@gfxstrand@social.treehouse.systems
Post #2290671
2026-05-08 06:12 UTC
@dotstdy@mastodon.social Oh, that’s just because AMD’s caching situation in bonkers.
But also, marking stuff appropriately RO will probably help NVIDIA, too, because it’ll enable their constant caching mode which is a lot faster. (Though it’s less bonkers than AMD’s scalar cache. 😂)
Replies (1)
-
@dotstdy@mastodon.social 2026-05-08 06:18
@gfxstrand@social.treehouse.systems Yeah the caching is bonkers, I also suppose it's a lot about whether making the analysis more complex is profitable in real shaders. Like if there's no potentially aliasing stores in a shader then you could theoretically promote the bda loads to NoAlias even if they're not marked as such. But you could also just argue that whatever generates the spir-v could do this itself and save everyone the trouble. (glsl has a pass which does this, but spir-v is a bit more "you break, you buy")