libmpcodecs/cmmx.h

   1 /*
   2  * x86 MMX and MMX2 packed byte operations in portable C.
   3  * Extra instructions: pdiffub, pcmpzb, psumbw, pcmpgtub
   4  * Author: Zoltan Hidvegi
   5  *
   6  * This file is part of MPlayer.
   7  *
   8  * MPlayer is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * MPlayer is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along
  19  * with MPlayer; if not, write to the Free Software Foundation, Inc.,
  20  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  21  */
  22
  23 #ifndef MPLAYER_CMMX_H
  24 #define MPLAYER_CMMX_H
  25
  26 typedef unsigned long cmmx_t;
  27
  28 #define ONE_BYTES (~(cmmx_t)0 / 255)
  29 #define SIGN_BITS (ONE_BYTES << 7)
  30 #define LOWBW_MASK (~(cmmx_t)0 / 257)
  31
  32 static inline cmmx_t
  33 paddb(cmmx_t a, cmmx_t b)
  34 {
  35     return ((a & ~SIGN_BITS) + (b & ~SIGN_BITS)) ^ ((a^b) & SIGN_BITS);
  36 }
  37
  38 static inline cmmx_t
  39 psubb(cmmx_t a, cmmx_t b)
  40 {
  41     return ((a | SIGN_BITS) - (b & ~SIGN_BITS)) ^ (~(a^b) & SIGN_BITS);
  42 }
  43
  44 static inline cmmx_t
  45 paddusb(cmmx_t a, cmmx_t b)
  46 {
  47     cmmx_t s = (a & ~SIGN_BITS) + (b & ~SIGN_BITS);
  48     cmmx_t abs = (a | b) & SIGN_BITS;
  49     cmmx_t c = abs & (s | (a & b));
  50     return s | abs | (abs - (c >> 7));
  51 }
  52
  53 static inline cmmx_t
  54 paddusb_s(cmmx_t a, cmmx_t b)
  55 {
  56     cmmx_t sum = a+b;
  57     cmmx_t ov = sum & SIGN_BITS;
  58     return sum + (sum ^ (ov - (ov>>7)));
  59 }
  60
  61 static inline cmmx_t
  62 psubusb(cmmx_t a, cmmx_t b)
  63 {
  64     cmmx_t s = (a | SIGN_BITS) - (b & ~SIGN_BITS);
  65     cmmx_t anb = a & ~b;
  66     cmmx_t c = (anb | (s & ~(a^b))) & SIGN_BITS;
  67     return s & ((c & anb) | (c - (c >> 7)));
  68 }
  69
  70 static inline cmmx_t
  71 psubusb_s(cmmx_t a, cmmx_t b)
  72 {
  73     cmmx_t d = (a|SIGN_BITS) - b;
  74     cmmx_t m = d & SIGN_BITS;
  75     return d & (m - (m>>7));
  76 }
  77
  78 static inline cmmx_t
  79 pcmpgtub(cmmx_t b, cmmx_t a)
  80 {
  81     cmmx_t s = (a | SIGN_BITS) - (b & ~SIGN_BITS);
  82     cmmx_t ret = ((~a & b) | (~s & ~(a ^ b))) & SIGN_BITS;
  83     return ret | (ret - (ret >> 7));
  84 }
  85
  86 static inline cmmx_t
  87 pdiffub(cmmx_t a, cmmx_t b)
  88 {
  89     cmmx_t xs = (~a ^ b) & SIGN_BITS;
  90     cmmx_t s = ((a | SIGN_BITS) - (b & ~SIGN_BITS)) ^ xs;
  91     cmmx_t gt = ((~a & b) | (s & xs)) & SIGN_BITS;
  92     cmmx_t gt7 = gt >> 7;
  93     return (s ^ gt ^ (gt - gt7)) + gt7;
  94 }
  95
  96 static inline cmmx_t
  97 pdiffub_s(cmmx_t a, cmmx_t b)
  98 {
  99     cmmx_t d = (a|SIGN_BITS) - b;
 100     cmmx_t g = (~d & SIGN_BITS) >> 7;
 101     return (d ^ (SIGN_BITS-g)) + g;
 102 }
 103
 104 static inline cmmx_t
 105 pmaxub(cmmx_t a, cmmx_t b)
 106 {
 107     return psubusb(a,b) + b;
 108 }
 109
 110 static inline cmmx_t
 111 pminub(cmmx_t a, cmmx_t b)
 112 {
 113     return paddusb(a,~b) - ~b;
 114 }
 115
 116 static inline cmmx_t
 117 pminub_s(cmmx_t a, cmmx_t b)
 118 {
 119     cmmx_t d = (a|SIGN_BITS) - b;
 120     cmmx_t m = ~SIGN_BITS + ((d&SIGN_BITS)>>7);
 121     return ((d&m) + b) & ~SIGN_BITS;
 122 }
 123
 124 static inline cmmx_t
 125 pavgb(cmmx_t a, cmmx_t b)
 126 {
 127     cmmx_t ao = a & ONE_BYTES;
 128     cmmx_t bo = b & ONE_BYTES;
 129     return ((a^ao)>>1) + ((b^bo)>>1) + (ao|bo);
 130 }
 131
 132 static inline cmmx_t
 133 pavgb_s(cmmx_t a, cmmx_t b)
 134 {
 135     return ((a+b+ONE_BYTES)>>1) & ~SIGN_BITS;
 136 }
 137
 138 static inline cmmx_t
 139 p31avgb(cmmx_t a, cmmx_t b)
 140 {
 141     cmmx_t ao = a & (3*ONE_BYTES);
 142     cmmx_t bo = b & (3*ONE_BYTES);
 143     return 3*((a^ao)>>2) + ((b^bo)>>2) +
 144         (((3*ao+bo+2*ONE_BYTES)>>2) & (3*ONE_BYTES));
 145 }
 146
 147 static inline cmmx_t
 148 p31avgb_s(cmmx_t a, cmmx_t b)
 149 {
 150     cmmx_t avg = ((a+b)>>1) & ~SIGN_BITS;
 151     return pavgb_s(avg, a);
 152 }
 153
 154 static inline unsigned long
 155 psumbw(cmmx_t a)
 156 {
 157     cmmx_t t = (a & LOWBW_MASK) + ((a>>8) & LOWBW_MASK);
 158     unsigned long ret =
 159         (unsigned long)t + (unsigned long)(t >> (4*sizeof(cmmx_t)));
 160     if (sizeof(cmmx_t) > 4)
 161         ret += ret >> 16;
 162     return ret & 0xffff;
 163 }
 164
 165 static inline unsigned long
 166 psumbw_s(cmmx_t a)
 167 {
 168     unsigned long ret =
 169         (unsigned long)a + (unsigned long)(a >> (4*sizeof(cmmx_t)));
 170     if (sizeof(cmmx_t) <= 4)
 171         return (ret & 0xff) + ((ret>>8) & 0xff);
 172     ret = (ret & 0xff00ff) + ((ret>>8) & 0xff00ff);
 173     ret += ret >> 16;
 174     return ret & 0xffff;
 175 }
 176
 177 static inline unsigned long
 178 psadbw(cmmx_t a, cmmx_t b)
 179 {
 180     return psumbw(pdiffub(a,b));
 181 }
 182
 183 static inline unsigned long
 184 psadbw_s(cmmx_t a, cmmx_t b)
 185 {
 186     return psumbw_s(pdiffub_s(a,b));
 187 }
 188
 189 static inline cmmx_t
 190 pcmpzb(cmmx_t a)
 191 {
 192     cmmx_t ret = (((a | SIGN_BITS) - ONE_BYTES) | a) & SIGN_BITS;
 193     return ~(ret | (ret - (ret >> 7)));
 194 }
 195
 196 static inline cmmx_t
 197 pcmpeqb(cmmx_t a, cmmx_t b)
 198 {
 199     return pcmpzb(a ^ b);
 200 }
 201
 202 #endif /* MPLAYER_CMMX_H */