torch-mlir/lib/Dialect/Torch/IR/TorchOps.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 "npcomp/Dialect/Torch/IR/TorchOps.h"

#include "mlir/Dialect/StandardOps/IR/Ops.h"
#include "mlir/IR/Builders.h"
#include "mlir/IR/BuiltinOps.h"
#include "mlir/IR/Matchers.h"
#include "mlir/IR/PatternMatch.h"
#include "mlir/IR/TypeUtilities.h"
#include "npcomp/Dialect/Basicpy/IR/BasicpyDialect.h"
#include "npcomp/Dialect/Basicpy/IR/BasicpyOps.h"
#include "llvm/ADT/StringMap.h"

using namespace mlir;
using namespace mlir::NPCOMP;
using namespace mlir::NPCOMP::Torch;

//===----------------------------------------------------------------------===//
// Utilities
//===----------------------------------------------------------------------===//

Value mlir::NPCOMP::Torch::copyTensorToType(OpBuilder &builder, Location loc,
                                            BaseTensorType newType,
                                            Value tensor) {
  auto originalType = tensor.getType().cast<BaseTensorType>();
  // Adjust the static information in the type to match between the original and
  // new types.
  if (!originalType.hasSameSizesAndDtype(newType)) {
    tensor = builder.create<TensorStaticInfoCastOp>(
        loc, originalType.getWithSizesAndDtypeFrom(newType), tensor);
  }
  // If both the original and new types already have value semantics, a copy is
  // pointless.
  if (originalType.isa<ValueTensorType>() && newType.isa<ValueTensorType>())
    return tensor;
  return builder.create<CopyTensorOp>(loc, newType, tensor);
}

//===----------------------------------------------------------------------===//
// MethodOp
//===----------------------------------------------------------------------===//

LogicalResult MethodOp::verifySymbolUses(SymbolTableCollection &symbolTable) {
  auto func = symbolTable.lookupNearestSymbolFrom<FuncOp>(*this, function());
  if (!func)
    return emitError() << "'@" << function()
                       << "' does not reference a valid function";
  if (func.getVisibility() != SymbolTable::Visibility::Private)
    return emitError() << "'@" << function()
                       << "' must reference a private function";
  if (func.isDeclaration())
    return emitError() << "'@" << function()
                       << "' must reference a function that is defined (not "
                          "merely declared)";
  auto expectedReceiverArgType = NnModuleType::get(
      getContext(), getOperation()->getParentOfType<ClassTypeOp>().getName());
  if (func.getType().getNumInputs() == 0 ||
      func.getType().getInput(0) != expectedReceiverArgType) {
    return emitError() << "the referenced function '" << function()
                       << "' must have a first argument of type "
                       << expectedReceiverArgType;
  }
  return success();
}

//===----------------------------------------------------------------------===//
// NnModuleOp
//===----------------------------------------------------------------------===//

static LogicalResult verify(NnModuleOp op) {
  for (Operation &child : *op.getBody())
    if (!isa<SlotOp, NnModuleTerminatorOp>(&child))
      return child.emitOpError() << "is not allowed inside 'torch.nn_module'";
  return success();
}

// PyTorch has a well-developed notion of subtyping.
//
// This is a restricted subset of it.
//
// TODO: Flesh this out.
// TODO: Decide / properly model the distinction between PEP 483 / Python
// subtyping vs "more static information".
bool isValidSubtype(Type subtype, Type type) {
  if (subtype == type)
    return true;
  if (auto optional = type.dyn_cast<OptionalType>())
    return subtype == optional.getContainedType() ||
           subtype.isa<Basicpy::NoneType>();
  // TODO: This is not subtyping according to PEP 483. See description
  // of NonValueTensorType.
  if (subtype.isa<NonValueTensorType>() && type.isa<NonValueTensorType>() &&
      type ==
          NonValueTensorType::getWithLeastStaticInformation(type.getContext()))
    return true;
  return false;
}

LogicalResult NnModuleOp::verifySymbolUses(SymbolTableCollection &symbolTable) {
  auto classType =
      symbolTable.lookupNearestSymbolFrom<ClassTypeOp>(*this, getClassName());
  if (!classType)
    return emitError() << "'" << getClassName()
                       << "' does not reference a valid class type";

  auto attrs = llvm::to_vector<6>(getBody()->getOps<SlotOp>());
  auto attrDefs = llvm::to_vector<6>(classType.getBody()->getOps<AttrOp>());
  if (attrs.size() != attrDefs.size())
    return emitError() << "number of 'torch.slot's in a 'torch.nn_module' must "
                          "match number of 'torch.attr's in "
                          "the corresponding 'torch.class_type'";
  for (int i = 0, e = attrs.size(); i != e; i++) {
    SlotOp attr = attrs[i];
    AttrOp attrDef = attrDefs[i];
    if (!isValidSubtype(attr.value().getType(), attrDef.type()) ||
        attr.name() != attrDef.name()) {
      return attr.emitOpError()
          .append("is expected to match type and name of '",
                  attrDef.getOperation(), "'")
          .attachNote(attrDef.getLoc())
          .append("see torch.attr at corresponding index ", i, " here");
    }
  }

  return success();
}

//===----------------------------------------------------------------------===//
// PrimListConstructOp
//===----------------------------------------------------------------------===//

static LogicalResult verify(PrimListConstructOp op) {
  auto resultType = op.getResult().getType();
  auto resultElementType = resultType.dyn_cast<ListType>().getContainedType();
  auto matchResultElementType = [&](Type type) {
    return type.getTypeID() == resultElementType.getTypeID();
  };
  if (llvm::all_of(op->getOperandTypes(), matchResultElementType))
    return success();
  else return failure();
}

//===----------------------------------------------------------------------===//
// ClassTypeOp
//===----------------------------------------------------------------------===//

static LogicalResult verify(ClassTypeOp op) {
  llvm::StringMap<Operation *> namesToOps;
  for (Operation &child : op.getBody()->without_terminator()) {
    if (!isa<AttrOp, MethodOp>(&child))
      return child.emitOpError() << "is not allowed inside `torch.class_type`";
    StringRef name;
    if (auto attr = dyn_cast<AttrOp>(child))
      name = attr.name();
    else
      name = cast<MethodOp>(child).name();
    auto itAndWasInserted = namesToOps.insert({name, &child});
    auto it = itAndWasInserted.first;
    bool wasInserted = itAndWasInserted.second;
    if (!wasInserted) {
      auto diag = op.emitOpError().append(
          "has duplicate attr/method with name '", name, "'");
      diag.attachNote(it->second->getLoc())
          .append("see first conflicting attr/method here");
      diag.attachNote(child.getLoc())
          .append("see second conflicting attr/method here");
      return failure();
    }
  }

  return success();
}

//===----------------------------------------------------------------------===//
// PrimLoopOp
//===----------------------------------------------------------------------===//

OperandRange PrimLoopOp::getSuccessorEntryOperands(unsigned index) {
  assert(index == 0);
  return iterArgsInit();
}

void PrimLoopOp::getSuccessorRegions(
    Optional<unsigned> index, ArrayRef<Attribute> operands,
    SmallVectorImpl<RegionSuccessor> &regions) {
  (void)operands;

  if (!index.hasValue()) {
    regions.emplace_back(&region(), region().getArguments().slice(1));
    return;
  }
  assert(*index == 0);
  regions.emplace_back(&region(), region().getArguments().slice(1));
  regions.emplace_back(getResults());
}

//===----------------------------------------------------------------------===//
// DerefineOp
//===----------------------------------------------------------------------===//

bool DerefineOp::areCastCompatible(mlir::TypeRange inputs,
                                   mlir::TypeRange outputs) {
  return isValidSubtype(inputs[0], outputs[0]);
}

void DerefineOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
                                             MLIRContext *context) {
  patterns.add(+[](DerefineOp op, PatternRewriter &rewriter) {
    // TODO: Extend RefineTypes for this case and delete this canonicalization,
    // since we don't want control flow or calls to randomly block this fold
    // (this canonicalization pattern makes the compiler brittle to control flow
    // and calls).
    bool allAllowRefinement =
        llvm::all_of(op.getResult().getUsers(), allowsTypeRefinement);
    if (!allAllowRefinement)
      return failure();
    rewriter.replaceOp(op, op.getOperand());
    return success();
  });
}

//===----------------------------------------------------------------------===//
// Aten__Is__Op
//===----------------------------------------------------------------------===//

OpFoldResult Aten__Is__Op::fold(ArrayRef<Attribute> operands) {
  auto lhsType = self().getType();
  auto rhsType = obj().getType();
  // If either type is a NoneType, make it be the lhsType.
  if (rhsType.isa<Basicpy::NoneType>())
    std::swap(lhsType, rhsType);
  // TODO: Implement and use subtype infra for this.
  // If neither type is a subtype of the other, then the result is false.
  if (lhsType.isa<Basicpy::NoneType>() && !rhsType.isa<Torch::OptionalType>())
    return IntegerAttr::get(IntegerType::get(getContext(), 1), 0);
  return nullptr;
}

//===----------------------------------------------------------------------===//
// AtenLenTOp
//===----------------------------------------------------------------------===//

OpFoldResult AtenDimOp::fold(ArrayRef<Attribute> operands) {
  if (auto tensorType = getOperand().getType().dyn_cast<BaseTensorType>()) {
    if (tensorType.hasSizes())
      return IntegerAttr::get(IntegerType::get(getContext(), 64),
                              tensorType.getSizes().size());
  }
  return nullptr;
}

//===----------------------------------------------------------------------===//
// AtenLenTOp
//===----------------------------------------------------------------------===//

OpFoldResult AtenLenTOp::fold(ArrayRef<Attribute> operands) {
  // `len([1,1,1])` -> `3`
  if (auto listConstruct = getOperand().getDefiningOp<Torch::PrimListConstructOp>()) {
    return IntegerAttr::get(IntegerType::get(getContext(), 64),
                            listConstruct.getNumOperands());
  }
  return nullptr;
}

void AtenLenTOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
                                             MLIRContext *context) {
  // `len(t.size())` -> `t.ndim`
  patterns.add(+[](AtenLenTOp op, PatternRewriter &rewriter) {
    auto size = op.getOperand().getDefiningOp<AtenSizeOp>();
    if (!size)
      return rewriter.notifyMatchFailure(op, "operand not AtenSizeOp");
    // TODO: Normalize all the torch scalar integer types to consistently use
    // a `!torch.int` type so that this op and others can automatically infer
    // their type. An additional benefit is that there's already enough of a
    // semantic gap between Python ints (which tend to be arbitrary precision)
    // and Torch/et-al ints (fixed bit depth, usually 64), it would be nice to
    // preserve the fact that we are working on a !torch.int and not just a
    // thing that was prematurely pinned to an `i64`.
    rewriter.replaceOpWithNewOp<AtenDimOp>(op, rewriter.getI64Type(),
                                           size.getOperand());
    return success();
  });
}

//===----------------------------------------------------------------------===//
// AtenSizeOp
//===----------------------------------------------------------------------===//

void AtenSizeOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
                                             MLIRContext *context) {
  patterns.add(+[](AtenSizeOp op, PatternRewriter &rewriter) {
    auto type = op.getOperand().getType().dyn_cast<BaseTensorType>();
    if (!type || !type.areAllSizesKnown())
      return rewriter.notifyMatchFailure(op, "all sizes not known");
    SmallVector<Value> listElements;
    for (int64_t size : type.getSizes()) {
      listElements.push_back(rewriter.create<::mlir::ConstantOp>(
          op->getLoc(), rewriter.getI64IntegerAttr(size)));
    }
    rewriter.replaceOpWithNewOp<Torch::PrimListConstructOp>(
        op, Torch::ListType::get(rewriter.getI64Type()), listElements);
    return success();
  });
  // One-off pattern to erase if dead.
  // TODO: Use the effects infra to express the semantics of this op and enable
  // a centralized "erase if dead" canonicalization.
  // Specifically, we need to mark the op as only MemoryEffects::Allocate
  // so that `mlir::wouldOpBeTriviallyDead` does the right thing.
  patterns.add(+[](AtenSizeOp op, PatternRewriter &rewriter) {
    if (!op.use_empty())
      return failure();
    rewriter.eraseOp(op);
    return failure();
  });
}

//===----------------------------------------------------------------------===//
// TensorOp
//===----------------------------------------------------------------------===//

LogicalResult
TensorOp::inferReturnTypes(MLIRContext *context, Optional<Location> location,
                           ValueRange operands, DictionaryAttr attributes,
                           RegionRange regions,
                           SmallVectorImpl<Type> &inferredReturnTypes) {
  auto attr = attributes.get("value").dyn_cast_or_null<ElementsAttr>();
  if (!attr)
    return failure();
  auto tensorType = attr.getType().cast<RankedTensorType>();
  inferredReturnTypes.push_back(NonValueTensorType::getFromShaped(tensorType));
  return success();
}

static bool areSizesAndDtypesCompatible(BaseTensorType a, BaseTensorType b) {
  if (a.hasSizes() && b.hasSizes()) {
    if (failed(verifyCompatibleShape(a.getSizes(), b.getSizes())))
      return false;
  }
  if (a.hasDtype() && b.hasDtype()) {
    if (a.getDtype() != b.getDtype())
      return false;
  }
  return true;
}

bool TensorOp::isCompatibleReturnTypes(TypeRange inferred, TypeRange actual) {
  if (!actual[0].isa<BaseTensorType>())
    return false;
  return areSizesAndDtypesCompatible(inferred[0].cast<BaseTensorType>(),
                                     actual[0].cast<BaseTensorType>());
}

//----------------------------------------------------------------------------//
// TensorStaticInfoCast
//----------------------------------------------------------------------------//

bool TensorStaticInfoCastOp::areCastCompatible(mlir::TypeRange inputs,
                                               mlir::TypeRange outputs) {
  return areSizesAndDtypesCompatible(inputs[0].cast<BaseTensorType>(),
                                     outputs[0].cast<BaseTensorType>());
}

//===----------------------------------------------------------------------===//
// CopyTensorOp
//===----------------------------------------------------------------------===//

static LogicalResult verify(CopyTensorOp op) {
  auto resultType = op.getResult().getType().cast<BaseTensorType>();
  auto operandType = op.getOperand().getType().cast<BaseTensorType>();
  if (!resultType.hasSameSizesAndDtype(operandType)) {
    return op.emitError()
           << "operand and result must have same sizes and dtype";
  }
  return success();
}

OpFoldResult CopyTensorOp::fold(ArrayRef<Attribute> operands) {
  // A copy between value semantic tensors is a no-op.
  if (getType().isa<ValueTensorType>() &&
      getOperand().getType().isa<ValueTensorType>()) {
    return getOperand();
  }
  return nullptr;
}

void CopyTensorOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
                                               MLIRContext *context) {
  // y = torch.copy.tensor(hasOneUse@torch.copy.tensor(x)) -> x
  // Only safe when `y` and `x` have value semantics.
  patterns.add(+[](CopyTensorOp op, PatternRewriter &rewriter) {
    auto otherCopy = op.getOperand().getDefiningOp<CopyTensorOp>();
    if (!otherCopy)
      return failure();
    if (otherCopy.getOperand().getType().isa<ValueTensorType>() &&
        op.getResult().getType().isa<ValueTensorType>() &&
        op.getOperand().hasOneUse()) {
      rewriter.replaceOp(op, {otherCopy.getOperand()});
      // TODO: Implement MemoryEffectOpInterface to handle the value/non-value
      // cases precisely. In this case, we specifically know that `otherCopy`
      // is dead so eagerly clean it up.
      rewriter.eraseOp(otherCopy);
      return success();
    }
    return failure();
  });
}

//===----------------------------------------------------------------------===//
// ToBuiltinTensorOp
//===----------------------------------------------------------------------===//

LogicalResult ToBuiltinTensorOp::inferReturnTypes(
    MLIRContext *context, Optional<Location> location, ValueRange operands,
    DictionaryAttr attributes, RegionRange regions,
    SmallVectorImpl<Type> &inferredReturnTypes) {
  auto resultType =
      operands[0].getType().cast<ValueTensorType>().toBuiltinTensor();
  if (!resultType)
    return failure();
  inferredReturnTypes.push_back(resultType);
  return success();
}

//===----------------------------------------------------------------------===//
// FromBuiltinTensorOp
//===----------------------------------------------------------------------===//

LogicalResult FromBuiltinTensorOp::inferReturnTypes(
    MLIRContext *context, Optional<Location> location, ValueRange operands,
    DictionaryAttr attributes, RegionRange regions,
    SmallVectorImpl<Type> &inferredReturnTypes) {
  inferredReturnTypes.push_back(
      ValueTensorType::getFromShaped(operands[0].getType().cast<TensorType>()));
  return success();
}

//===----------------------------------------------------------------------===//
// Aten__Getitem__TOp
//===----------------------------------------------------------------------===//

void Aten__Getitem__TOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
                                                     MLIRContext *context) {
  patterns.add(+[](Aten__Getitem__TOp op, PatternRewriter &rewriter) {
    auto torchList = op.getOperand(0);
    if(!torchList.hasOneUse())
      return failure();

    auto listConstruct = torchList.getDefiningOp<Torch::PrimListConstructOp>();
    if (!listConstruct)
      return failure();

    APInt indexAP;
    if (!matchPattern(op.getOperand(1), m_ConstantInt(&indexAP)))
      return failure();

    auto index = indexAP.getSExtValue();
    rewriter.replaceOp(op, {listConstruct.getOperand(index)});
    return success();
  });
}

#define GET_OP_CLASSES
#include "npcomp/Dialect/Torch/IR/TorchOps.cpp.inc"