xref: /linux/arch/m68k/lib/udivsi3.S (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
1/* libgcc1 routines for 68000 w/o floating-point hardware.
2   Copyright (C) 1994, 1996, 1997, 1998 Free Software Foundation, Inc.
3
4This file is part of GNU CC.
5
6GNU CC is free software; you can redistribute it and/or modify it
7under the terms of the GNU General Public License as published by the
8Free Software Foundation; either version 2, or (at your option) any
9later version.
10
11In addition to the permissions in the GNU General Public License, the
12Free Software Foundation gives you unlimited permission to link the
13compiled version of this file with other programs, and to distribute
14those programs without any restriction coming from the use of this
15file.  (The General Public License restrictions do apply in other
16respects; for example, they cover modification of the file, and
17distribution when not linked into another program.)
18
19This file is distributed in the hope that it will be useful, but
20WITHOUT ANY WARRANTY; without even the implied warranty of
21MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
22General Public License for more details.
23
24You should have received a copy of the GNU General Public License
25along with this program; see the file COPYING.  If not, write to
26the Free Software Foundation, 59 Temple Place - Suite 330,
27Boston, MA 02111-1307, USA.  */
28
29/* As a special exception, if you link this library with files
30   compiled with GCC to produce an executable, this does not cause
31   the resulting executable to be covered by the GNU General Public License.
32   This exception does not however invalidate any other reasons why
33   the executable file might be covered by the GNU General Public License.  */
34
35/* Use this one for any 680x0; assumes no floating point hardware.
36   The trailing " '" appearing on some lines is for ANSI preprocessors.  Yuk.
37   Some of this code comes from MINIX, via the folks at ericsson.
38   D. V. Henkel-Wallace (gumby@cygnus.com) Fete Bastille, 1992
39*/
40
41/* These are predefined by new versions of GNU cpp.  */
42
43#ifndef __USER_LABEL_PREFIX__
44#define __USER_LABEL_PREFIX__ _
45#endif
46
47#ifndef __REGISTER_PREFIX__
48#define __REGISTER_PREFIX__
49#endif
50
51#ifndef __IMMEDIATE_PREFIX__
52#define __IMMEDIATE_PREFIX__ #
53#endif
54
55/* ANSI concatenation macros.  */
56
57#define CONCAT1(a, b) CONCAT2(a, b)
58#define CONCAT2(a, b) a ## b
59
60/* Use the right prefix for global labels.  */
61
62#define SYM(x) CONCAT1 (__USER_LABEL_PREFIX__, x)
63
64/* Use the right prefix for registers.  */
65
66#define REG(x) CONCAT1 (__REGISTER_PREFIX__, x)
67
68/* Use the right prefix for immediate values.  */
69
70#define IMM(x) CONCAT1 (__IMMEDIATE_PREFIX__, x)
71
72#define d0 REG (d0)
73#define d1 REG (d1)
74#define d2 REG (d2)
75#define d3 REG (d3)
76#define d4 REG (d4)
77#define d5 REG (d5)
78#define d6 REG (d6)
79#define d7 REG (d7)
80#define a0 REG (a0)
81#define a1 REG (a1)
82#define a2 REG (a2)
83#define a3 REG (a3)
84#define a4 REG (a4)
85#define a5 REG (a5)
86#define a6 REG (a6)
87#define fp REG (fp)
88#define sp REG (sp)
89
90	.text
91	.proc
92	.globl	SYM (__udivsi3)
93SYM (__udivsi3):
94#if !(defined(__mcf5200__) || defined(__mcoldfire__))
95	movel	d2, sp@-
96	movel	sp@(12), d1	/* d1 = divisor */
97	movel	sp@(8), d0	/* d0 = dividend */
98
99	cmpl	IMM (0x10000), d1 /* divisor >= 2 ^ 16 ?   */
100	jcc	L3		/* then try next algorithm */
101	movel	d0, d2
102	clrw	d2
103	swap	d2
104	divu	d1, d2          /* high quotient in lower word */
105	movew	d2, d0		/* save high quotient */
106	swap	d0
107	movew	sp@(10), d2	/* get low dividend + high rest */
108	divu	d1, d2		/* low quotient */
109	movew	d2, d0
110	jra	L6
111
112L3:	movel	d1, d2		/* use d2 as divisor backup */
113L4:	lsrl	IMM (1), d1	/* shift divisor */
114	lsrl	IMM (1), d0	/* shift dividend */
115	cmpl	IMM (0x10000), d1 /* still divisor >= 2 ^ 16 ?  */
116	jcc	L4
117	divu	d1, d0		/* now we have 16 bit divisor */
118	andl	IMM (0xffff), d0 /* mask out divisor, ignore remainder */
119
120/* Multiply the 16 bit tentative quotient with the 32 bit divisor.  Because of
121   the operand ranges, this might give a 33 bit product.  If this product is
122   greater than the dividend, the tentative quotient was too large. */
123	movel	d2, d1
124	mulu	d0, d1		/* low part, 32 bits */
125	swap	d2
126	mulu	d0, d2		/* high part, at most 17 bits */
127	swap	d2		/* align high part with low part */
128	tstw	d2		/* high part 17 bits? */
129	jne	L5		/* if 17 bits, quotient was too large */
130	addl	d2, d1		/* add parts */
131	jcs	L5		/* if sum is 33 bits, quotient was too large */
132	cmpl	sp@(8), d1	/* compare the sum with the dividend */
133	jls	L6		/* if sum > dividend, quotient was too large */
134L5:	subql	IMM (1), d0	/* adjust quotient */
135
136L6:	movel	sp@+, d2
137	rts
138
139#else /* __mcf5200__ || __mcoldfire__ */
140
141/* Coldfire implementation of non-restoring division algorithm from
142   Hennessy & Patterson, Appendix A. */
143	link	a6,IMM (-12)
144	moveml	d2-d4,sp@
145	movel	a6@(8),d0
146	movel	a6@(12),d1
147	clrl	d2		| clear p
148	moveq	IMM (31),d4
149L1:	addl	d0,d0		| shift reg pair (p,a) one bit left
150	addxl	d2,d2
151	movl	d2,d3		| subtract b from p, store in tmp.
152	subl	d1,d3
153	jcs	L2		| if no carry,
154	bset	IMM (0),d0	| set the low order bit of a to 1,
155	movl	d3,d2		| and store tmp in p.
156L2:	subql	IMM (1),d4
157	jcc	L1
158	moveml	sp@,d2-d4	| restore data registers
159	unlk	a6		| and return
160	rts
161#endif /* __mcf5200__ || __mcoldfire__ */
162
163