
89                           "output the MLIR dump after llvm lowering")),
90     cl::values(clEnumValN(DumpLLVMIR, "llvm", "output the LLVM IR dump")),
91     cl::values(
92         clEnumValN(RunJIT, "jit",
93                    "JIT the code and run it by invoking the main function")));
94 
95 static cl::opt<bool> enableOpt("opt", cl::desc("Enable optimizations"));
96 
97 /// Returns a Toy AST resulting from parsing the file or a nullptr on error.
98 std::unique_ptr<toy::ModuleAST> parseInputFile(llvm::StringRef filename) {
99   llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> fileOrErr =
100       llvm::MemoryBuffer::getFileOrSTDIN(filename);
101   if (std::error_code ec = fileOrErr.getError()) {
102     llvm::errs() << "Could not open input file: " << ec.message() << "\n";
103     return nullptr;
104   }
105   auto buffer = fileOrErr.get()->getBuffer();
106   LexerBuffer lexer(buffer.begin(), buffer.end(), std::string(filename));
107   Parser parser(lexer);
108   return parser.parseModule();
109 }
110 
111 int loadMLIR(mlir::MLIRContext &context,
112              mlir::OwningOpRef<mlir::ModuleOp> &module) {
113   // Handle '.toy' input to the compiler.
114   if (inputType != InputType::MLIR &&
115       !llvm::StringRef(inputFilename).ends_with(".mlir")) {
116     auto moduleAST = parseInputFile(inputFilename);
117     if (!moduleAST)
118       return 6;
119     module = mlirGen(context, *moduleAST);
120     return !module ? 1 : 0;
121   }
122 
123   // Otherwise, the input is '.mlir'.
124   llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> fileOrErr =
125       llvm::MemoryBuffer::getFileOrSTDIN(inputFilename);
126   if (std::error_code ec = fileOrErr.getError()) {
127     llvm::errs() << "Could not open input file: " << ec.message() << "\n";
128     return -1;
129   }
130 
131   // Parse the input mlir.
132   llvm::SourceMgr sourceMgr;
133   sourceMgr.AddNewSourceBuffer(std::move(*fileOrErr), llvm::SMLoc());
134   module = mlir::parseSourceFile<mlir::ModuleOp>(sourceMgr, &context);
135   if (!module) {
136     llvm::errs() << "Error can't load file " << inputFilename << "\n";
137     return 3;
138   }
139   return 0;
140 }
141 
142 int loadAndProcessMLIR(mlir::MLIRContext &context,
143                        mlir::OwningOpRef<mlir::ModuleOp> &module) {
144   if (int error = loadMLIR(context, module))
145     return error;
146 
147   mlir::PassManager pm(module.get()->getName());
148   // Apply any generic pass manager command line options and run the pipeline.
149   if (mlir::failed(mlir::applyPassManagerCLOptions(pm)))
150     return 4;
151 
152   // Check to see what granularity of MLIR we are compiling to.
153   bool isLoweringToAffine = emitAction >= Action::DumpMLIRAffine;
154   bool isLoweringToLLVM = emitAction >= Action::DumpMLIRLLVM;
155 
156   if (enableOpt || isLoweringToAffine) {
157     // Inline all functions into main and then delete them.
158     pm.addPass(mlir::createInlinerPass());
159 
160     // Now that there is only one function, we can infer the shapes of each of
161     // the operations.
162     mlir::OpPassManager &optPM = pm.nest<mlir::toy::FuncOp>();
163     optPM.addPass(mlir::createCanonicalizerPass());
164     optPM.addPass(mlir::toy::createShapeInferencePass());
165     optPM.addPass(mlir::createCanonicalizerPass());
166     optPM.addPass(mlir::createCSEPass());
167   }
168 
169   if (isLoweringToAffine) {
170     // Partially lower the toy dialect.
171     pm.addPass(mlir::toy::createLowerToAffinePass());
172 
173     // Add a few cleanups post lowering.
174     mlir::OpPassManager &optPM = pm.nest<mlir::func::FuncOp>();
175     optPM.addPass(mlir::createCanonicalizerPass());
176     optPM.addPass(mlir::createCSEPass());
177 
178     // Add optimizations if enabled.
179     if (enableOpt) {
180       optPM.addPass(mlir::affine::createLoopFusionPass());
181       optPM.addPass(mlir::affine::createAffineScalarReplacementPass());
182     }
183   }
184 
185   if (isLoweringToLLVM) {
186     // Finish lowering the toy IR to the LLVM dialect.
187     pm.addPass(mlir::toy::createLowerToLLVMPass());
188     // This is necessary to have line tables emitted and basic
189     // debugger working. In the future we will add proper debug information
190     // emission directly from our frontend.
191     pm.addPass(mlir::LLVM::createDIScopeForLLVMFuncOpPass());
192   }
193 
194   if (mlir::failed(pm.run(*module)))
195     return 4;
196   return 0;
197 }
198 
199 int dumpAST() {
200   if (inputType == InputType::MLIR) {
201     llvm::errs() << "Can't dump a Toy AST when the input is MLIR\n";
202     return 5;
203   }
204 
205   auto moduleAST = parseInputFile(inputFilename);
206   if (!moduleAST)
207     return 1;
208 
209   dump(*moduleAST);
210   return 0;
211 }
212 
213 int dumpLLVMIR(mlir::ModuleOp module) {
214   // Register the translation to LLVM IR with the MLIR context.
215   mlir::registerBuiltinDialectTranslation(*module->getContext());
216   mlir::registerLLVMDialectTranslation(*module->getContext());
217 
218   // Convert the module to LLVM IR in a new LLVM IR context.
219   llvm::LLVMContext llvmContext;
220   auto llvmModule = mlir::translateModuleToLLVMIR(module, llvmContext);
221   if (!llvmModule) {
222     llvm::errs() << "Failed to emit LLVM IR\n";
223     return -1;
224   }
225 
226   // Initialize LLVM targets.
227   llvm::InitializeNativeTarget();
228   llvm::InitializeNativeTargetAsmPrinter();
229 
230   // Create target machine and configure the LLVM Module
231   auto tmBuilderOrError = llvm::orc::JITTargetMachineBuilder::detectHost();
232   if (!tmBuilderOrError) {
233     llvm::errs() << "Could not create JITTargetMachineBuilder\n";
234     return -1;
235   }
236 
237   auto tmOrError = tmBuilderOrError->createTargetMachine();
238   if (!tmOrError) {
239     llvm::errs() << "Could not create TargetMachine\n";
240     return -1;
241   }
242   mlir::ExecutionEngine::setupTargetTripleAndDataLayout(llvmModule.get(),
243                                                         tmOrError.get().get());
244 
245   /// Optionally run an optimization pipeline over the llvm module.
246   auto optPipeline = mlir::makeOptimizingTransformer(
247       /*optLevel=*/enableOpt ? 3 : 0, /*sizeLevel=*/0,
248       /*targetMachine=*/nullptr);
249   if (auto err = optPipeline(llvmModule.get())) {
250     llvm::errs() << "Failed to optimize LLVM IR " << err << "\n";
251     return -1;
252   }
253   llvm::errs() << *llvmModule << "\n";
254   return 0;
255 }
256 
257 int runJit(mlir::ModuleOp module) {
258   // Initialize LLVM targets.
259   llvm::InitializeNativeTarget();
260   llvm::InitializeNativeTargetAsmPrinter();
261 
262   // Register the translation from MLIR to LLVM IR, which must happen before we
263   // can JIT-compile.
264   mlir::registerBuiltinDialectTranslation(*module->getContext());
265   mlir::registerLLVMDialectTranslation(*module->getContext());
266 
267   // An optimization pipeline to use within the execution engine.
268   auto optPipeline = mlir::makeOptimizingTransformer(
269       /*optLevel=*/enableOpt ? 3 : 0, /*sizeLevel=*/0,
270       /*targetMachine=*/nullptr);
271 
272   // Create an MLIR execution engine. The execution engine eagerly JIT-compiles
273   // the module.
274   mlir::ExecutionEngineOptions engineOptions;
275   engineOptions.transformer = optPipeline;
276   auto maybeEngine = mlir::ExecutionEngine::create(module, engineOptions);
277   assert(maybeEngine && "failed to construct an execution engine");
278   auto &engine = maybeEngine.get();
279 
280   // Invoke the JIT-compiled function.
281   auto invocationResult = engine->invokePacked("main");
282   if (invocationResult) {
283     llvm::errs() << "JIT invocation failed\n";
284     return -1;
285   }
286 
287   return 0;
288 }
289 
290 int main(int argc, char **argv) {
291   // Register any command line options.
292   mlir::registerAsmPrinterCLOptions();
293   mlir::registerMLIRContextCLOptions();
294   mlir::registerPassManagerCLOptions();
295 
296   cl::ParseCommandLineOptions(argc, argv, "toy compiler\n");
297 
298   if (emitAction == Action::DumpAST)
299     return dumpAST();
300 
301   // If we aren't dumping the AST, then we are compiling with/to MLIR.
302   mlir::DialectRegistry registry;
303   mlir::func::registerAllExtensions(registry);
304   mlir::LLVM::registerInlinerInterface(registry);
305 
306   mlir::MLIRContext context(registry);
307   // Load our Dialect in this MLIR Context.
308   context.getOrLoadDialect<mlir::toy::ToyDialect>();
309 
310   mlir::OwningOpRef<mlir::ModuleOp> module;
311   if (int error = loadAndProcessMLIR(context, module))
312     return error;
313 
314   // If we aren't exporting to non-mlir, then we are done.
315   bool isOutputingMLIR = emitAction <= Action::DumpMLIRLLVM;
316   if (isOutputingMLIR) {
317     module->dump();
318     return 0;
319   }
320 
321   // Check to see if we are compiling to LLVM IR.
322   if (emitAction == Action::DumpLLVMIR)
323     return dumpLLVMIR(*module);
324 
325   // Otherwise, we must be running the jit.
326   if (emitAction == Action::RunJIT)
327     return runJit(*module);
328 
329   llvm::errs() << "No action specified (parsing only?), use -emit=<action>\n";
330   return -1;
331 }
332