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/gemm_matrix_add | |
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/gemm_matrix_add')
-rw-r--r-- | src/cpu/kernels/gemm_matrix_add/generic/neon/fp16.cpp | 44 | ||||
-rw-r--r-- | src/cpu/kernels/gemm_matrix_add/generic/neon/impl.cpp | 49 |
2 files changed, 49 insertions, 44 deletions
diff --git a/src/cpu/kernels/gemm_matrix_add/generic/neon/fp16.cpp b/src/cpu/kernels/gemm_matrix_add/generic/neon/fp16.cpp index 505a37174e..4d7507a5da 100644 --- a/src/cpu/kernels/gemm_matrix_add/generic/neon/fp16.cpp +++ b/src/cpu/kernels/gemm_matrix_add/generic/neon/fp16.cpp @@ -48,30 +48,32 @@ void matrix_addition_f16(const ITensor *src, ITensor *dst, const Window &window, Iterator in(src, win); Iterator out(dst, win); - execute_window_loop(win, [&](const Coordinates &) - { - const auto in_ptr = reinterpret_cast<const float16_t *>(in.ptr()); - const auto out_ptr = reinterpret_cast<float16_t *>(out.ptr()); - - int x = window_start_x; - for(; x < (window_end_x - window_step_x); x += window_step_x) + execute_window_loop( + win, + [&](const Coordinates &) { - float16x8x2_t alpha_ab = vld2q_f16(out_ptr + x); - const float16x8x2_t c = vld2q_f16(in_ptr + x); - // Multiply matrix C by its weight and accumulate - alpha_ab.val[0] = vaddq_f16(alpha_ab.val[0], vmulq_f16(c.val[0], beta_f16)); - alpha_ab.val[1] = vaddq_f16(alpha_ab.val[1], vmulq_f16(c.val[1], beta_f16)); + const auto in_ptr = reinterpret_cast<const float16_t *>(in.ptr()); + const auto out_ptr = reinterpret_cast<float16_t *>(out.ptr()); - vst2q_f16(out_ptr + x, alpha_ab); - } + int x = window_start_x; + for (; x < (window_end_x - window_step_x); x += window_step_x) + { + float16x8x2_t alpha_ab = vld2q_f16(out_ptr + x); + const float16x8x2_t c = vld2q_f16(in_ptr + x); + // Multiply matrix C by its weight and accumulate + alpha_ab.val[0] = vaddq_f16(alpha_ab.val[0], vmulq_f16(c.val[0], beta_f16)); + alpha_ab.val[1] = vaddq_f16(alpha_ab.val[1], vmulq_f16(c.val[1], beta_f16)); - // Left-over loop - for(; x < window_end_x; ++x) - { - *(out_ptr + x) += *(in_ptr + x) * static_cast<float16_t>(beta); - } - }, - in, out); + vst2q_f16(out_ptr + x, alpha_ab); + } + + // Left-over loop + for (; x < window_end_x; ++x) + { + *(out_ptr + x) += *(in_ptr + x) * static_cast<float16_t>(beta); + } + }, + in, out); } } // namespace void neon_fp16_gemm_matrix_add(const ITensor *src, ITensor *dst, const Window &window, float beta) diff --git a/src/cpu/kernels/gemm_matrix_add/generic/neon/impl.cpp b/src/cpu/kernels/gemm_matrix_add/generic/neon/impl.cpp index dd0384ca13..47de0f3928 100644 --- a/src/cpu/kernels/gemm_matrix_add/generic/neon/impl.cpp +++ b/src/cpu/kernels/gemm_matrix_add/generic/neon/impl.cpp @@ -23,6 +23,7 @@ */ #include "src/cpu/kernels/gemm_matrix_add/generic/neon/impl.h" + #include <arm_neon.h> namespace arm_compute @@ -44,33 +45,35 @@ void matrix_addition_f32(const ITensor *src, ITensor *dst, const Window &window, Iterator in(src, win); Iterator out(dst, win); - execute_window_loop(win, [&](const Coordinates &) - { - const auto in_ptr = reinterpret_cast<const float *>(in.ptr()); - const auto out_ptr = reinterpret_cast<float *>(out.ptr()); - - int x = window_start_x; - for(; x < (window_end_x - window_step_x); x += window_step_x) + execute_window_loop( + win, + [&](const Coordinates &) { - float32x4x4_t alpha_ab = vld4q_f32(out_ptr + x); - const float32x4x4_t c = vld4q_f32(in_ptr + x); + const auto in_ptr = reinterpret_cast<const float *>(in.ptr()); + const auto out_ptr = reinterpret_cast<float *>(out.ptr()); - // Multiply matrix C by its weight and accumulate - alpha_ab.val[0] = vmlaq_f32(alpha_ab.val[0], c.val[0], beta_f32); - alpha_ab.val[1] = vmlaq_f32(alpha_ab.val[1], c.val[1], beta_f32); - alpha_ab.val[2] = vmlaq_f32(alpha_ab.val[2], c.val[2], beta_f32); - alpha_ab.val[3] = vmlaq_f32(alpha_ab.val[3], c.val[3], beta_f32); + int x = window_start_x; + for (; x < (window_end_x - window_step_x); x += window_step_x) + { + float32x4x4_t alpha_ab = vld4q_f32(out_ptr + x); + const float32x4x4_t c = vld4q_f32(in_ptr + x); - vst4q_f32(out_ptr + x, alpha_ab); - } + // Multiply matrix C by its weight and accumulate + alpha_ab.val[0] = vmlaq_f32(alpha_ab.val[0], c.val[0], beta_f32); + alpha_ab.val[1] = vmlaq_f32(alpha_ab.val[1], c.val[1], beta_f32); + alpha_ab.val[2] = vmlaq_f32(alpha_ab.val[2], c.val[2], beta_f32); + alpha_ab.val[3] = vmlaq_f32(alpha_ab.val[3], c.val[3], beta_f32); - // Left-over loop - for(; x < window_end_x; ++x) - { - *(out_ptr + x) += *(in_ptr + x) * beta; - } - }, - in, out); + vst4q_f32(out_ptr + x, alpha_ab); + } + + // Left-over loop + for (; x < window_end_x; ++x) + { + *(out_ptr + x) += *(in_ptr + x) * beta; + } + }, + in, out); } } // namespace cpu } // namespace arm_compute |