Re: [hatari-devel] Re: Profiler - long history

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On maanantai 18 toukokuu 2015, Nicolas Pomarède wrote:
> when laurent imported the old winuae cpu core into Hatari some years
> ago, he added his own code to compute cycles based on whether
> instruction was in cache or not :
> 
> CpuInstruction.iCacheMisses++
> 
> but this is not part of recent winuae by Toni.
> 
> I could add something similar,

Could you add that before next release, so that we don't regress?

(Don't care what's the variable name or whether it's some struct,
main thing is just to have the information available in the profiler.)


> but this would need to be improved anyway, because this code
> only counted instruction cache misses, but not data cache misses.

Yes, that data cache misses would be good to have also.
This will require some changes both to Hatari and profiler
data post-processing tools.

Btw. I thought old WinUAE core supported only icache, and only
the new one supports data cache?  If I remember correctly, Douglas
had mentioned that the old icache misses weren't accurate, real
machine cached few instructions more (or less) than what Hatari
claimed.


	- Eero