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: 04 November 2024
25 * $Revision: V.3.6.1
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 NULL,
111 output_dims,
112 output_data);
113 }
114
115 const uint32_t num_elem_left = kernel_x * input_ch;
116 const uint32_t num_elem_right = num_elem_left - input_ch;
117
118 for (int i_batch = 0; i_batch < input_dims->n; i_batch++)
119 {
120 /* Handle left padded sections */
121 int32_t lhs_rows = left_pad_num;
122 int8_t *im2col = ctx->buf;
123
124 arm_memset_s8(im2col, (int8_t)-conv_params->input_offset, sizeof(int8_t) * (uint32_t)pad_size_left);
125 im2col += pad_size_left;
126 arm_memcpy_s8(im2col, input_data, sizeof(int8_t) * num_elem_left);
127
128 arm_nn_mat_mult_nt_t_s8((int8_t *)ctx->buf,
129 filter_data,
130 bias_data,
131 output_data,
132 quant_params->multiplier,
133 quant_params->shift,
134 lhs_rows,
135 rhs_rows,
136 rhs_cols,
137 conv_params->input_offset,
138 conv_params->output_offset,
139 conv_params->activation.min,
140 conv_params->activation.max,
141 rhs_rows,
142 lhs_offset);
143
144 output_data += lhs_rows * rhs_rows;
145
146 /* Non padded elements */
147 int32_t out_idx = lhs_rows;
148 int32_t input_start = stride_x * lhs_rows - pad_x;
149
150 if (input_start < 0)
151 {
152 return ARM_CMSIS_NN_FAILURE;
153 }
154
155 input_start *= input_ch;
156 lhs_rows = no_pad_num;
157
158 arm_nn_mat_mult_nt_t_s8(input_data + input_start,
159 filter_data,
160 bias_data,
161 output_data,
162 quant_params->multiplier,
163 quant_params->shift,
164 lhs_rows,
165 rhs_rows,
166 rhs_cols,
167 conv_params->input_offset,
168 conv_params->output_offset,
169 conv_params->activation.min,
170 conv_params->activation.max,
171 rhs_rows,
172 lhs_offset);
173
174 output_data += lhs_rows * rhs_rows;
175 out_idx += lhs_rows;
176
177 /* Right padded elements */
178 lhs_rows = output_x - out_idx;
179
180 if (lhs_rows < 0)
181 {
182 return ARM_CMSIS_NN_FAILURE;
183 }
184
185 im2col = ctx->buf;
186 input_start = (stride_x * (left_pad_num + no_pad_num) - pad_x) * input_ch;
187
188 arm_memcpy_s8(im2col, input_data + input_start, sizeof(int8_t) * num_elem_right);
189 im2col += num_elem_right;
190 arm_memset_s8(im2col, (int8_t)-conv_params->input_offset, sizeof(int8_t) * (uint32_t)pad_size_right);
191
192 arm_nn_mat_mult_nt_t_s8((int8_t *)ctx->buf,
193 filter_data,
194 bias_data,
195 output_data,
196 quant_params->multiplier,
197 quant_params->shift,
198 lhs_rows,
199 rhs_rows,
200 rhs_cols,
201 conv_params->input_offset,
202 conv_params->output_offset,
203 conv_params->activation.min,
204 conv_params->activation.max,
205 rhs_rows,
206 lhs_offset);
207
208 output_data += lhs_rows * rhs_rows;
209
210 /* Advance to the next batch */
211 input_data += (input_x * input_ch);
212 }
213 #else
214 status = arm_convolve_s8(ctx,
215 conv_params,
216 quant_params,
217 input_dims,
218 input_data,
219 filter_dims,
220 filter_data,
221 bias_dims,
222 bias_data,
223 NULL,
224 output_dims,
225 output_data);
226
227 #endif
228
229 /* Return to application */
230 return status;
231 }
232
233 /**
234 * @} end of NNConv group
235 */
236
