xref: /linux/arch/m68k/lib/udivsi3.S (revision cc04a46f11ea046ed53e2c832ae29e4790f7e35f)
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
24/* As a special exception, if you link this library with files
25   compiled with GCC to produce an executable, this does not cause
26   the resulting executable to be covered by the GNU General Public License.
27   This exception does not however invalidate any other reasons why
28   the executable file might be covered by the GNU General Public License.  */
29
30/* Use this one for any 680x0; assumes no floating point hardware.
31   The trailing " '" appearing on some lines is for ANSI preprocessors.  Yuk.
32   Some of this code comes from MINIX, via the folks at ericsson.
33   D. V. Henkel-Wallace (gumby@cygnus.com) Fete Bastille, 1992
34*/
35
36/* These are predefined by new versions of GNU cpp.  */
37
38#ifndef __USER_LABEL_PREFIX__
39#define __USER_LABEL_PREFIX__ _
40#endif
41
42#ifndef __REGISTER_PREFIX__
43#define __REGISTER_PREFIX__
44#endif
45
46#ifndef __IMMEDIATE_PREFIX__
47#define __IMMEDIATE_PREFIX__ #
48#endif
49
50/* ANSI concatenation macros.  */
51
52#define CONCAT1(a, b) CONCAT2(a, b)
53#define CONCAT2(a, b) a ## b
54
55/* Use the right prefix for global labels.  */
56
57#define SYM(x) CONCAT1 (__USER_LABEL_PREFIX__, x)
58
59/* Use the right prefix for registers.  */
60
61#define REG(x) CONCAT1 (__REGISTER_PREFIX__, x)
62
63/* Use the right prefix for immediate values.  */
64
65#define IMM(x) CONCAT1 (__IMMEDIATE_PREFIX__, x)
66
67#define d0 REG (d0)
68#define d1 REG (d1)
69#define d2 REG (d2)
70#define d3 REG (d3)
71#define d4 REG (d4)
72#define d5 REG (d5)
73#define d6 REG (d6)
74#define d7 REG (d7)
75#define a0 REG (a0)
76#define a1 REG (a1)
77#define a2 REG (a2)
78#define a3 REG (a3)
79#define a4 REG (a4)
80#define a5 REG (a5)
81#define a6 REG (a6)
82#define fp REG (fp)
83#define sp REG (sp)
84
85	.text
86	.proc
87	.globl	SYM (__udivsi3)
88SYM (__udivsi3):
89#if !(defined(__mcf5200__) || defined(__mcoldfire__))
90	movel	d2, sp@-
91	movel	sp@(12), d1	/* d1 = divisor */
92	movel	sp@(8), d0	/* d0 = dividend */
93
94	cmpl	IMM (0x10000), d1 /* divisor >= 2 ^ 16 ?   */
95	jcc	L3		/* then try next algorithm */
96	movel	d0, d2
97	clrw	d2
98	swap	d2
99	divu	d1, d2          /* high quotient in lower word */
100	movew	d2, d0		/* save high quotient */
101	swap	d0
102	movew	sp@(10), d2	/* get low dividend + high rest */
103	divu	d1, d2		/* low quotient */
104	movew	d2, d0
105	jra	L6
106
107L3:	movel	d1, d2		/* use d2 as divisor backup */
108L4:	lsrl	IMM (1), d1	/* shift divisor */
109	lsrl	IMM (1), d0	/* shift dividend */
110	cmpl	IMM (0x10000), d1 /* still divisor >= 2 ^ 16 ?  */
111	jcc	L4
112	divu	d1, d0		/* now we have 16 bit divisor */
113	andl	IMM (0xffff), d0 /* mask out divisor, ignore remainder */
114
115/* Multiply the 16 bit tentative quotient with the 32 bit divisor.  Because of
116   the operand ranges, this might give a 33 bit product.  If this product is
117   greater than the dividend, the tentative quotient was too large. */
118	movel	d2, d1
119	mulu	d0, d1		/* low part, 32 bits */
120	swap	d2
121	mulu	d0, d2		/* high part, at most 17 bits */
122	swap	d2		/* align high part with low part */
123	tstw	d2		/* high part 17 bits? */
124	jne	L5		/* if 17 bits, quotient was too large */
125	addl	d2, d1		/* add parts */
126	jcs	L5		/* if sum is 33 bits, quotient was too large */
127	cmpl	sp@(8), d1	/* compare the sum with the dividend */
128	jls	L6		/* if sum > dividend, quotient was too large */
129L5:	subql	IMM (1), d0	/* adjust quotient */
130
131L6:	movel	sp@+, d2
132	rts
133
134#else /* __mcf5200__ || __mcoldfire__ */
135
136/* Coldfire implementation of non-restoring division algorithm from
137   Hennessy & Patterson, Appendix A. */
138	link	a6,IMM (-12)
139	moveml	d2-d4,sp@
140	movel	a6@(8),d0
141	movel	a6@(12),d1
142	clrl	d2		| clear p
143	moveq	IMM (31),d4
144L1:	addl	d0,d0		| shift reg pair (p,a) one bit left
145	addxl	d2,d2
146	movl	d2,d3		| subtract b from p, store in tmp.
147	subl	d1,d3
148	jcs	L2		| if no carry,
149	bset	IMM (0),d0	| set the low order bit of a to 1,
150	movl	d3,d2		| and store tmp in p.
151L2:	subql	IMM (1),d4
152	jcc	L1
153	moveml	sp@,d2-d4	| restore data registers
154	unlk	a6		| and return
155	rts
156#endif /* __mcf5200__ || __mcoldfire__ */
157
158