UniformSupport.h

Go to the documentation of this file.

//===- UniformSupport.h - Support utilities for uniform quant ---*- 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
//
//===----------------------------------------------------------------------===//

#ifndef MLIR_DIALECT_QUANT_UNIFORMSUPPORT_H_
#define MLIR_DIALECT_QUANT_UNIFORMSUPPORT_H_

#include <utility>

#include "mlir/Dialect/Quant/QuantTypes.h"
#include "mlir/IR/BuiltinTypes.h"
#include "mlir/IR/Types.h"
#include "llvm/ADT/APFloat.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/APSInt.h"

namespace mlir {
namespace quant {

/// Performs type conversion from an arbitrary input type to a type
/// that is expressed by a QuantizedType.
///
/// This handles cases where the inputType is a supported primitive type
/// (i.e. f32, bf16, etc) or a vector/tensor type based on a supported
/// elemental type.
///
/// Since conversion often involves introspecting some attributes of the
/// input type in order to determine how to represent it, this is a two step
/// process.
struct ExpressedToQuantizedConverter {
  /// Creates a converter for the given input type.
  static ExpressedToQuantizedConverter forInputType(Type inputType);

  /// Converts the inputType to be based on the given elemental type,
  /// returning the new type (or nullptr and emit an error on failure).
  Type convert(QuantizedType elementalType) const;

  /// Whether the conversion is legal.
  explicit operator bool() const { return (bool)expressedType; }

  /// The input type that is being converted from.
  /// This may be an elemental or composite type.
  const Type inputType;

  /// Supported, elemental expressed type (i.e. f32).
  /// Will be nullptr if conversion is not supported.
  const Type expressedType;
};

/// Reference implementation of converting between real numbers and values
/// represented by a UniformQuantizedType.
/// Note that this is not expected to be speedy and may be superseded eventually
/// by a more optimal implementation.
/// Also, the interface assumes that quantization is done per-layer and will
/// need to be wider for various per-channel schemes. As such, this is a
/// placeholder.
class UniformQuantizedValueConverter {
public:
  explicit UniformQuantizedValueConverter(UniformQuantizedType uniformType)
      : UniformQuantizedValueConverter(
            uniformType.getScale(),
            static_cast<double>(uniformType.getZeroPoint()),
            static_cast<double>(uniformType.getStorageTypeMin()),
            static_cast<double>(uniformType.getStorageTypeMax()),
            uniformType.getStorageTypeIntegralWidth(), uniformType.isSigned()) {
    assert(uniformType.getExpressedType().isa<FloatType>());
    assert(uniformType.getStorageType().isSignlessInteger());
  }

  UniformQuantizedValueConverter(double scale, double zeroPoint,
                                 double clampMin, double clampMax,
                                 uint32_t storageBitWidth, bool isSigned)
      : scale(scale), zeroPoint(zeroPoint), clampMin(clampMin),
        clampMax(clampMax), scaleDouble(scale), zeroPointDouble(zeroPoint),
        clampMinDouble(clampMin), clampMaxDouble(clampMax),
        storageBitWidth(storageBitWidth), isSigned(isSigned),
        roundMode(APFloat::rmNearestTiesToAway) {}

  UniformQuantizedValueConverter(double scale, double zeroPoint,
                                 const APFloat &clampMin,
                                 const APFloat &clampMax,
                                 uint32_t storageBitWidth, bool isSigned)
      : scale(scale), zeroPoint(zeroPoint), clampMin(clampMin),
        clampMax(clampMax), scaleDouble(scale), zeroPointDouble(zeroPoint),
        clampMinDouble(clampMin.convertToDouble()),
        clampMaxDouble(clampMax.convertToDouble()),
        storageBitWidth(storageBitWidth), isSigned(isSigned),
        roundMode(APFloat::rmNearestTiesToAway) {}

  virtual APInt quantizeFloatToInt(APFloat expressedValue) const {
    // This function is a performance critical code path in quantization
    // since it runs for each single float parameter value.

    // Specialize f32->u8/i8 case to optimize performance.
    if (&expressedValue.getSemantics() == &APFloat::IEEEsingle() &&
        storageBitWidth == 8 &&
        roundMode == llvm::APFloatBase::rmNearestTiesToAway) {
      return quantizeF32ToInt8(expressedValue);
    }

    bool lossy;
    expressedValue.convert(scale.getSemantics(), roundMode, &lossy);
    // fixedpoint = clamp(clampMin, clampMax, (
    //   roundHalfToEven(expressed / scale) + zeroPoint))
    APFloat scaled = (expressedValue / scale);
    scaled.roundToIntegral(roundMode);
    scaled.add(zeroPoint, roundMode);
    APFloat fixedpoint = llvm::minimum(scaled, clampMax);
    fixedpoint = llvm::maximum(fixedpoint, clampMin);

    llvm::APSInt result(storageBitWidth, !isSigned);
    fixedpoint.convertToInteger(result, roundMode, &lossy);

    return std::move(result);
  }

  int64_t quantizeFloatToInt64(APFloat expressedValue) const {
    APInt qValue = quantizeFloatToInt(std::move(expressedValue));
    return isSigned ? qValue.getSExtValue() : qValue.getZExtValue();
  }

  virtual ~UniformQuantizedValueConverter() = default;

private:
  // An optimized implementation to quantize f32 to i8/u8 with C++ native
  // arithmetic.
  virtual APInt quantizeF32ToInt8(APFloat expressedValue) const {
    assert(&expressedValue.getSemantics() == &APFloat::IEEEsingle());
    assert(storageBitWidth == 8);
    assert(roundMode == llvm::APFloatBase::rmNearestTiesToAway);

    const float realValue = expressedValue.convertToFloat();

    const double scaled = realValue / scaleDouble + zeroPointDouble;
    // Round to nearest integer with halfway cases rounded away from zero.
    const double scaledRounded = std::round(scaled);
    const double clamped =
        std::min(std::max(scaledRounded, clampMinDouble), clampMaxDouble);

    uint64_t signlessResult;
    if (isSigned) {
      int64_t clampedInt = static_cast<int8_t>(clamped);
      memcpy(&signlessResult, &clampedInt, sizeof(clampedInt));
    } else {
      signlessResult = static_cast<uint8_t>(clamped);
    }
    return APInt(storageBitWidth, signlessResult);
  }

  // Keep both APFloat and double versions of the quantization parameters
  // around since they will be used in generic and specialized arithmetic,
  // respectively.
  const APFloat scale;
  const APFloat zeroPoint;
  const APFloat clampMin;
  const APFloat clampMax;

  const double scaleDouble;
  const double zeroPointDouble;
  const double clampMinDouble;
  const double clampMaxDouble;

  const uint32_t storageBitWidth;
  const bool isSigned;
  const llvm::APFloat::roundingMode roundMode;
};

/// An utility class to quantize an attribute by the per-axis quantization
/// parameters. The size of the quantization dim in the converted elements
/// attribute should matche the size of of scales/zeroPoints vectors in the
/// quantization parameters.
class UniformQuantizedPerAxisValueConverter {
public:
  explicit UniformQuantizedPerAxisValueConverter(
      UniformQuantizedPerAxisType uniformType)
      : scales(uniformType.getScales()),
        zeroPoints(uniformType.getZeroPoints()),
        clampMin(static_cast<double>(uniformType.getStorageTypeMin())),
        clampMax(static_cast<double>(uniformType.getStorageTypeMax())),
        storageBitWidth(uniformType.getStorageTypeIntegralWidth()),
        isSigned(uniformType.isSigned()),
        quantizationDim(uniformType.getQuantizedDimension()) {
    assert(uniformType.getExpressedType().isa<FloatType>());
    assert(uniformType.getStorageType().isSignlessInteger());
    assert(scales.size() == zeroPoints.size());
  }

  /// Quantize an Attribute by the quantization parameters. Return nullptr if
  /// the conversion fails or the input array isn't an ElementsAttr.
  ElementsAttr convert(Attribute realValue);

private:
  /// Quantize an DenseFPElementsAttr by the quantization parameters.
  DenseElementsAttr convert(DenseFPElementsAttr attr);

  /// Get a uniform converter for the index-th chunk along the quantizationDim.
  /// All the elements in this chunk is quantized by the returned converter.
  UniformQuantizedValueConverter getPerChunkConverter(int index) const {
    UniformQuantizedValueConverter converter(scales[index], zeroPoints[index],
                                             clampMin, clampMax,
                                             storageBitWidth, isSigned);
    return converter;
  }

  const ArrayRef<double> scales;
  const ArrayRef<int64_t> zeroPoints;
  const APFloat clampMin;
  const APFloat clampMax;
  const uint32_t storageBitWidth;
  const bool isSigned;
  int32_t quantizationDim;
};

} // namespace quant
} // namespace mlir

#endif // MLIR_DIALECT_QUANT_UNIFORMSUPPORT_H_