[SYCL] Adds regression test for buffer allocation for failed SYCL streams #467
Conversation
…eams SYCL streams that throws an exception during construction would still create buffers and accessors. This adds a test to ensure that no device memory is allocated for streams if they fail during construction. Signed-off-by: Steffen Larsen <[email protected]>
steffenlarsen
requested review from
bader,
romanovvlad and
vladimirlaz
as code owners
| sycl::queue q; | ||
| q.submit([&](sycl::handler &cgh) { | ||
| try { | ||
| // Try to create stream with invalid workItemBufferSize parameter. |
There was a problem hiding this comment.
Can we have a test in intel/llvm sycl/unittests instead?
There was a problem hiding this comment.
I was considering it, but it was simpler to have single_task try and manifest the memory.
One way to do it in the unittests would be to check that StreamBufferPool remains empty, but since it it private to the scheduler I would need to use the mock scheduler. Do you reckon that would work?
There was a problem hiding this comment.
You can have the same code + PIImages mocks + redefine call of piMemBufferCreate to a function which sets global flag if it is called. Then is "main" check that the flag is not set. Will it work?
There was a problem hiding this comment.
You are absolutely right, that works. I have moved this to a unittest in intel/llvm#4594 and will close this.
SYCL streams that throws an exception during construction would still create buffers and accessors. This adds a test to ensure that no device memory is allocated for streams if they fail during construction.
This test passes with intel/llvm#4594.