VAAPI: green artifacts/stutter in mpeg2 mkv movies

[Memesa]

Recently I've build my new m-itx HTPC box, and it was also the first time I used openelec (3.0 RC1). I'm really astonished by the ease of installation, speed and stability. 720p and 1080p HD content work right away.
Some of the films however showed a lot of green artifacts and usually stutter or even locks up the system altogether. These were all obtained by converting a DVD into an mkv using makemkv. Most of the films converted like this are absolutely fine, so I did some comparison and found out that the only difference in mediainfo were the following lines:

Code:

Color primaries : BT.601 NTSC
Transfer characteristics : BT.601
Matrix coefficients : BT.601

movie with errors: http://pastebin.com/927i00xk movie without errors: http://pastebin.com/KDNru0Zu

The artifacts only show up when vaapi is enabled, without hardware acceleration everything is perfectly fine.
I've tried XBMCbuntu, however vaapi seems to be broken (onscreen info doesn't list vaapi being active). In addition I've tried Fedora 18, with vaapi enabled it has the same issues in XBMC and in VLC (can't find vaapi enabled mplayer for Fedora). When I inserted the USB disk drive where openelec temporarily is installed to, into a ivy bridge (i5 3570K) computer, thesame artifacts were there.

I've cut the first 15 seconds of the the movie with errors and the one without for reference. It would be great if someone can test them on their vaapi accelerated openelec.
movie with errors:
http://rapidshare.com/files/3729165933/4...0m15s).mkv
movie without errors
http://rapidshare.com/files/392775373/12...0m20s).mkv

I've boiled down the problems to the video part, when extracting the m2v from the mkv it still show the artifacts. I tried to modify the header, however I just can't seem to get any changes to what mediainfo shows me. Does anybody have any idea how to solve the issue, other than reencoding...? So by either changing the header, or even better by fixing this through a change in vaapi?

My HTPC configuration:
Intel Pentium G630
Asrock H61M-ITX
Corsair 4GB DDR3 dual channel
Lian-Li Lian Li PC-Q02
Intel postville SSD 80GB

Regards,
Memesa

[Memesa]

Some additions:
So far I've tried the following:
Openelec 3.0 RC1: artifacts
Openelec 3.0 RC2: playback completely broken
Openelec dev 5 feb: artifacts
Openelec 2.0: artifacts
Fedora 18 XBMC 3.0 RC3: artifacts
Fedora 18 VLC: artifacts
XBMCbuntu 3.0: vaapi broken; no artifacts
Ubuntu 12.10: artifacts

Fedora 18 vainfo:

Code:

libva info: VA-API version 0.33.0
libva info: va_getDriverName() returns 0
libva info: Trying to open /usr/lib64/dri/i965_drv_video.so
libva info: Found init function __vaDriverInit_0_33
libva info: va_openDriver() returns 0
vainfo: VA-API version: 0.33 (libva 1.1.0)
vainfo: Driver version: Intel i965 driver - 1.0.19
vainfo: Supported profile and entrypoints
VAProfileMPEG2Simple : VAEntrypointVLD
VAProfileMPEG2Main : VAEntrypointVLD
VAProfileH264Baseline : VAEntrypointVLD
VAProfileH264Baseline : VAEntrypointEncSlice
VAProfileH264Main : VAEntrypointVLD
VAProfileH264Main : VAEntrypointEncSlice
VAProfileH264High : VAEntrypointVLD
VAProfileH264High : VAEntrypointEncSlice
VAProfileVC1Simple : VAEntrypointVLD
VAProfileVC1Main : VAEntrypointVLD
VAProfileVC1Advanced : VAEntrypointVLD

This is how the artifacts looks like:

[cncb]

I can't help at all other than to say "me too". I get this behavior on both my Pentium G620 and Core i3-2100. I have also seen more subtle green artifacts with Blu-ray sourced MKVs (VC1) which also go away when VAAPI is disabled. This along with a constant black screen via HDMI when I wake my PC from sleep (discussed in another thread in this forum) leads me to believe XBMC and Intel graphics aren't very compatible at the moment.

[Memesa]

It's really weird, the artifacts. Have you been able to analyse some of the Blu-Ray sourced MKVs using mediainfo, to see if it also holds any info about colour primaries, transfer characteristics or matrix coefficients?

The black screen issue was something that I was facing as well. It seems to be resolved after connecting XBMC directly to my TV by HDMI, instead of routing it through the receiver. So I assume it needs EDID info to be able to start rendering. I noticed that in xrandr, all info was properly set (including EDID info) while the screen was still black.

Do you maybe also have an Asrock motherboard, maybe that is what is causing the issues? Both the pc's I tried here have an Asrock H61 motherboard.

And in my opinion its not XBMC and Intel that are not very compatible at the moment, it's rather the combination of Linux and Intel what is causing the issues (as VLC in Fedora shows the same issues). Which is really a shame for the beautiful combination of hardware, which is hampered by the support of their own manufacturers.

[cncb]

Sorry, I really haven't investigated this at all. I just noted the problems and stopped using it as the whole reason for using OpenELEC for me is to have it just work like an appliance and not have to deal with these "PC-like" issues. I have Intel H67 motherboards so it doesn't appear to just be an Asrock issue.

[MikeSeddon]

Hi Memesa,
There is a thread going on in the issues list about Openelec Intel video issues:
https://github.com/OpenELEC/OpenELEC.tv/issues/1687

I suggest you access it and download the build which has been posted by itechatmxc (dropbox link). Basically the build dependency list was modified to use the most recent source code for the Intel VAAPI (graphics) Linux drivers. The guys developing them are part of Intel so hopefully it works.

Can you post your outcome in that thread as it appears to be working for a lot of people and if we get enough we can get Openelec to merge the change for the Frodo final.

Cheers,
Mike

[Memesa]

cncb, if I understand correct, you're now running Windows to use XBMC. And thereby I completely agree with you, running XBMC under windows makes it feel slower and less optimised. One of it being necessary to change the remote commands to prevent Windows Media Center from starting.

MikeSeddon, thanks for pointing me to the topic. I've tried the build and noted my results in the thread. The issue however is not fixed by using his build.
For RC2, is it true that vaapi is used from the master branch as well? (seeing similar results)

[Endymion]

His Image fixed my VAAPI problems, but i cant play DTS-HD anymore on my Intel NUC (HD4000) DC3217IYE. I get only noise.

[Memesa]

My receiver is connected through SPDIF, which is thereby also unable to output HD audio (doesn't support it either).

But his image shows exactly the same issues as OpenELEC 3.0 RC2 shows; gives playback sometimes, other moments it just crashes directly. These crashes gave me the following info in one of the logs:
On a h264 mkv:

Code:

Feb 13 22:15:04 openelec user.crit kernel: [ 160.788075] kernel BUG at drivers/gpu/drm/i915/i915_gem_execbuffer.c:1002!
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788146] invalid opcode: 0000 [#1] SMP
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788191] Modules linked in: xhci_hcd iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat ip_tables ir_mce_kbd_decoder ir_sanyo_decoder ir_sony_decoder ir_jvc_decoder ir_rc6_decoder ir_rc5_decoder ir_nec_decode
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788584] CPU 1
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788608] Pid: 2130, comm: xbmc.bin Not tainted 3.7.4 #1 To Be Filled By O.E.M. To Be Filled By O.E.M./H61M-ITX
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788704] RIP: 0010:[<ffffffff81349b38>] [<ffffffff81349b38>] i915_gem_do_execbuffer.isra.14+0xa54/0xd6a
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788803] RSP: 0018:ffff88009e8bfc68 EFLAGS: 00010206
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788855] RAX: 0000000000000270 RBX: ffff88011eb896d8 RCX: ffff88011eb89678
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788922] RDX: ffff88011eb89600 RSI: ffff88011eb8abe0 RDI: ffff88011eb89810
Feb 13 22:15:04 openelec user.warn kernel: [ 160.788988] RBP: ffff88009e8bfdf0 R08: ffff88011eb895a0 R09: 0000000000000000
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789055] R10: ffff88011eb895a0 R11: ffff880119ed4940 R12: ffff88011eb88000
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789121] R13: 0000000000000aec R14: 0000000000000000 R15: ffff88009e8bfc40
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789189] FS: 00007f8e22e75700(0000) GS:ffff88011f300000(0000) knlGS:0000000000000000
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789265] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789318] CR2: 00007f8e2c3fa2a8 CR3: 000000010da33000 CR4: 00000000000407a0
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789384] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789451] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789518] Process xbmc.bin (pid: 2130, threadinfo ffff88009e8be000, task ffff88010c6620d0)
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789596] Stack:
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789616] ffff880119cde5d0 0000000000000002 ffff88011ebc2b08 0000000000000007
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789693] 0000000000000008 ffff88011e800800 ffff88011ab71800 ffff880000000000
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789771] ffff88011ab71800 000000000da737c0 ffff880119f07740 ffff88009e8bfd18
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789849] Call Trace:
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789878] [<ffffffff8134a1ae>] ? i915_gem_execbuffer2+0x105/0x173
Feb 13 22:15:04 openelec user.warn kernel: [ 160.789939] [<ffffffff81344ae5>] ? i915_gem_sw_finish_ioctl+0x5c/0x67
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790002] [<ffffffff813278bf>] ? drm_ioctl+0x28c/0x36a
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790057] [<ffffffff810864c1>] ? sync_mm_rss+0x35/0x5f
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790110] [<ffffffff8134a0a9>] ? i915_gem_execbuffer+0x25b/0x25b
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790172] [<ffffffff8101b95d>] ? __do_page_fault+0x28c/0x32a
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790230] [<ffffffff810a66de>] ? vfs_ioctl+0x1e/0x31
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790281] [<ffffffff810a6f09>] ? do_vfs_ioctl+0x3ed/0x42f
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790337] [<ffffffff8109ae0c>] ? vfs_write+0xab/0xd4
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790388] [<ffffffff8109bf08>] ? fput+0xf/0xac
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790434] [<ffffffff810a6f86>] ? sys_ioctl+0x3b/0x65
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790487] [<ffffffff815c8212>] ? system_call_fastpath+0x16/0x1b
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790544] Code: e3 a9 ff ff 83 7b 6c 00 75 21 8b 94 24 90 00 00 00 48 89 df 48 8b 74 24 78 e8 d5 e7 ff ff 85 c0 41 89 c6 0f 85 34 02 00 00 eb 02 <0f> 0b 48 3b 9c 24 88 00 00 00 75 6e 8b 74 24 4c 41 3b 74 24 10
Feb 13 22:15:04 openelec user.alert kernel: [ 160.790858] RIP [<ffffffff81349b38>] i915_gem_do_execbuffer.isra.14+0xa54/0xd6a
Feb 13 22:15:04 openelec user.warn kernel: [ 160.790931] RSP <ffff88009e8bfc68>

Mpeg2 mkv's are more forgiving and don't crash the system, but hang instead;

Code:

Feb 13 22:49:06 openelec user.err kernel: [ 37.777671] [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung
Feb 13 22:49:13 openelec user.err kernel: [ 44.152313] [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung
Feb 13 22:49:16 openelec user.notice Boot: ### syncing hardware clock with system clock ###
Feb 13 22:49:20 openelec user.err kernel: [ 51.350030] [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung

One thing I guess it's clear; it's the GPU that crashes. When VAAPI is disabled, the above crashes do not occur.

I didn't test the build in the github for its logs to find the origin of its crashes, but the screen looks exactly thesame (plain green) and thereby presume it has the same issues.

[Johan]

I have the same problem with the movie with "errors" using the Intel x86_64 Version:2.99.4 (3.0 RC 4) version. However XBMCbuntu plays the movie without artifacts, it uses 3.0 Mesa 9.0.
An other problem is stuttering when I play my own movies made by my Panasonic HDC-SDT750, 1920X1080 50p MPEG-4 AVC/H.264 , also in this case XBMCbuntu does everything well.

Configuration:
Asrock B75M-ITX
Intel i3-3225.
Kingston 4GB DDR3 dual channel.