root/arch/sparc/lib/udiv.S

/* [previous][next][first][last][top][bottom][index][help] */ 
   1 /* udiv.S:      This routine was taken from glibc-1.09 and is covered
   2  *              by the GNU Library General Public License Version 2.
   3  */
   4 
   5 
   6 /* This file is generated from divrem.m4; DO NOT EDIT! */
   7 /*
   8  * Division and remainder, from Appendix E of the Sparc Version 8
   9  * Architecture Manual, with fixes from Gordon Irlam.
  10  */
  11 
  12 /*
  13  * Input: dividend and divisor in %o0 and %o1 respectively.
  14  *
  15  * m4 parameters:
  16  *  .udiv       name of function to generate
  17  *  div         div=div => %o0 / %o1; div=rem => %o0 % %o1
  18  *  false               false=true => signed; false=false => unsigned
  19  *
  20  * Algorithm parameters:
  21  *  N           how many bits per iteration we try to get (4)
  22  *  WORDSIZE    total number of bits (32)
  23  *
  24  * Derived constants:
  25  *  TOPBITS     number of bits in the top decade of a number
  26  *
  27  * Important variables:
  28  *  Q           the partial quotient under development (initially 0)
  29  *  R           the remainder so far, initially the dividend
  30  *  ITER        number of main division loop iterations required;
  31  *              equal to ceil(log2(quotient) / N).  Note that this
  32  *              is the log base (2^N) of the quotient.
  33  *  V           the current comparand, initially divisor*2^(ITER*N-1)
  34  *
  35  * Cost:
  36  *  Current estimate for non-large dividend is
  37  *      ceil(log2(quotient) / N) * (10 + 7N/2) + C
  38  *  A large dividend is one greater than 2^(31-TOPBITS) and takes a
  39  *  different path, as the upper bits of the quotient must be developed
  40  *  one bit at a time.
  41  */
  42 
  43 
  44         .globl .udiv
  45 .udiv:
  46 
  47         ! Ready to divide.  Compute size of quotient; scale comparand.
  48         orcc    %o1, %g0, %o5
  49         bne     1f
  50         mov     %o0, %o3
  51 
  52                 ! Divide by zero trap.  If it returns, return 0 (about as
  53                 ! wrong as possible, but that is what SunOS does...).
  54                 ta      ST_DIV0
  55                 retl
  56                 clr     %o0
  57 
  58 1:
  59         cmp     %o3, %o5                        ! if %o1 exceeds %o0, done
  60         blu     Lgot_result             ! (and algorithm fails otherwise)
  61         clr     %o2
  62         sethi   %hi(1 << (32 - 4 - 1)), %g1
  63         cmp     %o3, %g1
  64         blu     Lnot_really_big
  65         clr     %o4
  66 
  67         ! Here the dividend is >= 2**(31-N) or so.  We must be careful here,
  68         ! as our usual N-at-a-shot divide step will cause overflow and havoc.
  69         ! The number of bits in the result here is N*ITER+SC, where SC <= N.
  70         ! Compute ITER in an unorthodox manner: know we need to shift V into
  71         ! the top decade: so do not even bother to compare to R.
  72         1:
  73                 cmp     %o5, %g1
  74                 bgeu    3f
  75                 mov     1, %g7
  76                 sll     %o5, 4, %o5
  77                 b       1b
  78                 add     %o4, 1, %o4
  79 
  80         ! Now compute %g7.
  81         2:      addcc   %o5, %o5, %o5
  82                 bcc     Lnot_too_big
  83                 add     %g7, 1, %g7
  84 
  85                 ! We get here if the %o1 overflowed while shifting.
  86                 ! This means that %o3 has the high-order bit set.
  87                 ! Restore %o5 and subtract from %o3.
  88                 sll     %g1, 4, %g1     ! high order bit
  89                 srl     %o5, 1, %o5             ! rest of %o5
  90                 add     %o5, %g1, %o5
  91                 b       Ldo_single_div
  92                 sub     %g7, 1, %g7
  93 
  94         Lnot_too_big:
  95         3:      cmp     %o5, %o3
  96                 blu     2b
  97                 nop
  98                 be      Ldo_single_div
  99                 nop
 100         /* NB: these are commented out in the V8-Sparc manual as well */
 101         /* (I do not understand this) */
 102         ! %o5 > %o3: went too far: back up 1 step
 103         !       srl     %o5, 1, %o5
 104         !       dec     %g7
 105         ! do single-bit divide steps
 106         !
 107         ! We have to be careful here.  We know that %o3 >= %o5, so we can do the
 108         ! first divide step without thinking.  BUT, the others are conditional,
 109         ! and are only done if %o3 >= 0.  Because both %o3 and %o5 may have the high-
 110         ! order bit set in the first step, just falling into the regular
 111         ! division loop will mess up the first time around.
 112         ! So we unroll slightly...
 113         Ldo_single_div:
 114                 subcc   %g7, 1, %g7
 115                 bl      Lend_regular_divide
 116                 nop
 117                 sub     %o3, %o5, %o3
 118                 mov     1, %o2
 119                 b       Lend_single_divloop
 120                 nop
 121         Lsingle_divloop:
 122                 sll     %o2, 1, %o2
 123                 bl      1f
 124                 srl     %o5, 1, %o5
 125                 ! %o3 >= 0
 126                 sub     %o3, %o5, %o3
 127                 b       2f
 128                 add     %o2, 1, %o2
 129         1:      ! %o3 < 0
 130                 add     %o3, %o5, %o3
 131                 sub     %o2, 1, %o2
 132         2:
 133         Lend_single_divloop:
 134                 subcc   %g7, 1, %g7
 135                 bge     Lsingle_divloop
 136                 tst     %o3
 137                 b,a     Lend_regular_divide
 138 
 139 Lnot_really_big:
 140 1:
 141         sll     %o5, 4, %o5
 142         cmp     %o5, %o3
 143         bleu    1b
 144         addcc   %o4, 1, %o4
 145         be      Lgot_result
 146         sub     %o4, 1, %o4
 147 
 148         tst     %o3     ! set up for initial iteration
 149 Ldivloop:
 150         sll     %o2, 4, %o2
 151                 ! depth 1, accumulated bits 0
 152         bl      L.1.16
 153         srl     %o5,1,%o5
 154         ! remainder is positive
 155         subcc   %o3,%o5,%o3
 156                         ! depth 2, accumulated bits 1
 157         bl      L.2.17
 158         srl     %o5,1,%o5
 159         ! remainder is positive
 160         subcc   %o3,%o5,%o3
 161                         ! depth 3, accumulated bits 3
 162         bl      L.3.19
 163         srl     %o5,1,%o5
 164         ! remainder is positive
 165         subcc   %o3,%o5,%o3
 166                         ! depth 4, accumulated bits 7
 167         bl      L.4.23
 168         srl     %o5,1,%o5
 169         ! remainder is positive
 170         subcc   %o3,%o5,%o3
 171                 b       9f
 172                 add     %o2, (7*2+1), %o2
 173         
 174 L.4.23:
 175         ! remainder is negative
 176         addcc   %o3,%o5,%o3
 177                 b       9f
 178                 add     %o2, (7*2-1), %o2
 179         
 180         
 181 L.3.19:
 182         ! remainder is negative
 183         addcc   %o3,%o5,%o3
 184                         ! depth 4, accumulated bits 5
 185         bl      L.4.21
 186         srl     %o5,1,%o5
 187         ! remainder is positive
 188         subcc   %o3,%o5,%o3
 189                 b       9f
 190                 add     %o2, (5*2+1), %o2
 191         
 192 L.4.21:
 193         ! remainder is negative
 194         addcc   %o3,%o5,%o3
 195                 b       9f
 196                 add     %o2, (5*2-1), %o2
 197         
 198         
 199         
 200 L.2.17:
 201         ! remainder is negative
 202         addcc   %o3,%o5,%o3
 203                         ! depth 3, accumulated bits 1
 204         bl      L.3.17
 205         srl     %o5,1,%o5
 206         ! remainder is positive
 207         subcc   %o3,%o5,%o3
 208                         ! depth 4, accumulated bits 3
 209         bl      L.4.19
 210         srl     %o5,1,%o5
 211         ! remainder is positive
 212         subcc   %o3,%o5,%o3
 213                 b       9f
 214                 add     %o2, (3*2+1), %o2
 215         
 216 L.4.19:
 217         ! remainder is negative
 218         addcc   %o3,%o5,%o3
 219                 b       9f
 220                 add     %o2, (3*2-1), %o2
 221         
 222         
 223 L.3.17:
 224         ! remainder is negative
 225         addcc   %o3,%o5,%o3
 226                         ! depth 4, accumulated bits 1
 227         bl      L.4.17
 228         srl     %o5,1,%o5
 229         ! remainder is positive
 230         subcc   %o3,%o5,%o3
 231                 b       9f
 232                 add     %o2, (1*2+1), %o2
 233         
 234 L.4.17:
 235         ! remainder is negative
 236         addcc   %o3,%o5,%o3
 237                 b       9f
 238                 add     %o2, (1*2-1), %o2
 239         
 240         
 241         
 242         
 243 L.1.16:
 244         ! remainder is negative
 245         addcc   %o3,%o5,%o3
 246                         ! depth 2, accumulated bits -1
 247         bl      L.2.15
 248         srl     %o5,1,%o5
 249         ! remainder is positive
 250         subcc   %o3,%o5,%o3
 251                         ! depth 3, accumulated bits -1
 252         bl      L.3.15
 253         srl     %o5,1,%o5
 254         ! remainder is positive
 255         subcc   %o3,%o5,%o3
 256                         ! depth 4, accumulated bits -1
 257         bl      L.4.15
 258         srl     %o5,1,%o5
 259         ! remainder is positive
 260         subcc   %o3,%o5,%o3
 261                 b       9f
 262                 add     %o2, (-1*2+1), %o2
 263         
 264 L.4.15:
 265         ! remainder is negative
 266         addcc   %o3,%o5,%o3
 267                 b       9f
 268                 add     %o2, (-1*2-1), %o2
 269         
 270         
 271 L.3.15:
 272         ! remainder is negative
 273         addcc   %o3,%o5,%o3
 274                         ! depth 4, accumulated bits -3
 275         bl      L.4.13
 276         srl     %o5,1,%o5
 277         ! remainder is positive
 278         subcc   %o3,%o5,%o3
 279                 b       9f
 280                 add     %o2, (-3*2+1), %o2
 281         
 282 L.4.13:
 283         ! remainder is negative
 284         addcc   %o3,%o5,%o3
 285                 b       9f
 286                 add     %o2, (-3*2-1), %o2
 287         
 288         
 289         
 290 L.2.15:
 291         ! remainder is negative
 292         addcc   %o3,%o5,%o3
 293                         ! depth 3, accumulated bits -3
 294         bl      L.3.13
 295         srl     %o5,1,%o5
 296         ! remainder is positive
 297         subcc   %o3,%o5,%o3
 298                         ! depth 4, accumulated bits -5
 299         bl      L.4.11
 300         srl     %o5,1,%o5
 301         ! remainder is positive
 302         subcc   %o3,%o5,%o3
 303                 b       9f
 304                 add     %o2, (-5*2+1), %o2
 305         
 306 L.4.11:
 307         ! remainder is negative
 308         addcc   %o3,%o5,%o3
 309                 b       9f
 310                 add     %o2, (-5*2-1), %o2
 311         
 312         
 313 L.3.13:
 314         ! remainder is negative
 315         addcc   %o3,%o5,%o3
 316                         ! depth 4, accumulated bits -7
 317         bl      L.4.9
 318         srl     %o5,1,%o5
 319         ! remainder is positive
 320         subcc   %o3,%o5,%o3
 321                 b       9f
 322                 add     %o2, (-7*2+1), %o2
 323         
 324 L.4.9:
 325         ! remainder is negative
 326         addcc   %o3,%o5,%o3
 327                 b       9f
 328                 add     %o2, (-7*2-1), %o2
 329         
 330         
 331         
 332         
 333         9:
 334 Lend_regular_divide:
 335         subcc   %o4, 1, %o4
 336         bge     Ldivloop
 337         tst     %o3
 338         bl,a    Lgot_result
 339         ! non-restoring fixup here (one instruction only!)
 340         sub     %o2, 1, %o2
 341 
 342 
 343 Lgot_result:
 344 
 345         retl
 346         mov %o2, %o0
/* [previous][next][first][last][top][bottom][index][help] */