1 /* $Id: urem.S,v 1.4 1996/09/30 02:22:42 davem Exp $
2 * urem.S: This routine was taken from glibc-1.09 and is covered
3 * by the GNU Library General Public License Version 2.
6 /* This file is generated from divrem.m4; DO NOT EDIT! */
8 * Division and remainder, from Appendix E of the Sparc Version 8
9 * Architecture Manual, with fixes from Gordon Irlam.
13 * Input: dividend and divisor in %o0 and %o1 respectively.
16 * .urem name of function to generate
17 * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1
18 * false false=true => signed; false=false => unsigned
20 * Algorithm parameters:
21 * N how many bits per iteration we try to get (4)
22 * WORDSIZE total number of bits (32)
25 * TOPBITS number of bits in the top decade of a number
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)
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
46 ! Ready to divide. Compute size of quotient; scale comparand.
51 ! Divide by zero trap. If it returns, return 0 (about as
52 ! wrong as possible, but that is what SunOS does...).
58 cmp %o3, %o5 ! if %o1 exceeds %o0, done
59 blu Lgot_result ! (and algorithm fails otherwise)
62 sethi %hi(1 << (32 - 4 - 1)), %g1
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.
89 ! We get here if the %o1 overflowed while shifting.
90 ! This means that %o3 has the high-order bit set.
91 ! Restore %o5 and subtract from %o3.
92 sll %g1, 4, %g1 ! high order bit
93 srl %o5, 1, %o5 ! rest of %o5
107 /* NB: these are commented out in the V8-Sparc manual as well */
108 /* (I do not understand this) */
109 ! %o5 > %o3: went too far: back up 1 step
112 ! do single-bit divide steps
114 ! We have to be careful here. We know that %o3 >= %o5, so we can do the
115 ! first divide step without thinking. BUT, the others are conditional,
116 ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high-
117 ! order bit set in the first step, just falling into the regular
118 ! division loop will mess up the first time around.
119 ! So we unroll slightly...
122 bl Lend_regular_divide
128 b Lend_single_divloop
147 b,a Lend_regular_divide
160 tst %o3 ! set up for initial iteration
163 ! depth 1, accumulated bits 0
166 ! remainder is positive
168 ! depth 2, accumulated bits 1
171 ! remainder is positive
173 ! depth 3, accumulated bits 3
176 ! remainder is positive
178 ! depth 4, accumulated bits 7
181 ! remainder is positive
184 add %o2, (7*2+1), %o2
187 ! remainder is negative
190 add %o2, (7*2-1), %o2
193 ! remainder is negative
195 ! depth 4, accumulated bits 5
198 ! remainder is positive
201 add %o2, (5*2+1), %o2
204 ! remainder is negative
207 add %o2, (5*2-1), %o2
210 ! remainder is negative
212 ! depth 3, accumulated bits 1
215 ! remainder is positive
217 ! depth 4, accumulated bits 3
220 ! remainder is positive
223 add %o2, (3*2+1), %o2
226 ! remainder is negative
229 add %o2, (3*2-1), %o2
232 ! remainder is negative
234 ! depth 4, accumulated bits 1
237 ! remainder is positive
240 add %o2, (1*2+1), %o2
243 ! remainder is negative
246 add %o2, (1*2-1), %o2
249 ! remainder is negative
251 ! depth 2, accumulated bits -1
254 ! remainder is positive
256 ! depth 3, accumulated bits -1
259 ! remainder is positive
261 ! depth 4, accumulated bits -1
264 ! remainder is positive
267 add %o2, (-1*2+1), %o2
270 ! remainder is negative
273 add %o2, (-1*2-1), %o2
276 ! remainder is negative
278 ! depth 4, accumulated bits -3
281 ! remainder is positive
284 add %o2, (-3*2+1), %o2
287 ! remainder is negative
290 add %o2, (-3*2-1), %o2
293 ! remainder is negative
295 ! depth 3, accumulated bits -3
298 ! remainder is positive
300 ! depth 4, accumulated bits -5
303 ! remainder is positive
306 add %o2, (-5*2+1), %o2
309 ! remainder is negative
312 add %o2, (-5*2-1), %o2
315 ! remainder is negative
317 ! depth 4, accumulated bits -7
320 ! remainder is positive
323 add %o2, (-7*2+1), %o2
326 ! remainder is negative
329 add %o2, (-7*2-1), %o2
338 ! non-restoring fixup here (one instruction only!)