73% tenuring rate on modified Codepen Demo (https://codepen.io/Mikhail-Bespalov/pen/dPyyZed?editors=1010 ), and Firefox uses 800MB-1.2GB RAM. Chrome uses 200MB
Categories
(Core :: JavaScript: GC, task, P2)
Tracking
()
People
(Reporter: mayankleoboy1, Unassigned)
References
(Blocks 1 open bug, )
Details
Attachments
(1 file)
|
19.95 KB,
text/html
|
Details |
Open the attached testcase (original at https://codepen.io/Mikhail-Bespalov/pen/dPyyZed?editors=1010. I changed the code so that the minimum grid size is 1)
Slide the grid-size slider all the way to the left ("1" value)
AR: https://share.firefox.dev/41dekuQ
We use 800MB-1.2gb RAM. Chrome takes 200MB.
| Reporter | ||
Comment 1•9 months ago
|
||
Comment 2•9 months ago
|
||
It looks like we have a huge number of very small nursery GCs that look like this:
Reason: NURSERY_MALLOC_BUFFERS
Bytes tenured: 192B / 280B (69%)
Cells tenured: 3.0 / 5.0 (60%)
Bytes used: 280B / 256KB (0.1%)
Nursery allocations since last minor GC: 5.0 / 9.0 (56%)
Looking at the code, I think we're allocating a big Float32Array in generateHalftone every frame, so we're hitting our mallocedBufferBytes capacity very quickly.
Updated•9 months ago
|
Comment 3•9 months ago
|
||
We don't resize the nursery based on malloc bytes usage in the same way as we do for space used in the nursery itself. This would likely help here.
Updated•9 months ago
|
| Reporter | ||
Comment 4•3 months ago
|
||
| Reporter | ||
Comment 5•3 months ago
|
||
(In reply to Jon Coppeard (:jonco) from comment #3)
We don't resize the nursery based on malloc bytes usage in the same way as we do for space used in the nursery itself. This would likely help here.
Worth filing a bug for posterity?
Description
•