llvm-project/mlir/lib/Transforms/Mem2Reg.cpp

//===- Mem2Reg.cpp - Promotes memory slots into values ----------*- 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
//
//===----------------------------------------------------------------------===//

#include "mlir/Transforms/Mem2Reg.h"
#include "mlir/Analysis/SliceAnalysis.h"
#include "mlir/IR/Builders.h"
#include "mlir/IR/Dominance.h"
#include "mlir/Interfaces/ControlFlowInterfaces.h"
#include "mlir/Interfaces/MemorySlotInterfaces.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
#include "mlir/Transforms/Passes.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/GenericIteratedDominanceFrontier.h"

namespace mlir {
#define GEN_PASS_DEF_MEM2REG
#include "mlir/Transforms/Passes.h.inc"
} // namespace mlir

#define DEBUG_TYPE "mem2reg"

using namespace mlir;

/// mem2reg
///
/// This pass turns unnecessary uses of automatically allocated memory slots
/// into direct Value-based operations. For example, it will simplify storing a
/// constant in a memory slot to immediately load it to a direct use of that
/// constant. In other words, given a memory slot addressed by a non-aliased
/// "pointer" Value, mem2reg removes all the uses of that pointer.
///
/// Within a block, this is done by following the chain of stores and loads of
/// the slot and replacing the results of loads with the values previously
/// stored. If a load happens before any other store, a poison value is used
/// instead.
///
/// Control flow can create situations where a load could be replaced by
/// multiple possible stores depending on the control flow path taken. As a
/// result, this pass must introduce new block arguments in some blocks to
/// accomodate for the multiple possible definitions. Each predecessor will
/// populate the block argument with the definition reached at its end. With
/// this, the value stored can be well defined at block boundaries, allowing
/// the propagation of replacement through blocks.
///
/// This pass computes this transformation in four main steps. The two first
/// steps are performed during an analysis phase that does not mutate IR.
///
/// The two steps of the analysis phase are the following:
/// - A first step computes the list of operations that transitively use the
/// memory slot we would like to promote. The purpose of this phase is to
/// identify which uses must be removed to promote the slot, either by rewiring
/// the user or deleting it. Naturally, direct uses of the slot must be removed.
/// Sometimes additional uses must also be removed: this is notably the case
/// when a direct user of the slot cannot rewire its use and must delete itself,
/// and thus must make its users no longer use it. If any of those uses cannot
/// be removed by their users in any way, promotion cannot continue: this is
/// decided at this step.
/// - A second step computes the list of blocks where a block argument will be
/// needed ("merge points") without mutating the IR. These blocks are the blocks
/// leading to a definition clash between two predecessors. Such blocks happen
/// to be the Iterated Dominance Frontier (IDF) of the set of blocks containing
/// a store, as they represent the point where a clear defining dominator stops
/// existing. Computing this information in advance allows making sure the
/// terminators that will forward values are capable of doing so (inability to
/// do so aborts promotion at this step).
///
/// At this point, promotion is guaranteed to happen, and the mutation phase can
/// begin with the following steps:
/// - A third step computes the reaching definition of the memory slot at each
/// blocking user. This is the core of the mem2reg algorithm, also known as
/// load-store forwarding. This analyses loads and stores and propagates which
/// value must be stored in the slot at each blocking user.  This is achieved by
/// doing a depth-first walk of the dominator tree of the function. This is
/// sufficient because the reaching definition at the beginning of a block is
/// either its new block argument if it is a merge block, or the definition
/// reaching the end of its immediate dominator (parent in the dominator tree).
/// We can therefore propagate this information down the dominator tree to
/// proceed with renaming within blocks.
/// - The final fourth step uses the reaching definition to remove blocking uses
/// in topological order.
///
/// For further reading, chapter three of SSA-based Compiler Design [1]
/// showcases SSA construction, where mem2reg is an adaptation of the same
/// process.
///
/// [1]: Rastello F. & Bouchez Tichadou F., SSA-based Compiler Design (2022),
///      Springer.

MemorySlotPromoter::MemorySlotPromoter(
    MemorySlot slot, PromotableAllocationOpInterface allocator,
    OpBuilder &builder, DominanceInfo &dominance, MemorySlotPromotionInfo info)
    : slot(slot), allocator(allocator), builder(builder), dominance(dominance),
      info(std::move(info)) {
#ifndef NDEBUG
  auto isResultOrNewBlockArgument = [&]() {
    if (BlockArgument arg = slot.ptr.dyn_cast<BlockArgument>())
      return arg.getOwner()->getParentOp() == allocator;
    return slot.ptr.getDefiningOp() == allocator;
  };

  assert(isResultOrNewBlockArgument() &&
         "a slot must be a result of the allocator or an argument of the child "
         "regions of the allocator");
#endif // NDEBUG
}

Value MemorySlotPromoter::getLazyDefaultValue() {
  if (defaultValue)
    return defaultValue;

  OpBuilder::InsertionGuard guard(builder);
  builder.setInsertionPointToStart(slot.ptr.getParentBlock());
  return defaultValue = allocator.getDefaultValue(slot, builder);
}

LogicalResult MemorySlotPromotionAnalyzer::computeBlockingUses(
    DenseMap<Operation *, SmallPtrSet<OpOperand *, 4>> &userToBlockingUses) {
  // The promotion of an operation may require the promotion of further
  // operations (typically, removing operations that use an operation that must
  // delete itself). We thus need to start from the use of the slot pointer and
  // propagate further requests through the forward slice.

  // First insert that all immediate users of the slot pointer must no longer
  // use it.
  for (OpOperand &use : slot.ptr.getUses()) {
    SmallPtrSet<OpOperand *, 4> &blockingUses =
        userToBlockingUses.getOrInsertDefault(use.getOwner());
    blockingUses.insert(&use);
  }

  // Then, propagate the requirements for the removal of uses. The
  // topologically-sorted forward slice allows for all blocking uses of an
  // operation to have been computed before it is reached. Operations are
  // traversed in topological order of their uses, starting from the slot
  // pointer.
  SetVector<Operation *> forwardSlice;
  mlir::getForwardSlice(slot.ptr, &forwardSlice);
  for (Operation *user : forwardSlice) {
    // If the next operation has no blocking uses, everything is fine.
    if (!userToBlockingUses.contains(user))
      continue;

    SmallPtrSet<OpOperand *, 4> &blockingUses = userToBlockingUses[user];

    SmallVector<OpOperand *> newBlockingUses;
    // If the operation decides it cannot deal with removing the blocking uses,
    // promotion must fail.
    if (auto promotable = dyn_cast<PromotableOpInterface>(user)) {
      if (!promotable.canUsesBeRemoved(blockingUses, newBlockingUses))
        return failure();
    } else if (auto promotable = dyn_cast<PromotableMemOpInterface>(user)) {
      if (!promotable.canUsesBeRemoved(slot, blockingUses, newBlockingUses))
        return failure();
    } else {
      // An operation that has blocking uses must be promoted. If it is not
      // promotable, promotion must fail.
      return failure();
    }

    // Then, register any new blocking uses for coming operations.
    for (OpOperand *blockingUse : newBlockingUses) {
      assert(llvm::is_contained(user->getResults(), blockingUse->get()));

      SmallPtrSetImpl<OpOperand *> &newUserBlockingUseSet =
          userToBlockingUses.getOrInsertDefault(blockingUse->getOwner());
      newUserBlockingUseSet.insert(blockingUse);
    }
  }

  // Because this pass currently only supports analysing the parent region of
  // the slot pointer, if a promotable memory op that needs promotion is outside
  // of this region, promotion must fail because it will be impossible to
  // provide a valid `reachingDef` for it.
  for (auto &[toPromote, _] : userToBlockingUses)
    if (isa<PromotableMemOpInterface>(toPromote) &&
        toPromote->getParentRegion() != slot.ptr.getParentRegion())
      return failure();

  return success();
}

SmallPtrSet<Block *, 16> MemorySlotPromotionAnalyzer::computeSlotLiveIn(
    SmallPtrSetImpl<Block *> &definingBlocks) {
  SmallPtrSet<Block *, 16> liveIn;

  // The worklist contains blocks in which it is known that the slot value is
  // live-in. The further blocks where this value is live-in will be inferred
  // from these.
  SmallVector<Block *> liveInWorkList;

  // Blocks with a load before any other store to the slot are the starting
  // points of the analysis. The slot value is definitely live-in in those
  // blocks.
  SmallPtrSet<Block *, 16> visited;
  for (Operation *user : slot.ptr.getUsers()) {
    if (visited.contains(user->getBlock()))
      continue;
    visited.insert(user->getBlock());

    for (Operation &op : user->getBlock()->getOperations()) {
      if (auto memOp = dyn_cast<PromotableMemOpInterface>(op)) {
        // If this operation loads the slot, it is loading from it before
        // ever writing to it, so the value is live-in in this block.
        if (memOp.loadsFrom(slot)) {
          liveInWorkList.push_back(user->getBlock());
          break;
        }

        // If we store to the slot, further loads will see that value.
        // Because we did not meet any load before, the value is not live-in.
        if (memOp.getStored(slot))
          break;
      }
    }
  }

  // The information is then propagated to the predecessors until a def site
  // (store) is found.
  while (!liveInWorkList.empty()) {
    Block *liveInBlock = liveInWorkList.pop_back_val();

    if (!liveIn.insert(liveInBlock).second)
      continue;

    // If a predecessor is a defining block, either:
    // - It has a load before its first store, in which case it is live-in but
    // has already been processed in the initialisation step.
    // - It has a store before any load, in which case it is not live-in.
    // We can thus at this stage insert to the worklist only predecessors that
    // are not defining blocks.
    for (Block *pred : liveInBlock->getPredecessors())
      if (!definingBlocks.contains(pred))
        liveInWorkList.push_back(pred);
  }

  return liveIn;
}

using IDFCalculator = llvm::IDFCalculatorBase<Block, false>;
void MemorySlotPromotionAnalyzer::computeMergePoints(
    SmallPtrSetImpl<Block *> &mergePoints) {
  if (slot.ptr.getParentRegion()->hasOneBlock())
    return;

  IDFCalculator idfCalculator(dominance.getDomTree(slot.ptr.getParentRegion()));

  SmallPtrSet<Block *, 16> definingBlocks;
  for (Operation *user : slot.ptr.getUsers())
    if (auto storeOp = dyn_cast<PromotableMemOpInterface>(user))
      if (storeOp.getStored(slot))
        definingBlocks.insert(user->getBlock());

  idfCalculator.setDefiningBlocks(definingBlocks);

  SmallPtrSet<Block *, 16> liveIn = computeSlotLiveIn(definingBlocks);
  idfCalculator.setLiveInBlocks(liveIn);

  SmallVector<Block *> mergePointsVec;
  idfCalculator.calculate(mergePointsVec);

  mergePoints.insert(mergePointsVec.begin(), mergePointsVec.end());
}

bool MemorySlotPromotionAnalyzer::areMergePointsUsable(
    SmallPtrSetImpl<Block *> &mergePoints) {
  for (Block *mergePoint : mergePoints)
    for (Block *pred : mergePoint->getPredecessors())
      if (!isa<BranchOpInterface>(pred->getTerminator()))
        return false;

  return true;
}

std::optional<MemorySlotPromotionInfo>
MemorySlotPromotionAnalyzer::computeInfo() {
  MemorySlotPromotionInfo info;

  // First, find the set of operations that will need to be changed for the
  // promotion to happen. These operations need to resolve some of their uses,
  // either by rewiring them or simply deleting themselves. If any of them
  // cannot find a way to resolve their blocking uses, we abort the promotion.
  if (failed(computeBlockingUses(info.userToBlockingUses)))
    return {};

  // Then, compute blocks in which two or more definitions of the allocated
  // variable may conflict. These blocks will need a new block argument to
  // accomodate this.
  computeMergePoints(info.mergePoints);

  // The slot can be promoted if the block arguments to be created can
  // actually be populated with values, which may not be possible depending
  // on their predecessors.
  if (!areMergePointsUsable(info.mergePoints))
    return {};

  return info;
}

Value MemorySlotPromoter::computeReachingDefInBlock(Block *block,
                                                    Value reachingDef) {
  for (Operation &op : block->getOperations()) {
    if (auto memOp = dyn_cast<PromotableMemOpInterface>(op)) {
      if (info.userToBlockingUses.contains(memOp))
        reachingDefs.insert({memOp, reachingDef});

      if (Value stored = memOp.getStored(slot))
        reachingDef = stored;
    }
  }

  return reachingDef;
}

void MemorySlotPromoter::computeReachingDefInRegion(Region *region,
                                                    Value reachingDef) {
  if (region->hasOneBlock()) {
    computeReachingDefInBlock(&region->front(), reachingDef);
    return;
  }

  struct DfsJob {
    llvm::DomTreeNodeBase<Block> *block;
    Value reachingDef;
  };

  SmallVector<DfsJob> dfsStack;

  auto &domTree = dominance.getDomTree(slot.ptr.getParentRegion());

  dfsStack.emplace_back<DfsJob>(
      {domTree.getNode(&region->front()), reachingDef});

  while (!dfsStack.empty()) {
    DfsJob job = dfsStack.pop_back_val();
    Block *block = job.block->getBlock();

    if (info.mergePoints.contains(block)) {
      BlockArgument blockArgument =
          block->addArgument(slot.elemType, slot.ptr.getLoc());
      builder.setInsertionPointToStart(block);
      allocator.handleBlockArgument(slot, blockArgument, builder);
      job.reachingDef = blockArgument;
    }

    job.reachingDef = computeReachingDefInBlock(block, job.reachingDef);

    if (auto terminator = dyn_cast<BranchOpInterface>(block->getTerminator())) {
      for (BlockOperand &blockOperand : terminator->getBlockOperands()) {
        if (info.mergePoints.contains(blockOperand.get())) {
          if (!job.reachingDef)
            job.reachingDef = getLazyDefaultValue();
          terminator.getSuccessorOperands(blockOperand.getOperandNumber())
              .append(job.reachingDef);
        }
      }
    }

    for (auto *child : job.block->children())
      dfsStack.emplace_back<DfsJob>({child, job.reachingDef});
  }
}

void MemorySlotPromoter::removeBlockingUses() {
  llvm::SetVector<Operation *> usersToRemoveUses;
  for (auto &user : llvm::make_first_range(info.userToBlockingUses))
    usersToRemoveUses.insert(user);
  SetVector<Operation *> sortedUsersToRemoveUses =
      mlir::topologicalSort(usersToRemoveUses);

  llvm::SmallVector<Operation *> toErase;
  for (Operation *toPromote : llvm::reverse(sortedUsersToRemoveUses)) {
    if (auto toPromoteMemOp = dyn_cast<PromotableMemOpInterface>(toPromote)) {
      Value reachingDef = reachingDefs.lookup(toPromoteMemOp);
      // If no reaching definition is known, this use is outside the reach of
      // the slot. The default value should thus be used.
      if (!reachingDef)
        reachingDef = getLazyDefaultValue();

      builder.setInsertionPointAfter(toPromote);
      if (toPromoteMemOp.removeBlockingUses(
              slot, info.userToBlockingUses[toPromote], builder, reachingDef) ==
          DeletionKind::Delete)
        toErase.push_back(toPromote);

      continue;
    }

    auto toPromoteBasic = cast<PromotableOpInterface>(toPromote);
    builder.setInsertionPointAfter(toPromote);
    if (toPromoteBasic.removeBlockingUses(info.userToBlockingUses[toPromote],
                                          builder) == DeletionKind::Delete)
      toErase.push_back(toPromote);
  }

  for (Operation *toEraseOp : toErase)
    toEraseOp->erase();

  assert(slot.ptr.use_empty() &&
         "after promotion, the slot pointer should not be used anymore");
}

void MemorySlotPromoter::promoteSlot() {
  computeReachingDefInRegion(slot.ptr.getParentRegion(), {});

  // Now that reaching definitions are known, remove all users.
  removeBlockingUses();

  // Update terminators in dead branches to forward default if they are
  // succeeded by a merge points.
  for (Block *mergePoint : info.mergePoints) {
    for (BlockOperand &use : mergePoint->getUses()) {
      auto user = cast<BranchOpInterface>(use.getOwner());
      SuccessorOperands succOperands =
          user.getSuccessorOperands(use.getOperandNumber());
      assert(succOperands.size() == mergePoint->getNumArguments() ||
             succOperands.size() + 1 == mergePoint->getNumArguments());
      if (succOperands.size() + 1 == mergePoint->getNumArguments())
        succOperands.append(getLazyDefaultValue());
    }
  }

  LLVM_DEBUG(llvm::dbgs() << "[mem2reg] Promoted memory slot: " << slot.ptr
                          << "\n");

  allocator.handlePromotionComplete(slot, defaultValue);
}

LogicalResult mlir::tryToPromoteMemorySlots(
    ArrayRef<PromotableAllocationOpInterface> allocators, OpBuilder &builder,
    DominanceInfo &dominance) {
  // Actual promotion may invalidate the dominance analysis, so slot promotion
  // is prepated in batches.
  SmallVector<MemorySlotPromoter> toPromote;
  for (PromotableAllocationOpInterface allocator : allocators) {
    for (MemorySlot slot : allocator.getPromotableSlots()) {
      if (slot.ptr.use_empty())
        continue;

      MemorySlotPromotionAnalyzer analyzer(slot, dominance);
      std::optional<MemorySlotPromotionInfo> info = analyzer.computeInfo();
      if (info)
        toPromote.emplace_back(slot, allocator, builder, dominance,
                               std::move(*info));
    }
  }

  for (MemorySlotPromoter &promoter : toPromote)
    promoter.promoteSlot();

  return success(!toPromote.empty());
}

LogicalResult Mem2RegPattern::matchAndRewrite(Operation *op,
                                              PatternRewriter &rewriter) const {
  hasBoundedRewriteRecursion();

  if (op->getNumRegions() == 0)
    return failure();

  DominanceInfo dominance;

  SmallVector<PromotableAllocationOpInterface> allocators;
  // Build a list of allocators to attempt to promote the slots of.
  for (Region &region : op->getRegions())
    for (auto allocator : region.getOps<PromotableAllocationOpInterface>())
      allocators.emplace_back(allocator);

  // Because pattern rewriters are normally not expressive enough to support a
  // transformation like mem2reg, this uses an escape hatch to mark modified
  // operations manually and operate outside of its context.
  rewriter.startRootUpdate(op);

  OpBuilder builder(rewriter.getContext());

  if (failed(tryToPromoteMemorySlots(allocators, builder, dominance))) {
    rewriter.cancelRootUpdate(op);
    return failure();
  }

  rewriter.finalizeRootUpdate(op);
  return success();
}

namespace {

struct Mem2Reg : impl::Mem2RegBase<Mem2Reg> {
  void runOnOperation() override {
    Operation *scopeOp = getOperation();
    bool changed = false;

    RewritePatternSet rewritePatterns(&getContext());
    rewritePatterns.add<Mem2RegPattern>(&getContext());
    FrozenRewritePatternSet frozen(std::move(rewritePatterns));
    (void)applyOpPatternsAndFold({scopeOp}, frozen, GreedyRewriteConfig(),
                                 &changed);

    if (!changed)
      markAllAnalysesPreserved();
  }
};

} // namespace