Benchmark in barebones

[riema...@gmail.com]

Hi
I'm a university student, using and studying coremark now. While I study this benchmark tool, I got two questions.
First I want to know the differences between performance, validation, profile generation run.
Only thing I found was the differences initial seed value which affects data structure for algorithm, and i can't understand the meaning of that.

And second, when i run this tools in simple linux platform by using command 'make', i could get two results the performance run and validation run as a default. But when i run in barebones, things i could only get was performance result even if i change the command line. I want to know how to get both results in barebone running.

Thanks D.S

[Shay Gal-On]

Hello D.S,

1. The difference between the 3 runs are as follows:
- Different seeds to change the input data being used.
- Different data size.
- Please check the readme for explanation of when to use the different seeds.
2. When you run barebones, you need to provide a way to give the benchmark inputs that are not known at compile time. There are various ways of doing that as mentioned in the documentation. Easily on bare bones is usually to use volatile input, and build a different binary for each seed that you wish to test.

Thanks,
- Shay Gal-On
Dir. Sw. Eng., EEMBC Technology Center
http://www.eembc.com/contact

> --
> You received this message because you are subscribed to the Google Groups
> "EEMBC CoreMark Public Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to coremark+u...@googlegroups.com.
> To post to this group, send email to core...@googlegroups.com.
> Visit this group at http://groups.google.com/group/coremark.
> For more options, visit https://groups.google.com/groups/opt_out.

[marek.b...@gmail.com]

Anyone,

Can you explain me what the "profile run" is?
I know that "performance run" is what the result of comprises the CoreMark score.
"validation run" should indicate whether validation passed or not.

I know their different in seeds and the total size but please give me the meaning of when the "profile run" could be of interest to me.

You may consider it a lazy way as I could probably infer that from the code but having your explanation could save me a time.

The primary reason the question arose from was that ARM for Cortex-M series recommends one runs 10x profile runs for the value to converge. I'm confused there...

Thanks,
Marek