arch/sparc/lib/udiv.S

/* */
   1 /* $Id: udiv.S,v 1.2 1995/11/25 00:59:06 davem Exp $
   2  * udiv.S:      This routine was taken from glibc-1.09 and is covered
   3  *              by the GNU Library General Public License Version 2.
   4  */
   5 
   6 
   7 /* This file is generated from divrem.m4; DO NOT EDIT! */
   8 /*
   9  * Division and remainder, from Appendix E of the Sparc Version 8
  10  * Architecture Manual, with fixes from Gordon Irlam.
  11  */
  12 
  13 /*
  14  * Input: dividend and divisor in %o0 and %o1 respectively.
  15  *
  16  * m4 parameters:
  17  *  .udiv       name of function to generate
  18  *  div         div=div => %o0 / %o1; div=rem => %o0 % %o1
  19  *  false               false=true => signed; false=false => unsigned
  20  *
  21  * Algorithm parameters:
  22  *  N           how many bits per iteration we try to get (4)
  23  *  WORDSIZE    total number of bits (32)
  24  *
  25  * Derived constants:
  26  *  TOPBITS     number of bits in the top decade of a number
  27  *
  28  * Important variables:
  29  *  Q           the partial quotient under development (initially 0)
  30  *  R           the remainder so far, initially the dividend
  31  *  ITER        number of main division loop iterations required;
  32  *              equal to ceil(log2(quotient) / N).  Note that this
  33  *              is the log base (2^N) of the quotient.
  34  *  V           the current comparand, initially divisor*2^(ITER*N-1)
  35  *
  36  * Cost:
  37  *  Current estimate for non-large dividend is
  38  *      ceil(log2(quotient) / N) * (10 + 7N/2) + C
  39  *  A large dividend is one greater than 2^(31-TOPBITS) and takes a
  40  *  different path, as the upper bits of the quotient must be developed
  41  *  one bit at a time.
  42  */
  43 
  44 
  45         .globl .udiv
  46 .udiv:
  47 
  48         ! Ready to divide.  Compute size of quotient; scale comparand.
  49         orcc    %o1, %g0, %o5
  50         bne     1f
  51         mov     %o0, %o3
  52 
  53                 ! Divide by zero trap.  If it returns, return 0 (about as
  54                 ! wrong as possible, but that is what SunOS does...).
  55                 ta      ST_DIV0
  56                 retl
  57                 clr     %o0
  58 
  59 1:
  60         cmp     %o3, %o5                        ! if %o1 exceeds %o0, done
  61         blu     Lgot_result             ! (and algorithm fails otherwise)
  62         clr     %o2
  63         sethi   %hi(1 << (32 - 4 - 1)), %g1
  64         cmp     %o3, %g1
  65         blu     Lnot_really_big
  66         clr     %o4
  67 
  68         ! Here the dividend is >= 2**(31-N) or so.  We must be careful here,
  69         ! as our usual N-at-a-shot divide step will cause overflow and havoc.
  70         ! The number of bits in the result here is N*ITER+SC, where SC <= N.
  71         ! Compute ITER in an unorthodox manner: know we need to shift V into
  72         ! the top decade: so do not even bother to compare to R.
  73         1:
  74                 cmp     %o5, %g1
  75                 bgeu    3f
  76                 mov     1, %g7
  77                 sll     %o5, 4, %o5
  78                 b       1b
  79                 add     %o4, 1, %o4
  80 
  81         ! Now compute %g7.
  82         2:      addcc   %o5, %o5, %o5
  83                 bcc     Lnot_too_big
  84                 add     %g7, 1, %g7
  85 
  86                 ! We get here if the %o1 overflowed while shifting.
  87                 ! This means that %o3 has the high-order bit set.
  88                 ! Restore %o5 and subtract from %o3.
  89                 sll     %g1, 4, %g1     ! high order bit
  90                 srl     %o5, 1, %o5             ! rest of %o5
  91                 add     %o5, %g1, %o5
  92                 b       Ldo_single_div
  93                 sub     %g7, 1, %g7
  94 
  95         Lnot_too_big:
  96         3:      cmp     %o5, %o3
  97                 blu     2b
  98                 nop
  99                 be      Ldo_single_div
 100                 nop
 101         /* NB: these are commented out in the V8-Sparc manual as well */
 102         /* (I do not understand this) */
 103         ! %o5 > %o3: went too far: back up 1 step
 104         !       srl     %o5, 1, %o5
 105         !       dec     %g7
 106         ! do single-bit divide steps
 107         !
 108         ! We have to be careful here.  We know that %o3 >= %o5, so we can do the
 109         ! first divide step without thinking.  BUT, the others are conditional,
 110         ! and are only done if %o3 >= 0.  Because both %o3 and %o5 may have the high-
 111         ! order bit set in the first step, just falling into the regular
 112         ! division loop will mess up the first time around.
 113         ! So we unroll slightly...
 114         Ldo_single_div:
 115                 subcc   %g7, 1, %g7
 116                 bl      Lend_regular_divide
 117                 nop
 118                 sub     %o3, %o5, %o3
 119                 mov     1, %o2
 120                 b       Lend_single_divloop
 121                 nop
 122         Lsingle_divloop:
 123                 sll     %o2, 1, %o2
 124                 bl      1f
 125                 srl     %o5, 1, %o5
 126                 ! %o3 >= 0
 127                 sub     %o3, %o5, %o3
 128                 b       2f
 129                 add     %o2, 1, %o2
 130         1:      ! %o3 < 0
 131                 add     %o3, %o5, %o3
 132                 sub     %o2, 1, %o2
 133         2:
 134         Lend_single_divloop:
 135                 subcc   %g7, 1, %g7
 136                 bge     Lsingle_divloop
 137                 tst     %o3
 138                 b,a     Lend_regular_divide
 139 
 140 Lnot_really_big:
 141 1:
 142         sll     %o5, 4, %o5
 143         cmp     %o5, %o3
 144         bleu    1b
 145         addcc   %o4, 1, %o4
 146         be      Lgot_result
 147         sub     %o4, 1, %o4
 148 
 149         tst     %o3     ! set up for initial iteration
 150 Ldivloop:
 151         sll     %o2, 4, %o2
 152                 ! depth 1, accumulated bits 0
 153         bl      L.1.16
 154         srl     %o5,1,%o5
 155         ! remainder is positive
 156         subcc   %o3,%o5,%o3
 157                         ! depth 2, accumulated bits 1
 158         bl      L.2.17
 159         srl     %o5,1,%o5
 160         ! remainder is positive
 161         subcc   %o3,%o5,%o3
 162                         ! depth 3, accumulated bits 3
 163         bl      L.3.19
 164         srl     %o5,1,%o5
 165         ! remainder is positive
 166         subcc   %o3,%o5,%o3
 167                         ! depth 4, accumulated bits 7
 168         bl      L.4.23
 169         srl     %o5,1,%o5
 170         ! remainder is positive
 171         subcc   %o3,%o5,%o3
 172                 b       9f
 173                 add     %o2, (7*2+1), %o2
 174         
 175 L.4.23:
 176         ! remainder is negative
 177         addcc   %o3,%o5,%o3
 178                 b       9f
 179                 add     %o2, (7*2-1), %o2
 180         
 181         
 182 L.3.19:
 183         ! remainder is negative
 184         addcc   %o3,%o5,%o3
 185                         ! depth 4, accumulated bits 5
 186         bl      L.4.21
 187         srl     %o5,1,%o5
 188         ! remainder is positive
 189         subcc   %o3,%o5,%o3
 190                 b       9f
 191                 add     %o2, (5*2+1), %o2
 192         
 193 L.4.21:
 194         ! remainder is negative
 195         addcc   %o3,%o5,%o3
 196                 b       9f
 197                 add     %o2, (5*2-1), %o2
 198         
 199         
 200         
 201 L.2.17:
 202         ! remainder is negative
 203         addcc   %o3,%o5,%o3
 204                         ! depth 3, accumulated bits 1
 205         bl      L.3.17
 206         srl     %o5,1,%o5
 207         ! remainder is positive
 208         subcc   %o3,%o5,%o3
 209                         ! depth 4, accumulated bits 3
 210         bl      L.4.19
 211         srl     %o5,1,%o5
 212         ! remainder is positive
 213         subcc   %o3,%o5,%o3
 214                 b       9f
 215                 add     %o2, (3*2+1), %o2
 216         
 217 L.4.19:
 218         ! remainder is negative
 219         addcc   %o3,%o5,%o3
 220                 b       9f
 221                 add     %o2, (3*2-1), %o2
 222         
 223         
 224 L.3.17:
 225         ! remainder is negative
 226         addcc   %o3,%o5,%o3
 227                         ! depth 4, accumulated bits 1
 228         bl      L.4.17
 229         srl     %o5,1,%o5
 230         ! remainder is positive
 231         subcc   %o3,%o5,%o3
 232                 b       9f
 233                 add     %o2, (1*2+1), %o2
 234         
 235 L.4.17:
 236         ! remainder is negative
 237         addcc   %o3,%o5,%o3
 238                 b       9f
 239                 add     %o2, (1*2-1), %o2
 240         
 241         
 242         
 243         
 244 L.1.16:
 245         ! remainder is negative
 246         addcc   %o3,%o5,%o3
 247                         ! depth 2, accumulated bits -1
 248         bl      L.2.15
 249         srl     %o5,1,%o5
 250         ! remainder is positive
 251         subcc   %o3,%o5,%o3
 252                         ! depth 3, accumulated bits -1
 253         bl      L.3.15
 254         srl     %o5,1,%o5
 255         ! remainder is positive
 256         subcc   %o3,%o5,%o3
 257                         ! depth 4, accumulated bits -1
 258         bl      L.4.15
 259         srl     %o5,1,%o5
 260         ! remainder is positive
 261         subcc   %o3,%o5,%o3
 262                 b       9f
 263                 add     %o2, (-1*2+1), %o2
 264         
 265 L.4.15:
 266         ! remainder is negative
 267         addcc   %o3,%o5,%o3
 268                 b       9f
 269                 add     %o2, (-1*2-1), %o2
 270         
 271         
 272 L.3.15:
 273         ! remainder is negative
 274         addcc   %o3,%o5,%o3
 275                         ! depth 4, accumulated bits -3
 276         bl      L.4.13
 277         srl     %o5,1,%o5
 278         ! remainder is positive
 279         subcc   %o3,%o5,%o3
 280                 b       9f
 281                 add     %o2, (-3*2+1), %o2
 282         
 283 L.4.13:
 284         ! remainder is negative
 285         addcc   %o3,%o5,%o3
 286                 b       9f
 287                 add     %o2, (-3*2-1), %o2
 288         
 289         
 290         
 291 L.2.15:
 292         ! remainder is negative
 293         addcc   %o3,%o5,%o3
 294                         ! depth 3, accumulated bits -3
 295         bl      L.3.13
 296         srl     %o5,1,%o5
 297         ! remainder is positive
 298         subcc   %o3,%o5,%o3
 299                         ! depth 4, accumulated bits -5
 300         bl      L.4.11
 301         srl     %o5,1,%o5
 302         ! remainder is positive
 303         subcc   %o3,%o5,%o3
 304                 b       9f
 305                 add     %o2, (-5*2+1), %o2
 306         
 307 L.4.11:
 308         ! remainder is negative
 309         addcc   %o3,%o5,%o3
 310                 b       9f
 311                 add     %o2, (-5*2-1), %o2
 312         
 313         
 314 L.3.13:
 315         ! remainder is negative
 316         addcc   %o3,%o5,%o3
 317                         ! depth 4, accumulated bits -7
 318         bl      L.4.9
 319         srl     %o5,1,%o5
 320         ! remainder is positive
 321         subcc   %o3,%o5,%o3
 322                 b       9f
 323                 add     %o2, (-7*2+1), %o2
 324         
 325 L.4.9:
 326         ! remainder is negative
 327         addcc   %o3,%o5,%o3
 328                 b       9f
 329                 add     %o2, (-7*2-1), %o2
 330         
 331         
 332         
 333         
 334         9:
 335 Lend_regular_divide:
 336         subcc   %o4, 1, %o4
 337         bge     Ldivloop
 338         tst     %o3
 339         bl,a    Lgot_result
 340         ! non-restoring fixup here (one instruction only!)
 341         sub     %o2, 1, %o2
 342 
 343 
 344 Lgot_result:
 345 
 346         retl
 347         mov %o2, %o0
/* */
root/arch/sparc/lib/udiv.S
