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