Conversation
stolk
force-pushed
the
fix/peer-copy-vmm-span
branch
2 times, most recently
from
September 16, 2026 19:45
6400f17 to
6b5e193
Compare
We cannot assume that the peer's blocks are contiguous. This fixes a real gpu crash, triggered in real-world use. The command will now fail cpu-side instead. A follow-up commit will address avoiding this case. The error now surfaces as ZE_RESULT_ERROR_OUT_OF_DEVICE_MEMORY. But an accurate zeDriverGetLastErrorDescription is set. The guard covers all fourteen resolveAlignedAllocation callers: appendMemoryCopyRegion appendMemoryFill appendImageCopy* appendSignalEvent appendWaitOnMemory appendWriteToMemory Related-To: GSD-13429 Signed-off-by: Bram Stolk <b.stolk@gmail.com>
Memory blocks on the peer are possibly not contiguous. This means that an integral copy may not be possible. Instead, split the copies at block-ends. This adds a helper function to find the splits. Verified as correct on real hardware (2x Intel Arc Pro B70) which crashed before, but now correctly copies. Resolves: GSD-13429 Signed-off-by: Bram Stolk <b.stolk@gmail.com>
stolk
force-pushed
the
fix/peer-copy-vmm-span
branch
from
September 25, 2026 05:27
6b5e193 to
35887ab
Compare
Author
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This MR contains two commits.
The earlier one detects when we are about the access beyond a block end. It will return a failure CPU side, instead of crashing the GPU. It adds robustness to the driver, and offers wide protection, also outside the copy-case.
The later commit avoids the incorrect copy by splitting it up where appropriate.
I have verified this to be correct on Dual Intel Arc Pro B70 setup.
Before, it would crash, with these changes, the copy succeeds.
This fixes: #995
See that issue for repro code to trigger the crash. (libze and SYCL versions available as repro.)