1/*
2 * include/asm-v850/unaligned.h -- Unaligned memory access
3 *
4 *  Copyright (C) 2001  NEC Corporation
5 *  Copyright (C) 2001  Miles Bader <miles@gnu.org>
6 *
7 * This file is subject to the terms and conditions of the GNU General
8 * Public License.  See the file COPYING in the main directory of this
9 * archive for more details.
10 *
11 * This file is a copy of the arm version, include/asm-arm/unaligned.h
12 *
13 * Note that some v850 chips support unaligned access, but it seems too
14 * annoying to use.
15 */
16
17#ifndef __V850_UNALIGNED_H__
18#define __V850_UNALIGNED_H__
19
20#include <asm/types.h>
21
22extern int __bug_unaligned_x(void *ptr);
23
24/*
25 * What is the most efficient way of loading/storing an unaligned value?
26 *
27 * That is the subject of this file.  Efficiency here is defined as
28 * minimum code size with minimum register usage for the common cases.
29 * It is currently not believed that long longs are common, so we
30 * trade efficiency for the chars, shorts and longs against the long
31 * longs.
32 *
33 * Current stats with gcc 2.7.2.2 for these functions:
34 *
35 *	ptrsize	get:	code	regs	put:	code	regs
36 *	1		1	1		1	2
37 *	2		3	2		3	2
38 *	4		7	3		7	3
39 *	8		20	6		16	6
40 *
41 * gcc 2.95.1 seems to code differently:
42 *
43 *	ptrsize	get:	code	regs	put:	code	regs
44 *	1		1	1		1	2
45 *	2		3	2		3	2
46 *	4		7	4		7	4
47 *	8		19	8		15	6
48 *
49 * which may or may not be more efficient (depending upon whether
50 * you can afford the extra registers).  Hopefully the gcc 2.95
51 * is inteligent enough to decide if it is better to use the
52 * extra register, but evidence so far seems to suggest otherwise.
53 *
54 * Unfortunately, gcc is not able to optimise the high word
55 * out of long long >> 32, or the low word from long long << 32
56 */
57
58#define __get_unaligned_2(__p)					\
59	(__p[0] | __p[1] << 8)
60
61#define __get_unaligned_4(__p)					\
62	(__p[0] | __p[1] << 8 | __p[2] << 16 | __p[3] << 24)
63
64#define get_unaligned(ptr)					\
65	({							\
66		__typeof__(*(ptr)) __v;				\
67		__u8 *__p = (__u8 *)(ptr);			\
68		switch (sizeof(*(ptr))) {			\
69		case 1:	__v = *(ptr);			break;	\
70		case 2: __v = __get_unaligned_2(__p);	break;	\
71		case 4: __v = __get_unaligned_4(__p);	break;	\
72		case 8: {					\
73				unsigned int __v1, __v2;	\
74				__v2 = __get_unaligned_4((__p+4)); \
75				__v1 = __get_unaligned_4(__p);	\
76				__v = ((unsigned long long)__v2 << 32 | __v1);	\
77			}					\
78			break;					\
79		default: __v = __bug_unaligned_x(__p);	break;	\
80		}						\
81		__v;						\
82	})
83
84
85static inline void __put_unaligned_2(__u32 __v, register __u8 *__p)
86{
87	*__p++ = __v;
88	*__p++ = __v >> 8;
89}
90
91static inline void __put_unaligned_4(__u32 __v, register __u8 *__p)
92{
93	__put_unaligned_2(__v >> 16, __p + 2);
94	__put_unaligned_2(__v, __p);
95}
96
97static inline void __put_unaligned_8(const unsigned long long __v, register __u8 *__p)
98{
99	/*
100	 * tradeoff: 8 bytes of stack for all unaligned puts (2
101	 * instructions), or an extra register in the long long
102	 * case - go for the extra register.
103	 */
104	__put_unaligned_4(__v >> 32, __p+4);
105	__put_unaligned_4(__v, __p);
106}
107
108/*
109 * Try to store an unaligned value as efficiently as possible.
110 */
111#define put_unaligned(val,ptr)					\
112	({							\
113		switch (sizeof(*(ptr))) {			\
114		case 1:						\
115			*(ptr) = (val);				\
116			break;					\
117		case 2: __put_unaligned_2((val),(__u8 *)(ptr));	\
118			break;					\
119		case 4:	__put_unaligned_4((val),(__u8 *)(ptr));	\
120			break;					\
121		case 8:	__put_unaligned_8((val),(__u8 *)(ptr)); \
122			break;					\
123		default: __bug_unaligned_x(ptr);		\
124			break;					\
125		}						\
126		(void) 0;					\
127	})
128
129
130#endif /* __V850_UNALIGNED_H__ */
131