1/*
2 * Copyright (c) 2007 Luca Barbato <lu_zero@gentoo.org>
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21/**
22 ** @file libavcodec/ppc/int_altivec.c
23 ** integer misc ops.
24 **/
25
26#include "libavcodec/dsputil.h"
27
28#include "gcc_fixes.h"
29
30#include "dsputil_altivec.h"
31
32#include "types_altivec.h"
33
34static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2,
35                                     int size) {
36    int i, size16;
37    vector signed char vpix1;
38    vector signed short vpix2, vdiff, vpix1l,vpix1h;
39    union { vector signed int vscore;
40            int32_t score[4];
41          } u;
42    u.vscore = vec_splat_s32(0);
43//
44//XXX lazy way, fix it later
45
46#define vec_unaligned_load(b) \
47    vec_perm(vec_ld(0,b),vec_ld(15,b),vec_lvsl(0, b));
48
49    size16 = size >> 4;
50    while(size16) {
51//        score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]);
52        //load pix1 and the first batch of pix2
53
54        vpix1 = vec_unaligned_load(pix1);
55        vpix2 = vec_unaligned_load(pix2);
56        pix2 += 8;
57        //unpack
58        vpix1h = vec_unpackh(vpix1);
59        vdiff  = vec_sub(vpix1h, vpix2);
60        vpix1l = vec_unpackl(vpix1);
61        // load another batch from pix2
62        vpix2 = vec_unaligned_load(pix2);
63        u.vscore = vec_msum(vdiff, vdiff, u.vscore);
64        vdiff  = vec_sub(vpix1l, vpix2);
65        u.vscore = vec_msum(vdiff, vdiff, u.vscore);
66        pix1 += 16;
67        pix2 += 8;
68        size16--;
69    }
70    u.vscore = vec_sums(u.vscore, vec_splat_s32(0));
71
72    size %= 16;
73    for (i = 0; i < size; i++) {
74        u.score[3] += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]);
75    }
76    return u.score[3];
77}
78
79static void add_int16_altivec(int16_t * v1, int16_t * v2, int order)
80{
81    int i;
82    register vec_s16 vec, *pv;
83
84    for(i = 0; i < order; i += 8){
85        pv = (vec_s16*)v2;
86        vec = vec_perm(pv[0], pv[1], vec_lvsl(0, v2));
87        vec_st(vec_add(vec_ld(0, v1), vec), 0, v1);
88        v1 += 8;
89        v2 += 8;
90    }
91}
92
93static void sub_int16_altivec(int16_t * v1, int16_t * v2, int order)
94{
95    int i;
96    register vec_s16 vec, *pv;
97
98    for(i = 0; i < order; i += 8){
99        pv = (vec_s16*)v2;
100        vec = vec_perm(pv[0], pv[1], vec_lvsl(0, v2));
101        vec_st(vec_sub(vec_ld(0, v1), vec), 0, v1);
102        v1 += 8;
103        v2 += 8;
104    }
105}
106
107static int32_t scalarproduct_int16_altivec(int16_t * v1, int16_t * v2, int order, const int shift)
108{
109    int i;
110    LOAD_ZERO;
111    register vec_s16 vec1, *pv;
112    register vec_s32 res = vec_splat_s32(0), t;
113    register vec_u32 shifts;
114    DECLARE_ALIGNED_16(int32_t, ires);
115
116    shifts = zero_u32v;
117    if(shift & 0x10) shifts = vec_add(shifts, vec_sl(vec_splat_u32(0x08), vec_splat_u32(0x1)));
118    if(shift & 0x08) shifts = vec_add(shifts, vec_splat_u32(0x08));
119    if(shift & 0x04) shifts = vec_add(shifts, vec_splat_u32(0x04));
120    if(shift & 0x02) shifts = vec_add(shifts, vec_splat_u32(0x02));
121    if(shift & 0x01) shifts = vec_add(shifts, vec_splat_u32(0x01));
122
123    for(i = 0; i < order; i += 8){
124        pv = (vec_s16*)v1;
125        vec1 = vec_perm(pv[0], pv[1], vec_lvsl(0, v1));
126        t = vec_msum(vec1, vec_ld(0, v2), zero_s32v);
127        t = vec_sr(t, shifts);
128        res = vec_sums(t, res);
129        v1 += 8;
130        v2 += 8;
131    }
132    res = vec_splat(res, 3);
133    vec_ste(res, 0, &ires);
134    return ires;
135}
136
137void int_init_altivec(DSPContext* c, AVCodecContext *avctx)
138{
139    c->ssd_int8_vs_int16 = ssd_int8_vs_int16_altivec;
140    c->add_int16 = add_int16_altivec;
141    c->sub_int16 = sub_int16_altivec;
142    c->scalarproduct_int16 = scalarproduct_int16_altivec;
143}
144