search

conda-forge / cuda-feedstock

A conda-smithy repository for cuda.
BSD 3-Clause "New" or "Revised" License
2 stars 8 forks source link

Switch to cross-compilation for `linux_aarch64` and `linux_ppc64le` #25

Closed jakirkham closed 4 months ago

jakirkham commented 5 months ago

Sometimes the natives builds of linux_aarch64 and linux_ppc64le on Travis CI don't start or fail in the queuing process, which needs restarting. This creates a bit of friction during the CTK release process

To reduce this friction, think we should switch linux_aarch64 and linux_ppc64le to "cross-compilation" on Azure. As the CTK packages are binary redists, we are not actually cross-compiling them. However the way the build architecture for Azure is setup is to use cross-compilers and different build_platform and target_platform. This would allow the builds to complete just as fast. Though it would mean that some kinds of tests (like running binaries for the target_platform) wouldn't work in cross-compilation (as they do in the native case). However this can already be an issue as we don't currently test with GPUs either

Making this change would amount to adding the following lines to the top of conda-forge.yml files for the feedstocks and re-rendering:

build_platform:
  linux_aarch64: linux_64
  linux_ppc64le: linux_64
jakirkham commented 4 months ago

Closing as completed