2006/3/27, Elias Pschernig <elias@xxxxxxxxxxxxxxxxxxxxx>:
On Sun, 2006-03-26 at 23:36 +0200, Milan Mimica wrote:
> Elias Pschernig wrote:
> >
> > And I also just noticed again in the test program that the non-asm
> > version is between 10% and 100% (!) faster in the draw_sprite tests for
> > my old pentium 4 here, so I propose to change the default ./configure
> > result to always disable asm, and give Allegro some speed up under unix
> > that way, the X11 port can need it :)
>
> It would break ABI compatibility.
> There are some functions that benefit from ASM code. I don't know
> exactly which one (some memory bitmaps rotation/stretching I think), I
> just noticed it while profiling my game. This things really need to be
> benchmarked.
>
Hm, it does break ABI? Oh well, it was badly designed then I guess.. a
good way would have been to hide the asm/non-asm in the driver. But
yeah, I faintly can remember something like this coming up before.
About benchmarking, I think there was a lot of benchmarking on
allegro.cc about a year ago, and non-asm clearly won.
I don't know if you can call running allegro test program on two computers under Linux "a lot of benchmarking" but that is all I can remember about that. Those benchmark stats are here:
P4 Northwood
http://www.allegro.cc/forums/thread/477775#targetP1 MMX
Unfortunately there are no benches of some functions like rotating and some others.
If I could find some time I could probably repeat those tests on a64 win/lin, P3 lin, athlon xp win/lin and P4 Prescott lin.