diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-07-02 09:22:14 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-07-08 19:13:10 +0000 |
commit | da816752cad76c8e1b367e8e9c648994a1af599a (patch) | |
tree | d6efc7e559e717c07c719037947b2533a835b583 /src/core/cpu/kernels/sub/neon | |
parent | 4411e1f1d49efb78fb07a3c183f386307f951bad (diff) | |
download | ComputeLibrary-da816752cad76c8e1b367e8e9c648994a1af599a.tar.gz |
Remove redundant implementations of Add/Sub operators
Allows only implementations where inputs/output are of the same data
type and removes legacy Computer Vision ones.
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: Ia2b3d23a04236aab682f0c36a1110a30f7c06d1c
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5900
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/cpu/kernels/sub/neon')
-rw-r--r-- | src/core/cpu/kernels/sub/neon/integer.cpp | 183 | ||||
-rw-r--r-- | src/core/cpu/kernels/sub/neon/list.h | 3 |
2 files changed, 0 insertions, 186 deletions
diff --git a/src/core/cpu/kernels/sub/neon/integer.cpp b/src/core/cpu/kernels/sub/neon/integer.cpp deleted file mode 100644 index bba73df1e8..0000000000 --- a/src/core/cpu/kernels/sub/neon/integer.cpp +++ /dev/null @@ -1,183 +0,0 @@ -/* - * Copyright (c) 2021 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/Helpers.h" -#include "arm_compute/core/ITensor.h" -#include "arm_compute/core/Types.h" -#include "arm_compute/core/utils/misc/Traits.h" -#include "src/core/NEON/wrapper/wrapper.h" -#include "src/core/helpers/WindowHelpers.h" - -namespace arm_compute -{ -namespace cpu -{ -namespace -{ -void sub_s16_u8_s16_impl(const ITensor *src0, const ITensor *src1, ITensor *dst, const ConvertPolicy &policy, const Window &window, bool is_swapped) -{ - // Create input windows - Window win = window; - Window input1_win = window.broadcast_if_dimension_le_one(src0->info()->tensor_shape()); - Window input2_win = window.broadcast_if_dimension_le_one(src1->info()->tensor_shape()); - - // Clear X Dimension on execution window as we handle manually - win.set(Window::DimX, Window::Dimension(0, 1, 1)); - input1_win.set(Window::DimX, Window::Dimension(0, 1, 1)); - input2_win.set(Window::DimX, Window::Dimension(0, 1, 1)); - - Iterator input1(src0, input1_win); - Iterator input2(src1, input2_win); - Iterator output(dst, win); - - const int window_step_x = 8; - const auto window_start_x = static_cast<int>(window.x().start()); - const auto window_end_x = static_cast<int>(window.x().end()); - - execute_window_loop(win, [&](const Coordinates &) - { - const auto input1_ptr = reinterpret_cast<const int16_t *>(input1.ptr()); - const auto input2_ptr = reinterpret_cast<const uint8_t *>(input2.ptr()); - const auto output_ptr = reinterpret_cast<int16_t *>(output.ptr()); - - if(policy == ConvertPolicy::WRAP) - { - // Compute S elements per iteration - int x = window_start_x; - for(; x <= (window_end_x - window_step_x); x += window_step_x) - { - const auto vin1 = wrapper::vloadq(input1_ptr + x); - const auto vin2 = vreinterpretq_s16_u16(wrapper::vmovl(wrapper::vload(input2_ptr + x))); - const auto res = is_swapped ? wrapper::vsub(vin2, vin1) : wrapper::vsub(vin1, vin2); - wrapper::vstore(output_ptr + x, res); - } - - // Compute left-over elements - for(; x < window_end_x; ++x) - { - const auto res = is_swapped ? static_cast<int16_t>(*(input2_ptr + x)) - *(input1_ptr + x) : *(input1_ptr + x) - static_cast<int16_t>(*(input2_ptr + x)); - *(output_ptr + x) = res; - } - } - else - { - // Compute S elements per iteration - int x = window_start_x; - for(; x <= (window_end_x - window_step_x); x += window_step_x) - { - const auto vin1 = wrapper::vloadq(input1_ptr + x); - const auto vin2 = vreinterpretq_s16_u16(wrapper::vmovl(wrapper::vload(input2_ptr + x))); - const auto res = is_swapped ? wrapper::vqsub(vin2, vin1) : wrapper::vqsub(vin1, vin2); - wrapper::vstore(output_ptr + x, res); - } - - // Compute left-over elements - for(; x < window_end_x; ++x) - { - const auto res = is_swapped ? wrapper::sub_sat(static_cast<int16_t>(*(input2_ptr + x)), *(input1_ptr + x)) : wrapper::sub_sat(*(input1_ptr + x), static_cast<int16_t>(*(input2_ptr + x))); - *(output_ptr + x) = res; - } - } - }, - input1, input2, output); -} -} - -void sub_s16_u8_s16_neon(const ITensor *src0, const ITensor *src1, ITensor *dst, const ConvertPolicy &policy, const Window &window) -{ - sub_s16_u8_s16_impl(src1, src0, dst, policy, window, false); -} - -void sub_u8_s16_s16_neon(const ITensor *src0, const ITensor *src1, ITensor *dst, const ConvertPolicy &policy, const Window &window) -{ - // Swap arguments - sub_s16_u8_s16_impl(src1, src0, dst, policy, window, true); -} - -void sub_u8_u8_s16_neon(const ITensor *src0, const ITensor *src1, ITensor *dst, const ConvertPolicy &policy, const Window &window) -{ - // Create input windows - Window win = window; - Window input1_win = window.broadcast_if_dimension_le_one(src0->info()->tensor_shape()); - Window input2_win = window.broadcast_if_dimension_le_one(src1->info()->tensor_shape()); - - // Clear X Dimension on execution window as we handle manually - win.set(Window::DimX, Window::Dimension(0, 1, 1)); - input1_win.set(Window::DimX, Window::Dimension(0, 1, 1)); - input2_win.set(Window::DimX, Window::Dimension(0, 1, 1)); - - Iterator input1(src0, input1_win); - Iterator input2(src1, input2_win); - Iterator output(dst, win); - - const int window_step_x = 8; - const auto window_start_x = static_cast<int>(window.x().start()); - const auto window_end_x = static_cast<int>(window.x().end()); - - execute_window_loop(win, [&](const Coordinates &) - { - const auto input1_ptr = reinterpret_cast<const uint8_t *>(input1.ptr()); - const auto input2_ptr = reinterpret_cast<const uint8_t *>(input2.ptr()); - const auto output_ptr = reinterpret_cast<int16_t *>(output.ptr()); - - if(policy == ConvertPolicy::WRAP) - { - // Compute S elements per iteration - int x = window_start_x; - for(; x <= (window_end_x - window_step_x); x += window_step_x) - { - const auto vin1 = vreinterpretq_s16_u16(wrapper::vmovl(wrapper::vload(input1_ptr + x))); - const auto vin2 = vreinterpretq_s16_u16(wrapper::vmovl(wrapper::vload(input2_ptr + x))); - wrapper::vstore(output_ptr + x, wrapper::vsub(vin1, vin2)); - } - - // Compute left-over elements - for(; x < window_end_x; ++x) - { - *(output_ptr + x) = static_cast<int16_t>(*(input1_ptr + x)) - static_cast<int16_t>(*(input2_ptr + x)); - } - } - else - { - // Compute S elements per iteration - int x = window_start_x; - for(; x <= (window_end_x - window_step_x); x += window_step_x) - { - const auto vin1 = vreinterpretq_s16_u16(wrapper::vmovl(wrapper::vload(input1_ptr + x))); - const auto vin2 = vreinterpretq_s16_u16(wrapper::vmovl(wrapper::vload(input2_ptr + x))); - wrapper::vstore(output_ptr + x, wrapper::vqsub(vin1, vin2)); - } - - // Compute left-over elements - for(; x < window_end_x; ++x) - { - *(output_ptr + x) = wrapper::sub_sat(static_cast<int16_t>(*(input1_ptr + x)), - static_cast<int16_t>(*(input2_ptr + x))); - } - } - }, - input1, input2, output); -} - -} // namespace cpu -} // namespace arm_compute
\ No newline at end of file diff --git a/src/core/cpu/kernels/sub/neon/list.h b/src/core/cpu/kernels/sub/neon/list.h index 1ab4e6367b..ac1346001a 100644 --- a/src/core/cpu/kernels/sub/neon/list.h +++ b/src/core/cpu/kernels/sub/neon/list.h @@ -38,9 +38,6 @@ namespace cpu DECLARE_SUB_KERNEL(sub_qasymm8_neon); DECLARE_SUB_KERNEL(sub_qasymm8_signed_neon); DECLARE_SUB_KERNEL(sub_qsymm16_neon); -DECLARE_SUB_KERNEL(sub_s16_u8_s16_neon); -DECLARE_SUB_KERNEL(sub_u8_s16_s16_neon); -DECLARE_SUB_KERNEL(sub_u8_u8_s16_neon); #undef DECLARE_SUB_KERNEL |