1262765 - Remove 16 bits fixed point integer sample support

Status: NEW

(Core :: Audio/Video: Playback, defect, P3)

Description

[Jean-Yves Avenard [:jya]]

The only place we are using S16 audio sample format is on Android/B2G.

That means that in various places we have two maintain two different code paths.

On modern ARM architecture, the speed gain of integers over floats have greatly reduced and that advantage keeps narrowing.

There's a few operations where we use fixed 16 bits arithmetic (such as downmixing and resampling) which ought to be noisy with such low precision.

We should stop using S16 samples on android platform and use floats.

Comment 1

[Jean-Yves Avenard [:jya]]

Eugen, can the android decode API use float?

or does that mean we have to convert the S16 to float and back to S16 before playing them?

Comment 2

[Eugen Sawin [:esawin]]

MediaCodec supports single-precision floating-point samples starting with API level 21 (Android 5) and most of our code implicitly or explicitly assumes 16 bit PCM.

We could provide an API level-based conversion fallback, assuming it would perform well on older/low-spec devices.

Comment 3

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Mass change P2 -> P3