Linus Torvalds | f68e556 | 2012-04-06 13:54:56 -0700 | [diff] [blame] | 1 | #ifndef _ASM_WORD_AT_A_TIME_H |
| 2 | #define _ASM_WORD_AT_A_TIME_H |
| 3 | |
| 4 | /* |
| 5 | * This is largely generic for little-endian machines, but the |
| 6 | * optimal byte mask counting is probably going to be something |
| 7 | * that is architecture-specific. If you have a reliably fast |
| 8 | * bit count instruction, that might be better than the multiply |
| 9 | * and shift, for example. |
| 10 | */ |
| 11 | |
| 12 | #ifdef CONFIG_64BIT |
| 13 | |
| 14 | /* |
| 15 | * Jan Achrenius on G+: microoptimized version of |
| 16 | * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56" |
| 17 | * that works for the bytemasks without having to |
| 18 | * mask them first. |
| 19 | */ |
| 20 | static inline long count_masked_bytes(unsigned long mask) |
| 21 | { |
| 22 | return mask*0x0001020304050608ul >> 56; |
| 23 | } |
| 24 | |
| 25 | #else /* 32-bit case */ |
| 26 | |
| 27 | /* Carl Chatfield / Jan Achrenius G+ version for 32-bit */ |
| 28 | static inline long count_masked_bytes(long mask) |
| 29 | { |
| 30 | /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */ |
| 31 | long a = (0x0ff0001+mask) >> 23; |
| 32 | /* Fix the 1 for 00 case */ |
| 33 | return a & mask; |
| 34 | } |
| 35 | |
| 36 | #endif |
| 37 | |
| 38 | #define REPEAT_BYTE(x) ((~0ul / 0xff) * (x)) |
| 39 | |
| 40 | /* Return the high bit set in the first byte that is a zero */ |
| 41 | static inline unsigned long has_zero(unsigned long a) |
| 42 | { |
| 43 | return ((a - REPEAT_BYTE(0x01)) & ~a) & REPEAT_BYTE(0x80); |
| 44 | } |
| 45 | |
| 46 | #endif /* _ASM_WORD_AT_A_TIME_H */ |