summaryrefslogtreecommitdiffstats
path: root/arch/arm/lib/div64.S
blob: 58eef66076293a923d00a81e03370bcc1c734f83 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
/*
 *  linux/arch/arm/lib/div64.S
 *
 *  Optimized computation of 64-bit dividend / 32-bit divisor
 *
 *  Author:	Nicolas Pitre
 *  Created:	Oct 5, 2003
 *  Copyright:	Monta Vista Software, Inc.
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License version 2 as
 *  published by the Free Software Foundation.
 */

#include <linux/linkage.h>

#ifdef __ARMEB__
#define xh r0
#define xl r1
#define yh r2
#define yl r3
#else
#define xl r0
#define xh r1
#define yl r2
#define yh r3
#endif

/*
 * __do_div64: perform a division with 64-bit dividend and 32-bit divisor.
 *
 * Note: Calling convention is totally non standard for optimal code.
 *       This is meant to be used by do_div() from include/asm/div64.h only.
 *
 * Input parameters:
 * 	xh-xl	= dividend (clobbered)
 * 	r4	= divisor (preserved)
 *
 * Output values:
 * 	yh-yl	= result
 * 	xh	= remainder
 *
 * Clobbered regs: xl, ip
 */

ENTRY(__do_div64)

	@ Test for easy paths first.
	subs	ip, r4, #1
	bls	9f			@ divisor is 0 or 1
	tst	ip, r4
	beq	8f			@ divisor is power of 2

	@ See if we need to handle upper 32-bit result.
	cmp	xh, r4
	mov	yh, #0
	blo	3f

	@ Align divisor with upper part of dividend.
	@ The aligned divisor is stored in yl preserving the original.
	@ The bit position is stored in ip.

#if __LINUX_ARM_ARCH__ >= 5

	clz	yl, r4
	clz	ip, xh
	sub	yl, yl, ip
	mov	ip, #1
	mov	ip, ip, lsl yl
	mov	yl, r4, lsl yl

#else

	mov	yl, r4
	mov	ip, #1
1:	cmp	yl, #0x80000000
	cmpcc	yl, xh
	movcc	yl, yl, lsl #1
	movcc	ip, ip, lsl #1
	bcc	1b

#endif

	@ The division loop for needed upper bit positions.
 	@ Break out early if dividend reaches 0.
2:	cmp	xh, yl
	orrcs	yh, yh, ip
	subcss	xh, xh, yl
	movnes	ip, ip, lsr #1
	mov	yl, yl, lsr #1
	bne	2b

	@ See if we need to handle lower 32-bit result.
3:	cmp	xh, #0
	mov	yl, #0
	cmpeq	xl, r4
	movlo	xh, xl
	movlo	pc, lr

	@ The division loop for lower bit positions.
	@ Here we shift remainer bits leftwards rather than moving the
	@ divisor for comparisons, considering the carry-out bit as well.
	mov	ip, #0x80000000
4:	movs	xl, xl, lsl #1
	adcs	xh, xh, xh
	beq	6f
	cmpcc	xh, r4
5:	orrcs	yl, yl, ip
	subcs	xh, xh, r4
	movs	ip, ip, lsr #1
	bne	4b
	mov	pc, lr

	@ The top part of remainder became zero.  If carry is set
	@ (the 33th bit) this is a false positive so resume the loop.
	@ Otherwise, if lower part is also null then we are done.
6:	bcs	5b
	cmp	xl, #0
	moveq	pc, lr

	@ We still have remainer bits in the low part.  Bring them up.

#if __LINUX_ARM_ARCH__ >= 5

	clz	xh, xl			@ we know xh is zero here so...
	add	xh, xh, #1
	mov	xl, xl, lsl xh
	mov	ip, ip, lsr xh

#else

7:	movs	xl, xl, lsl #1
	mov	ip, ip, lsr #1
	bcc	7b

#endif

	@ Current remainder is now 1.  It is worthless to compare with
	@ divisor at this point since divisor can not be smaller than 3 here.
	@ If possible, branch for another shift in the division loop.
	@ If no bit position left then we are done.
	movs	ip, ip, lsr #1
	mov	xh, #1
	bne	4b
	mov	pc, lr

8:	@ Division by a power of 2: determine what that divisor order is
	@ then simply shift values around

#if __LINUX_ARM_ARCH__ >= 5

	clz	ip, r4
	rsb	ip, ip, #31

#else

	mov	yl, r4
	cmp	r4, #(1 << 16)
	mov	ip, #0
	movhs	yl, yl, lsr #16
	movhs	ip, #16

	cmp	yl, #(1 << 8)
	movhs	yl, yl, lsr #8
	addhs	ip, ip, #8

	cmp	yl, #(1 << 4)
	movhs	yl, yl, lsr #4
	addhs	ip, ip, #4

	cmp	yl, #(1 << 2)
	addhi	ip, ip, #3
	addls	ip, ip, yl, lsr #1

#endif

	mov	yh, xh, lsr ip
	mov	yl, xl, lsr ip
	rsb	ip, ip, #32
	orr	yl, yl, xh, lsl ip
	mov	xh, xl, lsl ip
	mov	xh, xh, lsr ip
	mov	pc, lr

	@ eq -> division by 1: obvious enough...
9:	moveq	yl, xl
	moveq	yh, xh
	moveq	xh, #0
	moveq	pc, lr

	@ Division by 0:
	str	lr, [sp, #-8]!
	bl	__div0

	@ as wrong as it could be...
	mov	yl, #0
	mov	yh, #0
	mov	xh, #0
	ldr	pc, [sp], #8

OpenPOWER on IntegriCloud