RabbitMash icon indicating copy to clipboard operation
RabbitMash copied to clipboard

support for aarch64 (Neon)?

Open jianshu93 opened this issue 2 years ago • 2 comments

Dear RabbitMash team,

Thanks for making Mash so fast! I am wondering whether it also supports aarch64 like cpu instructions, for example, macOS M1/M2 series and Linux aarch64 (e.g., Huawei Kunpeng 920). I have no problems for other instructions.

Thanks,

Jianshu

jianshu93 avatar Mar 20 '23 01:03 jianshu93

Hi, I'm pretty sure most of the SIMD kernels can be implemented using the 128bit NEON intrinsics. I think the easiest way to do this is just porting the SSE code to NEON. Unfortunately, I don't have an aarch64 device at the moment, and I'm a little bit busy with another project recently. I think I can try to port the SSE version to NEON version at the end of April.

Best, Zekun

ZekunYin avatar Mar 20 '23 13:03 ZekunYin

Hello Zekun,

Many thanks for the quick response. I would love to test on macOS aarch64 and I also think the sse2neon will be very useful to port sse to neon,as you described. For Linux aarch64,I will ask for a friend for help if needed.

Thanks,

Jianshu

jianshu93 avatar Mar 20 '23 14:03 jianshu93