git.sesse.net Git - ffmpeg/blob - libavcodec/ps2/idct_mmi.c

   1 /*
   2  * Originally provided by Intel at Application Note AP-922.
   3  *
   4  * Column code adapted from Peter Gubanov.
   5  * Copyright (c) 2000-2001 Peter Gubanov <peter@elecard.net.ru>
   6  * http://www.elecard.com/peter/idct.shtml
   7  * rounding trick copyright (c) 2000 Michel Lespinasse <walken@zoy.org>
   8  *
   9  * MMI port and (c) 2002 by Leon van Stuivenberg
  10  *
  11  * This file is part of FFmpeg.
  12  *
  13  * FFmpeg is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU Lesser General Public
  15  * License as published by the Free Software Foundation; either
  16  * version 2.1 of the License, or (at your option) any later version.
  17  *
  18  * FFmpeg is distributed in the hope that it will be useful,
  19  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  20  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  21  * Lesser General Public License for more details.
  22  *
  23  * You should have received a copy of the GNU Lesser General Public
  24  * License along with FFmpeg; if not, write to the Free Software
  25  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  26  *
  27 */
  28
  29 #include "../common.h"
  30 #include "../dsputil.h"
  31 #include "mmi.h"
  32
  33 #define BITS_INV_ACC    5       // 4 or 5 for IEEE
  34 #define SHIFT_INV_ROW   (16 - BITS_INV_ACC)
  35 #define SHIFT_INV_COL   (1 + BITS_INV_ACC)
  36
  37 #define TG1             6518
  38 #define TG2             13573
  39 #define TG3             21895
  40 #define CS4             23170
  41
  42 #define ROUNDER_0       0
  43 #define ROUNDER_1       16
  44
  45 #define TAB_i_04        (32+0)
  46 #define TAB_i_17        (32+64)
  47 #define TAB_i_26        (32+128)
  48 #define TAB_i_35        (32+192)
  49
  50 #define TG_1_16         (32+256+0)
  51 #define TG_2_16         (32+256+16)
  52 #define TG_3_16         (32+256+32)
  53 #define COS_4_16        (32+256+48)
  54
  55 #define CLIPMAX         (32+256+64+0)
  56
  57 static short consttable[] align16 = {
  58 /* rounder 0*/  // assume SHIFT_INV_ROW == 11
  59  0x3ff, 1, 0x3ff, 1, 0x3ff, 1, 0x3ff, 1,
  60 /* rounder 1*/
  61  0x3ff, 0, 0x3ff, 0, 0x3ff, 0, 0x3ff, 0,
  62 /* row 0/4*/
  63  16384,  21407, -16384, -21407,  22725,  19266, -22725, -12873,
  64   8867,  16384,   8867,  16384,   4520,  12873,  -4520,  19266,
  65  16384,  -8867,  16384,  -8867,  12873, -22725,  19266, -22725,
  66  21407, -16384, -21407,  16384,  19266,   4520, -12873,   4520,
  67 /* row 1/7*/
  68  22725,  29692, -22725, -29692,  31521,  26722, -31521, -17855,
  69  12299,  22725,  12299,  22725,   6270,  17855,  -6270,  26722,
  70  22725, -12299,  22725, -12299,  17855, -31521,  26722, -31521,
  71  29692, -22725, -29692,  22725,  26722,   6270, -17855,   6270,
  72 /* row 2/6*/
  73  21407,  27969, -21407, -27969,  29692,  25172, -29692, -16819,
  74  11585,  21407,  11585,  21407,   5906,  16819,  -5906,  25172,
  75  21407, -11585,  21407, -11585,  16819, -29692,  25172, -29692,
  76  27969, -21407, -27969,  21407,  25172,   5906, -16819,   5906,
  77 /*row 3/5*/
  78  19266,  25172, -19266, -25172,  26722,  22654, -26722, -15137,
  79  10426,  19266,  10426,  19266,   5315,  15137,  -5315,  22654,
  80  19266, -10426,  19266, -10426,  15137, -26722,  22654, -26722,
  81  25172, -19266, -25172,  19266,  22654,   5315, -15137,   5315,
  82 /*column constants*/
  83  TG1, TG1, TG1, TG1, TG1, TG1, TG1, TG1,
  84  TG2, TG2, TG2, TG2, TG2, TG2, TG2, TG2,
  85  TG3, TG3, TG3, TG3, TG3, TG3, TG3, TG3,
  86  CS4, CS4, CS4, CS4, CS4, CS4, CS4, CS4,
  87 /* clamp */
  88  255, 255, 255, 255, 255, 255, 255, 255
  89 };
  90
  91
  92 #define DCT_8_INV_ROW1(blk, rowoff, taboff, rnd, outreg) { \
  93         lq(blk, rowoff, $16);   /* r16 = x7  x5  x3  x1  x6  x4  x2  x0 */ \
  94         /*slot*/ \
  95         lq($24, 0+taboff, $17); /* r17 = w */ \
  96         /*delay slot $16*/ \
  97         lq($24, 16+taboff, $18);/* r18 = w */ \
  98         prevh($16, $2);         /* r2  = x1  x3  x5  x7  x0  x2  x4  x6 */ \
  99         lq($24, 32+taboff, $19);/* r19 = w */ \
 100         phmadh($17, $16, $17);  /* r17 = b1"b0'a1"a0' */ \
 101         lq($24, 48+taboff, $20);/* r20 = w */ \
 102         phmadh($18, $2, $18);   /* r18 = b1'b0"a1'a0" */ \
 103         phmadh($19, $16, $19);  /* r19 = b3"b2'a3"a2' */ \
 104         phmadh($20, $2, $20);   /* r20 = b3'b2"a3'a2" */ \
 105         paddw($17, $18, $17);   /* r17 = (b1)(b0)(a1)(a0) */ \
 106         paddw($19, $20, $19);   /* r19 = (b3)(b2)(a3)(a2) */ \
 107         pcpyld($19, $17, $18);  /* r18 = (a3)(a2)(a1)(a0) */ \
 108         pcpyud($17, $19, $20);  /* r20 = (b3)(b2)(b1)(b0) */ \
 109         paddw($18, rnd, $18);   /* r18 = (a3)(a2)(a1)(a0) */\
 110         paddw($18, $20, $17);   /* r17 = ()()()(a0+b0) */ \
 111         psubw($18, $20, $20);   /* r20 = ()()()(a0-b0) */ \
 112         psraw($17, SHIFT_INV_ROW, $17); /* r17 = (y3 y2 y1 y0) */ \
 113         psraw($20, SHIFT_INV_ROW, $20); /* r20 = (y4 y5 y6 y7) */ \
 114         ppach($20, $17, outreg);/* out = y4 y5 y6 y7 y3 y2 y1 y0  Note order */ \
 115 \
 116         prevh(outreg, $2);        \
 117         pcpyud($2, $2, $2);        \
 118         pcpyld($2, outreg, outreg);        \
 119 }
 120
 121
 122 #define DCT_8_INV_COL8() \
 123 \
 124         lq($24, TG_3_16, $2);   /* r2  = tn3 */         \
 125 \
 126         pmulth($11, $2, $17);   /* r17 = x3 * tn3 (6420) */ \
 127         psraw($17, 15, $17);    \
 128         pmfhl_uw($3);           /* r3  = 7531 */        \
 129         psraw($3, 15, $3);      \
 130         pinteh($3, $17, $17);   /* r17 = x3 * tn3 */    \
 131         psubh($17, $13, $17);   /* r17 = tm35 */        \
 132 \
 133         pmulth($13, $2, $18);   /* r18 = x5 * tn3 (6420) */ \
 134         psraw($18, 15, $18);    \
 135         pmfhl_uw($3);           /* r3  = 7531 */        \
 136         psraw($3, 15, $3);      \
 137         pinteh($3, $18, $18);   /* r18 = x5 * tn3 */    \
 138         paddh($18, $11, $18);   /* r18 = tp35 */        \
 139 \
 140         lq($24, TG_1_16, $2);   /* r2  = tn1 */         \
 141 \
 142         pmulth($15, $2, $19);   /* r19 = x7 * tn1 (6420) */ \
 143         psraw($19, 15, $19);    \
 144         pmfhl_uw($3);           /* r3  = 7531 */        \
 145         psraw($3, 15, $3);      \
 146         pinteh($3, $19, $19);   /* r19 = x7 * tn1 */    \
 147         paddh($19, $9, $19);    /* r19 = tp17 */        \
 148 \
 149         pmulth($9, $2, $20);    /* r20 = x1 * tn1 (6420) */ \
 150         psraw($20, 15, $20);    \
 151         pmfhl_uw($3);           /* r3  = 7531 */        \
 152         psraw($3, 15, $3);      \
 153         pinteh($3, $20, $20);   /* r20 = x1 * tn1 */    \
 154         psubh($20, $15, $20);   /* r20 = tm17 */        \
 155 \
 156         psubh($19, $18, $3);    /* r3  = t1 */          \
 157         paddh($20, $17, $16);   /* r16 = t2 */          \
 158         psubh($20, $17, $23);   /* r23 = b3 */          \
 159         paddh($19, $18, $20);   /* r20 = b0 */          \
 160 \
 161         lq($24, COS_4_16, $2);  /* r2  = cs4 */         \
 162 \
 163         paddh($3, $16, $21);    /* r21 = t1+t2 */       \
 164         psubh($3, $16, $22);    /* r22 = t1-t2 */       \
 165 \
 166         pmulth($21, $2, $21);   /* r21 = cs4 * (t1+t2) 6420 */ \
 167         psraw($21, 15, $21);    \
 168         pmfhl_uw($3);           /* r3  = 7531 */        \
 169         psraw($3, 15, $3);      \
 170         pinteh($3, $21, $21);   /* r21 = b1 */          \
 171 \
 172         pmulth($22, $2, $22);   /* r22 = cs4 * (t1-t2) 6420 */ \
 173         psraw($22, 15, $22);    \
 174         pmfhl_uw($3);           /* r3  = 7531 */        \
 175         psraw($3, 15, $3);      \
 176         pinteh($3, $22, $22);   /* r22 = b2 */          \
 177 \
 178         lq($24, TG_2_16, $2);   /* r2  = tn2 */         \
 179 \
 180         pmulth($10, $2, $17);   /* r17 = x2 * tn2 (6420) */ \
 181         psraw($17, 15, $17);    \
 182         pmfhl_uw($3);           /* r3  = 7531 */        \
 183         psraw($3, 15, $3);      \
 184         pinteh($3, $17, $17);   /* r17 = x3 * tn3 */    \
 185         psubh($17, $14, $17);   /* r17 = tm26 */        \
 186 \
 187         pmulth($14, $2, $18);   /* r18 = x6 * tn2 (6420) */ \
 188         psraw($18, 15, $18);    \
 189         pmfhl_uw($3);           /* r3  = 7531 */        \
 190         psraw($3, 15, $3);      \
 191         pinteh($3, $18, $18);   /* r18 = x6 * tn2 */    \
 192         paddh($18, $10, $18);   /* r18 = tp26 */        \
 193 \
 194         paddh($8, $12, $2);     /* r2  = tp04 */        \
 195         psubh($8, $12, $3);     /* r3  = tm04 */        \
 196 \
 197         paddh($2, $18, $16);    /* r16 = a0 */          \
 198         psubh($2, $18, $19);    /* r19 = a3 */          \
 199         psubh($3, $17, $18);    /* r18 = a2 */          \
 200         paddh($3, $17, $17);    /* r17 = a1 */
 201
 202
 203 #define DCT_8_INV_COL8_STORE(blk) \
 204 \
 205         paddh($16, $20, $2);    /* y0  a0+b0 */ \
 206         psubh($16, $20, $16);   /* y7  a0-b0 */ \
 207         psrah($2, SHIFT_INV_COL, $2);           \
 208         psrah($16, SHIFT_INV_COL, $16);         \
 209         sq($2, 0, blk);                         \
 210         sq($16, 112, blk);                      \
 211 \
 212         paddh($17, $21, $3);    /* y1  a1+b1 */ \
 213         psubh($17, $21, $17);   /* y6  a1-b1 */ \
 214         psrah($3, SHIFT_INV_COL, $3);           \
 215         psrah($17, SHIFT_INV_COL, $17);         \
 216         sq($3, 16, blk);                        \
 217         sq($17, 96, blk);                       \
 218 \
 219         paddh($18, $22, $2);    /* y2  a2+b2 */ \
 220         psubh($18, $22, $18);   /* y5  a2-b2 */ \
 221         psrah($2, SHIFT_INV_COL, $2);           \
 222         psrah($18, SHIFT_INV_COL, $18);         \
 223         sq($2, 32, blk);                        \
 224         sq($18, 80, blk);                       \
 225 \
 226         paddh($19, $23, $3);    /* y3  a3+b3 */ \
 227         psubh($19, $23, $19);   /* y4  a3-b3 */ \
 228         psrah($3, SHIFT_INV_COL, $3);           \
 229         psrah($19, SHIFT_INV_COL, $19);         \
 230         sq($3, 48, blk);                        \
 231         sq($19, 64, blk);
 232
 233
 234
 235 #define DCT_8_INV_COL8_PMS() \
 236         paddh($16, $20, $2);    /* y0  a0+b0 */ \
 237         psubh($16, $20, $20);   /* y7  a0-b0 */ \
 238         psrah($2, SHIFT_INV_COL, $16);          \
 239         psrah($20, SHIFT_INV_COL, $20);         \
 240 \
 241         paddh($17, $21, $3);    /* y1  a1+b1 */ \
 242         psubh($17, $21, $21);   /* y6  a1-b1 */ \
 243         psrah($3, SHIFT_INV_COL, $17);          \
 244         psrah($21, SHIFT_INV_COL, $21);         \
 245 \
 246         paddh($18, $22, $2);    /* y2  a2+b2 */ \
 247         psubh($18, $22, $22);   /* y5  a2-b2 */ \
 248         psrah($2, SHIFT_INV_COL, $18);          \
 249         psrah($22, SHIFT_INV_COL, $22);         \
 250 \
 251         paddh($19, $23, $3);    /* y3  a3+b3 */ \
 252         psubh($19, $23, $23);   /* y4  a3-b3 */ \
 253         psrah($3, SHIFT_INV_COL, $19);          \
 254         psrah($23, SHIFT_INV_COL, $23);
 255
 256 #define PUT(rs)                 \
 257         pminh(rs, $11, $2);     \
 258         pmaxh($2, $0, $2);      \
 259         ppacb($0, $2, $2);      \
 260         sd3(2, 0, 4);           \
 261         __asm__ __volatile__ ("add $4, $5, $4");
 262
 263 #define DCT_8_INV_COL8_PUT() \
 264         PUT($16);        \
 265         PUT($17);        \
 266         PUT($18);        \
 267         PUT($19);        \
 268         PUT($23);        \
 269         PUT($22);        \
 270         PUT($21);        \
 271         PUT($20);
 272
 273 #define ADD(rs)          \
 274         ld3(4, 0, 2);        \
 275         pextlb($0, $2, $2);  \
 276         paddh($2, rs, $2);   \
 277         pminh($2, $11, $2);  \
 278         pmaxh($2, $0, $2);   \
 279         ppacb($0, $2, $2);   \
 280         sd3(2, 0, 4); \
 281         __asm__ __volatile__ ("add $4, $5, $4");
 282
 283 /*fixme: schedule*/
 284 #define DCT_8_INV_COL8_ADD() \
 285         ADD($16);        \
 286         ADD($17);        \
 287         ADD($18);        \
 288         ADD($19);        \
 289         ADD($23);        \
 290         ADD($22);        \
 291         ADD($21);        \
 292         ADD($20);
 293
 294
 295 void ff_mmi_idct(int16_t * block)
 296 {
 297         /* $4 = block */
 298         __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
 299         lq($24, ROUNDER_0, $8);
 300         lq($24, ROUNDER_1, $7);
 301         DCT_8_INV_ROW1($4, 0, TAB_i_04, $8, $8);
 302         DCT_8_INV_ROW1($4, 16, TAB_i_17, $7, $9);
 303         DCT_8_INV_ROW1($4, 32, TAB_i_26, $7, $10);
 304         DCT_8_INV_ROW1($4, 48, TAB_i_35, $7, $11);
 305         DCT_8_INV_ROW1($4, 64, TAB_i_04, $7, $12);
 306         DCT_8_INV_ROW1($4, 80, TAB_i_35, $7, $13);
 307         DCT_8_INV_ROW1($4, 96, TAB_i_26, $7, $14);
 308         DCT_8_INV_ROW1($4, 112, TAB_i_17, $7, $15);
 309         DCT_8_INV_COL8();
 310         DCT_8_INV_COL8_STORE($4);
 311
 312         //let savedtemp regs be saved
 313         __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
 314 }
 315
 316
 317 void ff_mmi_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
 318 {
 319         /* $4 = dest, $5 = line_size, $6 = block */
 320         __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
 321         lq($24, ROUNDER_0, $8);
 322         lq($24, ROUNDER_1, $7);
 323         DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8);
 324         DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9);
 325         DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10);
 326         DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11);
 327         DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12);
 328         DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13);
 329         DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14);
 330         DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15);
 331         DCT_8_INV_COL8();
 332         lq($24, CLIPMAX, $11);
 333         DCT_8_INV_COL8_PMS();
 334         DCT_8_INV_COL8_PUT();
 335
 336         //let savedtemp regs be saved
 337         __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
 338 }
 339
 340
 341 void ff_mmi_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
 342 {
 343         /* $4 = dest, $5 = line_size, $6 = block */
 344         __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
 345         lq($24, ROUNDER_0, $8);
 346         lq($24, ROUNDER_1, $7);
 347         DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8);
 348         DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9);
 349         DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10);
 350         DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11);
 351         DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12);
 352         DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13);
 353         DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14);
 354         DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15);
 355         DCT_8_INV_COL8();
 356         lq($24, CLIPMAX, $11);
 357         DCT_8_INV_COL8_PMS();
 358         DCT_8_INV_COL8_ADD();
 359
 360         //let savedtemp regs be saved
 361         __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
 362 }
 363