Does the NDK Support the Oryon CPU Architecture?

[chenghuaWang]

Qualcomm's Snapdragon 8 Elite has been widely adopted, featuring a CPU core based on the Oryon architecture instead of the Cortex architecture. I've noticed that LLVM has specific optimizations for Oryon. Could you please inform me when the NDK will support the '-mcpu=oryon-1' architecture?

https://www.mail-archive.com/[email protected]/msg433532.html

[DanAlbert]

Sounds like it's still an armv8 compatible thing, so whenever we ship an LLVM new enough to include that commit. I know that's not a useful answer but it's the most accurate one I can give you. Some time in the next year, presumably.

New CPU microarchictectures aren't usually all that interesting for app developers. Your app can't assume those features will exist at runtime because it'll be a decade or more before you can rely on your entire user base having CPUs that new. You can use things like https://clang.llvm.org/docs/AttributeReference.html#target-clones for individual hot functions, but that complicates your test matrix and I suspect most developers don't do that.

[pirama-arumuga-nainar]

+1 this should largely be irrelevant to app developers. But since you asked, this commit merged in June 2024 - it should be in NDK r29.

[chenghuaWang]

Thank you for your answer. I'll try to see if r29 has this arch option, as the r28 I'm currently using doesn't have. We are currently writing some high-performance operators for specific devices and may need to utilize certain architecture-specific capabilities. I think the best approach for now might still be to implement it directly with inline assembly.