-
Notifications
You must be signed in to change notification settings - Fork 796
[SYCL][Test] Make testsuite respect SYCL_LIB_DUMPS_ONLY #14879
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
|
An alternative to this is to teach the standalone header format about suffixes: diff --git a/sycl/test/format.py b/sycl/test/format.py
index 19d16a263426..b64f41844222 100644
--- a/sycl/test/format.py
+++ b/sycl/test/format.py
@@ -23,6 +23,8 @@ class SYCLHeadersTest(lit.formats.TestFormat):
for filename in filenames:
if not filename.endswith(".hpp"):
continue
+ if os.path.splitext(filename)[1] not in litConfig.suffixes:
+ continue
filepath = os.path.join(dirpath, filename)
if headers_filter is not None: |
Running the check-sycl-dumps target sets the SYCL_LIB_DUMPS_ONLY param for lit which should only enable the ABI dump tests. However, since 1fde656 we've added a non `lit.formats.ShTest()` which doesn't respect config.suffixes, so we need to account for this.
0d382af to
1d5a436
Compare
This might actually be a better approach, because instead of introducing a custom logic for skipping certain tests, it would mean that tests won't launch themselves thanks to existing/standard/built-in mechanism of filters. The latter is a cleaner design, I think |
|
On Thu Aug 1, 2024 at 10:46 AM BST, Alexey Sachkov wrote:
The latter is a cleaner design, I think
Yeah. Let's go with that. I'll redo it.
|
AlexeySachkov
left a comment
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.
Thanks!
Done :) |
Thanks Alexey. |
We use these tags at least for creating release notes. @bader can provide more reasons if any. |
|
The original idea is these patches are supposed to be upstreamed to llvm/llvm-project repository. In order to do that we are trying to follow LLVM's guidelines and policies. This requirement is specified in https://llvm.org/docs/DeveloperPolicy.html#commit-messages. |
|
CUDA failures seem down to device detection unrelated to this patch @intel/llvm-gatekeepers please merge |
sure thing ;) |
| for filename in filenames: | ||
| if not filename.endswith(".hpp"): | ||
| suffix = os.path.splitext(filename)[1] | ||
| if suffix not in SUFFIXES or suffix not in litConfig.suffixes: |
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.
@ldrumm, I've just realized that this patch had completely disable the testing suite :)
We don't have .hpp registered as a suffix and therefore we always skip all those tests. This is more of a FYI, I'm working on a revamp/expansion of the suite in scope of #16012, so I will fix it as part of that effort
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.
Addressed this in #16117
We have a suite to check that every header contains enough `#include` directives and forward declarations so that it can be included standalone and used. However, that suite got accidentally disabled some time ago in #14879 This PR re-enabled the suite and fixed all issues it detected. Besides that, the suite is upgraded: - it now scans source directory, meaning that it won't complain about headers which were removed from the codebase, but left in build folder - it is now possible to have add a custom test instead of auto-generated one: this is useful when you want to trigger certain template instantiations to make sure that all code paths are covered
Running the check-sycl-dumps target sets the SYCL_LIB_DUMPS_ONLY param for lit which should only enable the ABI dump tests. However, since 1fde656 we've added a non
lit.formats.ShTest()which doesn't respect config.suffixes, so we need to account for this.