apps/codecs/Tremor/asm_arm.h

   1 /********************************************************************
   2  *                                                                  *
   3  * THIS FILE IS PART OF THE OggVorbis 'TREMOR' CODEC SOURCE CODE.   *
   4  *                                                                  *
   5  * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS     *
   6  * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
   7  * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING.       *
   8  *                                                                  *
   9  * THE OggVorbis 'TREMOR' SOURCE CODE IS (C) COPYRIGHT 1994-2002    *
  10  * BY THE Xiph.Org FOUNDATION http://www.xiph.org/                  *
  11  *                                                                  *
  12  ********************************************************************
  13
  14  function: arm7 and later wide math functions
  15
  16  ********************************************************************/
  17
  18 #ifdef _ARM_ASSEM_
  19
  20 #if !defined(_V_WIDE_MATH) && !defined(_LOW_ACCURACY_)
  21 #define _V_WIDE_MATH
  22
  23 static inline ogg_int32_t MULT32(ogg_int32_t x, ogg_int32_t y) {
  24   int lo,hi;
  25   asm volatile("smull\t%0, %1, %2, %3"
  26                : "=&r"(lo),"=&r"(hi)
  27                : "%r"(x),"r"(y)
  28                : "cc");
  29   return(hi);
  30 }
  31
  32 static inline ogg_int32_t MULT31(ogg_int32_t x, ogg_int32_t y) {
  33   return MULT32(x,y)<<1;
  34 }
  35
  36 static inline ogg_int32_t MULT31_SHIFT15(ogg_int32_t x, ogg_int32_t y) {
  37   int lo,hi;
  38   asm volatile("smull   %0, %1, %2, %3\n\t"
  39                "movs    %0, %0, lsr #15\n\t"
  40                "adc     %1, %0, %1, lsl #17\n\t"
  41                : "=&r"(lo),"=&r"(hi)
  42                : "%r"(x),"r"(y)
  43                : "cc");
  44   return(hi);
  45 }
  46
  47 #define MB() asm volatile ("" : : : "memory")
  48
  49 #define XPROD32(a, b, t, v, x, y) \
  50 { \
  51   long l; \
  52   asm(  "smull  %0, %1, %4, %6\n\t" \
  53         "smlal  %0, %1, %5, %7\n\t" \
  54         "rsb    %3, %4, #0\n\t" \
  55         "smull  %0, %2, %5, %6\n\t" \
  56         "smlal  %0, %2, %3, %7" \
  57         : "=&r" (l), "=&r" (x), "=&r" (y), "=r" ((a)) \
  58         : "3" ((a)), "r" ((b)), "r" ((t)), "r" ((v)) \
  59         : "cc" ); \
  60 }
  61
  62 static inline void XPROD31(ogg_int32_t  a, ogg_int32_t  b,
  63                            ogg_int32_t  t, ogg_int32_t  v,
  64                            ogg_int32_t *x, ogg_int32_t *y)
  65 {
  66   int x1, y1, l;
  67   asm(  "smull  %0, %1, %4, %6\n\t"
  68         "smlal  %0, %1, %5, %7\n\t"
  69         "rsb    %3, %4, #0\n\t"
  70         "smull  %0, %2, %5, %6\n\t"
  71         "smlal  %0, %2, %3, %7"
  72         : "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a)
  73         : "3" (a), "r" (b), "r" (t), "r" (v)
  74         : "cc" );
  75   *x = x1 << 1;
  76   MB();
  77   *y = y1 << 1;
  78 }
  79
  80 static inline void XNPROD31(ogg_int32_t  a, ogg_int32_t  b,
  81                             ogg_int32_t  t, ogg_int32_t  v,
  82                             ogg_int32_t *x, ogg_int32_t *y)
  83 {
  84   int x1, y1, l;
  85   asm(  "rsb    %2, %4, #0\n\t"
  86         "smull  %0, %1, %3, %5\n\t"
  87         "smlal  %0, %1, %2, %6\n\t"
  88         "smull  %0, %2, %4, %5\n\t"
  89         "smlal  %0, %2, %3, %6"
  90         : "=&r" (l), "=&r" (x1), "=&r" (y1)
  91         : "r" (a), "r" (b), "r" (t), "r" (v)
  92         : "cc" );
  93   *x = x1 << 1;
  94   MB();
  95   *y = y1 << 1;
  96 }
  97
  98 #ifndef _V_VECT_OPS
  99 #define _V_VECT_OPS
 100
 101 /* asm versions of vector operations for block.c, window.c */
 102 static inline
 103 void vect_add(ogg_int32_t *x, ogg_int32_t *y, int n)
 104 {
 105   while (n>=4) {
 106     asm volatile ("ldmia %[x], {r0, r1, r2, r3};"
 107                   "ldmia %[y]!, {r4, r5, r6, r7};"
 108                   "add r0, r0, r4;"
 109                   "add r1, r1, r5;"
 110                   "add r2, r2, r6;"
 111                   "add r3, r3, r7;"
 112                   "stmia %[x]!, {r0, r1, r2, r3};"
 113                   : [x] "+r" (x), [y] "+r" (y)
 114                   : : "r0", "r1", "r2", "r3",
 115                   "r4", "r5", "r6", "r7",
 116                   "memory");
 117     n -= 4;
 118   }
 119   /* add final elements */
 120   while (n>0) {
 121     *x++ += *y++;
 122     n--;
 123   }
 124 }
 125
 126 static inline
 127 void vect_copy(ogg_int32_t *x, ogg_int32_t *y, int n)
 128 {
 129   while (n>=4) {
 130     asm volatile ("ldmia %[y]!, {r0, r1, r2, r3};"
 131                   "stmia %[x]!, {r0, r1, r2, r3};"
 132                   : [x] "+r" (x), [y] "+r" (y)
 133                   : : "r0", "r1", "r2", "r3",
 134                   "memory");
 135     n -= 4;
 136   }
 137   /* copy final elements */
 138   while (n>0) {
 139     *x++ = *y++;
 140     n--;
 141   }
 142 }
 143
 144 static inline
 145 void vect_mult_fw(ogg_int32_t *data, LOOKUP_T *window, int n)
 146 {
 147   while (n>=4) {
 148     asm volatile ("ldmia %[d], {r0, r1, r2, r3};"
 149                   "ldmia %[w]!, {r4, r5, r6, r7};"
 150                   "smull r8, r9, r0, r4;"
 151                   "mov   r0, r9, lsl #1;"
 152                   "smull r8, r9, r1, r5;"
 153                   "mov   r1, r9, lsl #1;"
 154                   "smull r8, r9, r2, r6;"
 155                   "mov   r2, r9, lsl #1;"
 156                   "smull r8, r9, r3, r7;"
 157                   "mov   r3, r9, lsl #1;"
 158                   "stmia %[d]!, {r0, r1, r2, r3};"
 159                   : [d] "+r" (data), [w] "+r" (window)
 160                   : : "r0", "r1", "r2", "r3",
 161                   "r4", "r5", "r6", "r7", "r8", "r9",
 162                   "memory", "cc");
 163     n -= 4;
 164   }
 165   while(n>0) {
 166     *data = MULT31(*data, *window);
 167     data++;
 168     window++;
 169     n--;
 170   }
 171 }
 172
 173 static inline
 174 void vect_mult_bw(ogg_int32_t *data, LOOKUP_T *window, int n)
 175 {
 176   while (n>=4) {
 177     asm volatile ("ldmia %[d], {r0, r1, r2, r3};"
 178                   "ldmda %[w]!, {r4, r5, r6, r7};"
 179                   "smull r8, r9, r0, r7;"
 180                   "mov   r0, r9, lsl #1;"
 181                   "smull r8, r9, r1, r6;"
 182                   "mov   r1, r9, lsl #1;"
 183                   "smull r8, r9, r2, r5;"
 184                   "mov   r2, r9, lsl #1;"
 185                   "smull r8, r9, r3, r4;"
 186                   "mov   r3, r9, lsl #1;"
 187                   "stmia %[d]!, {r0, r1, r2, r3};"
 188                   : [d] "+r" (data), [w] "+r" (window)
 189                   : : "r0", "r1", "r2", "r3",
 190                   "r4", "r5", "r6", "r7", "r8", "r9",
 191                   "memory", "cc");
 192     n -= 4;
 193   }
 194   while(n>0) {
 195     *data = MULT31(*data, *window);
 196     data++;
 197     window--;
 198     n--;
 199   }
 200 }
 201
 202 #endif
 203
 204 #endif
 205
 206 #ifndef _V_CLIP_MATH
 207 #define _V_CLIP_MATH
 208
 209 static inline ogg_int32_t CLIP_TO_15(ogg_int32_t x) {
 210   int tmp;
 211   asm volatile("subs    %1, %0, #32768\n\t"
 212                "movpl   %0, #0x7f00\n\t"
 213                "orrpl   %0, %0, #0xff\n"
 214                "adds    %1, %0, #32768\n\t"
 215                "movmi   %0, #0x8000"
 216                : "+r"(x),"=r"(tmp)
 217                :
 218                : "cc");
 219   return(x);
 220 }
 221
 222 #endif
 223
 224 #ifndef _V_LSP_MATH_ASM
 225 #define _V_LSP_MATH_ASM
 226
 227 static inline void lsp_loop_asm(ogg_uint32_t *qip,ogg_uint32_t *pip,
 228                                 ogg_int32_t *qexpp,
 229                                 ogg_int32_t *ilsp,ogg_int32_t wi,
 230                                 ogg_int32_t m){
 231
 232   ogg_uint32_t qi=*qip,pi=*pip;
 233   ogg_int32_t qexp=*qexpp;
 234
 235   asm("mov     r0,%3;"
 236       "mov     r1,%5,asr#1;"
 237       "add     r0,r0,r1,lsl#3;"
 238       "1:"
 239
 240       "ldmdb   r0!,{r1,r3};"
 241       "subs    r1,r1,%4;"          //ilsp[j]-wi
 242       "rsbmi   r1,r1,#0;"          //labs(ilsp[j]-wi)
 243       "umull   %0,r2,r1,%0;"       //qi*=labs(ilsp[j]-wi)
 244
 245       "subs    r1,r3,%4;"          //ilsp[j+1]-wi
 246       "rsbmi   r1,r1,#0;"          //labs(ilsp[j+1]-wi)
 247       "umull   %1,r3,r1,%1;"       //pi*=labs(ilsp[j+1]-wi)
 248
 249       "cmn     r2,r3;"             // shift down 16?
 250       "beq     0f;"
 251       "add     %2,%2,#16;"
 252       "mov     %0,%0,lsr #16;"
 253       "orr     %0,%0,r2,lsl #16;"
 254       "mov     %1,%1,lsr #16;"
 255       "orr     %1,%1,r3,lsl #16;"
 256       "0:"
 257       "cmp     r0,%3;\n"
 258       "bhi     1b;\n"
 259
 260       // odd filter assymetry
 261       "ands    r0,%5,#1;\n"
 262       "beq     2f;\n"
 263       "add     r0,%3,%5,lsl#2;\n"
 264
 265       "ldr     r1,[r0,#-4];\n"
 266       "mov     r0,#0x4000;\n"
 267
 268       "subs    r1,r1,%4;\n"          //ilsp[j]-wi
 269       "rsbmi   r1,r1,#0;\n"          //labs(ilsp[j]-wi)
 270       "umull   %0,r2,r1,%0;\n"       //qi*=labs(ilsp[j]-wi)
 271       "umull   %1,r3,r0,%1;\n"       //pi*=labs(ilsp[j+1]-wi)
 272
 273       "cmn     r2,r3;\n"             // shift down 16?
 274       "beq     2f;\n"
 275       "add     %2,%2,#16;\n"
 276       "mov     %0,%0,lsr #16;\n"
 277       "orr     %0,%0,r2,lsl #16;\n"
 278       "mov     %1,%1,lsr #16;\n"
 279       "orr     %1,%1,r3,lsl #16;\n"
 280
 281       //qi=(pi>>shift)*labs(ilsp[j]-wi);
 282       //pi=(qi>>shift)*labs(ilsp[j+1]-wi);
 283       //qexp+=shift;
 284
 285       //}
 286
 287       /* normalize to max 16 sig figs */
 288       "2:"
 289       "mov     r2,#0;"
 290       "orr     r1,%0,%1;"
 291       "tst     r1,#0xff000000;"
 292       "addne   r2,r2,#8;"
 293       "movne   r1,r1,lsr #8;"
 294       "tst     r1,#0x00f00000;"
 295       "addne   r2,r2,#4;"
 296       "movne   r1,r1,lsr #4;"
 297       "tst     r1,#0x000c0000;"
 298       "addne   r2,r2,#2;"
 299       "movne   r1,r1,lsr #2;"
 300       "tst     r1,#0x00020000;"
 301       "addne   r2,r2,#1;"
 302       "movne   r1,r1,lsr #1;"
 303       "tst     r1,#0x00010000;"
 304       "addne   r2,r2,#1;"
 305       "mov     %0,%0,lsr r2;"
 306       "mov     %1,%1,lsr r2;"
 307       "add     %2,%2,r2;"
 308
 309       : "+r"(qi),"+r"(pi),"+r"(qexp)
 310       : "r"(ilsp),"r"(wi),"r"(m)
 311       : "r0","r1","r2","r3","cc");
 312
 313   *qip=qi;
 314   *pip=pi;
 315   *qexpp=qexp;
 316 }
 317
 318 static inline void lsp_norm_asm(ogg_uint32_t *qip,ogg_int32_t *qexpp){
 319
 320   ogg_uint32_t qi=*qip;
 321   ogg_int32_t qexp=*qexpp;
 322
 323   asm("tst     %0,#0x0000ff00;"
 324       "moveq   %0,%0,lsl #8;"
 325       "subeq   %1,%1,#8;"
 326       "tst     %0,#0x0000f000;"
 327       "moveq   %0,%0,lsl #4;"
 328       "subeq   %1,%1,#4;"
 329       "tst     %0,#0x0000c000;"
 330       "moveq   %0,%0,lsl #2;"
 331       "subeq   %1,%1,#2;"
 332       "tst     %0,#0x00008000;"
 333       "moveq   %0,%0,lsl #1;"
 334       "subeq   %1,%1,#1;"
 335       : "+r"(qi),"+r"(qexp)
 336       :
 337       : "cc");
 338   *qip=qi;
 339   *qexpp=qexp;
 340 }
 341
 342 #endif
 343 #endif
 344