Question Regarding Optimizing FFMpeg on Blackfin

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Question Regarding Optimizing FFMpeg on Blackfin

Hasseeb Ashfaq
Hello,
  First I have some problem regarding login in.
  I registered and then confirmed the confirmation email and I recieved the welcome message as well. But then i wanted to login to list i kept recieving error message which is am pasting below.
  Error  ffmpeg-user roster authentication failed.
   
---------------------------------
 
  ffmpeg-user list run by michaelni at users.sourceforge.net, baptiste.coudurier at free.fr
ffmpeg-user administrative interface (requires authorization)
Overview of all mplayerhq.hu mailing lists        so could you plz help me in this regard.        Second I have a question regarding FFMpeg implementation on Blackfin. I am mailing it to you if you could post it on the list.        I am trying to implement FFMPEG 2 Decoder on BF561.
My BF561 is running at 600MHZ, my decoder is outputing 100 frames in 13
seconds. If i enable caching it takes 3.8 seconds (just less than real time as
real time requires 4secs to decode 100 frames) but if I enable PPI with it then
it takes 6 seconds.
So any idea how could i make it run faster then this?
worth mentioning that if i use mannual caching(then DMA runs at 133MHZ which is
slower than processor and net result, DMA slows the processor when enabled)
If i enable CPLB option in caching, and select it for write back, I recieve
exception.... why? I dont know
I have already turned on optimization and set it to 100%.
I have used Profiling, and profiling shows bulk of time is used by functions
1. ff_bfin_idct (it is written in assembly and all its parameters are in L1
memory so i believe it cant go any faster than this.
2. ff_bfin_put_pixels8uc
3. ff_bfin_put_pixels16uc
4. ff_bfin_z_put_pixels16_xy2
5. avg_pixels8_xy2_c
6. avg_pixels8_12
(functions number to 2 to 6 takes around 50 % of time, they copy some
attributes of previous picture, with current picture perform some mathematical
function like addition or averaging and then copy it back. Now the profiling
says that both the pictures are in sdram and i believe reading and writing from
sdram take alot of time. I have caching enabled but I am not reaching that
desired position. any helps? p.s. these functions are written in assembly and
considered to be optimized for blackfin architecture.)
So any help or ideas how could i make my system run alittle more faster than it
is running right now?
Thanx alot
Regards
Hasseeb

       
---------------------------------
Looking for last minute shopping deals?  Find them fast with Yahoo! Search.
_______________________________________________
ffmpeg-user mailing list
[hidden email]
http://lists.mplayerhq.hu/mailman/listinfo/ffmpeg-user