commit 7f876c8930a1732c898b897739455a95424dc976
author Emil Lundmark <lndmrk@webrtc.org> Fri Sep 10 10:36:31 2021
committer WebRTC LUCI CQ <webrtc-scoped@luci-project-accounts.iam.gserviceaccount.com> Mon Sep 13 15:52:42 2021
tree 82fcf31478892cf7d7818e564af123930396b885
parent 9103e8efaf4f4ec825e1783f28bf319382b32dd4

Allow full SVC to reference T0 frame only after it has been encoded

The VP9 encoder may drop a frame internally which will not advance the
frame pattern. Consider the following scenario where only spatial layer
0 and temporal layer 0 is active:

1. Key frame encoded
2. Spatial layer 1 is activated
3. Delta T0 dropped
4. Delta T0 encoded

No S1T0 frame is encoded in (1) since it's not active. When
NextFrameConfig is called in (3) it will say that future frames may
reference T0 on both S0 and S1, but it's then dropped.

On step (4), the SVC controller essentially thinks it's encoding a new
picture and will happily reference the T0 on what it thinks is the first
delta frame. However, this is actually still the key frame and since
there was no S1T0 frame produced it will reference an invalid buffer.

To fix this, only say it's possible to reference a T0 frame after it has
been successfully encoded.

Bug: webrtc:11999, webrtc:13142, chromium:1178444
Change-Id: Iab3d2042ce0b3fa7d952b2831d1a36b1a6613a86
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/231695
Reviewed-by: Erik Språng <sprang@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Commit-Queue: Emil Lundmark <lndmrk@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#34982}