]>
Commit | Line | Data |
---|---|---|
36126f8f LT |
1 | #ifndef _ASM_WORD_AT_A_TIME_H |
2 | #define _ASM_WORD_AT_A_TIME_H | |
3 | ||
36126f8f | 4 | #include <linux/kernel.h> |
a6e2f029 CM |
5 | #include <asm/byteorder.h> |
6 | ||
7 | #ifdef __BIG_ENDIAN | |
36126f8f LT |
8 | |
9 | struct word_at_a_time { | |
10 | const unsigned long high_bits, low_bits; | |
11 | }; | |
12 | ||
13 | #define WORD_AT_A_TIME_CONSTANTS { REPEAT_BYTE(0xfe) + 1, REPEAT_BYTE(0x7f) } | |
14 | ||
15 | /* Bit set in the bytes that have a zero */ | |
16 | static inline long prep_zero_mask(unsigned long val, unsigned long rhs, const struct word_at_a_time *c) | |
17 | { | |
18 | unsigned long mask = (val & c->low_bits) + c->low_bits; | |
19 | return ~(mask | rhs); | |
20 | } | |
21 | ||
22 | #define create_zero_mask(mask) (mask) | |
23 | ||
24 | static inline long find_zero(unsigned long mask) | |
25 | { | |
26 | long byte = 0; | |
27 | #ifdef CONFIG_64BIT | |
28 | if (mask >> 32) | |
29 | mask >>= 32; | |
30 | else | |
31 | byte = 4; | |
32 | #endif | |
33 | if (mask >> 16) | |
34 | mask >>= 16; | |
35 | else | |
36 | byte += 2; | |
37 | return (mask >> 8) ? byte : byte + 1; | |
38 | } | |
39 | ||
40 | static inline bool has_zero(unsigned long val, unsigned long *data, const struct word_at_a_time *c) | |
41 | { | |
42 | unsigned long rhs = val | c->low_bits; | |
43 | *data = rhs; | |
44 | return (val + c->high_bits) & ~rhs; | |
45 | } | |
46 | ||
11ec50ca | 47 | #ifndef zero_bytemask |
789ce9dc | 48 | #define zero_bytemask(mask) (~1ul << __fls(mask)) |
ec6931b2 | 49 | #endif |
11ec50ca | 50 | |
a6e2f029 CM |
51 | #else |
52 | ||
53 | /* | |
54 | * The optimal byte mask counting is probably going to be something | |
55 | * that is architecture-specific. If you have a reliably fast | |
56 | * bit count instruction, that might be better than the multiply | |
57 | * and shift, for example. | |
58 | */ | |
59 | struct word_at_a_time { | |
60 | const unsigned long one_bits, high_bits; | |
61 | }; | |
62 | ||
63 | #define WORD_AT_A_TIME_CONSTANTS { REPEAT_BYTE(0x01), REPEAT_BYTE(0x80) } | |
64 | ||
65 | #ifdef CONFIG_64BIT | |
66 | ||
67 | /* | |
68 | * Jan Achrenius on G+: microoptimized version of | |
69 | * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56" | |
70 | * that works for the bytemasks without having to | |
71 | * mask them first. | |
72 | */ | |
73 | static inline long count_masked_bytes(unsigned long mask) | |
74 | { | |
75 | return mask*0x0001020304050608ul >> 56; | |
76 | } | |
77 | ||
78 | #else /* 32-bit case */ | |
79 | ||
80 | /* Carl Chatfield / Jan Achrenius G+ version for 32-bit */ | |
81 | static inline long count_masked_bytes(long mask) | |
82 | { | |
83 | /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */ | |
84 | long a = (0x0ff0001+mask) >> 23; | |
85 | /* Fix the 1 for 00 case */ | |
86 | return a & mask; | |
87 | } | |
88 | ||
89 | #endif | |
90 | ||
91 | /* Return nonzero if it has a zero */ | |
92 | static inline unsigned long has_zero(unsigned long a, unsigned long *bits, const struct word_at_a_time *c) | |
93 | { | |
94 | unsigned long mask = ((a - c->one_bits) & ~a) & c->high_bits; | |
95 | *bits = mask; | |
96 | return mask; | |
97 | } | |
98 | ||
99 | static inline unsigned long prep_zero_mask(unsigned long a, unsigned long bits, const struct word_at_a_time *c) | |
100 | { | |
101 | return bits; | |
102 | } | |
103 | ||
104 | static inline unsigned long create_zero_mask(unsigned long bits) | |
105 | { | |
106 | bits = (bits - 1) & ~bits; | |
107 | return bits >> 7; | |
108 | } | |
109 | ||
110 | /* The mask we created is directly usable as a bytemask */ | |
111 | #define zero_bytemask(mask) (mask) | |
112 | ||
113 | static inline unsigned long find_zero(unsigned long mask) | |
114 | { | |
115 | return count_masked_bytes(mask); | |
116 | } | |
117 | ||
118 | #endif /* __BIG_ENDIAN */ | |
119 | ||
36126f8f | 120 | #endif /* _ASM_WORD_AT_A_TIME_H */ |