apps/codecs/lib/asm_mcf5249.h

   1 /***************************************************************************
   2  *             __________               __   ___.
   3  *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
   4  *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
   5  *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
   6  *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
   7  *                     \/            \/     \/    \/            \/
   8  *
   9  * Copyright (C) 2005 by Pedro Vasconcelos
  10  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU General Public License
  13  * as published by the Free Software Foundation; either version 2
  14  * of the License, or (at your option) any later version.
  15  *
  16  * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
  17  * KIND, either express or implied.
  18  *
  19  ****************************************************************************/
  20 /* asm routines for wide math on the MCF5249 */
  21
  22 #if defined(CPU_COLDFIRE)
  23
  24 /* attribute for 16-byte alignment */
  25 #define LINE_ATTR   __attribute__ ((aligned (16)))
  26
  27 #ifndef _V_WIDE_MATH
  28 #define _V_WIDE_MATH
  29
  30 #define MB()
  31
  32 static inline int32_t MULT32(int32_t x, int32_t y) {
  33
  34   asm volatile ("mac.l %[x], %[y], %%acc0;"    /* multiply & shift  */
  35                 "movclr.l %%acc0, %[x];"       /* move & clear acc */
  36                 "asr.l #1, %[x];"              /* no overflow test */
  37                 : [x] "+&d" (x)
  38                 : [y] "r" (y)
  39                 : "cc");
  40   return x;
  41 }
  42
  43 static inline int32_t MULT31(int32_t x, int32_t y) {
  44   asm volatile ("mac.l %[x], %[y], %%acc0;" /* multiply */
  45                 "movclr.l %%acc0, %[x];"    /* move and clear */
  46                 : [x] "+&r" (x)
  47                 : [y] "r" (y)
  48                 : "cc");
  49   return x;
  50 }
  51
  52 static inline int32_t MULT31_SHIFT15(int32_t x, int32_t y) {
  53   int32_t r;
  54
  55   asm volatile ("mac.l %[x], %[y], %%acc0;"  /* multiply */
  56                 "mulu.l %[y], %[x];"         /* get lower half, avoid emac stall */
  57                 "movclr.l %%acc0, %[r];"     /* get higher half */
  58                 "asl.l #8, %[r];"            /* hi<<16, plus one free */
  59                 "asl.l #8, %[r];"
  60                 "lsr.l #8, %[x];"            /* (unsigned)lo >> 15 */
  61                 "lsr.l #7, %[x];"
  62                 "or.l %[x], %[r];"           /* logical-or results */
  63                 : [r] "=&d" (r), [x] "+d" (x)
  64                 : [y] "d" (y)
  65                 : "cc");
  66   return r;
  67 }
  68
  69 static inline
  70 void XPROD31(int32_t  a, int32_t  b,
  71              int32_t  t, int32_t  v,
  72              int32_t *x, int32_t *y)
  73 {
  74   asm volatile ("mac.l %[a], %[t], %%acc0;"
  75                 "mac.l %[b], %[v], %%acc0;"
  76                 "mac.l %[b], %[t], %%acc1;"
  77                 "msac.l %[a], %[v], %%acc1;"
  78                 "movclr.l %%acc0, %[a];"
  79                 "move.l %[a], (%[x]);"
  80                 "movclr.l %%acc1, %[a];"
  81                 "move.l %[a], (%[y]);"
  82                 : [a] "+&r" (a)
  83                 : [x] "a" (x), [y] "a" (y),
  84                   [b] "r" (b), [t] "r" (t), [v] "r" (v)
  85                 : "cc", "memory");
  86 }
  87
  88 static inline
  89 void XNPROD31(int32_t  a, int32_t  b,
  90               int32_t  t, int32_t  v,
  91               int32_t *x, int32_t *y)
  92 {
  93   asm volatile ("mac.l %[a], %[t], %%acc0;"
  94                 "msac.l %[b], %[v], %%acc0;"
  95                 "mac.l %[b], %[t], %%acc1;"
  96                 "mac.l %[a], %[v], %%acc1;"
  97                 "movclr.l %%acc0, %[a];"
  98                 "move.l %[a], (%[x]);"
  99                 "movclr.l %%acc1, %[a];"
 100                 "move.l %[a], (%[y]);"
 101                 : [a] "+&r" (a)
 102                 : [x] "a" (x), [y] "a" (y),
 103                   [b] "r" (b), [t] "r" (t), [v] "r" (v)
 104                 : "cc", "memory");
 105 }
 106
 107 #if 0    /* canonical Tremor definition */
 108 #define XPROD32(_a, _b, _t, _v, _x, _y)         \
 109   { (_x)=MULT32(_a,_t)+MULT32(_b,_v);           \
 110     (_y)=MULT32(_b,_t)-MULT32(_a,_v); }
 111 #endif
 112
 113 /* this could lose the LSB by overflow, but i don't think it'll ever happen.
 114    if anyone think they can hear a bug caused by this, please try the above
 115    version. */
 116 #define XPROD32(_a, _b, _t, _v, _x, _y)     \
 117   asm volatile ("mac.l %[a], %[t], %%acc0;" \
 118                 "mac.l %[b], %[v], %%acc0;" \
 119                 "mac.l %[b], %[t], %%acc1;" \
 120                 "msac.l %[a], %[v], %%acc1;" \
 121                 "movclr.l %%acc0, %[x];" \
 122                 "asr.l #1, %[x];" \
 123                 "movclr.l %%acc1, %[y];" \
 124                 "asr.l #1, %[y];" \
 125                 : [x] "=&d" (_x), [y] "=&d" (_y) \
 126                 : [a] "r" (_a), [b] "r" (_b), \
 127                   [t] "r" (_t), [v] "r" (_v) \
 128                 : "cc");
 129
 130 #ifndef _V_VECT_OPS
 131 #define _V_VECT_OPS
 132
 133 /* asm versions of vector operations for block.c, window.c */
 134 /* assumes MAC is initialized & accumulators cleared */
 135 static inline
 136 void vect_add(int32_t *x, const int32_t *y, int n)
 137 {
 138   /* align to 16 bytes */
 139   while(n>0 && (int)x&15) {
 140     *x++ += *y++;
 141     n--;
 142   }
 143   asm volatile ("bra 1f;"
 144                 "0:"                          /* loop start */
 145                 "movem.l (%[x]), %%d0-%%d3;"  /* fetch values */
 146                 "movem.l (%[y]), %%a0-%%a3;"
 147                 /* add */
 148                 "add.l %%a0, %%d0;"
 149                 "add.l %%a1, %%d1;"
 150                 "add.l %%a2, %%d2;"
 151                 "add.l %%a3, %%d3;"
 152                 /* store and advance */
 153                 "movem.l %%d0-%%d3, (%[x]);"
 154                 "lea.l (4*4, %[x]), %[x];"
 155                 "lea.l (4*4, %[y]), %[y];"
 156                 "subq.l #4, %[n];"     /* done 4 elements */
 157                 "1: cmpi.l #4, %[n];"
 158                 "bge 0b;"
 159                 : [n] "+d" (n), [x] "+a" (x), [y] "+a" (y)
 160                 : : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
 161                     "cc", "memory");
 162   /* add final elements */
 163   while (n>0) {
 164     *x++ += *y++;
 165     n--;
 166   }
 167 }
 168
 169 static inline
 170 void vect_copy(int32_t *x, int32_t *y, int n)
 171 {
 172   /* align to 16 bytes */
 173   while(n>0 && (int)x&15) {
 174     *x++ = *y++;
 175     n--;
 176   }
 177   asm volatile ("bra 1f;"
 178                 "0:"                                    /* loop start */
 179                 "movem.l (%[y]), %%d0-%%d3;"            /* fetch values */
 180                 "movem.l %%d0-%%d3, (%[x]);"            /* store */
 181                 "lea.l (4*4, %[x]), %[x];"              /* advance */
 182                 "lea.l (4*4, %[y]), %[y];"
 183                 "subq.l #4, %[n];"                      /* done 4 elements */
 184                 "1: cmpi.l #4, %[n];"
 185                 "bge 0b;"
 186                 : [n] "+d" (n), [x] "+a" (x), [y] "+a" (y)
 187                 : : "%d0", "%d1", "%d2", "%d3", "cc", "memory");
 188   /* copy final elements */
 189   while (n>0) {
 190     *x++ = *y++;
 191     n--;
 192   }
 193 }
 194
 195 static inline
 196 void vect_mult_fw(int32_t *data, int32_t *window, int n)
 197 {
 198   /* ensure data is aligned to 16-bytes */
 199   while(n>0 && (int)data&15) {
 200     *data = MULT31(*data, *window);
 201     data++;
 202     window++;
 203     n--;
 204   }
 205   asm volatile ("movem.l (%[d]), %%d0-%%d3;"  /* loop start */
 206                 "movem.l (%[w]), %%a0-%%a3;"  /* pre-fetch registers */
 207                 "lea.l (4*4, %[w]), %[w];"
 208                 "bra 1f;"               /* jump to loop condition */
 209                 "0:" /* loop body */
 210                 /* multiply and load next window values */
 211                 "mac.l %%d0, %%a0, (%[w])+, %%a0, %%acc0;"
 212                 "mac.l %%d1, %%a1, (%[w])+, %%a1, %%acc1;"
 213                 "mac.l %%d2, %%a2, (%[w])+, %%a2, %%acc2;"
 214                 "mac.l %%d3, %%a3, (%[w])+, %%a3, %%acc3;"
 215                 "movclr.l %%acc0, %%d0;"  /* get the products */
 216                 "movclr.l %%acc1, %%d1;"
 217                 "movclr.l %%acc2, %%d2;"
 218                 "movclr.l %%acc3, %%d3;"
 219                 /* store and advance */
 220                 "movem.l %%d0-%%d3, (%[d]);"
 221                 "lea.l (4*4, %[d]), %[d];"
 222                 "movem.l (%[d]), %%d0-%%d3;"
 223                 "subq.l #4, %[n];"     /* done 4 elements */
 224                 "1: cmpi.l #4, %[n];"
 225                 "bge 0b;"
 226                 /* multiply final elements */
 227                 "tst.l %[n];"
 228                 "beq 1f;"      /* n=0 */
 229                 "mac.l %%d0, %%a0, %%acc0;"
 230                 "movclr.l %%acc0, %%d0;"
 231                 "move.l %%d0, (%[d])+;"
 232                 "subq.l #1, %[n];"
 233                 "beq 1f;"     /* n=1 */
 234                 "mac.l %%d1, %%a1, %%acc0;"
 235                 "movclr.l %%acc0, %%d1;"
 236                 "move.l %%d1, (%[d])+;"
 237                 "subq.l #1, %[n];"
 238                 "beq 1f;"     /* n=2 */
 239                 /* otherwise n = 3 */
 240                 "mac.l %%d2, %%a2, %%acc0;"
 241                 "movclr.l %%acc0, %%d2;"
 242                 "move.l %%d2, (%[d])+;"
 243                 "1:"
 244                 : [n] "+d" (n), [d] "+a" (data), [w] "+a" (window)
 245                 : : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
 246                     "cc", "memory");
 247 }
 248
 249 static inline
 250 void vect_mult_bw(int32_t *data, int32_t *window, int n)
 251 {
 252   /* ensure at least data is aligned to 16-bytes */
 253   while(n>0 && (int)data&15) {
 254     *data = MULT31(*data, *window);
 255     data++;
 256     window--;
 257     n--;
 258   }
 259   asm volatile ("lea.l (-3*4, %[w]), %[w];"     /* loop start */
 260                 "movem.l (%[d]), %%d0-%%d3;"    /* pre-fetch registers */
 261                 "movem.l (%[w]), %%a0-%%a3;"
 262                 "bra 1f;"               /* jump to loop condition */
 263                 "0:" /* loop body */
 264                 /* multiply and load next window value */
 265                 "mac.l %%d0, %%a3, -(%[w]), %%a3, %%acc0;"
 266                 "mac.l %%d1, %%a2, -(%[w]), %%a2, %%acc1;"
 267                 "mac.l %%d2, %%a1, -(%[w]), %%a1, %%acc2;"
 268                 "mac.l %%d3, %%a0, -(%[w]), %%a0, %%acc3;"
 269                 "movclr.l %%acc0, %%d0;"  /* get the products */
 270                 "movclr.l %%acc1, %%d1;"
 271                 "movclr.l %%acc2, %%d2;"
 272                 "movclr.l %%acc3, %%d3;"
 273                 /* store and advance */
 274                 "movem.l %%d0-%%d3, (%[d]);"
 275                 "lea.l (4*4, %[d]), %[d];"
 276                 "movem.l (%[d]), %%d0-%%d3;"
 277                 "subq.l #4, %[n];"     /* done 4 elements */
 278                 "1: cmpi.l #4, %[n];"
 279                 "bge 0b;"
 280                 /* multiply final elements */
 281                 "tst.l %[n];"
 282                 "beq 1f;"      /* n=0 */
 283                 "mac.l %%d0, %%a3, %%acc0;"
 284                 "movclr.l %%acc0, %%d0;"
 285                 "move.l %%d0, (%[d])+;"
 286                 "subq.l #1, %[n];"
 287                 "beq 1f;"     /* n=1 */
 288                 "mac.l %%d1, %%a2, %%acc0;"
 289                 "movclr.l %%acc0, %%d1;"
 290                 "move.l %%d1, (%[d])+;"
 291                 "subq.l #1, %[n];"
 292                 "beq 1f;"     /* n=2 */
 293                 /* otherwise n = 3 */
 294                 "mac.l %%d2, %%a1, %%acc0;"
 295                 "movclr.l %%acc0, %%d2;"
 296                 "move.l %%d2, (%[d])+;"
 297                 "1:"
 298                 : [n] "+d" (n), [d] "+a" (data), [w] "+a" (window)
 299                 : : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
 300                     "cc", "memory");
 301 }
 302
 303 #endif
 304
 305 #endif
 306
 307 #ifndef _V_CLIP_MATH
 308 #define _V_CLIP_MATH
 309
 310 /* this is portable C and simple; why not use this as default? */
 311 static inline int32_t CLIP_TO_15(register int32_t x) {
 312   register int32_t hi=32767, lo=-32768;
 313   return (x>=hi ? hi : (x<=lo ? lo : x));
 314 }
 315
 316 #endif
 317 #else
 318 #define LINE_ATTR
 319 #endif
 320