[libc][SVE] consider implement SVE acceleration for memset

[SchrodingerZhu]

llvm-project/libc/src/string/memory_utils/aarch64/inline_memset.h

Line 58 in 114f3b5

if (count == 0)

Similar to memcpy, memset also have the initial branches pattern. It can be interesting to check if predicated SVE can speed up the initial part.

[llvmbot]

@llvm/issue-subscribers-libc

Author: Schrodinger ZHU Yifan (SchrodingerZhu)

Details

https://github.com/llvm/llvm-project/blob/114f3b530bae5a195234e3bd1c1328b38b39a000/libc/src/string/memory_utils/aarch64/inline_memset.h#L58

Similar to memcpy, memset also have the initial branches pattern. It can be interesting to check if predicated SVE can speed the initial part.

[SchrodingerZhu]

This should be rather easy to do but requires devices to benchmark it.

cc @gchatelet @michaelrj-google