-
Notifications
You must be signed in to change notification settings - Fork 130
[SYCL][CUDA] Add tests for asynchronous barrier #737
Conversation
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.
The tests look good. I've added some small comments.
One concern is that the the interface
test case includes a lot in a single kernel. If that test fails it may be hard to determine which part of the test caused the failure. Would it make sense to either split it up or save and validate intermediate results to help isolate the failing operations?
I can split the test into multiple kernels, but a large amount of code (more or less the whole basic test) would need to be duplicated in each test. |
Is there no good way of reducing common code for the cases? If it isn't worth it, I am okay with keeping the test as-is. |
I think at best I can reduce the amount of code duplication (but not eliminate it) at cost of making the code a bit harder to read. I would prefer to leave this as is, as it is unlikely that implementation of something changes in a way that would break semantics but not interface. |
Adds extension proposal and implementation for asynchronous barrier (for now the implementation is for CUDA backend sm 80+ only). Tests for this are here: intel/llvm-test-suite#737
Adds extension proposal and implementation for asynchronous barrier (for now the implementation is for CUDA backend sm 80+ only). Tests for this are here: intel/llvm-test-suite#737
Adds tests for intel/llvm#5303