[d3d12/vk] Implement out of memory detection #7472
Open
+606
−42
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
teoxoy
commented
Apr 3, 2025
| @@ -2435,6 +2519,10 @@ impl crate::Device for super::Device { | |||
| &self, | |||
| desc: &wgt::QuerySetDescriptor<crate::Label>, | |||
| ) -> Result<super::QuerySet, crate::DeviceError> { | |||
| // Assume each query is 256 bytes. | |||
| // On an AMD W6800 with driver version 32.0.12030.9, occlusion queries are 256. | |||
| self.check_for_oom(true, desc.count as u64 * 256)?; | |||
There was a problem hiding this comment.
I'm not 100% sure about this, there doesn't seem to be any indication in the Vulkan API that query sets reside in a host visible heap (even if this is the case on the AMD card I tested).
There was a problem hiding this comment.
I was actually surprised they are in the host visible heap rather than the device local one (as they are on D3D12).
teoxoy
force-pushed
the
oom-detection
branch
2 times, most recently
from
4b29a2d to
7dd2aa9
Compare
…and acceleration structures The D3D12 API doesn't guarantee that it returns `E_OUTOFMEMORY` in high memory pressure situations; drivers/kernel will happily start swapping objects that were in VRAM to RAM and then RAM to DISK, slowing down the system to a crawl if done in a loop.
…re, query set and acceleration structure creation
… if we are over 95% of our budget
This removes the possibility of deadlocks happening since `release_gpu_resources` tries to lock resources (trackers, snatchable_lock, pending_writes, life_tracker) while they might be already locked; `handle_hal_error` is called in lots of places. Removing the call only delays destruction since `release_gpu_resources` is still called in `maintain`.
This is to preserve the current behavior as tested by the `SAMPLER_CREATION_FAILURE` test. This is not spec compliant but it's unclear what we should do instead. I opened gpuweb/gpuweb#5142 to figure out what we should do.
… acceleration structures
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Connections
Resolves D3D12 and Vulkan parts of #7460.
Description
Implements OOM detection. See #7460 for details.
Testing
I added a new crate
oom-testfor this.Squash or Rebase?
Rebase