Re: [MiNT] gcc-performace, was: Samba working

[Eero Tamminen <oak@helsinkinet.fi>]

Hi,

On keskiviikko 11 joulukuu 2013, Alan Hourihane wrote:
> On 12/11/13 21:23, Helmut Karlowski wrote:
> > Someone with a "fast" real atari should compare compile-speeds to rule
> > out any jit-influence. My estimation is that the atari in aranym has
> > about 2GHz, at least 1GHz (I think have 2x2.5), some things run even
> > faster on it than on native cygwin, except gcc. So if the firebee runs
> > at 266 MHz compilation of that 100 kB C-source with -O3 should not
> > need more than 20s or so. All very inexact of course, but before I'd
> > start looking for a reason in mintlib or somewhere else I'd want to be
> > sure if it's worth the time. Maybe Mirkos  gcc is faster, haven't
> > tried that. All in all gcc on atari seems not the best thing to do for
> > medium projects, unfortunately, at least on aranym.
> > 
> > That would be a nice task for Eero's profiling tool ;-)
> 
> Is there documentation on how to run this someplace ?

Sure, here:
http://hg.tuxfamily.org/mercurialroot/hatari/hatari/raw-
file/tip/doc/manual.html#Profiling

I've never tried it under MiNT though, because by default MiNT overwrites
the GEMDOS vector set up by Hatari (to call Hatari's GEMDOS emulation).
This makes MiNT usage fairly awkward because you need to deal with HD
images as Hatari doesn't support file system NatFeats like Aranym does.

Maybe using TOSFS (Helmut has patches for that), would help with that.

While MiNT doesn't affect the profiler itself, to load symbols to right
address (program TEXT section start) and to set breakpoints at convenient
places (e.g. Pexec() and Pterm()) in the debugger, some GEMDOS support
would be needed.


	- Eero