9170fa5808
Using MLIR attributes instead of metadata has many advantages: * No indirection: Attributes can simply refer to each other seemlessly without having to use the indirection of `SymbolRefAttr`. This also gives us correctness by construction in a lot of places as well * Multithreading safe: The Attribute infrastructure gives us thread-safety for free. Creating operations and inserting them into a block is not thread-safe. This is a major use case for e.g. the inliner in MLIR which runs in parallel * Easier to create: There is no need for a builder or a metadata region This patch therefore does exactly that. It leverages the new distinct attributes to create distinct access groups in a deterministic and threadsafe manner. Differential Revision: https://reviews.llvm.org/D155285
75 lines
2.8 KiB
C++
75 lines
2.8 KiB
C++
//===- LoopAnnotationTranslation.h ------------------------------*- 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 implements the translation between an MLIR loop annotations and
|
|
// the corresponding LLVMIR metadata representation.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#ifndef MLIR_LIB_TARGET_LLVMIR_LOOPANNOTATIONTRANSLATION_H_
|
|
#define MLIR_LIB_TARGET_LLVMIR_LOOPANNOTATIONTRANSLATION_H_
|
|
|
|
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
|
|
#include "mlir/Target/LLVMIR/ModuleTranslation.h"
|
|
|
|
namespace mlir {
|
|
namespace LLVM {
|
|
namespace detail {
|
|
|
|
/// A helper class that converts LoopAnnotationAttrs and AccessGroupMetadataOps
|
|
/// into a corresponding llvm::MDNodes.
|
|
class LoopAnnotationTranslation {
|
|
public:
|
|
LoopAnnotationTranslation(ModuleTranslation &moduleTranslation,
|
|
llvm::Module &llvmModule)
|
|
: moduleTranslation(moduleTranslation), llvmModule(llvmModule) {}
|
|
|
|
llvm::MDNode *translateLoopAnnotation(LoopAnnotationAttr attr, Operation *op);
|
|
|
|
/// Returns the LLVM metadata corresponding to an mlir LLVM dialect access
|
|
/// group attribute.
|
|
llvm::MDNode *getAccessGroup(AccessGroupAttr accessGroupAttr);
|
|
|
|
/// Returns the LLVM metadata corresponding to the access group attribute
|
|
/// referenced by the AccessGroupOpInterface or null if there are none.
|
|
llvm::MDNode *getAccessGroups(AccessGroupOpInterface op);
|
|
|
|
/// The ModuleTranslation owning this instance.
|
|
ModuleTranslation &moduleTranslation;
|
|
|
|
private:
|
|
/// Returns the LLVM metadata corresponding to a llvm loop metadata attribute.
|
|
llvm::MDNode *lookupLoopMetadata(Attribute options) const {
|
|
return loopMetadataMapping.lookup(options);
|
|
}
|
|
|
|
void mapLoopMetadata(Attribute options, llvm::MDNode *metadata) {
|
|
auto result = loopMetadataMapping.try_emplace(options, metadata);
|
|
(void)result;
|
|
assert(result.second &&
|
|
"attempting to map loop options that was already mapped");
|
|
}
|
|
|
|
/// Mapping from an attribute describing loop metadata to its LLVM metadata.
|
|
/// The metadata is attached to Latch block branches with this attribute.
|
|
DenseMap<Attribute, llvm::MDNode *> loopMetadataMapping;
|
|
|
|
/// Mapping from an access group attribute to its LLVM metadata.
|
|
/// This map is populated on module entry and is used to annotate loops (as
|
|
/// identified via their branches) and contained memory accesses.
|
|
DenseMap<AccessGroupAttr, llvm::MDNode *> accessGroupMetadataMapping;
|
|
|
|
llvm::Module &llvmModule;
|
|
};
|
|
|
|
} // namespace detail
|
|
} // namespace LLVM
|
|
} // namespace mlir
|
|
|
|
#endif // MLIR_LIB_TARGET_LLVMIR_LOOPANNOTATIONTRANSLATION_H_
|