[Mplayer-dev-eng] MP3LIB: new stuff

[Nick Kurshev]

Hi, Felix Buenemann!

On 2001-06-23 21:50:05 you wrote:

>> I've done applying of MMX optimized decore to mp3lib (PLEASE SEE ATTACH,
>> NOT CVS). There are big changes:
>> 1. I move out all constructions like #ifdef HAVE_3DNOW and so on.
>> Now all version of decoders are present in mp3lib and MP3_Init chooses them
>> automatically. (It's even faster because old synth_1to1 code contained too
>> many unnecessary if else constructions)
>> 2. Probably I have found and fixed error in SSE stuff! (Arpi, Felix please
>> your attention) Bug was in tabinit when SSE decoder uses modified 3dnow
>> stuff but right initialization was limited only by detecting AMD cpu.
>> 3. I've found that MMX decoder is faster for K7 (probably it will faster
>> and for P3) Please test it - and maybe it would be better to move out
>> decode_k7 and decode_sse! 4. It would be better to test this new stuff by
>> all developers who has different cpus: (K6, Pent-MMX, k6-2, (k7 - tested),
>> P3). After that it can be applyed to mplayer.
>>
>> Waiting for testing and reports!
>Applied patch, gives me sort of white noise with sse (with decode_sse.s from 
>cvs, will try my vast bunch of others too), how can I disable sse (->try 
>mmx)? (haven't looked at the sources yet, so maybe I'll have found it out the 
>time you answer).
I mean that it would be better to move out sse code.
Now it requires additional investigation. IMHO you with me tested about 20-40
different versions of decode_see isn't? We should to find out workable from
those. But for now you can simply disable sse code by:
_isse = 0
In MP3_Init (sr1.c source after _isse=isse() function call)

Best regards! Nick (2001-06-24 01:29:41)


_______________________________________________
Mplayer-dev-eng mailing list
Mplayer-dev-eng at lists.sourceforge.net
http://lists.sourceforge.net/lists/listinfo/mplayer-dev-eng