[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#697041: atlas: FTBFS: tune/sysinfo/GetSysSum.c:47: GetL1CacheSize: Assertion `system(ln) == 0' failed



Hi,

On sunnuntai 28 heinäkuu 2013, Geert Uytterhoeven wrote:
> On Sun, Jul 28, 2013 at 10:35 PM, Thorsten Glaser <tg@mirbsd.de> wrote:
> >>> Calculated L1 cache size = 8kb; Correct=0
> >>
> >>Hmm, 68060...
> >>
> > I think ARAnyM emulates an 040 though. It is
> > trying for 4 KiB right now, after the 8 KiB
> > attempt seems to have failed…
> 
> So you hope ARAnyM is sufficiently accurate in emulating the cache to
> make these measurements valid?

AFAIK Aranym doesn't emulate CPU cache.  It doesn't aim to be cycle
accurate, like e.g. Hatari does [1], as that can slow down the emulation
a LOT.


	- Eero

[1] Hatari emulator emulates 68000 - 68040 CPU caches to some extent, but
for now its cache emulation can be considered accurate enough only for 68000
(instruction pairing).

Hatari's MMU emulation isn't yet good enough to run Linux or BSD either.
After somebody re-ports latest WinUAE CPU core to Hatari, it will though,
and then it could be used to do so real Debian port debugging &
optimization.

(Currently Hatari can accurately profile CPU instruction cache misses
and DSP cycle counts, that's used in optimizing Doom I/II for Atari
Falcon 030 + DSP.)


Reply to: