mbutrovich commented on PR #3366: URL: https://github.com/apache/datafusion-comet/pull/3366#issuecomment-3838423224
> > I propose just changing the `target-cpu` on ARM builds to `neoverse-n1` which gets you LSE (atomics) and is a reasonable baseline for ARM architectures available in 2026 from cloud vendors. It's also a subset of what Apple Silicon provides. Newer `neoverse` targets would be great, but I don't know if we want to require new Graviton and eliminate stuff like Ampere Altra. I'd need to do a survey of GCP and Azure ARM cores. > > One reason for doing it this way was that building for LSE and running on an ARM version that doesn't support LSE causes an instant crash. I don't have any real world data on how many people may be on older versions of ARM but didn't want to take the chance. As best as I can tell, LSE was added in ARMv-8.1-A, which predates anything I think we'd find in modern data centers. Azure and GCP use something newer than `neoverse-n1`, and `n1` seems like a reasonable baselines for current `Graviton` offerings. In the same way probably don't want to ship binaries for anything that doesn't have AVX2, I think it's reasonable to cut off ARM without LSE, at least for shipped binaries. Folks are free to compile their own if they need, in the same way that they'll get a faster build if they compile on their target architecture. I summarized my thoughts in https://github.com/apache/datafusion-comet/pull/3368 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
