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_convolve_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_convolve_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, const uint16_t out_shift, q7_t * Im_out,    // output image
31                              const uint16_t dim_im_out, // output image dimension
32                              q15_t * bufferA,   //buffer space for input
33                              q7_t * bufferB //buffer space for output
34     )
35 {
36     int       i, j, k, l, m, n;
37     int       conv_out;
38     int       in_row, in_col;
39 
40     for (i = 0; i < ch_im_out; i++)
41     {
42         for (j = 0; j < dim_im_out; j++)
43         {
44             for (k = 0; k < dim_im_out; k++)
45             {
46 #ifndef ARM_NN_TRUNCATE
47                 conv_out = ((q31_t) (bias[i]) << bias_shift) + (0x1 << (out_shift - 1));
48 #else
49                 conv_out = bias[i] << bias_shift;
50 #endif
51                 for (m = 0; m < dim_kernel; m++)
52                 {
53                     for (n = 0; n < dim_kernel; n++)
54                     {
55                         // if-for implementation
56                         in_row = stride * j + m - padding;
57                         in_col = stride * k + n - padding;
58                         if (in_row >= 0 && in_col >= 0 && in_row < dim_im_in && in_col < dim_im_in)
59                         {
60                             for (l = 0; l < ch_im_in; l++)
61                             {
62                                 conv_out += Im_in[(in_row * dim_im_in + in_col) * ch_im_in + l] *
63                                     wt[i * ch_im_in * dim_kernel * dim_kernel + (m * dim_kernel + n) * ch_im_in + l];
64                             }
65                         }
66                     }
67                 }
68                 Im_out[i + (j * dim_im_out + k) * ch_im_out] = (q7_t) __SSAT((conv_out >> out_shift), 8);
69             }
70         }
71     }
72 }
73