|
sl@0
|
1 |
/*
|
|
sl@0
|
2 |
* Copyright (c) 2005
|
|
sl@0
|
3 |
* Eric Anholt. All rights reserved.
|
|
sl@0
|
4 |
*
|
|
sl@0
|
5 |
* Redistribution and use in source and binary forms, with or without
|
|
sl@0
|
6 |
* modification, are permitted provided that the following conditions
|
|
sl@0
|
7 |
* are met:
|
|
sl@0
|
8 |
* 1. Redistributions of source code must retain the above copyright
|
|
sl@0
|
9 |
* notice, this list of conditions and the following disclaimer.
|
|
sl@0
|
10 |
* 2. Redistributions in binary form must reproduce the above copyright
|
|
sl@0
|
11 |
* notice, this list of conditions and the following disclaimer in the
|
|
sl@0
|
12 |
* documentation and/or other materials provided with the distribution.
|
|
sl@0
|
13 |
*
|
|
sl@0
|
14 |
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
|
|
sl@0
|
15 |
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
sl@0
|
16 |
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
sl@0
|
17 |
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE
|
|
sl@0
|
18 |
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
sl@0
|
19 |
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
sl@0
|
20 |
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
sl@0
|
21 |
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
sl@0
|
22 |
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
sl@0
|
23 |
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
sl@0
|
24 |
* SUCH DAMAGE.
|
|
sl@0
|
25 |
*/
|
|
sl@0
|
26 |
|
|
sl@0
|
27 |
#ifdef HAVE_CONFIG_H
|
|
sl@0
|
28 |
#include "config.h"
|
|
sl@0
|
29 |
#endif
|
|
sl@0
|
30 |
#include <liboil/liboilclasses.h>
|
|
sl@0
|
31 |
#include <liboil/liboilfunction.h>
|
|
sl@0
|
32 |
#include <emmintrin.h>
|
|
sl@0
|
33 |
|
|
sl@0
|
34 |
#define SSE_FUNCTION __attribute__((force_align_arg_pointer))
|
|
sl@0
|
35 |
|
|
sl@0
|
36 |
#ifdef ENABLE_BROKEN_IMPLS
|
|
sl@0
|
37 |
union m128_int {
|
|
sl@0
|
38 |
__m128i m128;
|
|
sl@0
|
39 |
uint32_t i[4];
|
|
sl@0
|
40 |
uint16_t s[8];
|
|
sl@0
|
41 |
};
|
|
sl@0
|
42 |
|
|
sl@0
|
43 |
SSE_FUNCTION static void
|
|
sl@0
|
44 |
sad8x8_u8_sse (uint32_t *dest, uint8_t *src1, int sstr1, uint8_t *src2,
|
|
sl@0
|
45 |
int sstr2)
|
|
sl@0
|
46 |
{
|
|
sl@0
|
47 |
int i;
|
|
sl@0
|
48 |
__m128i sum = _mm_setzero_si128();
|
|
sl@0
|
49 |
union m128_int sumi;
|
|
sl@0
|
50 |
|
|
sl@0
|
51 |
for (i = 0; i < 4; i++) {
|
|
sl@0
|
52 |
__m128i xmm0, xmm1, xmm2, xmm3;
|
|
sl@0
|
53 |
xmm0 = _mm_loadl_epi64((__m128i *)src1);
|
|
sl@0
|
54 |
xmm1 = _mm_loadl_epi64((__m128i *)(src1 + sstr1));
|
|
sl@0
|
55 |
xmm2 = _mm_loadl_epi64((__m128i *)src2);
|
|
sl@0
|
56 |
xmm3 = _mm_loadl_epi64((__m128i *)(src2 + sstr2));
|
|
sl@0
|
57 |
xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);
|
|
sl@0
|
58 |
xmm2 = _mm_unpacklo_epi8(xmm2, xmm3);
|
|
sl@0
|
59 |
sum = _mm_add_epi64(sum, _mm_sad_epu8(xmm0, xmm2));
|
|
sl@0
|
60 |
src1 += 2 * sstr1;
|
|
sl@0
|
61 |
src2 += 2 * sstr2;
|
|
sl@0
|
62 |
}
|
|
sl@0
|
63 |
sumi.m128 = sum;
|
|
sl@0
|
64 |
*dest = sumi.i[0] + sumi.i[2];
|
|
sl@0
|
65 |
}
|
|
sl@0
|
66 |
OIL_DEFINE_IMPL_FULL_WRAPPER(sad8x8_u8_sse, sad8x8_u8, OIL_IMPL_FLAG_SSE2);
|
|
sl@0
|
67 |
#endif
|
|
sl@0
|
68 |
|