arch/sparc/lib/rem.S

/* */
   1 /* rem.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  *  .rem        name of function to generate
  17  *  rem         rem=div => %o0 / %o1; rem=rem => %o0 % %o1
  18  *  true                true=true => signed; true=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 .rem
  45 .rem:
  46         ! compute sign of result; if neither is negative, no problem
  47         orcc    %o1, %o0, %g0   ! either negative?
  48         bge     2f                      ! no, go do the divide
  49         xor     %o1, %o0, %g6   ! compute sign in any case
  50         tst     %o1
  51         bge     1f
  52         tst     %o0
  53         ! %o1 is definitely negative; %o0 might also be negative
  54         bge     2f                      ! if %o0 not negative...
  55         sub     %g0, %o1, %o1   ! in any case, make %o1 nonneg
  56 1:      ! %o0 is negative, %o1 is nonnegative
  57         sub     %g0, %o0, %o0   ! make %o0 nonnegative
  58 2:
  59 
  60         ! Ready to divide.  Compute size of quotient; scale comparand.
  61         orcc    %o1, %g0, %o5
  62         bne     1f
  63         mov     %o0, %o3
  64 
  65                 ! Divide by zero trap.  If it returns, return 0 (about as
  66                 ! wrong as possible, but that is what SunOS does...).
  67                 ta      ST_DIV0
  68                 retl
  69                 clr     %o0
  70 
  71 1:
  72         cmp     %o3, %o5                        ! if %o1 exceeds %o0, done
  73         blu     Lgot_result             ! (and algorithm fails otherwise)
  74         clr     %o2
  75         sethi   %hi(1 << (32 - 4 - 1)), %g1
  76         cmp     %o3, %g1
  77         blu     Lnot_really_big
  78         clr     %o4
  79 
  80         ! Here the dividend is >= 2**(31-N) or so.  We must be careful here,
  81         ! as our usual N-at-a-shot divide step will cause overflow and havoc.
  82         ! The number of bits in the result here is N*ITER+SC, where SC <= N.
  83         ! Compute ITER in an unorthodox manner: know we need to shift V into
  84         ! the top decade: so do not even bother to compare to R.
  85         1:
  86                 cmp     %o5, %g1
  87                 bgeu    3f
  88                 mov     1, %g7
  89                 sll     %o5, 4, %o5
  90                 b       1b
  91                 add     %o4, 1, %o4
  92 
  93         ! Now compute %g7.
  94         2:      addcc   %o5, %o5, %o5
  95                 bcc     Lnot_too_big
  96                 add     %g7, 1, %g7
  97 
  98                 ! We get here if the %o1 overflowed while shifting.
  99                 ! This means that %o3 has the high-order bit set.
 100                 ! Restore %o5 and subtract from %o3.
 101                 sll     %g1, 4, %g1     ! high order bit
 102                 srl     %o5, 1, %o5             ! rest of %o5
 103                 add     %o5, %g1, %o5
 104                 b       Ldo_single_div
 105                 sub     %g7, 1, %g7
 106 
 107         Lnot_too_big:
 108         3:      cmp     %o5, %o3
 109                 blu     2b
 110                 nop
 111                 be      Ldo_single_div
 112                 nop
 113         /* NB: these are commented out in the V8-Sparc manual as well */
 114         /* (I do not understand this) */
 115         ! %o5 > %o3: went too far: back up 1 step
 116         !       srl     %o5, 1, %o5
 117         !       dec     %g7
 118         ! do single-bit divide steps
 119         !
 120         ! We have to be careful here.  We know that %o3 >= %o5, so we can do the
 121         ! first divide step without thinking.  BUT, the others are conditional,
 122         ! and are only done if %o3 >= 0.  Because both %o3 and %o5 may have the high-
 123         ! order bit set in the first step, just falling into the regular
 124         ! division loop will mess up the first time around.
 125         ! So we unroll slightly...
 126         Ldo_single_div:
 127                 subcc   %g7, 1, %g7
 128                 bl      Lend_regular_divide
 129                 nop
 130                 sub     %o3, %o5, %o3
 131                 mov     1, %o2
 132                 b       Lend_single_divloop
 133                 nop
 134         Lsingle_divloop:
 135                 sll     %o2, 1, %o2
 136                 bl      1f
 137                 srl     %o5, 1, %o5
 138                 ! %o3 >= 0
 139                 sub     %o3, %o5, %o3
 140                 b       2f
 141                 add     %o2, 1, %o2
 142         1:      ! %o3 < 0
 143                 add     %o3, %o5, %o3
 144                 sub     %o2, 1, %o2
 145         2:
 146         Lend_single_divloop:
 147                 subcc   %g7, 1, %g7
 148                 bge     Lsingle_divloop
 149                 tst     %o3
 150                 b,a     Lend_regular_divide
 151 
 152 Lnot_really_big:
 153 1:
 154         sll     %o5, 4, %o5
 155         cmp     %o5, %o3
 156         bleu    1b
 157         addcc   %o4, 1, %o4
 158         be      Lgot_result
 159         sub     %o4, 1, %o4
 160 
 161         tst     %o3     ! set up for initial iteration
 162 Ldivloop:
 163         sll     %o2, 4, %o2
 164                 ! depth 1, accumulated bits 0
 165         bl      L.1.16
 166         srl     %o5,1,%o5
 167         ! remainder is positive
 168         subcc   %o3,%o5,%o3
 169                         ! depth 2, accumulated bits 1
 170         bl      L.2.17
 171         srl     %o5,1,%o5
 172         ! remainder is positive
 173         subcc   %o3,%o5,%o3
 174                         ! depth 3, accumulated bits 3
 175         bl      L.3.19
 176         srl     %o5,1,%o5
 177         ! remainder is positive
 178         subcc   %o3,%o5,%o3
 179                         ! depth 4, accumulated bits 7
 180         bl      L.4.23
 181         srl     %o5,1,%o5
 182         ! remainder is positive
 183         subcc   %o3,%o5,%o3
 184                 b       9f
 185                 add     %o2, (7*2+1), %o2
 186         
 187 L.4.23:
 188         ! remainder is negative
 189         addcc   %o3,%o5,%o3
 190                 b       9f
 191                 add     %o2, (7*2-1), %o2
 192         
 193         
 194 L.3.19:
 195         ! remainder is negative
 196         addcc   %o3,%o5,%o3
 197                         ! depth 4, accumulated bits 5
 198         bl      L.4.21
 199         srl     %o5,1,%o5
 200         ! remainder is positive
 201         subcc   %o3,%o5,%o3
 202                 b       9f
 203                 add     %o2, (5*2+1), %o2
 204         
 205 L.4.21:
 206         ! remainder is negative
 207         addcc   %o3,%o5,%o3
 208                 b       9f
 209                 add     %o2, (5*2-1), %o2
 210         
 211         
 212         
 213 L.2.17:
 214         ! remainder is negative
 215         addcc   %o3,%o5,%o3
 216                         ! depth 3, accumulated bits 1
 217         bl      L.3.17
 218         srl     %o5,1,%o5
 219         ! remainder is positive
 220         subcc   %o3,%o5,%o3
 221                         ! depth 4, accumulated bits 3
 222         bl      L.4.19
 223         srl     %o5,1,%o5
 224         ! remainder is positive
 225         subcc   %o3,%o5,%o3
 226                 b       9f
 227                 add     %o2, (3*2+1), %o2
 228         
 229 L.4.19:
 230         ! remainder is negative
 231         addcc   %o3,%o5,%o3
 232                 b       9f
 233                 add     %o2, (3*2-1), %o2
 234         
 235         
 236 L.3.17:
 237         ! remainder is negative
 238         addcc   %o3,%o5,%o3
 239                         ! depth 4, accumulated bits 1
 240         bl      L.4.17
 241         srl     %o5,1,%o5
 242         ! remainder is positive
 243         subcc   %o3,%o5,%o3
 244                 b       9f
 245                 add     %o2, (1*2+1), %o2
 246         
 247 L.4.17:
 248         ! remainder is negative
 249         addcc   %o3,%o5,%o3
 250                 b       9f
 251                 add     %o2, (1*2-1), %o2
 252         
 253         
 254         
 255         
 256 L.1.16:
 257         ! remainder is negative
 258         addcc   %o3,%o5,%o3
 259                         ! depth 2, accumulated bits -1
 260         bl      L.2.15
 261         srl     %o5,1,%o5
 262         ! remainder is positive
 263         subcc   %o3,%o5,%o3
 264                         ! depth 3, accumulated bits -1
 265         bl      L.3.15
 266         srl     %o5,1,%o5
 267         ! remainder is positive
 268         subcc   %o3,%o5,%o3
 269                         ! depth 4, accumulated bits -1
 270         bl      L.4.15
 271         srl     %o5,1,%o5
 272         ! remainder is positive
 273         subcc   %o3,%o5,%o3
 274                 b       9f
 275                 add     %o2, (-1*2+1), %o2
 276         
 277 L.4.15:
 278         ! remainder is negative
 279         addcc   %o3,%o5,%o3
 280                 b       9f
 281                 add     %o2, (-1*2-1), %o2
 282         
 283         
 284 L.3.15:
 285         ! remainder is negative
 286         addcc   %o3,%o5,%o3
 287                         ! depth 4, accumulated bits -3
 288         bl      L.4.13
 289         srl     %o5,1,%o5
 290         ! remainder is positive
 291         subcc   %o3,%o5,%o3
 292                 b       9f
 293                 add     %o2, (-3*2+1), %o2
 294         
 295 L.4.13:
 296         ! remainder is negative
 297         addcc   %o3,%o5,%o3
 298                 b       9f
 299                 add     %o2, (-3*2-1), %o2
 300         
 301         
 302         
 303 L.2.15:
 304         ! remainder is negative
 305         addcc   %o3,%o5,%o3
 306                         ! depth 3, accumulated bits -3
 307         bl      L.3.13
 308         srl     %o5,1,%o5
 309         ! remainder is positive
 310         subcc   %o3,%o5,%o3
 311                         ! depth 4, accumulated bits -5
 312         bl      L.4.11
 313         srl     %o5,1,%o5
 314         ! remainder is positive
 315         subcc   %o3,%o5,%o3
 316                 b       9f
 317                 add     %o2, (-5*2+1), %o2
 318         
 319 L.4.11:
 320         ! remainder is negative
 321         addcc   %o3,%o5,%o3
 322                 b       9f
 323                 add     %o2, (-5*2-1), %o2
 324         
 325         
 326 L.3.13:
 327         ! remainder is negative
 328         addcc   %o3,%o5,%o3
 329                         ! depth 4, accumulated bits -7
 330         bl      L.4.9
 331         srl     %o5,1,%o5
 332         ! remainder is positive
 333         subcc   %o3,%o5,%o3
 334                 b       9f
 335                 add     %o2, (-7*2+1), %o2
 336         
 337 L.4.9:
 338         ! remainder is negative
 339         addcc   %o3,%o5,%o3
 340                 b       9f
 341                 add     %o2, (-7*2-1), %o2
 342         
 343         
 344         
 345         
 346         9:
 347 Lend_regular_divide:
 348         subcc   %o4, 1, %o4
 349         bge     Ldivloop
 350         tst     %o3
 351         bl,a    Lgot_result
 352         ! non-restoring fixup here (one instruction only!)
 353         add     %o3, %o1, %o3
 354 
 355 
 356 Lgot_result:
 357 
 358         retl
 359         mov %o3, %o0
/* */
root/arch/sparc/lib/rem.S
