[PATCH v1] x86: Remove SSE3 instruction for broadcast in memset.S (SSE2 Only)

[H.J. Lu]

On Sun, Feb 6, 2022 at 10:39 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
>
> commit b62ace2740a106222e124cc86956448fa07abf4d
> Author: Noah Goldstein <goldstein.w.n@gmail.com>
> Date:   Sun Feb 6 00:54:18 2022 -0600
>
>     x86: Improve vec generation in memset-vec-unaligned-erms.S
>
> Revert usage of 'pshufb' in broadcast logic as it is an SSE3
> instruction and memset.S is restricted to only SSE2 instructions.
> ---
>  sysdeps/x86_64/memset.S | 19 ++++++++++---------
>  1 file changed, 10 insertions(+), 9 deletions(-)
>
> diff --git a/sysdeps/x86_64/memset.S b/sysdeps/x86_64/memset.S
> index ccf036be53..148553cf3d 100644
> --- a/sysdeps/x86_64/memset.S
> +++ b/sysdeps/x86_64/memset.S
> @@ -28,22 +28,23 @@
>  #define VMOVU     movups
>  #define VMOVA     movaps
>
> -# define MEMSET_SET_VEC0_AND_SET_RETURN(d, r) \
> +#define MEMSET_SET_VEC0_AND_SET_RETURN(d, r) \
>    movd d, %xmm0; \
> -  pxor %xmm1, %xmm1; \
> -  pshufb %xmm1, %xmm0; \
> -  movq r, %rax
> +  movq r, %rax; \
> +  punpcklbw %xmm0, %xmm0; \
> +  punpcklwd %xmm0, %xmm0; \
> +  pshufd $0, %xmm0, %xmm0
>
> -# define WMEMSET_SET_VEC0_AND_SET_RETURN(d, r) \
> +#define WMEMSET_SET_VEC0_AND_SET_RETURN(d, r) \
>    movd d, %xmm0; \
>    pshufd $0, %xmm0, %xmm0; \
>    movq r, %rax
>
> -# define MEMSET_VDUP_TO_VEC0_HIGH()
> -# define MEMSET_VDUP_TO_VEC0_LOW()
> +#define MEMSET_VDUP_TO_VEC0_HIGH()
> +#define MEMSET_VDUP_TO_VEC0_LOW()

What are these changes for?

> -# define WMEMSET_VDUP_TO_VEC0_HIGH()
> -# define WMEMSET_VDUP_TO_VEC0_LOW()
> +#define WMEMSET_VDUP_TO_VEC0_HIGH()
> +#define WMEMSET_VDUP_TO_VEC0_LOW()

What are these changes for?

>  #define SECTION(p)             p
>
> --
> 2.25.1
>


-- 
H.J.