1851755 - Reduce number of expf() calls for setTargetAtTime()

Status: RESOLVED FIXED

(Core :: Web Audio, enhancement)

Description

[Karl Tomlinson (:karlt)]

Bug 1358149 comment 55 indicates that expf() evaluation is having a significant effect on the "Synth" and "Substractive synth" performance benchmarks, which use setTargetAtTime().

Using a linear recurrence relation to skip most of the expf() calls improves performance on "Substractive synth" by 13-23% and on "Synth" by 10-19% .

Comment 1

[Karl Tomlinson (:karlt)]

On top of other recent performance improvements, this now reduces time by 25 - 30% on "Substractive synth".

Comment 2

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 remove unused AudioEventTime methods r?padenot

AudioParam uses ValidateEvent() and InsertEvent().

Comment 3

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 use separate constructor for SetValueCurve AudioTimelineEvent r?padenot

To simplify parameter sets and clarify which parameters may be in use.

Depends on D189463

Comment 4

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 clarify that AudioParam tracks are handled by AudioParamTimeline not AudioEventTimeline r?padenot

Depends on D189464

Comment 5

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 move ConvertAudioTimelineEventToTicks() to AudioTimelineEvent r?padenot

This is a more natural place to manage the AudioTimelineEvent members.
It will also be a better place for pre-computing a constant exponential for
SetTarget curves in a subsequent patch.

Depends on D189465

Comment 6

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 add getters to AudioTimelineEvent::Value() to assert that members are not accessed for the wrong event types r?padenot

Depends on D189466

Comment 7

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 use a union for more AudioTimelineEvent members r?padenot

where SetValueCurve requires different parameters.

These objects are copied when inserted into and moved within the
AudioEventTimeline array.

Depends on D189467

Comment 8

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 use assert_approx_equals for comparison with tolerance r?padenot

This changes failure messages from
assert_true: sample 1471 should equal 0.24924172461032867 expected true got false
to
assert_approx_equals: sample 1471 expected 0.24924172461032867 +/- 0.000001342 but got 0.24924308061599731

Depends on D189468

Comment 9

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 pre-compute constant exponential for inter-tick ratio in setTargetAtTime() curves r?padenot

Depends on D189469

Comment 10

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 clarify that mCurve and mCurveLength are set in the AudioTimelineEvent constructor r?padenot

addressing a cppcoreguidelines-pro-type-member-init warning.

Depends on D189468

Comment 11

[Karl Tomlinson (:karlt)]

Attachment: Bug 1851755 suppress cppcoreguidelines-pro-type-member-init warning for AudioEventTimeline copy constructor r?padenot

Depends on D189628

Comment 12

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/648bf0239d99
remove unused AudioEventTime methods r=padenot
https://hg.mozilla.org/integration/autoland/rev/7139b608a37e
use separate constructor for SetValueCurve AudioTimelineEvent r=padenot

Comment 13

[Cristian Tuns]

https://hg.mozilla.org/mozilla-central/rev/648bf0239d99
https://hg.mozilla.org/mozilla-central/rev/7139b608a37e

Comment 14

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/4f5e812e3ab7
clarify that AudioParam tracks are handled by AudioParamTimeline not AudioEventTimeline r=padenot

Comment 15

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/f1f858a5c8c2
move ConvertAudioTimelineEventToTicks() to AudioTimelineEvent r=padenot

Comment 16

[Natalia Csoregi [:nataliaCs]]

https://hg.mozilla.org/mozilla-central/rev/4f5e812e3ab7

Comment 17

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/d7b66b8f2e34
add getters to AudioTimelineEvent::Value() to assert that members are not accessed for the wrong event types r=padenot

Comment 18

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/19c7e9c983a2
use a union for more AudioTimelineEvent members r=padenot
https://hg.mozilla.org/integration/autoland/rev/b781e655236d
use assert_approx_equals for comparison with tolerance r=padenot

Comment 19

[Web Platform Test Sync Bot [:wpt-sync] (Matrix: #interop:mozilla.org)]

Created web-platform-tests PR https://github.com/web-platform-tests/wpt/pull/42267 for changes under testing/web-platform/tests

Comment 20

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/66bd7c78781a
clarify that mCurve and mCurveLength are set in the AudioTimelineEvent constructor r=padenot
https://hg.mozilla.org/integration/autoland/rev/f90a7a1b6c24
suppress cppcoreguidelines-pro-type-member-init warning for AudioEventTimeline copy constructor r=padenot

Comment 21

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/1fd044c18e14
pre-compute constant exponential for inter-tick ratio in setTargetAtTime() curves r=padenot

Comment 22

[Sandor Molnar[:smolnar]]

https://hg.mozilla.org/mozilla-central/rev/f1f858a5c8c2
https://hg.mozilla.org/mozilla-central/rev/d7b66b8f2e34
https://hg.mozilla.org/mozilla-central/rev/19c7e9c983a2
https://hg.mozilla.org/mozilla-central/rev/b781e655236d

Comment 23

[Web Platform Test Sync Bot [:wpt-sync] (Matrix: #interop:mozilla.org)]

Upstream PR merged by moz-wptsync-bot

Comment 24

[Serban Stanca [:SerbanS]]

https://hg.mozilla.org/mozilla-central/rev/66bd7c78781a
https://hg.mozilla.org/mozilla-central/rev/f90a7a1b6c24
https://hg.mozilla.org/mozilla-central/rev/1fd044c18e14