XviD DSF

[Nic]

I hope your attempt wasn't too "abortive", optimising the SAD code is a really good idea....

YASM is here (as h pointed out)
http://www.tortall.net/projects/yasm/

But doesn't look any better really than NASM....(i.e. not fully done)

But I hope you do get somewhere with it...even if its just inline for Win32 for now ;)

Cheers,
-Nic

[avih]

tom, i was using Masn and not Nasm. i didn't use any Nasm yet.

i didn't refer to greedy/hma in any of my posts (except indirectly when asking about dscaler status)

food for the mind:
1. could vc6 disasm be corrupted for (some) sse2 instructions?
2. can u just 'pop in' the correct op-code directly to the source? i'm sure Nasm has such an option. its going back to the good old days, but it could work if u have only few glitches.

cheers
avi.

[robUx4]

Originally posted by Nic
Ill look into doing the tooltips, never done them before (outside of MFC) :)

-Nic

I suggest that you have a look at the VirtualDub source code. Since it uses tooltips.

[ChristianHJW]

Originally posted by trbarry
But there is certainly no shortage of things to do. Xvid is one of the most interesting things going these days. ... especially in combination with Vorbis audio in an Ogg ( or soon MCF ) container ;) ..

[avih]

tom-

"To code the conditional move, MMX, XMM instructions etc. on an assembler that doesn't have these instructions use the macros at www.agner.org/assem/macros.zip
"

taken from 'pentopt.htm' (from same web site). it might help u to remain within nasm

cheers
avi

[trbarry]

"To code the conditional move, MMX, XMM instructions etc. on an assembler that doesn't have these instructions use the macros at www.agner.org/assem/macros.zip "

avih -

Thanks, I'll go get that. But someone also suggested another version of nasm I can try. There seem to be no shortage of them. ;)

nic -

I haven't given up completely. But I think much of SSE2's advantage over SSE is only for well aligned data. There should be some of that in XviD somewhere and maybe soon more if some aligned_malloc() is used.

Meanwhile I'll just sit on the SSE2 sad16 code until I can make it actually do something usefull. But I'll still look for other places where it might already help.

Since you have VS7, can you tell me:

- Does VS7 offers any extra support for 16 byte alignment or SSE2 instructions?

- Does it still need a processor pack for SSE2?

- Can it display the value of SSE2 XMM registers in the IDE?

Thanks,

- Tom