-
Notifications
You must be signed in to change notification settings - Fork 419
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
Add/improve errors for .localAccess()
calls to non-local elements
#25754
Add/improve errors for .localAccess()
calls to non-local elements
#25754
Commits on Aug 14, 2024
-
Add .localAccess() errors for remote single-locale arrays
--- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 22618dd - Browse repository at this point
Copy the full SHA 22618ddView commit details -
Refactor to support better errors for block-distributed arrays as well
--- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 4128a23 - Browse repository at this point
Copy the full SHA 4128a23View commit details
Commits on Aug 16, 2024
-
Change localCheck to a param to avoid module init circularities
I considered doing this in the original draft since the value is always known and the different callsites need different things, but shied away from it worrying about code size increases. However, I'm getting problems with module initialization orders for tests like: arrays/ferguson/array-initialization-patterns/array-init-replicated.chpl --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for f55e69d - Browse repository at this point
Copy the full SHA f55e69dView commit details -
Extend locality checks to handle arrays with multiple blocks per locale
--- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for b71e027 - Browse repository at this point
Copy the full SHA b71e027View commit details -
Skip new tests for a single locale; they're designed for multiple
(not to say they wouldn't be interesting to test on a single as well to make sure there's no error, but getting the .good files right doesn't seem worth the effort given that they'll vary from case to case. --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for ef3e202 - Browse repository at this point
Copy the full SHA ef3e202View commit details
Commits on Sep 4, 2024
-
Merge branch 'main' of https://github.com/chapel-lang/chapel into loc…
…alaccess-error-improvements
Configuration menu - View commit details
-
Copy full SHA for 1175602 - Browse repository at this point
Copy the full SHA 1175602View commit details
Commits on Sep 6, 2024
-
Fix localSubdomains() calls for main distributions in the face of dup…
… locales It seems these have never worked, and that the tests written to see whether they did were getting away with things? Also, add a test showing why my implementation so far is insufficient to make locality checks for Stencils work. --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 34a636e - Browse repository at this point
Copy the full SHA 34a636eView commit details -
Rework local access checks to permit a new optional interface for Ste…
…ncilDist This adds a new iterator similar to localSubdomains() yet which returns all stored local indices, whether part of the locale's subdomain or not, to permit fluff to be considered "in bounds" for local accesses. --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for adabcef - Browse repository at this point
Copy the full SHA adabcefView commit details
Commits on Sep 7, 2024
-
Merge branch 'main' of https://github.com/chapel-lang/chapel into loc…
…alaccess-error-improvements
Configuration menu - View commit details
-
Copy full SHA for 802c72a - Browse repository at this point
Copy the full SHA 802c72aView commit details -
Update a sensitive-to-uid changes .bad entries
--- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 9a30dd7 - Browse repository at this point
Copy the full SHA 9a30dd7View commit details
Commits on Sep 9, 2024
-
--- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 12f63a0 - Browse repository at this point
Copy the full SHA 12f63a0View commit details -
Clean up skipifs and generalize for --no-local testing
--- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 5df5a4a - Browse repository at this point
Copy the full SHA 5df5a4aView commit details -
Update test to work for comm=none as well; add more working + error c…
…ases --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 266fb3f - Browse repository at this point
Copy the full SHA 266fb3fView commit details -
Clean up localAccess indexing expressions for 1D and nD arrays
Also, add new test of multidim localAccess bounds violation to lock that case in. --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 435c1bf - Browse repository at this point
Copy the full SHA 435c1bfView commit details -
Clean up / simplify code and diff
--- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for f7270bd - Browse repository at this point
Copy the full SHA f7270bdView commit details
Commits on Sep 10, 2024
-
Act on part of Engin's feedback
The main piece of feedback I'm acting on here is to update the comment in StencilDom.doiLocalStoredSubdomains(), which had fallen out of date when I removed the 'locale' argument to the iterator in a late-breaking commit. Here, I've updated it to note that the current call should be communication-free since we are always answering the question for 'here'. I've also updated the stencil localAccess.chpl test to wrap the localAccess calls in a `local` block to lock in that no communication is required to implement the calls. Then I took that same principle to the localAccess-block*.chpl tests to lock in that the block localAccess calls do not require communication. Note that I used an unusual formatting scheme here because I didn't want to have to update the line numbers of the error messages... :( --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for f76bf02 - Browse repository at this point
Copy the full SHA f76bf02View commit details -
Add illegal local access back in to capture error
Engin pointed out, rightly, that in updating these tests to make the localaccess calls legal, I aso pointed out that we had a pre-existing bug specific to the case of oversubsctibed locales that we didn't have any tests catching. So here, I've added the illegal accesses back at the end of the test to lock in the remote accesses. The reason that these cases are different is that our dsiLocalAccess() calls on Block and Cyclic use different (arguably incorrect) logic when used in an oversubscribed manner. --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 24b927e - Browse repository at this point
Copy the full SHA 24b927eView commit details -
Update stencil oversubscription test for multiple locales
While reviewing the block and cyclic versions of this test, it wasn't clear to me why this stencil version of the test didn't require changes when the others did. It turns out that the answer is that it wasn't ever enabled for multi-locale runs... Fixing that here, and updating it similar to the block and stencil versions. --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 629421f - Browse repository at this point
Copy the full SHA 629421fView commit details -
Add --checks flags to localAccess error test cases
This adds a --checks flag to all these tests I've been working on for localAccess() errors to ensure that they don't fail in --fast testing. This may be the first time in years I've thought of this before having the nightly tests point out my mistake (?). --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 05ec799 - Browse repository at this point
Copy the full SHA 05ec799View commit details -
With new error-generating localAccess cases added, specialize .goods …
…for comm=none --- Signed-off-by: Brad Chamberlain <[email protected]>
Configuration menu - View commit details
-
Copy full SHA for 2f9db17 - Browse repository at this point
Copy the full SHA 2f9db17View commit details