[MLIR][IRDL] Added IRDL to C++ Translation (llvm#133982)

This PR introduces a new tool, mlir-irdl-to-cpp, that converts IRDL to
C++ definitions.

The C++ definitions allow use of the IRDL-defined dialect in MLIR C++
infrastructure, enabling the use of conversion patterns with IRDL
dialects for example. This PR also adds CMake utilities to easily
integrate the IRDL dialects into MLIR projects.

Note that most IRDL features are not supported. In general, we are only
able to define simple types and operations.
- The only type constraint supported is `irdl.any`.
- Variadic operands and results are not supported.
- Verifiers for the IRDL constraints are not generated.
- Attributes are not supported.

---------

Co-authored-by: Théo Degioanni <[email protected]>
Co-authored-by: Fehr Mathieu <[email protected]>

Author: 3 people

mlir/CMakeLists.txt

@@ -63,6 +63,7 @@ list(INSERT CMAKE_MODULE_PATH 0
   )
 
 include(AddMLIR)
+include(IRDLToCpp)
 
 # -BSymbolic is incompatible with TypeID
 if("${CMAKE_SHARED_LINKER_FLAGS}" MATCHES "-Bsymbolic[^-]")
@@ -206,6 +207,7 @@ include_directories(BEFORE
 # Adding tools/mlir-tblgen here as calling add_tablegen sets some variables like
 # MLIR_TABLEGEN_EXE in PARENT_SCOPE which gets lost if that folder is included
 # from another directory like tools
+add_subdirectory(tools/mlir-irdl-to-cpp)
 add_subdirectory(tools/mlir-linalg-ods-gen)
 add_subdirectory(tools/mlir-pdll)
 add_subdirectory(tools/mlir-tblgen)

mlir/cmake/modules/CMakeLists.txt

@@ -112,6 +112,8 @@ set(MLIR_CONFIG_TABLEGEN_EXE mlir-tblgen)
 set(MLIR_CONFIG_PDLL_TABLEGEN_EXE mlir-pdll)
 set(MLIR_CONFIG_SRC_SHARDER_TABLEGEN_EXE mlir-src-sharder)
 
+set(MLIR_CONFIG_IRDL_TO_CPP_EXE mlir-irdl-to-cpp)
+
 configure_file(
   ${CMAKE_CURRENT_SOURCE_DIR}/MLIRConfig.cmake.in
   ${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/MLIRConfig.cmake
@@ -135,6 +137,7 @@ if (NOT LLVM_INSTALL_TOOLCHAIN_ONLY)
     ${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/MLIRConfigVersion.cmake
     ${CMAKE_CURRENT_SOURCE_DIR}/AddMLIR.cmake
     ${CMAKE_CURRENT_SOURCE_DIR}/AddMLIRPython.cmake
+    ${CMAKE_CURRENT_SOURCE_DIR}/IRDLToCpp.cmake
     ${CMAKE_CURRENT_SOURCE_DIR}/MLIRDetectPythonEnv.cmake
     DESTINATION ${MLIR_INSTALL_PACKAGE_DIR}
     COMPONENT mlir-cmake-exports)

mlir/cmake/modules/IRDLToCpp.cmake

@@ -0,0 +1,12 @@
+function(add_irdl_to_cpp_target target irdl_file)
+  add_custom_command(
+    OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${irdl_file}.cpp.inc
+    COMMAND ${MLIR_IRDL_TO_CPP_EXE} ${CMAKE_CURRENT_SOURCE_DIR}/${irdl_file} -o ${CMAKE_CURRENT_BINARY_DIR}/${irdl_file}.cpp.inc
+
+    # The command output depends on the executable to ensure IRDL sources are properly rebuilt
+    # if the tool changes.
+    DEPENDS ${MLIR_IRDL_TO_CPP_EXE} ${CMAKE_CURRENT_SOURCE_DIR}/${irdl_file}
+    COMMENT "Building ${irdl_file}..."
+  )
+  add_custom_target(${target} DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/${irdl_file}.cpp.inc)
+endfunction()

mlir/cmake/modules/MLIRConfig.cmake.in

@@ -12,6 +12,7 @@ set(MLIR_INCLUDE_DIRS "@MLIR_CONFIG_INCLUDE_DIRS@")
 set(MLIR_TABLEGEN_EXE "@MLIR_CONFIG_TABLEGEN_EXE@")
 set(MLIR_PDLL_TABLEGEN_EXE "@MLIR_CONFIG_PDLL_TABLEGEN_EXE@")
 set(MLIR_SRC_SHARDER_TABLEGEN_EXE "@MLIR_CONFIG_SRC_SHARDER_TABLEGEN_EXE@")
+set(MLIR_IRDL_TO_CPP_EXE "@MLIR_CONFIG_IRDL_TO_CPP_EXE@")
 set(MLIR_INSTALL_AGGREGATE_OBJECTS "@MLIR_INSTALL_AGGREGATE_OBJECTS@")
 set(MLIR_ENABLE_BINDINGS_PYTHON "@MLIR_ENABLE_BINDINGS_PYTHON@")
 set(MLIR_ENABLE_EXECUTION_ENGINE "@MLIR_ENABLE_EXECUTION_ENGINE@")

mlir/include/mlir/Dialect/IRDL/IR/IRDLOps.td

@@ -91,6 +91,7 @@ def IRDL_TypeOp : IRDL_Op<"type",
   let regions = (region SizedRegion<1>:$body);
   let assemblyFormat =
     "$sym_name attr-dict-with-keyword custom<SingleBlockRegion>($body)";
+  let hasVerifier = 1;
 }
 
 def IRDL_AttributeOp : IRDL_Op<"attribute",
@@ -126,6 +127,8 @@ def IRDL_AttributeOp : IRDL_Op<"attribute",
   let regions = (region SizedRegion<1>:$body);
   let assemblyFormat =
     "$sym_name attr-dict-with-keyword custom<SingleBlockRegion>($body)";
+
+  let hasVerifier = 1;
 }
 
 def IRDL_ParametersOp : IRDL_Op<"parameters",
@@ -202,6 +205,7 @@ def IRDL_OperationOp : IRDL_Op<"operation",
   let assemblyFormat =
     "$sym_name attr-dict-with-keyword custom<SingleBlockRegion>($body)";
   let hasRegionVerifier = true;
+  let hasVerifier = 1;
 }
 
 def IRDL_OperandsOp : IRDL_Op<"operands", [HasParent<"OperationOp">]> {

mlir/include/mlir/InitAllTranslations.h

@@ -14,6 +14,8 @@
 #ifndef MLIR_INITALLTRANSLATIONS_H
 #define MLIR_INITALLTRANSLATIONS_H
 
+#include "mlir/Target/IRDLToCpp/TranslationRegistration.h"
+
 namespace mlir {
 
 void registerFromLLVMIRTranslation();
@@ -33,6 +35,7 @@ inline void registerAllTranslations() {
   static bool initOnce = []() {
     registerFromLLVMIRTranslation();
     registerFromSPIRVTranslation();
+    registerIRDLToCppTranslation();
     registerToCppTranslation();
     registerToLLVMIRTranslation();
     registerToSPIRVTranslation();

mlir/include/mlir/Target/IRDLToCpp/IRDLToCpp.h

@@ -0,0 +1,38 @@
+//===- IRDLToCpp.h - Register translation -----------------------*- C++ -*-===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the registration function for the IRDL to C++ translation.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_TARGET_IRDLTOCPP_IRDLTOCPP_H
+#define MLIR_TARGET_IRDLTOCPP_IRDLTOCPP_H
+
+#include "mlir/Dialect/IRDL/IR/IRDL.h"
+
+namespace mlir {
+namespace irdl {
+
+/// Translates an IRDL dialect definition to a C++ definition that can be used
+/// with MLIR.
+///
+/// The following preprocessor macros will generate the following code:
+///
+///  // This define generates code for the dialect's class declarations
+///  #define GEN_DIALECT_DECL_HEADER
+///
+///  // This define generates code for the dialect's class definitions
+///  #define GEN_DIALECT_DEF
+LogicalResult
+translateIRDLDialectToCpp(llvm::ArrayRef<irdl::DialectOp> dialects,
+                          raw_ostream &output);
+
+} // namespace irdl
+} // namespace mlir
+
+#endif // MLIR_TARGET_IRDLTOCPP_IRDLTOCPP_H

mlir/include/mlir/Target/IRDLToCpp/TranslationRegistration.h

@@ -0,0 +1,22 @@
+//===- TranslationRegistration.h - Register translation ---------*- C++ -*-===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the registration function for the IRDL to C++ translation.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_TARGET_IRDLTOCPP_TRANSLATIONREGISTRATION_H
+#define MLIR_TARGET_IRDLTOCPP_TRANSLATIONREGISTRATION_H
+
+namespace mlir {
+
+void registerIRDLToCppTranslation();
+
+} // namespace mlir
+
+#endif // MLIR_TARGET_IRDLTOCPP_TRANSLATIONREGISTRATION_H

mlir/lib/Dialect/IRDL/IR/IRDL.cpp

@@ -74,13 +74,62 @@ static void printSingleBlockRegion(OpAsmPrinter &p, Operation *op,
   if (!region.getBlocks().front().empty())
     p.printRegion(region);
 }
+static llvm::LogicalResult isValidName(llvm::StringRef in, mlir::Operation *loc,
+                                       const Twine &label) {
+  if (in.empty())
+    return loc->emitError("name of ") << label << " is empty";
+
+  bool allowUnderscore = false;
+  for (auto &elem : in) {
+    if (elem == '_') {
+      if (!allowUnderscore)
+        return loc->emitError("name of ")
+               << label << " should not contain leading or double underscores";
+    } else {
+      if (!isalnum(elem))
+        return loc->emitError("name of ")
+               << label
+               << " must contain only lowercase letters, digits and "
+                  "underscores";
+
+      if (llvm::isUpper(elem))
+        return loc->emitError("name of ")
+               << label << " should not contain uppercase letters";
+    }
+
+    allowUnderscore = elem != '_';
+  }
+
+  return success();
+}
 
 LogicalResult DialectOp::verify() {
   if (!Dialect::isValidNamespace(getName()))
     return emitOpError("invalid dialect name");
+  if (failed(isValidName(getSymName(), getOperation(), "dialect")))
+    return failure();
+
   return success();
 }
 
+LogicalResult OperationOp::verify() {
+  return isValidName(getSymName(), getOperation(), "operation");
+}
+
+LogicalResult TypeOp::verify() {
+  auto symName = getSymName();
+  if (symName.front() == '!')
+    symName = symName.substr(1);
+  return isValidName(symName, getOperation(), "type");
+}
+
+LogicalResult AttributeOp::verify() {
+  auto symName = getSymName();
+  if (symName.front() == '#')
+    symName = symName.substr(1);
+  return isValidName(symName, getOperation(), "attribute");
+}
+
 LogicalResult OperationOp::verifyRegions() {
   // Stores pairs of value kinds and the list of names of values of this kind in
   // the operation.
@@ -133,18 +182,10 @@ static LogicalResult verifyNames(Operation *op, StringRef kindName,
   DenseMap<StringRef, size_t> nameMap;
   for (auto [i, name] : llvm::enumerate(names)) {
     StringRef nameRef = llvm::cast<StringAttr>(name).getValue();
-    if (nameRef.empty())
-      return op->emitOpError()
-             << "name of " << kindName << " #" << i << " is empty";
-    if (!llvm::isAlpha(nameRef[0]) && nameRef[0] != '_')
-      return op->emitOpError()
-             << "name of " << kindName << " #" << i
-             << " must start with either a letter or an underscore";
-    if (llvm::any_of(nameRef,
-                     [](char c) { return !llvm::isAlnum(c) && c != '_'; }))
-      return op->emitOpError()
-             << "name of " << kindName << " #" << i
-             << " must contain only letters, digits and underscores";
+
+    if (failed(isValidName(nameRef, op, Twine(kindName) + " #" + Twine(i))))
+      return failure();
+
     if (nameMap.contains(nameRef))
       return op->emitOpError() << "name of " << kindName << " #" << i
                                << " is a duplicate of the name of " << kindName

mlir/lib/Target/CMakeLists.txt

@@ -1,4 +1,5 @@
 add_subdirectory(Cpp)
+add_subdirectory(IRDLToCpp)
 add_subdirectory(SPIRV)
 add_subdirectory(LLVMIR)
 add_subdirectory(LLVM)

mlir/lib/Target/IRDLToCpp/CMakeLists.txt

@@ -0,0 +1,18 @@
+add_mlir_translation_library(MLIRTargetIRDLToCpp
+  TranslationRegistration.cpp
+  IRDLToCpp.cpp
+  Templates/DialectDecl.txt
+  Templates/DialectDef.txt
+  Templates/Header.txt
+  Templates/PerOperationDecl.txt
+  Templates/PerOperationDef.txt
+  Templates/TypeDecl.txt
+  Templates/TypeDef.txt
+  Templates/TypeHeaderDecl.txt
+  Templates/TypeHeaderDef.txt
+
+  LINK_LIBS PUBLIC
+  MLIRIR
+  MLIRIRDL
+  MLIRTranslateLib
+)