You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It would be nice to use the AMD AOCC packages for building to be able to take advantage of the hardware-optimizations: https://developer.amd.com/amd-aocc/
as well as other use cases where there are flang/clang optimizations on linux, such as AMD's ROCM OpenMP packages.
I saw this ticket here: #13462
but also wanted to request a way to specify clang as well as flang, or if there might be a way to do it via pip that just isn't in the documentation
It seems unlikely that the 3ff9227flang included there would work for the current LLVM Flang "F18". Classic flang which was based off of pgifortran might not be a very long term solution so I wouldn't count on it being used. The flang/clang optimizations are all F18 based.
Also since np.distutils are deprecated and distutils will be removed in Python 3.12.0 in October 2023, it might not be an area worth exploring.
clang is very well supported already, so I'm not sure what needs to be done for it. Another issue (or an edit) with a focus on AOCC would be good. :)
Activity
HaoZeke commentedon Oct 30, 2021
It seems unlikely that the 3ff9227
flangincluded there would work for the current LLVM Flang "F18". Classicflangwhich was based off ofpgifortranmight not be a very long term solution so I wouldn't count on it being used. Theflang/clangoptimizations are allF18based.Also since
np.distutilsare deprecated anddistutilswill be removed in Python 3.12.0 in October 2023, it might not be an area worth exploring.clangis very well supported already, so I'm not sure what needs to be done for it. Another issue (or an edit) with a focus on AOCC would be good. :)rgommers commentedon Sep 16, 2023
We won't be accepting new features for
numpy.distutilsanymore - I suggest switching to either Meson or CMake. Closing, thanks all