Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
2 | mjames | 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 | #ifndef _REF_FUNCTIONS_H_ |
||
20 | #define _REF_FUNCTIONS_H_ |
||
21 | |||
22 | #include "arm_math.h" |
||
23 | #include "arm_nnfunctions.h" |
||
24 | //#include "arm_nnsupportfunctions.h" |
||
25 | #include "fully_connected_testing_weights.h" |
||
26 | |||
27 | #ifdef __cplusplus |
||
28 | extern "C" |
||
29 | { |
||
30 | #endif |
||
31 | |||
32 | /* |
||
33 | * |
||
34 | * Convolution reference implemenation |
||
35 | * |
||
36 | */ |
||
37 | |||
38 | void arm_convolve_HWC_q7_ref(const q7_t * Im_in, // input image |
||
39 | const uint16_t dim_im_in, // input image dimention |
||
40 | const uint16_t ch_im_in, // number of input image channels |
||
41 | const q7_t * wt, // kernel weights |
||
42 | const uint16_t ch_im_out, // number of filters, i.e., output image channels |
||
43 | const uint16_t dim_kernel, // filter kernel size |
||
44 | const uint16_t padding, // padding sizes |
||
45 | const uint16_t stride, // stride |
||
46 | const q7_t * bias, // bias |
||
47 | const uint16_t bias_shift, const uint16_t out_shift, q7_t * Im_out, // output image |
||
48 | const uint16_t dim_im_out, // output image dimension |
||
49 | q15_t * bufferA, //buffer space for input |
||
50 | q7_t * bufferB //buffer space for output |
||
51 | ); |
||
52 | |||
53 | void arm_convolve_HWC_q7_ref_nonsquare(const q7_t * Im_in, // input image |
||
54 | const uint16_t dim_im_in_x, // input image dimention x |
||
55 | const uint16_t dim_im_in_y, // input image dimention y |
||
56 | const uint16_t ch_im_in, // number of input image channels |
||
57 | const q7_t * wt, // kernel weights |
||
58 | const uint16_t ch_im_out, // number of filters, i.e., output image channels |
||
59 | const uint16_t dim_kernel_x, // filter kernel size x |
||
60 | const uint16_t dim_kernel_y, // filter kernel size y |
||
61 | const uint16_t padding_x, // padding sizes x |
||
62 | const uint16_t padding_y, // padding sizes y |
||
63 | const uint16_t stride_x, // stride x |
||
64 | const uint16_t stride_y, // stride y |
||
65 | const q7_t * bias, // bias |
||
66 | const uint16_t bias_shift, const uint16_t out_shift, q7_t * Im_out, // output image |
||
67 | const uint16_t dim_im_out_x, // output image dimension x |
||
68 | const uint16_t dim_im_out_y, // output image dimension y |
||
69 | q15_t * bufferA, //buffer space for input |
||
70 | q7_t * bufferB //buffer space for output |
||
71 | ); |
||
72 | |||
73 | void arm_convolve_HWC_q15_ref(const q15_t * Im_in, // input image |
||
74 | const uint16_t dim_im_in, // input image dimention |
||
75 | const uint16_t ch_im_in, // number of input image channels |
||
76 | const q15_t * wt, // kernel weights |
||
77 | const uint16_t ch_im_out, // number of filters, i.e., output image channels |
||
78 | const uint16_t dim_kernel, // filter kernel size |
||
79 | const uint16_t padding, // padding sizes |
||
80 | const uint16_t stride, // stride |
||
81 | const q15_t * bias, // bias |
||
82 | const uint16_t bias_shift, const uint16_t out_shift, q15_t * Im_out, // output image |
||
83 | const uint16_t dim_im_out, // output image dimension |
||
84 | q15_t * bufferA, //buffer space for input |
||
85 | q7_t * bufferB //buffer space for output |
||
86 | ); |
||
87 | void arm_convolve_HWC_q15_nonsquare_ref(const q15_t * Im_in, |
||
88 | const uint16_t dim_im_in_x, |
||
89 | const uint16_t dim_im_in_y, |
||
90 | const uint16_t ch_im_in, |
||
91 | const q15_t * wt, |
||
92 | const uint16_t ch_im_out, |
||
93 | const uint16_t dim_kernel_x, |
||
94 | const uint16_t dim_kernel_y, |
||
95 | const uint16_t padding_x, |
||
96 | const uint16_t padding_y, |
||
97 | const uint16_t stride_x, |
||
98 | const uint16_t stride_y, |
||
99 | const q15_t * bias, |
||
100 | const uint16_t bias_shift, |
||
101 | const uint16_t out_shift, |
||
102 | q15_t * Im_out, |
||
103 | const uint16_t dim_im_out_x, |
||
104 | const uint16_t dim_im_out_y, |
||
105 | q15_t * bufferA, |
||
106 | q7_t * bufferB); |
||
107 | |||
108 | void arm_depthwise_separable_conv_HWC_q7_ref(const q7_t * Im_in, // input image |
||
109 | const uint16_t dim_im_in, // input image dimention |
||
110 | const uint16_t ch_im_in, // number of input image channels |
||
111 | const q7_t * wt, // kernel weights |
||
112 | const uint16_t ch_im_out, // number of filters, i.e., output image channels |
||
113 | const uint16_t dim_kernel, // filter kernel size |
||
114 | const uint16_t padding, // padding sizes |
||
115 | const uint16_t stride, // stride |
||
116 | const q7_t * bias, // bias |
||
117 | const uint16_t bias_shift, // amount of left-shift for bias |
||
118 | const uint16_t out_shift, // amount of right-shift for output |
||
119 | q7_t * Im_out, // output image |
||
120 | const uint16_t dim_im_out, // output image dimension |
||
121 | q15_t * bufferA, //buffer space for input |
||
122 | q7_t * bufferB //buffer space for output |
||
123 | ); |
||
124 | void arm_depthwise_separable_conv_HWC_q7_ref_nonsquare(const q7_t * Im_in, // input image |
||
125 | const uint16_t dim_im_in_x, // input image dimention x |
||
126 | const uint16_t dim_im_in_y, // input image dimention y |
||
127 | const uint16_t ch_im_in, // number of input image channels |
||
128 | const q7_t * wt, // kernel weights |
||
129 | const uint16_t ch_im_out, // number of filters, i.e., output image channels |
||
130 | const uint16_t dim_kernel_x, // filter kernel size x |
||
131 | const uint16_t dim_kernel_y, // filter kernel size y |
||
132 | const uint16_t padding_x, // padding sizes x |
||
133 | const uint16_t padding_y, // padding sizes y |
||
134 | const uint16_t stride_x, // stride x |
||
135 | const uint16_t stride_y, // stride y |
||
136 | const q7_t * bias, // bias |
||
137 | const uint16_t bias_shift, // amount of left-shift for bias |
||
138 | const uint16_t out_shift, // amount of right-shift for output |
||
139 | q7_t * Im_out, // output image |
||
140 | const uint16_t dim_im_out_x, // output image dimension x |
||
141 | const uint16_t dim_im_out_y, // output image dimension y |
||
142 | q15_t * bufferA, //buffer space for input |
||
143 | q7_t * bufferB //buffer space for output |
||
144 | ); |
||
145 | |||
146 | /* |
||
147 | * |
||
148 | * Fully-connected reference implemenation |
||
149 | * |
||
150 | */ |
||
151 | |||
152 | void arm_fully_connected_q7_ref(const q7_t * pV, // pointer to vector |
||
153 | const q7_t * pM, // pointer to matrix |
||
154 | const uint16_t dim_vec, // length of the vector |
||
155 | const uint16_t num_of_rows, // numCol of A |
||
156 | const uint16_t bias_shift, // amount of left-shift for bias |
||
157 | const uint16_t out_shift, // amount of right-shift for output |
||
158 | const q7_t * bias, q7_t * pOut, // output operand |
||
159 | q15_t * vec_buffer); |
||
160 | |||
161 | void arm_fully_connected_q15_ref(const q15_t * pV, // pointer to vector |
||
162 | const q15_t * pM, // pointer to matrix |
||
163 | const uint16_t dim_vec, // length of the vector |
||
164 | const uint16_t num_of_rows, // numCol of A |
||
165 | const uint16_t bias_shift, // amount of left-shift for bias |
||
166 | const uint16_t out_shift, // amount of right-shift for output |
||
167 | const q15_t * bias, q15_t * pOut, // output operand |
||
168 | q15_t * vec_buffer); |
||
169 | |||
170 | void arm_fully_connected_mat_q7_vec_q15_ref(const q15_t * pV, // pointer to vector |
||
171 | const q7_t * pM, // pointer to matrix |
||
172 | const uint16_t dim_vec, // length of the vector |
||
173 | const uint16_t num_of_rows, // numCol of A |
||
174 | const uint16_t bias_shift, // amount of left-shift for bias |
||
175 | const uint16_t out_shift, // amount of right-shift for output |
||
176 | const q7_t * bias, q15_t * pOut, // output operand |
||
177 | q15_t * vec_buffer); |
||
178 | |||
179 | void arm_fully_connected_q7_opt_ref(const q7_t * pV, // pointer to vector |
||
180 | const q7_t * pM, // pointer to matrix |
||
181 | const uint16_t dim_vec, // length of the vector |
||
182 | const uint16_t num_of_rows, // numCol of A |
||
183 | const uint16_t bias_shift, // amount of left-shift for bias |
||
184 | const uint16_t out_shift, // amount of right-shift for output |
||
185 | const q7_t * bias, q7_t * pOut, // output operand |
||
186 | q15_t * vec_buffer); |
||
187 | |||
188 | void arm_fully_connected_q15_opt_ref(const q15_t * pV, // pointer to vector |
||
189 | const q15_t * pM, // pointer to matrix |
||
190 | const uint16_t dim_vec, // length of the vector |
||
191 | const uint16_t num_of_rows, // numCol of A |
||
192 | const uint16_t bias_shift, // amount of left-shift for bias |
||
193 | const uint16_t out_shift, // amount of right-shift for output |
||
194 | const q15_t * bias, q15_t * pOut, // output operand |
||
195 | q15_t * vec_buffer); |
||
196 | |||
197 | void arm_fully_connected_mat_q7_vec_q15_opt_ref(const q15_t * pV, // pointer to vector |
||
198 | const q7_t * pM, // pointer to matrix |
||
199 | const uint16_t dim_vec, // length of the vector |
||
200 | const uint16_t num_of_rows, // numCol of A |
||
201 | const uint16_t bias_shift, // amount of left-shift for bias |
||
202 | const uint16_t out_shift, // amount of right-shift for output |
||
203 | const q7_t * bias, q15_t * pOut, // output operand |
||
204 | q15_t * vec_buffer); |
||
205 | |||
206 | /* |
||
207 | * |
||
208 | * Pooling reference implemenation |
||
209 | * |
||
210 | */ |
||
211 | |||
212 | void arm_avepool_q7_HWC_ref(const q7_t * Im_in, // input image |
||
213 | const uint16_t dim_im_in, // input image dimension |
||
214 | const uint16_t ch_im_in, // number of input image channels |
||
215 | const uint16_t dim_kernel, // window kernel size |
||
216 | const uint16_t padding, // padding sizes |
||
217 | const uint16_t stride, // stride |
||
218 | const uint16_t dim_im_out, // output image dimension |
||
219 | q7_t * bufferA, // a buffer for local storage |
||
220 | q7_t * Im_out); |
||
221 | |||
222 | void arm_maxpool_q7_HWC_ref(const q7_t * Im_in, // input image |
||
223 | const uint16_t dim_im_in, // input image dimension |
||
224 | const uint16_t ch_im_in, // number of input image channels |
||
225 | const uint16_t dim_kernel, // window kernel size |
||
226 | const uint16_t padding, // padding sizes |
||
227 | const uint16_t stride, // stride |
||
228 | const uint16_t dim_im_out, // output image dimension |
||
229 | q7_t * bufferA, // a buffer for local storage |
||
230 | q7_t * Im_out); |
||
231 | |||
232 | /* |
||
233 | * |
||
234 | * Other reference implemenation |
||
235 | * |
||
236 | */ |
||
237 | |||
238 | void arm_relu_q7_ref(q7_t * data, uint16_t size); |
||
239 | |||
240 | void arm_relu_q15_ref(q15_t * data, uint16_t size); |
||
241 | |||
242 | void arm_nn_mult_q7_ref(q7_t * pSrcA, q7_t * pSrcB, q7_t * pDst, const uint16_t out_shift, uint32_t blockSize); |
||
243 | |||
244 | void arm_nn_mult_q15_ref(q15_t * pSrcA, q15_t * pSrcB, q15_t * pDst, const uint16_t out_shift, uint32_t blockSize); |
||
245 | |||
246 | #ifdef __cplusplus |
||
247 | } |
||
248 | #endif |
||
249 | |||
250 | #endif |