llvm-project/mlir/lib/Conversion/ShapeToStandard/ShapeToStandard.cpp

//===- ShapeToStandard.cpp - conversion from Shape to Standard dialect ----===//
//
// 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/Conversion/ShapeToStandard/ShapeToStandard.h"

#include "../PassDetail.h"
#include "mlir/Dialect/SCF/SCF.h"
#include "mlir/Dialect/Shape/IR/Shape.h"
#include "mlir/Dialect/StandardOps/IR/Ops.h"
#include "mlir/IR/BlockAndValueMapping.h"
#include "mlir/Transforms/DialectConversion.h"

using namespace mlir;
using namespace mlir::shape;
using namespace mlir::scf;

/// Conversion patterns.
namespace {
class AnyOpConversion : public OpConversionPattern<AnyOp> {
public:
  using OpConversionPattern<AnyOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(AnyOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override;
};
} // namespace

LogicalResult
AnyOpConversion::matchAndRewrite(AnyOp op, ArrayRef<Value> operands,
                                 ConversionPatternRewriter &rewriter) const {
  AnyOp::Adaptor transformed(operands);

  // Replace `any` with its first operand.
  // Any operand would be a valid substitution.
  rewriter.replaceOp(op, {transformed.inputs().front()});
  return success();
}

namespace {
template <typename SrcOpTy, typename DstOpTy>
class BinaryOpConversion : public OpConversionPattern<SrcOpTy> {
public:
  using OpConversionPattern<SrcOpTy>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(SrcOpTy op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    typename SrcOpTy::Adaptor transformed(operands);

    // For now, only error-free types are supported by this lowering.
    if (op.getType().template isa<SizeType>())
      return failure();

    rewriter.replaceOpWithNewOp<DstOpTy>(op, transformed.lhs(),
                                         transformed.rhs());
    return success();
  }
};
} // namespace

namespace {
struct BroadcastOpConverter : public OpConversionPattern<BroadcastOp> {
  using OpConversionPattern<BroadcastOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(BroadcastOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override;
};
} // namespace

LogicalResult BroadcastOpConverter::matchAndRewrite(
    BroadcastOp op, ArrayRef<Value> operands,
    ConversionPatternRewriter &rewriter) const {
  // For now, this lowering is only defined on `tensor<?xindex>` operands, not
  // on shapes.
  if (op.getType().isa<ShapeType>())
    return failure();

  assert(!op.lhs().getType().isa<ShapeType>() &&
         !op.rhs().getType().isa<ShapeType>());
  auto loc = op.getLoc();
  BroadcastOp::Adaptor transformed(operands);
  Value zero = rewriter.create<ConstantIndexOp>(loc, 0);
  Value one = rewriter.create<ConstantIndexOp>(loc, 1);

  // Find smaller and greater rank and extent tensor.
  Value lhsRank = rewriter.create<DimOp>(loc, transformed.lhs(), zero);
  Value rhsRank = rewriter.create<DimOp>(loc, transformed.rhs(), zero);
  Value lhsSmaller =
      rewriter.create<CmpIOp>(loc, CmpIPredicate::ule, lhsRank, rhsRank);
  Type indexTy = rewriter.getIndexType();
  Type extentTensorTy = op.getType();
  auto ifOp = rewriter.create<IfOp>(
      loc, TypeRange{indexTy, extentTensorTy, indexTy, extentTensorTy},
      lhsSmaller,
      [&](OpBuilder &b, Location loc) {
        b.create<scf::YieldOp>(loc, ValueRange{lhsRank, transformed.lhs(),
                                               rhsRank, transformed.rhs()});
      },
      [&](OpBuilder &b, Location loc) {
        b.create<scf::YieldOp>(loc, ValueRange{rhsRank, transformed.rhs(),
                                               lhsRank, transformed.lhs()});
      });
  Value smallerRank = ifOp.getResult(0);
  Value smallerOperand = ifOp.getResult(1);
  Value greaterRank = ifOp.getResult(2);
  Value greaterOperand = ifOp.getResult(3);

  // Allocate stack memory for the broadcasted extent tensor.
  Type memTy = MemRefType::get({ShapedType::kDynamicSize}, indexTy);
  Value mem = rewriter.create<AllocaOp>(loc, memTy, ValueRange{greaterRank});

  // Copy extents from greater operand that are not challenged.
  Value rankDiff =
      rewriter.create<SubIOp>(loc, indexTy, greaterRank, smallerRank);
  rewriter.create<ForOp>(loc, zero, rankDiff, one, llvm::None,
                         [&](OpBuilder &b, Location loc, Value iv, ValueRange) {
                           Value extent = b.create<ExtractElementOp>(
                               loc, greaterOperand, ValueRange{iv});
                           b.create<StoreOp>(loc, extent, mem, ValueRange{iv});
                           b.create<scf::YieldOp>(loc);
                         });

  // Determine remaining broadcasted extents.
  rewriter.create<ForOp>(
      loc, rankDiff, greaterRank, one, llvm::None,
      [&](OpBuilder &b, Location loc, Value iv, ValueRange) {
        Value greaterOperandExtent =
            b.create<ExtractElementOp>(loc, greaterOperand, ValueRange{iv});
        Value greaterOperandExtentIsOne =
            b.create<CmpIOp>(loc, CmpIPredicate::eq, greaterOperandExtent, one);
        auto ifOp = b.create<IfOp>(
            loc, TypeRange{indexTy}, greaterOperandExtentIsOne,
            [&](OpBuilder &b, Location loc) {
              Value ivShifted = b.create<SubIOp>(loc, indexTy, iv, rankDiff);
              Value smallerOperandExtent = b.create<ExtractElementOp>(
                  loc, smallerOperand, ValueRange{ivShifted});
              b.create<scf::YieldOp>(loc, smallerOperandExtent);
            },
            [&](OpBuilder &b, Location loc) {
              b.create<scf::YieldOp>(loc, greaterOperandExtent);
            });
        Value extent = ifOp.getResult(0);
        b.create<StoreOp>(loc, extent, mem, ValueRange{iv});
        b.create<scf::YieldOp>(loc);
      });

  // Load broadcasted shape as an extent tensor.
  rewriter.replaceOpWithNewOp<TensorLoadOp>(op, mem);
  return success();
}

namespace {
class ConstShapeOpConverter : public OpConversionPattern<ConstShapeOp> {
public:
  using OpConversionPattern<ConstShapeOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(ConstShapeOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override;
};
} // namespace

LogicalResult ConstShapeOpConverter::matchAndRewrite(
    ConstShapeOp op, ArrayRef<Value> operands,
    ConversionPatternRewriter &rewriter) const {

  // For now, this lowering supports only extent tensors, not `shape.shape`
  // types.
  if (op.getType().isa<ShapeType>())
    return failure();

  auto loc = op.getLoc();
  SmallVector<Value, 4> extentOperands;
  for (auto extent : op.shape()) {
    extentOperands.push_back(
        rewriter.create<ConstantIndexOp>(loc, extent.getLimitedValue()));
  }
  Type indexTy = rewriter.getIndexType();
  Value tensor =
      rewriter.create<TensorFromElementsOp>(loc, indexTy, extentOperands);
  Type resultTy = RankedTensorType::get({ShapedType::kDynamicSize}, indexTy);
  rewriter.replaceOpWithNewOp<TensorCastOp>(op, tensor, resultTy);
  return success();
}

namespace {
class ConstSizeOpConversion : public OpConversionPattern<ConstSizeOp> {
public:
  using OpConversionPattern<ConstSizeOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(ConstSizeOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override;
};
} // namespace

LogicalResult ConstSizeOpConversion::matchAndRewrite(
    ConstSizeOp op, ArrayRef<Value> operands,
    ConversionPatternRewriter &rewriter) const {
  rewriter.replaceOpWithNewOp<ConstantIndexOp>(op, op.value().getSExtValue());
  return success();
}

namespace {
class GetExtentOpConverter : public OpConversionPattern<GetExtentOp> {
  using OpConversionPattern<GetExtentOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(GetExtentOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override;
};
} // namespace

LogicalResult GetExtentOpConverter::matchAndRewrite(
    GetExtentOp op, ArrayRef<Value> operands,
    ConversionPatternRewriter &rewriter) const {
  GetExtentOp::Adaptor transformed(operands);

  // For now, only error-free types are supported by this lowering.
  if (op.getType().isa<SizeType>())
    return failure();

  // Derive shape extent directly from shape origin if possible. This
  // circumvents the necessity to materialize the shape in memory.
  if (auto shapeOfOp = op.shape().getDefiningOp<ShapeOfOp>()) {
    if (shapeOfOp.arg().getType().isa<ShapedType>()) {
      rewriter.replaceOpWithNewOp<DimOp>(op, shapeOfOp.arg(),
                                         transformed.dim());
      return success();
    }
  }

  rewriter.replaceOpWithNewOp<ExtractElementOp>(op, rewriter.getIndexType(),
                                                transformed.shape(),
                                                ValueRange{transformed.dim()});
  return success();
}

namespace {
class RankOpConverter : public OpConversionPattern<shape::RankOp> {
public:
  using OpConversionPattern<shape::RankOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(shape::RankOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override;
};
} // namespace

LogicalResult
RankOpConverter::matchAndRewrite(shape::RankOp op, ArrayRef<Value> operands,
                                 ConversionPatternRewriter &rewriter) const {
  // For now, this lowering supports only error-free types.
  if (op.getType().isa<SizeType>())
    return failure();

  shape::RankOp::Adaptor transformed(operands);
  rewriter.replaceOpWithNewOp<DimOp>(op, transformed.shape(), 0);
  return success();
}

namespace {
/// Converts `shape.reduce` to `scf.for`.
struct ReduceOpConverter : public OpConversionPattern<shape::ReduceOp> {
public:
  using OpConversionPattern::OpConversionPattern;

  LogicalResult
  matchAndRewrite(shape::ReduceOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const final;
};
} // namespace

LogicalResult
ReduceOpConverter::matchAndRewrite(shape::ReduceOp op, ArrayRef<Value> operands,
                                   ConversionPatternRewriter &rewriter) const {
  // For now, this lowering is only defined on `tensor<?xindex>` operands.
  if (op.shape().getType().isa<ShapeType>())
    return failure();

  auto loc = op.getLoc();
  shape::ReduceOp::Adaptor transformed(operands);

  Value zero = rewriter.create<ConstantIndexOp>(loc, 0);
  Value one = rewriter.create<ConstantIndexOp>(loc, 1);
  Type indexTy = rewriter.getIndexType();
  Value rank = rewriter.create<DimOp>(loc, indexTy, transformed.shape(), zero);

  auto loop = rewriter.create<scf::ForOp>(
      loc, zero, rank, one, op.initVals(),
      [&](OpBuilder &b, Location loc, Value iv, ValueRange args) {
        Value extent = b.create<ExtractElementOp>(loc, transformed.shape(), iv);

        SmallVector<Value, 2> mappedValues{iv, extent};
        mappedValues.append(args.begin(), args.end());

        BlockAndValueMapping mapping;
        Block *reduceBody = op.getBody();
        mapping.map(reduceBody->getArguments(), mappedValues);
        for (auto &nested : reduceBody->without_terminator())
          b.clone(nested, mapping);

        SmallVector<Value, 2> mappedResults;
        for (auto result : reduceBody->getTerminator()->getOperands())
          mappedResults.push_back(mapping.lookup(result));
        b.create<scf::YieldOp>(loc, mappedResults);
      });

  rewriter.replaceOp(op, loop.getResults());
  return success();
}

namespace {
/// Converts `shape.shape_eq` to an `scf.for` loop. For now, the lowering is
/// only defined on `tensor<?xindex>` operands. The test for equality first
/// compares their size and, if equal, checks every extent for equality.
///
/// Example:
///
/// %result = shape.shape_eq %a, %b : tensor<?xindex>, tensor<?xindex>
///
/// becomes
///
/// %c0 = constant 0 : index
/// %0 = dim %arg0, %c0 : tensor<?xindex>
/// %1 = dim %arg1, %c0 : tensor<?xindex>
/// %2 = cmpi "eq", %0, %1 : index
/// %result = scf.if %2 -> (i1) {
///   %c1 = constant 1 : index
///   %true = constant true
///   %4 = scf.for %arg2 = %c0 to %0 step %c1 iter_args(%arg3 = %true) -> (i1) {
///     %5 = extract_element %arg0[%arg2] : tensor<?xindex>
///     %6 = extract_element %arg1[%arg2] : tensor<?xindex>
///     %7 = cmpi "eq", %5, %6 : index
///     %8 = and %arg3, %7 : i1
///     scf.yield %8 : i1
///   }
///   scf.yield %4 : i1
/// } else {
///   %false = constant false
///   scf.yield %false : i1
/// }
///
struct ShapeEqOpConverter : public OpConversionPattern<ShapeEqOp> {
  using OpConversionPattern<ShapeEqOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(ShapeEqOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override;
};
} // namespace

LogicalResult
ShapeEqOpConverter::matchAndRewrite(ShapeEqOp op, ArrayRef<Value> operands,
                                    ConversionPatternRewriter &rewriter) const {
  // For now, this lowering is only defined on `tensor<?xindex>` operands, not
  // on shapes.
  if (op.lhs().getType().isa<ShapeType>() ||
      op.rhs().getType().isa<ShapeType>()) {
    return failure();
  }

  ShapeEqOp::Adaptor transformed(operands);
  auto loc = op.getLoc();
  Type indexTy = rewriter.getIndexType();
  Value zero = rewriter.create<ConstantIndexOp>(loc, 0);
  Value lhsRank = rewriter.create<DimOp>(loc, indexTy, transformed.lhs(), zero);
  Value rhsRank = rewriter.create<DimOp>(loc, indexTy, transformed.rhs(), zero);
  Value eqRank =
      rewriter.create<CmpIOp>(loc, CmpIPredicate::eq, lhsRank, rhsRank);
  Type i1Ty = rewriter.getI1Type();
  rewriter.replaceOpWithNewOp<IfOp>(
      op, i1Ty, eqRank,
      [&](OpBuilder &b, Location loc) {
        Value one = b.create<ConstantIndexOp>(loc, 1);
        Value init = b.create<ConstantOp>(loc, i1Ty, b.getBoolAttr(true));
        auto loop = b.create<scf::ForOp>(
            loc, zero, lhsRank, one, ValueRange{init},
            [&](OpBuilder &b, Location nestedLoc, Value iv, ValueRange args) {
              Value conj = args[0];
              Value lhsExtent =
                  b.create<ExtractElementOp>(loc, transformed.lhs(), iv);
              Value rhsExtent =
                  b.create<ExtractElementOp>(loc, transformed.rhs(), iv);
              Value eqExtent = b.create<CmpIOp>(loc, CmpIPredicate::eq,
                                                lhsExtent, rhsExtent);
              Value conjNext = b.create<AndOp>(loc, conj, eqExtent);
              b.create<scf::YieldOp>(loc, ValueRange({conjNext}));
            });
        b.create<scf::YieldOp>(loc, loop.getResults());
      },
      [&](OpBuilder &b, Location loc) {
        Value result = b.create<ConstantOp>(loc, i1Ty, b.getBoolAttr(false));
        b.create<scf::YieldOp>(loc, result);
      });
  return success();
}

namespace {
class ShapeOfOpConversion : public OpConversionPattern<ShapeOfOp> {
public:
  using OpConversionPattern<ShapeOfOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(ShapeOfOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override;
};
} // namespace

LogicalResult ShapeOfOpConversion::matchAndRewrite(
    ShapeOfOp op, ArrayRef<Value> operands,
    ConversionPatternRewriter &rewriter) const {

  // For now, only error-free types are supported by this lowering.
  if (op.getType().isa<ShapeType>())
    return failure();

  // For ranked tensor arguments, lower to `tensor_from_elements`.
  auto loc = op.getLoc();
  ShapeOfOp::Adaptor transformed(operands);
  Value tensor = transformed.arg();
  Type tensorTy = tensor.getType();
  if (tensorTy.isa<RankedTensorType>()) {

    // Build values for individual extents.
    SmallVector<Value, 8> extentValues;
    RankedTensorType rankedTensorTy = tensorTy.cast<RankedTensorType>();
    int64_t rank = rankedTensorTy.getRank();
    for (int64_t i = 0; i < rank; i++) {
      if (rankedTensorTy.isDynamicDim(i)) {
        Value extent = rewriter.create<DimOp>(loc, tensor, i);
        extentValues.push_back(extent);
      } else {
        Value extent =
            rewriter.create<ConstantIndexOp>(loc, rankedTensorTy.getDimSize(i));
        extentValues.push_back(extent);
      }
    }

    // Materialize extent tensor.
    Value staticExtentTensor = rewriter.create<TensorFromElementsOp>(
        loc, rewriter.getIndexType(), extentValues);
    rewriter.replaceOpWithNewOp<TensorCastOp>(op, staticExtentTensor,
                                              op.getType());
    return success();
  }

  // Lower to `dynamic_tensor_from_elements` otherwise.
  auto *ctx = rewriter.getContext();
  Value rank = rewriter.create<mlir::RankOp>(loc, tensor);
  rewriter.replaceOpWithNewOp<DynamicTensorFromElementsOp>(
      op, getExtentTensorType(ctx), ValueRange{rank},
      [&](OpBuilder &b, Location loc, ValueRange args) {
        Value dim = args.front();
        Value extent = b.create<DimOp>(loc, tensor, dim);
        b.create<mlir::YieldOp>(loc, extent);
      });

  return success();
}

namespace {
class ToExtentTensorOpConversion
    : public OpConversionPattern<ToExtentTensorOp> {
public:
  using OpConversionPattern<ToExtentTensorOp>::OpConversionPattern;

  LogicalResult
  matchAndRewrite(ToExtentTensorOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    ToExtentTensorOpAdaptor adaptor(operands);

    if (!adaptor.input().getType().isa<RankedTensorType>())
      return rewriter.notifyMatchFailure(op, "input needs to be a tensor");

    rewriter.replaceOpWithNewOp<TensorCastOp>(op, adaptor.input(),
                                              op.getType());
    return success();
  }
};
} // namespace

namespace {
/// Conversion pass.
class ConvertShapeToStandardPass
    : public ConvertShapeToStandardBase<ConvertShapeToStandardPass> {

  void runOnOperation() override;
};
} // namespace

void ConvertShapeToStandardPass::runOnOperation() {
  // Setup target legality.
  MLIRContext &ctx = getContext();
  ConversionTarget target(ctx);
  target.addLegalDialect<StandardOpsDialect, SCFDialect>();
  target.addLegalOp<FuncOp, ModuleOp, ModuleTerminatorOp>();

  // Setup conversion patterns.
  OwningRewritePatternList patterns;
  populateShapeToStandardConversionPatterns(patterns, &ctx);

  // Apply conversion.
  auto module = getOperation();
  if (failed(applyPartialConversion(module, target, patterns)))
    signalPassFailure();
}

void mlir::populateShapeToStandardConversionPatterns(
    OwningRewritePatternList &patterns, MLIRContext *ctx) {
  // clang-format off
  patterns.insert<
      AnyOpConversion,
      BinaryOpConversion<AddOp, AddIOp>,
      BinaryOpConversion<MulOp, MulIOp>,
      BroadcastOpConverter,
      ConstShapeOpConverter,
      ConstSizeOpConversion,
      GetExtentOpConverter,
      RankOpConverter,
      ReduceOpConverter,
      ShapeEqOpConverter,
      ShapeOfOpConversion,
      ToExtentTensorOpConversion>(ctx);
  // clang-format on
}

std::unique_ptr<OperationPass<ModuleOp>>
mlir::createConvertShapeToStandardPass() {
  return std::make_unique<ConvertShapeToStandardPass>();
}