diff options
author | Viet-Hoa Do <viet-hoa.do@arm.com> | 2023-11-13 17:20:45 +0000 |
---|---|---|
committer | Viet-Hoa Do <viet-hoa.do@arm.com> | 2023-12-07 09:42:47 +0000 |
commit | 47370943471c98b5ed4c954b350b925d368a810e (patch) | |
tree | f5fd7d4c4a51b538874632607887b93d8ece3c07 /src/cpu/kernels/depth_to_space/nchw/any | |
parent | 17e116e90e6b962a09c133c646b6ad7884e94693 (diff) | |
download | ComputeLibrary-47370943471c98b5ed4c954b350b925d368a810e.tar.gz |
Optimize CPU depth-to-space
Resolves: COMPMID-6622
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: Ibac276618bdda125dcbb9c851c547f12739b15b4
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10749
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/cpu/kernels/depth_to_space/nchw/any')
-rw-r--r-- | src/cpu/kernels/depth_to_space/nchw/any/impl.cpp | 123 |
1 files changed, 123 insertions, 0 deletions
diff --git a/src/cpu/kernels/depth_to_space/nchw/any/impl.cpp b/src/cpu/kernels/depth_to_space/nchw/any/impl.cpp new file mode 100644 index 0000000000..0277690112 --- /dev/null +++ b/src/cpu/kernels/depth_to_space/nchw/any/impl.cpp @@ -0,0 +1,123 @@ +/* + * Copyright (c) 2023 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/Error.h" + +#include <cstdint> +#include <cstring> + +namespace arm_compute +{ +namespace cpu +{ + +void depth_to_space_nchw_any( // + const uint8_t *src, + uint8_t *dst, + const uintptr_t src_shape[4], + const uintptr_t src_strides[4], + const uintptr_t dst_strides[4], + uintptr_t element_size, + uintptr_t block_size) +{ + ARM_COMPUTE_ERROR_ON(src_strides[0] != element_size); + ARM_COMPUTE_ERROR_ON(dst_strides[0] != element_size); + + const auto dst_channels = src_shape[2] / (block_size * block_size); + const auto src_block_col_stride = dst_channels * src_strides[2]; + const auto src_block_row_stride = block_size * dst_channels * src_strides[2]; + + auto *src_batch_ptr = src; + auto *dst_batch_ptr = dst; + + for (uintptr_t batch = 0; batch < src_shape[3]; ++batch) + { + auto *src_channel_ptr = src_batch_ptr; + auto *dst_channel_ptr = dst_batch_ptr; + + for (uintptr_t channel = 0; channel < dst_channels; ++channel) + { + auto *src_height_block_ptr = src_channel_ptr; + auto *dst_row_ptr = dst_channel_ptr; + + for (uintptr_t height_block = 0; height_block < src_shape[1]; ++height_block) + { + auto *src_block_row_ptr = src_height_block_ptr; + + for (uintptr_t block_row = 0; block_row < block_size; ++block_row) + { + auto *src_width_block_ptr = src_block_row_ptr; + auto *dst_col_ptr = dst_row_ptr; + + for (uintptr_t width_block = 0; width_block < src_shape[0]; ++width_block) + { + auto *src_block_col_ptr = src_width_block_ptr; + + for (uintptr_t block_col = 0; block_col < block_size; ++block_col) + { + // The source pointer is accumulated as: + // + // src_block_col_ptr = + // src + + // batch * dst_strides[3] + + // (channel + (block_row * block_size + block_col) * dst_channels) * src_strides[2] + + // height_block * src_strides[1] + + // width_block * element_size; + // + // The destination pointer is accumuated as: + // + // dst_col_ptr = + // dst + + // batch * dst_strides[3] + + // channel * dst_strides[2] + + // (height_block * block_size + block_row) * dst_strides[1] + + // (width_block * block_size + block_col) * element_size + + std::memcpy(dst_col_ptr, src_block_col_ptr, element_size); + + src_block_col_ptr += src_block_col_stride; + dst_col_ptr += element_size; + } + + src_width_block_ptr += element_size; + } + + src_block_row_ptr += src_block_row_stride; + dst_row_ptr += dst_strides[1]; + } + + src_height_block_ptr += src_strides[1]; + } + + src_channel_ptr += src_strides[2]; + dst_channel_ptr += dst_strides[2]; + } + + src_batch_ptr += src_strides[3]; + dst_batch_ptr += dst_strides[3]; + } +} + +} // namespace cpu +} // namespace arm_compute |