Branch: refs/heads/main
Home: https://github.com/WebKit/WebKit
Commit: 66035f0df72081e8a186bbcfe25042046f586069
https://github.com/WebKit/WebKit/commit/66035f0df72081e8a186bbcfe25042046f586069
Author: Geoffrey Garen <[email protected]>
Date: 2025-11-18 (Tue, 18 Nov 2025)
Changed paths:
M Source/WebGPU/WebGPU/BindGroup.mm
M Source/WebGPU/WebGPU/Sampler.h
M Source/WebGPU/WebGPU/Sampler.mm
Log Message:
-----------
[WebGPU] Clean up MTLSamplerState cache logic
https://bugs.webkit.org/show_bug.cgi?id=302729
rdar://164991782
Reviewed by Mike Wyrzykowski.
This patch does two things:
1. Remove the reinterpret_cast<uintptr_t> I just added for SaferCPP. Pointer =>
integer cast is technically memory-safe so long as you don't cast back. But it's
still sketchy. This patch replaces explicit pointer tracking with a use count.
2. Fix the cache eviction logic regression in 294328@main. I realized that the
reinterpret_cast logic was a no-op, so I considered just removing it, but Mike
and I discovered that the no-op was a regression. So let's fix it instead.
In extreme cases, we want to evict live entries from the cache and regenerate
them on demand. For example, if a program is churning Sampler allocations, and
GC is not keeping up -- similar to a problem we sometimes have with <canvas> --
we want to reclaim live MTLSamplers instead of just failing. We model this using
a strong + weak reference in the cache. We use our weak reference to detect
whether reclamation succeeds when we clear our strong reference.
Canonical link: https://commits.webkit.org/303218@main
To unsubscribe from these emails, change your notification settings at
https://github.com/WebKit/WebKit/settings/notifications
