[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
Elias Pschernig wrote:
> I"d really like to see an analysis of why the libc implementation is
> faster than the MMX asm we use currently.
I can't say, but here is libc i386 implementation:
http://www.globalnet.hr/~mmimica/memset.c
I run the test also on XP on the same machine. The improvement is even
more notable:
15&16-bit: 1.0705/1.672=0.6429
24-bit: 2.086/32.398=0.87
32-bit: 3.07/3.1955=0.96
Note that I get the highest difference with 15&16 bit, not 24. How can
you explain that?
I forgot to put 8-bit test within, but it behaves like 32-bit: mekes no
difference.
--
Milan Mimica