Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
36126f8f LT |
2 | #ifndef _ASM_WORD_AT_A_TIME_H |
3 | #define _ASM_WORD_AT_A_TIME_H | |
4 | ||
36126f8f | 5 | #include <linux/kernel.h> |
a6e2f029 CM |
6 | #include <asm/byteorder.h> |
7 | ||
8 | #ifdef __BIG_ENDIAN | |
36126f8f LT |
9 | |
10 | struct word_at_a_time { | |
11 | const unsigned long high_bits, low_bits; | |
12 | }; | |
13 | ||
14 | #define WORD_AT_A_TIME_CONSTANTS { REPEAT_BYTE(0xfe) + 1, REPEAT_BYTE(0x7f) } | |
15 | ||
16 | /* Bit set in the bytes that have a zero */ | |
17 | static inline long prep_zero_mask(unsigned long val, unsigned long rhs, const struct word_at_a_time *c) | |
18 | { | |
19 | unsigned long mask = (val & c->low_bits) + c->low_bits; | |
20 | return ~(mask | rhs); | |
21 | } | |
22 | ||
23 | #define create_zero_mask(mask) (mask) | |
24 | ||
25 | static inline long find_zero(unsigned long mask) | |
26 | { | |
27 | long byte = 0; | |
28 | #ifdef CONFIG_64BIT | |
29 | if (mask >> 32) | |
30 | mask >>= 32; | |
31 | else | |
32 | byte = 4; | |
33 | #endif | |
34 | if (mask >> 16) | |
35 | mask >>= 16; | |
36 | else | |
37 | byte += 2; | |
38 | return (mask >> 8) ? byte : byte + 1; | |
39 | } | |
40 | ||
41 | static inline bool has_zero(unsigned long val, unsigned long *data, const struct word_at_a_time *c) | |
42 | { | |
43 | unsigned long rhs = val | c->low_bits; | |
44 | *data = rhs; | |
45 | return (val + c->high_bits) & ~rhs; | |
46 | } | |
47 | ||
11ec50ca | 48 | #ifndef zero_bytemask |
789ce9dc | 49 | #define zero_bytemask(mask) (~1ul << __fls(mask)) |
ec6931b2 | 50 | #endif |
11ec50ca | 51 | |
a6e2f029 CM |
52 | #else |
53 | ||
54 | /* | |
55 | * The optimal byte mask counting is probably going to be something | |
56 | * that is architecture-specific. If you have a reliably fast | |
57 | * bit count instruction, that might be better than the multiply | |
58 | * and shift, for example. | |
59 | */ | |
60 | struct word_at_a_time { | |
61 | const unsigned long one_bits, high_bits; | |
62 | }; | |
63 | ||
64 | #define WORD_AT_A_TIME_CONSTANTS { REPEAT_BYTE(0x01), REPEAT_BYTE(0x80) } | |
65 | ||
66 | #ifdef CONFIG_64BIT | |
67 | ||
68 | /* | |
69 | * Jan Achrenius on G+: microoptimized version of | |
70 | * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56" | |
71 | * that works for the bytemasks without having to | |
72 | * mask them first. | |
73 | */ | |
74 | static inline long count_masked_bytes(unsigned long mask) | |
75 | { | |
76 | return mask*0x0001020304050608ul >> 56; | |
77 | } | |
78 | ||
79 | #else /* 32-bit case */ | |
80 | ||
81 | /* Carl Chatfield / Jan Achrenius G+ version for 32-bit */ | |
82 | static inline long count_masked_bytes(long mask) | |
83 | { | |
84 | /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */ | |
85 | long a = (0x0ff0001+mask) >> 23; | |
86 | /* Fix the 1 for 00 case */ | |
87 | return a & mask; | |
88 | } | |
89 | ||
90 | #endif | |
91 | ||
92 | /* Return nonzero if it has a zero */ | |
93 | static inline unsigned long has_zero(unsigned long a, unsigned long *bits, const struct word_at_a_time *c) | |
94 | { | |
95 | unsigned long mask = ((a - c->one_bits) & ~a) & c->high_bits; | |
96 | *bits = mask; | |
97 | return mask; | |
98 | } | |
99 | ||
100 | static inline unsigned long prep_zero_mask(unsigned long a, unsigned long bits, const struct word_at_a_time *c) | |
101 | { | |
102 | return bits; | |
103 | } | |
104 | ||
105 | static inline unsigned long create_zero_mask(unsigned long bits) | |
106 | { | |
107 | bits = (bits - 1) & ~bits; | |
108 | return bits >> 7; | |
109 | } | |
110 | ||
111 | /* The mask we created is directly usable as a bytemask */ | |
112 | #define zero_bytemask(mask) (mask) | |
113 | ||
114 | static inline unsigned long find_zero(unsigned long mask) | |
115 | { | |
116 | return count_masked_bytes(mask); | |
117 | } | |
118 | ||
119 | #endif /* __BIG_ENDIAN */ | |
120 | ||
36126f8f | 121 | #endif /* _ASM_WORD_AT_A_TIME_H */ |