Home
last modified time | relevance | path

Searched refs:output_mult (Results 1 – 15 of 15) sorted by relevance

/cmsis-nn-latest/Source/NNSupportFunctions/
Darm_nn_mat_mul_core_1x_s8.c59 int32_t *output_mult = quant_params->multiplier; in arm_nn_mat_mul_core_1x_s8() local
110 res = arm_requantize_mve_32x4(res, vldrwq_s32(output_mult), vldrwq_s32(output_shift)); in arm_nn_mat_mul_core_1x_s8()
111 output_mult += 4; in arm_nn_mat_mul_core_1x_s8()
129 acc_n0 = arm_nn_requantize(acc_n0, output_mult[i], output_shift[i]); in arm_nn_mat_mul_core_1x_s8()
Darm_nn_mat_mul_core_1x_s4.c59 int32_t *output_mult = quant_params->multiplier; in arm_nn_mat_mul_core_1x_s4() local
105 res = arm_requantize_mve_32x4(res, vldrwq_s32(output_mult), vldrwq_s32(output_shift)); in arm_nn_mat_mul_core_1x_s4()
106 output_mult += 4; in arm_nn_mat_mul_core_1x_s4()
125 acc_n0 = arm_nn_requantize(acc_n0, output_mult[i], output_shift[i]); in arm_nn_mat_mul_core_1x_s4()
/cmsis-nn-latest/Source/ConvolutionFunctions/
Darm_depthwise_conv_fast_s16.c94 const int32_t *output_mult = quant_params->multiplier; in arm_depthwise_conv_fast_s16() local
140 output_mult, 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()
223 const int32_t *const out_mult_start_pos = output_mult; in arm_depthwise_conv_fast_s16()
282 output_mult = out_mult_start_pos; in arm_depthwise_conv_fast_s16()
292 int32_t output_mult_1 = REDUCE_MULTIPLIER(output_mult[0]); in arm_depthwise_conv_fast_s16()
293 int32_t output_mult_2 = REDUCE_MULTIPLIER(output_mult[1]); in arm_depthwise_conv_fast_s16()
[all …]
Darm_depthwise_conv_s8.c63 const int32_t *output_mult, in depthwise_conv_s8_mult_4() argument
72 const int32_t *mult_base = output_mult; in depthwise_conv_s8_mult_4()
81 output_mult = mult_base; in depthwise_conv_s8_mult_4()
117 … res = arm_requantize_mve_32x4(res, vldrwq_s32(output_mult), vldrwq_s32(output_shift)); in depthwise_conv_s8_mult_4()
118 output_mult += 4; in depthwise_conv_s8_mult_4()
127 out_buff[0] = arm_nn_requantize(out_buff[0], *output_mult++, *output_shift++); in depthwise_conv_s8_mult_4()
128 out_buff[1] = arm_nn_requantize(out_buff[1], *output_mult++, *output_shift++); in depthwise_conv_s8_mult_4()
129 out_buff[2] = arm_nn_requantize(out_buff[2], *output_mult++, *output_shift++); in depthwise_conv_s8_mult_4()
130 out_buff[3] = arm_nn_requantize(out_buff[3], *output_mult++, *output_shift++); in depthwise_conv_s8_mult_4()
171 const int32_t *output_mult, in depthwise_conv_s8_generic() argument
[all …]
Darm_convolve_s16.c90 int32_t *output_mult = quant_params->multiplier; in arm_convolve_s16() local
143 output_mult, in arm_convolve_s16()
163 output_mult, in arm_convolve_s16()
191 output_mult, in arm_convolve_s16()
257 sum = arm_nn_requantize(sum, output_mult[i], output_shift[i]); in arm_convolve_s16()
268 int32_t reduced_multiplier = REDUCE_MULTIPLIER(output_mult[i]); in arm_convolve_s16()
Darm_depthwise_conv_s4_opt.c88 const int32_t *output_mult = quant_params->multiplier; in arm_depthwise_conv_s4_opt() local
147 output_mult + block_offset, in arm_depthwise_conv_s4_opt()
233 const int32x4_t mult = vldrwq_s32(&output_mult[offset]); in arm_depthwise_conv_s4_opt()
255 const int32_t *const out_mult_start_pos = output_mult; in arm_depthwise_conv_s4_opt()
314 output_mult = out_mult_start_pos; in arm_depthwise_conv_s4_opt()
436 sum = arm_nn_requantize(sum, *output_mult++, *output_shift++); in arm_depthwise_conv_s4_opt()
442 sum_2 = arm_nn_requantize(sum_2, *output_mult++, *output_shift++); in arm_depthwise_conv_s4_opt()
447 sum_3 = arm_nn_requantize(sum_3, *output_mult++, *output_shift++); in arm_depthwise_conv_s4_opt()
453 sum_4 = arm_nn_requantize(sum_4, *output_mult++, *output_shift++); in arm_depthwise_conv_s4_opt()
504 sum = arm_nn_requantize(sum, *output_mult++, *output_shift++); in arm_depthwise_conv_s4_opt()
[all …]
Darm_depthwise_conv_s8_opt.c87 const int32_t *output_mult = quant_params->multiplier; in arm_depthwise_conv_s8_opt() local
147 output_mult + block_offset, in arm_depthwise_conv_s8_opt()
191 const int32x4_t mult = vldrwq_s32(&output_mult[offset]); in arm_depthwise_conv_s8_opt()
215 const int32_t *const out_mult_start_pos = output_mult; in arm_depthwise_conv_s8_opt()
271 output_mult = out_mult_start_pos; in arm_depthwise_conv_s8_opt()
348 sum = arm_nn_requantize(sum, *output_mult++, *output_shift++); in arm_depthwise_conv_s8_opt()
354 sum_2 = arm_nn_requantize(sum_2, *output_mult++, *output_shift++); in arm_depthwise_conv_s8_opt()
359 sum_3 = arm_nn_requantize(sum_3, *output_mult++, *output_shift++); in arm_depthwise_conv_s8_opt()
365 sum_4 = arm_nn_requantize(sum_4, *output_mult++, *output_shift++); in arm_depthwise_conv_s8_opt()
391 sum = arm_nn_requantize(sum, *output_mult++, *output_shift++); in arm_depthwise_conv_s8_opt()
Darm_nn_mat_mult_s8.c44 const int32_t *output_mult, in arm_nn_mat_mult_s8() argument
109 res = arm_requantize_mve(res, output_mult[i_out_ch], output_shift[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()
169 (void)output_mult; in arm_nn_mat_mult_s8()
Darm_depthwise_conv_s16.c59 const int32_t *output_mult, in depthwise_conv_s16_mult_4_s16() argument
74 int32_t out_buff32[4] = {REDUCE_MULTIPLIER(output_mult[out_ch + 0 + mult_tile]), in depthwise_conv_s16_mult_4_s16()
75 REDUCE_MULTIPLIER(output_mult[out_ch + 1 + mult_tile]), in depthwise_conv_s16_mult_4_s16()
76 REDUCE_MULTIPLIER(output_mult[out_ch + 2 + mult_tile]), in depthwise_conv_s16_mult_4_s16()
77 … REDUCE_MULTIPLIER(output_mult[out_ch + 3 + mult_tile])}; in depthwise_conv_s16_mult_4_s16()
149 const int32_t *output_mult, in depthwise_conv_s16_generic_s16() argument
172 … const int32_t reduced_multiplier = REDUCE_MULTIPLIER(output_mult[idx_out_ch]); in depthwise_conv_s16_generic_s16()
Darm_convolve_s4.c92 int32_t *output_mult = quant_params->multiplier; in arm_convolve_s4() local
140 output_mult, in arm_convolve_s4()
170 output_mult, in arm_convolve_s4()
226 output_mult, in arm_convolve_s4()
316 sum = arm_nn_requantize(sum, output_mult[i], output_shift[i]); in arm_convolve_s4()
Darm_depthwise_conv_3x3_s8.c76 const int32_t *output_mult = quant_params->multiplier; in arm_depthwise_conv_3x3_s8() local
217 … out_buff0 = arm_nn_requantize(out_buff0, output_mult[in_ch + 0], output_shift[in_ch + 0]); in arm_depthwise_conv_3x3_s8()
218 … out_buff1 = arm_nn_requantize(out_buff1, output_mult[in_ch + 1], output_shift[in_ch + 1]); in arm_depthwise_conv_3x3_s8()
219 … out_buff2 = arm_nn_requantize(out_buff2, output_mult[in_ch + 2], output_shift[in_ch + 2]); in arm_depthwise_conv_3x3_s8()
220 … out_buff3 = arm_nn_requantize(out_buff3, output_mult[in_ch + 3], output_shift[in_ch + 3]); in arm_depthwise_conv_3x3_s8()
268 out_buff = arm_nn_requantize(out_buff, output_mult[in_ch], output_shift[in_ch]); in arm_depthwise_conv_3x3_s8()
Darm_depthwise_conv_s4.c60 const int32_t *output_mult, in depthwise_conv_s4_generic() argument
161 … acc_0 = arm_nn_requantize(acc_0, output_mult[i_input_ch], output_shift[i_input_ch]); in depthwise_conv_s4_generic()
167 … acc_1 = arm_nn_requantize(acc_1, output_mult[i_input_ch + 1], output_shift[i_input_ch + 1]); in depthwise_conv_s4_generic()
259 … acc_0 = arm_nn_requantize(acc_0, output_mult[idx_out_ch], output_shift[idx_out_ch]); in depthwise_conv_s4_generic()
346 … acc_0 = arm_nn_requantize(acc_0, output_mult[idx_out_ch], output_shift[idx_out_ch]); in depthwise_conv_s4_generic()
353 … arm_nn_requantize(acc_1, output_mult[idx_out_ch + 1], output_shift[idx_out_ch + 1]); in depthwise_conv_s4_generic()
468 … acc_0 = arm_nn_requantize(acc_0, output_mult[idx_out_ch], output_shift[idx_out_ch]); in depthwise_conv_s4_generic()
Darm_convolve_s8.c95 int32_t *output_mult = quant_params->multiplier; in arm_convolve_s8() local
123 const int32_t *output_mult_ptr = &output_mult[0]; in arm_convolve_s8()
/cmsis-nn-latest/Tests/UnitTest/
Dadd_mul_settings.py124 (self.output_mult, self.output_shift) = self.quantize_scale(actual_output_scale)
153 f.write("#define {}_OUTPUT_MULT {}\n".format(prefix, self.output_mult))
/cmsis-nn-latest/Include/
Darm_nnsupportfunctions.h272 const int32_t *output_mult,