[PATCH v1 2/3] riscv: Add RVV implementation __strcpy_vector

[Sergei Lewis]

>
> In previous measurements, reading the vl CSR appears to add noticeable
> latency
>

Why do we ever need to read the vl CSR at all? Sure, vle8ff adjusts it if
there is a fault. However, consider: strcpy() behaviour is undefined if the
buffer is not null terminated, so no particular behaviour is required in
that situation and we are free to choose whatever is cheapest; and if it
/is/ null terminated, the null byte will be available to subsequent
operations as normal regardless of what else vle8ff does or does not do,
and the subsequent comparison and store operations will behave correctly
without us needing to explicitly interact with VL.

Compare:

    mv          t0, a0
    vsetvli     t1, zero, e8, m8, ta, ma
1:  vle8ff.v    v8, (a1)
    vmseq.vi    v0, v8, 0
    vfirst.m    a2, v0
    bgez        a2, 1f
    vse8.v      v8, (t0)
    add         t0, t0, t1
    add         a1, a1, t1
    bltz        a2, 1b

1:  addi        t1, a2, 1
    vsetvli     zero, t1, e8, m8, ta, ma
    vse8.v      v8, (t0)
    ret

Also, I note the submitted patch uses m4. Why not m8?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/libc-alpha/attachments/20250919/1e14f1e5/attachment.htm>