add av1parse element for low rank GStreamer av1 decoders

[mdimopoulos]

• affects av1dec, dav1ddec and vaapiav1dec
• workaround for typefind GStreamer issue when decoder has low rank

[mdimopoulos]

@ndufresne this should solve the issues happening for av1 decoders with low rank.
I didn't touch the ones that have primary or above

[ndufresne]

I'd say go for it, its a safe workaround to the negotiation issue. What we could perhaps do is link to the upstream issue, so this tiny hack have a context: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/4578

[mdimopoulos]

fyi @ndufresne
Initial solution didn't work, parsebin ! av1parse still results in failures due to improper negotiation. For example with some AV1 vectors h263 parser is plugged-in and the pipeline of course fails.

I have now removed parsebin element from 3 relevant AV1 decoder pipelines. I tested it and it works fine

[ndufresne]

That will break the AOM and Chromium tests that requires matroska demuxing.

[mdimopoulos]

That will break the AOM and Chromium tests that requires matroska demuxing.

I executed a quick run with the test suites i think you are referring to and everything seems to be working perfectly.

Generating summary for test suite AV1-TEST-VECTORS and decoders GStreamer-AV1-libaom-Gst1.0:
|Test|GStreamer-AV1-libaom-Gst1.0|
|-|-|
|TOTAL|242/242|
Generating summary for test suite CHROMIUM-8bit-AV1-TEST-VECTORS and decoders GStreamer-AV1-libaom-Gst1.0:
|Test|GStreamer-AV1-libaom-Gst1.0|
|-|-|
|TOTAL|13/13|
Generating summary for test suite CHROMIUM-10bit-AV1-TEST-VECTORS and decoders GStreamer-AV1-libaom-Gst1.0:
|-|-|
|Test|GStreamer-AV1-libaom-Gst1.0|
|TOTAL|23/23|

Do you refer to other test suites or decoders 🤔? @ndufresne

[mdimopoulos]

@ndufresne based on the last test results i believe we are good. what do you think?

[ndufresne]

Didn't have time to test, but I'm very suspicious of your change, perhaps you didn't test exactly what you wrote here. There is IVF and MKV files, and to read those, you need to plug ivfparse/matroskademux, if you drop parsebin and only place av1parse, it cannot work.

[mdimopoulos]

Didn't have time to test, but I'm very suspicious of your change, perhaps you didn't test exactly what you wrote here. There is IVF and MKV files, and to read those, you need to plug ivfparse/matroskademux, if you drop parsebin and only place av1parse, it cannot work.

I see, i will take another look just in case.

[mdimopoulos]

Didn't have time to test, but I'm very suspicious of your change, perhaps you didn't test exactly what you wrote here. There is IVF and MKV files, and to read those, you need to plug ivfparse/matroskademux, if you drop parsebin and only place av1parse, it cannot work.

I see, i will take another look just in case.

It's good you were suspicious, myself was a bit surprised to see 100% success but didn't think much of it.
After a closer look, AV1-TEST-VECTORS is failing miserably. I am looking for solutions.
Thanks again @ndufresne

[mdimopoulos]

I had to go back to the initial solution parsebin ! av1parse, it seems to be the best tradeoff for now. Below the test pass count to show the point
GStreamer 1.26.2 and av1dec

Test Suite	parsebin	av1parse	parsebin ! av1parse
AV1-ARGON-PROFILE0-CORE-ANNEX-B	0	0	27
AV1-ARGON-PROFILE0-NON-ANNEX-B	35	35	36
AV1-TEST-VECTORS	242	0	242

[ndufresne]

Great, note that typefind is heuristic base, I can increase the AV1 parser confidence so that it overrule the broken cases, but to do that I'll need to get a list of miss-identification.

[mdimopoulos]

Great, note that typefind is heuristic base, I can increase the AV1 parser confidence so that it overrule the broken cases, but to do that I'll need to get a list of miss-identification.

Ok. What would be of help to you, debug logs, the name of the parser element that is plugged in by parsebin, something else?

[ndufresne]

Some help sorting out the errors that are due to miss-identication is always welcome. There is no magic in fixing it, I need to pick one, and figure-out if improving the typefinding that miss-identified it to now reject it, or changing around some typefinder "confidence" level.

[rsanchez87]

LGTM, @mdimopoulos, a rebase is necessary before merging to master