From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <libc-ports-return-3951-listarch-libc-ports=sources.redhat.com@sourceware.org>
Received: (qmail 5652 invoked by alias); 25 Mar 2013 15:33:06 -0000
Mailing-List: contact libc-ports-help@sourceware.org; run by ezmlm
Precedence: bulk
List-Id: <libc-ports.sourceware.org>
List-Subscribe: <mailto:libc-ports-subscribe@sourceware.org>
List-Post: <mailto:libc-ports@sourceware.org>
List-Help: <mailto:libc-ports-help@sourceware.org>, <http://sourceware.org/lists.html#faqs>
Sender: libc-ports-owner@sourceware.org
Received: (qmail 5228 invoked by uid 89); 25 Mar 2013 15:32:58 -0000
Received: from popelka.ms.mff.cuni.cz (HELO popelka.ms.mff.cuni.cz) (195.113.20.131)    by sourceware.org (qpsmtpd/0.84/v0.84-167-ge50287c) with ESMTP; Mon, 25 Mar 2013 15:32:58 +0000
Received: from domone.kolej.mff.cuni.cz (popelka.ms.mff.cuni.cz [195.113.20.131])	by popelka.ms.mff.cuni.cz (Postfix) with ESMTPS id CDEBC62A3A;	Mon, 25 Mar 2013 16:32:50 +0100 (CET)
Received: by domone.kolej.mff.cuni.cz (Postfix, from userid 1000)	id 586925FC15; Mon, 25 Mar 2013 16:28:23 +0100 (CET)
Date: Mon, 25 Mar 2013 15:33:00 -0000
From: =?utf-8?B?T25kxZllaiBCw61sa2E=?= <neleai@seznam.cz>
To: marcus.shawcroft@linaro.org
Cc: libc-ports@sourceware.org
Subject: memset on aarch64
Message-ID: <20130325152822.GA23513@domone.kolej.mff.cuni.cz>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.5.20 (2009-06-14)
X-SW-Source: 2013-03/txt/msg00158.txt.bz2

Hi Marcus,

could you try how following memset implementation works on 
aarch64.

It could be faster for small n, I found that on x64 overlapping stores
are best way how handle end conditions. I do not know how arm could handle that.

I do not know how good loop will gcc generate, if this is faster header
you can replace loop in assembly.

#include <stdint.h>
#include <stdlib.h>

/* Align VALUE down by ALIGN bytes.  */
#define ALIGN_DOWN(value, align) \
       ALIGN_DOWN_M1(value, align - 1)
/* Align VALUE down by ALIGN_M1 + 1 bytes.
   Useful if you have precomputed ALIGN - 1.  */
#define ALIGN_DOWN_M1(value, align_m1) \
       (void *)((uintptr_t)(value) \
                & ~(uintptr_t)(align_m1))

/* Align VALUE up by ALIGN bytes.  */
#define ALIGN_UP(value, align) \
       ALIGN_UP_M1(value, align - 1)
/* Align VALUE up by ALIGN_M1 + 1 bytes.
   Useful if you have precomputed ALIGN - 1.  */
#define ALIGN_UP_M1(value, align_m1) \
       (void *)(((uintptr_t)(value) + (uintptr_t)(align_m1)) \
                & ~(uintptr_t)(align_m1))

#define STOREU(x,y) STORE(x,y)
#define STORE(x,y) ((uint64_t*)(x))[0]=y; ((uint64_t*)(x))[1]=y;

static char *memset_small (char *dest, uint64_t c, size_t no, char *ret);
void *memset_new(char *dest, int _c, size_t n)
{
  int i;
  unsigned char c = _c;
  uint64_t vc = 0x0101010101010101ULL*c;
  if (n < 16)
    {
      return memset_small(dest, vc, n, dest);
    }
  else
    {
      STOREU(dest, vc);
      STOREU(dest + n - 16, vc);
      char *to   = ALIGN_DOWN(dest + n, 16);
      dest = ALIGN_DOWN(dest + 16, 16);
      while (dest != to)
        {
          STORE(dest,vc);
          dest += 16;
        }
    }
  return dest;
}

static char *memset_small (char *dest, uint64_t c, size_t no, char *ret)
{
  if (no & (8))
    {
      ((uint64_t *) dest)[0] = c;
      ((uint64_t *)(dest + no - 8))[0] = c;
      return ret;
    }
  if (no & 4)
    {
      ((uint32_t *) dest)[0] = c;
      ((uint32_t *)(dest + no - 4))[0] = c;
      return ret;
    }
  if (no & 1)
    {
      dest[0] = c;
    }
  if (no & 2)
    {
      ((uint16_t *)(dest + no - 2))[0] = c;
    }
  return ret;
}