[PATCH v5] aarch64: Optimize memcpy for Kunpeng 950 processor

[Wilco Dijkstra]

Hi Weihong,

> For copies ¡Ü64 bytes, the implementation remains consistent with memcpy_sve.
> For 65¨C128 bytes, it removes the 96-byte branch and reorders instructions, improving performance by 18¨C32%.
> For >128 bytes, it aligns the destination to a 32-byte boundary and uses Pre-indexed load/store instructions to reduce address-update overhead.

> I¡¯ve switched from my personal QQ email (used in v1~v4) to my Huawei corporate email for this v5 submission.
> Huawei has a valid copyright assignment agreement with the FSF, and I¡¯m contributing this patch under that agreement.

OK thanks.

I've now committed it for you (9399998f72921d58f56023cce0d706358f3449c0).

Cheers,
Wilco