Re: [hatari-devel] Feature request/idea: cycle counting

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On 01/30/2018 04:03 AM, Miro Kropáček wrote:
> I've been having this idea to create my own cycle counter and just
> yesterday I've realised that Hatari can do that for me with nearly zero
> effort. :)
>
> The idea is that you have a m68k code in memory, disassemble it and you see
> not only each instruction's size but also cycles it eats. And since it
> would be in Hatari, you could immediately calculate whether it's a cache
> hit/miss, influence of the instruction head/tail, literally anything needed.
>
> Since Hatari already tries to calculate Falcon's bus activity and CPU
> cycles, how hard this would be? Virtually just one more column with the
> cycles number and some indication about instr/data cache state.

Hatari already provides that info:
https://hg.tuxfamily.org/mercurialroot/hatari/hatari/raw-file/tip/doc/manual.html#Investigating_the_profile_data


For Bad Mood, I had done Hatari debugger scripts which automatically
find out the frames that are heaviest in regards to CPU instruction
and DSP instruction[1] count wise, and automatically profile them &
provide call-graphs for those.

Those scripts could have done that also based on cycle counts,
but back then Hatari didn't yet have good enough cache emulation
so I trusted instruction counts more.

(Scripts for that are included to Bad Mood sources.  All one needed
to do, is make a recording of Doom I / II game play and specify that
on the Bad Mood command line, to automatically get separate CPU & DSP
callgraphs of the heaviest frames during game startup & actual game
play.)


	- Eero