Thanks a lot for your very complete answer !
I doubt that is possible. Those open source projects are very slow to support even new Mac operating system versions. This is a new platform altogether. Much of that parallel scientific software stack may not exist yet.
I have no problem with OpenMP and with the performance measurement algorithm (written by my teacher).
Why? Clock speed is not necessarily the most important factor.
To be honest, I don't know how my teacher wrote his program. But it works and it prints some measurements in GFLOP/S for each sort of operations (for example : additions, copy of vectors...). My work is just to write different types of algorithms and operation, and measure their performance with my teacher's program. I just have one thing to add to my teacher's program : my processor's nominal frequency.
How do you know that if you don't know what those speeds are yet?
I know their maximal frequencies, but you're right, I don't know their working frequency (of course firestorms cores aren't working all the time at their max. frequency).
A better question is to ask what you are trying to do, in an overall sense. What's the "big picture"?
Sorry for being very confused... As I just resumed, I need a fixed frequency when my program is executing, otherwise, the measurements will be wrong. But I didn't find anything about my cores frequency, so an idea was to force the execution on an Icestorm core and take its max frequency (2.06 GHz according to Wikipedia) for our measurements. By this way, we could be at least sure that the threads will be executed on same-frequencies cores.
I think you could probably get the clock speeds of the cores using IOKit. When I check using "ioreg -l", on an M1, I see multiple "frequency" fields. Which one is the one you need? And is it fixed? There is a "fixed-frequency" and a "clock-frequency", and 3 other "frequencies". There is no documentation for what any of those values mean or how they might change.
I just tried this one. I didn't know it so thank you ! I will show this to my teacher. But I think that it is a good way to solve our problem. You're right, there is a "general" clock-frequency, one fixed-frequency and 8 clock-frequency (I suppose that there is one frequency per core). And all have the same value : <00366e01>. After few searches, I found that this value was in kHz, and presented as little-endian ! So this value is in fact : 0x016e3600 kHz = 2,4 GHz.
This is strange, it is more than the max Icestorm cores frequency, and all cores have this frequency. Maybe it is an "average" frequency of all cores ? And maybe I can use this value for my problem... I will discuss about that with my teacher.
I'm afraid that Apple platforms are a better area of study for social scientists than computer scientists.
Haha I think that's the general opinion. But I like Apple and I like challenges...
Anyways, thank you a lot, maybe the ioreg output is what my teacher is searching for. I mark this problem as "resolved" as you answered my question :
As far as I know, it is impossible to "force" the execution on a single specific core.
Have a nice day !
