71 lines
3.3 KiB
C++
71 lines
3.3 KiB
C++
/* Copyright 2024 The TensorFlow Authors. All Rights Reserved.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
==============================================================================*/
|
|
#ifndef TENSORFLOW_LITE_EXPERIMENTAL_SHLO_QUANTIZE_H_
|
|
#define TENSORFLOW_LITE_EXPERIMENTAL_SHLO_QUANTIZE_H_
|
|
|
|
#include <algorithm>
|
|
|
|
#include "tensorflow/lite/experimental/shlo/data_type.h"
|
|
|
|
namespace shlo_ref {
|
|
|
|
// Converts floating-point values of the expressed type into corresponding
|
|
// integer values of the storage type using the zero point and scale associated
|
|
// with the quantized element type.
|
|
template <typename StorageT, typename ExpressedT>
|
|
inline constexpr StorageT Quantize(ExpressedT expressed_value,
|
|
StorageT zero_point, ExpressedT scale_inv,
|
|
StorageT min_value, StorageT max_value) {
|
|
const ExpressedT rounding_extra =
|
|
(expressed_value > 0) ? ExpressedT(0.5f) : ExpressedT(-0.5f);
|
|
ExpressedT tmp = expressed_value * scale_inv + rounding_extra;
|
|
|
|
// Clamp the value in case of overflow/underflow. This is needed to avoid
|
|
// getting a SIGILL exception when casting down below.
|
|
tmp = std::clamp(tmp, static_cast<ExpressedT>(min_value),
|
|
static_cast<ExpressedT>(max_value));
|
|
auto rounded_value = static_cast<StorageT>(tmp);
|
|
StorageT storage_value(rounded_value + zero_point);
|
|
|
|
// Clamp again using the min & max values.
|
|
return std::clamp(storage_value, min_value, max_value);
|
|
}
|
|
|
|
// A DataType dispatched version of Quantize, this allows for leveraging the min
|
|
// and max values of the DataType, which may not necessarily be the same as the
|
|
// min and max values of the underlying C data type. Ie, a 4-bit integer can be
|
|
// stored in an int8_t, but the value range is from -8 to 7.
|
|
template <DataType storage_type, DataType expressed_type>
|
|
inline constexpr StorageType<storage_type> Quantize(
|
|
StorageType<expressed_type> expressed_value,
|
|
StorageType<storage_type> zero_point, StorageType<expressed_type> scale_inv,
|
|
StorageType<storage_type> min_value = Storage<storage_type>::kMinValue,
|
|
StorageType<storage_type> max_value = Storage<storage_type>::kMaxValue) {
|
|
return Quantize(expressed_value, zero_point, scale_inv, min_value, max_value);
|
|
}
|
|
|
|
// Converts quantized elements which represent integer values of the storage
|
|
// type into corresponding floating-point values of the expressed type using
|
|
// the zero point and scale associated with the quantized element type.
|
|
template <typename StorageT, typename ExpressedT>
|
|
inline constexpr ExpressedT Dequantize(StorageT quantized_value,
|
|
StorageT zero_point, ExpressedT scale) {
|
|
auto sub = quantized_value - zero_point;
|
|
return static_cast<ExpressedT>(sub) * scale;
|
|
}
|
|
} // namespace shlo_ref
|
|
|
|
#endif // TENSORFLOW_LITE_EXPERIMENTAL_SHLO_QUANTIZE_H_
|