就在本周六，成都站 Gitee Talk | 模力方舟 AI 应用开发沙龙抢占最后席位！聚焦 AI 开发全链路，结识技术同频人~

加入 Gitee

与超过 1200万开发者一起发现、参与优秀开源项目，私有仓库也完全免费：）

文件

克隆/下载

esp-dsp-benchmarks.rst 5.24 KB

Espressif DSP Library Benchmarks
================================

The table bellow contains benchmarks of functions provided by ESP-DSP library. The values are CPU cycle counts taken to execute each of the functions. Values in "ESP32" column are for the optimized (assembly) implementation, values in "ANSI C" column are for the non-optimized implementation.

+----------------------------------------------------------+----------+----------+
| Function name and arguments                              | CPU cycles          |
+----------------------------------------------------------+----------+----------+
|                                                          | ESP32    | ANSI C   |
+==========================================================+==========+==========+
|                                                          |          |          |
+----------------------------------------------------------+----------+----------+
| **Dot Product**                                          |          |          |
+----------------------------------------------------------+----------+----------+
| dsps_dotprod_f32 for N=256 points                        |     1057 |     2597 |
+----------------------------------------------------------+----------+----------+
| dsps_dotprode_f32 for N=256 points, with step 1          |     1318 |     2601 |
+----------------------------------------------------------+----------+----------+
| dsps_dotprod_s16 for N=256 points                        |      448 |     5185 |
+----------------------------------------------------------+----------+----------+
|                                                          |          |          |
+----------------------------------------------------------+----------+----------+
| **FIR Filters**                                          |          |          |
+----------------------------------------------------------+----------+----------+
| dsps_fir_f32 1024 input samples and 256 coefficients     |  1338418 |  3583556 |
+----------------------------------------------------------+----------+----------+
| dsps_fird_f32 1024 samples, 256 coeffs and decimation 4  |    37582 |    82535 |
+----------------------------------------------------------+----------+----------+
|                                                          |          |          |
+----------------------------------------------------------+----------+----------+
| **FFTs**                                                 |          |          |
+----------------------------------------------------------+----------+----------+
| dsps_fft2r_fc32 for  64 complex points                   |     5451 |     8187 |
+----------------------------------------------------------+----------+----------+
| dsps_fft2r_fc32 for 128 complex points                   |    12400 |    18756 |
+----------------------------------------------------------+----------+----------+
| dsps_fft2r_fc32 for 256 complex points                   |    27829 |    42381 |
+----------------------------------------------------------+----------+----------+
| dsps_fft2r_fc32 for 512 complex points                   |    61755 |    94616 |
+----------------------------------------------------------+----------+----------+
| dsps_fft2r_fc32 for 1024 complex points                  |   135745 |   209058 |
+----------------------------------------------------------+----------+----------+
|                                                          |          |          |
+----------------------------------------------------------+----------+----------+
| **IIR Filters**                                          |          |          |
+----------------------------------------------------------+----------+----------+
| dsps_biquad_f32 - biquad filter for 1024 input samples   |    17451 |    31778 |
+----------------------------------------------------------+----------+----------+
|                                                          |          |          |
+----------------------------------------------------------+----------+----------+
| **Matrix Multiplication**                                |          |          |
+----------------------------------------------------------+----------+----------+
| dspm_mult_f32 - C[16,16] = A[16,16]*B[16,16];            |    24669 |    59690 |
+----------------------------------------------------------+----------+----------+
| dspm_mult_s16 - C[16,16] = A[16,16]*B[16,16];            |    24964 |   114150 |
+----------------------------------------------------------+----------+----------+
| dspm_mult_3x3x1_f32 - C[3,1] = A[3,3]*B[3,1];            |       80 |      242 |
+----------------------------------------------------------+----------+----------+
| dspm_mult_3x3x3_f32 - C[3,3] = A[3,3]*B[3,3];            |      212 |      541 |
+----------------------------------------------------------+----------+----------+
| dspm_mult_4x4x1_f32 - C[4,1] = A[4,4]*B[4,1];            |      112 |      362 |
+----------------------------------------------------------+----------+----------+
| dspm_mult_4x4x4_f32 - C[4,4] = A[4,4]*B[4,4];            |      404 |     1130 |
+----------------------------------------------------------+----------+----------+

The benchmark test could be reproduced by executing test cases found in :repo_file:`test/test_dsp.c`.

一键复制编辑原始数据按行查看历史

提交于 2019-02-11 14:23 +08:00 . docs: remove duplicated benchmark line

Espressif DSP Library Benchmarks

The table bellow contains benchmarks of functions provided by ESP-DSP library. The values are CPU cycle counts taken to execute each of the functions. Values in "ESP32" column are for the optimized (assembly) implementation, values in "ANSI C" column are for the non-optimized implementation.

Function name and arguments	CPU cycles
	ESP32	ANSI C

Dot Product
dsps_dotprod_f32 for N=256 points	1057	2597
dsps_dotprode_f32 for N=256 points, with step 1	1318	2601
dsps_dotprod_s16 for N=256 points	448	5185

FIR Filters
dsps_fir_f32 1024 input samples and 256 coefficients	1338418	3583556
dsps_fird_f32 1024 samples, 256 coeffs and decimation 4	37582	82535

FFTs
dsps_fft2r_fc32 for 64 complex points	5451	8187
dsps_fft2r_fc32 for 128 complex points	12400	18756
dsps_fft2r_fc32 for 256 complex points	27829	42381
dsps_fft2r_fc32 for 512 complex points	61755	94616
dsps_fft2r_fc32 for 1024 complex points	135745	209058

IIR Filters
dsps_biquad_f32 - biquad filter for 1024 input samples	17451	31778

Matrix Multiplication
dspm_mult_f32 - C[16,16] = A[16,16]*B[16,16];	24669	59690
dspm_mult_s16 - C[16,16] = A[16,16]*B[16,16];	24964	114150
dspm_mult_3x3x1_f32 - C[3,1] = A[3,3]*B[3,1];	80	242
dspm_mult_3x3x3_f32 - C[3,3] = A[3,3]*B[3,3];	212	541
dspm_mult_4x4x1_f32 - C[4,1] = A[4,4]*B[4,1];	112	362
dspm_mult_4x4x4_f32 - C[4,4] = A[4,4]*B[4,4];	404	1130

The benchmark test could be reproduced by executing test cases found in :repo_file:`test/test_dsp.c`.

Loading...

马建仓 AI 助手

尝试更多

代码解读

代码找茬

代码优化

1

https://gitee.com/espressif-systems/esp-dsp.git

git@gitee.com:espressif-systems/esp-dsp.git

espressif-systems

esp-dsp

esp-dsp

master