RE: [RFC] Clean up SSE variable shifts

["Lu, Hongjiu" <hongjiu dot lu at intel dot com>]

Hi,

Here are TSC deltas between different implementations. It is
hard to tell which one is faster.

H.J.


> -----Original Message-----
> From: Ulrich Drepper [mailto:drepper@redhat.com]
> Sent: Monday, August 23, 2010 10:08 AM
> To: Richard Henderson
> Cc: libc-alpha@sourceware.org; Lu, Hongjiu
> Subject: Re: [RFC] Clean up SSE variable shifts
> 
> ----- "Richard Henderson" <rth@twiddle.net> wrote:
> > And for good measure, here's a third patch which implements a
> > variable shift via pshufb.
> 
> As a forth variant try this patch with the following change.
> Considering the reduced memory footprint (one cache line instead of
> four) and the improved capabilities for unaligned SSE memory
> operations in modern CPUs it might be even faster.
> 
> 
> diff -u b/sysdeps/x86_64/multiarch/varshift.S
> b/sysdeps/x86_64/multiarch/varshift.S
> --- b/sysdeps/x86_64/multiarch/varshift.S
> +++ b/sysdeps/x86_64/multiarch/varshift.S
> @@ -26,20 +26,5 @@
> -	.size	___m128i_shift_right, 256
> +	.size	___m128i_shift_right, 31
> 
> -	.balign 16
>  ___m128i_shift_right:
>  	.byte	  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13,
> 14, 15
> -	.byte	  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14,
> 15, -1
> -	.byte	  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, -
> 1, -1
> -	.byte	  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, -1, -
> 1, -1
> -	.byte	  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, -1, -1, -
> 1, -1
> -	.byte	  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, -1, -1, -1, -
> 1, -1
> -	.byte	  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -
> 1, -1
> -	.byte	  7,  8,  9, 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -
> 1, -1
> -	.byte	  8,  9, 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -
> 1, -1
> -	.byte	  9, 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -
> 1, -1
> -	.byte	 10, 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -
> 1, -1
> -	.byte	 11, 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, -
> 1, -1
> -	.byte	 12, 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -
> 1, -1
> -	.byte	 13, 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -
> 1, -1
> -	.byte	 14, 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -
> 1, -1
> -	.byte	 15, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -
> 1, -1
> +	.byte	 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -
> 1
> diff -u b/sysdeps/x86_64/multiarch/varshift.h
> b/sysdeps/x86_64/multiarch/varshift.h
> --- b/sysdeps/x86_64/multiarch/varshift.h
> +++ b/sysdeps/x86_64/multiarch/varshift.h
> @@ -20,8 +20,8 @@
> 
> -extern __m128i ___m128i_shift_right[16]
> __attribute__((visibility("hidden")));
> +extern char ___m128i_shift_right[31]
> __attribute__((visibility("hidden")));
> 
>  static __inline__ __m128i
>  __m128i_shift_right (__m128i value, unsigned long offset)
>  {
> -  return _mm_shuffle_epi8 (value, ___m128i_shift_right[offset]);
> +  return _mm_shuffle_epi8 (value, _mm_loadu_si128 ((__m128 *)
> (___m128i_shift_right + offset)));
>  }
> 
> 
> --
> â Ulrich Drepper â Red Hat, Inc. â 444 Castro St â Mountain View, CA â

Attachment: strcspn-2010-08-23.xlsx
Description: strcspn-2010-08-23.xlsx