Summary: | vpp timing issue on hsw | ||
---|---|---|---|
Product: | libva | Reporter: | Rainer Hochecker <fernetmenta> |
Component: | intel | Assignee: | haihao <haihao.xiang> |
Status: | RESOLVED FIXED | QA Contact: | Sean V Kelley <seanvk> |
Severity: | normal | ||
Priority: | medium | CC: | fritsch, gb.devel, holger.k |
Version: | unspecified | ||
Hardware: | x86-64 (AMD64) | ||
OS: | Linux (All) | ||
Whiteboard: | |||
i915 platform: | i915 features: |
Description
Rainer Hochecker
2014-06-05 18:49:56 UTC
Sorry, actually it never works on hsw. The breakpoint made vaapi time out the player open sw codec. I did some more tests: - if I don't set the forward ref, ever 2nd picture is green - if I set the ref, the pics look temporarily misordered. How can we verify that the drivers picks the correct reference? As you mentioned in https://bugs.freedesktop.org/show_bug.cgi?id=79528, gst-vaapi + MADI works with MPEG-2 video on HSW, so do you mean XBMC + MADI doesn't work with MPEG-2 video on HSW ? Could you enable LIBVA_TRACE to dump VA parameters to a log file ? e.g. export LIBVA_TRACE=libva then a log file named libva.xxx.xxx will be generated when you run xbmc with vaapi enabled. The problem here has the following differences to the other bug: 1) xbmc is an openGL application 2) VPP is done in a different thread. So there is one thread for decoding and another one for doing the post processing. here is the trace file: http://paste.ubuntu.com/7599764/ after havin gdone some fixes to my code I created a new trace file: http://paste.ubuntu.com/7606829/ Apart from the filter_flags it looks similar to the one I created with gst. sometimes the trace looks ugly, maybe a lock missing somewhere? [54772.996222] element[28] = [54772.996222] --VASliceParameterBufferType [54772.996225] element[29] = [54772.996225] --VASliceParameterBufferType [54772.996149] [54772.996227] element[30] = [54772.996228] x = 0 --VASliceParameterBufferType [54772.996231] element[31] = [54772.996232] --VASliceParameterBufferType [54772.996234] element[32] = [54772.996235] --VASliceParameterBufferType [54772.996237] element[33] = [54772.996238] --VASliceParameterBufferType [54772.996229] [54772.996239] y = 0 [54772.996242] width = 720 element[34] = [54772.996245] --VASliceParameterBufferType [54772.996250] element[35] = [54772.996250] --VASliceParameterBufferType [54772.996255] --------------------------- [54772.996243] [54772.996257] height = 576 buffers[1] = 0x08000003 [54772.996259] output_background_color = 0xff000000 [54772.996260] [54772.996260] output_color_standard = 0 type = VASliceDataBufferType [54772.996262] [54772.996262] size = 17062 [54772.996263] pipeline_flags = 0x00000000 num_elements = 1 [54772.996265] filter_flags = 0x00000001 This seems to be fixed with Gwenole's vpp fixes https://github.com/gbeauchesne/libva-intel-driver/tree/17.vpp.vebox Can you reproduce this issue with the latest driver? I remember all patches have been merged into master branch. fixed in latest driver Thanks for your confirmation. |
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.