1 /*
2 * Copyright (C) 2010-2018 Arm Limited or its affiliates. All rights reserved.
3 *
4 * SPDX-License-Identifier: Apache-2.0
5 *
6 * Licensed under the Apache License, Version 2.0 (the License); you may
7 * not use this file except in compliance with the License.
8 * You may obtain a copy of the License at
9 *
10 * www.apache.org/licenses/LICENSE-2.0
11 *
12 * Unless required by applicable law or agreed to in writing, software
13 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
14 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15 * See the License for the specific language governing permissions and
16 * limitations under the License.
17 */
18
19 #include "ref_functions.h"
20
arm_depthwise_separable_conv_HWC_q7_ref(const q7_t * Im_in,const uint16_t dim_im_in,const uint16_t ch_im_in,const q7_t * wt,const uint16_t ch_im_out,const uint16_t dim_kernel,const uint16_t padding,const uint16_t stride,const q7_t * bias,const uint16_t bias_shift,const uint16_t out_shift,q7_t * Im_out,const uint16_t dim_im_out,q15_t * bufferA,q7_t * bufferB)21 void arm_depthwise_separable_conv_HWC_q7_ref(const q7_t * Im_in, // input image
22 const uint16_t dim_im_in, // input image dimention
23 const uint16_t ch_im_in, // number of input image channels
24 const q7_t * wt, // kernel weights
25 const uint16_t ch_im_out, // number of filters, i.e., output image channels
26 const uint16_t dim_kernel, // filter kernel size
27 const uint16_t padding, // padding sizes
28 const uint16_t stride, // stride
29 const q7_t * bias, // bias
30 const uint16_t bias_shift, // amount of left-shift for bias
31 const uint16_t out_shift, // amount of right-shift for output
32 q7_t * Im_out, // output image
33 const uint16_t dim_im_out, // output image dimension
34 q15_t * bufferA, //buffer space for input
35 q7_t * bufferB //buffer space for output
36 )
37 {
38 int i_out_y, i_out_x, i_ch_out;
39 int i_ker_y, i_ker_x;
40 for (i_out_y = 0; i_out_y < dim_im_out; i_out_y++)
41 {
42 for (i_out_x = 0; i_out_x < dim_im_out; i_out_x++)
43 {
44 for (i_ch_out = 0; i_ch_out < ch_im_out; i_ch_out++)
45 {
46 // for each output
47 #ifndef ARM_NN_TRUNCATE
48 int conv_out = (bias[i_ch_out] << bias_shift) + (0x1 << (out_shift - 1));
49 #else
50 int conv_out = bias[i_ch_out] << bias_shift;
51 #endif
52 for (i_ker_y = 0; i_ker_y < dim_kernel; i_ker_y++)
53 {
54 for (i_ker_x = 0; i_ker_x < dim_kernel; i_ker_x++)
55 {
56 int in_row = stride * i_out_y + i_ker_y - padding;
57 int in_col = stride * i_out_x + i_ker_x - padding;
58 if (in_row >= 0 && in_col >= 0 && in_row < dim_im_in && in_col < dim_im_in)
59 {
60 conv_out += Im_in[(in_row * dim_im_in + in_col) * ch_im_in + i_ch_out] *
61 wt[(i_ker_y * dim_kernel + i_ker_x) * ch_im_out + i_ch_out];
62 }
63 }
64 }
65 Im_out[(i_out_y * dim_im_out + i_out_x) * ch_im_out + i_ch_out] =
66 (q7_t) __SSAT((conv_out >> out_shift), 8);
67 }
68 }
69 }
70 }
71