1/*
2 * High quality image resampling with polyphase filters
3 * Copyright (c) 2001 Fabrice Bellard
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22/**
23 * @file libavcodec/ppc/imgresample_altivec.c
24 * High quality image resampling with polyphase filters - AltiVec bits
25 */
26
27#include "util_altivec.h"
28#define FILTER_BITS   8
29
30typedef         union {
31    vector signed short v;
32    signed short s[8];
33} vec_ss;
34
35void v_resample16_altivec(uint8_t *dst, int dst_width, const uint8_t *src,
36                          int wrap, int16_t *filter)
37{
38    int sum, i;
39    const uint8_t *s;
40    vector unsigned char *tv, tmp, dstv, zero;
41    vec_ss srchv[4], srclv[4], fv[4];
42    vector signed short zeros, sumhv, sumlv;
43    s = src;
44
45    for(i=0;i<4;i++) {
46        /*
47           The vec_madds later on does an implicit >>15 on the result.
48           Since FILTER_BITS is 8, and we have 15 bits of magnitude in
49           a signed short, we have just enough bits to pre-shift our
50           filter constants <<7 to compensate for vec_madds.
51        */
52        fv[i].s[0] = filter[i] << (15-FILTER_BITS);
53        fv[i].v = vec_splat(fv[i].v, 0);
54    }
55
56    zero = vec_splat_u8(0);
57    zeros = vec_splat_s16(0);
58
59
60    /*
61       When we're resampling, we'd ideally like both our input buffers,
62       and output buffers to be 16-byte aligned, so we can do both aligned
63       reads and writes. Sadly we can't always have this at the moment, so
64       we opt for aligned writes, as unaligned writes have a huge overhead.
65       To do this, do enough scalar resamples to get dst 16-byte aligned.
66    */
67    i = (-(int)dst) & 0xf;
68    while(i>0) {
69        sum = s[0 * wrap] * filter[0] +
70        s[1 * wrap] * filter[1] +
71        s[2 * wrap] * filter[2] +
72        s[3 * wrap] * filter[3];
73        sum = sum >> FILTER_BITS;
74        if (sum<0) sum = 0; else if (sum>255) sum=255;
75        dst[0] = sum;
76        dst++;
77        s++;
78        dst_width--;
79        i--;
80    }
81
82    /* Do our altivec resampling on 16 pixels at once. */
83    while(dst_width>=16) {
84        /* Read 16 (potentially unaligned) bytes from each of
85           4 lines into 4 vectors, and split them into shorts.
86           Interleave the multipy/accumulate for the resample
87           filter with the loads to hide the 3 cycle latency
88           the vec_madds have. */
89        tv = (vector unsigned char *) &s[0 * wrap];
90        tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[i * wrap]));
91        srchv[0].v = (vector signed short) vec_mergeh(zero, tmp);
92        srclv[0].v = (vector signed short) vec_mergel(zero, tmp);
93        sumhv = vec_madds(srchv[0].v, fv[0].v, zeros);
94        sumlv = vec_madds(srclv[0].v, fv[0].v, zeros);
95
96        tv = (vector unsigned char *) &s[1 * wrap];
97        tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[1 * wrap]));
98        srchv[1].v = (vector signed short) vec_mergeh(zero, tmp);
99        srclv[1].v = (vector signed short) vec_mergel(zero, tmp);
100        sumhv = vec_madds(srchv[1].v, fv[1].v, sumhv);
101        sumlv = vec_madds(srclv[1].v, fv[1].v, sumlv);
102
103        tv = (vector unsigned char *) &s[2 * wrap];
104        tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[2 * wrap]));
105        srchv[2].v = (vector signed short) vec_mergeh(zero, tmp);
106        srclv[2].v = (vector signed short) vec_mergel(zero, tmp);
107        sumhv = vec_madds(srchv[2].v, fv[2].v, sumhv);
108        sumlv = vec_madds(srclv[2].v, fv[2].v, sumlv);
109
110        tv = (vector unsigned char *) &s[3 * wrap];
111        tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[3 * wrap]));
112        srchv[3].v = (vector signed short) vec_mergeh(zero, tmp);
113        srclv[3].v = (vector signed short) vec_mergel(zero, tmp);
114        sumhv = vec_madds(srchv[3].v, fv[3].v, sumhv);
115        sumlv = vec_madds(srclv[3].v, fv[3].v, sumlv);
116
117        /* Pack the results into our destination vector,
118           and do an aligned write of that back to memory. */
119        dstv = vec_packsu(sumhv, sumlv) ;
120        vec_st(dstv, 0, (vector unsigned char *) dst);
121
122        dst+=16;
123        s+=16;
124        dst_width-=16;
125    }
126
127    /* If there are any leftover pixels, resample them
128       with the slow scalar method. */
129    while(dst_width>0) {
130        sum = s[0 * wrap] * filter[0] +
131        s[1 * wrap] * filter[1] +
132        s[2 * wrap] * filter[2] +
133        s[3 * wrap] * filter[3];
134        sum = sum >> FILTER_BITS;
135        if (sum<0) sum = 0; else if (sum>255) sum=255;
136        dst[0] = sum;
137        dst++;
138        s++;
139        dst_width--;
140    }
141}
142
143