-
Notifications
You must be signed in to change notification settings - Fork 1.6k
Fix and document the two HUGETLB options for buffer allocation in Makefile.rule #4662
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
Regarding the option |
I won't merge this without further testing, this is here so I won't forget. |
Seems one also needs to define ALLOC_SHM, but after that the new code path is actually taken. |
Indeed, it is necessary to define |
Commit
Using |
Yes, and I'm getting the impression that the PR566 was bogus, there should be no dependency on ALLOC_SHM |
Hello, this PR removes the test for compiler options -march=loongarch64 on loongarch64, which will cause compilation failure when using clang. |
Another issue is that when COMPILE_TLS is defined, the alloc_hugetlb interface is still controlled by the ALLOC_SHM macro, which prevents the use of large page functionality. (Please correct me if I'm mistaken.) |
Oops, sorry, missed that difference in the merge conflict resolution. Restoring in #4679. I will look into cleaning up the USE_TLS side of memory.c later today. (And maybe memory.o should be conditionally compiled from either of two separate C files instead of being the giant Frankenstein monster I created when the TLS PRs landed) |
While I'm not intimately familiar with the structure of memory.c, it's clear that it's currently quite bloated with a lot of redundant code. Splitting it into two files seems like a good choice based on what I can see. |
prompted by a recent question - these are undocumented holdovers from GotoBLAS