[PATCH v5 2/2] x86: Optimize strlen-avx2.S

[Aurelien Jarno]

On 2022-10-04 18:10, Sunil Pandey via Libc-alpha wrote:
> On Tue, Oct 4, 2022 at 2:20 PM Aurelien Jarno <aurelien@aurel32.net> wrote:
> >
> > On 2022-09-28 06:54, Sunil Pandey via Libc-stable wrote:
> > > Attached patch fixes BZ# 29611.
> > >
> > > I would like to backport it to 2.32,2.31,2.30,2.29 and 2.29. Let me know
> > > if there is any objection.
> >
> > Sorry to be late on this. I have a few comments about that patch:
> >
> > > From 86e1d88e1a3c126597ef39165275ada7564cfce9 Mon Sep 17 00:00:00 2001
> > > From: "H.J. Lu" <hjl.tools@gmail.com>
> > > Date: Mon, 19 Apr 2021 10:45:07 -0700
> > > Subject: [PATCH] x86-64: Require BMI2 for strchr-avx2.S
> > >
> > > Since strchr-avx2.S updated by
> > >
> > > commit 1f745ecc2109890886b161d4791e1406fdfc29b8
> > > Author: noah <goldstein.w.n@gmail.com>
> > > Date:   Wed Feb 3 00:38:59 2021 -0500
> > >
> > >     x86-64: Refactor and improve performance of strchr-avx2.S
> > >
> > > uses sarx:
> > >
> > > c4 e2 72 f7 c0        sarx   %ecx,%eax,%eax
> > >
> > > for strchr-avx2 family functions, require BMI2 in ifunc-impl-list.c and
> > > ifunc-avx2.h.
> > >
> > > (cherry picked from commit 83c5b368226c34a2f0a5287df40fc290b2b34359)
> > > ---
> > >  sysdeps/x86_64/multiarch/ifunc-avx2.h      |  4 ++--
> > >  sysdeps/x86_64/multiarch/ifunc-impl-list.c | 12 +++++++++---
> > >  2 files changed, 11 insertions(+), 5 deletions(-)
> >
> > First of all 1f745ecc2109890886b161d4791e1406fdfc29b8 never got
> > backported to 2.32 and older branches, and strchr-avx2.S in those
> > branches do not use BMI2 instructions. So it doesn't make sense to
> > backport it.
> >
> > That said the change in ifunc-avx2.h fixes:
> >
> > - memchr and rawmemchr, broken by the backport of acfd088a1963 ("x86:
> >   Optimize memchr-avx2.S")
> > - strlen and strnlen, broken by the backport of aaa23c350715 ("x86:
> >   Optimize strlen-avx2.S")
> >
> > So the issues are fixed, but mostly by chance.
> 
> How do you know it is a "by chance" fix, do you have any evidence to back
> your claim?

My point is that the commit that has been backported is fixing a bug
that doesn't exist in 2.32 branches. strchr-avx2.S does not the sarx
instruction as the commit claims, and does not use other BMI2
instructions either.

However following the backport of commit acfd088a1963 and aaa23c350715
in these branches, memchr-avx2.S and strlen-avx2.S use BMI2
instructions, and as they use ifunc-avx2.h, this actually fixes the bug.

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net