1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/os/ossrv/genericopenlibs/liboil/src/sad8x8_sse.c	Fri Jun 15 03:10:57 2012 +0200
     1.3 @@ -0,0 +1,68 @@
     1.4 +/*
     1.5 + * Copyright (c) 2005
     1.6 + *	Eric Anholt.  All rights reserved.
     1.7 + *
     1.8 + * Redistribution and use in source and binary forms, with or without
     1.9 + * modification, are permitted provided that the following conditions
    1.10 + * are met:
    1.11 + * 1. Redistributions of source code must retain the above copyright
    1.12 + *    notice, this list of conditions and the following disclaimer.
    1.13 + * 2. Redistributions in binary form must reproduce the above copyright
    1.14 + *    notice, this list of conditions and the following disclaimer in the
    1.15 + *    documentation and/or other materials provided with the distribution.
    1.16 + *
    1.17 + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
    1.18 + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
    1.19 + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
    1.20 + * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE
    1.21 + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
    1.22 + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
    1.23 + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
    1.24 + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
    1.25 + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
    1.26 + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
    1.27 + * SUCH DAMAGE.
    1.28 + */
    1.29 +
    1.30 +#ifdef HAVE_CONFIG_H
    1.31 +#include "config.h"
    1.32 +#endif
    1.33 +#include <liboil/liboilclasses.h>
    1.34 +#include <liboil/liboilfunction.h>
    1.35 +#include <emmintrin.h>
    1.36 +
    1.37 +#define SSE_FUNCTION __attribute__((force_align_arg_pointer))
    1.38 +
    1.39 +#ifdef ENABLE_BROKEN_IMPLS
    1.40 +union m128_int {
    1.41 +  __m128i m128;
    1.42 +  uint32_t i[4];
    1.43 +  uint16_t s[8];
    1.44 +};
    1.45 +
    1.46 +SSE_FUNCTION static void
    1.47 +sad8x8_u8_sse (uint32_t *dest, uint8_t *src1, int sstr1, uint8_t *src2,
    1.48 +    int sstr2)
    1.49 +{
    1.50 +  int i;
    1.51 +  __m128i sum = _mm_setzero_si128();
    1.52 +  union m128_int sumi;
    1.53 +
    1.54 +  for (i = 0; i < 4; i++) {
    1.55 +    __m128i xmm0, xmm1, xmm2, xmm3;
    1.56 +    xmm0 = _mm_loadl_epi64((__m128i *)src1);
    1.57 +    xmm1 = _mm_loadl_epi64((__m128i *)(src1 + sstr1));
    1.58 +    xmm2 = _mm_loadl_epi64((__m128i *)src2);
    1.59 +    xmm3 = _mm_loadl_epi64((__m128i *)(src2 + sstr2));
    1.60 +    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);
    1.61 +    xmm2 = _mm_unpacklo_epi8(xmm2, xmm3);
    1.62 +    sum = _mm_add_epi64(sum, _mm_sad_epu8(xmm0, xmm2));
    1.63 +    src1 += 2 * sstr1;
    1.64 +    src2 += 2 * sstr2;
    1.65 +  }
    1.66 +  sumi.m128 = sum;
    1.67 +  *dest = sumi.i[0] + sumi.i[2];
    1.68 +}
    1.69 +OIL_DEFINE_IMPL_FULL_WRAPPER(sad8x8_u8_sse, sad8x8_u8, OIL_IMPL_FLAG_SSE2);
    1.70 +#endif
    1.71 +