| /cmsis-nn-latest/Source/ConvolutionFunctions/ |
| D | arm_convolve_1_x_n_s4.c | 74 const uint16_t kernel_x = filter_dims->w; in arm_convolve_1_x_n_s4() local
82 const int32_t total_pad = ((output_x - 1) * stride_x + kernel_x - input_x); in arm_convolve_1_x_n_s4()
103 const int32_t rhs_cols = kernel_x * input_dims->c; in arm_convolve_1_x_n_s4()
113 const int32_t actual_kernel_len = kernel_x - ker_begin_idx; in arm_convolve_1_x_n_s4()
164 const int32_t ker_end_idx = MIN(kernel_x, input_x - est_input_x_idx); in arm_convolve_1_x_n_s4()
166 (kernel_x - ker_end_idx) * input_ch, in arm_convolve_1_x_n_s4()
|
| D | arm_depthwise_conv_s4.c | 51 const int32_t kernel_x, in depthwise_conv_s4_generic() argument
107 ker_x_end = MIN(kernel_x, end_min_x); in depthwise_conv_s4_generic()
112 ker_x_end = MIN(kernel_x, input_x - base_idx_x); in depthwise_conv_s4_generic()
141 … (i_ker_y * kernel_x + ker_x_start) * kernel_index_offset + idx_out_ch_s4; in depthwise_conv_s4_generic()
199 ker_x_end = MIN(kernel_x, end_min_x); in depthwise_conv_s4_generic()
204 ker_x_end = MIN(kernel_x, input_x - base_idx_x); in depthwise_conv_s4_generic()
232 … (i_ker_y * kernel_x + ker_x_start) * (kernel_index_offset * ch_mult) + in depthwise_conv_s4_generic()
291 ker_x_end = MIN(kernel_x, end_min_x); in depthwise_conv_s4_generic()
296 ker_x_end = MIN(kernel_x, input_x - base_idx_x); in depthwise_conv_s4_generic()
325 … (i_ker_y * kernel_x + ker_x_start) * (kernel_index_offset * ch_mult) + in depthwise_conv_s4_generic()
[all …]
|
| D | arm_depthwise_conv_s8_opt.c | 80 const int32_t kernel_x = filter_dims->w; in arm_depthwise_conv_s8_opt() local
101 const int32_t kernel_size = kernel_x * kernel_y; in arm_depthwise_conv_s8_opt()
119 for (int i_ker_x = base_idx_x; i_ker_x < base_idx_x + kernel_x; i_ker_x++) in arm_depthwise_conv_s8_opt()
236 … memset(&col_buffer[index], 0, (kernel_x * input_ch) * ker_y_start * sizeof(int16_t)); in arm_depthwise_conv_s8_opt()
237 index += (kernel_x * input_ch) * ker_y_start; in arm_depthwise_conv_s8_opt()
244 for (int i_ker_x = 0; i_ker_x < kernel_x; i_ker_x++) in arm_depthwise_conv_s8_opt()
265 memset(&col_buffer[index], 0, (kernel_x * input_ch) * diff * sizeof(int16_t)); in arm_depthwise_conv_s8_opt()
288 uint16_t col_count = (kernel_x * kernel_y) / 2; in arm_depthwise_conv_s8_opt()
335 col_count = (kernel_x * kernel_y) & 0x1; in arm_depthwise_conv_s8_opt()
384 const uint16_t col_count = (kernel_x * kernel_y); in arm_depthwise_conv_s8_opt()
|
| D | arm_depthwise_conv_s16.c | 50 const int32_t kernel_x, in depthwise_conv_s16_mult_4_s16() argument
91 … int32_t ker_idx = ker_h * (output_ch * kernel_x) + ker_w_start * output_ch + out_ch; in depthwise_conv_s16_mult_4_s16()
96 for (int32_t ker_w = ker_w_start; ker_w < MIN(kernel_x, input_x - in_w); in depthwise_conv_s16_mult_4_s16()
140 const uint16_t kernel_x, in depthwise_conv_s16_generic_s16() argument
185 ker_x_end = MIN(kernel_x, end_min_x); in depthwise_conv_s16_generic_s16()
190 ker_x_end = MIN(kernel_x, input_x - base_idx_x); in depthwise_conv_s16_generic_s16()
218 … int32_t ker_idx_0 = (i_ker_y * kernel_x + i_ker_x) * (input_ch * ch_mult) + idx_out_ch; in depthwise_conv_s16_generic_s16()
|
| D | arm_depthwise_conv_fast_s16.c | 87 const int32_t kernel_x = filter_dims->w; in arm_depthwise_conv_fast_s16() local
105 const int32_t kernel_size = kernel_x * kernel_y; in arm_depthwise_conv_fast_s16()
116 for (int i_ker_x = base_idx_x; i_ker_x < base_idx_x + kernel_x; i_ker_x++) in arm_depthwise_conv_fast_s16()
247 … memset(&col_buffer[index], 0, (kernel_x * input_ch) * ker_y_start * sizeof(int16_t)); in arm_depthwise_conv_fast_s16()
248 index += (kernel_x * input_ch) * ker_y_start; in arm_depthwise_conv_fast_s16()
255 for (int i_ker_x = 0; i_ker_x < kernel_x; i_ker_x++) in arm_depthwise_conv_fast_s16()
276 memset(&col_buffer[index], 0, (kernel_x * input_ch) * diff * sizeof(int16_t)); in arm_depthwise_conv_fast_s16()
298 uint16_t col_count = (kernel_x * kernel_y) / 2; in arm_depthwise_conv_fast_s16()
345 col_count = (kernel_x * kernel_y) & 0x1; in arm_depthwise_conv_fast_s16()
401 const uint16_t col_count = (kernel_x * kernel_y); in arm_depthwise_conv_fast_s16()
|
| D | arm_convolve_1_x_n_s8.c | 73 const int32_t kernel_x = filter_dims->w; in arm_convolve_1_x_n_s8() local
80 const int32_t total_pad = ((output_x - 1) * stride_x + kernel_x - input_x); in arm_convolve_1_x_n_s8()
95 const int32_t rhs_cols = kernel_x * input_ch; in arm_convolve_1_x_n_s8()
114 const uint32_t num_elem_left = kernel_x * input_ch; in arm_convolve_1_x_n_s8()
|
| D | arm_depthwise_conv_s8.c | 54 const int32_t kernel_x, in depthwise_conv_s8_mult_4() argument
99 … int32_t ker_idx = ker_h * (output_ch * kernel_x) + ker_w_start * output_ch + out_ch; in depthwise_conv_s8_mult_4()
105 for (int32_t ker_w = ker_w_start; ker_w < MIN(kernel_x, input_x - in_w); in depthwise_conv_s8_mult_4()
162 const uint16_t kernel_x, in depthwise_conv_s8_generic() argument
211 ker_x_end = MIN(kernel_x, end_min_x); in depthwise_conv_s8_generic()
216 ker_x_end = MIN(kernel_x, input_x - base_idx_x); in depthwise_conv_s8_generic()
244 … int32_t ker_idx_0 = (i_ker_y * kernel_x + i_ker_x) * (input_ch * ch_mult) + idx_out_ch; in depthwise_conv_s8_generic()
|
| D | arm_transpose_conv_s8.c | 75 const uint16_t kernel_x = filter_dims->w; in arm_transpose_conv_s8() local
148 const int32_t col_x = (output_x + pad_x_offset + pad_x - kernel_x) / stride_x + 1; in arm_transpose_conv_s8()
159 for (int32_t i_ker_x = i_pad_x; i_ker_x < i_pad_x + kernel_x; i_ker_x++) in arm_transpose_conv_s8()
171 dst_data += (output_x - kernel_x) * output_ch; in arm_transpose_conv_s8()
|
| D | arm_convolve_get_buffer_sizes_s8.c | 62 const int32_t kernel_x = filter_dims->w; in arm_convolve_1_x_n_s8_get_buffer_size_mve() local
65 const int32_t total_pad = ((output_x - 1) * stride_x + kernel_x - input_x); in arm_convolve_1_x_n_s8_get_buffer_size_mve()
79 const int32_t num_elem_left = kernel_x * input_dims->c; in arm_convolve_1_x_n_s8_get_buffer_size_mve()
|
| D | arm_convolve_s4.c | 74 const uint16_t kernel_x = filter_dims->w; in arm_convolve_s4() local
89 const int32_t rhs_cols = kernel_x * kernel_y * input_ch; in arm_convolve_s4()
115 for (int32_t i_ker_x = 0; i_ker_x < kernel_x; i_ker_x++) in arm_convolve_s4()
199 for (int32_t i_ker_x = 0; i_ker_x < kernel_x; i_ker_x++) in arm_convolve_s4()
|
| D | arm_convolve_s16.c | 74 const int32_t kernel_x = filter_dims->w; in arm_convolve_s16() local
79 const int32_t rhs_cols = input_ch * kernel_y * kernel_x; in arm_convolve_s16()
114 for (int32_t i_ker_x = 0; i_ker_x < kernel_x; i_ker_x++) in arm_convolve_s16()
|
| D | arm_convolve_get_buffer_sizes_s4.c | 62 const int32_t kernel_x = filter_dims->w; in arm_convolve_1_x_n_s4_get_buffer_size_mve() local
65 const int32_t total_pad = ((output_x - 1) * stride_x + kernel_x - input_x); in arm_convolve_1_x_n_s4_get_buffer_size_mve()
|
| D | arm_depthwise_conv_s4_opt.c | 81 const int32_t kernel_x = filter_dims->w; in arm_depthwise_conv_s4_opt() local
102 const int32_t kernel_size = kernel_x * kernel_y; in arm_depthwise_conv_s4_opt()
120 for (int i_ker_x = base_idx_x; i_ker_x < base_idx_x + kernel_x; i_ker_x++) in arm_depthwise_conv_s4_opt()
257 const uint16_t num_cols = kernel_x * kernel_y; in arm_depthwise_conv_s4_opt()
278 … memset(&col_buffer[index], 0, (kernel_x * input_ch) * ker_y_start * sizeof(int16_t)); in arm_depthwise_conv_s4_opt()
279 index += (kernel_x * input_ch) * ker_y_start; in arm_depthwise_conv_s4_opt()
286 for (int i_ker_x = 0; i_ker_x < kernel_x; i_ker_x++) in arm_depthwise_conv_s4_opt()
307 memset(&col_buffer[index], 0, (kernel_x * input_ch) * diff * sizeof(int16_t)); in arm_depthwise_conv_s4_opt()
|
| D | arm_convolve_s8.c | 73 const uint16_t kernel_x = filter_dims->w; in arm_convolve_s8() local
92 const int32_t rhs_cols = kernel_x * kernel_y * kernel_ch; in arm_convolve_s8()
139 for (int32_t i_ker_x = 0; i_ker_x < kernel_x; i_ker_x++) in arm_convolve_s8()
|
| /cmsis-nn-latest/Source/PoolingFunctions/ |
| D | arm_avgpool_s8.c | 90 const int32_t kernel_x = filter_dims->w; in arm_avgpool_s8() local
115 const int32_t k_x_end = MIN(i_x * stride_x - pad_x + kernel_x, input_x); in arm_avgpool_s8()
251 const int32_t kernel_x = filter_dims->w; in arm_avgpool_s8() local
289 const int32_t kernel_x_end = MIN(kernel_x, input_x - idx_x); in arm_avgpool_s8()
351 … for (int k_x = i_x * stride_x - pad_x; k_x < i_x * stride_x - pad_x + kernel_x; k_x++) in arm_avgpool_s8()
|
| D | arm_avgpool_s16.c | 88 const int32_t kernel_x = filter_dims->w; in arm_avgpool_s16() local
117 const int32_t k_x_end = MIN(i_x * stride_x - pad_x + kernel_x, input_x); in arm_avgpool_s16()
225 const int32_t kernel_x_end = MIN(kernel_x, input_x - idx_x); in arm_avgpool_s16()
286 const int32_t kernel_x_end = MIN(kernel_x, input_x - base_idx_x); in arm_avgpool_s16()
|
| D | arm_max_pool_s16.c | 161 const int32_t kernel_x = filter_dims->w; in arm_max_pool_s16() local
190 const int32_t kernel_x_end = MIN(kernel_x, input_x - base_idx_x); in arm_max_pool_s16()
|
| D | arm_max_pool_s8.c | 178 const int32_t kernel_x = filter_dims->w; in arm_max_pool_s8() local
206 const int32_t kernel_x_end = MIN(kernel_x, input_x - base_idx_x); in arm_max_pool_s8()
|
