test/nextgen/quantization/quantizer.hpp
| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | // | ||
| 2 | // SPDX-FileCopyrightText: Copyright 2025 Arm Limited and/or its affiliates <open-source-office@arm.com> | ||
| 3 | // | ||
| 4 | // SPDX-License-Identifier: Apache-2.0 | ||
| 5 | // | ||
| 6 | |||
| 7 | #pragma once | ||
| 8 | |||
| 9 | #include <cstddef> | ||
| 10 | |||
| 11 | #include "test/common/buffer.hpp" | ||
| 12 | #include "test/common/data_type.hpp" | ||
| 13 | #include "test/common/span.hpp" | ||
| 14 | #include "test/nextgen/harness/tensor.hpp" | ||
| 15 | |||
| 16 | namespace kai::test { | ||
| 17 | |||
| 18 | /// Quantizes floating-point data to lower-precision data types. | ||
| 19 | class Quantizer { | ||
| 20 | public: | ||
| 21 | 24 | Quantizer() = default; ///< Default constructor. | |
| 22 | 24 | virtual ~Quantizer() = default; ///< Destructor. | |
| 23 | Quantizer(const Quantizer&) = delete; ///< No copy constructor. | ||
| 24 | Quantizer& operator=(const Quantizer&) = delete; ///< No copy assignment. | ||
| 25 | Quantizer(Quantizer&&) = default; ///< Move constructor. | ||
| 26 | Quantizer& operator=(Quantizer&&) = default; ///< Move assignment. | ||
| 27 | |||
| 28 | /// Dynamically quantizes the data. | ||
| 29 | /// | ||
| 30 | /// This method determines the quantization information automatically from the input data. | ||
| 31 | /// | ||
| 32 | /// @param[in] fp_dtype The floating-point data type. | ||
| 33 | /// @param[in] shape The size of multidimensional array. | ||
| 34 | /// @param[in] fp_data The floating-point data buffer. | ||
| 35 | /// @param[out] qdata The quantized data. | ||
| 36 | /// @param[out] qscale The quantization scale. | ||
| 37 | /// @param[out] qzp The quantization zero-point. | ||
| 38 | virtual void dynamic_quantize( | ||
| 39 | DataType fp_dtype, Span<const size_t> shape, Span<const std::byte> fp_data, Tensor& qdata, Tensor& qscale, | ||
| 40 | Tensor& qzp) const = 0; | ||
| 41 | |||
| 42 | /// Dequantizes the data. | ||
| 43 | /// | ||
| 44 | /// @param[in] fp_dtype The dequantized data type. | ||
| 45 | /// @param[in] shape The size of multidimensional array. | ||
| 46 | /// @param[in] qdata The quantized data. | ||
| 47 | /// @param[in] qscale The quantization scale. | ||
| 48 | /// @param[in] qzp The quantization zero-point. | ||
| 49 | /// | ||
| 50 | /// @return The dequantized data. | ||
| 51 | [[nodiscard]] virtual Buffer dequantize( | ||
| 52 | DataType fp_dtype, Span<const size_t> shape, Span<const std::byte> qdata, Span<const std::byte> qscale, | ||
| 53 | Span<const std::byte> qzp) const = 0; | ||
| 54 | }; | ||
| 55 | |||
| 56 | } // namespace kai::test | ||
| 57 |