search

conda-forge / openblas-feedstock

A conda-smithy repository for openblas.
BSD 3-Clause "New" or "Revised" License
9 stars 38 forks source link

aarch64 linux: fix docker image used in linux_aarch64_USE_OPENMPx build configs #158

Closed snadampal closed 5 months ago

snadampal commented 5 months ago

updated the aarch64 linux build config yaml files to pull aarch64 condaforge docker image instead of x86 version. with this change, the local builds are working fine on aarch64 platform. fixes this issue: https://github.com/conda-forge/openblas-feedstock/issues/157

Checklist

This PR fixes the issue# https://github.com/conda-forge/openblas-feedstock/issues/157

conda-forge-webservices[bot] commented 5 months ago

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

isuruf commented 5 months ago

We are using cross compiling in CI. You should set export DOCKER_IMAGE=quay.io/condaforge/linux-anvil-aarch64 before running build-locally.py

snadampal commented 5 months ago

thanks @isuruf , I have a question about the conda openblas package naming: Conda has two openblas packages for aarch64 linux platform. (1) pthread version and (2) openmp version pthread version is the default one, and gets installed when we use conda install openblas

to install openmp version, we need to specify it explicitly

conda install libopenblas=*=*openmp*

(1) Is there any specific reason for making the pthread version the default? (2) I found this information in a hard way by building the conda-forge from sources and seeing the perf difference by myself. Is this information documented somewhere in conda-forge? if not, appreciate if this is documented clearly to avoid performance issues due to mismatched runtime (e.g. openmp runtime installing pthread version and vice versa).