160 …for (int i_loop_cnt = 0, offset = 0; i_loop_cnt < loop_count; num_ch_to_process -= 4, offset += 4,…  in arm_depthwise_conv_fast_s16()  local
162             const int8_t *row_0 = kernel + offset;  in arm_depthwise_conv_fast_s16()
163             const int16_t *col_0 = lhs_buffer + (kernel_size * input_ch * i_buf) + offset;  in arm_depthwise_conv_fast_s16()
185                 in_requantize_0 += bias[offset];  in arm_depthwise_conv_fast_s16()
186                 in_requantize_1 += bias[offset + 1];  in arm_depthwise_conv_fast_s16()
187                 in_requantize_2 += bias[offset + 2];  in arm_depthwise_conv_fast_s16()
188                 in_requantize_3 += bias[offset + 3];  in arm_depthwise_conv_fast_s16()
191             int32_t reduced_multiplier_0 = REDUCE_MULTIPLIER(output_mult[offset]);  in arm_depthwise_conv_fast_s16()
192             int32_t reduced_multiplier_1 = REDUCE_MULTIPLIER(output_mult[offset + 1]);  in arm_depthwise_conv_fast_s16()
193             int32_t reduced_multiplier_2 = REDUCE_MULTIPLIER(output_mult[offset + 2]);  in arm_depthwise_conv_fast_s16()
194             int32_t reduced_multiplier_3 = REDUCE_MULTIPLIER(output_mult[offset + 3]);  in arm_depthwise_conv_fast_s16()
196 …     out_0[0] = arm_nn_requantize_s64(in_requantize_0, reduced_multiplier_0, output_shift[offset]);  in arm_depthwise_conv_fast_s16()
197 … out_0[1] = arm_nn_requantize_s64(in_requantize_1, reduced_multiplier_1, output_shift[offset + 1]);  in arm_depthwise_conv_fast_s16()
198 … out_0[2] = arm_nn_requantize_s64(in_requantize_2, reduced_multiplier_2, output_shift[offset + 2]);  in arm_depthwise_conv_fast_s16()
199 … out_0[3] = arm_nn_requantize_s64(in_requantize_3, reduced_multiplier_3, output_shift[offset + 3]);  in arm_depthwise_conv_fast_s16()