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.
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 *  .urem	name of function to generate
17 *  rem		rem=div => %o0 / %o1; rem=rem => %o0 % %o1
18 *  false		false=true => signed; false=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	.globl .urem
44.urem:
45
46	! Ready to divide.  Compute size of quotient; scale comparand.
47	orcc	%o1, %g0, %o5
48	bne	1f
49	 mov	%o0, %o3
50
51		! Divide by zero trap.  If it returns, return 0 (about as
52		! wrong as possible, but that is what SunOS does...).
53		ta	ST_DIV0
54		retl
55		 clr	%o0
56
571:
58	cmp	%o3, %o5			! if %o1 exceeds %o0, done
59	blu	Lgot_result		! (and algorithm fails otherwise)
60	 clr	%o2
61
62	sethi	%hi(1 << (32 - 4 - 1)), %g1
63
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
78		sll	%o5, 4, %o5
79
80		b	1b
81		 add	%o4, 1, %o4
82
83	! Now compute %g7.
84	2:
85		addcc	%o5, %o5, %o5
86		bcc	Lnot_too_big
87		 add	%g7, 1, %g7
88
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
94		add	%o5, %g1, %o5
95
96		b	Ldo_single_div
97		 sub	%g7, 1, %g7
98
99	Lnot_too_big:
100	3:
101		cmp	%o5, %o3
102		blu	2b
103		 nop
104
105		be	Ldo_single_div
106		 nop
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
110	!	srl	%o5, 1, %o5
111	!	dec	%g7
112	! do single-bit divide steps
113	!
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...
120	Ldo_single_div:
121		subcc	%g7, 1, %g7
122		bl	Lend_regular_divide
123		 nop
124
125		sub	%o3, %o5, %o3
126		mov	1, %o2
127
128		b	Lend_single_divloop
129		 nop
130	Lsingle_divloop:
131		sll	%o2, 1, %o2
132		bl	1f
133		 srl	%o5, 1, %o5
134		! %o3 >= 0
135		sub	%o3, %o5, %o3
136		b	2f
137		 add	%o2, 1, %o2
138	1:	! %o3 < 0
139		add	%o3, %o5, %o3
140		sub	%o2, 1, %o2
141	2:
142	Lend_single_divloop:
143		subcc	%g7, 1, %g7
144		bge	Lsingle_divloop
145		 tst	%o3
146
147		b,a	Lend_regular_divide
148
149Lnot_really_big:
1501:
151	sll	%o5, 4, %o5
152
153	cmp	%o5, %o3
154	bleu	1b
155	 addcc	%o4, 1, %o4
156
157	be	Lgot_result
158	 sub	%o4, 1, %o4
159
160	tst	%o3	! set up for initial iteration
161Ldivloop:
162	sll	%o2, 4, %o2
163		! depth 1, accumulated bits 0
164	bl	L.1.16
165	 srl	%o5,1,%o5
166	! remainder is positive
167	subcc	%o3,%o5,%o3
168			! depth 2, accumulated bits 1
169	bl	L.2.17
170	 srl	%o5,1,%o5
171	! remainder is positive
172	subcc	%o3,%o5,%o3
173			! depth 3, accumulated bits 3
174	bl	L.3.19
175	 srl	%o5,1,%o5
176	! remainder is positive
177	subcc	%o3,%o5,%o3
178			! depth 4, accumulated bits 7
179	bl	L.4.23
180	 srl	%o5,1,%o5
181	! remainder is positive
182	subcc	%o3,%o5,%o3
183	b	9f
184	 add	%o2, (7*2+1), %o2
185
186L.4.23:
187	! remainder is negative
188	addcc	%o3,%o5,%o3
189	b	9f
190	 add	%o2, (7*2-1), %o2
191
192L.3.19:
193	! remainder is negative
194	addcc	%o3,%o5,%o3
195			! depth 4, accumulated bits 5
196	bl	L.4.21
197	 srl	%o5,1,%o5
198	! remainder is positive
199	subcc	%o3,%o5,%o3
200	b	9f
201	 add	%o2, (5*2+1), %o2
202
203L.4.21:
204	! remainder is negative
205	addcc	%o3,%o5,%o3
206	b	9f
207	 add	%o2, (5*2-1), %o2
208
209L.2.17:
210	! remainder is negative
211	addcc	%o3,%o5,%o3
212			! depth 3, accumulated bits 1
213	bl	L.3.17
214	 srl	%o5,1,%o5
215	! remainder is positive
216	subcc	%o3,%o5,%o3
217			! depth 4, accumulated bits 3
218	bl	L.4.19
219	 srl	%o5,1,%o5
220	! remainder is positive
221	subcc	%o3,%o5,%o3
222	b	9f
223	 add	%o2, (3*2+1), %o2
224
225L.4.19:
226	! remainder is negative
227	addcc	%o3,%o5,%o3
228	b	9f
229	 add	%o2, (3*2-1), %o2
230
231L.3.17:
232	! remainder is negative
233	addcc	%o3,%o5,%o3
234			! depth 4, accumulated bits 1
235	bl	L.4.17
236	 srl	%o5,1,%o5
237	! remainder is positive
238	subcc	%o3,%o5,%o3
239	b	9f
240	 add	%o2, (1*2+1), %o2
241
242L.4.17:
243	! remainder is negative
244	addcc	%o3,%o5,%o3
245	b	9f
246	 add	%o2, (1*2-1), %o2
247
248L.1.16:
249	! remainder is negative
250	addcc	%o3,%o5,%o3
251			! depth 2, accumulated bits -1
252	bl	L.2.15
253	 srl	%o5,1,%o5
254	! remainder is positive
255	subcc	%o3,%o5,%o3
256			! depth 3, accumulated bits -1
257	bl	L.3.15
258	 srl	%o5,1,%o5
259	! remainder is positive
260	subcc	%o3,%o5,%o3
261			! depth 4, accumulated bits -1
262	bl	L.4.15
263	 srl	%o5,1,%o5
264	! remainder is positive
265	subcc	%o3,%o5,%o3
266	b	9f
267	 add	%o2, (-1*2+1), %o2
268
269L.4.15:
270	! remainder is negative
271	addcc	%o3,%o5,%o3
272	b	9f
273	 add	%o2, (-1*2-1), %o2
274
275L.3.15:
276	! remainder is negative
277	addcc	%o3,%o5,%o3
278			! depth 4, accumulated bits -3
279	bl	L.4.13
280	 srl	%o5,1,%o5
281	! remainder is positive
282	subcc	%o3,%o5,%o3
283	b	9f
284	 add	%o2, (-3*2+1), %o2
285
286L.4.13:
287	! remainder is negative
288	addcc	%o3,%o5,%o3
289	b	9f
290	 add	%o2, (-3*2-1), %o2
291
292L.2.15:
293	! remainder is negative
294	addcc	%o3,%o5,%o3
295			! depth 3, accumulated bits -3
296	bl	L.3.13
297	 srl	%o5,1,%o5
298	! remainder is positive
299	subcc	%o3,%o5,%o3
300			! depth 4, accumulated bits -5
301	bl	L.4.11
302	 srl	%o5,1,%o5
303	! remainder is positive
304	subcc	%o3,%o5,%o3
305	b	9f
306	 add	%o2, (-5*2+1), %o2
307
308L.4.11:
309	! remainder is negative
310	addcc	%o3,%o5,%o3
311	b	9f
312	 add	%o2, (-5*2-1), %o2
313
314L.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
325L.4.9:
326	! remainder is negative
327	addcc	%o3,%o5,%o3
328	b	9f
329	 add	%o2, (-7*2-1), %o2
330
331	9:
332Lend_regular_divide:
333	subcc	%o4, 1, %o4
334	bge	Ldivloop
335	 tst	%o3
336
337	bl,a	Lgot_result
338	! non-restoring fixup here (one instruction only!)
339	add	%o3, %o1, %o3
340
341Lgot_result:
342
343	retl
344	 mov %o3, %o0
345
346	.globl	.urem_patch
347.urem_patch:
348	wr	%g0, 0x0, %y
349	nop
350	nop
351	nop
352	udiv	%o0, %o1, %o2
353	umul	%o2, %o1, %o2
354	retl
355	 sub	%o0, %o2, %o0
356