1783181 - Regression in StereoPannerNode on M1 Macs

Status: VERIFIED FIXED

(Core :: Web Audio, defect, P1)

Description

[chrisguttandin]

Steps to reproduce:

I ran the following script:

const length = 5;
const sampleRate = 44100;
const buffer = new AudioBuffer({ length, numberOfChannels: 2, sampleRate });

buffer.copyToChannel(new Float32Array([1, 0.5, 0, -0.5, -1]), 0);
buffer.copyToChannel(new Float32Array([-0.5, -0.25, 0, 0.25, 0.5]), 1);

const offlineAudioContext = new OfflineAudioContext({ length, sampleRate });
const audioBufferSourceNode = new AudioBufferSourceNode(offlineAudioContext, { buffer });
const stereoPannerNode = new StereoPannerNode(offlineAudioContext, { channelCount: 2 });

audioBufferSourceNode.connect(stereoPannerNode).connect(offlineAudioContext.destination);

stereoPannerNode.pan.setValueAtTime(0.5, 0);
stereoPannerNode.pan.setValueAtTime(0, 2 / offlineAudioContext.sampleRate);
stereoPannerNode.pan.linearRampToValueAtTime(1, 5 / offlineAudioContext.sampleRate);
stereoPannerNode.pan.cancelScheduledValues(3 / offlineAudioContext.sampleRate);

audioBufferSourceNode.start(0);

offlineAudioContext.startRendering().then((renderedBuffer) => {
const channelData = renderedBuffer.getChannelData(0);

console.log(Array.from(channelData));

});

Actual results:

It logs [ 0.4571067690849304, 0.2285533845424652, 0, -0.125, -0.0732232928276062 ] when running it on an M1 Mac. The last number seems to be wrong.

Expected results:

It should log [ 0.4571067690849304, 0.2285533845424652, 0, -0.125, -0.25].

It does work on an older Intel Mac in Firefox v103. It also works in Chrome or Safari on any Apple processor architecture.

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

The Bugbug bot thinks this bug should belong to the 'Core::JavaScript Engine' component, and is moving the bug to that component. Please correct in case you think the bot is wrong.

Comment 2

[Paul Adenot (:padenot)]

Attachment: panner-m1.html

I can repro. The testcase in comment 0 is now attached.

Comment 3

[Paul Adenot (:padenot)]

Attachment: Bug 1783181 - Use correct scaling vector when panning the left channel values using NEON intrinsics. r?karlt

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/1facaa986f25
Use correct scaling vector when panning the left channel values using NEON intrinsics. r=karlt

Comment 5

[Norisz Fay [:noriszfay]]

https://hg.mozilla.org/mozilla-central/rev/1facaa986f25

Comment 6

[Alexandru Trif, Desktop Test Engineering [:atrif]]

Reproduced the issue with Firefox 103.0.2 (20220808125904) on macOS 11 mac M1 mini. After loading the test case the console displayed the following results: Array(5) [ 0.4571067690849304, 0.2285533845424652, 0, -0.125, -0.0732232928276062 ]
The issue is verified fixed with Firefox 105.0b8 on macOS11 mac M1 mini. The console displays the correct results as stated in comment 0: Array(5) [ 0.4571067690849304, 0.2285533845424652, 0, -0.125, -0.25 ]