-
Notifications
You must be signed in to change notification settings - Fork 795
[DevASAN] Move memory alloc info into DeviceInfo #20611
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: sycl
Are you sure you want to change the base?
Conversation
Since CPU/GPU device both support USM indirect access, we need to poison shadow of whole allocated memory in the device instead of only one context.
|
I think this PR is causing UR hangs, please make sure it's fixed before running CI again, thanks. |
pbalcer
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
lgtm, just two small comments.
| } | ||
| }; | ||
|
|
||
| struct USMLaunchInfo { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
no need to address this now, but it might be useful to unify this with the tsan launch info.
| } | ||
| // UR doesn't allow GlobalWorkOffset is null, we need to construct a zero | ||
| // value array if user doesn't specify its value. | ||
| if (GlobalWorkOffset) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
global work offset can be null, see:
llvm/unified-runtime/include/ur_api.h
Lines 7857 to 7859 in d10b37e
| /// [in][optional] pointer to an array of workDim unsigned values that | |
| /// specify the offset used to calculate the global ID of a work-item | |
| const size_t *pGlobalWorkOffset, |
If something is segfaulting when null is passed to launch kernel, please file a bug or let me know.
Since CPU/GPU device both support USM indirect access, we need to poison shadow of whole allocated memory in the device instead of only one context.