edge defects to play wmv7/8/9 or vc1 files on beagleboard

[kaijun61]

When I play wmv7/8/9 or vc1 files on cortex-a8 based beagleboard with recent ffmpeg based gstreamer, there are random green color at the top edge of screen and mirrored image at the left edge of screen. But no problem to play h264, mpeg4/2/1 files

[cehoyos]

Is this also reproducible with the following?
ffmpeg -i input -an -qscale 2 out.avi
If yes,. please provide complete, uncut console output and please also test with --disable-asm (and report).

Do you think this is a regression?

[kaijun61]

Hi, after ffmpeg conversion, the defects are gone. I don't think it is a regression.

[cehoyos]

I meant: Is the problem is reproducible if you install current ffmpeg on your beagleboard and try above command line (on the beagleboard) - does the output file show the artefacts you saw at playback?

[kaijun61]

After I converted wmv9 into mpeg4 on beagleboard by ffmpeg as suggested, then I see no artifacts at playback.

[cehoyos]

Then why do you think there is an issue?

[kaijun61]

As explained, if I play the original wmv9 file, there are artifacts at the edges of screen. But after it is converted into mpeg4 file as suggested, then I play the mpeg4 file. Those artifacts disappear!

[cehoyos]

Can you reproduce the artefacts if you compile FFmpeg with --disable-asm?

[kaijun61]

No, not reproduceable. Those artefacts disappear in the mpeg4 after I compile FFmpeg with --disable-asm.

[cehoyos]

I thought the artefacts are only visible with WMV, not with mpeg4?

[kaijun61]

yes, that is issue.

[cehoyos]

So if you see artefacts playing a wmv9 file, do you also see the artefacts if you compile libavcodec (FFmpeg) with --disable-asm?

[kaijun61]

yes, I do. It seems wmv9 doesn't use neon.

[cehoyos]

Is the problem also reproducible on other platforms?

[kaijun61]

I cannot see them in PC.

[cehoyos]

Do you suspect a compiler issue?

[kaijun61]

Why?

[cehoyos]

If the same code works fine on one platform but not on another (and you disabled all platform-specific optimisations), what other reason for the failure is likely?

Could you explain how you are compiling current GStreamer with current FFmpeg? Perhaps I can try to reproduce the issue...

[kaijun61]

all are compiled with OE (angstrom tool chains).

[kaijun61]

I sorted out the issue. It is my bug. Since the decoded frame buffers are allocated in physical contiguous memory for post-processing by DSP, h264, mpeg2/4 increase decoded frame size for CODEC_FLAG_EMU_EDGE, but wmv9 doesn't. That is why I see artefacts of edges.