vulkan: optimize UMA buffer operations and fix driver hangs#16059
Merged
0cc4m merged 2 commits intoggml-org:masterfrom Sep 21, 2025
Merged
vulkan: optimize UMA buffer operations and fix driver hangs#160590cc4m merged 2 commits intoggml-org:masterfrom
0cc4m merged 2 commits intoggml-org:masterfrom
Conversation
github-actions
Bot
added
Vulkan
0cc4m
reviewed
Sep 18, 2025
Contributor
0cc4m
left a comment
0cc4m
left a comment
There was a problem hiding this comment.
I forgot to implement this for the memset function, thank you. If it would help to be able to do this asynchronously as well, we could instead implement a "deferred_memset" (see deferred_memcpy for reference), but this would require more changes.
Contributor
Author
I've added a new commit to implement |
The previous implementation was blocking the GPU for extended periods, causing the i915 driver to reset the context due to the hangcheck protection. [32628.443070] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in llama-server [194114] [32628.443091] i915 0000:00:02.0: [drm] llama-server[194114] context reset due to GPU hang Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
giuseppe
force-pushed
the
access-directly-mem-with-uma
branch
from
34ba782 to
87d3cd0
Compare
struct
pushed a commit
to struct/llama.cpp
that referenced
this pull request
Sep 26, 2025
…#16059) * vulkan: optimize UMA buffer operations and fix driver hangs The previous implementation was blocking the GPU for extended periods, causing the i915 driver to reset the context due to the hangcheck protection. [32628.443070] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in llama-server [194114] [32628.443091] i915 0000:00:02.0: [drm] llama-server[194114] context reset due to GPU hang * vulkan: implement deferred_memset on UMA --------- Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
pwilkin
pushed a commit
to pwilkin/llama.cpp
that referenced
this pull request
Oct 23, 2025
…#16059) * vulkan: optimize UMA buffer operations and fix driver hangs The previous implementation was blocking the GPU for extended periods, causing the i915 driver to reset the context due to the hangcheck protection. [32628.443070] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in llama-server [194114] [32628.443091] i915 0000:00:02.0: [drm] llama-server[194114] context reset due to GPU hang * vulkan: implement deferred_memset on UMA --------- Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Anico2
added a commit
to Anico2/llama.cpp
that referenced
this pull request
Jan 15, 2026
…#16059) * vulkan: optimize UMA buffer operations and fix driver hangs The previous implementation was blocking the GPU for extended periods, causing the i915 driver to reset the context due to the hangcheck protection. [32628.443070] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in llama-server [194114] [32628.443091] i915 0000:00:02.0: [drm] llama-server[194114] context reset due to GPU hang * vulkan: implement deferred_memset on UMA --------- Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
blime4
referenced
this pull request
in blime4/llama.cpp
Feb 5, 2026
* vulkan: optimize UMA buffer operations and fix driver hangs The previous implementation was blocking the GPU for extended periods, causing the i915 driver to reset the context due to the hangcheck protection. [32628.443070] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in llama-server [194114] [32628.443091] i915 0000:00:02.0: [drm] llama-server[194114] context reset due to GPU hang * vulkan: implement deferred_memset on UMA --------- Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Seunghhon
pushed a commit
to Seunghhon/llama.cpp
that referenced
this pull request
Apr 26, 2026
…#16059) * vulkan: optimize UMA buffer operations and fix driver hangs The previous implementation was blocking the GPU for extended periods, causing the i915 driver to reset the context due to the hangcheck protection. [32628.443070] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in llama-server [194114] [32628.443091] i915 0000:00:02.0: [drm] llama-server[194114] context reset due to GPU hang * vulkan: implement deferred_memset on UMA --------- Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
The previous implementation was blocking the GPU for extended periods, causing the i915 driver to reset the context due to the hangcheck protection.
[32628.443070] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in llama-server [194114]
[32628.443091] i915 0000:00:02.0: [drm] llama-server[194114] context reset due to GPU hang
Make sure to read the contributing guidelines before submitting a PR