Any altivec experts? (H.264 codec)

@feanor
Ok, a difference between theory and reality
I did read things like that about that in Altivec related documents but only store instructions (dstst) were discouraged (potentially using dcbz to clear memory aimed to be written, avoiding fetching data from RAM).

@tommysammy
How do you get those numbers? I only get 17fps according to output.

Quote:

@Antique

Which computer and which graphicard do you use?
I have a X1000+Radeon6870+FE

@tommysammy

X1000 with FE. Tried with a 5450 and a 6970. Same result on both. Playing with no parameters.

@Antique
Do you use WBCandi in background? That cost 2fps

@tommysammy

Nothing in the background.

@Tommysammy

What command line options are you using when testing? Antique is using none, which may be part of the problem. For example, I just had a video play back rather jerkily with the default audio output. That went away as soon as I switched to the AHI output (-ao ao_ahi).

@Antique
Quote:
While waiting for Tommysammy to reply with his command line options, you might want to experiment with this yourself.

Hans

@Hans

Output from MPlayerGui
LiveForIt-MPlayer-6.4 SVN-r37230-snapshot-1.1.1 (C) 2000-2014 MPlayer Team

Playing Media:Filme/Prometheus.mp4.
Cache fill: 0.00% (0 bytes)
Select error: No such file or directory
AmigaOS4.1 Bug select() don't reset or set errno value
Cache fill: 19.84% (2031616 bytes)

libavformat version 55.33.100 (internal)
libavformat file format detected.
[mov,mp4,m4a,3gp,3g2,mj2 @ 0x4a6f0bf8]stream 0, timescale not set
Cache not responding! [performance issue]
[lavf] stream 0: video (h264), -vid 0
[lavf] stream 1: audio (aac), -aid 0, -alang eng
[lavf] stream 2: video (mjpeg), -vid 1
VIDEO: [H264] 1920x816 24bpp 23.976 fps 6828.3 kbps (833.5 kbyte/s)
Clip info:
major_brand: mp42
minor_version: 0
compatible_brands: mp42isomavc1
creation_time: 2011-12-23 08:24:51
genre: Trailer
artist: 20th Fox
title: Prometheus - Trailer
encoder: HandBrake 4344svn 2011111001
date: 2012
Load subtitles in Media:Filme/
Select error: No such file or directory
AmigaOS4.1 Bug select() don't reset or set errno value
VO: [comp_yuv2] Welcome man !.
==========================================================================
Opening video decoder: [ffmpeg] FFmpeg's libavcodec codec family
libavcodec version 55.52.102 (internal)
Selected video codec: [ffh264] vfm: ffmpeg (FFmpeg H.264)
==========================================================================
==========================================================================
Opening audio decoder: [ffmpeg] FFmpeg/libavcodec audio decoders
AUDIO: 44100 Hz, 2 ch, floatbe, 154.4 kbit/5.47% (ratio: 19305->352800)
Selected audio codec: [ffaac] afm: ffmpeg (FFmpeg AAC (MPEG-2/MPEG-4 Audio))
==========================================================================
Not supported audio format 0000001c
Endian: BE
Signed: Yes
Float: Yes
Bits: 32
Special Type: 0
Using Format AF_FORMAT_S16_BE
AHI: Sound format 28 not supported by this driver.
AO: [ahi_dev] 44100Hz 2ch s16be (2 bytes per sample)
Starting playback...
libmpcodecs/vd.c:mpcodecs_config_vo:d
Movie-Aspect is 1.78:1 - prescaling to correct movie aspect.
VO: [comp_yuv2] 1920x816 => 1920x1080 Planar YV12
**
your forceing aspect to 1.78:1
WARNING: workbench screen resolution suggests 1.78:1
VO: Disabling blanker
alloc_output_buffer:155 - format 19
ANS_LENGTH=69.40
Internal COMP YUV FPS 21
VO: Enabling blanker
gfx_ReleaseArg:422

Exiting... (Quit)


Curious, sometimes this trailer runs with 21fps sometimes with 19fps

@tommysammy

3 times 21fps, this is with skiploopfilter. 17fps without it.


LiveForIt-MPlayer-6.4 SVN-r37230-snapshot-1.1.1 (C) 2000-2014 MPlayer Team

Playing RAM Disk:Prometheus - Trailer.mp4.
libavformat version 55.33.100 (internal)
libavformat file format detected.
[mov,mp4,m4a,3gp,3g2,mj2 @ 0x58f17bf8]stream 0, timescale not set
[lavf] stream 0: video (h264), -vid 0
[lavf] stream 1: audio (aac), -aid 0, -alang eng
[lavf] stream 2: video (mjpeg), -vid 1
VIDEO: [H264] 1920x816 24bpp 23.976 fps 6828.3 kbps (833.5 kbyte/s)
Clip info:
major_brand: mp42
minor_version: 0
compatible_brands: mp42isomavc1
creation_time: 2011-12-23 08:24:51
genre: Trailer
artist: 20th Fox
title: Prometheus - Trailer
encoder: HandBrake 4344svn 2011111001
date: 2012
Load subtitles in RAM Disk:
VO: [comp_yuv2] Welcome man !.
==========================================================================
Opening video decoder: [ffmpeg] FFmpeg's libavcodec codec family
libavcodec version 55.52.102 (internal)
Selected video codec: [ffh264] vfm: ffmpeg (FFmpeg H.264)
==========================================================================
==========================================================================
Opening audio decoder: [ffmpeg] FFmpeg/libavcodec audio decoders
AUDIO: 44100 Hz, 2 ch, floatbe, 154.4 kbit/5.47% (ratio: 19305->352800)
Selected audio codec: [ffaac] afm: ffmpeg (FFmpeg AAC (MPEG-2/MPEG-4 Audio))
==========================================================================
Not supported audio format 0000001c
Endian: BE
Signed: Yes
Float: Yes
Bits: 32
Special Type: 0
Using Format AF_FORMAT_S16_BE
AHI: Sound format 28 not supported by this driver.
AO: [ahi_dev] 44100Hz 2ch s16be (2 bytes per sample)
Starting playback...
libmpcodecs/vd.c:mpcodecs_config_vo:d
Movie-Aspect is 2.35:1 - prescaling to correct movie aspect.
VO: [comp_yuv2] 1920x816 => 1920x816 Planar YV12
screen w 1920 h 1200 aspect 1.60:1
VO: Disabling blanker
alloc_output_buffer:155 - format 19
New_Face failed. Maybe the font path is wrong.
Please supply the text font file (~/.mplayer/subfont.ttf).
subtitle font: load_sub_face failed.
New_Face failed. Maybe the font path is wrong.
Please supply the text font file (~/.mplayer/subfont.ttf).
subtitle font: load_sub_face failed.
A: 0.6 V: 0.0 A-V: 0.554 ct: 0.000 0/ 0 ??% ??% ??,?% 2 0
Select error: No such file or directory
AmigaOS4.1 Bug select() don't reset or set errno value
A: 0.5 V: 0.1 A-V: 0.437 ct: 0.004 0/ 0 ??% ??% ??,?% 2 0
ANS_LENGTH=69.40
A: 11.1 V: 8.8 A-V: 2.295 ct: 0.065 0/ 0 71% 6% 1.6% 50 0


************************************************
**** Your system is too SLOW to play this! ****
************************************************

Possible reasons, problems, workarounds:
- Most common: broken/buggy _audio_ driver
- Try -ao sdl or use the OSS emulation of ALSA.
- Experiment with different values for -autosync, 30 is a good start.
- Slow video output
- Try a different -vo driver (-vo help for a list) or try -framedrop!
- Slow CPU
- Don't try to play a big DVD/DivX on a slow CPU! Try some of the lavdopts,
e.g. -vfm ffmpeg -lavdopts lowres=1:fast:skiploopfilter=all.
- Broken file
- Try various combinations of -nobps -ni -forceidx -mc 0.
- Slow media (NFS/SMB mounts, DVD, VCD etc)
- Try -cache 8192.
- Are you using -cache to play a non-interleaved AVI file?
- Try -nocache.
Read DOCS/HTML/en/video.html for tuning/speedup tips.
If none of this helps you, read DOCS/HTML/en/bugreports.html.

A: 81.4 V: 69.3 A-V: 12.170 ct: 1.137 0/ 0 93% 6% 1.4% 1231 0

Internal COMP YUV FPS 21
VO: Enabling blanker
gfx_ReleaseArg:422

Exiting... (End of file)

@feanor
any news to report?
You wanted to commit some code until tuesday

@tommysammy

I did last night, if you check the github page[1], I am a bit sick with the flu atm, but I wanted to run some benchmarks first, but currently this version gives me another fps in some cases and also helps performance with other formats as well (like mpeg/ogg) not just h264. I'll post some benchmarks soon.

[1]: https://github.com/markos/FFmpeg/commits/master

@feanor
Thanks I have flu too. I am since monday at home and sleep very much. Sleeping is the best medicine.

@all

Updated results (up to commit 4766373):



I now want to work on optimizing AAC and AC3, plus some others not so vital components, but every little bit helps. I'll post updates here.

Really nice work ! Thanks for sharing these results

@feanor
Many thanks :)

@all

darn, just found a bug that corrupts the video partially, need to find exactly which commit broke it, so stay tuned. :-/

@feanor

Just som leyman's question out of curiosity about f.ex:


Is there really a gain in unrolling this loop considering more code to cache vs. one inc+branch?

How good is the compiler in optimizing 0*stride -> 0, 1*stride -> stride, 2*stride -> stride<<1, etc?

Is x<<1,2,3 really faster than 2,4,8*x on this CPU?

Just wondering, since you're the expert

@Deniil

in my experience, the more branches you eliminate the better, code can be fully decoded in the pipeline and the compiler usually does a pretty good job of scheduling the instructions to minimize stalls, etc. Plus in case of small loops (this is after all a very small loop), there is no reason not to unroll, as we have plenty of registers to use. What I'm undecided is whether I should just increase src += stride or use the i*stride form. But looking at the asm output, I'm guessing this is a better form, as the src += creates a dependency for each store, whereas the i*stride steps can be precalculated and just used when needed.

@feanor

Yes I also believe i*stride and loop unrolling is better, I was just thinking out loud. But what code does it generate regarding 0* 1* 2* 4* 8*?

@Deniil

asm output of the i*stride is of the form:

mulli r12,r4,7

which produces 7*stride product and puts it in r12. Later on,

stvx v0, 0, r3 (so 0*stride properly gets eliminated)
...
stvx v0,r12,r3