diff options
Diffstat (limited to 'src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp')
| -rw-r--r-- | src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp | 14 |
1 files changed, 7 insertions, 7 deletions
diff --git a/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp b/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp index 162c4b1ace..d557cfa1bd 100644 --- a/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp +++ b/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -192,12 +192,12 @@ public: execute_window_loop(window_out, [&](const Coordinates & id) { - const uint8_t *input_ptr = in.ptr() - conv_pad_left * input_stride_x - conv_pad_top * input_stride_y; - uint8_t *out_ptr = out.ptr(); - int ih = 0; - int oh = 0; - float32x4_t accum0[small_tensor_size_optim] = { vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0) }; - float32x4_t accum1[small_tensor_size_optim] = { vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0) }; + const uint8_t *input_ptr = in.ptr() - conv_pad_left * input_stride_x - conv_pad_top * input_stride_y; + uint8_t *out_ptr = out.ptr(); + int ih = 0; + int oh = 0; + std::array<float32x4_t, 8> accum0 = { vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0) }; + std::array<float32x4_t, 8> accum1 = { vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0), vdupq_n_f32(0) }; for(int oz = 0; oz < range_z; ++oz) { accum0[0] = accum0[1] = accum0[2] = accum0[3] = accum0[4] = accum0[5] = accum0[6] = accum0[7] = vdupq_n_f32(0.f); |