Hi,

KH KH schrieb:
2008/1/14, Burkhard Plaum <[EMAIL PROTECTED]>:
Hi,

KH KH schrieb:
[...]
[EMAIL PROTECTED] video]$ nasm -f elf64 reconmmx.s -o reconmmx.o
nasm: fatal: unrecognised output format `elf64' - use -hf for a list
AFAIK, nasm cannot produce 64 bit binaries.
Both yasm or gcc inline asm work here.
IT seems it do not work easily even with a newest svn snapshot of cinelera...
I'm exploring this with the aim to produce a patch that could be merged.
Unless you have already one ?

No.

I will try with gcc inline

AFAIK, gcc inline asm has a different syntax than nasm.

I had a different case (forgot which) where replacing the nasm command
in the Makefile by the yasm equivalent worked.

Also it would be interesting to have this into the system libmpeg3.
One good reason would be to be able to preload this library when OSes
use this functionality massively.

Don't know what this has to do with compiling MMX code

By the way. I wonder if all this stuff will be necessary:
quote of the doc/index.htm from the libmpeg3 pacakge:
--------------------
Call <CODE>mpeg3_set_mmx(mpeg3_t *file, int use_mmx)</CODE> to set if
MMX is used for video.  Disabling MMX is mandatory for low bitrate
streams since it is very lossy.  By the way, lately the compiled MMX
output has been producing corrupted video.  This is a change in the way
modern compilers and CPU's handle MMX from the way it was done 4 years
ago but since modern CPU's are so fast, you're better off not using MMX
at all.<P>
--------------------
mpeg3_set_mmx function isn't still available in 1.7. So i wonder it it
really important to have it mmx compiled on modern OSes ?

I don't know much about the MMX routines used for MPEG decoding, but usually
MMX has a decreased precision (e.g. 7 bit instead of 8 bit). If you write
"lossless" MMX code, it will be much slower because e.g. you process 4 uint16_t
instead of 8 uint8_t at once.

IMO the best would be to implement MPEG-1/2 decoding with libmpeg2 (not 3!).
It's known to be very stable, fast (with MMX code) and 64-bit clean.

The speedup of MMX is in the order of 200%...700%, but MPEG-decoding is still
pretty fast compared to the other operations done ine cinelerra (e.g. by
filters).

Burkhard

_______________________________________________
Cinelerra mailing list
[email protected]
https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra

Reply via email to