Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* $Id: udiv.S,v 1.4 1996/09/30 02:22:38 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 |
Al Viro | 7caaeab | 2005-09-11 20:14:07 -0700 | [diff] [blame^] | 46 | .globl _Udiv |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | .udiv: |
Al Viro | 7caaeab | 2005-09-11 20:14:07 -0700 | [diff] [blame^] | 48 | _Udiv: /* needed for export */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | |
| 50 | ! Ready to divide. Compute size of quotient; scale comparand. |
| 51 | orcc %o1, %g0, %o5 |
| 52 | bne 1f |
| 53 | mov %o0, %o3 |
| 54 | |
| 55 | ! Divide by zero trap. If it returns, return 0 (about as |
| 56 | ! wrong as possible, but that is what SunOS does...). |
| 57 | ta ST_DIV0 |
| 58 | retl |
| 59 | clr %o0 |
| 60 | |
| 61 | 1: |
| 62 | cmp %o3, %o5 ! if %o1 exceeds %o0, done |
| 63 | blu Lgot_result ! (and algorithm fails otherwise) |
| 64 | clr %o2 |
| 65 | |
| 66 | sethi %hi(1 << (32 - 4 - 1)), %g1 |
| 67 | |
| 68 | cmp %o3, %g1 |
| 69 | blu Lnot_really_big |
| 70 | clr %o4 |
| 71 | |
| 72 | ! Here the dividend is >= 2**(31-N) or so. We must be careful here, |
| 73 | ! as our usual N-at-a-shot divide step will cause overflow and havoc. |
| 74 | ! The number of bits in the result here is N*ITER+SC, where SC <= N. |
| 75 | ! Compute ITER in an unorthodox manner: know we need to shift V into |
| 76 | ! the top decade: so do not even bother to compare to R. |
| 77 | 1: |
| 78 | cmp %o5, %g1 |
| 79 | bgeu 3f |
| 80 | mov 1, %g7 |
| 81 | |
| 82 | sll %o5, 4, %o5 |
| 83 | |
| 84 | b 1b |
| 85 | add %o4, 1, %o4 |
| 86 | |
| 87 | ! Now compute %g7. |
| 88 | 2: |
| 89 | addcc %o5, %o5, %o5 |
| 90 | bcc Lnot_too_big |
| 91 | add %g7, 1, %g7 |
| 92 | |
| 93 | ! We get here if the %o1 overflowed while shifting. |
| 94 | ! This means that %o3 has the high-order bit set. |
| 95 | ! Restore %o5 and subtract from %o3. |
| 96 | sll %g1, 4, %g1 ! high order bit |
| 97 | srl %o5, 1, %o5 ! rest of %o5 |
| 98 | add %o5, %g1, %o5 |
| 99 | |
| 100 | b Ldo_single_div |
| 101 | sub %g7, 1, %g7 |
| 102 | |
| 103 | Lnot_too_big: |
| 104 | 3: |
| 105 | cmp %o5, %o3 |
| 106 | blu 2b |
| 107 | nop |
| 108 | |
| 109 | be Ldo_single_div |
| 110 | nop |
| 111 | /* NB: these are commented out in the V8-Sparc manual as well */ |
| 112 | /* (I do not understand this) */ |
| 113 | ! %o5 > %o3: went too far: back up 1 step |
| 114 | ! srl %o5, 1, %o5 |
| 115 | ! dec %g7 |
| 116 | ! do single-bit divide steps |
| 117 | ! |
| 118 | ! We have to be careful here. We know that %o3 >= %o5, so we can do the |
| 119 | ! first divide step without thinking. BUT, the others are conditional, |
| 120 | ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- |
| 121 | ! order bit set in the first step, just falling into the regular |
| 122 | ! division loop will mess up the first time around. |
| 123 | ! So we unroll slightly... |
| 124 | Ldo_single_div: |
| 125 | subcc %g7, 1, %g7 |
| 126 | bl Lend_regular_divide |
| 127 | nop |
| 128 | |
| 129 | sub %o3, %o5, %o3 |
| 130 | mov 1, %o2 |
| 131 | |
| 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 | |
| 151 | b,a Lend_regular_divide |
| 152 | |
| 153 | Lnot_really_big: |
| 154 | 1: |
| 155 | sll %o5, 4, %o5 |
| 156 | |
| 157 | cmp %o5, %o3 |
| 158 | bleu 1b |
| 159 | addcc %o4, 1, %o4 |
| 160 | |
| 161 | be Lgot_result |
| 162 | sub %o4, 1, %o4 |
| 163 | |
| 164 | tst %o3 ! set up for initial iteration |
| 165 | Ldivloop: |
| 166 | sll %o2, 4, %o2 |
| 167 | ! depth 1, accumulated bits 0 |
| 168 | bl L.1.16 |
| 169 | srl %o5,1,%o5 |
| 170 | ! remainder is positive |
| 171 | subcc %o3,%o5,%o3 |
| 172 | ! depth 2, accumulated bits 1 |
| 173 | bl L.2.17 |
| 174 | srl %o5,1,%o5 |
| 175 | ! remainder is positive |
| 176 | subcc %o3,%o5,%o3 |
| 177 | ! depth 3, accumulated bits 3 |
| 178 | bl L.3.19 |
| 179 | srl %o5,1,%o5 |
| 180 | ! remainder is positive |
| 181 | subcc %o3,%o5,%o3 |
| 182 | ! depth 4, accumulated bits 7 |
| 183 | bl L.4.23 |
| 184 | srl %o5,1,%o5 |
| 185 | ! remainder is positive |
| 186 | subcc %o3,%o5,%o3 |
| 187 | b 9f |
| 188 | add %o2, (7*2+1), %o2 |
| 189 | |
| 190 | L.4.23: |
| 191 | ! remainder is negative |
| 192 | addcc %o3,%o5,%o3 |
| 193 | b 9f |
| 194 | add %o2, (7*2-1), %o2 |
| 195 | |
| 196 | L.3.19: |
| 197 | ! remainder is negative |
| 198 | addcc %o3,%o5,%o3 |
| 199 | ! depth 4, accumulated bits 5 |
| 200 | bl L.4.21 |
| 201 | srl %o5,1,%o5 |
| 202 | ! remainder is positive |
| 203 | subcc %o3,%o5,%o3 |
| 204 | b 9f |
| 205 | add %o2, (5*2+1), %o2 |
| 206 | |
| 207 | L.4.21: |
| 208 | ! remainder is negative |
| 209 | addcc %o3,%o5,%o3 |
| 210 | b 9f |
| 211 | add %o2, (5*2-1), %o2 |
| 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 | L.3.17: |
| 236 | ! remainder is negative |
| 237 | addcc %o3,%o5,%o3 |
| 238 | ! depth 4, accumulated bits 1 |
| 239 | bl L.4.17 |
| 240 | srl %o5,1,%o5 |
| 241 | ! remainder is positive |
| 242 | subcc %o3,%o5,%o3 |
| 243 | b 9f |
| 244 | add %o2, (1*2+1), %o2 |
| 245 | |
| 246 | L.4.17: |
| 247 | ! remainder is negative |
| 248 | addcc %o3,%o5,%o3 |
| 249 | b 9f |
| 250 | add %o2, (1*2-1), %o2 |
| 251 | |
| 252 | L.1.16: |
| 253 | ! remainder is negative |
| 254 | addcc %o3,%o5,%o3 |
| 255 | ! depth 2, accumulated bits -1 |
| 256 | bl L.2.15 |
| 257 | srl %o5,1,%o5 |
| 258 | ! remainder is positive |
| 259 | subcc %o3,%o5,%o3 |
| 260 | ! depth 3, accumulated bits -1 |
| 261 | bl L.3.15 |
| 262 | srl %o5,1,%o5 |
| 263 | ! remainder is positive |
| 264 | subcc %o3,%o5,%o3 |
| 265 | ! depth 4, accumulated bits -1 |
| 266 | bl L.4.15 |
| 267 | srl %o5,1,%o5 |
| 268 | ! remainder is positive |
| 269 | subcc %o3,%o5,%o3 |
| 270 | b 9f |
| 271 | add %o2, (-1*2+1), %o2 |
| 272 | |
| 273 | L.4.15: |
| 274 | ! remainder is negative |
| 275 | addcc %o3,%o5,%o3 |
| 276 | b 9f |
| 277 | add %o2, (-1*2-1), %o2 |
| 278 | |
| 279 | L.3.15: |
| 280 | ! remainder is negative |
| 281 | addcc %o3,%o5,%o3 |
| 282 | ! depth 4, accumulated bits -3 |
| 283 | bl L.4.13 |
| 284 | srl %o5,1,%o5 |
| 285 | ! remainder is positive |
| 286 | subcc %o3,%o5,%o3 |
| 287 | b 9f |
| 288 | add %o2, (-3*2+1), %o2 |
| 289 | |
| 290 | L.4.13: |
| 291 | ! remainder is negative |
| 292 | addcc %o3,%o5,%o3 |
| 293 | b 9f |
| 294 | add %o2, (-3*2-1), %o2 |
| 295 | |
| 296 | L.2.15: |
| 297 | ! remainder is negative |
| 298 | addcc %o3,%o5,%o3 |
| 299 | ! depth 3, accumulated bits -3 |
| 300 | bl L.3.13 |
| 301 | srl %o5,1,%o5 |
| 302 | ! remainder is positive |
| 303 | subcc %o3,%o5,%o3 |
| 304 | ! depth 4, accumulated bits -5 |
| 305 | bl L.4.11 |
| 306 | srl %o5,1,%o5 |
| 307 | ! remainder is positive |
| 308 | subcc %o3,%o5,%o3 |
| 309 | b 9f |
| 310 | add %o2, (-5*2+1), %o2 |
| 311 | |
| 312 | L.4.11: |
| 313 | ! remainder is negative |
| 314 | addcc %o3,%o5,%o3 |
| 315 | b 9f |
| 316 | add %o2, (-5*2-1), %o2 |
| 317 | |
| 318 | L.3.13: |
| 319 | ! remainder is negative |
| 320 | addcc %o3,%o5,%o3 |
| 321 | ! depth 4, accumulated bits -7 |
| 322 | bl L.4.9 |
| 323 | srl %o5,1,%o5 |
| 324 | ! remainder is positive |
| 325 | subcc %o3,%o5,%o3 |
| 326 | b 9f |
| 327 | add %o2, (-7*2+1), %o2 |
| 328 | |
| 329 | L.4.9: |
| 330 | ! remainder is negative |
| 331 | addcc %o3,%o5,%o3 |
| 332 | b 9f |
| 333 | add %o2, (-7*2-1), %o2 |
| 334 | |
| 335 | 9: |
| 336 | Lend_regular_divide: |
| 337 | subcc %o4, 1, %o4 |
| 338 | bge Ldivloop |
| 339 | tst %o3 |
| 340 | |
| 341 | bl,a Lgot_result |
| 342 | ! non-restoring fixup here (one instruction only!) |
| 343 | sub %o2, 1, %o2 |
| 344 | |
| 345 | Lgot_result: |
| 346 | |
| 347 | retl |
| 348 | mov %o2, %o0 |
| 349 | |
| 350 | .globl .udiv_patch |
| 351 | .udiv_patch: |
| 352 | wr %g0, 0x0, %y |
| 353 | nop |
| 354 | nop |
| 355 | retl |
| 356 | udiv %o0, %o1, %o0 |
| 357 | nop |