[FFmpeg-devel] H.264 CABAC Decoder on IBM Cell
Thu May 7 22:40:01 CEST 2009
2009/5/7 Loren Merritt <lorenm at u.washington.edu>:
> On Thu, 7 May 2009, Jason Garrett-Glaser wrote:
>> I'm going to blame the fact that the primary core of the Cell (stupidly)
>> doesn't have a barrel shift unit, resulting in variable shifts, such as
>> those in renormalization, being implemented in microcode (and
>> correspondingly being extraordinarily slow).
>> But that's just a guess. ?It needs to be profiled.
> Those shifts could all be multiplies if that's faster. But a quick google
> says that PPU sucks at integer multiplies too, just not quite as much as
> c->range could be LUTed, but c->low would take quite a big LUT.
you need to achive a lot more than 2x todays throughput (1080p etc)
In my honest opinion i think the spu thing its a step in the right direction.
It completely offloads the task rather than try to squeeze something
out of something not meant to deliver anything.
The ppu's task in a cell is only meant to be a manager of tasks.
Not a executioner.
And eventually we should aim to offload everything
> --Loren Merritt
> ffmpeg-devel mailing list
> ffmpeg-devel at mplayerhq.hu
Email: kristian.jerpetjoen at gmail.com
More information about the ffmpeg-devel