| 1 |
/* Searching in a string. |
| 2 |
Copyright (C) 2008-2024 Free Software Foundation, Inc. |
| 3 |
|
| 4 |
This file is free software: you can redistribute it and/or modify |
| 5 |
it under the terms of the GNU Lesser General Public License as |
| 6 |
published by the Free Software Foundation; either version 2.1 of the |
| 7 |
License, or (at your option) any later version. |
| 8 |
|
| 9 |
This file is distributed in the hope that it will be useful, |
| 10 |
but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 |
GNU Lesser General Public License for more details. |
| 13 |
|
| 14 |
You should have received a copy of the GNU Lesser General Public License |
| 15 |
along with this program. If not, see <https://www.gnu.org/licenses/>. */ |
| 16 |
|
| 17 |
#include <config.h> |
| 18 |
|
| 19 |
/* Specification. */ |
| 20 |
#include <string.h> |
| 21 |
|
| 22 |
/* A function definition is needed only if HAVE_RAWMEMCHR is not defined. */ |
| 23 |
#if !HAVE_RAWMEMCHR |
| 24 |
|
| 25 |
# include <limits.h> |
| 26 |
# include <stdint.h> |
| 27 |
|
| 28 |
|
| 29 |
/* Find the first occurrence of C in S. */ |
| 30 |
void * |
| 31 |
rawmemchr (const void *s, int c_in) |
| 32 |
{ |
| 33 |
# ifdef __CHERI_PURE_CAPABILITY__ |
| 34 |
/* Most architectures let you read an aligned word, |
| 35 |
even if the unsigned char array at S ends in the middle of the word. |
| 36 |
However CHERI does not, so call memchr |
| 37 |
with the underlying object's remaining length. |
| 38 |
This cannot return NULL if S points to a C_IN-terminated array. |
| 39 |
Use builtins rather than including <cheri.h> which is less stable. */ |
| 40 |
return memchr (s, c_in, (__builtin_cheri_length_get (s) |
| 41 |
- __builtin_cheri_offset_get (s))); |
| 42 |
# else |
| 43 |
|
| 44 |
/* You can change this typedef to experiment with performance. */ |
| 45 |
typedef uintptr_t longword; |
| 46 |
/* Verify that the longword type lacks padding bits. */ |
| 47 |
static_assert (UINTPTR_WIDTH == UCHAR_WIDTH * sizeof (uintptr_t)); |
| 48 |
|
| 49 |
const unsigned char *char_ptr; |
| 50 |
unsigned char c = c_in; |
| 51 |
|
| 52 |
/* Handle the first few bytes by reading one byte at a time. |
| 53 |
Do this until CHAR_PTR is aligned on a natural longword boundary, |
| 54 |
as using alignof (longword) might be slower. */ |
| 55 |
for (char_ptr = (const unsigned char *) s; |
| 56 |
(uintptr_t) char_ptr % sizeof (longword) != 0; |
| 57 |
++char_ptr) |
| 58 |
if (*char_ptr == c) |
| 59 |
return (void *) char_ptr; |
| 60 |
|
| 61 |
longword const *longword_ptr = s = char_ptr; |
| 62 |
|
| 63 |
/* Compute auxiliary longword values: |
| 64 |
repeated_one is a value which has a 1 in every byte. |
| 65 |
repeated_c has c in every byte. */ |
| 66 |
longword repeated_one = (longword) -1 / UCHAR_MAX; |
| 67 |
longword repeated_c = repeated_one * c; |
| 68 |
longword repeated_hibit = repeated_one * (UCHAR_MAX / 2 + 1); |
| 69 |
|
| 70 |
/* Instead of the traditional loop which tests each byte, we will |
| 71 |
test a longword at a time. The tricky part is testing if any of |
| 72 |
the bytes in the longword in question are equal to |
| 73 |
c. We first use an xor with repeated_c. This reduces the task |
| 74 |
to testing whether any of the bytes in longword1 is zero. |
| 75 |
|
| 76 |
(The following comments assume 8-bit bytes, as POSIX requires; |
| 77 |
the code's use of UCHAR_MAX should work even if bytes have more |
| 78 |
than 8 bits.) |
| 79 |
|
| 80 |
We compute tmp = |
| 81 |
((longword1 - repeated_one) & ~longword1) & (repeated_one * 0x80). |
| 82 |
That is, we perform the following operations: |
| 83 |
1. Subtract repeated_one. |
| 84 |
2. & ~longword1. |
| 85 |
3. & a mask consisting of 0x80 in every byte. |
| 86 |
Consider what happens in each byte: |
| 87 |
- If a byte of longword1 is zero, step 1 and 2 transform it into 0xff, |
| 88 |
and step 3 transforms it into 0x80. A carry can also be propagated |
| 89 |
to more significant bytes. |
| 90 |
- If a byte of longword1 is nonzero, let its lowest 1 bit be at |
| 91 |
position k (0 <= k <= 7); so the lowest k bits are 0. After step 1, |
| 92 |
the byte ends in a single bit of value 0 and k bits of value 1. |
| 93 |
After step 2, the result is just k bits of value 1: 2^k - 1. After |
| 94 |
step 3, the result is 0. And no carry is produced. |
| 95 |
So, if longword1 has only non-zero bytes, tmp is zero. |
| 96 |
Whereas if longword1 has a zero byte, call j the position of the least |
| 97 |
significant zero byte. Then the result has a zero at positions 0, ..., |
| 98 |
j-1 and a 0x80 at position j. We cannot predict the result at the more |
| 99 |
significant bytes (positions j+1..3), but it does not matter since we |
| 100 |
already have a non-zero bit at position 8*j+7. |
| 101 |
|
| 102 |
The test whether any byte in longword1 is zero is equivalent |
| 103 |
to testing whether tmp is nonzero. |
| 104 |
|
| 105 |
This test can read beyond the end of a string, depending on where |
| 106 |
C_IN is encountered. However, this is considered safe since the |
| 107 |
initialization phase ensured that the read will be aligned, |
| 108 |
therefore, the read will not cross page boundaries and will not |
| 109 |
cause a fault. */ |
| 110 |
|
| 111 |
while (1) |
| 112 |
{ |
| 113 |
longword longword1 = *longword_ptr ^ repeated_c; |
| 114 |
|
| 115 |
if ((((longword1 - repeated_one) & ~longword1) & repeated_hibit) != 0) |
| 116 |
break; |
| 117 |
longword_ptr++; |
| 118 |
} |
| 119 |
|
| 120 |
char_ptr = s = longword_ptr; |
| 121 |
|
| 122 |
/* At this point, we know that one of the sizeof (longword) bytes |
| 123 |
starting at char_ptr is == c. If we knew endianness, we |
| 124 |
could determine the first such byte without any further memory |
| 125 |
accesses, just by looking at the tmp result from the last loop |
| 126 |
iteration. However, the following simple and portable code does |
| 127 |
not attempt this potential optimization. */ |
| 128 |
|
| 129 |
while (*char_ptr != c) |
| 130 |
char_ptr++; |
| 131 |
return (void *) char_ptr; |
| 132 |
# endif |
| 133 |
} |
| 134 |
|
| 135 |
#endif |