1198656 - Avoid sometimes unnecessary initial AudioBuffer allocations

Status: RESOLVED FIXED

(Core :: Web Audio, defect)

Description

[Karl Tomlinson (:karlt)]

These changes save 50% on the "Empty testcase" benchmark when increasing
buffer period to 2000s for semi-repeatable results.
The changes appear to remove most of the variability in this test's results. 
https://treeherder.mozilla.org/#/jobs?repo=try&revision=304a78625ab2

Comment 1

[Karl Tomlinson (:karlt)]

Bug 1199559 tracked the buffer copy removal in the changes in comment 0.

Comment 2

[Karl Tomlinson (:karlt)]

Attachment: MozReview Request: bug 1198656 remove unnecessary reinterpret_casts r?padenot

bug 1198656 remove unnecessary reinterpret_casts r?padenot

JS_StealArrayBufferContents() returns void*.

Comment 3

[Karl Tomlinson (:karlt)]

Attachment: MozReview Request: bug 1198656 interpret null ConvolverNode mBuffer as a buffer of zeros r?padenot

bug 1198656 interpret null ConvolverNode mBuffer as a buffer of zeros r?padenot

Zero output even when normalize is set is consistent with
http://webaudio.github.io/web-audio-api/#widl-ConvolverNode-normalize
due to the specified isinf() test.

Comment 4

[Karl Tomlinson (:karlt)]

Attachment: MozReview Request: bug 1198656 refactor acquiring the content into an object method r?

bug 1198656 refactor acquiring the content into an object method r?

This makes it clearer that the algorithm is intentionally aborted when any of
the buffers have been neutered and that the stolen data has the correct
length.

It also makes mJSChannels available for clearing in a subsequent changeset.

Comment 5

[Karl Tomlinson (:karlt)]

Attachment: MozReview Request: bug 1198656 clear references in mJSChannels on successful content acquire r?

bug 1198656 clear references in mJSChannels on successful content acquire r?

The array buffers are no longer available and mJSChannels will be overwritten
in RestoreJSChannelData() before it is used again.  This is consistent with
"Attach ArrayBuffers containing copies of the data to the AudioBuffer, to be
returned by the next call to getChannelData."

Comment 6

[Karl Tomlinson (:karlt)]

Attachment: MozReview Request: bug 1198656 delay AudioBuffer allocation until required r?padenot

bug 1198656 delay AudioBuffer allocation until required r?padenot

This saves an allocation and zeroing for buffers generated by AudioNodes and
avoids allocation altogether for empty buffers.

Incidentally, RestoreJSChannelData() now avoids unnecessary recreation of
Float32Arrays if they already exist after a previous call failed.

Comment 7

[Paul Adenot (:padenot)]

Comment on attachment 8657093 [details]
MozReview Request: bug 1198656 remove unnecessary reinterpret_casts r?padenot

https://reviewboard.mozilla.org/r/18303/#review16495

Comment 8

[Paul Adenot (:padenot)]

Comment on attachment 8657094 [details]
MozReview Request: bug 1198656 interpret null ConvolverNode mBuffer as a buffer of zeros r?padenot

https://reviewboard.mozilla.org/r/18305/#review16497

Comment 9

[Paul Adenot (:padenot)]

Comment on attachment 8657095 [details]
MozReview Request: bug 1198656 refactor acquiring the content into an object method r?

https://reviewboard.mozilla.org/r/18307/#review16499

Comment 10

[Paul Adenot (:padenot)]

Comment on attachment 8657096 [details]
MozReview Request: bug 1198656 clear references in mJSChannels on successful content acquire r?

https://reviewboard.mozilla.org/r/18309/#review16501

Comment 11

[Paul Adenot (:padenot)]

Comment on attachment 8657097 [details]
MozReview Request: bug 1198656 delay AudioBuffer allocation until required r?padenot

https://reviewboard.mozilla.org/r/18311/#review16503

Comment 12

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/2b9a5c8a0a4a
https://hg.mozilla.org/integration/mozilla-inbound/rev/b27cf1d88807
https://hg.mozilla.org/integration/mozilla-inbound/rev/5376e4042cf3
https://hg.mozilla.org/integration/mozilla-inbound/rev/d325ca7fa94c
https://hg.mozilla.org/integration/mozilla-inbound/rev/e4fb01e4e3cd

Comment 13

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/2b9a5c8a0a4a
https://hg.mozilla.org/mozilla-central/rev/b27cf1d88807
https://hg.mozilla.org/mozilla-central/rev/5376e4042cf3
https://hg.mozilla.org/mozilla-central/rev/d325ca7fa94c
https://hg.mozilla.org/mozilla-central/rev/e4fb01e4e3cd

Comment 14

[Karl Tomlinson (:karlt)]

Could add a test for ConvolverNode output with null buffer.