Add mechanism to dump JIT-compiled objects to files

This commit introduces the bits to be able to dump JIT-compile
objects to external files by passing an object cache to OrcJit.
The new functionality is tested in mlir-cpu-runner under the flag
`dump-object-file`.

Closes tensorflow/mlir#95

PiperOrigin-RevId: 266439265

diff --git a/mlir/include/mlir/ExecutionEngine/ExecutionEngine.h b/mlir/include/mlir/ExecutionEngine/ExecutionEngine.h
index e3ba490..72aacd0 100644 (file)
--- a/mlir/include/mlir/ExecutionEngine/ExecutionEngine.h
+++ b/mlir/include/mlir/ExecutionEngine/ExecutionEngine.h
@@ -49,8 +49,11 @@ public:
                             llvm::MemoryBufferRef ObjBuffer) override;
   std::unique_ptr<llvm::MemoryBuffer> getObject(const llvm::Module *M) override;
 
+  /// Dump cached object to output file `filename`.
+  void dumpToObjectFile(llvm::StringRef filename);
+
 private:
-  llvm::StringMap<std::unique_ptr<llvm::MemoryBuffer>> CachedObjects;
+  llvm::StringMap<std::unique_ptr<llvm::MemoryBuffer>> cachedObjects;
 };
 
 /// JIT-backed execution engine for MLIR modules.  Assumes the module can be
@@ -65,15 +68,18 @@ private:
 /// be used to invoke the JIT-compiled function.
 class ExecutionEngine {
 public:
+  ExecutionEngine(bool enableObjectCache);
+
   /// Creates an execution engine for the given module.  If `transformer` is
   /// provided, it will be called on the LLVM module during JIT-compilation and
   /// can be used, e.g., for reporting or optimization.
   /// If `sharedLibPaths` are provided, the underlying JIT-compilation will open
   /// and link the shared libraries for symbol resolution.
-  static llvm::Expected<std::unique_ptr<ExecutionEngine>>
-  create(ModuleOp m,
-         std::function<llvm::Error(llvm::Module *)> transformer = {},
-         ArrayRef<StringRef> sharedLibPaths = {});
+  /// If `objectCache` is provided, JIT compiler will use it to store the object
+  /// generated for the given module.
+  static llvm::Expected<std::unique_ptr<ExecutionEngine>> create(
+      ModuleOp m, std::function<llvm::Error(llvm::Module *)> transformer = {},
+      ArrayRef<StringRef> sharedLibPaths = {}, bool enableObjectCache = false);
 
   /// Looks up a packed-argument function with the given name and returns a
   /// pointer to it.  Propagates errors in case of failure.
@@ -94,6 +100,9 @@ public:
   /// the engine.
   static bool setupTargetTriple(llvm::Module *llvmModule);
 
+  /// Dump object code to output file `filename`.
+  void dumpToObjectFile(llvm::StringRef filename);
+
 private:
   // Ordering of llvmContext and jit is important for destruction purposes: the
   // jit must be destroyed before the context.

diff --git a/mlir/lib/ExecutionEngine/ExecutionEngine.cpp b/mlir/lib/ExecutionEngine/ExecutionEngine.cpp
index dbc59d0..08b8086 100644 (file)
--- a/mlir/lib/ExecutionEngine/ExecutionEngine.cpp
+++ b/mlir/lib/ExecutionEngine/ExecutionEngine.cpp
@@ -22,6 +22,7 @@
 #include "mlir/ExecutionEngine/ExecutionEngine.h"
 #include "mlir/IR/Function.h"
 #include "mlir/IR/Module.h"
+#include "mlir/Support/FileUtilities.h"
 #include "mlir/Target/LLVMIR.h"
 
 #include "llvm/Bitcode/BitcodeReader.h"
@@ -37,6 +38,7 @@
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/Support/Error.h"
 #include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/ToolOutputFile.h"
 
 using namespace mlir;
 using llvm::dbgs;
@@ -68,13 +70,13 @@ namespace mlir {
 
 void SimpleObjectCache::notifyObjectCompiled(const Module *M,
                                              MemoryBufferRef ObjBuffer) {
-  CachedObjects[M->getModuleIdentifier()] = MemoryBuffer::getMemBufferCopy(
+  cachedObjects[M->getModuleIdentifier()] = MemoryBuffer::getMemBufferCopy(
       ObjBuffer.getBuffer(), ObjBuffer.getBufferIdentifier());
 }
 
 std::unique_ptr<MemoryBuffer> SimpleObjectCache::getObject(const Module *M) {
-  auto I = CachedObjects.find(M->getModuleIdentifier());
-  if (I == CachedObjects.end()) {
+  auto I = cachedObjects.find(M->getModuleIdentifier());
+  if (I == cachedObjects.end()) {
     dbgs() << "No object for " << M->getModuleIdentifier()
            << " in cache. Compiling.\n";
     return nullptr;
@@ -84,6 +86,26 @@ std::unique_ptr<MemoryBuffer> SimpleObjectCache::getObject(const Module *M) {
   return MemoryBuffer::getMemBuffer(I->second->getMemBufferRef());
 }
 
+void SimpleObjectCache::dumpToObjectFile(llvm::StringRef outputFilename) {
+  // Set up the output file.
+  std::string errorMessage;
+  auto file = openOutputFile(outputFilename, &errorMessage);
+  if (!file) {
+    llvm::errs() << errorMessage << "\n";
+    return;
+  }
+
+  // Dump the object generated for a single module to the output file.
+  assert(cachedObjects.size() == 1 && "Expected only one object entry.");
+  auto &cachedObject = cachedObjects.begin()->second;
+  file->os() << cachedObject->getBuffer();
+  file->keep();
+}
+
+void ExecutionEngine::dumpToObjectFile(llvm::StringRef filename) {
+  cache->dumpToObjectFile(filename);
+}
+
 // Setup LLVM target triple from the current machine.
 bool ExecutionEngine::setupTargetTriple(Module *llvmModule) {
   // Setup the machine properties from the current architecture.
@@ -168,11 +190,13 @@ void packFunctionArguments(Module *module) {
   }
 }
 
-Expected<std::unique_ptr<ExecutionEngine>>
-ExecutionEngine::create(ModuleOp m,
-                        std::function<Error(llvm::Module *)> transformer,
-                        ArrayRef<StringRef> sharedLibPaths) {
-  auto engine = std::make_unique<ExecutionEngine>();
+ExecutionEngine::ExecutionEngine(bool enableObjectCache)
+    : cache(enableObjectCache ? nullptr : new SimpleObjectCache()) {}
+
+Expected<std::unique_ptr<ExecutionEngine>> ExecutionEngine::create(
+    ModuleOp m, std::function<Error(llvm::Module *)> transformer,
+    ArrayRef<StringRef> sharedLibPaths, bool enableObjectCache) {
+  auto engine = std::make_unique<ExecutionEngine>(enableObjectCache);
 
   std::unique_ptr<llvm::LLVMContext> ctx(new llvm::LLVMContext);
   auto llvmModule = translateModuleToLLVMIR(m);
@@ -280,5 +304,4 @@ Error ExecutionEngine::invoke(StringRef name, MutableArrayRef<void *> args) {
 
   return Error::success();
 }
-
 } // end namespace mlir

diff --git a/mlir/lib/Support/JitRunner.cpp b/mlir/lib/Support/JitRunner.cpp
index 40f1292..c79a55f 100644 (file)
--- a/mlir/lib/Support/JitRunner.cpp
+++ b/mlir/lib/Support/JitRunner.cpp
@@ -95,6 +95,16 @@ static llvm::cl::list<std::string>
                  llvm::cl::ZeroOrMore, llvm::cl::MiscFlags::CommaSeparated,
                  llvm::cl::cat(clOptionsCategory));
 
+// CLI variables for debugging.
+static llvm::cl::opt<bool> dumpObjectFile(
+    "dump-object-file",
+    llvm::cl::desc("Dump JITted-compiled object to file specified with "
+                   "-object-filename (<input file>.o by default)."));
+
+static llvm::cl::opt<std::string> objectFilename(
+    "object-filename",
+    llvm::cl::desc("Dump JITted-compiled object to file <input file>.o"));
+
 static OwningModuleRef parseMLIRInput(StringRef inputFilename,
                                       MLIRContext *context) {
   // Set up the input file.
@@ -181,6 +191,11 @@ compileAndExecute(ModuleOp module, StringRef entryPoint,
   auto expectedFPtr = engine->lookup(entryPoint);
   if (!expectedFPtr)
     return expectedFPtr.takeError();
+
+  if (dumpObjectFile)
+    engine->dumpToObjectFile(objectFilename.empty() ? inputFilename + ".o"
+                                                    : objectFilename);
+
   void (*fptr)(void **) = *expectedFPtr;
   (*fptr)(args);
 

diff --git a/mlir/test/mlir-cpu-runner/simple.mlir b/mlir/test/mlir-cpu-runner/simple.mlir
index 5fcbd22..6f8d0db 100644 (file)
--- a/mlir/test/mlir-cpu-runner/simple.mlir
+++ b/mlir/test/mlir-cpu-runner/simple.mlir
@@ -4,6 +4,15 @@
 // RUN: mlir-cpu-runner %s -O3 -loop-distribute -loop-vectorize | FileCheck %s
 // RUN: mlir-cpu-runner %s -loop-distribute -loop-vectorize | FileCheck %s
 
+// RUN: cp %s %t
+// RUN: mlir-cpu-runner %t -dump-object-file | FileCheck %t
+// RUN: ls %t.o
+// RUN: rm %t.o
+
+// RUN: mlir-cpu-runner %s -dump-object-file -object-filename=%T/test.o | FileCheck %s
+// RUN: ls %T/test.o
+// RUN: rm %T/test.o
+
 func @fabsf(f32) -> f32
 
 func @main(%a : memref<2xf32>, %b : memref<1xf32>) {