-
Notifications
You must be signed in to change notification settings - Fork 769
[SYCL] Turn on -fsycl-id-queries-fit-in-int by default #3427
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
Conversation
This patch enables SYCL ID queries fit within MAX_INT option to true on default mode. Signed-off-by: Soumi Manna <[email protected]>
Signed-off-by: Soumi Manna <[email protected]>
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.
Seems ok to me.
Thanks for the reviews. I am seeing 35 ESIMD tests fail with the patch. The same patch does not cause any failure on downstream. Not sure why? @DenisBakhvalov or @bader, is there any way i could reproduce the failures locally? |
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 aside from precommit failures.
@DenisBakhvalov mentions that, it's because of the llvm.assume intrinsic in device code. BE compiler cannot handle them. So, we cannot enable SYCL_ID_QUERIES_FIT_IN_INT for ESIMD. |
Yes, and I'm not sure we can introduce any ESIMD-specific workaround (like we did previously) since we unified the SYCL-ESIMD compilation. I think the way to proceed here is to have vector GPU backend (used by ESIMD code) accept |
ID as int is a big performance improvement for DPC++ in many cases, AFAIK. So, it would be good to have a w/a until ESIMD BE is ready to accepts these. It should be safe w/a to remove llvm.assume in ESIMD kernels for now. I will create a PR shortly. |
|
@smanna12, please rebase to latest llvm/sycl and rerun, ESIMD failures should be gone |
All ESIMD tests are passing now. This patch is causing regressions for two SubGroup tests. SYCL :: SubGroup/sub_group_as.cpp I have created PR for fixing the failing Subgroup tests: |
The changes are due to the PR: intel/llvm#3427 Signed-off-by: Soumi Manna <[email protected]>
The changes are due to the PR: intel/llvm#3427 Signed-off-by: Soumi Manna <[email protected]>
Rebasing and retesting PR after intel/llvm-test-suite#220 merge. |
@bader All pre-commit test failures are fixed. I did not make any change to any of the files on this PR. I am not sure why buildbot/Lit_with_Cuda is failing with warning like this: This worked fine couple of days ago with same patch. Is this expected? |
@tfzhu, is the machine in a broken state? |
|
The changes are due to the PR: intel/llvm#3427 Signed-off-by: Soumi Manna <[email protected]>
The changes are due to the PR: intel/llvm#3427 Signed-off-by: Soumi Manna <[email protected]>
…ng (#431) The changes are due to the PR: intel/llvm#3427 Signed-off-by: Soumi Manna <[email protected]>
The changes are due to the PR: intel#3427 Signed-off-by: Soumi Manna <[email protected]>
…ng (intel/llvm-test-suite#431) The changes are due to the PR: intel#3427 Signed-off-by: Soumi Manna <[email protected]>
This patch enables sycl_id_queries_fit_in_int macro to true on default mode to fix regressions and performance issue.
Signed-off-by: Soumi Manna [email protected]