@smf
1-2 hours work isn't anywhere near enough.
@tlosm
Yes, please contact them, and point them to this thread. If anything, it would be nice to get an opinion about the work from someone who knows more about altivec.
@corto
I'd like to add one more item to your list
Quote:
I am ok to ask to markos (from freevec.org) but whatever we choose, we have to:
5 - Be confident that we can raise at least a few thousand Euros in a bounty.
However, that should be done after benchmarking.
Quote:
2. Measure the performance baseline with ffmpeg on selected machines
I'd like to see results both with and without SIMD on x86, ARM, & PowerPC. These should be with both audio and video output disabled, so that it's just the codec. I'm not in a position to run these tests myself, though.
Quote:
3. Profile to know if bottlenecks are really where we suppose
4. Define targets / expected improvements
They're not bottlenecks, but a set of operations that are SIMD optimized on x86 & ARM, but not on PowerPC. In particular, there is no altivec code at all for videos with a 10-bit depth.**
AFAICT, operations such as iDCT have altivec versions for videos with a bit-depth of 8. However, it's hard to tell what the status is for other operations like the computationally-intensive qpel motion compensation. Figuring that out would require going through the code in more detail than I have time for.
I was hoping that a developer more familiar with altivec (and more time available than me) would be interested in taking a closer look. Links to the relevant PowerPC & x86 code are in the first post of this thread.
Hans
** NOTE: I have no idea how common 10-bit videos are. The prometheus trailer is 8-bit.