Re: [hatari-devel] New option --disable-video and --benchmark

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 24/02/2017 à 19:50, Eero Tamminen a écrit :
> Hi,
>
> On 02/24/2017 08:02 PM, Thomas Huth wrote:
> > On 23.02.2017 23:03, Nicolas Pomarède wrote:
> > > I thought about running Hatari without any window at all, but then more
> > > code would be needed as it's SDL that handles keyboard shortcut (for
> > > example handle ctrl+c to exit Hatari).
> >
> > FYI, you can also run Hatari without window by setting the
> > SDL_VIDEODRIVER variable to "dummy".
>
> Sounds good, one-liner implementation for the --benchmark option:
>     setenv("SDL_VIDEODRIVER", "dummy", 1);
> :-)

I made some tests, it seems the dummy value doesn't disable SDL calls 
that much : you don't get the video on screen, but you don't get that 
much speed up.

example :

default behaviour :
hatari --memstate a1.sav --benchmark --disable-video 0 --run-vbls 2000
SPEED: 200.6 VBL/s (2000/10.0s)

disable SDL calls (mainly SDL_Copy / SDL_Render) :
hatari --memstate a1.sav --benchmark --disable-video 1 --run-vbls 2000
SPEED: 270 to 279.7 VBL/s (2000/7.2s)

use dummy driver :
SDL_VIDEODRIVER=dummy hatari --memstate a1.sav --benchmark 
--disable-video 0 --run-vbls 2000
SPEED: 210.5 to 219.5 VBL/s

use dummy driver + disable SDL calls :
SDL_VIDEODRIVER=dummy hatari --memstate a1.sav --benchmark 
--disable-video 1 --run-vbls 2000
SPEED: 272.5 to 285 VBL/s


So, dummy driver might be good if you want to run an SDL program on a 
system without windows capabilities, but for benchmarking it doesn't 
disable the cost of SDL calls like Copy and Render, the gain is really 
low (5-10% or less) compared to 35-40% by really disabling SDL calls.

Nicolas