[MPlayer-dev-eng] Re: PATCH [0/12] CoreAVCDecoder support

[Alan Nisota]

Compn writes:
> i wonder if you have any mingw mplayer binaries with the coreavc patchset 
> applied? or if you know that it builds on win32 ok?

1st let me appologize for breaking the threads in my past couple of posts.  I 
had suspended my mail delivery, and so didn't have an easy way to respond.  

It built fine for me on mingw, however, my binaries would be useless, as I 
wasn't able to get display to work at any resonable frame-rate (even for normal 
mplayer)  If I use '-vo null' I can benchmark in windows, and it seems to be 
about right.  I haven't found a way to compare to Native windows though...Using 
graphedit w/ Chegepuga gives almost identical performance to mplayer with '-vo 
null'  Using the Haali 'TimeCodec' utility gives performance 20% better than 
mplayer with '-vo null'

> 
> also do you have any current benchmarks for us ? i know i'm asking a lot, but 
> providing specific samples where coreavc is faster will also be appreciated.
> 

Here are some benchmarks on Linux using a 2.8GHz P4 w/ HT enabled (adds a few 
fps with the dual-cpu patch).  I don't have a dual-cpu system to benchmark 
against, but others have confirmed that a dual-cpu system just about doubles 
performance with this codec.  I also have a 1.8 GHz Pentium-M which tested 
anywhere from the same to 20% slower on Windows compared to the P4.
Using CoreAVC 1.2.0
mplayer -demuxer lavf -vo null -nosound -benchmark <stream>
Need to add '-demuxer lavf' when using TS streams, as mplayer's demuxer doesn't 
supply 'format' during codec initialization.
                                        ffh264   coreAVC
sample1 h264 1280x720x60fields/sec:      82       117    fields/sec
sample2 mbaff 1440x1080x25frames/sec:    13        17    frames/sec
sample3 avc1  1280x720x30frames/sec:     48        72    frames/sec
sample4 paff  1440x1080x30frames/sec     N/A       45    frames/sec

here is an archive containing segments of the above streams.  In most cases, it 
is just a small section of a larger file which I used for benchmarking.
http://www.zshare.net/download/samples-tgz.html

Last time I did this, I generated a bunch of data by encoding with x264 and 
various options, I'm not going to redo that.  The results shouldn't have 
changed much.