Conversation
Fixes a number of things including a LPMS crash, choppy video quality, green screens during rotation, inconsistent frame counts vs software decoding, etc. This is a massive change: we can no longer add outputs up front due to the ffmpeg hwaccel API, so we have to wait until we receive a decoded video frame in order to add outputs. This also means properly queuing up audio and draining things in the same order.
leszko
left a comment
leszko
left a comment
There was a problem hiding this comment.
I skimmed through the PR and looks good. Saying that, I don't have enough LPMS knowledge to review it in depth, so if you need a "better" review, then you can wait for approves from some other folks.
ffmpeg/nvidia_test.go
Outdated
| fname := dir + "/mpeg2.ts" | ||
| oname := dir + "/out.ts" | ||
| fname := dir + "/test.flv" | ||
| //oname := dir + "/out.ts" |
There was a problem hiding this comment.
Nit: Remove commented-out code
There was a problem hiding this comment.
Thanks for the spot, fixed and cleaned up in 9e51cee
| tc := NewTranscoder() | ||
| idx := []int{1, 0, 3, 2} | ||
| for _, i := range idx { | ||
| // TODO this breaks with nvidia acceleration on the input! |
There was a problem hiding this comment.
Note that this is not a new problem with this PR; the old code had the same issue
Maybe @emranemran you could take a look at the C code? Can't remember if you've reviewed much of lpms in the past.. |
mjh1
left a comment
mjh1
left a comment
There was a problem hiding this comment.
Similar to Rafal I'm afraid I don't know enough about this stuff, maybe we could do a knowledge transfer call or something?
3 participants
Fixes a number of things with the nvidia built-in parser (cuvid) including a LPMS crash, choppy video quality, green screen flashing during rotation, inconsistent frame counts vs software decoding, etc. We also apparently gained GPU support for MPEG2 decoding.
This is a massive change: we can no longer add outputs up front due to the ffmpeg hwaccel API, so we have to wait until we receive a decoded video frame in order to add outputs. This also means properly queuing up audio and draining things in the same order.
This also unfortunately adds ~13 MB of new samples to the repo for test cases, since these are hard to generate synthetically.