diff options
author | Felix Thomasmathibalan <felixjohnny.thomasmathibalan@arm.com> | 2023-09-27 17:46:17 +0100 |
---|---|---|
committer | felixjohnny.thomasmathibalan <felixjohnny.thomasmathibalan@arm.com> | 2023-09-28 12:08:05 +0000 |
commit | afd38f0c617d6f89b2b4532c6c44f116617e2b6f (patch) | |
tree | 03bc7d5a762099989b16a656fa8d397b490ed70e /src/cpu/kernels/CpuConcatenateBatchKernel.cpp | |
parent | bdcb4c148ee2fdeaaddf4cf1e57bbb0de02bb894 (diff) | |
download | ComputeLibrary-afd38f0c617d6f89b2b4532c6c44f116617e2b6f.tar.gz |
Apply clang-format on repository
Code is formatted as per a revised clang format configuration
file(not part of this delivery). Version 14.0.6 is used.
Exclusion List:
- files with .cl extension
- files that are not strictly C/C++ (e.g. Android.bp, Sconscript ...)
And the following directories
- compute_kernel_writer/validation/
- tests/
- include/
- src/core/NEON/kernels/convolution/
- src/core/NEON/kernels/arm_gemm/
- src/core/NEON/kernels/arm_conv/
- data/
There will be a follow up for formatting of .cl files and the
files under tests/ and compute_kernel_writer/validation/.
Signed-off-by: Felix Thomasmathibalan <felixjohnny.thomasmathibalan@arm.com>
Change-Id: Ib7eb1fcf4e7537b9feaefcfc15098a804a3fde0a
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10391
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Diffstat (limited to 'src/cpu/kernels/CpuConcatenateBatchKernel.cpp')
-rw-r--r-- | src/cpu/kernels/CpuConcatenateBatchKernel.cpp | 126 |
1 files changed, 67 insertions, 59 deletions
diff --git a/src/cpu/kernels/CpuConcatenateBatchKernel.cpp b/src/cpu/kernels/CpuConcatenateBatchKernel.cpp index 29d40f0e52..8c290173e8 100644 --- a/src/cpu/kernels/CpuConcatenateBatchKernel.cpp +++ b/src/cpu/kernels/CpuConcatenateBatchKernel.cpp @@ -30,10 +30,11 @@ #include "arm_compute/core/Utils.h" #include "arm_compute/core/Validate.h" #include "arm_compute/core/Window.h" -#include "src/core/NEON/NEAsymm.h" -#include "src/core/NEON/wrapper/wrapper.h" + #include "src/core/helpers/AutoConfiguration.h" #include "src/core/helpers/WindowHelpers.h" +#include "src/core/NEON/NEAsymm.h" +#include "src/core/NEON/wrapper/wrapper.h" namespace arm_compute { @@ -50,13 +51,14 @@ void batch_concat(const ITensor *src, ITensor *dst, unsigned int batch_offset, c uint8_t *src_ptr = src->buffer() + src->info()->offset_first_element_in_bytes(); // Offset dst - uint8_t *dst_ptr = dst->buffer() + dst->info()->offset_first_element_in_bytes() + batch_offset * dst->info()->strides_in_bytes()[3]; + uint8_t *dst_ptr = dst->buffer() + dst->info()->offset_first_element_in_bytes() + + batch_offset * dst->info()->strides_in_bytes()[3]; const auto window_start_x = static_cast<int>(window.x().start()); const auto window_end_x = static_cast<int>(window.x().end()); const int window_step_x = 16 / dst->info()->element_size(); - Window win{ window }; + Window win{window}; win.set(Window::DimX, Window::Dimension(0, 1, 1)); win.set(3, Window::Dimension(0, src->info()->tensor_shape()[3], 1)); @@ -66,66 +68,74 @@ void batch_concat(const ITensor *src, ITensor *dst, unsigned int batch_offset, c const DataType dt = src->info()->data_type(); const UniformQuantizationInfo src_qinfo = src->info()->quantization_info().uniform(); const UniformQuantizationInfo dst_qinfo = dst->info()->quantization_info().uniform(); - if(dt == DataType::QASYMM8 && src_qinfo != dst_qinfo) + if (dt == DataType::QASYMM8 && src_qinfo != dst_qinfo) { - execute_window_loop(win, [&](const Coordinates &) - { - const auto in_ptr = reinterpret_cast<const uint8_t *>(src_ptr + src_it.offset()); - const auto out_ptr = reinterpret_cast<uint8_t *>(dst_ptr + dst_it.offset()); - - int x = window_start_x; - for(; x <= (window_end_x - window_step_x); x += window_step_x) + execute_window_loop( + win, + [&](const Coordinates &) { - wrapper::vstore(out_ptr, vquantize(vdequantize(wrapper::vloadq(in_ptr), src_qinfo), dst_qinfo)); - } - - // Compute left-over elements - for(; x < window_end_x; ++x) - { - *(out_ptr + x) = quantize_qasymm8(dequantize_qasymm8(*(in_ptr + x), src_qinfo), dst_qinfo); - } - }, - src_it, dst_it); + const auto in_ptr = reinterpret_cast<const uint8_t *>(src_ptr + src_it.offset()); + const auto out_ptr = reinterpret_cast<uint8_t *>(dst_ptr + dst_it.offset()); + + int x = window_start_x; + for (; x <= (window_end_x - window_step_x); x += window_step_x) + { + wrapper::vstore(out_ptr, vquantize(vdequantize(wrapper::vloadq(in_ptr), src_qinfo), dst_qinfo)); + } + + // Compute left-over elements + for (; x < window_end_x; ++x) + { + *(out_ptr + x) = quantize_qasymm8(dequantize_qasymm8(*(in_ptr + x), src_qinfo), dst_qinfo); + } + }, + src_it, dst_it); } - else if(dt == DataType::QASYMM8_SIGNED && src_qinfo != dst_qinfo) + else if (dt == DataType::QASYMM8_SIGNED && src_qinfo != dst_qinfo) { - execute_window_loop(win, [&](const Coordinates &) - { - const auto in_ptr = reinterpret_cast<const int8_t *>(src_ptr + src_it.offset()); - const auto out_ptr = reinterpret_cast<int8_t *>(dst_ptr + dst_it.offset()); - int x = window_start_x; - for(; x <= (window_end_x - window_step_x); x += window_step_x) - { - wrapper::vstore(out_ptr, vquantize_signed(vdequantize(wrapper::vloadq(in_ptr), src_qinfo), dst_qinfo)); - } - // Compute left-over elements - for(; x < window_end_x; ++x) + execute_window_loop( + win, + [&](const Coordinates &) { - *(out_ptr + x) = quantize_qasymm8_signed(dequantize_qasymm8_signed(*(in_ptr + x), src_qinfo), dst_qinfo); - } - }, - src_it, dst_it); + const auto in_ptr = reinterpret_cast<const int8_t *>(src_ptr + src_it.offset()); + const auto out_ptr = reinterpret_cast<int8_t *>(dst_ptr + dst_it.offset()); + int x = window_start_x; + for (; x <= (window_end_x - window_step_x); x += window_step_x) + { + wrapper::vstore(out_ptr, + vquantize_signed(vdequantize(wrapper::vloadq(in_ptr), src_qinfo), dst_qinfo)); + } + // Compute left-over elements + for (; x < window_end_x; ++x) + { + *(out_ptr + x) = + quantize_qasymm8_signed(dequantize_qasymm8_signed(*(in_ptr + x), src_qinfo), dst_qinfo); + } + }, + src_it, dst_it); } else { - execute_window_loop(win, [&](const Coordinates &) - { - const auto in_ptr = reinterpret_cast<const T *>(src_ptr + src_it.offset()); - const auto out_ptr = reinterpret_cast<T *>(dst_ptr + dst_it.offset()); - - int x = window_start_x; - for(; x <= (window_end_x - window_step_x); x += window_step_x) - { - wrapper::vstore(out_ptr + x, wrapper::vloadq(in_ptr + x)); - } - - // Compute left-over elements - for(; x < window_end_x; ++x) + execute_window_loop( + win, + [&](const Coordinates &) { - *(out_ptr + x) = *(in_ptr + x); - } - }, - src_it, dst_it); + const auto in_ptr = reinterpret_cast<const T *>(src_ptr + src_it.offset()); + const auto out_ptr = reinterpret_cast<T *>(dst_ptr + dst_it.offset()); + + int x = window_start_x; + for (; x <= (window_end_x - window_step_x); x += window_step_x) + { + wrapper::vstore(out_ptr + x, wrapper::vloadq(in_ptr + x)); + } + + // Compute left-over elements + for (; x < window_end_x; ++x) + { + *(out_ptr + x) = *(in_ptr + x); + } + }, + src_it, dst_it); } } @@ -154,7 +164,7 @@ void CpuConcatenateBatchKernel::configure(const ITensorInfo *src, unsigned int b _func = nullptr; _batch_offset = batch_offset; - switch(src->data_type()) + switch (src->data_type()) { case DataType::S8: case DataType::U8: @@ -196,9 +206,7 @@ void CpuConcatenateBatchKernel::run_op(ITensorPack &tensors, const Window &windo ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window); ARM_COMPUTE_ERROR_ON(_func == nullptr); - (*_func)(tensors.get_const_tensor(TensorType::ACL_SRC), - tensors.get_tensor(TensorType::ACL_DST), - _batch_offset, + (*_func)(tensors.get_const_tensor(TensorType::ACL_SRC), tensors.get_tensor(TensorType::ACL_DST), _batch_offset, window); } |