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 //#include "os_types.h"
  23
  24 #if defined(CPU_COLDFIRE)
  25
  26 /* attribute for 16-byte alignment */
  27 #define LINE_ATTR   __attribute__ ((aligned (16)))
  28
  29 #ifndef _V_WIDE_MATH
  30 #define _V_WIDE_MATH
  31
  32 #define MB()
  33
  34 static inline int32_t MULT32(int32_t x, int32_t y) {
  35
  36   asm volatile ("mac.l %[x], %[y], %%acc0;"    /* multiply & shift  */
  37                 "movclr.l %%acc0, %[x];"       /* move & clear acc */
  38                 "asr.l #1, %[x];"              /* no overflow test */
  39                 : [x] "+&d" (x)
  40                 : [y] "r" (y)
  41                 : "cc");
  42   return x;
  43 }
  44
  45 static inline int32_t MULT31(int32_t x, int32_t y) {
  46
  47   asm volatile ("mac.l %[x], %[y], %%acc0;" /* multiply */
  48                 "movclr.l %%acc0, %[x];"    /* move and clear */
  49                 : [x] "+&r" (x)
  50                 : [y] "r" (y)
  51                 : "cc");
  52   return x;
  53 }
  54
  55
  56 static inline int32_t MULT31_SHIFT15(int32_t x, int32_t y) {
  57   int32_t r;
  58
  59   asm volatile ("mac.l %[x], %[y], %%acc0;"  /* multiply */
  60                 "mulu.l %[y], %[x];"         /* get lower half, avoid emac stall */
  61                 "movclr.l %%acc0, %[r];"     /* get higher half */
  62                 "asl.l #8, %[r];"            /* hi<<16, plus one free */
  63                 "asl.l #8, %[r];"
  64                 "lsr.l #8, %[x];"            /* (unsigned)lo >> 15 */
  65                 "lsr.l #7, %[x];"
  66                 "or.l %[x], %[r];"           /* logical-or results */
  67                 : [r] "=&d" (r), [x] "+d" (x)
  68                 : [y] "d" (y)
  69                 : "cc");
  70   return r;
  71 }
  72
  73
  74 static inline
  75 void XPROD31(int32_t  a, int32_t  b,
  76              int32_t  t, int32_t  v,
  77              int32_t *x, int32_t *y)
  78 {
  79   asm volatile ("mac.l %[a], %[t], %%acc0;"
  80                 "mac.l %[b], %[v], %%acc0;"
  81                 "mac.l %[b], %[t], %%acc1;"
  82                 "msac.l %[a], %[v], %%acc1;"
  83                 "movclr.l %%acc0, %[a];"
  84                 "move.l %[a], (%[x]);"
  85                 "movclr.l %%acc1, %[a];"
  86                 "move.l %[a], (%[y]);"
  87                 : [a] "+&r" (a)
  88                 : [x] "a" (x), [y] "a" (y),
  89                   [b] "r" (b), [t] "r" (t), [v] "r" (v)
  90                 : "cc", "memory");
  91 }
  92
  93
  94 static inline
  95 void XNPROD31(int32_t  a, int32_t  b,
  96               int32_t  t, int32_t  v,
  97               int32_t *x, int32_t *y)
  98 {
  99   asm volatile ("mac.l %[a], %[t], %%acc0;"
 100                 "msac.l %[b], %[v], %%acc0;"
 101                 "mac.l %[b], %[t], %%acc1;"
 102                 "mac.l %[a], %[v], %%acc1;"
 103                 "movclr.l %%acc0, %[a];"
 104                 "move.l %[a], (%[x]);"
 105                 "movclr.l %%acc1, %[a];"
 106                 "move.l %[a], (%[y]);"
 107                 : [a] "+&r" (a)
 108                 : [x] "a" (x), [y] "a" (y),
 109                   [b] "r" (b), [t] "r" (t), [v] "r" (v)
 110                 : "cc", "memory");
 111 }
 112
 113
 114 #if 0    /* canonical Tremor definition */
 115 #define XPROD32(_a, _b, _t, _v, _x, _y)         \
 116   { (_x)=MULT32(_a,_t)+MULT32(_b,_v);           \
 117     (_y)=MULT32(_b,_t)-MULT32(_a,_v); }
 118 #endif
 119
 120 /* this could lose the LSB by overflow, but i don't think it'll ever happen.
 121    if anyone think they can hear a bug caused by this, please try the above
 122    version. */
 123 #define XPROD32(_a, _b, _t, _v, _x, _y)     \
 124   asm volatile ("mac.l %[a], %[t], %%acc0;" \
 125                 "mac.l %[b], %[v], %%acc0;" \
 126                 "mac.l %[b], %[t], %%acc1;" \
 127                 "msac.l %[a], %[v], %%acc1;" \
 128                 "movclr.l %%acc0, %[x];" \
 129                 "asr.l #1, %[x];" \
 130                 "movclr.l %%acc1, %[y];" \
 131                 "asr.l #1, %[y];" \
 132                 : [x] "=&d" (_x), [y] "=&d" (_y) \
 133                 : [a] "r" (_a), [b] "r" (_b), \
 134                   [t] "r" (_t), [v] "r" (_v) \
 135                 : "cc");
 136
 137 #ifndef _V_VECT_OPS
 138 #define _V_VECT_OPS
 139
 140 /* asm versions of vector operations for block.c, window.c */
 141 /* assumes MAC is initialized & accumulators cleared */
 142 static inline
 143 void vect_add(int32_t *x, int32_t *y, int n)
 144 {
 145   /* align to 16 bytes */
 146   while(n>0 && (int)x&16) {
 147     *x++ += *y++;
 148     n--;
 149   }
 150   asm volatile ("bra 1f;"
 151                 "0:"                          /* loop start */
 152                 "movem.l (%[x]), %%d0-%%d3;"  /* fetch values */
 153                 "movem.l (%[y]), %%a0-%%a3;"
 154                 /* add */
 155                 "add.l %%a0, %%d0;"
 156                 "add.l %%a1, %%d1;"
 157                 "add.l %%a2, %%d2;"
 158                 "add.l %%a3, %%d3;"
 159                 /* store and advance */
 160                 "movem.l %%d0-%%d3, (%[x]);"
 161                 "lea.l (4*4, %[x]), %[x];"
 162                 "lea.l (4*4, %[y]), %[y];"
 163                 "subq.l #4, %[n];"     /* done 4 elements */
 164                 "1: cmpi.l #4, %[n];"
 165                 "bge 0b;"
 166                 : [n] "+d" (n), [x] "+a" (x), [y] "+a" (y)
 167                 : : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
 168                     "cc", "memory");
 169   /* add final elements */
 170   while (n>0) {
 171     *x++ += *y++;
 172     n--;
 173   }
 174 }
 175
 176 static inline
 177 void vect_copy(int32_t *x, int32_t *y, int n)
 178 {
 179   /* align to 16 bytes */
 180   while(n>0 && (int)x&16) {
 181     *x++ = *y++;
 182     n--;
 183   }
 184   asm volatile ("bra 1f;"
 185                 "0:"                                    /* loop start */
 186                 "movem.l (%[y]), %%d0-%%d3;"            /* fetch values */
 187                 "movem.l %%d0-%%d3, (%[x]);"            /* store */
 188                 "lea.l (4*4, %[x]), %[x];"              /* advance */
 189                 "lea.l (4*4, %[y]), %[y];"
 190                 "subq.l #4, %[n];"                      /* done 4 elements */
 191                 "1: cmpi.l #4, %[n];"
 192                 "bge 0b;"
 193                 : [n] "+d" (n), [x] "+a" (x), [y] "+a" (y)
 194                 : : "%d0", "%d1", "%d2", "%d3", "cc", "memory");
 195   /* copy final elements */
 196   while (n>0) {
 197     *x++ = *y++;
 198     n--;
 199   }
 200 }
 201
 202
 203 static inline
 204 void vect_mult_fw(int32_t *data, int32_t *window, int n)
 205 {
 206   /* ensure data is aligned to 16-bytes */
 207   while(n>0 && (int)data%16) {
 208     *data = MULT31(*data, *window);
 209     data++;
 210     window++;
 211     n--;
 212   }
 213   asm volatile ("movem.l (%[d]), %%d0-%%d3;"  /* loop start */
 214                 "movem.l (%[w]), %%a0-%%a3;"  /* pre-fetch registers */
 215                 "lea.l (4*4, %[w]), %[w];"
 216                 "bra 1f;"               /* jump to loop condition */
 217                 "0:" /* loop body */
 218                 /* multiply and load next window values */
 219                 "mac.l %%d0, %%a0, (%[w])+, %%a0, %%acc0;"
 220                 "mac.l %%d1, %%a1, (%[w])+, %%a1, %%acc1;"
 221                 "mac.l %%d2, %%a2, (%[w])+, %%a2, %%acc2;"
 222                 "mac.l %%d3, %%a3, (%[w])+, %%a3, %%acc3;"
 223                 "movclr.l %%acc0, %%d0;"  /* get the products */
 224                 "movclr.l %%acc1, %%d1;"
 225                 "movclr.l %%acc2, %%d2;"
 226                 "movclr.l %%acc3, %%d3;"
 227                 /* store and advance */
 228                 "movem.l %%d0-%%d3, (%[d]);"
 229                 "lea.l (4*4, %[d]), %[d];"
 230                 "movem.l (%[d]), %%d0-%%d3;"
 231                 "subq.l #4, %[n];"     /* done 4 elements */
 232                 "1: cmpi.l #4, %[n];"
 233                 "bge 0b;"
 234                 /* multiply final elements */
 235                 "tst.l %[n];"
 236                 "beq 1f;"      /* n=0 */
 237                 "mac.l %%d0, %%a0, %%acc0;"
 238                 "movclr.l %%acc0, %%d0;"
 239                 "move.l %%d0, (%[d])+;"
 240                 "subq.l #1, %[n];"
 241                 "beq 1f;"     /* n=1 */
 242                 "mac.l %%d1, %%a1, %%acc0;"
 243                 "movclr.l %%acc0, %%d1;"
 244                 "move.l %%d1, (%[d])+;"
 245                 "subq.l #1, %[n];"
 246                 "beq 1f;"     /* n=2 */
 247                 /* otherwise n = 3 */
 248                 "mac.l %%d2, %%a2, %%acc0;"
 249                 "movclr.l %%acc0, %%d2;"
 250                 "move.l %%d2, (%[d])+;"
 251                 "1:"
 252                 : [n] "+d" (n), [d] "+a" (data), [w] "+a" (window)
 253                 : : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
 254                     "cc", "memory");
 255 }
 256
 257 static inline
 258 void vect_mult_bw(int32_t *data, int32_t *window, int n)
 259 {
 260   /* ensure at least data is aligned to 16-bytes */
 261   while(n>0 && (int)data%16) {
 262     *data = MULT31(*data, *window);
 263     data++;
 264     window--;
 265     n--;
 266   }
 267   asm volatile ("lea.l (-3*4, %[w]), %[w];"     /* loop start */
 268                 "movem.l (%[d]), %%d0-%%d3;"    /* pre-fetch registers */
 269                 "movem.l (%[w]), %%a0-%%a3;"
 270                 "bra 1f;"               /* jump to loop condition */
 271                 "0:" /* loop body */
 272                 /* multiply and load next window value */
 273                 "mac.l %%d0, %%a3, -(%[w]), %%a3, %%acc0;"
 274                 "mac.l %%d1, %%a2, -(%[w]), %%a2, %%acc1;"
 275                 "mac.l %%d2, %%a1, -(%[w]), %%a1, %%acc2;"
 276                 "mac.l %%d3, %%a0, -(%[w]), %%a0, %%acc3;"
 277                 "movclr.l %%acc0, %%d0;"  /* get the products */
 278                 "movclr.l %%acc1, %%d1;"
 279                 "movclr.l %%acc2, %%d2;"
 280                 "movclr.l %%acc3, %%d3;"
 281                 /* store and advance */
 282                 "movem.l %%d0-%%d3, (%[d]);"
 283                 "lea.l (4*4, %[d]), %[d];"
 284                 "movem.l (%[d]), %%d0-%%d3;"
 285                 "subq.l #4, %[n];"     /* done 4 elements */
 286                 "1: cmpi.l #4, %[n];"
 287                 "bge 0b;"
 288                 /* multiply final elements */
 289                 "tst.l %[n];"
 290                 "beq 1f;"      /* n=0 */
 291                 "mac.l %%d0, %%a3, %%acc0;"
 292                 "movclr.l %%acc0, %%d0;"
 293                 "move.l %%d0, (%[d])+;"
 294                 "subq.l #1, %[n];"
 295                 "beq 1f;"     /* n=1 */
 296                 "mac.l %%d1, %%a2, %%acc0;"
 297                 "movclr.l %%acc0, %%d1;"
 298                 "move.l %%d1, (%[d])+;"
 299                 "subq.l #1, %[n];"
 300                 "beq 1f;"     /* n=2 */
 301                 /* otherwise n = 3 */
 302                 "mac.l %%d2, %%a1, %%acc0;"
 303                 "movclr.l %%acc0, %%d2;"
 304                 "move.l %%d2, (%[d])+;"
 305                 "1:"
 306                 : [n] "+d" (n), [d] "+a" (data), [w] "+a" (window)
 307                 : : "%d0", "%d1", "%d2", "%d3", "%a0", "%a1", "%a2", "%a3",
 308                     "cc", "memory");
 309 }
 310
 311 #endif
 312
 313 #endif
 314
 315 #ifndef _V_CLIP_MATH
 316 #define _V_CLIP_MATH
 317
 318 /* this is portable C and simple; why not use this as default? */
 319 static inline int32_t CLIP_TO_15(register int32_t x) {
 320   register int32_t hi=32767, lo=-32768;
 321   return (x>=hi ? hi : (x<=lo ? lo : x));
 322 }
 323
 324 #endif
 325 #else
 326 #define LINE_ATTR
 327 #endif