[MEncoder-users] new doom9 codec comparission
Corey Hickey
bugfood-ml at fatooh.org
Thu Dec 8 07:14:21 CET 2005
Michael Niedermayer wrote:
>>>>Unfortunately, the quality drop from using threads=2 is, in my opinion,
>>>>unacceptable.
>>>
>>>Definitely. I'm against using threads.
>>
>>Last night, on a whim, I decided to test qpel. Tests published by Rémi
>>Guyomarch a few years ago showed a definite PSNR improvement from using
>>qpel. My own tests confirmed that. Later on, I found that qpel was
>>decreasing PSNR, and tests by a few other people confirmed that.
>>
>>Now, oddly enough, adding qpel raised the PSNR from 42.66 to 42.96, with
>>a definite corresponding quality improvement. I can't say whether qpel
>>should be recommended in general, but it obviously helps under these
>>specific circumstances.
>>
>>Meanwhile, Michael had fixed the PSNR calculation for multithreaded
>>encoding (thanks, Michael), so I ran a test with threads=2 and qpel.
>>This brought PSNR down to 42.92, and I can see only a very very slight
>>decrease in overall quality. Although "qpel" is the best,
>>"qpel:threads=2" still looks a little better than omitting both options.
>>
>>Unfortunately, adding qpel brings the second pass' fps down from 29 to
>>18. The first pass is the same (92), so the average drops from 44 to 30.
>
> you could try subq if you still have time ...
Good idea. I graphed the results of some old data I have on subq
testing. subq=5 is probably a reasonable value for a significant
performance increase and a fairly small PSNR decrease.
I'll test that and the surrounding subq values tonight. Doom9 wants
results by the weekend at the latest, but earlier is better so I'd
really like to get them out by tomorrow.
>>Can anyone give me a general estimate of how much of a performance
>>improvement to expect from using threads=2 on a dual-core machine? I
>>don't have access to any multiprocessor/multicore setups so I can't test
>>it myself. If the improvement is better than 40% or so, then I think
>>qpel and threads=2 is worth it. Otherwise, I think it would be good to
>>leave them both out. I'm going to wait a bit longer before I send the
>>results to Doom9, but I would appreciate it if anybody gave me some
>>feedback about this ASAP.
>
>
> i dunno but my feeling is that the threads=2 speedup will be below 40%
Just now I used some old data from tests of material that probably isn't
very similar to make some rough calculations that I probably screwed up,
but if everything is correct than the speedup from using threads=2
would only have to be 13% in order to, along with qpel:subq=5, give
equal quality to using neither threads=2 nor qpel.
So, now the question becomes: do you think the performance improvement
from threads=2 is greater than 13%?
Alternatively, I could drop the idea of threads=2 but retain
qpel:subq=5. This should give a mild but noticeable quality improvement
at the cost of lowering average fps from 44 to 39 (versus 30 for not
using subq).
Again, this is based on rough calculations that might or might not be
close; I'll have real numbers tomorrow.
-Corey
More information about the MEncoder-users
mailing list