Re: [PATCH] riscv: avoid vrgather in RVV memcmp mismatch path

--000000000000a20d120653fad3fd
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Ack, the patch seems good to me, plan to put to my test queue then push :)

Pincheng Wang <[email protected]> =E6=96=BC 2026=E5=B9=B46=E6=
=9C=8811=E6=97=A5=E9=80=B1=E5=9B=9B =E4=B8=8B=E5=8D=8810:01=E5=AF=AB=E9=81=
=93=EF=BC=9A

> Hi all,
>
> Gentle ping. :)
>
> BR,
> Pincheng Wang
>
> On 2026/5/29 9:29, Pincheng Wang wrote:
> > vfirst.m already returns the byte offset of the first mismatch in the
> > current vector chunk. Use that offset to reload the two differing bytes
> > with lbu instead of extracting them with vrgather.vx and vmv.x.s.
> >
> > The vector gather path can be more expensive on some implementations an=
d
> > also increases vector register pressure. This keeps the mismatch path
> > shorter while preserving the memcmp result.
> >
> > Signed-off-by: Pincheng Wang <[email protected]>
> > ---
> >   newlib/libc/machine/riscv/memcmp-asm.S | 10 ++++------
> >   1 file changed, 4 insertions(+), 6 deletions(-)
> >
> > diff --git a/newlib/libc/machine/riscv/memcmp-asm.S
> b/newlib/libc/machine/riscv/memcmp-asm.S
> > index b05df9521..1cf1680e2 100644
> > --- a/newlib/libc/machine/riscv/memcmp-asm.S
> > +++ b/newlib/libc/machine/riscv/memcmp-asm.S
> > @@ -28,12 +28,10 @@ memcmp:
> >     li a0, 0
> >     ret
> >   .Lfound:
> > -  vrgather.vx v16, v0, a4
> > -  vrgather.vx v24, v8, a4
> > -  vmv.x.s a0, v16
> > -  vmv.x.s a4, v24
> > -  andi a0, a0, 0xff
> > -  andi a4, a4, 0xff
> > +  add a0, a0, a4
> > +  add a1, a1, a4
> > +  lbu a0, 0(a0)
> > +  lbu a4, 0(a1)
> >     sub a0, a0, a4
> >     ret
> >   .size memcmp, .-memcmp
>
>

--000000000000a20d120653fad3fd
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">Ack, the patch seems good to me, plan=C2=A0to put to my te=
st queue then push :)</div><br><div class=3D"gmail_quote gmail_quote_contai=
ner"><div dir=3D"ltr" class=3D"gmail_attr">Pincheng Wang &lt;<a href=3D"mai=
lto:[email protected]">[email protected]</a>&gt; =
=E6=96=BC 2026=E5=B9=B46=E6=9C=8811=E6=97=A5=E9=80=B1=E5=9B=9B =E4=B8=8B=E5=
=8D=8810:01=E5=AF=AB=E9=81=93=EF=BC=9A<br></div><blockquote class=3D"gmail_=
quote" style=3D"margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,=
204);padding-left:1ex">Hi all,<br>
<br>
Gentle ping. :)<br>
<br>
BR,<br>
Pincheng Wang<br>
<br>
On 2026/5/29 9:29, Pincheng Wang wrote:<br>
&gt; vfirst.m already returns the byte offset of the first mismatch in the<=
br>
&gt; current vector chunk. Use that offset to reload the two differing byte=
s<br>
&gt; with lbu instead of extracting them with vrgather.vx and vmv.x.s.<br>
&gt; <br>
&gt; The vector gather path can be more expensive on some implementations a=
nd<br>
&gt; also increases vector register pressure. This keeps the mismatch path<=
br>
&gt; shorter while preserving the memcmp result.<br>
&gt; <br>
&gt; Signed-off-by: Pincheng Wang &lt;<a href=3D"mailto:pincheng.plct@isrc.=
iscas.ac.cn" target=3D"_blank">[email protected]</a>&gt;<br>
&gt; ---<br>
&gt;=C2=A0 =C2=A0newlib/libc/machine/riscv/memcmp-asm.S | 10 ++++------<br>
&gt;=C2=A0 =C2=A01 file changed, 4 insertions(+), 6 deletions(-)<br>
&gt; <br>
&gt; diff --git a/newlib/libc/machine/riscv/memcmp-asm.S b/newlib/libc/mach=
ine/riscv/memcmp-asm.S<br>
&gt; index b05df9521..1cf1680e2 100644<br>
&gt; --- a/newlib/libc/machine/riscv/memcmp-asm.S<br>
&gt; +++ b/newlib/libc/machine/riscv/memcmp-asm.S<br>
&gt; @@ -28,12 +28,10 @@ memcmp:<br>
&gt;=C2=A0 =C2=A0 =C2=A0li a0, 0<br>
&gt;=C2=A0 =C2=A0 =C2=A0ret<br>
&gt;=C2=A0 =C2=A0.Lfound:<br>
&gt; -=C2=A0 vrgather.vx v16, v0, a4<br>
&gt; -=C2=A0 vrgather.vx v24, v8, a4<br>
&gt; -=C2=A0 vmv.x.s a0, v16<br>
&gt; -=C2=A0 vmv.x.s a4, v24<br>
&gt; -=C2=A0 andi a0, a0, 0xff<br>
&gt; -=C2=A0 andi a4, a4, 0xff<br>
&gt; +=C2=A0 add a0, a0, a4<br>
&gt; +=C2=A0 add a1, a1, a4<br>
&gt; +=C2=A0 lbu a0, 0(a0)<br>
&gt; +=C2=A0 lbu a4, 0(a1)<br>
&gt;=C2=A0 =C2=A0 =C2=A0sub a0, a0, a4<br>
&gt;=C2=A0 =C2=A0 =C2=A0ret<br>
&gt;=C2=A0 =C2=A0.size memcmp, .-memcmp<br>
<br>
</blockquote></div>

--000000000000a20d120653fad3fd--