#2: The root-cause is that the videoconvert plugin (converting I420 -> BGRA) has significantly lower performance in GST 1.2.3. This apparently is fixed in newer version of GST (1.3.1). Here is a bugzilla report on the problem: https://bugzilla.gnome.org/show_bug.cgi?id=721701 and commits showing where the fixes are: http://code.metager.de/source/history/freedesktop/gstreamer/gst/plugins-base/gst/videoconvert/
In summary, decode-display demos aren't real-time/usable on AM3, AM4 with this GST upgrade. So, for the short-term, I'm thinking that we should stay with gst-0.10 for AM3, AM4 and upgrade GST only for AM5. In the mid-longer term, we should either back port this fixes to 1.2.3 or have a plan to migrate to a newer version of GST (and the newer version would bring in HEVC plugins as well). #1: If we revert AM3, AM4 to gst-0.10, then we should no longer see the build issues.. #3, #5: These issues appear due to incomplete/missing up-streaming of recipes. We will have a call to sort this out. #4: As we were missing kmssink in the filesystem, we were just trying to play to a fakesink (just to make sure accelerated decode works fine) and that was showing problem. This is an issue with GLSDK as well. Only playbin seems to work with GLSDK and not the basic pipeline as described in the problem report below. We can file an OMAPS IR for this and it can be lower priority if issue #3 is resolved.. Vivek From: Balagopalakrishnan, Anand Sent: Monday, March 23, 2015 10:50 AM To: Chengalvala, Vivek; Cooper Jr., Franklin; Stiffler, Jacob Cc: [email protected] Subject: RE: [meta-arago] [PATCH] gstreamer1.0-plugins-bad: add DEPENDS on libdrm and libdce due to KMSsink Vivek, GST 1.x is a major upgrade from GST 0.10. There are design changes involved especially from a buffer management perspective. The list below doesn't give a clear picture. Can you add the issues to a table and mark the the owner, debug history and next actions. For example, I am assuming issues #3, 5 will be owned by you. Issues #1,2 - not sure who will look at these. Can you let me know what is pipeline #4 trying to achieve? Regards, Anand ________________________________ From: Chengalvala, Vivek Sent: Monday, March 23, 2015 7:49 PM To: Cooper Jr., Franklin; Stiffler, Jacob Cc: [email protected]<mailto:[email protected]>; Balagopalakrishnan, Anand Subject: RE: [meta-arago] [PATCH] gstreamer1.0-plugins-bad: add DEPENDS on libdrm and libdce due to KMSsink Karthik, Anand, Any updates on this? We're checking regarding #2 on AM57x using ARM codecs and will post our observations.. Thanks, Vivek From: [email protected]<mailto:[email protected]> [mailto:[email protected]] On Behalf Of Chengalvala, Vivek Sent: Friday, March 20, 2015 2:17 PM To: Cooper Jr., Franklin; Stiffler, Jacob Cc: [email protected]<mailto:[email protected]>; Balagopalakrishnan, Anand Subject: Re: [meta-arago] [PATCH] gstreamer1.0-plugins-bad: add DEPENDS on libdrm and libdce due to KMSsink Karthik, Anand, Denys, Upgrade to GST 1.0 (using commit b5747ce8caa13e4528e2d3055fa81cc9ca7b48fd on meta-arago) is causing the following problems: 1. Broken build for AM3. Not sure how AM4 build succeeds. 2. AM4x: Performance drop with ARM only decode of H264 (not sure if it is due to a) base GST upgrade b) ffmpeg_dec --> avdec_h264 or c) ffmpeg_color_space --> videoconvert) 3. AM57x: playbin pipeline internally triggers arm video decoder (e.g., avdec_h264) instead of ducati decoder 4. AM57x: cannot run pipeline with video parse + decode, e.g., "gst-launch-1.0 filesrc location=HistoryOfTI-480p.264 ! h264parse ! ducatih264dec ! fakesink" All the above issues are not present with gst-0.10. Please note that Issue #3 is not observed on Vayu with GLSDK EA release 7.0.0.4. Issue #4 is also happening on Vayu with GLSDK EA release 7.0.0.4. In addition to above, 5. kmssink is missing from AM57x file system. We wanted to do an SDF today for PROCESSOR-SDK on AM3, AM4 and due to above failures, we are forced to freeze @ commit #5c2f4eea32afda6e68681470be664eb5a46fd984, and not pick up any further changes on meta-arago. There is another release targeted for next week and we'd like to be get back to the latest of meta-arago as soon as possible. Appreciate your help in addressing the above issues. Thanks, Vivek -----Original Message----- From: [email protected]<mailto:[email protected]> [mailto:[email protected]] On Behalf Of Cooper Jr., Franklin Sent: Friday, March 20, 2015 9:58 AM To: Stiffler, Jacob Cc: [email protected]<mailto:[email protected]> Subject: Re: [meta-arago] [PATCH] gstreamer1.0-plugins-bad: add DEPENDS on libdrm and libdce due to KMSsink Well that's not good. > On Mar 20, 2015, at 8:42 AM, Stiffler, Jacob > <[email protected]<mailto:[email protected]>> wrote: > > I ran the demo while running "top". It appears that CPU is near 100%. > However, this demo was working fine with gstreamer 0.10. > > -----Original Message----- > From: Cooper Jr., Franklin > Sent: Friday, March 20, 2015 9:34 AM > To: Stiffler, Jacob > Cc: Dmytriyenko, Denys; > [email protected]<mailto:[email protected]> > Subject: Re: [meta-arago] [PATCH] gstreamer1.0-plugins-bad: add DEPENDS on > libdrm and libdce due to KMSsink > > > > >> On Mar 20, 2015, at 8:17 AM, Stiffler, Jacob >> <[email protected]<mailto:[email protected]>> wrote: >> >> I tried the matrix h264 decode demo on am437x with gstreamer 1.0. >> >> To get it to work, I had to modify runH264Dec.sh to use the new pipelines. >> The new command is the following: >> >> gst-launch-1.0 -v filesrc location=$filename ! h264parse ! avdec_h264 ! >> videoconvert ! fbdevsink device=/dev/fb0 >> >> >> Once this was done, I was able to run the demo, but the framerate is >> extremely slow. Is this the correct command? Or is this behavior expected? > If slow as in stuttering and choppy then no. If cpu utilization is high then > that is expected. >> >> Thanks, >> Jake >> >> >> -----Original Message----- >> From: Stiffler, Jacob >> Sent: Friday, March 20, 2015 6:56 AM >> To: Dmytriyenko, Denys; >> [email protected]<mailto:[email protected]> >> Subject: RE: [meta-arago] [PATCH] gstreamer1.0-plugins-bad: add DEPENDS on >> libdrm and libdce due to KMSsink >> >> It appears I got lucky with a successful am437x-evm build. The package >> libdrm was installed and libdce was not. >> >> However, I got a build failure for am335x-evm due to the missing libdrm >> dependency: >> >> In file included from gstdrmbufferpool.c:42:0: >> gstdrmbufferpool.h:42:22: fatal error: omap_drm.h: No such file or directory >> compilation terminated. >> make[4]: *** [libgstdrm_1.0_la-gstdrmbufferpool.lo] Error 1 >> make[4]: Leaving directory >> `/home/gtbldadm/ti/oe-layersetup/build-CORTEX_1/arago-tmp-external-linaro-toolchain/work/cortexa8t2hf-vfp-neon-oe-linux-gnueabi/gstreamer1.0-plugins-bad/1.2.3-r0-arago0/gst-plugins-bad-1.2.3/gst-libs/gst/drm' >> make[3]: *** [all-recursive] Error 1 >> make[3]: Leaving directory >> `/home/gtbldadm/ti/oe-layersetup/build-CORTEX_1/arago-tmp-external-linaro-toolchain/work/cortexa8t2hf-vfp-neon-oe-linux-gnueabi/gstreamer1.0-plugins-bad/1.2.3-r0-arago0/gst-plugins-bad-1.2.3/gst-libs/gst' >> make[2]: *** [all-recursive] Error 1 >> make[2]: Leaving directory >> `/home/gtbldadm/ti/oe-layersetup/build-CORTEX_1/arago-tmp-external-linaro-toolchain/work/cortexa8t2hf-vfp-neon-oe-linux-gnueabi/gstreamer1.0-plugins-bad/1.2.3-r0-arago0/gst-plugins-bad-1.2.3/gst-libs' >> make[1]: *** [all-recursive] Error 1 >> make[1]: Leaving directory >> `/home/gtbldadm/ti/oe-layersetup/build-CORTEX_1/arago-tmp-external-linaro-toolchain/work/cortexa8t2hf-vfp-neon-oe-linux-gnueabi/gstreamer1.0-plugins-bad/1.2.3-r0-arago0/gst-plugins-bad-1.2.3' >> make: *** [all] Error 2 >> ERROR: oe_runmake failed >> >> - Jake >> >> -----Original Message----- >> From: >> [email protected]<mailto:[email protected]> >> [mailto:[email protected]] On Behalf Of Dmytriyenko, >> Denys >> Sent: Thursday, March 19, 2015 6:37 PM >> To: [email protected]<mailto:[email protected]> >> Subject: Re: [meta-arago] [PATCH] gstreamer1.0-plugins-bad: add DEPENDS on >> libdrm and libdce due to KMSsink >> >> Karthik, >> >> I had a question - I see that KMSsink in plugins-bad checks for drm and dce >> headers. I was trying to set the dependency accordingly, but it would break >> non dce platforms. Can you please elaborate on that a bit? Thanks. >> >> -- >> Denys >> >> >>> On Thu, Mar 19, 2015 at 05:29:38PM -0400, Denys Dmytriyenko wrote: >>> From: Denys Dmytriyenko <[email protected]<mailto:[email protected]>> >>> >>> Signed-off-by: Denys Dmytriyenko <[email protected]<mailto:[email protected]>> >>> --- >>> .../gstreamer/gstreamer1.0-plugins-bad_1.2.3.bbappend | 4 >>> +++- >>> 1 file changed, 3 insertions(+), 1 deletion(-) >>> >>> diff --git >>> a/meta-arago-extras/recipes-multimedia/gstreamer/gstreamer1.0-plugins- >>> bad_1.2.3.bbappend >>> b/meta-arago-extras/recipes-multimedia/gstreamer/gstreamer1.0-plugins- >>> bad_1.2.3.bbappend >>> index b1cacac..bd141a8 100644 >>> --- >>> a/meta-arago-extras/recipes-multimedia/gstreamer/gstreamer1.0-plugins- >>> bad_1.2.3.bbappend >>> +++ b/meta-arago-extras/recipes-multimedia/gstreamer/gstreamer1.0-plug >>> +++ ins-bad_1.2.3.bbappend >>> @@ -2,6 +2,8 @@ FILESEXTRAPATHS_prepend := "${THISDIR}/${PN}:" >>> >>> PACKAGECONFIG = "faad wayland" >>> >>> +DEPENDS += "libdrm libdce" >>> + >>> SRC_URI_append = " \ >>> >>> file://0001-Added-GstDRMBufferPool-support.patch<file:///\\0001-Added-GstDRMBufferPool-support.patch> >>> \ >>> >>> file://0002-Modified-waylandsink-to-accept-NV12-format.patch<file:///\\0002-Modified-waylandsink-to-accept-NV12-format.patch> >>> \ @@ -9,4 +11,4 @@ SRC_URI_append = " \ >>> >>> file://0004-waylandsink-Removed-dependency-on-dri2.patch<file:///\\0004-waylandsink-Removed-dependency-on-dri2.patch> >>> \ >>> >>> file://0005-vc1parse-and-jpegparse-Fixes-plugin-ranks.patch<file:///\\0005-vc1parse-and-jpegparse-Fixes-plugin-ranks.patch>" >>> >>> -PR_append = "-arago0" >>> +PR_append = "-arago1" >>> -- >>> 2.2.0 >> _______________________________________________ >> meta-arago mailing list >> [email protected]<mailto:[email protected]> >> http://arago-project.org/cgi-bin/mailman/listinfo/meta-arago >> _______________________________________________ >> meta-arago mailing list >> [email protected]<mailto:[email protected]> >> http://arago-project.org/cgi-bin/mailman/listinfo/meta-arago _______________________________________________ meta-arago mailing list [email protected]<mailto:[email protected]> http://arago-project.org/cgi-bin/mailman/listinfo/meta-arago
_______________________________________________ meta-arago mailing list [email protected] http://arago-project.org/cgi-bin/mailman/listinfo/meta-arago
