[SYCL] Do not internalize kernels when supporting dynamic linking #15307
Conversation
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
I privately enabled -fsycl-allow-device-dependencies by default and made a list of all new failures. These were the new failures fixed by this change. |
Are we planning to flip the default in future? If so, we need a comment near these new RUN lines that they should be dropped once that happens. And if not, I'm not sure this is a good strategy for the tests for a feature... |
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
Removed new RUN lines. Added new testcase in DeviceDependencies directory to exercise this fix. |
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
LU-JOHN
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
LU-JOHN
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
LU-JOHN
force-pushed
the
dont_internalize_kernels
branch
from
3ea898d
to
41879e1
Compare
LU-JOHN
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
| }); | ||
| Queue.wait(); | ||
| bool PassA = checkUSM(usmPtr, Range, Result); | ||
| std::cout << "Test 1a: " << (PassA ? "PASS" : "FAIL") << std::endl; |
There was a problem hiding this comment.
I don't think that our E2E tests should print anything in a positive scenario when they pass. The motivation for that is to reduce amount of I/O performed by tests and less data that has to be processed by lit framework. Logs for passed tests are hidden anyways, so they are of no use in our CI.
There was a problem hiding this comment.
I would rather immediately drop (almost) all std::cout lines, but I will leave that up to SYCL RT reviewers, who are code owners of the file
LU-JOHN
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
LU-JOHN
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
LU-JOHN
force-pushed
the
dont_internalize_kernels
branch
from
0d4a00d
to
a1d3543
Compare
LU-JOHN
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
LU-JOHN
force-pushed
the
dont_internalize_kernels
branch
from
a1d3543
to
0a98fa1
Compare
LU-JOHN
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
LU-JOHN
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
Signed-off-by: Lu, John <[email protected]>
Signed-off-by: Lu, John <[email protected]>
…a new testcase Signed-off-by: Lu, John <[email protected]>
Signed-off-by: Lu, John <[email protected]>
Signed-off-by: Lu, John <[email protected]>
Signed-off-by: Lu, John <[email protected]>
LU-JOHN
force-pushed
the
dont_internalize_kernels
branch
from
0a98fa1
to
4993992
Compare
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
LU-JOHN
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
Do not internalize kernels when supporting dynamic linking. Kernels must be visible so that host code can find them.