include/asm-generic/word-at-a-time.h

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