Replies: 1 comment
-
Closing this because this was a bug report that someone changed into discussion. We have users complaining pipeline switches are very slow, making our library unusable in production, and we can't group calls and avoid pipeline switches because |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Hi,
Some time ago I asked about the validation step which was auto-injected by
wgpu
(for the sake of browsers security or stability if I recall) before each indirect dispatch, and was messing with bindings (lots of unnecessary bind calls).As I recall I was told this would be made optional in a later version (it's mandatory in 24.x). Reading the latest docs in 26.x, I think this is what
InstanceFlag::VALIDATION_INDIRECT_CALL
corresponds to?The problem is that there's a section in that doc saying that on the D3D12 backend, disabling this flag makes
@builtin(instance_index)
not work. I checked the WGSL spec, I didn't see anything that would allow a compliant implementation to choose not to takefirst_instance
into account.So my understanding is that you have to choose between enabling a debug feature which adds extra dispatches and unnecessary bindings (and so impacts performance), and being able to use the
first_instance
feature (or re-implementing it yourself). Is that correct, or am I misunderstanding something? Can someone explain what's going on here and why only D3D12 has this issue?Thanks!
Beta Was this translation helpful? Give feedback.
All reactions