1229475 - Update libopus to 1.1.1

Status: RESOLVED FIXED

(Core :: WebRTC: Audio/Video, defect, P1)

Description

[NVD]

https://www.opus-codec.org/

"This Opus 1.1.1 release brings many optimizations to the encoder and decoder, including:

    x86 SSE, SSE2 and SSE4.1 intrinsics optimizations with run-time CPU detection contributed by Cisco Systems,
    MIPS intrinsics optimizations contributed by Imagination Technologies,
    ARM Neon optimizations contributed by Linaro and ARM,
    many architecture-independent optimizations and memory footprint reductions that should improve performance on all platforms, and
    several minor bug fixes.

The quality of the encoder should be mostly unchanged compared to version 1.1. The code is available from the downloads page."

Comment 1

[Chris Peterson [:cpeterson]]

In Firefox 45 (bug 1229605), jya allowed Opus MSE to be enabled independently of VP9 MSE. YouTube prefers Opus over AAC, so they automatically began serving Opus to Firefox 45+. We are now working with them to A/B test Firefox's Opus vs AAC. They will dial back Opus until we're ready to ship it.

Since libopus 1.1.1 has added many decoder optimizations since 1.1 was release two years ago (!), we should probably update to 1.1.1 before we ask YouTube to serve Opus to Firefox release users. Since they are already serving Opus to Firefox users with libopus 1.1, they will be able to compare user stats for 1.1 vs 1.1.1.

Comment 2

[Chris Peterson [:cpeterson]]

Ralph, can you update libopus? Anthony volunteered you. :)

Comment 3

[Chris Peterson [:cpeterson]]

We should double check whether enabling the new intrinsics in 1.1.1 requires new #defines or Makefile magic be added to moz.build.

Comment 5

[Ralph Giles (:rillian)]

Attachment: Fix gen-sources.py for opus 1.1.1

WIP. This doensn't work, and I won't be able to work on it again until Dec 29.

 If someone wants to rewrite mshal's parser so filtering is easier in the meantime, feel free to take the bug.

Comment 6

[Ralph Giles (:rillian)]

Attachment: Fixup run_analysis calls for webrtc

1.1.1 changed this (private) API. Make it more like a public api.

Comment 7

[Ralph Giles (:rillian)]

Attachment: Part 2 - update to libopus 1.1.1

Comment 8

[Ralph Giles (:rillian)]

Attachment: Part 3 - Patch out arch parameter from run_analysis

Patch out the new arch parameter added to run_analysis() in 1.1.1, and replace it with tonality_analysis_init() which can be called to intern the same value inside the state struct.

This makes it sane for the webrtc code to call in the next patch.

Comment 9

[Ralph Giles (:rillian)]

Attachment: Part 4 - Call tonality_analysis_init() before run_analysis.

Call the new initializer function to set asm capability flags before calling run_analysis() in the webrtc code.

Comment 10

[Ralph Giles (:rillian)]

Attachment: Part 5 - Hack unified build changes back in.

Bug 1146204 put this in without fixing update.sh to maintain it. In the interest of landing the update, this just patches the changes back in, so at least we won't lose them next time.

Comment 11

[Ralph Giles (:rillian)]

Attachment: Part 1 - Fix gen_sources for libopus 1.1.1 v2

Corresponding punt for gen_sources.py.

Comment 12

[Ralph Giles (:rillian)]

Attachment: Part 3 - Patch out run_analysis arch parameter v2

Update run_analysis patch to handle decoder reset properly.

Thanks to Jean-Marc and Mark Harris for review comments. See also https://review.xiph.org/1185/

Comment 13

[Ralph Giles (:rillian)]

Attachment: Part 2 - Update libopus source to 1.1.1 v2

Add new files missing from the previous patch.

Comment 14

[Ralph Giles (:rillian)]

Attachment: Part 3 - Patch out run_analysis arch parameter v3

run_analysis patch backported from the changes which landed upstream.

Comment 15

[Ralph Giles (:rillian)]

Attachment: Part 5 - Hack unified build changed back in v2

Update unified build patch to add two more conflicting files from 1.1.1.

I only found these because of a redefined cpp symbol warning, which left me nervous that there could be other collisions affecting compression. How trustworthy do you think the warning are to catch everything like this?

Comment 16

[Jean-Marc Valin]

Comment on attachment 8703092 [details] [diff] [review]
Part 2 - Update libopus source to 1.1.1 v2

Review of attachment 8703092 [details] [diff] [review]:
-----------------------------------------------------------------

r+ on the assumption that this gigantic patch basically brings libopus to 1.1.1.

Comment 17

[Jean-Marc Valin]

Comment on attachment 8703096 [details] [diff] [review]
Part 3 - Patch out run_analysis arch parameter v3

Review of attachment 8703096 [details] [diff] [review]:
-----------------------------------------------------------------

Why does the patch include a patch in addition to changing the files? Also, are these changes identical to what you just landed in Opus master?

Comment 18

[Ralph Giles (:rillian)]

(In reply to Jean-Marc Valin (:jmspeex) from comment #17)

> Why does the patch include a patch in addition to changing the files?

We use the update.sh script to pull new releases and apply any patches we're carrying relative to official opus repo. Including a copy of the patch makes this reproducible. We'll be able to remove it when we upgrade to a release with the changes included.

> are these changes identical to what you just landed in Opus master?

Yes. This is a squashed backport of the two patches I landed on master which applies to 1.1.1.

Comment 19

[Jean-Marc Valin]

Comment on attachment 8703096 [details] [diff] [review]
Part 3 - Patch out run_analysis arch parameter v3

Review of attachment 8703096 [details] [diff] [review]:
-----------------------------------------------------------------

r+

Comment 20

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a2033de17d55
https://hg.mozilla.org/integration/mozilla-inbound/rev/75fae6fb36ca
https://hg.mozilla.org/integration/mozilla-inbound/rev/6ef8fe75a53e
https://hg.mozilla.org/integration/mozilla-inbound/rev/b6f9c97c2c8d
https://hg.mozilla.org/integration/mozilla-inbound/rev/4b44a3e343c8

Comment 21

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/a2033de17d55
https://hg.mozilla.org/mozilla-central/rev/75fae6fb36ca
https://hg.mozilla.org/mozilla-central/rev/6ef8fe75a53e
https://hg.mozilla.org/mozilla-central/rev/b6f9c97c2c8d
https://hg.mozilla.org/mozilla-central/rev/4b44a3e343c8