-
Notifications
You must be signed in to change notification settings - Fork 787
[SPIR-V] Cherry-pick of "Add SPIR-V 1.4 checks" #7493
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
/verify |
/summary:run |
trying to trigger windows testing |
/verify with intel/llvm-test-suite#1414 |
/verify with intel/llvm-test-suite#1414 |
/start Windows testing |
44de063
to
0c17303
Compare
/start Windows testing |
/start Windows testing |
/verify with intel/llvm-test-suite#1477 |
/testwin |
/verify with intel/llvm-test-suite#1479 |
select_device.cpp is unrelated test failure, reported internally SubGroupMask/Basic.cpp on windows to investigate |
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 overall. Just some nitpicks on some changes. But I suppose I am also not looking at the full commit here. Ok to merge. Thanks. And yes, sorry for delay in review.
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.
Please, resolve merge conflicts.
One of the patches missing in intel/llvm Signed-off-by: Sidorov, Dmitry dmitry.sidorov@intel.com Original commit: KhronosGroup/SPIRV-LLVM-Translator@c5b3c8e3283b Second attempt
0c17303
to
082b2fe
Compare
/summary:run |
Other issues are found - seems like CPU and GPU SPIR-V consumers restrict max to 1.2, but I'm not sure why. Also for some reasons it happens only for KernelFusion tests (only they use SPIR-V 1.4 functionality?). Need to have extra investigation. Converted to draft for now. |
It seems like that before going to the device's driver, the code is being put to the translator one more (or two more?) time(s). And when it's done, |
Hi @MrSidims, We initially chose 1.2 as the maximum value, as this is a safe default for OpenCL 2.2 devices. You can remove setting the maximum version there, so kernel fusion would use the tool default. |
@sommerlukas thanks! I've created #8358 . One of the biggest reasons to move to 1.4 is that partial unrolling started to be supported only in this version, and we would lack performance in some workloads if we wouldn't switch. Also our compiler implementation heavily relies on |
@MrSidims: Thanks for the explanation! From our side, 1.2 was only the initial choice, it should be fine to switch to 1.4. |
/summary:run |
Alexey's requested change was to merge sycl branch into this one |
One of the patches missing in intel/llvm
Signed-off-by: Sidorov, Dmitry dmitry.sidorov@intel.com
Original commit:
KhronosGroup/SPIRV-LLVM-Translator@c5b3c8e3283b
Second attempt