1962838 - Testcase embedding N js modules via blobs is 2.2x slower in Firefox compared to Chrome (N=500K)

Status: NEW

(Core :: JavaScript Engine, task, P3)

Description

[Mayank Bansal]

Attachment: js modules with blob.HTML

js modules in blobs

N=500000

• Firefox: https://share.firefox.dev/4cLFPzQ (38s)
• Chrome: https://share.firefox.dev/4iwb6rJ (18s)

js modules

N=500000

• Chrome: https://share.firefox.dev/42zrPG8 (13s)
• Firefox: https://share.firefox.dev/42PsPEX (14s)

Comment 1

[Mayank Bansal]

Attachment: js modules.HTML

Comment 2

[Jon Coppeard (:jonco)]

The module refactoring in bug 1820594 needs to happen before we can work on optimizing module loading and should unlock some potential improvements.

Comment 3

[Mayank Bansal]

js modules in blob
500000 loopcount
Firefox: https://share.firefox.dev/4p03v8A (30 seconds)
Chrome: https://share.firefox.dev/3JTvmID (15 seconds)
Firefox gecko profiler (500000): https://share.firefox.dev/43n1Xgy

Bug 1820594 is fixed now. Can/should this bug be reassessed?

Comment 4

[Jon Coppeard (:jonco)]

I'm seeing 16% of time in nsCSPContext::ReportInlineViolation and 7% of time in mozilla::dom::CollectScriptTelemetry locally.

The first is a bit strange. Is the test case working as expected?

Comment 5

[Mayank Bansal]

I may have run the test as a BMO attachment, i dont remember. BMO has weird CSP restrictions.

Here is a profile where i run the testcase from a local HTTP server: https://share.firefox.dev/3WUGcRx