| /cmsis-nn-latest/Source/NNSupportFunctions/ |
| D | arm_nn_vec_mat_mul_result_acc_s8_s16.c | 76 int32_t acc_0 = *effective_bias_ptr++; in arm_nn_vec_mat_mul_result_acc_s8_s16() local
91 acc_0 = vmladavaq_s8(acc_0, ker_0, input); in arm_nn_vec_mat_mul_result_acc_s8_s16()
109 int32x4_t acc = {acc_0, acc_1, acc_2, acc_3}; in arm_nn_vec_mat_mul_result_acc_s8_s16()
123 int32_t acc_0 = *effective_bias_ptr++; in arm_nn_vec_mat_mul_result_acc_s8_s16() local
137 acc_0 = vmladavaq_s8(acc_0, ker_0, input); in arm_nn_vec_mat_mul_result_acc_s8_s16()
144 acc_0 = arm_nn_requantize(acc_0, dst_multiplier, dst_shift); in arm_nn_vec_mat_mul_result_acc_s8_s16()
145 acc_0 += *dst; in arm_nn_vec_mat_mul_result_acc_s8_s16()
148 acc_0 = MAX(acc_0, NN_Q15_MIN); in arm_nn_vec_mat_mul_result_acc_s8_s16()
149 acc_0 = MIN(acc_0, NN_Q15_MAX); in arm_nn_vec_mat_mul_result_acc_s8_s16()
150 *dst++ = (int16_t)acc_0; in arm_nn_vec_mat_mul_result_acc_s8_s16()
[all …]
|
| D | arm_nn_vec_mat_mult_t_s8.c | 82 int32_t acc_0 = 0; in arm_nn_vec_mat_mult_t_s8() local
99 acc_0 = *bias++; in arm_nn_vec_mat_mult_t_s8()
116 acc_0 = vmladavaq_s8(acc_0, ker_0, input); in arm_nn_vec_mat_mult_t_s8()
135 int32x4_t acc = {acc_0, acc_1, acc_2, acc_3}; in arm_nn_vec_mat_mult_t_s8()
157 int32_t acc_0 = 0; in arm_nn_vec_mat_mult_t_s8() local
172 acc_0 = vmladavaq_s8(acc_0, ker_0, input); in arm_nn_vec_mat_mult_t_s8()
181 acc_0 += *bias; in arm_nn_vec_mat_mult_t_s8()
185 acc_0 += rhs_sum * lhs_offset; in arm_nn_vec_mat_mult_t_s8()
186 acc_0 += lhs_sum * rhs_offset; in arm_nn_vec_mat_mult_t_s8()
187 acc_0 += rhs_cols * lhs_offset * rhs_offset; in arm_nn_vec_mat_mult_t_s8()
[all …]
|
| D | arm_nn_vec_mat_mult_t_svdf_s8.c | 71 int32_t acc_0 = 0; in arm_nn_vec_mat_mult_t_svdf_s8() local
97 acc_0 = vmladavaq_s8(acc_0, ker_0, input); in arm_nn_vec_mat_mult_t_svdf_s8()
114 int32x4_t acc = {acc_0, acc_1, acc_2, 0}; in arm_nn_vec_mat_mult_t_svdf_s8()
130 int32_t acc_0 = 0; in arm_nn_vec_mat_mult_t_svdf_s8() local
145 acc_0 = vmladavaq_s8(acc_0, ker_0, input); in arm_nn_vec_mat_mult_t_svdf_s8()
153 acc_0 = QADD(acc_0, offsets); in arm_nn_vec_mat_mult_t_svdf_s8()
154 acc_0 = arm_nn_requantize(acc_0, dst_multiplier, dst_shift); in arm_nn_vec_mat_mult_t_svdf_s8()
157 acc_0 = MAX(acc_0, activation_min); in arm_nn_vec_mat_mult_t_svdf_s8()
158 *dst = (int16_t)MIN(acc_0, activation_max); in arm_nn_vec_mat_mult_t_svdf_s8()
170 int32_t acc_0 = 0; in arm_nn_vec_mat_mult_t_svdf_s8() local
[all …]
|
| D | arm_nn_vec_mat_mult_t_s16.c | 225 int32_t acc_0 = 0; in arm_nn_vec_mat_mult_t_s16() local
244 acc_0 = SMLAD(ker_0, vec_part_0, acc_0); in arm_nn_vec_mat_mult_t_s16()
245 acc_0 = SMLAD(ker_1, vec_part_1, acc_0); in arm_nn_vec_mat_mult_t_s16()
253 acc_64_0 += acc_0; in arm_nn_vec_mat_mult_t_s16()
287 int32_t acc_0 = 0; in arm_nn_vec_mat_mult_t_s16() local
299 acc_0 = SMLAD(ker_0, vec, acc_0); in arm_nn_vec_mat_mult_t_s16()
302 acc_0 = SMLAD(ker_1, vec, acc_0); in arm_nn_vec_mat_mult_t_s16()
305 acc_64_0 += acc_0; in arm_nn_vec_mat_mult_t_s16()
|
| D | arm_nn_vec_mat_mul_result_acc_s16.c | 224 int32_t acc_0 = 0; in arm_nn_vec_mat_mul_result_acc_s16() local
244 acc_0 = SMLAD(ker_0, vec_part_0, acc_0); in arm_nn_vec_mat_mul_result_acc_s16()
245 acc_0 = SMLAD(ker_1, vec_part_1, acc_0); in arm_nn_vec_mat_mul_result_acc_s16()
253 acc_64_0 += acc_0; in arm_nn_vec_mat_mul_result_acc_s16()
286 int32_t acc_0 = 0; in arm_nn_vec_mat_mul_result_acc_s16() local
298 acc_0 = SMLAD(ker_0, vec, acc_0); in arm_nn_vec_mat_mul_result_acc_s16()
301 acc_0 = SMLAD(ker_1, vec, acc_0); in arm_nn_vec_mat_mul_result_acc_s16()
304 acc_64_0 += acc_0; in arm_nn_vec_mat_mul_result_acc_s16()
|
| /cmsis-nn-latest/Source/ConvolutionFunctions/ |
| D | arm_nn_mat_mult_s8.c | 67 int32_t acc_0 = 0; in arm_nn_mat_mult_s8() local
98 acc_0 = vmladavaq_p_s16(acc_0, r0, c0, p); in arm_nn_mat_mult_s8()
104 int32x4_t res = {acc_0, acc_1, acc_2, acc_3}; in arm_nn_mat_mult_s8()
130 int32_t acc_0 = 0; in arm_nn_mat_mult_s8() local
145 acc_0 = vmladavaq_p_s16(acc_0, r0, c0, p); in arm_nn_mat_mult_s8()
150 acc_0 += bias[i_out_ch]; in arm_nn_mat_mult_s8()
152 acc_0 = arm_nn_requantize(acc_0, output_mult[i_out_ch], output_shift[i_out_ch]); in arm_nn_mat_mult_s8()
153 acc_0 += out_offset; in arm_nn_mat_mult_s8()
154 acc_0 = MAX(acc_0, activation_min); in arm_nn_mat_mult_s8()
155 acc_0 = MIN(acc_0, activation_max); in arm_nn_mat_mult_s8()
[all …]
|
| D | arm_depthwise_conv_s4.c | 94 int32_t acc_0 = 0; in depthwise_conv_s4_generic() local
130 acc_0 = bias[i_input_ch]; in depthwise_conv_s4_generic()
150 acc_0 += (input[idx_0] + input_offset) * ker_val0; in depthwise_conv_s4_generic()
161 … acc_0 = arm_nn_requantize(acc_0, output_mult[i_input_ch], output_shift[i_input_ch]); in depthwise_conv_s4_generic()
162 acc_0 += output_offset; in depthwise_conv_s4_generic()
163 acc_0 = MAX(acc_0, output_activation_min); in depthwise_conv_s4_generic()
164 acc_0 = MIN(acc_0, output_activation_max); in depthwise_conv_s4_generic()
165 output[i_out++] = acc_0; in depthwise_conv_s4_generic()
187 int32_t acc_0 = 0; in depthwise_conv_s4_generic() local
222 acc_0 = bias[idx_out_ch]; in depthwise_conv_s4_generic()
[all …]
|
| D | arm_depthwise_conv_s8.c | 199 int32_t acc_0 = 0; in depthwise_conv_s8_generic() local
234 acc_0 = bias[idx_out_ch]; in depthwise_conv_s8_generic()
246 acc_0 += (input[idx_0] + input_offset) * kernel[ker_idx_0]; in depthwise_conv_s8_generic()
251 … acc_0 = arm_nn_requantize(acc_0, output_mult[idx_out_ch], output_shift[idx_out_ch]); in depthwise_conv_s8_generic()
252 acc_0 += output_offset; in depthwise_conv_s8_generic()
253 acc_0 = MAX(acc_0, output_activation_min); in depthwise_conv_s8_generic()
254 acc_0 = MIN(acc_0, output_activation_max); in depthwise_conv_s8_generic()
256 output[i_out++] = acc_0; in depthwise_conv_s8_generic()
|
| D | arm_depthwise_conv_s16.c | 173 int64_t acc_0 = 0; in depthwise_conv_s16_generic_s16() local
208 acc_0 = bias[idx_out_ch]; in depthwise_conv_s16_generic_s16()
220 acc_0 += input[idx_0] * kernel[ker_idx_0]; in depthwise_conv_s16_generic_s16()
225 … int32_t result = arm_nn_requantize_s64(acc_0, reduced_multiplier, output_shift[idx_out_ch]); in depthwise_conv_s16_generic_s16()
|
