1982758 - Testcase using 1 worker to decode 1 png N times in a loop decodes thread on mainthread and uses shmem to pass data between threads

Status: NEW

(Core :: Graphics: Canvas2D, defect)

Description

[Mayank Bansal]

Attachment: image-decode-using-workers-OMT.html

(splitting the image decode bits from bug 1982510)

N=50000
Firefox: https://share.firefox.dev/4fy6em5
Chrome: https://share.firefox.dev/45Mc4wW

Comment 1

[Mayank Bansal]

Attachment: blob-image-decode-driven-by-MT.html

N=50000
https://share.firefox.dev/3HcA6Ia

Comment 2

[Timothy Nikkel (:tnikkel)]

I'm not sure I parse the bug title. Can you write out in a comment using a few more words what the issue you are filing this for? Thanks

Comment 3

[Mayank Bansal]

I created this artificial testcase that generates a PNG and then uses workers to decode it.
Markus pointed out on matrix group that "looks like the decoding isn't happening in the worker" and "converting the blob to an imagebitmap does the CPU-intensive work on the main thread, and the fact that it's using a shmem to pass the data between threads".
I have filed this bug to track this.

Comment 4

[Erich Gubler [:ErichDonGubler] (he/him)]

This seems like an unfortunate breaking of expectation of what workloads run on specific threads, but I'm not sure how much applications will suffer if they run into this. Triaging as S3 for now, and NI'ing Lee.