[2/2] x86-64: Remove Prefer_AVX2_STRCMP

  Remove Prefer_AVX2_STRCMP to enable EVEX strcmp.  When comparing 2 32-byte
strings, EVEX strcmp has been improved to require 1 load, 1 VPTESTM, 1
VPCMP, 1 KMOVD and 1 INCL instead of 2 loads, 3 VPCMPs, 2 KORDs, 1 KMOVD
and 1 TESTL while AVX2 strcmp requires 1 load, 2 VPCMPEQs, 1 VPMINU, 1
VPMOVMSKB and 1 TESTL.  EVEX strcmp is now faster than AVX2 strcmp by up
to 40% on Tiger Lake and Ice Lake.
---
 sysdeps/x86/cpu-features.c                                | 8 --------
 sysdeps/x86/cpu-tunables.c                                | 2 --
 .../include/cpu-features-preferred_feature_index_1.def    | 1 -
 sysdeps/x86_64/multiarch/strcmp.c                         | 3 +--
 sysdeps/x86_64/multiarch/strncmp.c                        | 3 +--
 5 files changed, 2 insertions(+), 15 deletions(-)
  

On Mon, Nov 1, 2021 at 5:54 AM H.J. Lu via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> Remove Prefer_AVX2_STRCMP to enable EVEX strcmp.  When comparing 2 32-byte
> strings, EVEX strcmp has been improved to require 1 load, 1 VPTESTM, 1
> VPCMP, 1 KMOVD and 1 INCL instead of 2 loads, 3 VPCMPs, 2 KORDs, 1 KMOVD
> and 1 TESTL while AVX2 strcmp requires 1 load, 2 VPCMPEQs, 1 VPMINU, 1
> VPMOVMSKB and 1 TESTL.  EVEX strcmp is now faster than AVX2 strcmp by up
> to 40% on Tiger Lake and Ice Lake.
> ---
>  sysdeps/x86/cpu-features.c                                | 8 --------
>  sysdeps/x86/cpu-tunables.c                                | 2 --
>  .../include/cpu-features-preferred_feature_index_1.def    | 1 -
>  sysdeps/x86_64/multiarch/strcmp.c                         | 3 +--
>  sysdeps/x86_64/multiarch/strncmp.c                        | 3 +--
>  5 files changed, 2 insertions(+), 15 deletions(-)
>
> diff --git a/sysdeps/x86/cpu-features.c b/sysdeps/x86/cpu-features.c
> index 645bba6314..be2498b2e7 100644
> --- a/sysdeps/x86/cpu-features.c
> +++ b/sysdeps/x86/cpu-features.c
> @@ -546,14 +546,6 @@ init_cpu_features (struct cpu_features *cpu_features)
>           if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
>             cpu_features->preferred[index_arch_Prefer_No_VZEROUPPER]
>               |= bit_arch_Prefer_No_VZEROUPPER;
> -
> -         /* Since to compare 2 32-byte strings, 256-bit EVEX strcmp
> -            requires 2 loads, 3 VPCMPs and 2 KORDs while AVX2 strcmp
> -            requires 1 load, 2 VPCMPEQs, 1 VPMINU and 1 VPMOVMSKB,
> -            AVX2 strcmp is faster than EVEX strcmp.  */
> -         if (CPU_FEATURE_USABLE_P (cpu_features, AVX2))
> -           cpu_features->preferred[index_arch_Prefer_AVX2_STRCMP]
> -             |= bit_arch_Prefer_AVX2_STRCMP;
>         }
>
>        /* Avoid avoid short distance REP MOVSB on processor with FSRM.  */
> diff --git a/sysdeps/x86/cpu-tunables.c b/sysdeps/x86/cpu-tunables.c
> index 00fe5045eb..61b05e5b1d 100644
> --- a/sysdeps/x86/cpu-tunables.c
> +++ b/sysdeps/x86/cpu-tunables.c
> @@ -239,8 +239,6 @@ TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *valp)
>               CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features,
>                                                 Fast_Copy_Backward,
>                                                 disable, 18);
> -             CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH
> -               (n, cpu_features, Prefer_AVX2_STRCMP, AVX2, disable, 18);
>             }
>           break;
>         case 19:
> diff --git a/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def b/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def
> index d7c93f00c5..1530d594b3 100644
> --- a/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def
> +++ b/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def
> @@ -32,5 +32,4 @@ BIT (Prefer_ERMS)
>  BIT (Prefer_No_AVX512)
>  BIT (MathVec_Prefer_No_AVX512)
>  BIT (Prefer_FSRM)
> -BIT (Prefer_AVX2_STRCMP)
>  BIT (Avoid_Short_Distance_REP_MOVSB)
> diff --git a/sysdeps/x86_64/multiarch/strcmp.c b/sysdeps/x86_64/multiarch/strcmp.c
> index 62b7abeeee..7c2901bf44 100644
> --- a/sysdeps/x86_64/multiarch/strcmp.c
> +++ b/sysdeps/x86_64/multiarch/strcmp.c
> @@ -43,8 +43,7 @@ IFUNC_SELECTOR (void)
>      {
>        if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
>           && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)
> -         && CPU_FEATURE_USABLE_P (cpu_features, BMI2)
> -         && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_AVX2_STRCMP))
> +         && CPU_FEATURE_USABLE_P (cpu_features, BMI2))
>         return OPTIMIZE (evex);
>
>        if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
> diff --git a/sysdeps/x86_64/multiarch/strncmp.c b/sysdeps/x86_64/multiarch/strncmp.c
> index 60ba0fe356..f94a421784 100644
> --- a/sysdeps/x86_64/multiarch/strncmp.c
> +++ b/sysdeps/x86_64/multiarch/strncmp.c
> @@ -43,8 +43,7 @@ IFUNC_SELECTOR (void)
>      {
>        if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
>           && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)
> -         && CPU_FEATURE_USABLE_P (cpu_features, BMI2)
> -         && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_AVX2_STRCMP))
> +         && CPU_FEATURE_USABLE_P (cpu_features, BMI2))
>         return OPTIMIZE (evex);
>
>        if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
> --
> 2.33.1
>

I would like to backport this patch to release branches.
Any comments or objections?

--Sunil