The greedy rewriter is used in many different flows and it has a lot of
convenience (work list management, debugging actions, tracing, etc). But
it combines two kinds of greedy behavior 1) how ops are matched, 2)
folding wherever it can.
These are independent forms of greedy and leads to inefficiency. E.g.,
cases where one need to create different phases in lowering and is
required to applying patterns in specific order split across different
passes. Using the driver one ends up needlessly retrying folding/having
multiple rounds of folding attempts, where one final run would have
sufficed.
Of course folks can locally avoid this behavior by just building their
own, but this is also a common requested feature that folks keep on
working around locally in suboptimal ways.
For downstream users, there should be no behavioral change. Updating
from the deprecated should just be a find and replace (e.g., `find ./
-type f -exec sed -i
's|applyPatternsAndFoldGreedily|applyPatternsGreedily|g' {} \;` variety)
as the API arguments hasn't changed between the two.
185 lines
6.7 KiB
C++
185 lines
6.7 KiB
C++
//===- ConstantArgumentGlobalisation.cpp ----------------------------------===//
|
|
//
|
|
// 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
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#include "flang/Optimizer/Builder/FIRBuilder.h"
|
|
#include "flang/Optimizer/Dialect/FIRDialect.h"
|
|
#include "flang/Optimizer/Dialect/FIROps.h"
|
|
#include "flang/Optimizer/Dialect/FIRType.h"
|
|
#include "flang/Optimizer/Transforms/Passes.h"
|
|
#include "mlir/Dialect/Func/IR/FuncOps.h"
|
|
#include "mlir/IR/Diagnostics.h"
|
|
#include "mlir/IR/Dominance.h"
|
|
#include "mlir/Pass/Pass.h"
|
|
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
|
|
|
|
namespace fir {
|
|
#define GEN_PASS_DEF_CONSTANTARGUMENTGLOBALISATIONOPT
|
|
#include "flang/Optimizer/Transforms/Passes.h.inc"
|
|
} // namespace fir
|
|
|
|
#define DEBUG_TYPE "flang-constant-argument-globalisation-opt"
|
|
|
|
namespace {
|
|
unsigned uniqueLitId = 1;
|
|
|
|
class CallOpRewriter : public mlir::OpRewritePattern<fir::CallOp> {
|
|
protected:
|
|
const mlir::DominanceInfo &di;
|
|
|
|
public:
|
|
using OpRewritePattern::OpRewritePattern;
|
|
|
|
CallOpRewriter(mlir::MLIRContext *ctx, const mlir::DominanceInfo &_di)
|
|
: OpRewritePattern(ctx), di(_di) {}
|
|
|
|
llvm::LogicalResult
|
|
matchAndRewrite(fir::CallOp callOp,
|
|
mlir::PatternRewriter &rewriter) const override {
|
|
LLVM_DEBUG(llvm::dbgs() << "Processing call op: " << callOp << "\n");
|
|
auto module = callOp->getParentOfType<mlir::ModuleOp>();
|
|
bool needUpdate = false;
|
|
fir::FirOpBuilder builder(rewriter, module);
|
|
llvm::SmallVector<mlir::Value> newOperands;
|
|
llvm::SmallVector<std::pair<mlir::Operation *, mlir::Operation *>> allocas;
|
|
for (const mlir::Value &a : callOp.getArgs()) {
|
|
auto alloca = mlir::dyn_cast_or_null<fir::AllocaOp>(a.getDefiningOp());
|
|
// We can convert arguments that are alloca, and that has
|
|
// the value by reference attribute. All else is just added
|
|
// to the argument list.
|
|
if (!alloca || !alloca->hasAttr(fir::getAdaptToByRefAttrName())) {
|
|
newOperands.push_back(a);
|
|
continue;
|
|
}
|
|
|
|
mlir::Type varTy = alloca.getInType();
|
|
assert(!fir::hasDynamicSize(varTy) &&
|
|
"only expect statically sized scalars to be by value");
|
|
|
|
// Find immediate store with const argument
|
|
mlir::Operation *store = nullptr;
|
|
for (mlir::Operation *s : alloca->getUsers()) {
|
|
if (mlir::isa<fir::StoreOp>(s) && di.dominates(s, callOp)) {
|
|
// We can only deal with ONE store - if already found one,
|
|
// set to nullptr and exit the loop.
|
|
if (store) {
|
|
store = nullptr;
|
|
break;
|
|
}
|
|
store = s;
|
|
}
|
|
}
|
|
|
|
// If we didn't find any store, or multiple stores, add argument as is
|
|
// and move on.
|
|
if (!store) {
|
|
newOperands.push_back(a);
|
|
continue;
|
|
}
|
|
|
|
LLVM_DEBUG(llvm::dbgs() << " found store " << *store << "\n");
|
|
|
|
mlir::Operation *definingOp = store->getOperand(0).getDefiningOp();
|
|
// If not a constant, add to operands and move on.
|
|
if (!mlir::isa<mlir::arith::ConstantOp>(definingOp)) {
|
|
// Unable to remove alloca arg
|
|
newOperands.push_back(a);
|
|
continue;
|
|
}
|
|
|
|
LLVM_DEBUG(llvm::dbgs() << " found define " << *definingOp << "\n");
|
|
|
|
std::string globalName =
|
|
"_global_const_." + std::to_string(uniqueLitId++);
|
|
assert(!builder.getNamedGlobal(globalName) &&
|
|
"We should have a unique name here");
|
|
|
|
if (llvm::none_of(allocas,
|
|
[alloca](auto x) { return x.first == alloca; })) {
|
|
allocas.push_back(std::make_pair(alloca, store));
|
|
}
|
|
|
|
auto loc = callOp.getLoc();
|
|
fir::GlobalOp global = builder.createGlobalConstant(
|
|
loc, varTy, globalName,
|
|
[&](fir::FirOpBuilder &builder) {
|
|
mlir::Operation *cln = definingOp->clone();
|
|
builder.insert(cln);
|
|
mlir::Value val =
|
|
builder.createConvert(loc, varTy, cln->getResult(0));
|
|
builder.create<fir::HasValueOp>(loc, val);
|
|
},
|
|
builder.createInternalLinkage());
|
|
mlir::Value addr = builder.create<fir::AddrOfOp>(loc, global.resultType(),
|
|
global.getSymbol());
|
|
newOperands.push_back(addr);
|
|
needUpdate = true;
|
|
}
|
|
|
|
if (needUpdate) {
|
|
auto loc = callOp.getLoc();
|
|
llvm::SmallVector<mlir::Type> newResultTypes;
|
|
newResultTypes.append(callOp.getResultTypes().begin(),
|
|
callOp.getResultTypes().end());
|
|
fir::CallOp newOp = builder.create<fir::CallOp>(
|
|
loc,
|
|
callOp.getCallee().has_value() ? callOp.getCallee().value()
|
|
: mlir::SymbolRefAttr{},
|
|
newResultTypes, newOperands);
|
|
// Copy all the attributes from the old to new op.
|
|
newOp->setAttrs(callOp->getAttrs());
|
|
rewriter.replaceOp(callOp, newOp);
|
|
|
|
for (auto a : allocas) {
|
|
if (a.first->hasOneUse()) {
|
|
// If the alloca is only used for a store and the call operand, the
|
|
// store is no longer required.
|
|
rewriter.eraseOp(a.second);
|
|
rewriter.eraseOp(a.first);
|
|
}
|
|
}
|
|
LLVM_DEBUG(llvm::dbgs() << "global constant for " << callOp << " as "
|
|
<< newOp << '\n');
|
|
return mlir::success();
|
|
}
|
|
|
|
// Failure here just means "we couldn't do the conversion", which is
|
|
// perfectly acceptable to the upper layers of this function.
|
|
return mlir::failure();
|
|
}
|
|
};
|
|
|
|
// this pass attempts to convert immediate scalar literals in function calls
|
|
// to global constants to allow transformations such as Dead Argument
|
|
// Elimination
|
|
class ConstantArgumentGlobalisationOpt
|
|
: public fir::impl::ConstantArgumentGlobalisationOptBase<
|
|
ConstantArgumentGlobalisationOpt> {
|
|
public:
|
|
ConstantArgumentGlobalisationOpt() = default;
|
|
|
|
void runOnOperation() override {
|
|
mlir::ModuleOp mod = getOperation();
|
|
mlir::DominanceInfo *di = &getAnalysis<mlir::DominanceInfo>();
|
|
auto *context = &getContext();
|
|
mlir::RewritePatternSet patterns(context);
|
|
mlir::GreedyRewriteConfig config;
|
|
config.enableRegionSimplification =
|
|
mlir::GreedySimplifyRegionLevel::Disabled;
|
|
config.strictMode = mlir::GreedyRewriteStrictness::ExistingOps;
|
|
|
|
patterns.insert<CallOpRewriter>(context, *di);
|
|
if (mlir::failed(
|
|
mlir::applyPatternsGreedily(mod, std::move(patterns), config))) {
|
|
mlir::emitError(mod.getLoc(),
|
|
"error in constant globalisation optimization\n");
|
|
signalPassFailure();
|
|
}
|
|
}
|
|
};
|
|
} // namespace
|