1909812 - Codepen demo (https://codepen.io/cantelope/pen/QWXKgzN) is slower in Nightly compared to Chrome (some sort of memory allocation/deallocation issue?)

Status: NEW

(Core :: JavaScript Engine, task, P3)

Description

[Mayank Bansal]

Go to https://codepen.io/cantelope/pen/QWXKgzN

Nightly: https://share.firefox.dev/4bSMxlC
Chrome: https://share.firefox.dev/3LIhaPO

Nightly is visually much slower compared to Chrome

Comment 1

[Mayank Bansal]

I am curious about this testcase and the profile - I have not previously seen the continuous memory stuff in the Taskcontroller thread.
Jan, would you mind taking a quick look please?

Comment 2

[Jan de Mooij [:jandem]]

(In reply to Mayank Bansal from comment #1)

I am curious about this testcase and the profile - I have not previously seen the continuous memory stuff in the Taskcontroller thread.
Jan, would you mind taking a quick look please?

There are a lot of memory allocations on the main thread for dense elements and that likely causes the memory freeing that's happening on the TaskController background thread. Each small spike that you see for the Taskcontroller thread corresponds with a GCMinor entry on the marker chart.

Jon is looking into changing how we do these memory allocations and that might improve this test case.

Comment 3

[Mayank Bansal]

Profile with latest Nightly: https://share.firefox.dev/409mjXZ

Comment 4

[Jon Coppeard (:jonco)]

(In reply to Mayank Bansal from comment #3)
This profile shows the proportion of time spent freeing memory on the background thread has been reduced from 30% to 12% of the total time.

This is due to bug 1911537.

Comment 5

[Mayank Bansal]

Latest profile: https://share.firefox.dev/4jj5cvM