llvm-project/mlir/lib/Transforms/Utils/LoopUtils.cpp

459 lines
18 KiB
C++

//===- LoopUtils.cpp ---- Misc utilities for loop transformation ----------===//
//
// Copyright 2019 The MLIR Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// =============================================================================
//
// This file implements miscellaneous loop transformation routines.
//
//===----------------------------------------------------------------------===//
#include "mlir/Transforms/LoopUtils.h"
#include "mlir/AffineOps/AffineOps.h"
#include "mlir/Analysis/LoopAnalysis.h"
#include "mlir/IR/AffineExpr.h"
#include "mlir/IR/AffineMap.h"
#include "mlir/IR/BlockAndValueMapping.h"
#include "mlir/IR/Builders.h"
#include "mlir/IR/BuiltinOps.h"
#include "mlir/IR/InstVisitor.h"
#include "mlir/IR/Instruction.h"
#include "mlir/StandardOps/StandardOps.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/Support/Debug.h"
#define DEBUG_TYPE "LoopUtils"
using namespace mlir;
/// Returns the upper bound of an unrolled loop with lower bound 'lb' and with
/// the specified trip count, stride, and unroll factor. Returns nullptr when
/// the trip count can't be expressed as an affine expression.
AffineMap mlir::getUnrolledLoopUpperBound(ConstOpPointer<AffineForOp> forOp,
unsigned unrollFactor,
FuncBuilder *builder) {
auto lbMap = forOp->getLowerBoundMap();
// Single result lower bound map only.
if (lbMap.getNumResults() != 1)
return AffineMap();
// Sometimes, the trip count cannot be expressed as an affine expression.
auto tripCount = getTripCountExpr(forOp);
if (!tripCount)
return AffineMap();
AffineExpr lb(lbMap.getResult(0));
unsigned step = forOp->getStep();
auto newUb = lb + (tripCount - tripCount % unrollFactor - 1) * step;
return builder->getAffineMap(lbMap.getNumDims(), lbMap.getNumSymbols(),
{newUb}, {});
}
/// Returns the lower bound of the cleanup loop when unrolling a loop with lower
/// bound 'lb' and with the specified trip count, stride, and unroll factor.
/// Returns an AffinMap with nullptr storage (that evaluates to false)
/// when the trip count can't be expressed as an affine expression.
AffineMap mlir::getCleanupLoopLowerBound(ConstOpPointer<AffineForOp> forOp,
unsigned unrollFactor,
FuncBuilder *builder) {
auto lbMap = forOp->getLowerBoundMap();
// Single result lower bound map only.
if (lbMap.getNumResults() != 1)
return AffineMap();
// Sometimes the trip count cannot be expressed as an affine expression.
AffineExpr tripCount(getTripCountExpr(forOp));
if (!tripCount)
return AffineMap();
AffineExpr lb(lbMap.getResult(0));
unsigned step = forOp->getStep();
auto newLb = lb + (tripCount - tripCount % unrollFactor) * step;
return builder->getAffineMap(lbMap.getNumDims(), lbMap.getNumSymbols(),
{newLb}, {});
}
/// Promotes the loop body of a forOp to its containing block if the forOp
/// was known to have a single iteration. Returns false otherwise.
// TODO(bondhugula): extend this for arbitrary affine bounds.
bool mlir::promoteIfSingleIteration(OpPointer<AffineForOp> forOp) {
Optional<uint64_t> tripCount = getConstantTripCount(forOp);
if (!tripCount.hasValue() || tripCount.getValue() != 1)
return false;
// TODO(mlir-team): there is no builder for a max.
if (forOp->getLowerBoundMap().getNumResults() != 1)
return false;
// Replaces all IV uses to its single iteration value.
auto *iv = forOp->getInductionVar();
OperationInst *forInst = forOp->getInstruction();
if (!iv->use_empty()) {
if (forOp->hasConstantLowerBound()) {
auto *mlFunc = forInst->getFunction();
FuncBuilder topBuilder(mlFunc);
auto constOp = topBuilder.create<ConstantIndexOp>(
forOp->getLoc(), forOp->getConstantLowerBound());
iv->replaceAllUsesWith(constOp);
} else {
const AffineBound lb = forOp->getLowerBound();
SmallVector<Value *, 4> lbOperands(lb.operand_begin(), lb.operand_end());
FuncBuilder builder(forInst->getBlock(), Block::iterator(forInst));
if (lb.getMap() == builder.getDimIdentityMap()) {
// No need of generating an affine_apply.
iv->replaceAllUsesWith(lbOperands[0]);
} else {
auto affineApplyOp = builder.create<AffineApplyOp>(
forInst->getLoc(), lb.getMap(), lbOperands);
iv->replaceAllUsesWith(affineApplyOp);
}
}
}
// Move the loop body instructions to the loop's containing block.
auto *block = forInst->getBlock();
block->getInstructions().splice(Block::iterator(forInst),
forOp->getBody()->getInstructions());
forOp->erase();
return true;
}
/// Promotes all single iteration for inst's in the Function, i.e., moves
/// their body into the containing Block.
void mlir::promoteSingleIterationLoops(Function *f) {
// Gathers all innermost loops through a post order pruned walk.
f->walkPostOrder([](OperationInst *inst) {
if (auto forOp = inst->dyn_cast<AffineForOp>())
promoteIfSingleIteration(forOp);
});
}
/// Generates a 'for' inst with the specified lower and upper bounds while
/// generating the right IV remappings for the shifted instructions. The
/// instruction blocks that go into the loop are specified in instGroupQueue
/// starting from the specified offset, and in that order; the first element of
/// the pair specifies the shift applied to that group of instructions; note
/// that the shift is multiplied by the loop step before being applied. Returns
/// nullptr if the generated loop simplifies to a single iteration one.
static OpPointer<AffineForOp>
generateLoop(AffineMap lbMap, AffineMap ubMap,
const std::vector<std::pair<uint64_t, ArrayRef<Instruction *>>>
&instGroupQueue,
unsigned offset, OpPointer<AffineForOp> srcForInst,
FuncBuilder *b) {
SmallVector<Value *, 4> lbOperands(srcForInst->getLowerBoundOperands());
SmallVector<Value *, 4> ubOperands(srcForInst->getUpperBoundOperands());
assert(lbMap.getNumInputs() == lbOperands.size());
assert(ubMap.getNumInputs() == ubOperands.size());
auto loopChunk =
b->create<AffineForOp>(srcForInst->getLoc(), lbOperands, lbMap,
ubOperands, ubMap, srcForInst->getStep());
loopChunk->createBody();
auto *loopChunkIV = loopChunk->getInductionVar();
auto *srcIV = srcForInst->getInductionVar();
BlockAndValueMapping operandMap;
for (auto it = instGroupQueue.begin() + offset, e = instGroupQueue.end();
it != e; ++it) {
uint64_t shift = it->first;
auto insts = it->second;
// All 'same shift' instructions get added with their operands being
// remapped to results of cloned instructions, and their IV used remapped.
// Generate the remapping if the shift is not zero: remappedIV = newIV -
// shift.
if (!srcIV->use_empty() && shift != 0) {
FuncBuilder b(loopChunk->getBody());
auto ivRemap = b.create<AffineApplyOp>(
srcForInst->getLoc(),
b.getSingleDimShiftAffineMap(
-static_cast<int64_t>(srcForInst->getStep() * shift)),
loopChunkIV);
operandMap.map(srcIV, ivRemap);
} else {
operandMap.map(srcIV, loopChunkIV);
}
for (auto *inst : insts) {
loopChunk->getBody()->push_back(inst->clone(operandMap, b->getContext()));
}
}
if (promoteIfSingleIteration(loopChunk))
return OpPointer<AffineForOp>();
return loopChunk;
}
/// Skew the instructions in the body of a 'for' instruction with the specified
/// instruction-wise shifts. The shifts are with respect to the original
/// execution order, and are multiplied by the loop 'step' before being applied.
/// A shift of zero for each instruction will lead to no change.
// The skewing of instructions with respect to one another can be used for
// example to allow overlap of asynchronous operations (such as DMA
// communication) with computation, or just relative shifting of instructions
// for better register reuse, locality or parallelism. As such, the shifts are
// typically expected to be at most of the order of the number of instructions.
// This method should not be used as a substitute for loop distribution/fission.
// This method uses an algorithm// in time linear in the number of instructions
// in the body of the for loop - (using the 'sweep line' paradigm). This method
// asserts preservation of SSA dominance. A check for that as well as that for
// memory-based depedence preservation check rests with the users of this
// method.
UtilResult mlir::instBodySkew(OpPointer<AffineForOp> forOp,
ArrayRef<uint64_t> shifts,
bool unrollPrologueEpilogue) {
if (forOp->getBody()->empty())
return UtilResult::Success;
// If the trip counts aren't constant, we would need versioning and
// conditional guards (or context information to prevent such versioning). The
// better way to pipeline for such loops is to first tile them and extract
// constant trip count "full tiles" before applying this.
auto mayBeConstTripCount = getConstantTripCount(forOp);
if (!mayBeConstTripCount.hasValue()) {
LLVM_DEBUG(llvm::dbgs() << "non-constant trip count loop\n";);
return UtilResult::Success;
}
uint64_t tripCount = mayBeConstTripCount.getValue();
assert(isInstwiseShiftValid(forOp, shifts) &&
"shifts will lead to an invalid transformation\n");
int64_t step = forOp->getStep();
unsigned numChildInsts = forOp->getBody()->getInstructions().size();
// Do a linear time (counting) sort for the shifts.
uint64_t maxShift = 0;
for (unsigned i = 0; i < numChildInsts; i++) {
maxShift = std::max(maxShift, shifts[i]);
}
// Such large shifts are not the typical use case.
if (maxShift >= numChildInsts) {
LLVM_DEBUG(llvm::dbgs() << "inst shifts too large - unexpected\n";);
return UtilResult::Success;
}
// An array of instruction groups sorted by shift amount; each group has all
// instructions with the same shift in the order in which they appear in the
// body of the 'for' inst.
std::vector<std::vector<Instruction *>> sortedInstGroups(maxShift + 1);
unsigned pos = 0;
for (auto &inst : *forOp->getBody()) {
auto shift = shifts[pos++];
sortedInstGroups[shift].push_back(&inst);
}
// Unless the shifts have a specific pattern (which actually would be the
// common use case), prologue and epilogue are not meaningfully defined.
// Nevertheless, if 'unrollPrologueEpilogue' is set, we will treat the first
// loop generated as the prologue and the last as epilogue and unroll these
// fully.
OpPointer<AffineForOp> prologue;
OpPointer<AffineForOp> epilogue;
// Do a sweep over the sorted shifts while storing open groups in a
// vector, and generating loop portions as necessary during the sweep. A block
// of instructions is paired with its shift.
std::vector<std::pair<uint64_t, ArrayRef<Instruction *>>> instGroupQueue;
auto origLbMap = forOp->getLowerBoundMap();
uint64_t lbShift = 0;
FuncBuilder b(forOp->getInstruction());
for (uint64_t d = 0, e = sortedInstGroups.size(); d < e; ++d) {
// If nothing is shifted by d, continue.
if (sortedInstGroups[d].empty())
continue;
if (!instGroupQueue.empty()) {
assert(d >= 1 &&
"Queue expected to be empty when the first block is found");
// The interval for which the loop needs to be generated here is:
// [lbShift, min(lbShift + tripCount, d)) and the body of the
// loop needs to have all instructions in instQueue in that order.
OpPointer<AffineForOp> res;
if (lbShift + tripCount * step < d * step) {
res = generateLoop(
b.getShiftedAffineMap(origLbMap, lbShift),
b.getShiftedAffineMap(origLbMap, lbShift + tripCount * step),
instGroupQueue, 0, forOp, &b);
// Entire loop for the queued inst groups generated, empty it.
instGroupQueue.clear();
lbShift += tripCount * step;
} else {
res = generateLoop(b.getShiftedAffineMap(origLbMap, lbShift),
b.getShiftedAffineMap(origLbMap, d), instGroupQueue,
0, forOp, &b);
lbShift = d * step;
}
if (!prologue && res)
prologue = res;
epilogue = res;
} else {
// Start of first interval.
lbShift = d * step;
}
// Augment the list of instructions that get into the current open interval.
instGroupQueue.push_back({d, sortedInstGroups[d]});
}
// Those instructions groups left in the queue now need to be processed (FIFO)
// and their loops completed.
for (unsigned i = 0, e = instGroupQueue.size(); i < e; ++i) {
uint64_t ubShift = (instGroupQueue[i].first + tripCount) * step;
epilogue = generateLoop(b.getShiftedAffineMap(origLbMap, lbShift),
b.getShiftedAffineMap(origLbMap, ubShift),
instGroupQueue, i, forOp, &b);
lbShift = ubShift;
if (!prologue)
prologue = epilogue;
}
// Erase the original for inst.
forOp->erase();
if (unrollPrologueEpilogue && prologue)
loopUnrollFull(prologue);
if (unrollPrologueEpilogue && !epilogue &&
epilogue->getInstruction() != prologue->getInstruction())
loopUnrollFull(epilogue);
return UtilResult::Success;
}
/// Unrolls this loop completely.
bool mlir::loopUnrollFull(OpPointer<AffineForOp> forOp) {
Optional<uint64_t> mayBeConstantTripCount = getConstantTripCount(forOp);
if (mayBeConstantTripCount.hasValue()) {
uint64_t tripCount = mayBeConstantTripCount.getValue();
if (tripCount == 1) {
return promoteIfSingleIteration(forOp);
}
return loopUnrollByFactor(forOp, tripCount);
}
return false;
}
/// Unrolls and jams this loop by the specified factor or by the trip count (if
/// constant) whichever is lower.
bool mlir::loopUnrollUpToFactor(OpPointer<AffineForOp> forOp,
uint64_t unrollFactor) {
Optional<uint64_t> mayBeConstantTripCount = getConstantTripCount(forOp);
if (mayBeConstantTripCount.hasValue() &&
mayBeConstantTripCount.getValue() < unrollFactor)
return loopUnrollByFactor(forOp, mayBeConstantTripCount.getValue());
return loopUnrollByFactor(forOp, unrollFactor);
}
/// Unrolls this loop by the specified factor. Returns true if the loop
/// is successfully unrolled.
bool mlir::loopUnrollByFactor(OpPointer<AffineForOp> forOp,
uint64_t unrollFactor) {
assert(unrollFactor >= 1 && "unroll factor should be >= 1");
if (unrollFactor == 1)
return promoteIfSingleIteration(forOp);
if (forOp->getBody()->empty())
return false;
auto lbMap = forOp->getLowerBoundMap();
auto ubMap = forOp->getUpperBoundMap();
// Loops with max/min expressions won't be unrolled here (the output can't be
// expressed as a Function in the general case). However, the right way to
// do such unrolling for a Function would be to specialize the loop for the
// 'hotspot' case and unroll that hotspot.
if (lbMap.getNumResults() != 1 || ubMap.getNumResults() != 1)
return false;
// Same operand list for lower and upper bound for now.
// TODO(bondhugula): handle bounds with different operand lists.
if (!forOp->matchingBoundOperandList())
return false;
Optional<uint64_t> mayBeConstantTripCount = getConstantTripCount(forOp);
// If the trip count is lower than the unroll factor, no unrolled body.
// TODO(bondhugula): option to specify cleanup loop unrolling.
if (mayBeConstantTripCount.hasValue() &&
mayBeConstantTripCount.getValue() < unrollFactor)
return false;
// Generate the cleanup loop if trip count isn't a multiple of unrollFactor.
OperationInst *forInst = forOp->getInstruction();
if (getLargestDivisorOfTripCount(forOp) % unrollFactor != 0) {
FuncBuilder builder(forInst->getBlock(), ++Block::iterator(forInst));
auto cleanupForInst =
cast<OperationInst>(builder.clone(*forInst))->cast<AffineForOp>();
auto clLbMap = getCleanupLoopLowerBound(forOp, unrollFactor, &builder);
assert(clLbMap &&
"cleanup loop lower bound map for single result bound maps can "
"always be determined");
cleanupForInst->setLowerBoundMap(clLbMap);
// Promote the loop body up if this has turned into a single iteration loop.
promoteIfSingleIteration(cleanupForInst);
// Adjust upper bound.
auto unrolledUbMap =
getUnrolledLoopUpperBound(forOp, unrollFactor, &builder);
assert(unrolledUbMap &&
"upper bound map can alwayys be determined for an unrolled loop "
"with single result bounds");
forOp->setUpperBoundMap(unrolledUbMap);
}
// Scale the step of loop being unrolled by unroll factor.
int64_t step = forOp->getStep();
forOp->setStep(step * unrollFactor);
// Builder to insert unrolled bodies right after the last instruction in the
// body of 'forOp'.
FuncBuilder builder(forOp->getBody(), forOp->getBody()->end());
// Keep a pointer to the last instruction in the original block so that we
// know what to clone (since we are doing this in-place).
Block::iterator srcBlockEnd = std::prev(forOp->getBody()->end());
// Unroll the contents of 'forOp' (append unrollFactor-1 additional copies).
auto *forOpIV = forOp->getInductionVar();
for (unsigned i = 1; i < unrollFactor; i++) {
BlockAndValueMapping operandMap;
// If the induction variable is used, create a remapping to the value for
// this unrolled instance.
if (!forOpIV->use_empty()) {
// iv' = iv + 1/2/3...unrollFactor-1;
auto d0 = builder.getAffineDimExpr(0);
auto bumpMap = builder.getAffineMap(1, 0, {d0 + i * step}, {});
auto ivUnroll =
builder.create<AffineApplyOp>(forOp->getLoc(), bumpMap, forOpIV);
operandMap.map(forOpIV, ivUnroll);
}
// Clone the original body of 'forOp'.
for (auto it = forOp->getBody()->begin(); it != std::next(srcBlockEnd);
it++) {
builder.clone(*it, operandMap);
}
}
// Promote the loop body up if this has turned into a single iteration loop.
promoteIfSingleIteration(forOp);
return true;
}