[SYCL] updating tests for SYCL2020 errc conformance #857
Conversation
…minor mistakes in our tests. fixed Signed-off-by: Chris Perkins <[email protected]>
cperkinsintel
changed the title
Signed-off-by: Chris Perkins <[email protected]>
Signed-off-by: Chris Perkins <[email protected]>
|
/verify with intel/llvm#5582 |
| @@ -394,7 +394,7 @@ int main() { | |||
| myQueue.submit([&](handler &cgh) { | |||
| accessor<int, 2, access::mode::write, access::target::device, | |||
| access::placeholder::false_t> | |||
| B(Buffer, cgh, range<2>(20, 20), id<2>(10, 10)); | |||
| B(Buffer, cgh, range<2>(10, 10), id<2>(10, 10)); | |||
There was a problem hiding this comment.
These changes look unrelated to SYCL2020 errc conformance. Can we remove them from this PR?
There was a problem hiding this comment.
One of the conformance checks is that out-of-bounds range args should trigger an exception in accessor construction. I added that. Without this change, this test triggers that error.
| @@ -3,6 +3,9 @@ | |||
| // RUN: %CPU_RUN_PLACEHOLDER %t.out | |||
| // RUN: %GPU_RUN_PLACEHOLDER %t.out | |||
| // RUN: %ACC_RUN_PLACEHOLDER %t.out | |||
|
|
|||
| // UNSUPPORTED: (opencl && gpu) | |||
There was a problem hiding this comment.
All changes in this file doesn't look related to SYCL2020 errc conformance, I couldn't find any errc checks in the file.
It is possible to move them to a separate PR if they are unrelated? Does it make sense to create a separate test for your modified test case (which shows a problem on opencl gpu) to avoid disabling whole test for opencl gpu?
There was a problem hiding this comment.
This test was using an out-of-bounds range argument to an accessor. And, oddly, that was masking an error that is occurring over on OCL:GPU. I reduced that down and reported it, there is a ticket against it now. But in the interim, this test won't pass on OCL:GPU, not due to any bug in the error system, but due to a real bug that OCL:GPU has handling offsets. Strangely, that is the only device with the problem. L0, OCL:CPU, OCL:ACC and Host are all fine.
The SYCL2020 specification has quite a few errc values it expects thrown exceptions to have. Most of the existing ones are fine, but there are several cases that need updating. After this there are some other changes that will need to be done for errc conformance, but those changes are slightly more involved. They will be done in a follow up PR s. The tests were updated as well. intel/llvm-test-suite#857 has the matching changes.
in intel/llvm#5582 we are adding an update that ensures that the errc of any exception is correct as specified in the SYCL2020 spec. Here we are updating and expanding the tests to accompany Signed-off-by: Chris Perkins [email protected]
…suite#857) in intel#5582 we are adding an update that ensures that the errc of any exception is correct as specified in the SYCL2020 spec. Here we are updating and expanding the tests to accompany Signed-off-by: Chris Perkins [email protected]
in intel/llvm#5582 we are adding an update that ensures that the errc of any exception is correct as specified in the SYCL2020 spec. Here we are updating and expanding the tests to accompany
Signed-off-by: Chris Perkins [email protected]