1 /*
2 * SPDX-FileCopyrightText: Copyright 2010-2024 Arm Limited and/or its affiliates <open-source-office@arm.com>
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 /* ----------------------------------------------------------------------
20 * Project: CMSIS NN Library
21 * Title: arm_convolve_1_x_n_s8.c
22 * Description: s8 version of 1xN convolution using symmetric quantization.
23 *
24 * $Date: 19 March 2024
25 * $Revision: V.3.6.0
26 *
27 * Target : Arm(R) M-Profile Architecture
28 *
29 * -------------------------------------------------------------------- */
30
31 #include "arm_nnfunctions.h"
32 #include "arm_nnsupportfunctions.h"
33 /**
34 * @ingroup Public
35 */
36
37 /**
38 * @addtogroup NNConv
39 * @{
40 */
41
42 /*
43 * 1xN s8 convolution function.
44 *
45 * Refer header file for details.
46 *
47 */
arm_convolve_1_x_n_s8(const cmsis_nn_context * ctx,const cmsis_nn_conv_params * conv_params,const cmsis_nn_per_channel_quant_params * quant_params,const cmsis_nn_dims * input_dims,const int8_t * input_data,const cmsis_nn_dims * filter_dims,const int8_t * filter_data,const cmsis_nn_dims * bias_dims,const int32_t * bias_data,const cmsis_nn_dims * output_dims,int8_t * output_data)48 arm_cmsis_nn_status arm_convolve_1_x_n_s8(const cmsis_nn_context *ctx,
49 const cmsis_nn_conv_params *conv_params,
50 const cmsis_nn_per_channel_quant_params *quant_params,
51 const cmsis_nn_dims *input_dims,
52 const int8_t *input_data,
53 const cmsis_nn_dims *filter_dims,
54 const int8_t *filter_data,
55 const cmsis_nn_dims *bias_dims,
56 const int32_t *bias_data,
57 const cmsis_nn_dims *output_dims,
58 int8_t *output_data)
59 {
60 arm_cmsis_nn_status status = ARM_CMSIS_NN_SUCCESS;
61
62 /* The wrapper API is the ultimate reference for argument check */
63 if ((input_dims->h != 1) || conv_params->dilation.w != 1 || ctx->buf == NULL || conv_params->stride.w == 0 ||
64 (conv_params->stride.w * input_dims->c % 4 != 0))
65 {
66 return ARM_CMSIS_NN_ARG_ERROR;
67 }
68
69 #if defined(ARM_MATH_MVEI)
70 (void)bias_dims;
71
72 const int32_t input_x = input_dims->w;
73 const int32_t kernel_x = filter_dims->w;
74 const int32_t output_x = output_dims->w;
75 const int32_t input_ch = input_dims->c;
76 const int32_t pad_x = conv_params->padding.w;
77 const int32_t stride_x = conv_params->stride.w;
78
79 // Total pad for dilation of 1
80 const int32_t total_pad = ((output_x - 1) * stride_x + kernel_x - input_x);
81 const int32_t asym_pad = total_pad % 2;
82
83 if (pad_x * 2 + asym_pad != total_pad)
84 {
85 return ARM_CMSIS_NN_FAILURE;
86 }
87
88 const int32_t right_pad_num = pad_x + asym_pad != 0 ? MAX(1, (pad_x + asym_pad + stride_x - 1) / stride_x) : 0;
89 const int32_t left_pad_num = pad_x != 0 ? MAX(1, (pad_x + stride_x - 1) / stride_x) : 0;
90 const int32_t no_pad_num = MAX(output_x - (right_pad_num + left_pad_num), 0);
91
92 const int32_t pad_size_left = pad_x * input_ch;
93 const int32_t pad_size_right = asym_pad ? right_pad_num * input_ch : pad_size_left;
94
95 const int32_t rhs_cols = kernel_x * input_ch;
96 const int32_t rhs_rows = output_dims->c;
97 const int32_t lhs_offset = input_ch * stride_x;
98
99 if (right_pad_num + no_pad_num + left_pad_num != output_x)
100 {
101 return arm_convolve_s8(ctx,
102 conv_params,
103 quant_params,
104 input_dims,
105 input_data,
106 filter_dims,
107 filter_data,
108 bias_dims,
109 bias_data,
110 output_dims,
111 output_data);
112 }
113
114 const uint32_t num_elem_left = kernel_x * input_ch;
115 const uint32_t num_elem_right = num_elem_left - input_ch;
116
117 for (int i_batch = 0; i_batch < input_dims->n; i_batch++)
118 {
119 /* Handle left padded sections */
120 int32_t lhs_rows = left_pad_num;
121 int8_t *im2col = ctx->buf;
122
123 arm_memset_s8(im2col, (int8_t)-conv_params->input_offset, sizeof(int8_t) * (uint32_t)pad_size_left);
124 im2col += pad_size_left;
125 arm_memcpy_s8(im2col, input_data, sizeof(int8_t) * num_elem_left);
126
127 arm_nn_mat_mult_nt_t_s8((int8_t *)ctx->buf,
128 filter_data,
129 bias_data,
130 output_data,
131 quant_params->multiplier,
132 quant_params->shift,
133 lhs_rows,
134 rhs_rows,
135 rhs_cols,
136 conv_params->input_offset,
137 conv_params->output_offset,
138 conv_params->activation.min,
139 conv_params->activation.max,
140 rhs_rows,
141 lhs_offset);
142
143 output_data += lhs_rows * rhs_rows;
144
145 /* Non padded elements */
146 int32_t out_idx = lhs_rows;
147 int32_t input_start = stride_x * lhs_rows - pad_x;
148
149 if (input_start < 0)
150 {
151 return ARM_CMSIS_NN_FAILURE;
152 }
153
154 input_start *= input_ch;
155 lhs_rows = no_pad_num;
156
157 arm_nn_mat_mult_nt_t_s8(input_data + input_start,
158 filter_data,
159 bias_data,
160 output_data,
161 quant_params->multiplier,
162 quant_params->shift,
163 lhs_rows,
164 rhs_rows,
165 rhs_cols,
166 conv_params->input_offset,
167 conv_params->output_offset,
168 conv_params->activation.min,
169 conv_params->activation.max,
170 rhs_rows,
171 lhs_offset);
172
173 output_data += lhs_rows * rhs_rows;
174 out_idx += lhs_rows;
175
176 /* Right padded elements */
177 lhs_rows = output_x - out_idx;
178
179 if (lhs_rows < 0)
180 {
181 return ARM_CMSIS_NN_FAILURE;
182 }
183
184 im2col = ctx->buf;
185 input_start = (stride_x * (left_pad_num + no_pad_num) - pad_x) * input_ch;
186
187 arm_memcpy_s8(im2col, input_data + input_start, sizeof(int8_t) * num_elem_right);
188 im2col += num_elem_right;
189 arm_memset_s8(im2col, (int8_t)-conv_params->input_offset, sizeof(int8_t) * (uint32_t)pad_size_right);
190
191 arm_nn_mat_mult_nt_t_s8((int8_t *)ctx->buf,
192 filter_data,
193 bias_data,
194 output_data,
195 quant_params->multiplier,
196 quant_params->shift,
197 lhs_rows,
198 rhs_rows,
199 rhs_cols,
200 conv_params->input_offset,
201 conv_params->output_offset,
202 conv_params->activation.min,
203 conv_params->activation.max,
204 rhs_rows,
205 lhs_offset);
206
207 output_data += lhs_rows * rhs_rows;
208
209 /* Advance to the next batch */
210 input_data += (input_x * input_ch);
211 }
212 #else
213 status = arm_convolve_s8(ctx,
214 conv_params,
215 quant_params,
216 input_dims,
217 input_data,
218 filter_dims,
219 filter_data,
220 bias_dims,
221 bias_data,
222 output_dims,
223 output_data);
224
225 #endif
226
227 /* Return to application */
228 return status;
229 }
230
231 /**
232 * @} end of NNConv group
233 */
234
