arch/sparc/lib/rem.S

/* */
   1 /* $Id: rem.S,v 1.2 1995/11/25 00:59:02 davem Exp $
   2  * rem.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  *  .rem        name of function to generate
  18  *  rem         rem=div => %o0 / %o1; rem=rem => %o0 % %o1
  19  *  true                true=true => signed; true=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 .rem
  46 .rem:
  47         ! compute sign of result; if neither is negative, no problem
  48         orcc    %o1, %o0, %g0   ! either negative?
  49         bge     2f                      ! no, go do the divide
  50         xor     %o1, %o0, %g6   ! compute sign in any case
  51         tst     %o1
  52         bge     1f
  53         tst     %o0
  54         ! %o1 is definitely negative; %o0 might also be negative
  55         bge     2f                      ! if %o0 not negative...
  56         sub     %g0, %o1, %o1   ! in any case, make %o1 nonneg
  57 1:      ! %o0 is negative, %o1 is nonnegative
  58         sub     %g0, %o0, %o0   ! make %o0 nonnegative
  59 2:
  60 
  61         ! Ready to divide.  Compute size of quotient; scale comparand.
  62         orcc    %o1, %g0, %o5
  63         bne     1f
  64         mov     %o0, %o3
  65 
  66                 ! Divide by zero trap.  If it returns, return 0 (about as
  67                 ! wrong as possible, but that is what SunOS does...).
  68                 ta      ST_DIV0
  69                 retl
  70                 clr     %o0
  71 
  72 1:
  73         cmp     %o3, %o5                        ! if %o1 exceeds %o0, done
  74         blu     Lgot_result             ! (and algorithm fails otherwise)
  75         clr     %o2
  76         sethi   %hi(1 << (32 - 4 - 1)), %g1
  77         cmp     %o3, %g1
  78         blu     Lnot_really_big
  79         clr     %o4
  80 
  81         ! Here the dividend is >= 2**(31-N) or so.  We must be careful here,
  82         ! as our usual N-at-a-shot divide step will cause overflow and havoc.
  83         ! The number of bits in the result here is N*ITER+SC, where SC <= N.
  84         ! Compute ITER in an unorthodox manner: know we need to shift V into
  85         ! the top decade: so do not even bother to compare to R.
  86         1:
  87                 cmp     %o5, %g1
  88                 bgeu    3f
  89                 mov     1, %g7
  90                 sll     %o5, 4, %o5
  91                 b       1b
  92                 add     %o4, 1, %o4
  93 
  94         ! Now compute %g7.
  95         2:      addcc   %o5, %o5, %o5
  96                 bcc     Lnot_too_big
  97                 add     %g7, 1, %g7
  98 
  99                 ! We get here if the %o1 overflowed while shifting.
 100                 ! This means that %o3 has the high-order bit set.
 101                 ! Restore %o5 and subtract from %o3.
 102                 sll     %g1, 4, %g1     ! high order bit
 103                 srl     %o5, 1, %o5             ! rest of %o5
 104                 add     %o5, %g1, %o5
 105                 b       Ldo_single_div
 106                 sub     %g7, 1, %g7
 107 
 108         Lnot_too_big:
 109         3:      cmp     %o5, %o3
 110                 blu     2b
 111                 nop
 112                 be      Ldo_single_div
 113                 nop
 114         /* NB: these are commented out in the V8-Sparc manual as well */
 115         /* (I do not understand this) */
 116         ! %o5 > %o3: went too far: back up 1 step
 117         !       srl     %o5, 1, %o5
 118         !       dec     %g7
 119         ! do single-bit divide steps
 120         !
 121         ! We have to be careful here.  We know that %o3 >= %o5, so we can do the
 122         ! first divide step without thinking.  BUT, the others are conditional,
 123         ! and are only done if %o3 >= 0.  Because both %o3 and %o5 may have the high-
 124         ! order bit set in the first step, just falling into the regular
 125         ! division loop will mess up the first time around.
 126         ! So we unroll slightly...
 127         Ldo_single_div:
 128                 subcc   %g7, 1, %g7
 129                 bl      Lend_regular_divide
 130                 nop
 131                 sub     %o3, %o5, %o3
 132                 mov     1, %o2
 133                 b       Lend_single_divloop
 134                 nop
 135         Lsingle_divloop:
 136                 sll     %o2, 1, %o2
 137                 bl      1f
 138                 srl     %o5, 1, %o5
 139                 ! %o3 >= 0
 140                 sub     %o3, %o5, %o3
 141                 b       2f
 142                 add     %o2, 1, %o2
 143         1:      ! %o3 < 0
 144                 add     %o3, %o5, %o3
 145                 sub     %o2, 1, %o2
 146         2:
 147         Lend_single_divloop:
 148                 subcc   %g7, 1, %g7
 149                 bge     Lsingle_divloop
 150                 tst     %o3
 151                 b,a     Lend_regular_divide
 152 
 153 Lnot_really_big:
 154 1:
 155         sll     %o5, 4, %o5
 156         cmp     %o5, %o3
 157         bleu    1b
 158         addcc   %o4, 1, %o4
 159         be      Lgot_result
 160         sub     %o4, 1, %o4
 161 
 162         tst     %o3     ! set up for initial iteration
 163 Ldivloop:
 164         sll     %o2, 4, %o2
 165                 ! depth 1, accumulated bits 0
 166         bl      L.1.16
 167         srl     %o5,1,%o5
 168         ! remainder is positive
 169         subcc   %o3,%o5,%o3
 170                         ! depth 2, accumulated bits 1
 171         bl      L.2.17
 172         srl     %o5,1,%o5
 173         ! remainder is positive
 174         subcc   %o3,%o5,%o3
 175                         ! depth 3, accumulated bits 3
 176         bl      L.3.19
 177         srl     %o5,1,%o5
 178         ! remainder is positive
 179         subcc   %o3,%o5,%o3
 180                         ! depth 4, accumulated bits 7
 181         bl      L.4.23
 182         srl     %o5,1,%o5
 183         ! remainder is positive
 184         subcc   %o3,%o5,%o3
 185                 b       9f
 186                 add     %o2, (7*2+1), %o2
 187         
 188 L.4.23:
 189         ! remainder is negative
 190         addcc   %o3,%o5,%o3
 191                 b       9f
 192                 add     %o2, (7*2-1), %o2
 193         
 194         
 195 L.3.19:
 196         ! remainder is negative
 197         addcc   %o3,%o5,%o3
 198                         ! depth 4, accumulated bits 5
 199         bl      L.4.21
 200         srl     %o5,1,%o5
 201         ! remainder is positive
 202         subcc   %o3,%o5,%o3
 203                 b       9f
 204                 add     %o2, (5*2+1), %o2
 205         
 206 L.4.21:
 207         ! remainder is negative
 208         addcc   %o3,%o5,%o3
 209                 b       9f
 210                 add     %o2, (5*2-1), %o2
 211         
 212         
 213         
 214 L.2.17:
 215         ! remainder is negative
 216         addcc   %o3,%o5,%o3
 217                         ! depth 3, accumulated bits 1
 218         bl      L.3.17
 219         srl     %o5,1,%o5
 220         ! remainder is positive
 221         subcc   %o3,%o5,%o3
 222                         ! depth 4, accumulated bits 3
 223         bl      L.4.19
 224         srl     %o5,1,%o5
 225         ! remainder is positive
 226         subcc   %o3,%o5,%o3
 227                 b       9f
 228                 add     %o2, (3*2+1), %o2
 229         
 230 L.4.19:
 231         ! remainder is negative
 232         addcc   %o3,%o5,%o3
 233                 b       9f
 234                 add     %o2, (3*2-1), %o2
 235         
 236         
 237 L.3.17:
 238         ! remainder is negative
 239         addcc   %o3,%o5,%o3
 240                         ! depth 4, accumulated bits 1
 241         bl      L.4.17
 242         srl     %o5,1,%o5
 243         ! remainder is positive
 244         subcc   %o3,%o5,%o3
 245                 b       9f
 246                 add     %o2, (1*2+1), %o2
 247         
 248 L.4.17:
 249         ! remainder is negative
 250         addcc   %o3,%o5,%o3
 251                 b       9f
 252                 add     %o2, (1*2-1), %o2
 253         
 254         
 255         
 256         
 257 L.1.16:
 258         ! remainder is negative
 259         addcc   %o3,%o5,%o3
 260                         ! depth 2, accumulated bits -1
 261         bl      L.2.15
 262         srl     %o5,1,%o5
 263         ! remainder is positive
 264         subcc   %o3,%o5,%o3
 265                         ! depth 3, accumulated bits -1
 266         bl      L.3.15
 267         srl     %o5,1,%o5
 268         ! remainder is positive
 269         subcc   %o3,%o5,%o3
 270                         ! depth 4, accumulated bits -1
 271         bl      L.4.15
 272         srl     %o5,1,%o5
 273         ! remainder is positive
 274         subcc   %o3,%o5,%o3
 275                 b       9f
 276                 add     %o2, (-1*2+1), %o2
 277         
 278 L.4.15:
 279         ! remainder is negative
 280         addcc   %o3,%o5,%o3
 281                 b       9f
 282                 add     %o2, (-1*2-1), %o2
 283         
 284         
 285 L.3.15:
 286         ! remainder is negative
 287         addcc   %o3,%o5,%o3
 288                         ! depth 4, accumulated bits -3
 289         bl      L.4.13
 290         srl     %o5,1,%o5
 291         ! remainder is positive
 292         subcc   %o3,%o5,%o3
 293                 b       9f
 294                 add     %o2, (-3*2+1), %o2
 295         
 296 L.4.13:
 297         ! remainder is negative
 298         addcc   %o3,%o5,%o3
 299                 b       9f
 300                 add     %o2, (-3*2-1), %o2
 301         
 302         
 303         
 304 L.2.15:
 305         ! remainder is negative
 306         addcc   %o3,%o5,%o3
 307                         ! depth 3, accumulated bits -3
 308         bl      L.3.13
 309         srl     %o5,1,%o5
 310         ! remainder is positive
 311         subcc   %o3,%o5,%o3
 312                         ! depth 4, accumulated bits -5
 313         bl      L.4.11
 314         srl     %o5,1,%o5
 315         ! remainder is positive
 316         subcc   %o3,%o5,%o3
 317                 b       9f
 318                 add     %o2, (-5*2+1), %o2
 319         
 320 L.4.11:
 321         ! remainder is negative
 322         addcc   %o3,%o5,%o3
 323                 b       9f
 324                 add     %o2, (-5*2-1), %o2
 325         
 326         
 327 L.3.13:
 328         ! remainder is negative
 329         addcc   %o3,%o5,%o3
 330                         ! depth 4, accumulated bits -7
 331         bl      L.4.9
 332         srl     %o5,1,%o5
 333         ! remainder is positive
 334         subcc   %o3,%o5,%o3
 335                 b       9f
 336                 add     %o2, (-7*2+1), %o2
 337         
 338 L.4.9:
 339         ! remainder is negative
 340         addcc   %o3,%o5,%o3
 341                 b       9f
 342                 add     %o2, (-7*2-1), %o2
 343         
 344         
 345         
 346         
 347         9:
 348 Lend_regular_divide:
 349         subcc   %o4, 1, %o4
 350         bge     Ldivloop
 351         tst     %o3
 352         bl,a    Lgot_result
 353         ! non-restoring fixup here (one instruction only!)
 354         add     %o3, %o1, %o3
 355 
 356 
 357 Lgot_result:
 358 
 359         retl
 360         mov %o3, %o0
/* */
root/arch/sparc/lib/rem.S
