| /cmsis-nn-latest/Tests/UnitTest/TestCases/Common/ |
| D | conv2d_s4_weights_template.json | 36 output_ch,
54 output_ch
72 output_ch
|
| D | dw_s4_weights_template.json | 38 output_ch
51 output_ch
67 output_ch
|
| /cmsis-nn-latest/Source/ConvolutionFunctions/ |
| D | arm_transpose_conv_s8.c | 80 const int32_t output_ch = output_dims->c; in arm_transpose_conv_s8() local
112 … arm_memset_s8((int8_t *)img_buf_ptr, 0, output_x * output_y * output_ch * sizeof(int32_t)); in arm_transpose_conv_s8()
120 memcpy(img_data, bias_data, output_ch * sizeof(int32_t)); in arm_transpose_conv_s8()
121 img_data += output_ch; in arm_transpose_conv_s8()
130 for (int i_output_ch = 0; i_output_ch < output_ch; i_output_ch++) in arm_transpose_conv_s8()
139 output_ch); in arm_transpose_conv_s8()
155 int32_t *dst_data = img_data + (i_pad_y * output_x + i_pad_x) * output_ch; in arm_transpose_conv_s8()
163 for (int i_output_ch = 0; i_output_ch < output_ch; i_output_ch++) in arm_transpose_conv_s8()
168 dst_data += output_ch; in arm_transpose_conv_s8()
169 col_buf += output_ch; in arm_transpose_conv_s8()
[all …]
|
| D | arm_nn_mat_mult_s8.c | 41 const uint16_t output_ch, in arm_nn_mat_mult_s8() argument
58 for (int i_out_ch = 0; i_out_ch < output_ch; i_out_ch++) in arm_nn_mat_mult_s8()
115 const uint32x4_t scatter_offset = {0, output_ch, output_ch * 2, output_ch * 3}; in arm_nn_mat_mult_s8()
118 out += 4 * output_ch; in arm_nn_mat_mult_s8()
124 for (int i_out_ch = 0; i_out_ch < output_ch; i_out_ch++) in arm_nn_mat_mult_s8()
158 out += output_ch; in arm_nn_mat_mult_s8()
166 (void)output_ch; in arm_nn_mat_mult_s8()
|
| D | arm_convolve_s8.c | 78 const uint16_t output_ch = output_dims->c; in arm_convolve_s8() local
93 const int32_t output_ch_per_group = output_ch / groups; in arm_convolve_s8()
98 if (input_ch % groups != 0 || output_ch % groups != 0) in arm_convolve_s8()
178 output_ch, in arm_convolve_s8()
181 out += lhs_rows * output_ch; in arm_convolve_s8()
214 output_ch, in arm_convolve_s8()
264 output_ch, in arm_convolve_s8()
267 out += lhs_rows * output_ch; in arm_convolve_s8()
337 output_data += (output_x * output_y * output_ch); in arm_convolve_s8()
|
| D | arm_convolve_1_x_n_s4.c | 76 const uint16_t output_ch = output_dims->c; in arm_convolve_1_x_n_s4() local
118 output_ch, in arm_convolve_1_x_n_s4()
123 output_data += output_ch; in arm_convolve_1_x_n_s4()
169 output_ch, in arm_convolve_1_x_n_s4()
174 output_data += output_ch; in arm_convolve_1_x_n_s4()
|
| D | arm_nn_mat_mult_kernel_s8_s16.c | 42 const uint16_t output_ch, in arm_nn_mat_mult_kernel_s8_s16() argument
55 int8_t *out_1 = out_0 + output_ch; in arm_nn_mat_mult_kernel_s8_s16()
58 uint16_t row_count = output_ch / 2; in arm_nn_mat_mult_kernel_s8_s16()
162 if (output_ch & 0x1) in arm_nn_mat_mult_kernel_s8_s16()
227 out_0 += output_ch; in arm_nn_mat_mult_kernel_s8_s16()
234 (void)output_ch; in arm_nn_mat_mult_kernel_s8_s16()
|
| D | arm_convolve_s16.c | 78 const int32_t output_ch = output_dims->c; in arm_convolve_s16() local
150 out += lhs_rows * output_ch; in arm_convolve_s16()
161 output_ch, in arm_convolve_s16()
209 for (i = 0; i < output_ch; i++) in arm_convolve_s16()
283 output_data += (output_x * output_y * output_ch); in arm_convolve_s16()
|
| D | arm_nn_mat_mult_kernel_row_offset_s8_s16.c | 42 const uint16_t output_ch, in arm_nn_mat_mult_kernel_row_offset_s8_s16() argument
61 uint16_t row_count = output_ch / 2; in arm_nn_mat_mult_kernel_row_offset_s8_s16()
167 if (output_ch & 0x1) in arm_nn_mat_mult_kernel_row_offset_s8_s16()
234 out_0 += 2 * row_address_offset - output_ch; in arm_nn_mat_mult_kernel_row_offset_s8_s16()
241 (void)output_ch; in arm_nn_mat_mult_kernel_row_offset_s8_s16()
|
| D | arm_depthwise_conv_s8.c | 52 const int32_t output_ch, in depthwise_conv_s8_mult_4() argument
83 for (int32_t in_ch = 0, out_ch = 0, ker_w_start = MAX(0, -in_w); out_ch < output_ch; in depthwise_conv_s8_mult_4()
99 … int32_t ker_idx = ker_h * (output_ch * kernel_x) + ker_w_start * output_ch + out_ch; in depthwise_conv_s8_mult_4()
106 ++ker_w, kernel += output_ch) in depthwise_conv_s8_mult_4()
160 const uint16_t output_ch, in depthwise_conv_s8_generic() argument
182 (void)output_ch; in depthwise_conv_s8_generic()
|
| D | arm_nn_mat_mult_kernel_s16.c | 51 const int32_t output_ch, in arm_nn_mat_mult_kernel_s16() argument
68 int16_t *out_1 = out_0 + output_ch; in arm_nn_mat_mult_kernel_s16()
69 int32_t row_count = output_ch / 2; in arm_nn_mat_mult_kernel_s16()
240 if (output_ch & 0x1) in arm_nn_mat_mult_kernel_s16()
347 out_0 += output_ch; in arm_nn_mat_mult_kernel_s16()
354 (void)output_ch; in arm_nn_mat_mult_kernel_s16()
|
| D | arm_depthwise_conv_s16.c | 48 const int32_t output_ch, in depthwise_conv_s16_mult_4_s16() argument
69 for (int32_t in_ch = 0, out_ch = 0, ker_w_start = MAX(0, -in_w); out_ch < output_ch; in depthwise_conv_s16_mult_4_s16()
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()
97 ++ker_w, ker_idx += output_ch) in depthwise_conv_s16_mult_4_s16()
|
| D | arm_convolve_s4.c | 78 const uint16_t output_ch = output_dims->c; in arm_convolve_s4() local
224 output_ch, in arm_convolve_s4()
252 for (i = 0; i < output_ch; i++) in arm_convolve_s4()
326 output_data += (output_x * output_y * output_ch); in arm_convolve_s4()
|
| D | arm_depthwise_conv_s8_opt.c | 64 const int32_t output_ch = output_dims->c; in arm_depthwise_conv_s8_opt() local
67 if (input_ch != output_ch) in arm_depthwise_conv_s8_opt()
104 int32_t remaining_ch = output_ch; in arm_depthwise_conv_s8_opt()
268 row_count = output_ch / 4; in arm_depthwise_conv_s8_opt()
374 row_count = output_ch & 0x3; in arm_depthwise_conv_s8_opt()
|
| D | arm_nn_mat_mult_kernel_s4_s16.c | 41 const uint16_t output_ch, in arm_nn_mat_mult_kernel_s4_s16() argument
53 int8_t *out_1 = out_0 + output_ch; in arm_nn_mat_mult_kernel_s4_s16()
56 uint16_t row_count = output_ch / 4; in arm_nn_mat_mult_kernel_s4_s16()
333 while (left_over_rows < output_ch % 4) in arm_nn_mat_mult_kernel_s4_s16()
435 out_0 += output_ch; in arm_nn_mat_mult_kernel_s4_s16()
|
| D | arm_depthwise_conv_s4_opt.c | 66 const int32_t output_ch = output_dims->c; in arm_depthwise_conv_s4_opt() local
69 if (input_ch != output_ch) in arm_depthwise_conv_s4_opt()
105 int32_t remaining_ch = output_ch; in arm_depthwise_conv_s4_opt()
310 row_count = output_ch / 4; in arm_depthwise_conv_s4_opt()
317 if (output_ch % 2) /* Uneven number of channels */ in arm_depthwise_conv_s4_opt()
462 row_count = output_ch & 0x3; in arm_depthwise_conv_s4_opt()
637 if (output_ch & 0x2) in arm_depthwise_conv_s4_opt()
|
| D | arm_depthwise_conv_fast_s16.c | 64 const int32_t output_ch = output_dims->c; in arm_depthwise_conv_fast_s16() local
67 if (input_ch != output_ch) in arm_depthwise_conv_fast_s16()
279 row_count = output_ch / 4; in arm_depthwise_conv_fast_s16()
395 row_count = output_ch & 0x3; in arm_depthwise_conv_fast_s16()
|
| D | arm_depthwise_conv_3x3_s8.c | 70 const int32_t output_ch = output_dims->c; in arm_depthwise_conv_3x3_s8() local
85 if (input_ch != output_ch) in arm_depthwise_conv_3x3_s8()
|
| D | arm_depthwise_conv_s4.c | 49 const int32_t output_ch, in depthwise_conv_s4_generic() argument
71 (void)output_ch; in depthwise_conv_s4_generic()
|
| /cmsis-nn-latest/Tests/UnitTest/ |
| D | conv_settings.py | 105 self.channel_multiplier = self.output_ch // self.input_ch
106 if self.output_ch % self.input_ch != 0:
147 num_channels = self.output_ch
167 self.scaling_factors = np.random.uniform(0.001, 0.01, [self.output_ch]).tolist()
219 out_channel = self.output_ch
243 bias_scale = [64751.269531] * self.output_ch
244 bias_zp = [0] * self.output_ch
253 bias_scale = [bias_scale] * self.output_ch
254 bias_zp = [bias_zp] * self.output_ch
259 scaling_factors = np.random.uniform(0.001, 0.01, [self.output_ch]).tolist()
[all …]
|
| D | fully_connected_settings.py | 165 fc_weights_format = [self.input_ch * self.y_input * self.x_input * self.output_ch]
189 fc_weights_format = [self.input_ch * self.y_input * self.x_input * self.output_ch]
227 fc_weights_format = [self.input_ch * self.y_input * self.x_input, self.output_ch]
243 …fully_connected_layer = keras.layers.Dense(self.output_ch, activation=None, use_bias=self.generate…
|
| D | model_extractor.py | 95 f.write("#define {}_OUT_CH {}\n".format(prefix, self.output_ch))
100 … self.x_output * self.y_output * self.output_ch * self.batches))
150 [output_ch, self.filter_y, self.filter_x, self.input_ch] = filter_shape
154 [self.input_ch, self.output_ch] = filter_shape
238 self.output_ch = len(scaling_factors)
|
| D | test_settings.py | 147 self.output_ch = out_ch
287 biases = tf.reshape(np.full([self.output_ch], 0), [self.output_ch])
289 biases = tf.reshape(biases, [self.output_ch])
291 biases = self.get_randomized_data([self.output_ch],
349 f.write("#define {}_OUT_CH {}\n".format(prefix, self.output_ch))
354 prefix, self.x_output * self.y_output * self.output_ch * self.batches))
|
| /cmsis-nn-latest/Include/ |
| D | arm_nnsupportfunctions.h | 269 const uint16_t output_ch,
306 const int32_t output_ch,
1154 const uint16_t output_ch,
1188 const uint16_t output_ch,
1228 const uint16_t output_ch,
|
