From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <skpgkp2@gmail.com>
Received: from mail-qt1-x82f.google.com (mail-qt1-x82f.google.com
 [IPv6:2607:f8b0:4864:20::82f])
 by sourceware.org (Postfix) with ESMTPS id A51803858C2C;
 Sat, 23 Apr 2022 01:45:09 +0000 (GMT)
DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org A51803858C2C
Received: by mail-qt1-x82f.google.com with SMTP id x24so6725150qtq.11;
 Fri, 22 Apr 2022 18:45:09 -0700 (PDT)
X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed;
 d=1e100.net; s=20210112;
 h=x-gm-message-state:mime-version:references:in-reply-to:from:date
 :message-id:subject:to:cc;
 bh=9Ma05GU67ELvVz4EKuBgmfoI37dTaEJ/TRmICBOTeL0=;
 b=YMXQl1WVYYGKyy+yiSWM2acAnmw4oF5jL3NmayCVmPYIm6d6/aNankFSSeltoPZcKw
 AlKCDos+P93g+JW1LK7K6fG2oR/OUx2vNZWiNmKv3K4WPUrlhFeWJVNK7iqVgSuG8ZWn
 D3SUsNY8YGG0I4xY0vRWls2JF6fElHJP0IUqu9qpNturvzAGLUw8IXVgx6QP433YQOLJ
 WvIGUyo1qn8CZJwIhBYBBWTrGnoeUheNqxJRwvNxSoRKGAvYaGauW7eRbJs4svTUJvCn
 cuG0yNvVC/PpgDx9q9BubzJnjtODIlmtiu0REWxHWVXGfTlf7f5bQLoIBAt9XHaWFVFL
 NEdg==
X-Gm-Message-State: AOAM533Lswmy904FvIBJI+S145INPyVGaEtUFlClV8/p5eCSTGV6y3wx
 TR8gAvdkA86lPlEqMO4zUYH7DLKBlYw41lXurtQRjVPu10g=
X-Google-Smtp-Source: ABdhPJyHXuZ2bqaIGo2rMx8k1p3gbWuzQwvti/Z/JyoU7YRC8s3iRhT1f5X6acNfWNzwD+H0yaW5P6DPIHRMiG7Edg4=
X-Received: by 2002:a05:622a:1492:b0:2f1:f0a6:df02 with SMTP id
 t18-20020a05622a149200b002f1f0a6df02mr5376102qtx.282.1650678308650; Fri, 22
 Apr 2022 18:45:08 -0700 (PDT)
MIME-Version: 1.0
References: <20211111002714.2527889-1-goldstein.w.n@gmail.com>
 <CAMe9rOrssV7DAf3gHwAapLYBcy_1orXt+BtxOixbYM6=pzuLWA@mail.gmail.com>
 <CAFUsyf+DGMa6STkhT1B6O4xRZDuyKP-bbD7PFObyLVFiwjU-yQ@mail.gmail.com>
In-Reply-To: <CAFUsyf+DGMa6STkhT1B6O4xRZDuyKP-bbD7PFObyLVFiwjU-yQ@mail.gmail.com>
From: Sunil Pandey <skpgkp2@gmail.com>
Date: Fri, 22 Apr 2022 18:44:32 -0700
Message-ID: <CAMAf5_er7ELRhJ7ryA0uMbaErTpNQC+MtV+-BmCtT10tyAEvvg@mail.gmail.com>
Subject: Re: [PATCH v1] x86: Shrink memcmp-sse4.S code size
To: Noah Goldstein <goldstein.w.n@gmail.com>, libc-stable@sourceware.org
Cc: "H.J. Lu" <hjl.tools@gmail.com>, GNU C Library <libc-alpha@sourceware.org>
Content-Type: text/plain; charset="UTF-8"
X-Spam-Status: No, score=-6.3 required=5.0 tests=BAYES_00, DKIM_SIGNED,
 DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_ENVFROM_END_DIGIT,
 FREEMAIL_FROM, GIT_PATCH_0, HK_RANDOM_ENVFROM, HK_RANDOM_FROM,
 RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS,
 TXREP autolearn=ham autolearn_force=no version=3.4.4
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on
 server2.sourceware.org
X-BeenThere: libc-stable@sourceware.org
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: Libc-stable mailing list <libc-stable.sourceware.org>
List-Unsubscribe: <https://sourceware.org/mailman/options/libc-stable>,
 <mailto:libc-stable-request@sourceware.org?subject=unsubscribe>
List-Archive: <https://sourceware.org/pipermail/libc-stable/>
List-Help: <mailto:libc-stable-request@sourceware.org?subject=help>
List-Subscribe: <https://sourceware.org/mailman/listinfo/libc-stable>,
 <mailto:libc-stable-request@sourceware.org?subject=subscribe>
X-List-Received-Date: Sat, 23 Apr 2022 01:45:14 -0000

On Wed, Nov 10, 2021 at 6:15 PM Noah Goldstein via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> On Wed, Nov 10, 2021 at 8:03 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > On Wed, Nov 10, 2021 at 4:27 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
> > >
> > > No bug.
> > >
> > > This implementation refactors memcmp-sse4.S primarily with minimizing
> > > code size in mind. It does this by removing the lookup table logic and
> > > removing the unrolled check from (256, 512] bytes.
> > >
> > > memcmp-sse4 code size reduction : -3487 bytes
> > > wmemcmp-sse4 code size reduction: -1472 bytes
> > >
> > > The current memcmp-sse4.S implementation has a large code size
> > > cost. This has serious adverse affects on the ICache / ITLB. While
> > > in micro-benchmarks the implementations appears fast, traces of
> > > real-world code have shown that the speed in micro benchmarks does not
> > > translate when the ICache/ITLB are not primed, and that the cost
> > > of the code size has measurable negative affects on overall
> > > application performance.
> > >
> > > See https://research.google/pubs/pub48320/ for more details.
> > > ---
> > >  sysdeps/x86_64/multiarch/memcmp-sse4.S | 2267 +++++++-----------------
> > >  1 file changed, 646 insertions(+), 1621 deletions(-)
> > >
> > > diff --git a/sysdeps/x86_64/multiarch/memcmp-sse4.S b/sysdeps/x86_64/multiarch/memcmp-sse4.S
> > > index b82adcd5fa..7686d1aa9e 100644
> > > --- a/sysdeps/x86_64/multiarch/memcmp-sse4.S
> > > +++ b/sysdeps/x86_64/multiarch/memcmp-sse4.S
> > > @@ -24,14 +24,14 @@
> > >  #  define MEMCMP       __memcmp_sse4_1
> > >  # endif
> > >
> > > -# define JMPTBL(I, B)  (I - B)
> > > +#ifdef USE_AS_WMEMCMP
> > > +# define CMPEQ pcmpeqd
> > > +# define CHAR_SIZE     4
> > > +#else
> > > +# define CMPEQ pcmpeqb
> > > +# define CHAR_SIZE     1
> > > +#endif
> > >
> > > -# define BRANCH_TO_JMPTBL_ENTRY(TABLE, INDEX, SCALE)           \
> > > -  lea          TABLE(%rip), %r11;                              \
> > > -  movslq       (%r11, INDEX, SCALE), %rcx;                     \
> > > -  add          %r11, %rcx;                                     \
> > > -  _CET_NOTRACK jmp *%rcx;                                      \
> > > -  ud2
> > >
> > >  /* Warning!
> > >             wmemcmp has to use SIGNED comparison for elements.
> > > @@ -46,33 +46,253 @@ ENTRY (MEMCMP)
> > >         /* Clear the upper 32 bits.  */
> > >         mov     %edx, %edx
> > >  # endif
> > > -       pxor    %xmm0, %xmm0
> > >         cmp     $79, %RDX_LP
> > >         ja      L(79bytesormore)
> > > +
> > > +       cmp     $CHAR_SIZE, %RDX_LP
> > > +       jbe     L(firstbyte)
> > > +
> > > +       /* N in (CHAR_SIZE, 79) bytes.  */
> > > +       cmpl    $32, %edx
> > > +       ja      L(more_32_bytes)
> > > +
> > > +       cmpl    $16, %edx
> > > +       jae     L(16_to_32_bytes)
> > > +
> > >  # ifndef USE_AS_WMEMCMP
> > > -       cmp     $1, %RDX_LP
> > > -       je      L(firstbyte)
> > > +       cmpl    $8, %edx
> > > +       jae     L(8_to_16_bytes)
> > > +
> > > +       cmpl    $4, %edx
> > > +       jb      L(2_to_3_bytes)
> > > +
> > > +       movl    (%rdi), %eax
> > > +       movl    (%rsi), %ecx
> > > +
> > > +       bswap   %eax
> > > +       bswap   %ecx
> > > +
> > > +       shlq    $32, %rax
> > > +       shlq    $32, %rcx
> > > +
> > > +       movl    -4(%rdi, %rdx), %edi
> > > +       movl    -4(%rsi, %rdx), %esi
> > > +
> > > +       bswap   %edi
> > > +       bswap   %esi
> > > +
> > > +       orq     %rdi, %rax
> > > +       orq     %rsi, %rcx
> > > +       subq    %rcx, %rax
> > > +       cmovne  %edx, %eax
> > > +       sbbl    %ecx, %ecx
> > > +       orl     %ecx, %eax
> > > +       ret
> > > +
> > > +       .p2align 4,, 8
> > > +L(2_to_3_bytes):
> > > +       movzwl  (%rdi), %eax
> > > +       movzwl  (%rsi), %ecx
> > > +       shll    $8, %eax
> > > +       shll    $8, %ecx
> > > +       bswap   %eax
> > > +       bswap   %ecx
> > > +       movzbl  -1(%rdi, %rdx), %edi
> > > +       movzbl  -1(%rsi, %rdx), %esi
> > > +       orl     %edi, %eax
> > > +       orl     %esi, %ecx
> > > +       subl    %ecx, %eax
> > > +       ret
> > > +
> > > +       .p2align 4,, 8
> > > +L(8_to_16_bytes):
> > > +       movq    (%rdi), %rax
> > > +       movq    (%rsi), %rcx
> > > +
> > > +       bswap   %rax
> > > +       bswap   %rcx
> > > +
> > > +       subq    %rcx, %rax
> > > +       jne     L(8_to_16_bytes_done)
> > > +
> > > +       movq    -8(%rdi, %rdx), %rax
> > > +       movq    -8(%rsi, %rdx), %rcx
> > > +
> > > +       bswap   %rax
> > > +       bswap   %rcx
> > > +
> > > +       subq    %rcx, %rax
> > > +
> > > +L(8_to_16_bytes_done):
> > > +       cmovne  %edx, %eax
> > > +       sbbl    %ecx, %ecx
> > > +       orl     %ecx, %eax
> > > +       ret
> > > +# else
> > > +       xorl    %eax, %eax
> > > +       movl    (%rdi), %ecx
> > > +       cmpl    (%rsi), %ecx
> > > +       jne     L(8_to_16_bytes_done)
> > > +       movl    4(%rdi), %ecx
> > > +       cmpl    4(%rsi), %ecx
> > > +       jne     L(8_to_16_bytes_done)
> > > +       movl    -4(%rdi, %rdx), %ecx
> > > +       cmpl    -4(%rsi, %rdx), %ecx
> > > +       jne     L(8_to_16_bytes_done)
> > > +       ret
> > >  # endif
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > >
> > > -# ifndef USE_AS_WMEMCMP
> > > -       .p2align 4
> > > +       .p2align 4,, 3
> > > +L(ret_zero):
> > > +       xorl    %eax, %eax
> > > +L(zero):
> > > +       ret
> > > +
> > > +       .p2align 4,, 8
> > >  L(firstbyte):
> > > +       jb      L(ret_zero)
> > > +# ifdef USE_AS_WMEMCMP
> > > +       xorl    %eax, %eax
> > > +       movl    (%rdi), %ecx
> > > +       cmpl    (%rsi), %ecx
> > > +       je      L(zero)
> > > +L(8_to_16_bytes_done):
> > > +       setg    %al
> > > +       leal    -1(%rax, %rax), %eax
> > > +# else
> > >         movzbl  (%rdi), %eax
> > >         movzbl  (%rsi), %ecx
> > >         sub     %ecx, %eax
> > > +# endif
> > >         ret
> > > +
> > > +       .p2align 4
> > > +L(vec_return_begin_48):
> > > +       addq    $16, %rdi
> > > +       addq    $16, %rsi
> > > +L(vec_return_begin_32):
> > > +       bsfl    %eax, %eax
> > > +# ifdef USE_AS_WMEMCMP
> > > +       movl    32(%rdi, %rax), %ecx
> > > +       xorl    %edx, %edx
> > > +       cmpl    32(%rsi, %rax), %ecx
> > > +       setg    %dl
> > > +       leal    -1(%rdx, %rdx), %eax
> > > +# else
> > > +       movzbl  32(%rsi, %rax), %ecx
> > > +       movzbl  32(%rdi, %rax), %eax
> > > +       subl    %ecx, %eax
> > > +# endif
> > > +       ret
> > > +
> > > +       .p2align 4
> > > +L(vec_return_begin_16):
> > > +       addq    $16, %rdi
> > > +       addq    $16, %rsi
> > > +L(vec_return_begin):
> > > +       bsfl    %eax, %eax
> > > +# ifdef USE_AS_WMEMCMP
> > > +       movl    (%rdi, %rax), %ecx
> > > +       xorl    %edx, %edx
> > > +       cmpl    (%rsi, %rax), %ecx
> > > +       setg    %dl
> > > +       leal    -1(%rdx, %rdx), %eax
> > > +# else
> > > +       movzbl  (%rsi, %rax), %ecx
> > > +       movzbl  (%rdi, %rax), %eax
> > > +       subl    %ecx, %eax
> > > +# endif
> > > +       ret
> > > +
> > > +       .p2align 4
> > > +L(vec_return_end_16):
> > > +       subl    $16, %edx
> > > +L(vec_return_end):
> > > +       bsfl    %eax, %eax
> > > +       addl    %edx, %eax
> > > +# ifdef USE_AS_WMEMCMP
> > > +       movl    -16(%rdi, %rax), %ecx
> > > +       xorl    %edx, %edx
> > > +       cmpl    -16(%rsi, %rax), %ecx
> > > +       setg    %dl
> > > +       leal    -1(%rdx, %rdx), %eax
> > > +# else
> > > +       movzbl  -16(%rsi, %rax), %ecx
> > > +       movzbl  -16(%rdi, %rax), %eax
> > > +       subl    %ecx, %eax
> > >  # endif
> > > +       ret
> > > +
> > > +       .p2align 4,, 8
> > > +L(more_32_bytes):
> > > +       movdqu  (%rdi), %xmm0
> > > +       movdqu  (%rsi), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqu  16(%rdi), %xmm0
> > > +       movdqu  16(%rsi), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       cmpl    $64, %edx
> > > +       jbe     L(32_to_64_bytes)
> > > +       movdqu  32(%rdi), %xmm0
> > > +       movdqu  32(%rsi), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_32)
> > > +
> > > +       .p2align 4,, 6
> > > +L(32_to_64_bytes):
> > > +       movdqu  -32(%rdi, %rdx), %xmm0
> > > +       movdqu  -32(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end_16)
> > > +
> > > +       movdqu  -16(%rdi, %rdx), %xmm0
> > > +       movdqu  -16(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end)
> > > +       ret
> > > +
> > > +       .p2align 4
> > > +L(16_to_32_bytes):
> > > +       movdqu  (%rdi), %xmm0
> > > +       movdqu  (%rsi), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqu  -16(%rdi, %rdx), %xmm0
> > > +       movdqu  -16(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end)
> > > +       ret
> > > +
> > >
> > >         .p2align 4
> > >  L(79bytesormore):
> > > +       movdqu  (%rdi), %xmm0
> > >         movdqu  (%rsi), %xmm1
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +
> > >         mov     %rsi, %rcx
> > >         and     $-16, %rsi
> > >         add     $16, %rsi
> > > @@ -85,1694 +305,499 @@ L(79bytesormore):
> > >
> > >         cmp     $128, %rdx
> > >         ja      L(128bytesormore)
> > > -L(less128bytes):
> > > -       sub     $64, %rdx
> > > -
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > >
> > > -       movdqu  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -
> > > -       movdqu  32(%rdi), %xmm2
> > > -       pxor    32(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(48bytesin256)
> > > -
> > > -       movdqu  48(%rdi), %xmm2
> > > -       pxor    48(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(64bytesin256)
> > > -       cmp     $32, %rdx
> > > -       jb      L(less32bytesin64)
> > > -
> > > -       movdqu  64(%rdi), %xmm2
> > > -       pxor    64(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(80bytesin256)
> > > -
> > > -       movdqu  80(%rdi), %xmm2
> > > -       pxor    80(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(96bytesin256)
> > > -       sub     $32, %rdx
> > > -       add     $32, %rdi
> > > -       add     $32, %rsi
> > > -L(less32bytesin64):
> > > -       add     $64, %rdi
> > > -       add     $64, %rsi
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > +       .p2align 4,, 6
> > > +L(less128bytes):
> > > +       movdqu  (%rdi), %xmm1
> > > +       CMPEQ   (%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqu  16(%rdi), %xmm1
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       movdqu  32(%rdi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_32)
> > > +
> > > +       movdqu  48(%rdi), %xmm1
> > > +       CMPEQ   48(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_48)
> > > +
> > > +       cmp     $96, %rdx
> > > +       jb      L(32_to_64_bytes)
> > > +
> > > +       addq    $64, %rdi
> > > +       addq    $64, %rsi
> > > +       subq    $64, %rdx
> > > +
> > > +       .p2align 4,, 6
> > > +L(last_64_bytes):
> > > +       movdqu  (%rdi), %xmm1
> > > +       CMPEQ   (%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqu  16(%rdi), %xmm1
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       movdqu  -32(%rdi, %rdx), %xmm0
> > > +       movdqu  -32(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end_16)
> > > +
> > > +       movdqu  -16(%rdi, %rdx), %xmm0
> > > +       movdqu  -16(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end)
> > > +       ret
> > >
> > > +       .p2align 4
> > >  L(128bytesormore):
> > > -       cmp     $512, %rdx
> > > -       ja      L(512bytesormore)
> > >         cmp     $256, %rdx
> > > -       ja      L(less512bytes)
> > > +       ja      L(unaligned_loop)
> > >  L(less256bytes):
> > > -       sub     $128, %rdx
> > > -
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqu  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -
> > > -       movdqu  32(%rdi), %xmm2
> > > -       pxor    32(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(48bytesin256)
> > > -
> > > -       movdqu  48(%rdi), %xmm2
> > > -       pxor    48(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(64bytesin256)
> > > -
> > > -       movdqu  64(%rdi), %xmm2
> > > -       pxor    64(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(80bytesin256)
> > > -
> > > -       movdqu  80(%rdi), %xmm2
> > > -       pxor    80(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(96bytesin256)
> > > -
> > > -       movdqu  96(%rdi), %xmm2
> > > -       pxor    96(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(112bytesin256)
> > > -
> > > -       movdqu  112(%rdi), %xmm2
> > > -       pxor    112(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(128bytesin256)
> > > -
> > > -       add     $128, %rsi
> > > -       add     $128, %rdi
> > > -
> > > -       cmp     $64, %rdx
> > > -       jae     L(less128bytes)
> > > -
> > > -       cmp     $32, %rdx
> > > -       jb      L(less32bytesin128)
> > > -
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqu  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -       sub     $32, %rdx
> > > -       add     $32, %rdi
> > > -       add     $32, %rsi
> > > -L(less32bytesin128):
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > -
> > > -L(less512bytes):
> > > -       sub     $256, %rdx
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqu  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -
> > > -       movdqu  32(%rdi), %xmm2
> > > -       pxor    32(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(48bytesin256)
> > > -
> > > -       movdqu  48(%rdi), %xmm2
> > > -       pxor    48(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(64bytesin256)
> > > -
> > > -       movdqu  64(%rdi), %xmm2
> > > -       pxor    64(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(80bytesin256)
> > > -
> > > -       movdqu  80(%rdi), %xmm2
> > > -       pxor    80(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(96bytesin256)
> > > -
> > > -       movdqu  96(%rdi), %xmm2
> > > -       pxor    96(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(112bytesin256)
> > > -
> > > -       movdqu  112(%rdi), %xmm2
> > > -       pxor    112(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(128bytesin256)
> > > -
> > > -       movdqu  128(%rdi), %xmm2
> > > -       pxor    128(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(144bytesin256)
> > > -
> > > -       movdqu  144(%rdi), %xmm2
> > > -       pxor    144(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(160bytesin256)
> > > -
> > > -       movdqu  160(%rdi), %xmm2
> > > -       pxor    160(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(176bytesin256)
> > > -
> > > -       movdqu  176(%rdi), %xmm2
> > > -       pxor    176(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(192bytesin256)
> > > -
> > > -       movdqu  192(%rdi), %xmm2
> > > -       pxor    192(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(208bytesin256)
> > > -
> > > -       movdqu  208(%rdi), %xmm2
> > > -       pxor    208(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(224bytesin256)
> > > -
> > > -       movdqu  224(%rdi), %xmm2
> > > -       pxor    224(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(240bytesin256)
> > > -
> > > -       movdqu  240(%rdi), %xmm2
> > > -       pxor    240(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(256bytesin256)
> > > -
> > > -       add     $256, %rsi
> > > -       add     $256, %rdi
> > > -
> > > -       cmp     $128, %rdx
> > > -       jae     L(less256bytes)
> > > +       movdqu  (%rdi), %xmm1
> > > +       CMPEQ   (%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqu  16(%rdi), %xmm1
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       movdqu  32(%rdi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_32)
> > > +
> > > +       movdqu  48(%rdi), %xmm1
> > > +       CMPEQ   48(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_48)
> > > +
> > > +       addq    $64, %rdi
> > > +       addq    $64, %rsi
> > > +
> > > +       movdqu  (%rdi), %xmm1
> > > +       CMPEQ   (%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqu  16(%rdi), %xmm1
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       movdqu  32(%rdi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_32)
> > > +
> > > +       movdqu  48(%rdi), %xmm1
> > > +       CMPEQ   48(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_48)
> > > +
> > > +       addq    $-128, %rdx
> > > +       subq    $-64, %rsi
> > > +       subq    $-64, %rdi
> > >
> > >         cmp     $64, %rdx
> > > -       jae     L(less128bytes)
> > > +       ja      L(less128bytes)
> > >
> > >         cmp     $32, %rdx
> > > -       jb      L(less32bytesin256)
> > > -
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqu  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -       sub     $32, %rdx
> > > -       add     $32, %rdi
> > > -       add     $32, %rsi
> > > -L(less32bytesin256):
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > +       ja      L(last_64_bytes)
> > > +
> > > +       movdqu  -32(%rdi, %rdx), %xmm0
> > > +       movdqu  -32(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end_16)
> > > +
> > > +       movdqu  -16(%rdi, %rdx), %xmm0
> > > +       movdqu  -16(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end)
> > > +       ret
> > >
> > >         .p2align 4
> > > -L(512bytesormore):
> > > +L(unaligned_loop):
> > >  # ifdef DATA_CACHE_SIZE_HALF
> > >         mov     $DATA_CACHE_SIZE_HALF, %R8_LP
> > >  # else
> > >         mov     __x86_data_cache_size_half(%rip), %R8_LP
> > >  # endif
> > > -       mov     %r8, %r9
> > > -       shr     $1, %r8
> > > -       add     %r9, %r8
> > > -       cmp     %r8, %rdx
> > > -       ja      L(L2_L3_cache_unaglined)
> > > +       movq    %r8, %r9
> > > +       addq    %r8, %r8
> > > +       addq    %r9, %r8
> > > +       cmpq    %r8, %rdx
> > > +       ja      L(L2_L3_cache_unaligned)
> > >         sub     $64, %rdx
> > >         .p2align 4
> > >  L(64bytesormore_loop):
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       movdqa  %xmm2, %xmm1
> > > +       movdqu  (%rdi), %xmm0
> > > +       movdqu  16(%rdi), %xmm1
> > > +       movdqu  32(%rdi), %xmm2
> > > +       movdqu  48(%rdi), %xmm3
> > >
> > > -       movdqu  16(%rdi), %xmm3
> > > -       pxor    16(%rsi), %xmm3
> > > -       por     %xmm3, %xmm1
> > > +       CMPEQ   (%rsi), %xmm0
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm2
> > > +       CMPEQ   48(%rsi), %xmm3
> > >
> > > -       movdqu  32(%rdi), %xmm4
> > > -       pxor    32(%rsi), %xmm4
> > > -       por     %xmm4, %xmm1
> > > +       pand    %xmm0, %xmm1
> > > +       pand    %xmm2, %xmm3
> > > +       pand    %xmm1, %xmm3
> > >
> > > -       movdqu  48(%rdi), %xmm5
> > > -       pxor    48(%rsi), %xmm5
> > > -       por     %xmm5, %xmm1
> > > +       pmovmskb %xmm3, %eax
> > > +       incw    %ax
> > > +       jnz     L(64bytesormore_loop_end)
> > >
> > > -       ptest   %xmm1, %xmm0
> > > -       jnc     L(64bytesormore_loop_end)
> > >         add     $64, %rsi
> > >         add     $64, %rdi
> > >         sub     $64, %rdx
> > > -       jae     L(64bytesormore_loop)
> > > +       ja      L(64bytesormore_loop)
> > >
> > > -       add     $64, %rdx
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > +       .p2align 4,, 6
> > > +L(loop_tail):
> > > +       addq    %rdx, %rdi
> > > +       movdqu  (%rdi), %xmm0
> > > +       movdqu  16(%rdi), %xmm1
> > > +       movdqu  32(%rdi), %xmm2
> > > +       movdqu  48(%rdi), %xmm3
> > > +
> > > +       addq    %rdx, %rsi
> > > +       movdqu  (%rsi), %xmm4
> > > +       movdqu  16(%rsi), %xmm5
> > > +       movdqu  32(%rsi), %xmm6
> > > +       movdqu  48(%rsi), %xmm7
> > > +
> > > +       CMPEQ   %xmm4, %xmm0
> > > +       CMPEQ   %xmm5, %xmm1
> > > +       CMPEQ   %xmm6, %xmm2
> > > +       CMPEQ   %xmm7, %xmm3
> > > +
> > > +       pand    %xmm0, %xmm1
> > > +       pand    %xmm2, %xmm3
> > > +       pand    %xmm1, %xmm3
> > > +
> > > +       pmovmskb %xmm3, %eax
> > > +       incw    %ax
> > > +       jnz     L(64bytesormore_loop_end)
> > > +       ret
> > >
> > > -L(L2_L3_cache_unaglined):
> > > -       sub     $64, %rdx
> > > +L(L2_L3_cache_unaligned):
> > > +       subq    $64, %rdx
> > >         .p2align 4
> > >  L(L2_L3_unaligned_128bytes_loop):
> > >         prefetchnta 0x1c0(%rdi)
> > >         prefetchnta 0x1c0(%rsi)
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       movdqa  %xmm2, %xmm1
> > >
> > > -       movdqu  16(%rdi), %xmm3
> > > -       pxor    16(%rsi), %xmm3
> > > -       por     %xmm3, %xmm1
> > > +       movdqu  (%rdi), %xmm0
> > > +       movdqu  16(%rdi), %xmm1
> > > +       movdqu  32(%rdi), %xmm2
> > > +       movdqu  48(%rdi), %xmm3
> > > +
> > > +       CMPEQ   (%rsi), %xmm0
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm2
> > > +       CMPEQ   48(%rsi), %xmm3
> > >
> > > -       movdqu  32(%rdi), %xmm4
> > > -       pxor    32(%rsi), %xmm4
> > > -       por     %xmm4, %xmm1
> > > +       pand    %xmm0, %xmm1
> > > +       pand    %xmm2, %xmm3
> > > +       pand    %xmm1, %xmm3
> > >
> > > -       movdqu  48(%rdi), %xmm5
> > > -       pxor    48(%rsi), %xmm5
> > > -       por     %xmm5, %xmm1
> > > +       pmovmskb %xmm3, %eax
> > > +       incw    %ax
> > > +       jnz     L(64bytesormore_loop_end)
> > >
> > > -       ptest   %xmm1, %xmm0
> > > -       jnc     L(64bytesormore_loop_end)
> > >         add     $64, %rsi
> > >         add     $64, %rdi
> > >         sub     $64, %rdx
> > > -       jae     L(L2_L3_unaligned_128bytes_loop)
> > > +       ja      L(L2_L3_unaligned_128bytes_loop)
> > > +       jmp     L(loop_tail)
> > >
> > > -       add     $64, %rdx
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > >
> > > -/*
> > > - * This case is for machines which are sensitive for unaligned instructions.
> > > - */
> > > +       /* This case is for machines which are sensitive for unaligned
> > > +        * instructions.  */
> > >         .p2align 4
> > >  L(2aligned):
> > >         cmp     $128, %rdx
> > >         ja      L(128bytesormorein2aligned)
> > >  L(less128bytesin2aligned):
> > > -       sub     $64, %rdx
> > > -
> > > -       movdqa  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqa  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -
> > > -       movdqa  32(%rdi), %xmm2
> > > -       pxor    32(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(48bytesin256)
> > > -
> > > -       movdqa  48(%rdi), %xmm2
> > > -       pxor    48(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(64bytesin256)
> > > -       cmp     $32, %rdx
> > > -       jb      L(less32bytesin64in2alinged)
> > > -
> > > -       movdqa  64(%rdi), %xmm2
> > > -       pxor    64(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(80bytesin256)
> > > -
> > > -       movdqa  80(%rdi), %xmm2
> > > -       pxor    80(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(96bytesin256)
> > > -       sub     $32, %rdx
> > > -       add     $32, %rdi
> > > -       add     $32, %rsi
> > > -L(less32bytesin64in2alinged):
> > > -       add     $64, %rdi
> > > -       add     $64, %rsi
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > +       movdqa  (%rdi), %xmm1
> > > +       CMPEQ   (%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqa  16(%rdi), %xmm1
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       movdqa  32(%rdi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_32)
> > > +
> > > +       movdqa  48(%rdi), %xmm1
> > > +       CMPEQ   48(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_48)
> > > +
> > > +       cmp     $96, %rdx
> > > +       jb      L(32_to_64_bytes)
> > > +
> > > +       addq    $64, %rdi
> > > +       addq    $64, %rsi
> > > +       subq    $64, %rdx
> > > +
> > > +       .p2align 4,, 6
> > > +L(aligned_last_64_bytes):
> > > +       movdqa  (%rdi), %xmm1
> > > +       CMPEQ   (%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqa  16(%rdi), %xmm1
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       movdqu  -32(%rdi, %rdx), %xmm0
> > > +       movdqu  -32(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end_16)
> > > +
> > > +       movdqu  -16(%rdi, %rdx), %xmm0
> > > +       movdqu  -16(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end)
> > > +       ret
> > >
> > >         .p2align 4
> > >  L(128bytesormorein2aligned):
> > > -       cmp     $512, %rdx
> > > -       ja      L(512bytesormorein2aligned)
> > >         cmp     $256, %rdx
> > > -       ja      L(256bytesormorein2aligned)
> > > +       ja      L(aligned_loop)
> > >  L(less256bytesin2alinged):
> > > -       sub     $128, %rdx
> > > -
> > > -       movdqa  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqa  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -
> > > -       movdqa  32(%rdi), %xmm2
> > > -       pxor    32(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(48bytesin256)
> > > -
> > > -       movdqa  48(%rdi), %xmm2
> > > -       pxor    48(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(64bytesin256)
> > > -
> > > -       movdqa  64(%rdi), %xmm2
> > > -       pxor    64(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(80bytesin256)
> > > -
> > > -       movdqa  80(%rdi), %xmm2
> > > -       pxor    80(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(96bytesin256)
> > > -
> > > -       movdqa  96(%rdi), %xmm2
> > > -       pxor    96(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(112bytesin256)
> > > -
> > > -       movdqa  112(%rdi), %xmm2
> > > -       pxor    112(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(128bytesin256)
> > > -
> > > -       add     $128, %rsi
> > > -       add     $128, %rdi
> > > +       movdqa  (%rdi), %xmm1
> > > +       CMPEQ   (%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqa  16(%rdi), %xmm1
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       movdqa  32(%rdi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_32)
> > > +
> > > +       movdqa  48(%rdi), %xmm1
> > > +       CMPEQ   48(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_48)
> > > +
> > > +       addq    $64, %rdi
> > > +       addq    $64, %rsi
> > > +
> > > +       movdqa  (%rdi), %xmm1
> > > +       CMPEQ   (%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin)
> > > +
> > > +       movdqa  16(%rdi), %xmm1
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_16)
> > > +
> > > +       movdqa  32(%rdi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_32)
> > > +
> > > +       movdqa  48(%rdi), %xmm1
> > > +       CMPEQ   48(%rsi), %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_begin_48)
> > > +
> > > +       addq    $-128, %rdx
> > > +       subq    $-64, %rsi
> > > +       subq    $-64, %rdi
> > >
> > >         cmp     $64, %rdx
> > > -       jae     L(less128bytesin2aligned)
> > > +       ja      L(less128bytesin2aligned)
> > >
> > >         cmp     $32, %rdx
> > > -       jb      L(less32bytesin128in2aligned)
> > > -
> > > -       movdqu  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqu  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -       sub     $32, %rdx
> > > -       add     $32, %rdi
> > > -       add     $32, %rsi
> > > -L(less32bytesin128in2aligned):
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > -
> > > -       .p2align 4
> > > -L(256bytesormorein2aligned):
> > > -
> > > -       sub     $256, %rdx
> > > -       movdqa  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqa  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -
> > > -       movdqa  32(%rdi), %xmm2
> > > -       pxor    32(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(48bytesin256)
> > > -
> > > -       movdqa  48(%rdi), %xmm2
> > > -       pxor    48(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(64bytesin256)
> > > -
> > > -       movdqa  64(%rdi), %xmm2
> > > -       pxor    64(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(80bytesin256)
> > > -
> > > -       movdqa  80(%rdi), %xmm2
> > > -       pxor    80(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(96bytesin256)
> > > -
> > > -       movdqa  96(%rdi), %xmm2
> > > -       pxor    96(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(112bytesin256)
> > > -
> > > -       movdqa  112(%rdi), %xmm2
> > > -       pxor    112(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(128bytesin256)
> > > -
> > > -       movdqa  128(%rdi), %xmm2
> > > -       pxor    128(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(144bytesin256)
> > > -
> > > -       movdqa  144(%rdi), %xmm2
> > > -       pxor    144(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(160bytesin256)
> > > -
> > > -       movdqa  160(%rdi), %xmm2
> > > -       pxor    160(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(176bytesin256)
> > > -
> > > -       movdqa  176(%rdi), %xmm2
> > > -       pxor    176(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(192bytesin256)
> > > -
> > > -       movdqa  192(%rdi), %xmm2
> > > -       pxor    192(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(208bytesin256)
> > > -
> > > -       movdqa  208(%rdi), %xmm2
> > > -       pxor    208(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(224bytesin256)
> > > -
> > > -       movdqa  224(%rdi), %xmm2
> > > -       pxor    224(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(240bytesin256)
> > > -
> > > -       movdqa  240(%rdi), %xmm2
> > > -       pxor    240(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(256bytesin256)
> > > -
> > > -       add     $256, %rsi
> > > -       add     $256, %rdi
> > > -
> > > -       cmp     $128, %rdx
> > > -       jae     L(less256bytesin2alinged)
> > > -
> > > -       cmp     $64, %rdx
> > > -       jae     L(less128bytesin2aligned)
> > > -
> > > -       cmp     $32, %rdx
> > > -       jb      L(less32bytesin256in2alinged)
> > > -
> > > -       movdqa  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytesin256)
> > > -
> > > -       movdqa  16(%rdi), %xmm2
> > > -       pxor    16(%rsi), %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(32bytesin256)
> > > -       sub     $32, %rdx
> > > -       add     $32, %rdi
> > > -       add     $32, %rsi
> > > -L(less32bytesin256in2alinged):
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > +       ja      L(aligned_last_64_bytes)
> > > +
> > > +       movdqu  -32(%rdi, %rdx), %xmm0
> > > +       movdqu  -32(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end_16)
> > > +
> > > +       movdqu  -16(%rdi, %rdx), %xmm0
> > > +       movdqu  -16(%rsi, %rdx), %xmm1
> > > +       CMPEQ   %xmm0, %xmm1
> > > +       pmovmskb %xmm1, %eax
> > > +       incw    %ax
> > > +       jnz     L(vec_return_end)
> > > +       ret
> > >
> > >         .p2align 4
> > > -L(512bytesormorein2aligned):
> > > +L(aligned_loop):
> > >  # ifdef DATA_CACHE_SIZE_HALF
> > >         mov     $DATA_CACHE_SIZE_HALF, %R8_LP
> > >  # else
> > >         mov     __x86_data_cache_size_half(%rip), %R8_LP
> > >  # endif
> > > -       mov     %r8, %r9
> > > -       shr     $1, %r8
> > > -       add     %r9, %r8
> > > -       cmp     %r8, %rdx
> > > -       ja      L(L2_L3_cache_aglined)
> > > +       movq    %r8, %r9
> > > +       addq    %r8, %r8
> > > +       addq    %r9, %r8
> > > +       cmpq    %r8, %rdx
> > > +       ja      L(L2_L3_cache_aligned)
> > >
> > >         sub     $64, %rdx
> > >         .p2align 4
> > >  L(64bytesormore_loopin2aligned):
> > > -       movdqa  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       movdqa  %xmm2, %xmm1
> > > -
> > > -       movdqa  16(%rdi), %xmm3
> > > -       pxor    16(%rsi), %xmm3
> > > -       por     %xmm3, %xmm1
> > > +       movdqa  (%rdi), %xmm0
> > > +       movdqa  16(%rdi), %xmm1
> > > +       movdqa  32(%rdi), %xmm2
> > > +       movdqa  48(%rdi), %xmm3
> > >
> > > -       movdqa  32(%rdi), %xmm4
> > > -       pxor    32(%rsi), %xmm4
> > > -       por     %xmm4, %xmm1
> > > +       CMPEQ   (%rsi), %xmm0
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm2
> > > +       CMPEQ   48(%rsi), %xmm3
> > >
> > > -       movdqa  48(%rdi), %xmm5
> > > -       pxor    48(%rsi), %xmm5
> > > -       por     %xmm5, %xmm1
> > > +       pand    %xmm0, %xmm1
> > > +       pand    %xmm2, %xmm3
> > > +       pand    %xmm1, %xmm3
> > >
> > > -       ptest   %xmm1, %xmm0
> > > -       jnc     L(64bytesormore_loop_end)
> > > +       pmovmskb %xmm3, %eax
> > > +       incw    %ax
> > > +       jnz     L(64bytesormore_loop_end)
> > >         add     $64, %rsi
> > >         add     $64, %rdi
> > >         sub     $64, %rdx
> > > -       jae     L(64bytesormore_loopin2aligned)
> > > -
> > > -       add     $64, %rdx
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > -L(L2_L3_cache_aglined):
> > > -       sub     $64, %rdx
> > > +       ja      L(64bytesormore_loopin2aligned)
> > > +       jmp     L(loop_tail)
> > >
> > > +L(L2_L3_cache_aligned):
> > > +       subq    $64, %rdx
> > >         .p2align 4
> > >  L(L2_L3_aligned_128bytes_loop):
> > >         prefetchnta 0x1c0(%rdi)
> > >         prefetchnta 0x1c0(%rsi)
> > > -       movdqa  (%rdi), %xmm2
> > > -       pxor    (%rsi), %xmm2
> > > -       movdqa  %xmm2, %xmm1
> > > -
> > > -       movdqa  16(%rdi), %xmm3
> > > -       pxor    16(%rsi), %xmm3
> > > -       por     %xmm3, %xmm1
> > > +       movdqa  (%rdi), %xmm0
> > > +       movdqa  16(%rdi), %xmm1
> > > +       movdqa  32(%rdi), %xmm2
> > > +       movdqa  48(%rdi), %xmm3
> > >
> > > -       movdqa  32(%rdi), %xmm4
> > > -       pxor    32(%rsi), %xmm4
> > > -       por     %xmm4, %xmm1
> > > +       CMPEQ   (%rsi), %xmm0
> > > +       CMPEQ   16(%rsi), %xmm1
> > > +       CMPEQ   32(%rsi), %xmm2
> > > +       CMPEQ   48(%rsi), %xmm3
> > >
> > > -       movdqa  48(%rdi), %xmm5
> > > -       pxor    48(%rsi), %xmm5
> > > -       por     %xmm5, %xmm1
> > > +       pand    %xmm0, %xmm1
> > > +       pand    %xmm2, %xmm3
> > > +       pand    %xmm1, %xmm3
> > >
> > > -       ptest   %xmm1, %xmm0
> > > -       jnc     L(64bytesormore_loop_end)
> > > -       add     $64, %rsi
> > > -       add     $64, %rdi
> > > -       sub     $64, %rdx
> > > -       jae     L(L2_L3_aligned_128bytes_loop)
> > > -
> > > -       add     $64, %rdx
> > > -       add     %rdx, %rsi
> > > -       add     %rdx, %rdi
> > > -       BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
> > > +       pmovmskb %xmm3, %eax
> > > +       incw    %ax
> > > +       jnz     L(64bytesormore_loop_end)
> > >
> > > +       addq    $64, %rsi
> > > +       addq    $64, %rdi
> > > +       subq    $64, %rdx
> > > +       ja      L(L2_L3_aligned_128bytes_loop)
> > > +       jmp     L(loop_tail)
> > >
> > >         .p2align 4
> > >  L(64bytesormore_loop_end):
> > > -       add     $16, %rdi
> > > -       add     $16, %rsi
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(16bytes)
> > > -
> > > -       add     $16, %rdi
> > > -       add     $16, %rsi
> > > -       ptest   %xmm3, %xmm0
> > > -       jnc     L(16bytes)
> > > -
> > > -       add     $16, %rdi
> > > -       add     $16, %rsi
> > > -       ptest   %xmm4, %xmm0
> > > -       jnc     L(16bytes)
> > > -
> > > -       add     $16, %rdi
> > > -       add     $16, %rsi
> > > -       jmp     L(16bytes)
> > > -
> > > -L(256bytesin256):
> > > -       add     $256, %rdi
> > > -       add     $256, %rsi
> > > -       jmp     L(16bytes)
> > > -L(240bytesin256):
> > > -       add     $240, %rdi
> > > -       add     $240, %rsi
> > > -       jmp     L(16bytes)
> > > -L(224bytesin256):
> > > -       add     $224, %rdi
> > > -       add     $224, %rsi
> > > -       jmp     L(16bytes)
> > > -L(208bytesin256):
> > > -       add     $208, %rdi
> > > -       add     $208, %rsi
> > > -       jmp     L(16bytes)
> > > -L(192bytesin256):
> > > -       add     $192, %rdi
> > > -       add     $192, %rsi
> > > -       jmp     L(16bytes)
> > > -L(176bytesin256):
> > > -       add     $176, %rdi
> > > -       add     $176, %rsi
> > > -       jmp     L(16bytes)
> > > -L(160bytesin256):
> > > -       add     $160, %rdi
> > > -       add     $160, %rsi
> > > -       jmp     L(16bytes)
> > > -L(144bytesin256):
> > > -       add     $144, %rdi
> > > -       add     $144, %rsi
> > > -       jmp     L(16bytes)
> > > -L(128bytesin256):
> > > -       add     $128, %rdi
> > > -       add     $128, %rsi
> > > -       jmp     L(16bytes)
> > > -L(112bytesin256):
> > > -       add     $112, %rdi
> > > -       add     $112, %rsi
> > > -       jmp     L(16bytes)
> > > -L(96bytesin256):
> > > -       add     $96, %rdi
> > > -       add     $96, %rsi
> > > -       jmp     L(16bytes)
> > > -L(80bytesin256):
> > > -       add     $80, %rdi
> > > -       add     $80, %rsi
> > > -       jmp     L(16bytes)
> > > -L(64bytesin256):
> > > -       add     $64, %rdi
> > > -       add     $64, %rsi
> > > -       jmp     L(16bytes)
> > > -L(48bytesin256):
> > > -       add     $16, %rdi
> > > -       add     $16, %rsi
> > > -L(32bytesin256):
> > > -       add     $16, %rdi
> > > -       add     $16, %rsi
> > > -L(16bytesin256):
> > > -       add     $16, %rdi
> > > -       add     $16, %rsi
> > > -L(16bytes):
> > > -       mov     -16(%rdi), %rax
> > > -       mov     -16(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -L(8bytes):
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(12bytes):
> > > -       mov     -12(%rdi), %rax
> > > -       mov     -12(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -L(4bytes):
> > > -       mov     -4(%rsi), %ecx
> > > -# ifndef USE_AS_WMEMCMP
> > > -       mov     -4(%rdi), %eax
> > > -       cmp     %eax, %ecx
> > > -# else
> > > -       cmp     -4(%rdi), %ecx
> > > -# endif
> > > -       jne     L(diffin4bytes)
> > > -L(0bytes):
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -# ifndef USE_AS_WMEMCMP
> > > -/* unreal case for wmemcmp */
> > > -       .p2align 4
> > > -L(65bytes):
> > > -       movdqu  -65(%rdi), %xmm1
> > > -       movdqu  -65(%rsi), %xmm2
> > > -       mov     $-65, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(49bytes):
> > > -       movdqu  -49(%rdi), %xmm1
> > > -       movdqu  -49(%rsi), %xmm2
> > > -       mov     $-49, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(33bytes):
> > > -       movdqu  -33(%rdi), %xmm1
> > > -       movdqu  -33(%rsi), %xmm2
> > > -       mov     $-33, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(17bytes):
> > > -       mov     -17(%rdi), %rax
> > > -       mov     -17(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -L(9bytes):
> > > -       mov     -9(%rdi), %rax
> > > -       mov     -9(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       movzbl  -1(%rdi), %eax
> > > -       movzbl  -1(%rsi), %edx
> > > -       sub     %edx, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(13bytes):
> > > -       mov     -13(%rdi), %rax
> > > -       mov     -13(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(5bytes):
> > > -       mov     -5(%rdi), %eax
> > > -       mov     -5(%rsi), %ecx
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin4bytes)
> > > -       movzbl  -1(%rdi), %eax
> > > -       movzbl  -1(%rsi), %edx
> > > -       sub     %edx, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(66bytes):
> > > -       movdqu  -66(%rdi), %xmm1
> > > -       movdqu  -66(%rsi), %xmm2
> > > -       mov     $-66, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(50bytes):
> > > -       movdqu  -50(%rdi), %xmm1
> > > -       movdqu  -50(%rsi), %xmm2
> > > -       mov     $-50, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(34bytes):
> > > -       movdqu  -34(%rdi), %xmm1
> > > -       movdqu  -34(%rsi), %xmm2
> > > -       mov     $-34, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(18bytes):
> > > -       mov     -18(%rdi), %rax
> > > -       mov     -18(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -L(10bytes):
> > > -       mov     -10(%rdi), %rax
> > > -       mov     -10(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       movzwl  -2(%rdi), %eax
> > > -       movzwl  -2(%rsi), %ecx
> > > -       cmp     %cl, %al
> > > -       jne     L(end)
> > > -       and     $0xffff, %eax
> > > -       and     $0xffff, %ecx
> > > -       sub     %ecx, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(14bytes):
> > > -       mov     -14(%rdi), %rax
> > > -       mov     -14(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(6bytes):
> > > -       mov     -6(%rdi), %eax
> > > -       mov     -6(%rsi), %ecx
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin4bytes)
> > > -L(2bytes):
> > > -       movzwl  -2(%rsi), %ecx
> > > -       movzwl  -2(%rdi), %eax
> > > -       cmp     %cl, %al
> > > -       jne     L(end)
> > > -       and     $0xffff, %eax
> > > -       and     $0xffff, %ecx
> > > -       sub     %ecx, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(67bytes):
> > > -       movdqu  -67(%rdi), %xmm2
> > > -       movdqu  -67(%rsi), %xmm1
> > > -       mov     $-67, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(51bytes):
> > > -       movdqu  -51(%rdi), %xmm2
> > > -       movdqu  -51(%rsi), %xmm1
> > > -       mov     $-51, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(35bytes):
> > > -       movdqu  -35(%rsi), %xmm1
> > > -       movdqu  -35(%rdi), %xmm2
> > > -       mov     $-35, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(19bytes):
> > > -       mov     -19(%rdi), %rax
> > > -       mov     -19(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -L(11bytes):
> > > -       mov     -11(%rdi), %rax
> > > -       mov     -11(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     -4(%rdi), %eax
> > > -       mov     -4(%rsi), %ecx
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin4bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(15bytes):
> > > -       mov     -15(%rdi), %rax
> > > -       mov     -15(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(7bytes):
> > > -       mov     -7(%rdi), %eax
> > > -       mov     -7(%rsi), %ecx
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin4bytes)
> > > -       mov     -4(%rdi), %eax
> > > -       mov     -4(%rsi), %ecx
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin4bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(3bytes):
> > > -       movzwl  -3(%rdi), %eax
> > > -       movzwl  -3(%rsi), %ecx
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin2bytes)
> > > -L(1bytes):
> > > -       movzbl  -1(%rdi), %eax
> > > -       movzbl  -1(%rsi), %ecx
> > > -       sub     %ecx, %eax
> > > -       ret
> > > -# endif
> > > -
> > > -       .p2align 4
> > > -L(68bytes):
> > > -       movdqu  -68(%rdi), %xmm2
> > > -       movdqu  -68(%rsi), %xmm1
> > > -       mov     $-68, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(52bytes):
> > > -       movdqu  -52(%rdi), %xmm2
> > > -       movdqu  -52(%rsi), %xmm1
> > > -       mov     $-52, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(36bytes):
> > > -       movdqu  -36(%rdi), %xmm2
> > > -       movdqu  -36(%rsi), %xmm1
> > > -       mov     $-36, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(20bytes):
> > > -       movdqu  -20(%rdi), %xmm2
> > > -       movdqu  -20(%rsi), %xmm1
> > > -       mov     $-20, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -4(%rsi), %ecx
> > > -
> > > -# ifndef USE_AS_WMEMCMP
> > > -       mov     -4(%rdi), %eax
> > > -       cmp     %eax, %ecx
> > > -# else
> > > -       cmp     -4(%rdi), %ecx
> > > -# endif
> > > -       jne     L(diffin4bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -# ifndef USE_AS_WMEMCMP
> > > -/* unreal cases for wmemcmp */
> > > -       .p2align 4
> > > -L(69bytes):
> > > -       movdqu  -69(%rsi), %xmm1
> > > -       movdqu  -69(%rdi), %xmm2
> > > -       mov     $-69, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(53bytes):
> > > -       movdqu  -53(%rsi), %xmm1
> > > -       movdqu  -53(%rdi), %xmm2
> > > -       mov     $-53, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(37bytes):
> > > -       movdqu  -37(%rsi), %xmm1
> > > -       movdqu  -37(%rdi), %xmm2
> > > -       mov     $-37, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(21bytes):
> > > -       movdqu  -21(%rsi), %xmm1
> > > -       movdqu  -21(%rdi), %xmm2
> > > -       mov     $-21, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(70bytes):
> > > -       movdqu  -70(%rsi), %xmm1
> > > -       movdqu  -70(%rdi), %xmm2
> > > -       mov     $-70, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(54bytes):
> > > -       movdqu  -54(%rsi), %xmm1
> > > -       movdqu  -54(%rdi), %xmm2
> > > -       mov     $-54, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(38bytes):
> > > -       movdqu  -38(%rsi), %xmm1
> > > -       movdqu  -38(%rdi), %xmm2
> > > -       mov     $-38, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(22bytes):
> > > -       movdqu  -22(%rsi), %xmm1
> > > -       movdqu  -22(%rdi), %xmm2
> > > -       mov     $-22, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(71bytes):
> > > -       movdqu  -71(%rsi), %xmm1
> > > -       movdqu  -71(%rdi), %xmm2
> > > -       mov     $-71, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(55bytes):
> > > -       movdqu  -55(%rdi), %xmm2
> > > -       movdqu  -55(%rsi), %xmm1
> > > -       mov     $-55, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(39bytes):
> > > -       movdqu  -39(%rdi), %xmm2
> > > -       movdqu  -39(%rsi), %xmm1
> > > -       mov     $-39, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(23bytes):
> > > -       movdqu  -23(%rdi), %xmm2
> > > -       movdqu  -23(%rsi), %xmm1
> > > -       mov     $-23, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -# endif
> > > -
> > > -       .p2align 4
> > > -L(72bytes):
> > > -       movdqu  -72(%rsi), %xmm1
> > > -       movdqu  -72(%rdi), %xmm2
> > > -       mov     $-72, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(56bytes):
> > > -       movdqu  -56(%rdi), %xmm2
> > > -       movdqu  -56(%rsi), %xmm1
> > > -       mov     $-56, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(40bytes):
> > > -       movdqu  -40(%rdi), %xmm2
> > > -       movdqu  -40(%rsi), %xmm1
> > > -       mov     $-40, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(24bytes):
> > > -       movdqu  -24(%rdi), %xmm2
> > > -       movdqu  -24(%rsi), %xmm1
> > > -       mov     $-24, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -
> > > -       mov     -8(%rsi), %rcx
> > > -       mov     -8(%rdi), %rax
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -# ifndef USE_AS_WMEMCMP
> > > -/* unreal cases for wmemcmp */
> > > -       .p2align 4
> > > -L(73bytes):
> > > -       movdqu  -73(%rsi), %xmm1
> > > -       movdqu  -73(%rdi), %xmm2
> > > -       mov     $-73, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(57bytes):
> > > -       movdqu  -57(%rdi), %xmm2
> > > -       movdqu  -57(%rsi), %xmm1
> > > -       mov     $-57, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(41bytes):
> > > -       movdqu  -41(%rdi), %xmm2
> > > -       movdqu  -41(%rsi), %xmm1
> > > -       mov     $-41, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(25bytes):
> > > -       movdqu  -25(%rdi), %xmm2
> > > -       movdqu  -25(%rsi), %xmm1
> > > -       mov     $-25, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -9(%rdi), %rax
> > > -       mov     -9(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       movzbl  -1(%rdi), %eax
> > > -       movzbl  -1(%rsi), %ecx
> > > -       sub     %ecx, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(74bytes):
> > > -       movdqu  -74(%rsi), %xmm1
> > > -       movdqu  -74(%rdi), %xmm2
> > > -       mov     $-74, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(58bytes):
> > > -       movdqu  -58(%rdi), %xmm2
> > > -       movdqu  -58(%rsi), %xmm1
> > > -       mov     $-58, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(42bytes):
> > > -       movdqu  -42(%rdi), %xmm2
> > > -       movdqu  -42(%rsi), %xmm1
> > > -       mov     $-42, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(26bytes):
> > > -       movdqu  -26(%rdi), %xmm2
> > > -       movdqu  -26(%rsi), %xmm1
> > > -       mov     $-26, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -10(%rdi), %rax
> > > -       mov     -10(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       movzwl  -2(%rdi), %eax
> > > -       movzwl  -2(%rsi), %ecx
> > > -       jmp     L(diffin2bytes)
> > > -
> > > -       .p2align 4
> > > -L(75bytes):
> > > -       movdqu  -75(%rsi), %xmm1
> > > -       movdqu  -75(%rdi), %xmm2
> > > -       mov     $-75, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(59bytes):
> > > -       movdqu  -59(%rdi), %xmm2
> > > -       movdqu  -59(%rsi), %xmm1
> > > -       mov     $-59, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(43bytes):
> > > -       movdqu  -43(%rdi), %xmm2
> > > -       movdqu  -43(%rsi), %xmm1
> > > -       mov     $-43, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(27bytes):
> > > -       movdqu  -27(%rdi), %xmm2
> > > -       movdqu  -27(%rsi), %xmm1
> > > -       mov     $-27, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -11(%rdi), %rax
> > > -       mov     -11(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     -4(%rdi), %eax
> > > -       mov     -4(%rsi), %ecx
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin4bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -# endif
> > > -       .p2align 4
> > > -L(76bytes):
> > > -       movdqu  -76(%rsi), %xmm1
> > > -       movdqu  -76(%rdi), %xmm2
> > > -       mov     $-76, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(60bytes):
> > > -       movdqu  -60(%rdi), %xmm2
> > > -       movdqu  -60(%rsi), %xmm1
> > > -       mov     $-60, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(44bytes):
> > > -       movdqu  -44(%rdi), %xmm2
> > > -       movdqu  -44(%rsi), %xmm1
> > > -       mov     $-44, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(28bytes):
> > > -       movdqu  -28(%rdi), %xmm2
> > > -       movdqu  -28(%rsi), %xmm1
> > > -       mov     $-28, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -12(%rdi), %rax
> > > -       mov     -12(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     -4(%rsi), %ecx
> > > -# ifndef USE_AS_WMEMCMP
> > > -       mov     -4(%rdi), %eax
> > > -       cmp     %eax, %ecx
> > > -# else
> > > -       cmp     -4(%rdi), %ecx
> > > -# endif
> > > -       jne     L(diffin4bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -# ifndef USE_AS_WMEMCMP
> > > -/* unreal cases for wmemcmp */
> > > -       .p2align 4
> > > -L(77bytes):
> > > -       movdqu  -77(%rsi), %xmm1
> > > -       movdqu  -77(%rdi), %xmm2
> > > -       mov     $-77, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(61bytes):
> > > -       movdqu  -61(%rdi), %xmm2
> > > -       movdqu  -61(%rsi), %xmm1
> > > -       mov     $-61, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(45bytes):
> > > -       movdqu  -45(%rdi), %xmm2
> > > -       movdqu  -45(%rsi), %xmm1
> > > -       mov     $-45, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(29bytes):
> > > -       movdqu  -29(%rdi), %xmm2
> > > -       movdqu  -29(%rsi), %xmm1
> > > -       mov     $-29, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -
> > > -       mov     -13(%rdi), %rax
> > > -       mov     -13(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(78bytes):
> > > -       movdqu  -78(%rsi), %xmm1
> > > -       movdqu  -78(%rdi), %xmm2
> > > -       mov     $-78, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(62bytes):
> > > -       movdqu  -62(%rdi), %xmm2
> > > -       movdqu  -62(%rsi), %xmm1
> > > -       mov     $-62, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(46bytes):
> > > -       movdqu  -46(%rdi), %xmm2
> > > -       movdqu  -46(%rsi), %xmm1
> > > -       mov     $-46, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(30bytes):
> > > -       movdqu  -30(%rdi), %xmm2
> > > -       movdqu  -30(%rsi), %xmm1
> > > -       mov     $-30, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -14(%rdi), %rax
> > > -       mov     -14(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(79bytes):
> > > -       movdqu  -79(%rsi), %xmm1
> > > -       movdqu  -79(%rdi), %xmm2
> > > -       mov     $-79, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(63bytes):
> > > -       movdqu  -63(%rdi), %xmm2
> > > -       movdqu  -63(%rsi), %xmm1
> > > -       mov     $-63, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(47bytes):
> > > -       movdqu  -47(%rdi), %xmm2
> > > -       movdqu  -47(%rsi), %xmm1
> > > -       mov     $-47, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(31bytes):
> > > -       movdqu  -31(%rdi), %xmm2
> > > -       movdqu  -31(%rsi), %xmm1
> > > -       mov     $-31, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -       mov     -15(%rdi), %rax
> > > -       mov     -15(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -# endif
> > > -       .p2align 4
> > > -L(64bytes):
> > > -       movdqu  -64(%rdi), %xmm2
> > > -       movdqu  -64(%rsi), %xmm1
> > > -       mov     $-64, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(48bytes):
> > > -       movdqu  -48(%rdi), %xmm2
> > > -       movdqu  -48(%rsi), %xmm1
> > > -       mov     $-48, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -L(32bytes):
> > > -       movdqu  -32(%rdi), %xmm2
> > > -       movdqu  -32(%rsi), %xmm1
> > > -       mov     $-32, %dl
> > > -       pxor    %xmm1, %xmm2
> > > -       ptest   %xmm2, %xmm0
> > > -       jnc     L(less16bytes)
> > > -
> > > -       mov     -16(%rdi), %rax
> > > -       mov     -16(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -
> > > -       mov     -8(%rdi), %rax
> > > -       mov     -8(%rsi), %rcx
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -
> > > -/*
> > > - * Aligned 8 bytes to avoid 2 branch "taken" in one 16 alinged code block.
> > > - */
> > > -       .p2align 3
> > > -L(less16bytes):
> > > -       movsbq  %dl, %rdx
> > > -       mov     (%rsi, %rdx), %rcx
> > > -       mov     (%rdi, %rdx), %rax
> > > -       cmp     %rax, %rcx
> > > -       jne     L(diffin8bytes)
> > > -       mov     8(%rsi, %rdx), %rcx
> > > -       mov     8(%rdi, %rdx), %rax
> > > -L(diffin8bytes):
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin4bytes)
> > > -       shr     $32, %rcx
> > > -       shr     $32, %rax
> > > -
> > > +       pmovmskb %xmm0, %ecx
> > > +       incw    %cx
> > > +       jnz     L(loop_end_ret)
> > > +
> > > +       pmovmskb %xmm1, %ecx
> > > +       notw    %cx
> > > +       sall    $16, %ecx
> > > +       jnz     L(loop_end_ret)
> > > +
> > > +       pmovmskb %xmm2, %ecx
> > > +       notw    %cx
> > > +       shlq    $32, %rcx
> > > +       jnz     L(loop_end_ret)
> > > +
> > > +       addq    $48, %rdi
> > > +       addq    $48, %rsi
> > > +       movq    %rax, %rcx
> > > +
> > > +       .p2align 4,, 6
> > > +L(loop_end_ret):
> > > +       bsfq    %rcx, %rcx
> > >  # ifdef USE_AS_WMEMCMP
> > > -/* for wmemcmp */
> > > -       cmp     %eax, %ecx
> > > -       jne     L(diffin4bytes)
> > > -       xor     %eax, %eax
> > > -       ret
> > > -# endif
> > > -
> > > -L(diffin4bytes):
> > > -# ifndef USE_AS_WMEMCMP
> > > -       cmp     %cx, %ax
> > > -       jne     L(diffin2bytes)
> > > -       shr     $16, %ecx
> > > -       shr     $16, %eax
> > > -L(diffin2bytes):
> > > -       cmp     %cl, %al
> > > -       jne     L(end)
> > > -       and     $0xffff, %eax
> > > -       and     $0xffff, %ecx
> > > -       sub     %ecx, %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(end):
> > > -       and     $0xff, %eax
> > > -       and     $0xff, %ecx
> > > -       sub     %ecx, %eax
> > > -       ret
> > > +       movl    (%rdi, %rcx), %eax
> > > +       xorl    %edx, %edx
> > > +       cmpl    (%rsi, %rcx), %eax
> > > +       setg    %dl
> > > +       leal    -1(%rdx, %rdx), %eax
> > >  # else
> > > -
> > > -/* for wmemcmp */
> > > -       mov     $1, %eax
> > > -       jl      L(nequal_bigger)
> > > -       neg     %eax
> > > -       ret
> > > -
> > > -       .p2align 4
> > > -L(nequal_bigger):
> > > -       ret
> > > -
> > > -L(unreal_case):
> > > -       xor     %eax, %eax
> > > -       ret
> > > +       movzbl  (%rdi, %rcx), %eax
> > > +       movzbl  (%rsi, %rcx), %ecx
> > > +       subl    %ecx, %eax
> > >  # endif
> > > -
> > > +       ret
> > >  END (MEMCMP)
> > > -
> > > -       .section .rodata.sse4.1,"a",@progbits
> > > -       .p2align 3
> > > -# ifndef USE_AS_WMEMCMP
> > > -L(table_64bytes):
> > > -       .int    JMPTBL (L(0bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(1bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(2bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(3bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(4bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(5bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(6bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(7bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(8bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(9bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(10bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(11bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(12bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(13bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(14bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(15bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(16bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(17bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(18bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(19bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(20bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(21bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(22bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(23bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(24bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(25bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(26bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(27bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(28bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(29bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(30bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(31bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(32bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(33bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(34bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(35bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(36bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(37bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(38bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(39bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(40bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(41bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(42bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(43bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(44bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(45bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(46bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(47bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(48bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(49bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(50bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(51bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(52bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(53bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(54bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(55bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(56bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(57bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(58bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(59bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(60bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(61bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(62bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(63bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(64bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(65bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(66bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(67bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(68bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(69bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(70bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(71bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(72bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(73bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(74bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(75bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(76bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(77bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(78bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(79bytes), L(table_64bytes))
> > > -# else
> > > -L(table_64bytes):
> > > -       .int    JMPTBL (L(0bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(4bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(8bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(12bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(16bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(20bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(24bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(28bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(32bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(36bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(40bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(44bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(48bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(52bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(56bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(60bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(64bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(68bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(72bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(76bytes), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -       .int    JMPTBL (L(unreal_case), L(table_64bytes))
> > > -# endif
> > >  #endif
> > > --
> > > 2.25.1
> > >
> >
> > LGTM.
> >
> > Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
> Thanks. Pushed.
> >
> > Thanks.
> >
> > --
> > H.J.

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

--Sunil