[Jihai APM32F407 Tiny Board] 3. Coremark running test

dql2016

[Jihai APM32F407 Tiny Board] 3. Coremark running test [Copy link]

 

Based on the previous serial port printing, the coremark running program is transplanted.

There are many transplantation tutorials on the Internet, so I won’t go into details here. The main references are:

如何将coremark程序移植到STM32上.pdf (1.12 MB, downloads: 3)

2023-5-25 22:12 上传

点击文件名下载附件

armcc v5 compiler results:

The SDK of Jiuhai APM32F407 supports the ARMCLANG compiler. I have heard that ARMCLANG is much more powerful than ARMCC.

armclang v6 compiler results:

It can be seen that the running points differ by almost 1 times... This gap is quite large

What does 444 mean? Compared with STM32 (STM32 should be tested by IAR, and the running score is close to ARMCLANG)

Test code:

GPIO_Toggle.rar (86.39 KB, downloads: 5)

2023-5-25 22:11 上传

点击文件名下载附件

nmg

It seems that the running score of the microcontroller alone does not mean anything, because the running score may be different in different environments and compilers, and the difference may be large.

dql2016

nmg posted on 2023-5-26 10:01 It seems that taking the running score of the microcontroller alone does not mean anything, because the running environment and compiler used may be different, and the gap will be very large

This explains the problem. No matter how much you optimize the M3 compiler, it will not beat the M7.

nmg

dql2016 posted on 2023-5-26 21:37 This explains the problem. No matter how much you optimize with the M3 compiler, you can't beat the M7

Even if it's a dimensionality reduction attack

For example, for the same level of core, if the manufacturer does not specify the environment when promoting the running score, it is not easy to use the official score as a comparison.

迈尔风随

Does the armcc V5 compiler have any code optimization enabled? Looking at the code size of the two compilers, V5 has 4K more code than clang

dql2016

Posted by Maier Fengsui on 2023-5-29 16:20 Does the armcc V5 compiler have any code optimization enabled? Looking at the size of the compiled code, the V5 code size is 4K more than clang

No optimization

dql2016

nmg posted on 2023-5-29 08:33 Forget about dimensionality reduction. For example, for the same level of core, if the manufacturer does not specify the environment when promoting the running score, it is not easy to directly use the official score...

The manufacturer is definitely the best, ST usually runs with IAR