From: "H.J. Lu" <hjl.tools@gmail.com>
To: Noah Goldstein <goldstein.w.n@gmail.com>
Cc: GNU C Library <libc-alpha@sourceware.org>,
"Carlos O'Donell" <carlos@systemhalted.org>
Subject: Re: [PATCH v1] x86: Add support for building strstr with explicit ISA level
Date: Tue, 28 Jun 2022 11:34:34 -0700 [thread overview]
Message-ID: <CAMe9rOrM7M7pzjEtBFPyG6a0Tap7f8hT8PzWNjWE-Qui_rRP8g@mail.gmail.com> (raw)
In-Reply-To: <CAFUsyfL18LwS01sSEPJT_=dEuaiez_zyBxS8t2fDhO5e27iLXQ@mail.gmail.com>
On Tue, Jun 28, 2022 at 11:24 AM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
>
> On Tue, Jun 28, 2022 at 11:21 AM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > On Tue, Jun 28, 2022 at 8:26 AM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
> > >
> > > Small changes for this function as the generic implementation remains
> > > the same for all ISA levels.
> > >
> > > Only changes are using the X86_ISA_CPU_FEATURE{S}_{USABLE|ARCH}_P
> > > macros so that some of the checks at least can constant evaluate
> > > and some comments explaining the ISA constraints on the function.
> > > ---
> > > sysdeps/x86_64/multiarch/ifunc-impl-list.c | 13 +++++++------
> > > sysdeps/x86_64/multiarch/strstr.c | 10 +++++-----
> > > 2 files changed, 12 insertions(+), 11 deletions(-)
> > >
> > > diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
> > > index 0d28319905..a1bff560bc 100644
> > > --- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
> > > +++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
> > > @@ -620,12 +620,13 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
> > >
> > > /* Support sysdeps/x86_64/multiarch/strstr.c. */
> > > IFUNC_IMPL (i, name, strstr,
> > > - IFUNC_IMPL_ADD (array, i, strstr,
> > > - (CPU_FEATURE_USABLE (AVX512VL)
> > > - && CPU_FEATURE_USABLE (AVX512BW)
> > > - && CPU_FEATURE_USABLE (AVX512DQ)
> > > - && CPU_FEATURE_USABLE (BMI2)),
> > > - __strstr_avx512)
> > > + /* All implementations of strstr are built at all ISA levels. */
> > > + IFUNC_IMPL_ADD (array, i, strstr,
> > > + (CPU_FEATURE_USABLE (AVX512VL)
> > > + && CPU_FEATURE_USABLE (AVX512BW)
> > > + && CPU_FEATURE_USABLE (AVX512DQ)
> > > + && CPU_FEATURE_USABLE (BMI2)),
> > > + __strstr_avx512)
> > > IFUNC_IMPL_ADD (array, i, strstr, 1, __strstr_sse2_unaligned)
> > > IFUNC_IMPL_ADD (array, i, strstr, 1, __strstr_generic))
> > >
> > > diff --git a/sysdeps/x86_64/multiarch/strstr.c b/sysdeps/x86_64/multiarch/strstr.c
> > > index 2b83199245..3f86bfa5f2 100644
> > > --- a/sysdeps/x86_64/multiarch/strstr.c
> > > +++ b/sysdeps/x86_64/multiarch/strstr.c
> > > @@ -49,13 +49,13 @@ IFUNC_SELECTOR (void)
> > > const struct cpu_features *cpu_features = __get_cpu_features ();
> > >
> > > if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_AVX512)
> > > - && CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
> > > - && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)
> > > - && CPU_FEATURE_USABLE_P (cpu_features, AVX512DQ)
> > > - && CPU_FEATURE_USABLE_P (cpu_features, BMI2))
> > > + && X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
> > > + && X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)
> > > + && X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, AVX512DQ)
> > > + && X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, BMI2))
> > > return __strstr_avx512;
> > >
> > > - if (CPU_FEATURES_ARCH_P (cpu_features, Fast_Unaligned_Load))
> > > + if (X86_ISA_CPU_FEATURES_ARCH_P (cpu_features, Fast_Unaligned_Load, ))
> >
> > Is Fast_Unaligned_Load set on all processors before? If not, we should
> > revert
>
> It's set at ISA level >= 2. AFAICT the reason the bit exists is so that that
> CPUs with slow sse42 can fallback on an unaligned sse2 implementation
> if it's available as opposed to the generic / often quite expensive aligned
> sse2 impl.
Is Fast_Unaligned_Load set on Zhaoxin processors?
> Example in strcmp
>
> >
> > /* Feature(s) enabled when ISA level >= 2. */
> > #define Fast_Unaligned_Load_X86_ISA_LEVEL 2
> >
> > > return __strstr_sse2_unaligned;
> > >
> > > return __strstr_generic;
> > > --
> > > 2.34.1
> > >
> >
> >
> > --
> > H.J.
--
H.J.
next prev parent reply other threads:[~2022-06-28 18:35 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-06-28 15:26 [PATCH v1] x86-64: Small improvements to dl-trampoline.S Noah Goldstein
2022-06-28 15:26 ` [PATCH v1] x86: Add support for building strstr with explicit ISA level Noah Goldstein
2022-06-28 18:20 ` H.J. Lu
2022-06-28 18:24 ` Noah Goldstein
2022-06-28 18:34 ` H.J. Lu [this message]
2022-06-28 18:38 ` Noah Goldstein
2022-06-28 15:26 ` [PATCH v1] x86: Add support for building {w}memcmp{eq} " Noah Goldstein
2022-06-29 18:52 ` H.J. Lu
2022-06-29 19:27 ` Noah Goldstein
2022-06-29 19:41 ` H.J. Lu
2022-06-29 19:44 ` Noah Goldstein
2022-06-29 19:48 ` H.J. Lu
2022-06-29 22:09 ` Noah Goldstein
2022-06-29 22:09 ` [PATCH v2] " Noah Goldstein
2022-06-29 22:49 ` H.J. Lu
2022-06-29 23:11 ` Noah Goldstein
2022-06-29 23:11 ` [PATCH v3] " Noah Goldstein
2022-07-01 22:56 ` H.J. Lu
2022-06-28 18:15 ` [PATCH v1] x86-64: Small improvements to dl-trampoline.S H.J. Lu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAMe9rOrM7M7pzjEtBFPyG6a0Tap7f8hT8PzWNjWE-Qui_rRP8g@mail.gmail.com \
--to=hjl.tools@gmail.com \
--cc=carlos@systemhalted.org \
--cc=goldstein.w.n@gmail.com \
--cc=libc-alpha@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).