MAINT: Further fixups to uint alignment checks #12706
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Backport of #12677.
A few more fixes following on #12626, based on testing on a ppc64be system.
First, on this system some einsum tests failed due typo in that PR: ; should have been &&. This wasn't caught by x64 unit tests, so I've added a test that should cover those code paths.
Second, I figured out that some of the fix in #12626 wasn't quite right: I had misunderstood what the NPY_ITER_ALIGNED flag is supposed to do. My understanding now is that is a request to guarantee that the nditer buffers are "true" aligned. (uint alignment is only an internal numpy thing which users shouldn't know about), which nditer it does by forcing a buffer/cast if that flag is requested, since all new buffers are true aligned. With that understanding, it now seems that the NPY_OP_ITFLAG_ALIGNED flag is not needed: Its only use was as a precalculated computation of the aligned parameter of PyArray_GetDTypeTransferFunction, but that is more safely computed on the spot, based on both uint and true alignment of the op. So I've removed all uses of NPY_OP_ITFLAG_ALIGNED.
This latter change was indirectly discovered from staring at the code while trying to bug-hunt failed alignment asserts on ppc64be. But those turned out to be irrelevant glibc problems from #10491. At least it got me to read a lot of code over more thoroughly: I also made sure that the ufunc code doesn't need alignment fixes.