diff options
Diffstat (limited to 'compute/ARMComputeEx/src/core/NEON/kernels/NEQuantizationSymmetricKernel.cpp')
| -rw-r--r-- | compute/ARMComputeEx/src/core/NEON/kernels/NEQuantizationSymmetricKernel.cpp | 224 |
1 files changed, 224 insertions, 0 deletions
diff --git a/compute/ARMComputeEx/src/core/NEON/kernels/NEQuantizationSymmetricKernel.cpp b/compute/ARMComputeEx/src/core/NEON/kernels/NEQuantizationSymmetricKernel.cpp new file mode 100644 index 000000000..acf0092eb --- /dev/null +++ b/compute/ARMComputeEx/src/core/NEON/kernels/NEQuantizationSymmetricKernel.cpp @@ -0,0 +1,224 @@ +/* + * Copyright (c) 2019 Samsung Electronics Co., Ltd. All Rights Reserved + * Copyright (c) 2017-2019 ARM Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include "arm_compute/core/NEON/kernels/NEQuantizationSymmetricKernel.h" + +#include "arm_compute/core/Error.h" +#include "arm_compute/core/Helpers.h" +#include "arm_compute/core/NEON/NEAsymm.h" +#include "arm_compute/core/NEON/wrapper/wrapper.h" +#include "arm_compute/core/Utils.h" +#include "arm_compute/core/Validate.h" +#include "arm_compute/core/Window.h" + +#include "arm_compute/core/CPP/Validate.h" + +#include <arm_neon.h> + +using namespace arm_compute; + +namespace +{ +Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, + const ITensorInfo *scale_factor) +{ + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output); + ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input); + ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 2); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32); + ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape().total_size() == 0); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S8); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(scale_factor, 1, DataType::F16, + DataType::F32); + ARM_COMPUTE_RETURN_ERROR_ON(scale_factor->tensor_shape().total_size() == 0); + ARM_COMPUTE_RETURN_ERROR_ON(scale_factor->num_dimensions() > 1); + ARM_COMPUTE_RETURN_ERROR_ON(scale_factor->dimension(0) != input->dimension(1)); + + return Status{}; +} + +inline float32x4x4_t load_value(const float *input_ptr) +{ + return {wrapper::vloadq(input_ptr), wrapper::vloadq(input_ptr + 4), + wrapper::vloadq(input_ptr + 8), wrapper::vloadq(input_ptr + 12)}; +} +#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC +inline const float32x4x4_t load_value(const float16_t *input_ptr) +{ + return {vcvt_f32_f16(wrapper::vload(input_ptr)), vcvt_f32_f16(wrapper::vload(input_ptr + 4)), + vcvt_f32_f16(wrapper::vload(input_ptr + 8)), + vcvt_f32_f16(wrapper::vload(input_ptr + 12))}; +} + +#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC + +inline float32x4_t round(const float32x4_t &fv) +{ + const float32x4_t point5_f32x4 = vdupq_n_f32(0.5f); + const float32x4_t zero_f32x4 = vdupq_n_f32(0.0f); + // If value < 0, mask = -1, else mask = 0 + int32x4_t mask_less_zero_ui32x4 = reinterpret_cast<int32x4_t>(vcltq_f32(fv, zero_f32x4)); + return vaddq_f32(fv, vaddq_f32(vcvtq_f32_s32(mask_less_zero_ui32x4), point5_f32x4)); +} + +inline int8x16_t vquantizeSymm(const float32x4x4_t &fv, float scale_factor_inv, int32_t max_scale) +{ + const float32x4_t vinvscale = vdupq_n_f32(scale_factor_inv); + const int32x4_t vposend = vdupq_n_s32(max_scale); + const int32x4_t vnagend = vdupq_n_s32(-max_scale); + + const int32x4x4_t rf = {{ +#ifdef __aarch64__ + vminq_s32(vposend, + vmaxq_s32(vnagend, vcvtnq_s32_f32(round(vmulq_f32(fv.val[0], vinvscale))))), + vminq_s32(vposend, + vmaxq_s32(vnagend, vcvtnq_s32_f32(round(vmulq_f32(fv.val[1], vinvscale))))), + vminq_s32(vposend, + vmaxq_s32(vnagend, vcvtnq_s32_f32(round(vmulq_f32(fv.val[2], vinvscale))))), + vminq_s32(vposend, + vmaxq_s32(vnagend, vcvtnq_s32_f32(round(vmulq_f32(fv.val[3], vinvscale))))), +#else //__aarch64__ + vminq_s32(vposend, vmaxq_s32(vnagend, vcvtq_s32_f32(round(vmulq_f32(fv.val[0], vinvscale))))), + vminq_s32(vposend, vmaxq_s32(vnagend, vcvtq_s32_f32(round(vmulq_f32(fv.val[1], vinvscale))))), + vminq_s32(vposend, vmaxq_s32(vnagend, vcvtq_s32_f32(round(vmulq_f32(fv.val[2], vinvscale))))), + vminq_s32(vposend, vmaxq_s32(vnagend, vcvtq_s32_f32(round(vmulq_f32(fv.val[3], vinvscale))))), +#endif //__aarch64__ + }}; + const int8x8_t pa = vqmovn_s16(vcombine_s16(vqmovn_s32(rf.val[0]), vqmovn_s32(rf.val[1]))); + const int8x8_t pb = vqmovn_s16(vcombine_s16(vqmovn_s32(rf.val[2]), vqmovn_s32(rf.val[3]))); + return vcombine_s8(pa, pb); +} +} // namespace + +NEQuantizationSymmetricKernel::NEQuantizationSymmetricKernel() + : _input(nullptr), _output(nullptr), _scale_factor(nullptr) +{ +} + +void NEQuantizationSymmetricKernel::configure(const ITensor *input, ITensor *output, + ITensor *scale_factor) +{ + ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); + ARM_COMPUTE_ERROR_THROW_ON( + validate_arguments(input->info(), output->info(), scale_factor->info())); + + _input = input; + _output = output; + _scale_factor = scale_factor; + + // Configure kernel window + Window win_config = calculate_max_window(*input->info(), Steps()); + + Coordinates coord; + coord.set_num_dimensions(output->info()->num_dimensions()); + output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape())); + + INEKernel::configure(win_config); +} + +Status NEQuantizationSymmetricKernel::validate(const ITensorInfo *input, const ITensorInfo *output, + const ITensorInfo *scale_factor) +{ + ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, scale_factor)); + + return Status{}; +} + +template <typename T> void NEQuantizationSymmetricKernel::quantize(const Window &window) +{ + constexpr auto window_step = 16; + const auto window_start_x = static_cast<int>(window.x().start()); + const auto window_end_x = static_cast<int>(window.x().end()); + +#ifdef __aarch64__ + constexpr RoundingPolicy rounding_policy = RoundingPolicy::TO_NEAREST_EVEN; +#else //__aarch64__ + constexpr RoundingPolicy rounding_policy = RoundingPolicy::TO_NEAREST_UP; +#endif //__aarch64__ + + // Collapse window and reset first dimension to handle tail calculations manually + // Support Only 2D input + Window win_collapsed = window; + Iterator input(_input, win_collapsed); + Iterator output(_output, win_collapsed); + const auto dim_x = _input->info()->dimension(0); + win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1)); + execute_window_loop( + win_collapsed, + [&](const Coordinates &id) { + const auto start = reinterpret_cast<const T *>(input.ptr()); + const auto min_max = std::minmax_element(start, start + dim_x); + const auto int8_scale = 127; + auto range = std::max(std::abs(*min_max.first), std::abs(*min_max.second)); + if (range == 0) + { + *reinterpret_cast<T *>(_scale_factor->ptr_to_element({id.y()})) = 1; + range = 1; + } + else + { + *reinterpret_cast<T *>(_scale_factor->ptr_to_element({id.y()})) = range / int8_scale; + } + const auto scale_factor_inv = int8_scale / range; + + auto input_ptr = reinterpret_cast<const T *>(input.ptr()); + auto output_ptr = reinterpret_cast<int8_t *>(output.ptr()); + int x = window_start_x; + for (; x <= (window_end_x - window_step); x += window_step) + { + wrapper::vstore(&output_ptr[x], + vquantizeSymm(load_value(&input_ptr[x]), scale_factor_inv, int8_scale)); + } + // Compute left-over elements + for (; x < window_end_x; ++x) + { + int quantized = arm_compute::round(input_ptr[x] * scale_factor_inv, rounding_policy); + quantized = std::min(int8_scale, std::max(quantized, -int8_scale)); + output_ptr[x] = static_cast<int8_t>(quantized); + } + }, + input, output); +} + +void NEQuantizationSymmetricKernel::run(const Window &window, const ThreadInfo &info) +{ + ARM_COMPUTE_UNUSED(info); + ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); + ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); + + switch (_input->info()->data_type()) + { + case DataType::F32: + NEQuantizationSymmetricKernel::quantize<float>(window); + break; +#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC + case DataType::F16: + NEQuantizationSymmetricKernel::quantize<float16_t>(window); + break; +#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC + default: + ARM_COMPUTE_ERROR("Unsupported data type."); + } +} |