mirror of https://github.com/llvm/torch-mlir
1673 lines
73 KiB
C++
1673 lines
73 KiB
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
|
||
// Also available under a BSD-style license. See LICENSE.
|
||
//
|
||
//===----------------------------------------------------------------------===//
|
||
|
||
#include "torch-mlir/Conversion/TorchToLinalg/TorchToLinalg.h"
|
||
|
||
#include "../PassDetail.h"
|
||
#include "PopulatePatterns.h"
|
||
#include "Utils.h"
|
||
#include "mlir/Dialect/Arithmetic/IR/Arithmetic.h"
|
||
#include "mlir/Dialect/ControlFlow/IR/ControlFlowOps.h"
|
||
#include "mlir/Dialect/Linalg/IR/Linalg.h"
|
||
#include "mlir/Dialect/Math/IR/Math.h"
|
||
#include "mlir/Dialect/Tensor/IR/Tensor.h"
|
||
#include "mlir/IR/Matchers.h"
|
||
#include "torch-mlir/Conversion/Utils/Utils.h"
|
||
#include "torch-mlir/Dialect/Torch/IR/TorchDialect.h"
|
||
#include "torch-mlir/Dialect/Torch/IR/TorchOps.h"
|
||
#include "torch-mlir/Dialect/Torch/Utils/TorchUpstream.h"
|
||
#include "torch-mlir/Dialect/Torch/Utils/Utils.h"
|
||
|
||
using namespace mlir;
|
||
using namespace mlir::torch;
|
||
using namespace mlir::torch::Torch;
|
||
|
||
// Check if a ranked-tensor has the specified element type.
|
||
template <typename elementType> static bool hasElementType(Value tensor) {
|
||
auto tensorType = tensor.getType().cast<RankedTensorType>();
|
||
Type tensorElementType = tensorType.getElementType();
|
||
return tensorElementType.isa<elementType>();
|
||
}
|
||
|
||
static Value createElementwiseLinalgGeneric(
|
||
OpBuilder &b, Location loc, ValueRange tensorOperands,
|
||
Type resultElementType,
|
||
function_ref<void(OpBuilder &, Location, ValueRange)> bodyBuild) {
|
||
// The overall error handling strategy here is best viewed by thinking about
|
||
// what happens for a single result dimension. This loop not structured that
|
||
// way because it is hard to create the affine maps for each operand unless
|
||
// we structure the loop to iterate over tensor operands as the outer loop
|
||
// instead of inner loop. This pseudocode gives better intuition:
|
||
// ```
|
||
// for each result dimension:
|
||
// for each tensor operand:
|
||
// if it doesn't even have high enough rank relative to the result:
|
||
// continue
|
||
// if it is a static size-1 along this result dimension:
|
||
// continue
|
||
// if this is the first tensor operand that didn't continue above:
|
||
// take its dimension size as the size of the non-broadcasted
|
||
// traversal along this dimension (this may include a dynamic size-1,
|
||
// **non-broadcasted** traversal!)
|
||
// emit error check "if the size does not match the non-broadcasted
|
||
// traversal size along this dimension, error"
|
||
// ```
|
||
SmallVector<int64_t> operandRanks;
|
||
operandRanks.resize(tensorOperands.size());
|
||
llvm::transform(tensorOperands, operandRanks.begin(), [](Value tensor) {
|
||
return tensor.getType().dyn_cast<RankedTensorType>().getRank();
|
||
});
|
||
|
||
auto resultRankIt =
|
||
std::max_element(operandRanks.begin(), operandRanks.end());
|
||
assert(resultRankIt != operandRanks.end() && "Unable to get result rank.");
|
||
int64_t resultRank = *resultRankIt;
|
||
|
||
// Initialize the resultShape to all 1's, as a fallback in case
|
||
// all sizes along that result dimension are statically 1.
|
||
auto c1 = b.create<arith::ConstantIndexOp>(loc, /*value=*/1);
|
||
SmallVector<Value> resultShape(resultRank, c1);
|
||
SmallVector<AffineMap> indexingMaps;
|
||
for (Value tensorOperand : tensorOperands) {
|
||
SmallVector<AffineExpr> exprs;
|
||
auto type = tensorOperand.getType().cast<RankedTensorType>();
|
||
for (auto size : llvm::enumerate(type.getShape())) {
|
||
// If the size is statically known to be 1, we don't want any
|
||
// error guards to be spuriously emitted, since we are specifically
|
||
// allowing size-1 broadcasts in this case, as they correspond to a
|
||
// constant-0 indexing map.
|
||
if (size.value() == 1) {
|
||
exprs.push_back(b.getAffineConstantExpr(0));
|
||
continue;
|
||
}
|
||
|
||
// The rank of this operand might be smaller than the overall rank of
|
||
// the broadcast. Add an offset to correlate it to the correct
|
||
// dimension of the result.
|
||
auto resultDim = size.index() + (resultRank - type.getRank());
|
||
|
||
// The generated linalg op will now be iterating along the full size
|
||
// of this dimension. Record that fact.
|
||
exprs.push_back(b.getAffineDimExpr(resultDim));
|
||
|
||
// Now, we need to ensure that such iteration is not going to trigger
|
||
// undefined behavior, by doing appropriate checks against the current
|
||
// dimension size.
|
||
auto currentDimSize = getDimOp(b, loc, tensorOperand, size.index());
|
||
|
||
// If the result size of this dimension has so far only hit the
|
||
// statically-known-to-be-1 case above (i.e., we have not yet assigned a
|
||
// new Value to `resultShape[resultDim]`), then we have no other dynamic
|
||
// values to check against, and merely need to record the current
|
||
// dimension size.
|
||
if (resultShape[resultDim] == c1) {
|
||
resultShape[resultDim] = currentDimSize;
|
||
continue;
|
||
}
|
||
|
||
// We prohibit the size-1 dynamic broadcasting scenario, so just check
|
||
// for exact equality with the running result size.
|
||
// This is the check which protects against the undefined behavior of
|
||
// the generated linalg op in the case of iterating two operands with
|
||
// dimensions sizes that are expected to match.
|
||
auto equalToRunning =
|
||
b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::eq,
|
||
resultShape[resultDim], currentDimSize);
|
||
b.create<cf::AssertOp>(loc, equalToRunning,
|
||
"mismatched size for broadcast");
|
||
}
|
||
indexingMaps.push_back(AffineMap::get(
|
||
/*dimCount=*/resultRank, /*symbolCount=*/0, exprs, b.getContext()));
|
||
}
|
||
|
||
SmallVector<StringRef> iteratorTypes(resultRank,
|
||
getParallelIteratorTypeName());
|
||
// Add the indexing map for the outs init tensor.
|
||
indexingMaps.push_back(b.getMultiDimIdentityMap(resultRank));
|
||
|
||
Value initTensor = b.create<linalg::InitTensorOp>(
|
||
loc, getAsOpFoldResult(resultShape), resultElementType);
|
||
return b
|
||
.create<linalg::GenericOp>(loc,
|
||
/*resultTensorTypes=*/initTensor.getType(),
|
||
/*inputs=*/tensorOperands,
|
||
/*outputs=*/initTensor, indexingMaps,
|
||
iteratorTypes, bodyBuild)
|
||
.getResult(0);
|
||
}
|
||
|
||
template <arith::CmpFPredicate fpred, arith::CmpIPredicate iupred,
|
||
arith::CmpIPredicate ispred>
|
||
static Value createComparisonTemplate(OpBuilder &b, Location loc, Type type,
|
||
Value lhs, Value rhs) {
|
||
if (type.isa<mlir::FloatType>())
|
||
return b.create<arith::CmpFOp>(loc, fpred, lhs, rhs);
|
||
if (IntegerType intType = type.dyn_cast<mlir::IntegerType>()) {
|
||
if (intType.isUnsigned())
|
||
return b.create<arith::CmpIOp>(loc, iupred, lhs, rhs);
|
||
if (intType.isSigned())
|
||
return b.create<arith::CmpIOp>(loc, ispred, lhs, rhs);
|
||
}
|
||
assert(false && "Unhandled element type for comparison");
|
||
}
|
||
|
||
static Value createGreaterThan(OpBuilder &b, Location loc, Type elementalType,
|
||
Value lhs, Value rhs) {
|
||
return createComparisonTemplate<arith::CmpFPredicate::UGT,
|
||
arith::CmpIPredicate::ugt,
|
||
arith::CmpIPredicate::sgt>(
|
||
b, loc, elementalType, lhs, rhs);
|
||
}
|
||
|
||
static Value createLessThan(OpBuilder &b, Location loc, Type elementalType,
|
||
Value lhs, Value rhs) {
|
||
return createComparisonTemplate<arith::CmpFPredicate::ULT,
|
||
arith::CmpIPredicate::ult,
|
||
arith::CmpIPredicate::slt>(
|
||
b, loc, elementalType, lhs, rhs);
|
||
}
|
||
|
||
static Value createEqual(OpBuilder &b, Location loc, Type elementalType,
|
||
Value lhs, Value rhs) {
|
||
return createComparisonTemplate<arith::CmpFPredicate::UEQ,
|
||
arith::CmpIPredicate::eq,
|
||
arith::CmpIPredicate::eq>(
|
||
b, loc, elementalType, lhs, rhs);
|
||
}
|
||
|
||
static Value createNotEqual(OpBuilder &b, Location loc, Type elementalType,
|
||
Value lhs, Value rhs) {
|
||
return createComparisonTemplate<arith::CmpFPredicate::UNE,
|
||
arith::CmpIPredicate::ne,
|
||
arith::CmpIPredicate::ne>(
|
||
b, loc, elementalType, lhs, rhs);
|
||
}
|
||
|
||
static Value buildNormalCdf(OpBuilder &b, Location &loc, Value x, Value mean,
|
||
Value sigma) {
|
||
Type elementType = x.getType();
|
||
Value xMinusMean = b.create<arith::SubFOp>(loc, x, mean);
|
||
Value two = b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, 2));
|
||
Value sqrt2 = b.create<math::SqrtOp>(loc, two);
|
||
Value erfArg = b.create<arith::DivFOp>(loc, xMinusMean, sqrt2);
|
||
Value erf = b.create<math::ErfOp>(loc, erfArg);
|
||
Value one = b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, 1));
|
||
Value erfPlus1 = b.create<arith::AddFOp>(loc, one, erf);
|
||
Value oneHalf =
|
||
b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, 0.5));
|
||
Value normalCdf = b.create<arith::MulFOp>(loc, oneHalf, erfPlus1);
|
||
return normalCdf;
|
||
}
|
||
|
||
static Value buildUnitNormalCdf(OpBuilder &b, Location &loc, Value x) {
|
||
Type elementType = x.getType();
|
||
Value zero = b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, 0));
|
||
Value one = b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, 1));
|
||
return buildNormalCdf(b, loc, x, zero, one);
|
||
}
|
||
|
||
template <typename MathOpTy>
|
||
static Value createCalculationForMathOpWithDtypeConversion(
|
||
OpBuilder &b, TypeConverter *converter, Value payloadArg, Operation *op) {
|
||
Type dtype = converter->convertType(op->getResult(0).getType())
|
||
.template cast<RankedTensorType>()
|
||
.getElementType();
|
||
Location loc = op->getLoc();
|
||
Value arg = convertScalarToDtype(b, loc, payloadArg, dtype);
|
||
return b.create<MathOpTy>(loc, arg);
|
||
}
|
||
|
||
static Value createLinalgPayloadCalculationForElementwiseOp(
|
||
OpBuilder &b, Location loc, TypeConverter *converter,
|
||
ValueRange payloadArgs, Operation *op, ArrayRef<Value> operands) {
|
||
if (isa<AtenFloorOp>(op))
|
||
return b.create<math::FloorOp>(loc, payloadArgs[0]);
|
||
if (isa<AtenCeilOp>(op))
|
||
return b.create<math::CeilOp>(loc, payloadArgs[0]);
|
||
if (isa<AtenTanhOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::TanhOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenExpOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::ExpOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenLogOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::LogOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenLog2Op>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::Log2Op>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenErfOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::ErfOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenSqrtOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::SqrtOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenRsqrtOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::RsqrtOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenNegOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<arith::NegFOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenSinOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::SinOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (isa<AtenCosOp>(op)) {
|
||
return createCalculationForMathOpWithDtypeConversion<math::CosOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
}
|
||
if (auto clone = dyn_cast<AtenCloneOp>(op)) {
|
||
int64_t memoryFormat;
|
||
if (!clone.memory_format().getType().isa<Torch::NoneType>() &&
|
||
(!matchPattern(clone.memory_format(),
|
||
m_TorchConstantInt(&memoryFormat)) ||
|
||
memoryFormat != torch_upstream::MemoryFormat::Contiguous)) {
|
||
clone.emitError("unimplemented: only default memory format is supported");
|
||
return nullptr;
|
||
}
|
||
return payloadArgs[0];
|
||
}
|
||
if (auto bitwiseAndTensor = dyn_cast<AtenBitwiseAndTensorOp>(op)) {
|
||
if (bitwiseAndTensor.getType()
|
||
.cast<ValueTensorType>()
|
||
.getDtype()
|
||
.isa<mlir::FloatType>()) {
|
||
bitwiseAndTensor.emitError(
|
||
"Bitwise_And does not support floating point dtype");
|
||
return nullptr;
|
||
}
|
||
Type dtype = converter->convertType(bitwiseAndTensor.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value rhs = convertScalarToDtype(b, loc, payloadArgs[1], dtype);
|
||
return b.create<arith::AndIOp>(loc, lhs, rhs);
|
||
}
|
||
if (isa<AtenAbsOp>(op))
|
||
return b.create<math::AbsOp>(loc, payloadArgs[0]);
|
||
if (isa<AtenSigmoidOp>(op)) {
|
||
auto negate = createCalculationForMathOpWithDtypeConversion<arith::NegFOp>(
|
||
b, converter, payloadArgs[0], op);
|
||
auto one =
|
||
b.create<arith::ConstantOp>(loc, FloatAttr::get(negate.getType(), 1));
|
||
auto exp = b.create<math::ExpOp>(loc, negate);
|
||
auto added = b.create<arith::AddFOp>(loc, exp, one);
|
||
return b.create<arith::DivFOp>(loc, one, added);
|
||
}
|
||
if (auto relu = dyn_cast<AtenReluOp>(op)) {
|
||
if (!relu.getType()
|
||
.cast<ValueTensorType>()
|
||
.getDtype()
|
||
.isa<mlir::FloatType>()) {
|
||
relu.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
Type elementType = payloadArgs[0].getType();
|
||
Value constZero =
|
||
b.create<arith::ConstantOp>(loc, b.getZeroAttr(elementType));
|
||
Value pred = b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::UGT,
|
||
payloadArgs[0], constZero);
|
||
return b.create<arith::SelectOp>(loc, pred, payloadArgs[0], constZero);
|
||
}
|
||
if (auto lrelu = dyn_cast<AtenLeakyReluOp>(op)) {
|
||
if (!lrelu.getType()
|
||
.cast<ValueTensorType>()
|
||
.getDtype()
|
||
.isa<mlir::FloatType>()) {
|
||
lrelu.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
Type elementType = payloadArgs[0].getType();
|
||
Value constZero =
|
||
b.create<arith::ConstantOp>(loc, b.getZeroAttr(elementType));
|
||
Value pred = b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::UGT,
|
||
payloadArgs[0], constZero);
|
||
Value positivePart =
|
||
b.create<arith::SelectOp>(loc, pred, payloadArgs[0], constZero);
|
||
Value negativePart =
|
||
b.create<arith::SelectOp>(loc, pred, constZero, payloadArgs[0]);
|
||
Value scale = convertScalarToDtype(b, loc, operands[1], elementType);
|
||
Value scaledNegativePart =
|
||
b.create<arith::MulFOp>(loc, negativePart, scale);
|
||
return b.create<arith::AddFOp>(loc, positivePart, scaledNegativePart);
|
||
}
|
||
if (auto gelu = dyn_cast<AtenGeluOp>(op)) {
|
||
if (!gelu.getType()
|
||
.cast<ValueTensorType>()
|
||
.getDtype()
|
||
.isa<mlir::FloatType>()) {
|
||
gelu.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
// TODO: Take approximation into account.
|
||
std::string approximate;
|
||
if (!matchPattern(gelu.approximate(), m_TorchConstantStr(approximate)) ||
|
||
approximate != "none")
|
||
return nullptr;
|
||
Value cdf = buildUnitNormalCdf(b, loc, payloadArgs[0]);
|
||
return b.create<arith::MulFOp>(loc, payloadArgs[0], cdf);
|
||
}
|
||
if (auto geluBackward = dyn_cast<AtenGeluBackwardOp>(op)) {
|
||
if (!geluBackward.getType()
|
||
.cast<ValueTensorType>()
|
||
.getDtype()
|
||
.isa<mlir::FloatType>()) {
|
||
geluBackward.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
// TODO: Take approximation into account.
|
||
std::string approximate;
|
||
if (!matchPattern(geluBackward.approximate(),
|
||
m_TorchConstantStr(approximate)) ||
|
||
approximate != "none")
|
||
return nullptr;
|
||
Type elementType = payloadArgs[1].getType();
|
||
Value cstAlpha0 = b.create<arith::ConstantOp>(
|
||
loc, FloatAttr::get(elementType, 1.12837916709551257390));
|
||
Value cstAlpha1 = b.create<arith::ConstantOp>(
|
||
loc, FloatAttr::get(elementType, 0.70710678118654752440));
|
||
Value oneHalf =
|
||
b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, 0.5));
|
||
Value kAlpha = b.create<arith::MulFOp>(loc, cstAlpha0, cstAlpha1);
|
||
Value kAlphaHalf = b.create<arith::MulFOp>(loc, kAlpha, oneHalf);
|
||
Value negOneHalf =
|
||
b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, -0.5));
|
||
Value inputSquared =
|
||
b.create<arith::MulFOp>(loc, payloadArgs[1], payloadArgs[1]);
|
||
Value negHalfInputSquared =
|
||
b.create<arith::MulFOp>(loc, inputSquared, negOneHalf);
|
||
Value dinput = b.create<math::ExpOp>(loc, negHalfInputSquared);
|
||
Value cdf = buildUnitNormalCdf(b, loc, payloadArgs[1]);
|
||
Value dinputInput = b.create<arith::MulFOp>(loc, dinput, payloadArgs[1]);
|
||
Value dinputInputAlpha =
|
||
b.create<arith::MulFOp>(loc, dinputInput, kAlphaHalf);
|
||
Value cdfExt = b.create<arith::AddFOp>(loc, dinputInputAlpha, cdf);
|
||
return b.create<arith::MulFOp>(loc, payloadArgs[0], cdfExt);
|
||
}
|
||
if (auto add = dyn_cast<AtenAddTensorOp>(op)) {
|
||
AtenAddTensorOp::Adaptor adaptor(operands);
|
||
Type dtype = converter->convertType(add.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value rhs = convertScalarToDtype(b, loc, payloadArgs[1], dtype);
|
||
Value alpha = convertScalarToDtype(b, loc, adaptor.alpha(), dtype);
|
||
if (dtype.isa<mlir::FloatType>()) {
|
||
Value scaled = b.create<arith::MulFOp>(loc, rhs, alpha);
|
||
return b.create<arith::AddFOp>(loc, lhs, scaled);
|
||
} else {
|
||
Value scaled = b.create<arith::MulIOp>(loc, rhs, alpha);
|
||
return b.create<arith::AddIOp>(loc, lhs, scaled);
|
||
}
|
||
}
|
||
if (auto sub = dyn_cast<AtenSubTensorOp>(op)) {
|
||
AtenSubTensorOp::Adaptor adaptor(operands);
|
||
Type dtype = converter->convertType(sub.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value rhs = convertScalarToDtype(b, loc, payloadArgs[1], dtype);
|
||
Value alpha = convertScalarToDtype(b, loc, adaptor.alpha(), dtype);
|
||
if (dtype.isa<mlir::FloatType>()) {
|
||
Value scaled = b.create<arith::MulFOp>(loc, rhs, alpha);
|
||
return b.create<arith::SubFOp>(loc, lhs, scaled);
|
||
} else {
|
||
Value scaled = b.create<arith::MulIOp>(loc, rhs, alpha);
|
||
return b.create<arith::SubIOp>(loc, lhs, scaled);
|
||
}
|
||
}
|
||
if (auto subScalar = dyn_cast<AtenSubScalarOp>(op)) {
|
||
Type dtype = converter->convertType(subScalar.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value self = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value other = convertScalarToDtype(b, loc, operands[1], dtype);
|
||
Value alpha = convertScalarToDtype(b, loc, operands[2], dtype);
|
||
if (dtype.isa<mlir::FloatType>()) {
|
||
Value mult = b.create<arith::MulFOp>(loc, other, alpha);
|
||
return b.create<arith::SubFOp>(loc, self, mult);
|
||
} else if (dtype.isa<mlir::IntegerType>()) {
|
||
Value mult = b.create<arith::MulIOp>(loc, other, alpha);
|
||
return b.create<arith::SubIOp>(loc, self, mult);
|
||
}
|
||
subScalar.emitError("unimplemented: dtype other than float and integer "
|
||
"types are not supported.");
|
||
return nullptr;
|
||
}
|
||
if (auto addScalar = dyn_cast<AtenAddScalarOp>(op)) {
|
||
Type dtype = converter->convertType(addScalar.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value self = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value other = convertScalarToDtype(b, loc, operands[1], dtype);
|
||
Value alpha = convertScalarToDtype(b, loc, operands[2], dtype);
|
||
if (dtype.isa<mlir::FloatType>()) {
|
||
Value mult = b.create<arith::MulFOp>(loc, other, alpha);
|
||
return b.create<arith::AddFOp>(loc, self, mult);
|
||
} else if (dtype.isa<mlir::IntegerType>()) {
|
||
Value mult = b.create<arith::MulIOp>(loc, other, alpha);
|
||
return b.create<arith::AddIOp>(loc, self, mult);
|
||
}
|
||
addScalar.emitError("unimplemented: dtype other than float and integer "
|
||
"types are not supported.");
|
||
return nullptr;
|
||
}
|
||
if (auto mul = dyn_cast<AtenMulTensorOp>(op)) {
|
||
AtenMulTensorOp::Adaptor adaptor(operands);
|
||
Type dtype = converter->convertType(mul.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value rhs = convertScalarToDtype(b, loc, payloadArgs[1], dtype);
|
||
if (dtype.isa<mlir::FloatType>()) {
|
||
return b.create<arith::MulFOp>(loc, lhs, rhs);
|
||
} else {
|
||
return b.create<arith::MulIOp>(loc, lhs, rhs);
|
||
}
|
||
}
|
||
if (auto gtTensor = dyn_cast<AtenGtTensorOp>(op)) {
|
||
AtenGtTensorOp::Adaptor adaptor(operands);
|
||
Type lhsDtype = payloadArgs[0].getType();
|
||
Type rhsDtype = payloadArgs[1].getType();
|
||
|
||
// TODO: Type promotion in case of different `lhsDtype` and `rhsDtype` needs
|
||
// to be handled.
|
||
if (lhsDtype != rhsDtype) {
|
||
gtTensor.emitError("unimplemented: different lhs and rhs dtype");
|
||
return nullptr;
|
||
}
|
||
|
||
Type elementalType =
|
||
gtTensor.self().getType().cast<BaseTensorType>().getDtype();
|
||
return createGreaterThan(b, loc, elementalType, payloadArgs[0],
|
||
payloadArgs[1]);
|
||
}
|
||
if (auto eqTensor = dyn_cast<AtenEqTensorOp>(op)) {
|
||
AtenEqTensorOp::Adaptor adaptor(operands);
|
||
Type lhsDtype = payloadArgs[0].getType();
|
||
Type rhsDtype = payloadArgs[1].getType();
|
||
|
||
// TODO: Type promotion in case of different `lhsDtype` and `rhsDtype` needs
|
||
// to be handled.
|
||
if (lhsDtype != rhsDtype) {
|
||
eqTensor.emitError("unimplemented: lhs and rhs dtype must be same");
|
||
return nullptr;
|
||
}
|
||
|
||
Type elementalType =
|
||
eqTensor.self().getType().cast<BaseTensorType>().getDtype();
|
||
|
||
if (elementalType.isa<mlir::FloatType>())
|
||
return b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::UEQ,
|
||
payloadArgs[0], payloadArgs[1]);
|
||
if (elementalType.isa<mlir::IntegerType>()) {
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::eq,
|
||
payloadArgs[0], payloadArgs[1]);
|
||
}
|
||
eqTensor.emitError("unimplemented: dtype isn't supported.");
|
||
return nullptr;
|
||
}
|
||
if (auto ltTensor = dyn_cast<AtenLtTensorOp>(op)) {
|
||
AtenLtTensorOp::Adaptor adaptor(operands);
|
||
Type lhsDtype = payloadArgs[0].getType();
|
||
Type rhsDtype = payloadArgs[1].getType();
|
||
|
||
// TODO: Type promotion in case of different `lhsDtype` and `rhsDtype` needs
|
||
// to be handled.
|
||
if (lhsDtype != rhsDtype) {
|
||
ltTensor.emitError("unimplemented: lhs and rhs dtype must be same");
|
||
return nullptr;
|
||
}
|
||
|
||
Type elementalType =
|
||
ltTensor.self().getType().cast<BaseTensorType>().getDtype();
|
||
return createLessThan(b, loc, elementalType, payloadArgs[0],
|
||
payloadArgs[1]);
|
||
}
|
||
if (auto div = dyn_cast<AtenDivTensorOp>(op)) {
|
||
AtenDivTensorOp::Adaptor adaptor(operands);
|
||
Type dtype = converter->convertType(div.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
if (!dtype.isa<mlir::FloatType>())
|
||
div.emitError("unimplemented: non-floating point dtype");
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value rhs = convertScalarToDtype(b, loc, payloadArgs[1], dtype);
|
||
return b.create<arith::DivFOp>(loc, lhs, rhs);
|
||
}
|
||
if (auto pow = dyn_cast<AtenPowTensorScalarOp>(op)) {
|
||
if (!pow.getType()
|
||
.cast<ValueTensorType>()
|
||
.getDtype()
|
||
.isa<mlir::FloatType>()) {
|
||
pow.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
Type dtype = pow.self().getType().cast<ValueTensorType>().getDtype();
|
||
Value expPromoted = convertScalarToDtype(b, loc, operands[1], dtype);
|
||
return b.create<math::PowFOp>(loc, payloadArgs[0], expPromoted);
|
||
}
|
||
|
||
if (auto gtScalar = dyn_cast<AtenGtScalarOp>(op)) {
|
||
Type dtype = gtScalar.self().getType().cast<BaseTensorType>().getDtype();
|
||
|
||
// TODO: `gtTensor` and `gtScalar` share similar code and can be called from
|
||
// one static function.
|
||
Value otherPromoted =
|
||
convertScalarToDtype(b, loc, operands[1], payloadArgs[0].getType());
|
||
|
||
if (dtype.isa<mlir::FloatType>())
|
||
return b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::UGT,
|
||
payloadArgs[0], otherPromoted);
|
||
if (IntegerType intType = dtype.dyn_cast<mlir::IntegerType>()) {
|
||
if (!operands[1].getType().isa<mlir::IntegerType>()) {
|
||
// TODO: Promote tensor args from integer to float.
|
||
gtScalar.emitError(
|
||
"unimplemented: type promotion from tensor to scalar.");
|
||
return nullptr;
|
||
}
|
||
|
||
if (intType.isUnsigned())
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::ugt,
|
||
payloadArgs[0], otherPromoted);
|
||
if (intType.isSigned())
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::sgt,
|
||
payloadArgs[0], otherPromoted);
|
||
}
|
||
gtScalar.emitError("unimplemented: dtype isn't supported.");
|
||
return nullptr;
|
||
}
|
||
|
||
if (auto geScalar = dyn_cast<AtenGeScalarOp>(op)) {
|
||
Type dtype = geScalar.self().getType().cast<BaseTensorType>().getDtype();
|
||
|
||
// TODO: The `AtenGeScalarOp` and `AtenGtScalarOp` share a lot of code that
|
||
// can be refactored.
|
||
Value otherPromoted =
|
||
convertScalarToDtype(b, loc, operands[1], payloadArgs[0].getType());
|
||
|
||
if (dtype.isa<mlir::FloatType>())
|
||
return b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::UGE,
|
||
payloadArgs[0], otherPromoted);
|
||
if (IntegerType intType = dtype.dyn_cast<mlir::IntegerType>()) {
|
||
if (!operands[1].getType().isa<mlir::IntegerType>()) {
|
||
// TODO: Promote tensor args from integer to float.
|
||
geScalar.emitError(
|
||
"unimplemented: type promotion from tensor to scalar.");
|
||
return nullptr;
|
||
}
|
||
|
||
if (intType.isUnsigned())
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::uge,
|
||
payloadArgs[0], otherPromoted);
|
||
if (intType.isSigned())
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::sge,
|
||
payloadArgs[0], otherPromoted);
|
||
}
|
||
geScalar.emitError("unimplemented: dtype isn't supported.");
|
||
return nullptr;
|
||
}
|
||
|
||
if (auto eqScalar = dyn_cast<AtenEqScalarOp>(op)) {
|
||
Type dtype = eqScalar.self().getType().cast<BaseTensorType>().getDtype();
|
||
Value otherPromoted =
|
||
convertScalarToDtype(b, loc, operands[1], payloadArgs[0].getType());
|
||
|
||
if (dtype.isa<mlir::IntegerType>()) {
|
||
if (!operands[1].getType().isa<mlir::IntegerType>()) {
|
||
// TODO: Promote tensor operand from integer to float.
|
||
eqScalar.emitError(
|
||
"unimplemented: type promotion from tensor to scalar");
|
||
return nullptr;
|
||
}
|
||
}
|
||
return createEqual(b, loc, dtype, payloadArgs[0], otherPromoted);
|
||
}
|
||
|
||
if (auto neScalar = dyn_cast<AtenNeScalarOp>(op)) {
|
||
Type dtype = neScalar.self().getType().cast<BaseTensorType>().getDtype();
|
||
Value otherPromoted =
|
||
convertScalarToDtype(b, loc, operands[1], payloadArgs[0].getType());
|
||
|
||
if (dtype.isa<mlir::IntegerType>()) {
|
||
if (!operands[1].getType().isa<mlir::IntegerType>()) {
|
||
// TODO: Promote tensor operand from integer to float.
|
||
neScalar.emitError(
|
||
"unimplemented: type promotion from tensor to scalar");
|
||
return nullptr;
|
||
}
|
||
}
|
||
return createNotEqual(b, loc, dtype, payloadArgs[0], otherPromoted);
|
||
}
|
||
|
||
if (auto ltScalar = dyn_cast<AtenLtScalarOp>(op)) {
|
||
Type dtype = ltScalar.self().getType().cast<BaseTensorType>().getDtype();
|
||
Value otherPromoted =
|
||
convertScalarToDtype(b, loc, operands[1], payloadArgs[0].getType());
|
||
|
||
// TODO: Both tensor and scalar variants of `aten.gt` and `aten.lt` share
|
||
// a lot of code that can be refactored.
|
||
if (dtype.isa<mlir::FloatType>())
|
||
return b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::ULT,
|
||
payloadArgs[0], otherPromoted);
|
||
if (IntegerType intType = dtype.dyn_cast<mlir::IntegerType>()) {
|
||
if (!operands[1].getType().isa<mlir::IntegerType>()) {
|
||
// TODO: Promote tensor operand from integer to float.
|
||
ltScalar.emitError(
|
||
"unimplemented: type promotion from tensor to scalar");
|
||
return nullptr;
|
||
}
|
||
if (intType.isUnsigned())
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::ult,
|
||
payloadArgs[0], otherPromoted);
|
||
if (intType.isSigned())
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::slt,
|
||
payloadArgs[0], otherPromoted);
|
||
}
|
||
ltScalar.emitError("unimplemented: dtype isn't supported.");
|
||
return nullptr;
|
||
}
|
||
|
||
if (auto leScalar = dyn_cast<AtenLeScalarOp>(op)) {
|
||
Type dtype = leScalar.self().getType().cast<BaseTensorType>().getDtype();
|
||
Value otherPromoted =
|
||
convertScalarToDtype(b, loc, operands[1], payloadArgs[0].getType());
|
||
|
||
// TODO: The `AtenLeScalarOp` and `AtenLtScalarOp` share a lot of code
|
||
// that can be refactored.
|
||
if (dtype.isa<mlir::FloatType>())
|
||
return b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::ULE,
|
||
payloadArgs[0], otherPromoted);
|
||
if (IntegerType intType = dtype.dyn_cast<mlir::IntegerType>()) {
|
||
if (!operands[1].getType().isa<mlir::IntegerType>()) {
|
||
// TODO: Promote tensor operand from integer to float.
|
||
leScalar.emitError(
|
||
"unimplemented: type promotion from tensor to scalar");
|
||
return nullptr;
|
||
}
|
||
if (intType.isUnsigned())
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::ule,
|
||
payloadArgs[0], otherPromoted);
|
||
if (intType.isSigned())
|
||
return b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::sle,
|
||
payloadArgs[0], otherPromoted);
|
||
}
|
||
leScalar.emitError("unimplemented: dtype isn't supported.");
|
||
return nullptr;
|
||
}
|
||
|
||
if (auto whereSelf = dyn_cast<AtenWhereSelfOp>(op)) {
|
||
Type dtype = converter->convertType(whereSelf.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[1], dtype);
|
||
Value rhs = convertScalarToDtype(b, loc, payloadArgs[2], dtype);
|
||
return b.create<arith::SelectOp>(loc, payloadArgs[0], lhs, rhs);
|
||
}
|
||
|
||
if (auto lerp = dyn_cast<AtenLerpTensorOp>(op)) {
|
||
if (!lerp.getType()
|
||
.cast<ValueTensorType>()
|
||
.getDtype()
|
||
.isa<mlir::FloatType>()) {
|
||
lerp.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
AtenLerpTensorOp::Adaptor adaptor(payloadArgs);
|
||
auto start = adaptor.self();
|
||
auto end = adaptor.end();
|
||
auto weight = adaptor.weight();
|
||
auto delta = b.create<arith::SubFOp>(loc, end, start);
|
||
auto weightedDelta = b.create<arith::MulFOp>(loc, delta, weight);
|
||
return b.create<arith::AddFOp>(loc, start, weightedDelta);
|
||
}
|
||
if (auto minimum = dyn_cast<AtenMinimumOp>(op)) {
|
||
Type dtype = minimum.getType().cast<BaseTensorType>().getDtype();
|
||
Type elemTy = converter->convertType(minimum.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[0], elemTy);
|
||
Value rhs = convertScalarToDtype(b, loc, payloadArgs[1], elemTy);
|
||
Value pred = createLessThan(b, loc, dtype, lhs, rhs);
|
||
return b.create<arith::SelectOp>(loc, pred, lhs, rhs);
|
||
}
|
||
if (auto maximum = dyn_cast<AtenMaximumOp>(op)) {
|
||
Type dtype = maximum.getType().cast<BaseTensorType>().getDtype();
|
||
Type elemTy = converter->convertType(maximum.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[0], elemTy);
|
||
Value rhs = convertScalarToDtype(b, loc, payloadArgs[1], elemTy);
|
||
Value pred = createGreaterThan(b, loc, dtype, lhs, rhs);
|
||
return b.create<arith::SelectOp>(loc, pred, lhs, rhs);
|
||
}
|
||
if (auto clamp = dyn_cast<AtenClampOp>(op)) {
|
||
Type dtype = converter->convertType(clamp.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
if (!dtype.isa<mlir::FloatType>()) {
|
||
clamp.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
AtenClampOp::Adaptor adaptor(operands);
|
||
auto min = adaptor.min();
|
||
auto max = adaptor.max();
|
||
if (min.getType().isa<Torch::OptionalType>() ||
|
||
max.getType().isa<Torch::OptionalType>()) {
|
||
clamp.emitError("unimplemented: runtime optional type");
|
||
return nullptr;
|
||
}
|
||
auto result = payloadArgs[0];
|
||
if (!min.getType().isa<Torch::NoneType>()) {
|
||
auto minPromoted = convertScalarToDtype(b, loc, min, dtype);
|
||
auto pred = b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::ULT,
|
||
result, minPromoted);
|
||
result = b.create<arith::SelectOp>(loc, pred, minPromoted, result);
|
||
}
|
||
if (!max.getType().isa<Torch::NoneType>()) {
|
||
auto maxPromoted = convertScalarToDtype(b, loc, max, dtype);
|
||
auto pred = b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::UGT,
|
||
result, maxPromoted);
|
||
result = b.create<arith::SelectOp>(loc, pred, maxPromoted, result);
|
||
}
|
||
return result;
|
||
}
|
||
if (auto rsub = dyn_cast<AtenRsubScalarOp>(op)) {
|
||
Type dtype = converter->convertType(rsub.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
if (!dtype.isa<mlir::FloatType>()) {
|
||
rsub.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
Value self = payloadArgs[0];
|
||
Value other = convertScalarToDtype(b, loc, operands[1], dtype);
|
||
Value alpha = convertScalarToDtype(b, loc, operands[2], dtype);
|
||
Value mult = b.create<arith::MulFOp>(loc, self, alpha);
|
||
return b.create<arith::SubFOp>(loc, other, mult);
|
||
}
|
||
if (auto mulScalar = dyn_cast<AtenMulScalarOp>(op)) {
|
||
Type dtype = converter->convertType(mulScalar.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value lhs = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value rhs = convertScalarToDtype(b, loc, operands[1], dtype);
|
||
if (dtype.isa<mlir::FloatType>())
|
||
return b.create<arith::MulFOp>(loc, lhs, rhs);
|
||
if (dtype.isa<mlir::IntegerType>())
|
||
return b.create<arith::MulIOp>(loc, lhs, rhs);
|
||
mulScalar.emitError("unimplemented: Only integer/float dtype supported");
|
||
return nullptr;
|
||
}
|
||
if (auto atenToDtype = dyn_cast<AtenToDtypeOp>(op)) {
|
||
Value input = payloadArgs[0];
|
||
Type dtype = converter->convertType(atenToDtype.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value result = convertScalarToDtype(b, loc, input, dtype);
|
||
return result;
|
||
}
|
||
if (auto divScalar = dyn_cast<AtenDivScalarOp>(op)) {
|
||
Type dtype = converter->convertType(divScalar.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
if (!dtype.isa<mlir::FloatType>()) {
|
||
divScalar.emitError("unimplemented: non-floating point dtype");
|
||
return nullptr;
|
||
}
|
||
Value self = payloadArgs[0];
|
||
Value other = convertScalarToDtype(b, loc, operands[1], dtype);
|
||
return b.create<arith::DivFOp>(loc, self, other);
|
||
}
|
||
if (auto reciprocal = dyn_cast<AtenReciprocalOp>(op)) {
|
||
Type dtype = converter->convertType(reciprocal.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
Value arg = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Type elementType = arg.getType();
|
||
// assert(element != 0)
|
||
auto zero =
|
||
b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, 0.0));
|
||
auto pred =
|
||
b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::ONE, arg, zero);
|
||
b.create<cf::AssertOp>(
|
||
loc, pred, b.getStringAttr("unimplemented: tensor with zero element"));
|
||
|
||
auto one =
|
||
b.create<arith::ConstantOp>(loc, FloatAttr::get(elementType, 1.0));
|
||
return b.create<arith::DivFOp>(loc, one, arg);
|
||
}
|
||
if (auto thresholdOp = dyn_cast<AtenThresholdOp>(op)) {
|
||
// The approach used here is as follows:
|
||
// result = self <= threshold ? value : self
|
||
AtenThresholdOp::Adaptor adaptor(operands);
|
||
Type dtype = converter->convertType(thresholdOp.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
|
||
Value self = payloadArgs[0];
|
||
Value threshold = convertScalarToDtype(b, loc, adaptor.threshold(), dtype);
|
||
Value value = convertScalarToDtype(b, loc, adaptor.value(), dtype);
|
||
|
||
Value predicate;
|
||
if (dtype.isa<mlir::FloatType>())
|
||
predicate = b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::ULE, self,
|
||
threshold);
|
||
else
|
||
predicate = b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::sle, self,
|
||
threshold);
|
||
return b.create<arith::SelectOp>(loc, predicate, value, self);
|
||
}
|
||
if (auto thresholdBackward = dyn_cast<AtenThresholdBackwardOp>(op)) {
|
||
// The approach used here is as follows:
|
||
// result = self <= threshold ? 0 : grad
|
||
AtenThresholdBackwardOp::Adaptor adaptor(operands);
|
||
Type dtype = converter->convertType(thresholdBackward.getType())
|
||
.cast<RankedTensorType>()
|
||
.getElementType();
|
||
|
||
Value grad = convertScalarToDtype(b, loc, payloadArgs[0], dtype);
|
||
Value self = convertScalarToDtype(b, loc, payloadArgs[1], dtype);
|
||
Value threshold = convertScalarToDtype(b, loc, adaptor.threshold(), dtype);
|
||
Value constantZero = b.create<arith::ConstantOp>(loc, b.getZeroAttr(dtype));
|
||
|
||
Value predicate;
|
||
if (dtype.isa<mlir::FloatType>())
|
||
predicate = b.create<arith::CmpFOp>(loc, arith::CmpFPredicate::ULE, self,
|
||
threshold);
|
||
else
|
||
predicate = b.create<arith::CmpIOp>(loc, arith::CmpIPredicate::sle, self,
|
||
threshold);
|
||
return b.create<arith::SelectOp>(loc, predicate, constantZero, grad);
|
||
}
|
||
|
||
op->emitError("unimplemented lowering in "
|
||
"createLinalgPayloadCalculationForElementwiseOp");
|
||
return nullptr;
|
||
}
|
||
|
||
namespace {
|
||
// Converts an elementwise op.
|
||
// This specifically includes:
|
||
// - converting elementwise ops of any tensor arity
|
||
// - converting elementwise ops with any number of scalar captures (such as a
|
||
// scalar alpha to torch.aten.Add)
|
||
// - broadcasting of static size-1 dimensions
|
||
//
|
||
// Currently, we adopt the behavior that "size 1" broadcasting is a runtime
|
||
// error if it happens dynamically.
|
||
//
|
||
// Looking forward a bit, eventually, it probably makes sense to have
|
||
// a "linalg.generic-like" op for modeling a fused subgraph of numpy-broadcasted
|
||
// operands. Modeling elementwise ops that way is potentially useful to allow a
|
||
// more centralized reasoning about multiversioning. However a cost model will
|
||
// be needed for "pre-fusing" elementwise ops that way, as it can potentially be
|
||
// a pessimization. A mild extension of this pattern should work for such a
|
||
// general op.
|
||
class ConvertElementwiseOp : public ConversionPattern {
|
||
public:
|
||
ConvertElementwiseOp(TypeConverter &typeConverter, MLIRContext *context)
|
||
: ConversionPattern(typeConverter, MatchAnyOpTypeTag(), /*benefit=*/1,
|
||
context) {}
|
||
|
||
LogicalResult
|
||
matchAndRewrite(Operation *op, ArrayRef<Value> operands,
|
||
ConversionPatternRewriter &rewriter) const override {
|
||
if (!isa<AtenTanhOp, AtenReluOp, AtenLeakyReluOp, AtenGeluOp,
|
||
AtenGeluBackwardOp, AtenAddTensorOp, AtenMulTensorOp,
|
||
AtenDivTensorOp, AtenSubTensorOp, AtenLerpTensorOp, AtenSigmoidOp,
|
||
AtenExpOp, AtenMinimumOp, AtenMaximumOp, AtenToDtypeOp,
|
||
AtenClampOp, AtenRsubScalarOp, AtenMulScalarOp, AtenLogOp,
|
||
AtenErfOp, AtenSqrtOp, AtenFloorOp, AtenPowTensorScalarOp,
|
||
AtenLog2Op, AtenRsqrtOp, AtenDivScalarOp, AtenAbsOp,
|
||
AtenReciprocalOp, AtenBitwiseAndTensorOp, AtenGtScalarOp,
|
||
AtenGeScalarOp, AtenEqScalarOp, AtenLtScalarOp, AtenLeScalarOp,
|
||
AtenWhereSelfOp, AtenCeilOp, AtenGtTensorOp, AtenEqTensorOp,
|
||
AtenLtTensorOp, AtenSubScalarOp, AtenAddScalarOp, AtenThresholdOp,
|
||
AtenThresholdBackwardOp, AtenCloneOp, AtenSinOp, AtenCosOp,
|
||
AtenNeScalarOp, AtenNegOp>(op))
|
||
return rewriter.notifyMatchFailure(op, "not a supported elementwise op");
|
||
|
||
if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
|
||
return failure();
|
||
|
||
Location loc = op->getLoc();
|
||
auto tensorOperands = llvm::to_vector<6>(llvm::make_filter_range(
|
||
operands, [](Value v) { return v.getType().isa<RankedTensorType>(); }));
|
||
auto resultType = getTypeConverter()
|
||
->convertType(op->getResult(0).getType())
|
||
.cast<RankedTensorType>();
|
||
bool hadErrorCreatingPayload = false;
|
||
Value generic = createElementwiseLinalgGeneric(
|
||
rewriter, loc, tensorOperands, resultType.getElementType(),
|
||
[&](OpBuilder &b, Location loc, ValueRange payloadArgs) {
|
||
Value result = createLinalgPayloadCalculationForElementwiseOp(
|
||
b, loc, getTypeConverter(), payloadArgs, op, operands);
|
||
if (!result) {
|
||
hadErrorCreatingPayload = true;
|
||
return;
|
||
}
|
||
b.create<linalg::YieldOp>(loc, result);
|
||
});
|
||
if (hadErrorCreatingPayload)
|
||
return failure();
|
||
rewriter.replaceOpWithNewOp<tensor::CastOp>(op, resultType, generic);
|
||
return success();
|
||
}
|
||
};
|
||
} // namespace
|
||
|
||
// Given `input`, `target`, `nll_loss_forward` is given by:
|
||
// for i in range(0, len(target)):
|
||
// indi = target[i];
|
||
// nll_loss_forward[i] = -(input[i][indi]);
|
||
// TODO: `weight`operand is still to be taken care of.
|
||
namespace {
|
||
class ConvertAtenNllLossForwardOp
|
||
: public OpConversionPattern<AtenNllLossForwardOp> {
|
||
public:
|
||
using OpConversionPattern::OpConversionPattern;
|
||
LogicalResult
|
||
matchAndRewrite(AtenNllLossForwardOp op, OpAdaptor adaptor,
|
||
ConversionPatternRewriter &rewriter) const override {
|
||
if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
|
||
return failure();
|
||
Location loc = op->getLoc();
|
||
Value input = adaptor.self();
|
||
Value target = adaptor.target();
|
||
Value weight = adaptor.weight();
|
||
|
||
int64_t reduction;
|
||
if (!matchPattern(op.reduction(), m_TorchConstantInt(&reduction)))
|
||
return rewriter.notifyMatchFailure(op, "dim must be constant");
|
||
|
||
// TODO: Incorporate the weight argument.
|
||
if (!weight.getType().isa<mlir::torch::Torch::NoneType>())
|
||
return rewriter.notifyMatchFailure(
|
||
op, "Unimplemented, the weight operand is not incorporated.");
|
||
|
||
Value ignoreIndex = adaptor.ignore_index();
|
||
Value ignoreIndexVal = castIntToIndex(rewriter, loc, ignoreIndex);
|
||
|
||
unsigned inputRank = input.getType().cast<RankedTensorType>().getRank();
|
||
unsigned targetRank = target.getType().cast<RankedTensorType>().getRank();
|
||
|
||
// TODO: Add support for k-dim loss.
|
||
if (inputRank > 2) {
|
||
return rewriter.notifyMatchFailure(
|
||
op, "expected input and target to be rank <= 2");
|
||
}
|
||
RankedTensorType resultType = getTypeConverter()
|
||
->convertType(op->getResult(0).getType())
|
||
.cast<RankedTensorType>();
|
||
Type elementType = resultType.getElementType();
|
||
|
||
Value zeroVal = rewriter.create<arith::ConstantOp>(
|
||
loc, rewriter.getZeroAttr(elementType));
|
||
|
||
Value finalRes = createElementwiseLinalgGeneric(
|
||
rewriter, loc, {target}, elementType,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value targetVal = args[0];
|
||
Value indTarget = rewriter.create<arith::IndexCastOp>(
|
||
loc, rewriter.getIndexType(), targetVal);
|
||
|
||
// The final result is given by:
|
||
// final_res = (indTarget == ignoreIndexVal) ? 0 :
|
||
// input[indI][IndTarget]
|
||
Value cmpEq = rewriter.create<arith::CmpIOp>(
|
||
loc, arith::CmpIPredicate::eq, indTarget, ignoreIndexVal);
|
||
|
||
SmallVector<Value> extractionIndices{indTarget};
|
||
if (inputRank == 2) {
|
||
Value indI = rewriter.create<linalg::IndexOp>(loc, 0);
|
||
extractionIndices.insert(extractionIndices.begin(), indI);
|
||
}
|
||
|
||
Value result =
|
||
rewriter.create<tensor::ExtractOp>(loc, input, extractionIndices);
|
||
|
||
Value negate =
|
||
rewriter.create<arith::NegFOp>(loc, elementType, result);
|
||
Value selectFinal =
|
||
rewriter.create<arith::SelectOp>(loc, cmpEq, zeroVal, negate);
|
||
b.create<linalg::YieldOp>(loc, selectFinal);
|
||
});
|
||
|
||
if (reduction == torch_upstream::Reduction::Sum ||
|
||
reduction == torch_upstream::Reduction::Mean) {
|
||
Value numOfElems = getTensorSize(rewriter, loc, finalRes);
|
||
numOfElems = convertScalarToDtype(rewriter, loc, numOfElems, elementType);
|
||
llvm::iota_range<int64_t> dimsToReduce(0, targetRank,
|
||
/*inclusive=*/false);
|
||
DenseSet<int64_t> dimSet(dimsToReduce.begin(), dimsToReduce.end());
|
||
|
||
finalRes = torch_to_linalg::createReductionLinalgGeneric(
|
||
rewriter, loc, finalRes, dimSet, /*keepDim=*/false,
|
||
/*initElem=*/zeroVal,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value newVal = args[0];
|
||
Value accumulator = args[1];
|
||
if (reduction == torch_upstream::Reduction::Mean)
|
||
newVal = b.create<arith::DivFOp>(loc, newVal, numOfElems);
|
||
Value result = b.create<arith::AddFOp>(loc, newVal, accumulator);
|
||
b.create<linalg::YieldOp>(loc, result);
|
||
});
|
||
}
|
||
|
||
// TODO: Update the second result tensor.
|
||
Value weightUpdated = createZeroInitTensor(rewriter, loc, {}, elementType);
|
||
rewriter.replaceOp(op, {finalRes, weightUpdated});
|
||
return success();
|
||
}
|
||
};
|
||
} // namespace
|
||
|
||
/// Inverted STD: rSTD = 1 / sqrt(var + eps).
|
||
static Value calculateRSTD(OpBuilder &b, Location loc, Type elemTy, Value eps,
|
||
Value var) {
|
||
// The eps is always f64.
|
||
Value truncatedEps = b.create<arith::TruncFOp>(loc, elemTy, eps);
|
||
Value varPlusEps = b.create<arith::AddFOp>(loc, var, truncatedEps);
|
||
Value rSTD = b.create<math::RsqrtOp>(loc, varPlusEps);
|
||
return rSTD;
|
||
}
|
||
|
||
// Normalization formula:
|
||
// ((input - mean) * rSTD * weight + bias
|
||
static Value createLinalgPayloadCalculationForNormOpsWithRSTD(
|
||
OpBuilder &b, Location loc, Type elemTy, Value input, Value mean,
|
||
Value rSTD, Value eps, Value weight, Value bias) {
|
||
Value inputSubMean = b.create<arith::SubFOp>(loc, input, mean);
|
||
Value temp = b.create<arith::MulFOp>(loc, inputSubMean, rSTD);
|
||
Value timesWeight = b.create<arith::MulFOp>(loc, temp, weight);
|
||
Value plusBias = b.create<arith::AddFOp>(loc, timesWeight, bias);
|
||
return plusBias;
|
||
}
|
||
|
||
static Value createLinalgPayloadCalculationForNormOpsWithVar(
|
||
OpBuilder &b, Location loc, Type elemTy, Value input, Value mean, Value var,
|
||
Value eps, Value weight, Value bias) {
|
||
Value rSTD = calculateRSTD(b, loc, elemTy, eps, var);
|
||
Value result = createLinalgPayloadCalculationForNormOpsWithRSTD(
|
||
b, loc, elemTy, input, mean, rSTD, eps, weight, bias);
|
||
return result;
|
||
}
|
||
|
||
namespace {
|
||
class ConvertAtenBatchNormOp : public OpConversionPattern<AtenBatchNormOp> {
|
||
public:
|
||
using OpConversionPattern::OpConversionPattern;
|
||
LogicalResult
|
||
matchAndRewrite(AtenBatchNormOp op, OpAdaptor adaptor,
|
||
ConversionPatternRewriter &rewriter) const override {
|
||
MLIRContext *context = op->getContext();
|
||
Location loc = op->getLoc();
|
||
Value input = adaptor.input();
|
||
Value weight = adaptor.weight();
|
||
Value bias = adaptor.bias();
|
||
Value runningMean = adaptor.running_mean();
|
||
Value runningVar = adaptor.running_var();
|
||
Value training = adaptor.training();
|
||
Value eps = adaptor.eps();
|
||
|
||
if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
|
||
return failure();
|
||
|
||
// TODO: Handle the None cases for the optional parameters:
|
||
// weight, bias.
|
||
if (failed(checkNotNone(rewriter, op, weight)) ||
|
||
failed(checkNotNone(rewriter, op, bias)) ||
|
||
failed(checkNotNone(rewriter, op, runningMean)) ||
|
||
failed(checkNotNone(rewriter, op, runningVar)))
|
||
return failure();
|
||
|
||
auto inputType = input.getType().cast<RankedTensorType>();
|
||
auto weightType = weight.getType().cast<RankedTensorType>();
|
||
auto biasType = bias.getType().cast<RankedTensorType>();
|
||
auto runningMeanType = runningMean.getType().cast<RankedTensorType>();
|
||
auto runningVarType = runningVar.getType().cast<RankedTensorType>();
|
||
|
||
auto inputRank = inputType.getRank();
|
||
if (inputRank <= 2)
|
||
return rewriter.notifyMatchFailure(
|
||
op, "input should have rank larger than 2");
|
||
|
||
if (weightType.getRank() != 1 || biasType.getRank() != 1 ||
|
||
runningMeanType.getRank() != 1 || runningVarType.getRank() != 1) {
|
||
return rewriter.notifyMatchFailure(
|
||
op, "expect weight, bias, running_mean and running_var to be rank 1");
|
||
}
|
||
|
||
// TODO: Add support for training.
|
||
auto constFalse = rewriter.create<arith::ConstantOp>(
|
||
loc, IntegerAttr::get(IntegerType::get(context, 1), 0));
|
||
auto trainingFalse = rewriter.create<arith::CmpIOp>(
|
||
loc, arith::CmpIPredicate::eq, training, constFalse);
|
||
rewriter.create<cf::AssertOp>(
|
||
loc, trainingFalse,
|
||
rewriter.getStringAttr("training is not supported for now"));
|
||
|
||
// num_features – C from an expected input of size (N,C,D,H,W ...)
|
||
Value numFeatures = rewriter.create<tensor::DimOp>(loc, input, 1);
|
||
auto contractingDim0EqualsNumFeatures = [&](Value v) {
|
||
auto dim0 = rewriter.create<tensor::DimOp>(loc, v, 0);
|
||
auto dim0Equal = rewriter.create<arith::CmpIOp>(
|
||
loc, arith::CmpIPredicate::eq, numFeatures, dim0);
|
||
rewriter.create<cf::AssertOp>(
|
||
loc, dim0Equal,
|
||
rewriter.getStringAttr(
|
||
"expect the size of dim 0 equal to the number of features"));
|
||
};
|
||
contractingDim0EqualsNumFeatures(weight);
|
||
contractingDim0EqualsNumFeatures(bias);
|
||
contractingDim0EqualsNumFeatures(runningMean);
|
||
contractingDim0EqualsNumFeatures(runningVar);
|
||
|
||
auto indexingMap = AffineMap::get(
|
||
/*dimCount=*/inputRank,
|
||
/*symbolCount=*/0, rewriter.getAffineDimExpr(1), context);
|
||
SmallVector<AffineMap> indexingMaps = {
|
||
rewriter.getMultiDimIdentityMap(inputRank), // input
|
||
indexingMap, // weight
|
||
indexingMap, // bias
|
||
indexingMap, // runningMean
|
||
indexingMap, // runningVar
|
||
rewriter.getMultiDimIdentityMap(inputRank), // output
|
||
};
|
||
SmallVector<StringRef> iteratorTypes(inputRank, "parallel");
|
||
Value batchNorm =
|
||
rewriter
|
||
.create<linalg::GenericOp>(
|
||
loc, input.getType(),
|
||
ValueRange{input, weight, bias, runningMean, runningVar}, input,
|
||
/*indexingMaps=*/indexingMaps,
|
||
/*iteratorTypes=*/iteratorTypes,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value input = args[0], weight = args[1], bias = args[2],
|
||
mean = args[3], var = args[4];
|
||
Value result =
|
||
createLinalgPayloadCalculationForNormOpsWithVar(
|
||
b, loc, var.getType(), input, mean, var, eps, weight,
|
||
bias);
|
||
b.create<linalg::YieldOp>(loc, result);
|
||
})
|
||
.getResult(0);
|
||
Type newResultType = getTypeConverter()->convertType(op.getType());
|
||
rewriter.replaceOpWithNewOp<tensor::CastOp>(op, newResultType, batchNorm);
|
||
return success();
|
||
}
|
||
};
|
||
} // namespace
|
||
|
||
// For layernorm, the mean and standard-deviation are calculated separately over
|
||
// the last certain number dimensions which have to be of the shape specified by
|
||
// normalized_shape.
|
||
//
|
||
// The shapes of different parts are as the following:
|
||
// +-------------------+--------------------+
|
||
// | meanAndVarShape | normalizedShape |
|
||
// +-------------------+---------------------
|
||
// <------------+ inputShape +-------------->
|
||
// There are the following steps:
|
||
// Step 1. Check if all the arguments meet the requirements.
|
||
// Step 2. Common parts to be used for getting mean and var.
|
||
// This includes elements count, affineMap and iteratorTypes.
|
||
// Step 3. Get mean.
|
||
// Step 4. Get rSTD.
|
||
// Step 5. Get layernorm.
|
||
namespace {
|
||
class ConvertAtenNativeLayerNormOp
|
||
: public OpConversionPattern<AtenNativeLayerNormOp> {
|
||
public:
|
||
using OpConversionPattern::OpConversionPattern;
|
||
LogicalResult
|
||
matchAndRewrite(AtenNativeLayerNormOp op, OpAdaptor adaptor,
|
||
ConversionPatternRewriter &rewriter) const override {
|
||
MLIRContext *context = op->getContext();
|
||
Location loc = op->getLoc();
|
||
Value input = adaptor.input();
|
||
Value weight = adaptor.weight();
|
||
Value bias = adaptor.bias();
|
||
Value eps = adaptor.eps();
|
||
Value normalizedShape = op.normalized_shape();
|
||
|
||
if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
|
||
return failure();
|
||
|
||
// TODO: Handle the None cases for the optional parameters:
|
||
// weight, bias.
|
||
if (failed(checkNotNone(rewriter, op, weight)) ||
|
||
failed(checkNotNone(rewriter, op, bias)))
|
||
return failure();
|
||
|
||
auto inputType = input.getType().cast<RankedTensorType>();
|
||
auto weightType = weight.getType().cast<RankedTensorType>();
|
||
auto biasType = bias.getType().cast<RankedTensorType>();
|
||
int64_t inputRank = inputType.getRank();
|
||
Type elemTy = inputType.getElementType();
|
||
|
||
// Step 1. Check if all the arguments meet the requirements.
|
||
SmallVector<Value> normalizedShapeSizesTorchInt;
|
||
if (!getListConstructElements(normalizedShape,
|
||
normalizedShapeSizesTorchInt)) {
|
||
return rewriter.notifyMatchFailure(op,
|
||
"Unimplemented normalized_shape not"
|
||
"constructed from ListConstruct");
|
||
}
|
||
SmallVector<Value> normalizedShapeSizesInt = getTypeConvertedValues(
|
||
rewriter, loc, getTypeConverter(), normalizedShapeSizesTorchInt);
|
||
int64_t normalizedShapeRank = normalizedShapeSizesInt.size();
|
||
if (weightType.getRank() != normalizedShapeRank ||
|
||
biasType.getRank() != normalizedShapeRank ||
|
||
inputRank < normalizedShapeRank || normalizedShapeRank < 1)
|
||
return rewriter.notifyMatchFailure(op, "Input or weight or bias shape or"
|
||
"normalized shape not compatible");
|
||
|
||
// Check all the dimensions match the normalized_shape
|
||
int64_t meanAndVarShapeRank = inputRank - normalizedShapeSizesInt.size();
|
||
for (auto en : enumerate((normalizedShapeSizesInt))) {
|
||
auto index = en.index();
|
||
auto inputDim =
|
||
getDimOp(rewriter, loc, input, index + meanAndVarShapeRank);
|
||
auto weightDim = getDimOp(rewriter, loc, weight, index);
|
||
auto biasDim = getDimOp(rewriter, loc, bias, index);
|
||
|
||
auto expectedSize = en.value();
|
||
checkDimEqualHelper(rewriter, loc, inputDim, expectedSize);
|
||
checkDimEqualHelper(rewriter, loc, weightDim, expectedSize);
|
||
checkDimEqualHelper(rewriter, loc, biasDim, expectedSize);
|
||
}
|
||
|
||
// Get iterator types for input shape.
|
||
SmallVector<StringRef> normalizedShapeIteratorTypes(
|
||
normalizedShapeRank, getReductionIteratorTypeName());
|
||
SmallVector<StringRef> meanAndVarIterationTypes(
|
||
meanAndVarShapeRank, getParallelIteratorTypeName());
|
||
SmallVector<StringRef> inputShapeIteratorTypes = meanAndVarIterationTypes;
|
||
inputShapeIteratorTypes.append(normalizedShapeIteratorTypes);
|
||
|
||
// Step 2. Common parts to be used for getting mean and var.
|
||
|
||
// Get sizes and affineMaps needed for mean and var.
|
||
AffineMap inputShapeAffineMap = rewriter.getMultiDimIdentityMap(inputRank);
|
||
SmallVector<AffineExpr> meanAndVarShapeExprs;
|
||
for (int i = 0; i < meanAndVarShapeRank; i++)
|
||
meanAndVarShapeExprs.push_back(mlir::getAffineDimExpr(i, context));
|
||
auto meanAndVarShapeAffineMap = AffineMap::get(
|
||
/*dimCount=*/inputRank,
|
||
/*symbolCount=*/0, meanAndVarShapeExprs, context);
|
||
SmallVector<Value> meanAndVarShapeSizes =
|
||
getTensorSizesUntilDim(rewriter, loc, input, meanAndVarShapeRank - 1);
|
||
|
||
// Get number of elements to be used for calculating mean and var.
|
||
Value elemCnts = normalizedShapeSizesInt[0];
|
||
for (int i = 1; i < normalizedShapeRank; i++) {
|
||
elemCnts = rewriter.create<arith::MulIOp>(loc, elemCnts,
|
||
normalizedShapeSizesInt[i]);
|
||
}
|
||
Value elemCntsFloat =
|
||
rewriter.create<arith::SIToFPOp>(loc, elemTy, elemCnts);
|
||
|
||
// Helper to calculate mean and var.
|
||
auto genMeanOrVarCalculation = [&](Value sumOrSquareSum) {
|
||
SmallVector<AffineMap> indexingMaps(
|
||
2, rewriter.getMultiDimIdentityMap(meanAndVarShapeRank));
|
||
Value initShapeTensor = rewriter.create<linalg::InitTensorOp>(
|
||
loc, meanAndVarShapeSizes, elemTy);
|
||
return rewriter
|
||
.create<linalg::GenericOp>(
|
||
loc, initShapeTensor.getType(), sumOrSquareSum, initShapeTensor,
|
||
/*indexingMaps=*/indexingMaps,
|
||
/*iteratorTypes=*/meanAndVarIterationTypes,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value sumOrSqureSum = args[0];
|
||
Value result =
|
||
b.create<arith::DivFOp>(loc, sumOrSqureSum, elemCntsFloat);
|
||
b.create<linalg::YieldOp>(loc, result);
|
||
})
|
||
.getResult(0);
|
||
};
|
||
|
||
// Step 3. Get mean.
|
||
|
||
// Get sum to be used for calculating mean.
|
||
SmallVector<AffineMap, 2> sumIndexingMaps = {
|
||
inputShapeAffineMap, // input
|
||
meanAndVarShapeAffineMap, // output
|
||
};
|
||
auto initSumTensor =
|
||
createZeroInitTensor(rewriter, loc, meanAndVarShapeSizes, elemTy);
|
||
Value sum = rewriter
|
||
.create<linalg::GenericOp>(
|
||
loc, initSumTensor.getType(), input, initSumTensor,
|
||
/*indexingMaps=*/sumIndexingMaps,
|
||
/*iteratorTypes=*/inputShapeIteratorTypes,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value input = args[0], sum = args[1];
|
||
Value result =
|
||
rewriter.create<arith::AddFOp>(loc, sum, input);
|
||
b.create<linalg::YieldOp>(loc, result);
|
||
})
|
||
.getResult(0);
|
||
Value mean = genMeanOrVarCalculation(sum);
|
||
|
||
// Step 4. Get rSTD.
|
||
|
||
// Calculate squareSum for the layer.
|
||
SmallVector<AffineMap> squareSumIndexingMaps{
|
||
inputShapeAffineMap,
|
||
meanAndVarShapeAffineMap,
|
||
meanAndVarShapeAffineMap,
|
||
};
|
||
auto initSquareSumTensor =
|
||
createZeroInitTensor(rewriter, loc, meanAndVarShapeSizes, elemTy);
|
||
Value squareSum =
|
||
rewriter
|
||
.create<linalg::GenericOp>(
|
||
loc, initSquareSumTensor.getType(), ValueRange{input, mean},
|
||
initSquareSumTensor,
|
||
/*indexingMaps=*/squareSumIndexingMaps,
|
||
/*iteratorTypes=*/inputShapeIteratorTypes,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value input = args[0], mean = args[1], squareSum = args[2];
|
||
Value sub = rewriter.create<arith::SubFOp>(loc, input, mean);
|
||
Value square = rewriter.create<arith::MulFOp>(loc, sub, sub);
|
||
Value result =
|
||
rewriter.create<arith::AddFOp>(loc, squareSum, square);
|
||
b.create<linalg::YieldOp>(loc, result);
|
||
})
|
||
.getResult(0);
|
||
Value var = genMeanOrVarCalculation(squareSum);
|
||
Value rSTDTensor = rewriter.create<linalg::InitTensorOp>(
|
||
loc, meanAndVarShapeSizes, elemTy);
|
||
SmallVector<AffineMap> rSTDIndexingMap(
|
||
2, rewriter.getMultiDimIdentityMap(meanAndVarShapeRank));
|
||
|
||
Value rSTD = rewriter
|
||
.create<linalg::GenericOp>(
|
||
loc, rSTDTensor.getType(), var, rSTDTensor,
|
||
rSTDIndexingMap, meanAndVarIterationTypes,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value result =
|
||
calculateRSTD(b, loc, elemTy, eps, args[0]);
|
||
b.create<linalg::YieldOp>(loc, result);
|
||
})
|
||
.getResult(0);
|
||
|
||
// Step 5. Get layernorm.
|
||
|
||
// Get affineMap for normalized shape.
|
||
SmallVector<AffineExpr> normalizedShapeExprs;
|
||
for (int i = meanAndVarShapeRank; i < inputRank; i++)
|
||
normalizedShapeExprs.push_back(mlir::getAffineDimExpr(i, context));
|
||
auto normalizedShapeAffineMap = AffineMap::get(
|
||
/*dimCount=*/inputRank,
|
||
/*symbolCount=*/0, normalizedShapeExprs, context);
|
||
auto inputSizes = getTensorSizes(rewriter, loc, input);
|
||
Value initLayerNormTensor =
|
||
rewriter.create<linalg::InitTensorOp>(loc, inputSizes, elemTy);
|
||
SmallVector<AffineMap> indexingMaps(1, inputShapeAffineMap);
|
||
indexingMaps.resize(3, meanAndVarShapeAffineMap);
|
||
indexingMaps.resize(5, normalizedShapeAffineMap);
|
||
indexingMaps.push_back(inputShapeAffineMap);
|
||
SmallVector<StringRef> layerNormIterationTypes(
|
||
inputRank, getParallelIteratorTypeName());
|
||
Value layerNorm =
|
||
rewriter
|
||
.create<linalg::GenericOp>(
|
||
loc, initLayerNormTensor.getType(),
|
||
ValueRange{input, mean, rSTD, weight, bias},
|
||
initLayerNormTensor,
|
||
/*indexingMaps=*/indexingMaps,
|
||
/*iteratorTypes=*/layerNormIterationTypes,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value input = args[0], mean = args[1], rSTD = args[2],
|
||
weight = args[3], bias = args[4];
|
||
Value result =
|
||
createLinalgPayloadCalculationForNormOpsWithRSTD(
|
||
b, loc, elemTy, input, mean, rSTD, eps, weight, bias);
|
||
b.create<linalg::YieldOp>(loc, result);
|
||
})
|
||
.getResult(0);
|
||
SmallVector<int64_t> expandShape(inputRank, 1);
|
||
for (int i = 0; i < meanAndVarShapeRank; i++) {
|
||
// `mean` and `rstd` are not yet casted, so they will be having dynamic
|
||
// shape. Hence to match them, for each dimension corresponding to `mean`
|
||
// or `rstd` assign -1.
|
||
expandShape[i] = -1;
|
||
}
|
||
auto expandShapeType = RankedTensorType::get(expandShape, elemTy);
|
||
SmallVector<ReassociationIndices> reassociation(meanAndVarShapeRank);
|
||
for (auto i : llvm::seq<int64_t>(0, meanAndVarShapeRank)) {
|
||
reassociation[i].push_back(i);
|
||
if (i == meanAndVarShapeRank - 1) {
|
||
for (auto j : llvm::seq<int64_t>(0, normalizedShapeRank))
|
||
reassociation[i].push_back(i + j + 1);
|
||
}
|
||
}
|
||
Value meanResult = rewriter.create<tensor::ExpandShapeOp>(
|
||
loc, expandShapeType, mean, reassociation);
|
||
Value rSTDResult = rewriter.create<tensor::ExpandShapeOp>(
|
||
loc, expandShapeType, rSTD, reassociation);
|
||
Type layerNormResultType = getTypeConverter()->convertType(op.getType(0));
|
||
Type meanResultType = getTypeConverter()->convertType(op.getType(1));
|
||
Type rSTDResultType = getTypeConverter()->convertType(op.getType(2));
|
||
Value layerNorm_ =
|
||
rewriter.create<tensor::CastOp>(loc, layerNormResultType, layerNorm);
|
||
Value mean_ =
|
||
rewriter.create<tensor::CastOp>(loc, meanResultType, meanResult);
|
||
Value var_ =
|
||
rewriter.create<tensor::CastOp>(loc, rSTDResultType, rSTDResult);
|
||
rewriter.replaceOp(op, {layerNorm_, mean_, var_});
|
||
return success();
|
||
}
|
||
};
|
||
} // namespace
|
||
|
||
namespace {
|
||
class ConvertAtenNllLossBackwardOp
|
||
: public OpConversionPattern<AtenNllLossBackwardOp> {
|
||
public:
|
||
using OpConversionPattern::OpConversionPattern;
|
||
LogicalResult
|
||
matchAndRewrite(AtenNllLossBackwardOp op, OpAdaptor adaptor,
|
||
ConversionPatternRewriter &rewriter) const override {
|
||
if (failed(verifyLinalgCompatibleTypes(op, rewriter)))
|
||
return failure();
|
||
|
||
Location loc = op->getLoc();
|
||
Value gradOutput = adaptor.grad_output();
|
||
Value input = adaptor.self();
|
||
Value target = adaptor.target();
|
||
Value weight = adaptor.weight();
|
||
bool weightIsNone = op.weight().getType().isa<Torch::NoneType>();
|
||
Value ignoreIndex = castIntToIndex(rewriter, loc, adaptor.ignore_index());
|
||
Value totalWeight = adaptor.total_weight();
|
||
|
||
auto inputType = input.getType().cast<RankedTensorType>();
|
||
int inputRank = inputType.getRank();
|
||
auto gradOutputType = gradOutput.getType().cast<RankedTensorType>();
|
||
Type resultElementType = gradOutputType.getElementType();
|
||
|
||
int64_t reduction;
|
||
if (!matchPattern(op.reduction(), m_TorchConstantInt(&reduction)))
|
||
return rewriter.notifyMatchFailure(op, "dim must be constant");
|
||
|
||
if (!hasElementType<mlir::FloatType>(gradOutput) ||
|
||
!hasElementType<mlir::FloatType>(gradOutput) ||
|
||
(!weightIsNone && !hasElementType<mlir::FloatType>(weight))) {
|
||
return rewriter.notifyMatchFailure(
|
||
op, "`gradOutput`, 'weight', and `totalWeight` must be tensors of "
|
||
"type float");
|
||
}
|
||
|
||
if (!hasElementType<mlir::IntegerType>(target)) {
|
||
return rewriter.notifyMatchFailure(
|
||
op, "`target` must be a tensor of integer type");
|
||
}
|
||
|
||
auto outputSize = getTensorSizes(rewriter, loc, input);
|
||
Value gradInputTensor =
|
||
createZeroInitTensor(rewriter, loc, outputSize, resultElementType);
|
||
|
||
auto getAffineMapForSingleElementTensor = [&](Value tensor) {
|
||
auto tensorType = tensor.getType().cast<RankedTensorType>();
|
||
SmallVector<AffineExpr> affineExprs(tensorType.getRank(),
|
||
rewriter.getAffineConstantExpr(0));
|
||
return AffineMap::get(inputRank, /*symbolCount=*/0, affineExprs,
|
||
op->getContext());
|
||
};
|
||
|
||
AffineMap gradOutMap = AffineMap::get(inputRank, /*symbolCount=*/0,
|
||
rewriter.getAffineDimExpr(0));
|
||
if (reduction != torch_upstream::Reduction::None || inputRank == 1)
|
||
gradOutMap = getAffineMapForSingleElementTensor(gradOutput);
|
||
AffineMap targetMap = AffineMap::get(inputRank, /*symbolCount=*/0,
|
||
rewriter.getAffineDimExpr(0));
|
||
if (inputRank == 1)
|
||
targetMap = getAffineMapForSingleElementTensor(target);
|
||
AffineMap totalWeightMap = getAffineMapForSingleElementTensor(totalWeight);
|
||
AffineMap resultMap = rewriter.getMultiDimIdentityMap(inputRank);
|
||
|
||
SmallVector<AffineMap> indexingMaps{gradOutMap, targetMap, totalWeightMap,
|
||
resultMap};
|
||
SmallVector<StringRef> iteratorTypes(inputRank,
|
||
getParallelIteratorTypeName());
|
||
|
||
// The code generation is equivalent to the following pseudo-code:
|
||
//
|
||
// for batch_index in len(input.size(0)):
|
||
// for class_index in len(input.size(1)):
|
||
// target_elem = target[batch_index]
|
||
//
|
||
// if reduction == None:
|
||
// grad_out_elem = grad_output[batchIndex]
|
||
// else:
|
||
// grad_out_elem = grad_output[0]
|
||
//
|
||
// if reduction == Mean:
|
||
// total_weight_elem = total_weight[0]
|
||
// grad_out_elem /= total_weight_elem
|
||
//
|
||
// weight_elem = weight[target_elem] if weight != None else 1
|
||
//
|
||
// if target_elem != class_index or target_elem == ignore_index:
|
||
// grad_input_elem = -weight_elem * grad_out_elem
|
||
// else:
|
||
// grad_input_elem = 0
|
||
// grad_input[batch_index, target_elem] = grad_input_elem
|
||
//
|
||
// NOTE: In the case of not batch dimension, `batch_index` essentially
|
||
// becomes zero.
|
||
Value gradInput =
|
||
rewriter
|
||
.create<linalg::GenericOp>(
|
||
loc, gradInputTensor.getType(),
|
||
ValueRange{gradOutput, target, totalWeight}, gradInputTensor,
|
||
indexingMaps, iteratorTypes,
|
||
[&](OpBuilder &b, Location loc, ValueRange args) {
|
||
Value gradOutElem = args[0];
|
||
Value targetElem = castIntToIndex(b, loc, args[1]);
|
||
Value totalWeightElem = args[2];
|
||
Value classIndex =
|
||
b.create<linalg::IndexOp>(loc, inputRank - 1);
|
||
|
||
if (reduction == torch_upstream::Reduction::Mean) {
|
||
gradOutElem = b.create<arith::DivFOp>(loc, gradOutElem,
|
||
totalWeightElem);
|
||
}
|
||
|
||
Value negGradOutElem =
|
||
b.create<arith::NegFOp>(loc, gradOutElem);
|
||
Value weightElem = getConstant(b, loc, 1, resultElementType);
|
||
if (!weightIsNone) {
|
||
weightElem =
|
||
b.create<tensor::ExtractOp>(loc, weight, targetElem);
|
||
}
|
||
Value weightedNegGradOutElem =
|
||
b.create<arith::MulFOp>(loc, weightElem, negGradOutElem);
|
||
|
||
Value targetNeqClassIndex = b.create<arith::CmpIOp>(
|
||
loc, arith::CmpIPredicate::ne, targetElem, classIndex);
|
||
Value targetEqIgnoreIndex = b.create<arith::CmpIOp>(
|
||
loc, arith::CmpIPredicate::eq, targetElem, ignoreIndex);
|
||
Value gradInputIsZero = b.create<arith::OrIOp>(
|
||
loc, targetNeqClassIndex, targetEqIgnoreIndex);
|
||
|
||
Value zero = getConstant(b, loc, 0, resultElementType);
|
||
Value gradInElem = b.create<arith::SelectOp>(
|
||
loc, gradInputIsZero, zero, weightedNegGradOutElem);
|
||
b.create<linalg::YieldOp>(loc, gradInElem);
|
||
})
|
||
->getResult(0);
|
||
|
||
RankedTensorType resultType = getTypeConverter()
|
||
->convertType(op->getResult(0).getType())
|
||
.cast<RankedTensorType>();
|
||
rewriter.replaceOpWithNewOp<tensor::CastOp>(op, resultType, gradInput);
|
||
return success();
|
||
}
|
||
};
|
||
} // namespace
|
||
|
||
namespace {
|
||
class ConvertTensorStaticInfoCastOp
|
||
: public OpConversionPattern<TensorStaticInfoCastOp> {
|
||
public:
|
||
using OpConversionPattern::OpConversionPattern;
|
||
LogicalResult
|
||
matchAndRewrite(TensorStaticInfoCastOp op, OpAdaptor adaptor,
|
||
ConversionPatternRewriter &rewriter) const override {
|
||
RankedTensorType resultType = getTypeConverter()
|
||
->convertType(op->getResult(0).getType())
|
||
.cast<RankedTensorType>();
|
||
rewriter.replaceOpWithNewOp<tensor::CastOp>(op, resultType,
|
||
adaptor.operand());
|
||
return success();
|
||
}
|
||
};
|
||
} // namespace
|
||
|
||
void mlir::torch::torch_to_linalg::populateUncategorizedPatternsAndLegality(
|
||
TypeConverter &typeConverter, RewritePatternSet &patterns,
|
||
ConversionTarget &target) {
|
||
MLIRContext *context = patterns.getContext();
|
||
target.addIllegalOp<
|
||
AtenTanhOp, AtenReluOp, AtenLeakyReluOp, AtenGeluOp, AtenGeluBackwardOp,
|
||
AtenAddTensorOp, AtenMulTensorOp, AtenDivTensorOp, AtenSubTensorOp,
|
||
AtenLerpTensorOp, AtenSigmoidOp, AtenMinimumOp, AtenMaximumOp,
|
||
AtenToDtypeOp, AtenClampOp, AtenRsubScalarOp, AtenLogOp, AtenErfOp,
|
||
AtenSqrtOp, AtenFloorOp, AtenCeilOp, AtenPowTensorScalarOp, AtenLog2Op,
|
||
AtenRsqrtOp, AtenAbsOp, AtenReciprocalOp, AtenBitwiseAndTensorOp,
|
||
AtenGtScalarOp, AtenGeScalarOp, AtenEqScalarOp, AtenLtScalarOp,
|
||
AtenLeScalarOp, AtenWhereSelfOp, AtenGtTensorOp, AtenEqTensorOp,
|
||
AtenLtTensorOp, AtenThresholdOp, AtenThresholdBackwardOp, AtenCloneOp,
|
||
AtenSinOp, AtenCosOp, AtenNeScalarOp>();
|
||
patterns.add<ConvertElementwiseOp>(typeConverter, context);
|
||
target.addIllegalOp<AtenNllLossForwardOp>();
|
||
patterns.add<ConvertAtenNllLossForwardOp>(typeConverter, context);
|
||
target.addIllegalOp<AtenBatchNormOp>();
|
||
patterns.add<ConvertAtenBatchNormOp>(typeConverter, context);
|
||
target.addIllegalOp<AtenNativeLayerNormOp>();
|
||
patterns.add<ConvertAtenNativeLayerNormOp>(typeConverter, context);
|
||
target.addIllegalOp<AtenNllLossBackwardOp>();
|
||
patterns.add<ConvertAtenNllLossBackwardOp>(typeConverter, context);
|
||
patterns.add<ConvertTensorStaticInfoCastOp>(typeConverter, context);
|
||
target.addIllegalOp<TensorStaticInfoCastOp>();
|
||
}
|