vulkan: Hybrid waitForFences/getFenceStatus to reduce fence latency #12630
Conversation
This file contains hidden or 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
github-actions
bot
added
Vulkan
jeffbolznv
commented
Mar 28, 2025
ggml/src/ggml-vulkan/ggml-vulkan.cpp
Outdated
| exit(1); | ||
| } | ||
| for (uint32_t i = 0; i < 100; ++i) { | ||
| _mm_pause(); |
Collaborator
Author
There was a problem hiding this comment.
I didn't expect this to pass CI, but I guess we only build Vulkan for x64 in CI. I'll generalize this, but it doesn't block perf testing.
There seems to be a bubble waking up from waitForFences, which costs a few percent performance and also increased variance in performance. This change inserts an "almost_ready" fence when the graph is about 80% complete and we waitForFences for the almost_ready fence and then spin (with _mm_pauses) waiting for the final fence to be signaled.
Collaborator
|
I ran some quick tests on this using my usual setup and performance is pretty much the same as master. Maybe there's a <1% difference but I can't tell. |
0cc4m
approved these changes
Apr 2, 2025
Collaborator
0cc4m
left a comment
0cc4m
left a comment
There was a problem hiding this comment.
I see a small positive effect on all vendors, LGTM
timwu
pushed a commit
to timwu/llama.cpp
that referenced
this pull request
May 5, 2025
…gml-org#12630) There seems to be a bubble waking up from waitForFences, which costs a few percent performance and also increased variance in performance. This change inserts an "almost_ready" fence when the graph is about 80% complete and we waitForFences for the almost_ready fence and then spin (with _mm_pauses) waiting for the final fence to be signaled.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
There seems to be a bubble waking up from waitForFences, which costs a few percent performance and also increases variance in performance. This change inserts an "almost_ready" fence when the graph is about 80% complete and we waitForFences for the almost_ready fence and then spin (with _mm_pauses) waiting for the final fence to be signaled.
I've seen up to 5% performance improvement from this on NVIDIA/Windows. I'm curious whether it helps on other IHV/OS combinations as well. Yesterday I did some power testing using the NZXT CAM software to measure the CPU power, and this hybrid spin/wait seemed pretty good. Today, my CPU power is higher even at idle and tokens/s is lower (though still an improvement over master), regardless of the spin/wait behavior, 🤷 computers are weird.