libavcodec/ppc/int_altivec.c

   1 /*
   2  * Copyright (c) 2007 Luca Barbato <lu_zero@gentoo.org>
   3  *
   4  * This file is part of Libav.
   5  *
   6  * Libav 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  * Libav 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 Libav; 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
  23  ** integer misc ops.
  24  **/
  25
  26 #include "config.h"
  27 #if HAVE_ALTIVEC_H
  28 #include <altivec.h>
  29 #endif
  30
  31 #include "libavutil/attributes.h"
  32 #include "libavutil/ppc/types_altivec.h"
  33 #include "libavcodec/dsputil.h"
  34
  35 #include "dsputil_altivec.h"
  36
  37 static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2,
  38                                      int size) {
  39     int i, size16;
  40     vector signed char vpix1;
  41     vector signed short vpix2, vdiff, vpix1l,vpix1h;
  42     union { vector signed int vscore;
  43             int32_t score[4];
  44           } u;
  45     u.vscore = vec_splat_s32(0);
  46 //
  47 //XXX lazy way, fix it later
  48
  49 #define vec_unaligned_load(b) \
  50     vec_perm(vec_ld(0,b),vec_ld(15,b),vec_lvsl(0, b));
  51
  52     size16 = size >> 4;
  53     while(size16) {
  54 //        score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]);
  55         //load pix1 and the first batch of pix2
  56
  57         vpix1 = vec_unaligned_load(pix1);
  58         vpix2 = vec_unaligned_load(pix2);
  59         pix2 += 8;
  60         //unpack
  61         vpix1h = vec_unpackh(vpix1);
  62         vdiff  = vec_sub(vpix1h, vpix2);
  63         vpix1l = vec_unpackl(vpix1);
  64         // load another batch from pix2
  65         vpix2 = vec_unaligned_load(pix2);
  66         u.vscore = vec_msum(vdiff, vdiff, u.vscore);
  67         vdiff  = vec_sub(vpix1l, vpix2);
  68         u.vscore = vec_msum(vdiff, vdiff, u.vscore);
  69         pix1 += 16;
  70         pix2 += 8;
  71         size16--;
  72     }
  73     u.vscore = vec_sums(u.vscore, vec_splat_s32(0));
  74
  75     size %= 16;
  76     for (i = 0; i < size; i++) {
  77         u.score[3] += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]);
  78     }
  79     return u.score[3];
  80 }
  81
  82 static int32_t scalarproduct_int16_altivec(const int16_t *v1, const int16_t *v2,
  83                                            int order)
  84 {
  85     int i;
  86     LOAD_ZERO;
  87     const vec_s16 *pv;
  88     register vec_s16 vec1;
  89     register vec_s32 res = vec_splat_s32(0), t;
  90     int32_t ires;
  91
  92     for(i = 0; i < order; i += 8){
  93         pv = (const vec_s16*)v1;
  94         vec1 = vec_perm(pv[0], pv[1], vec_lvsl(0, v1));
  95         t = vec_msum(vec1, vec_ld(0, v2), zero_s32v);
  96         res = vec_sums(t, res);
  97         v1 += 8;
  98         v2 += 8;
  99     }
 100     res = vec_splat(res, 3);
 101     vec_ste(res, 0, &ires);
 102     return ires;
 103 }
 104
 105 static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul)
 106 {
 107     LOAD_ZERO;
 108     vec_s16 *pv1 = (vec_s16*)v1;
 109     register vec_s16 muls = {mul,mul,mul,mul,mul,mul,mul,mul};
 110     register vec_s16 t0, t1, i0, i1, i4;
 111     register vec_s16 i2 = vec_ld(0, v2), i3 = vec_ld(0, v3);
 112     register vec_s32 res = zero_s32v;
 113     register vec_u8 align = vec_lvsl(0, v2);
 114     int32_t ires;
 115     order >>= 4;
 116     do {
 117         i1 = vec_ld(16, v2);
 118         t0 = vec_perm(i2, i1, align);
 119         i2 = vec_ld(32, v2);
 120         t1 = vec_perm(i1, i2, align);
 121         i0 = pv1[0];
 122         i1 = pv1[1];
 123         res = vec_msum(t0, i0, res);
 124         res = vec_msum(t1, i1, res);
 125         i4 = vec_ld(16, v3);
 126         t0 = vec_perm(i3, i4, align);
 127         i3 = vec_ld(32, v3);
 128         t1 = vec_perm(i4, i3, align);
 129         pv1[0] = vec_mladd(t0, muls, i0);
 130         pv1[1] = vec_mladd(t1, muls, i1);
 131         pv1 += 2;
 132         v2  += 8;
 133         v3  += 8;
 134     } while(--order);
 135     res = vec_splat(vec_sums(res, zero_s32v), 3);
 136     vec_ste(res, 0, &ires);
 137     return ires;
 138 }
 139
 140 av_cold void ff_int_init_altivec(DSPContext *c, AVCodecContext *avctx)
 141 {
 142     c->ssd_int8_vs_int16 = ssd_int8_vs_int16_altivec;
 143     c->scalarproduct_int16 = scalarproduct_int16_altivec;
 144     c->scalarproduct_and_madd_int16 = scalarproduct_and_madd_int16_altivec;
 145 }