llvm-project/clang/lib/CIR/Dialect/Transforms/CIRCanonicalize.cpp
Henrich Lauko 95fdfcca9b
[CIR] Move BrOp block merging to canonicalize method (#178053)
Move the redundant branch removal pattern from CIRCanonicalize pass
to BrOp::canonicalize. This follows  MLIR's convention where
canonicalization logic lives with the operation definition.

Changes:
- Add hasCanonicalizeMethod to BrOp in CIROps.td
- Implement BrOp::canonicalize in CIRDialect.cpp
- Auto-collect canonicalization patterns from all registered ops
  in CIRCanonicalizePass (matching MLIR's Canonicalizer approach)
- Remove the standalone RemoveRedundantBranches pattern class

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-27 08:45:15 +01:00

91 lines
3.3 KiB
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 pass that canonicalizes CIR operations, eliminating
// redundant branches, empty scopes, and other unnecessary operations.
//
//===----------------------------------------------------------------------===//
#include "PassDetail.h"
#include "mlir/Dialect/Func/IR/FuncOps.h"
#include "mlir/IR/Block.h"
#include "mlir/IR/Operation.h"
#include "mlir/IR/PatternMatch.h"
#include "mlir/IR/Region.h"
#include "mlir/Support/LogicalResult.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
#include "clang/CIR/Dialect/IR/CIRDialect.h"
#include "clang/CIR/Dialect/Passes.h"
#include "clang/CIR/MissingFeatures.h"
using namespace mlir;
using namespace cir;
namespace mlir {
#define GEN_PASS_DEF_CIRCANONICALIZE
#include "clang/CIR/Dialect/Passes.h.inc"
} // namespace mlir
namespace {
//===----------------------------------------------------------------------===//
// CIRCanonicalizePass
//===----------------------------------------------------------------------===//
struct CIRCanonicalizePass
: public impl::CIRCanonicalizeBase<CIRCanonicalizePass> {
using CIRCanonicalizeBase::CIRCanonicalizeBase;
// The same operation rewriting done here could have been performed
// by CanonicalizerPass (adding hasCanonicalizer for target Ops and
// implementing the same from above in CIRDialects.cpp). However, it's
// currently too aggressive for static analysis purposes, since it might
// remove things where a diagnostic can be generated.
//
// FIXME: perhaps we can add one more mode to GreedyRewriteConfig to
// disable this behavior.
void runOnOperation() override;
};
void CIRCanonicalizePass::runOnOperation() {
RewritePatternSet patterns(&getContext());
// Collect canonicalization patterns from CIR ops.
mlir::Dialect *cir = getContext().getLoadedDialect<cir::CIRDialect>();
for (mlir::RegisteredOperationName op :
getContext().getRegisteredOperations())
if (&op.getDialect() == cir)
op.getCanonicalizationPatterns(patterns, &getContext());
// Collect operations to apply patterns.
llvm::SmallVector<Operation *, 16> ops;
getOperation()->walk([&](Operation *op) {
assert(!cir::MissingFeatures::tryOp());
assert(!cir::MissingFeatures::callOp());
// Many operations are here to perform a manual `fold` in
// applyOpPatternsGreedily.
if (isa<BrOp, BrCondOp, CastOp, ScopeOp, SwitchOp, SelectOp, UnaryOp,
ComplexCreateOp, ComplexImagOp, ComplexRealOp, VecCmpOp,
VecCreateOp, VecExtractOp, VecShuffleOp, VecShuffleDynamicOp,
VecTernaryOp, BitClrsbOp, BitClzOp, BitCtzOp, BitFfsOp, BitParityOp,
BitPopcountOp, BitReverseOp, ByteSwapOp, RotateOp, ConstantOp>(op))
ops.push_back(op);
});
// Apply patterns.
if (applyOpPatternsGreedily(ops, std::move(patterns)).failed())
signalPassFailure();
}
} // namespace
std::unique_ptr<Pass> mlir::createCIRCanonicalizePass() {
return std::make_unique<CIRCanonicalizePass>();
}