[PATCH v2] x86-64: Use 32-bit zero idiom for shorter encoding

[Sunil Pandey]

On Sat, Feb 7, 2026 at 8:07 AM George Hu <integral@archlinux.org> wrote:

> Replace the 64-bit zero idiom with the 32-bit form. In 64-bit mode,
> zeroing the lower 32 bits of a GPR implicitly clears the entire
> register. The 32-bit encoding is one byte shorter while preserving
> identical semantics.
> ---
>  sysdeps/mach/hurd/x86_64/static-start.S          | 2 +-
>  sysdeps/mach/hurd/x86_64/tst-sig-redzone.c       | 2 +-
>  sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S | 2 +-
>  sysdeps/x86_64/multiarch/strchr-sse2-no-bsf.S    | 2 +-
>  sysdeps/x86_64/multiarch/strcpy-sse2-unaligned.S | 4 ++--
>  sysdeps/x86_64/multiarch/strlen-sse2.S           | 2 +-
>  sysdeps/x86_64/multiarch/wcschr-sse2.S           | 2 +-
>  sysdeps/x86_64/multiarch/wcscmp-sse2.S           | 2 +-
>  8 files changed, 9 insertions(+), 9 deletions(-)
>
> diff --git a/sysdeps/mach/hurd/x86_64/static-start.S
> b/sysdeps/mach/hurd/x86_64/static-start.S
> index ec9c50239e..d670d45210 100644
> --- a/sysdeps/mach/hurd/x86_64/static-start.S
> +++ b/sysdeps/mach/hurd/x86_64/static-start.S
> @@ -27,7 +27,7 @@ _start:
>
>         movq %rsp, %rdi
>         call _hurd_stack_setup
> -       xorq %rdx, %rdx
> +       xorl %edx, %edx
>         jmp _start1
>
>  #define _start _start1
> diff --git a/sysdeps/mach/hurd/x86_64/tst-sig-redzone.c
> b/sysdeps/mach/hurd/x86_64/tst-sig-redzone.c
> index 32c7ab90f4..24b510c0aa 100644
> --- a/sysdeps/mach/hurd/x86_64/tst-sig-redzone.c
> +++ b/sysdeps/mach/hurd/x86_64/tst-sig-redzone.c
> @@ -103,7 +103,7 @@ asm (
>  "repe  scasq\n"
>  "      jne     fail\n"
>
> -"      xor     %rax,%rax\n"
> +"      xor     %eax,%eax\n"
>  "      ret\n"
>  "fail:\n"
>  "      movq    $1,%rax\n"
> diff --git a/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
> b/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
> index 1c04d611be..3c3f039047 100644
> --- a/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
> +++ b/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
> @@ -45,7 +45,7 @@ ENTRY (STRCAT)
>  /* Inline corresponding strlen file, temporary until new strcpy
>     implementation gets merged.  */
>
> -       xor     %rax, %rax
> +       xor     %eax, %eax
>         mov     %edi, %ecx
>         and     $0x3f, %ecx
>         pxor    %xmm0, %xmm0
> diff --git a/sysdeps/x86_64/multiarch/strchr-sse2-no-bsf.S
> b/sysdeps/x86_64/multiarch/strchr-sse2-no-bsf.S
> index 9521cd4dc4..7d2677b4fd 100644
> --- a/sysdeps/x86_64/multiarch/strchr-sse2-no-bsf.S
> +++ b/sysdeps/x86_64/multiarch/strchr-sse2-no-bsf.S
> @@ -69,7 +69,7 @@ L(loop):
>  /* Return NULL.  */
>         .p2align 4
>  L(return_null):
> -       xor     %rax, %rax
> +       xor     %eax, %eax
>         ret
>
>  L(matches):
> diff --git a/sysdeps/x86_64/multiarch/strcpy-sse2-unaligned.S
> b/sysdeps/x86_64/multiarch/strcpy-sse2-unaligned.S
> index 3dfee3039a..59ff7bc85e 100644
> --- a/sysdeps/x86_64/multiarch/strcpy-sse2-unaligned.S
> +++ b/sysdeps/x86_64/multiarch/strcpy-sse2-unaligned.S
> @@ -1657,7 +1657,7 @@ L(CopyFrom1To16BytesXmmExit):
>         .p2align 4
>  L(StrncpyFillTailWithZero):
>         pxor    %xmm0, %xmm0
> -       xor     %rdx, %rdx
> +       xor     %edx, %edx
>         sub     $16, %r8
>         jbe     L(StrncpyFillExit)
>
> @@ -1736,7 +1736,7 @@ L(Unaligned64LeaveCase3):
>
>         .p2align 4
>  L(Unaligned64LeaveCase2):
> -       xor     %rcx, %rcx
> +       xor     %ecx, %ecx
>         pcmpeqb %xmm4, %xmm0
>         pmovmskb %xmm0, %rdx
>         add     $48, %r8
> diff --git a/sysdeps/x86_64/multiarch/strlen-sse2.S
> b/sysdeps/x86_64/multiarch/strlen-sse2.S
> index ca6e0812be..6085ad1236 100644
> --- a/sysdeps/x86_64/multiarch/strlen-sse2.S
> +++ b/sysdeps/x86_64/multiarch/strlen-sse2.S
> @@ -77,7 +77,7 @@ ENTRY(STRLEN)
>  /* Do not read anything when n==0.  */
>         test    %RSI_LP, %RSI_LP
>         jne     L(n_nonzero)
> -       xor     %rax, %rax
> +       xor     %eax, %eax
>         ret
>  L(n_nonzero):
>  #  ifdef AS_WCSLEN
> diff --git a/sysdeps/x86_64/multiarch/wcschr-sse2.S
> b/sysdeps/x86_64/multiarch/wcschr-sse2.S
> index ec11a16a73..f0069f026a 100644
> --- a/sysdeps/x86_64/multiarch/wcschr-sse2.S
> +++ b/sysdeps/x86_64/multiarch/wcschr-sse2.S
> @@ -155,7 +155,7 @@ L(match):
>
>         .p2align 4
>  L(return_null):
> -       xor     %rax, %rax
> +       xor     %eax, %eax
>         ret
>
>  END (WCSCHR)
> diff --git a/sysdeps/x86_64/multiarch/wcscmp-sse2.S
> b/sysdeps/x86_64/multiarch/wcscmp-sse2.S
> index 5ab124e0a7..a54615a9ae 100644
> --- a/sysdeps/x86_64/multiarch/wcscmp-sse2.S
> +++ b/sysdeps/x86_64/multiarch/wcscmp-sse2.S
> @@ -950,7 +950,7 @@ L(nequal_bigger):
>
>         .p2align 4
>  L(equal):
> -       xor     %rax, %rax
> +       xor     %eax, %eax
>         ret
>
>  END (STRCMP)
> --
> 2.52.0
>
>
Can you please split this commit into multiple commits.
Each one of them needs benchmark data to properly evaluate this change.

--Sunil
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/libc-alpha/attachments/20260209/6aa5b40a/attachment-0001.htm>