1 /* _dl_new_hash for elf symbol lookup
2    Copyright (C) 2022 Free Software Foundation, Inc.
3    This file is part of the GNU C Library.
4 
5    The GNU C Library is free software; you can redistribute it and/or
6    modify it under the terms of the GNU Lesser General Public
7    License as published by the Free Software Foundation; either
8    version 2.1 of the License, or (at your option) any later version.
9 
10    The GNU C Library is distributed in the hope that it will be useful,
11    but WITHOUT ANY WARRANTY; without even the implied warranty of
12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13    Lesser General Public License for more details.
14 
15    You should have received a copy of the GNU Lesser General Public
16    License along with the GNU C Library; if not, see
17    <https://www.gnu.org/licenses/>.  */
18 
19 #ifndef _DL_NEW_HASH_H
20 #define _DL_NEW_HASH_H 1
21 
22 #include <stdint.h>
23 /* For __always_inline and __glibc_unlikely.  */
24 #include <sys/cdefs.h>
25 
26 /* The simplest implementation of _dl_new_hash is:
27 
28    _dl_new_hash (const char *s)
29    {
30       uint32_t h = 5381;
31       for (unsigned char c = *s; c != '\0'; c = *++s)
32         h = h * 33 + c;
33       return h;
34    }
35 
36    We can get better performance by slightly unrolling the loop to
37    pipeline the multiples, which gcc cannot easily do due to
38    dependencies across iterations.
39 
40    As well, as an architecture specific option we add asm statements
41    to explicitly specify order of operations and prevent reassociation
42    of instructions that lengthens the loop carried dependency. This
43    may have no affect as the compiler may have ordered instructions
44    the same way without it but in testing this has not been the case
45    for GCC. Improving GCC to reliably schedule instructions ideally
46    cannot be easily done.
47 
48    Architecture(s) that use the reassociation barriers are:
49    x86
50 
51    Note it is very unlikely the reassociation barriers would
52    de-optimize performance on any architecture and with an imperfect
53    compiler it may help performance, especially on out-of-order cpus,
54    so it is suggested that the respective maintainers add them.
55 
56    Architecture maintainers are encouraged to benchmark this with
57    __asm_reassociation_barrier defined to __asm__ like it is in x86.
58 */
59 
60 
61 #ifndef __asm_reassociation_barrier
62 # define __asm_reassociation_barrier(...)
63 #endif
64 
65 static __always_inline uint32_t
66 __attribute__ ((unused))
_dl_new_hash(const char * str)67 _dl_new_hash (const char *str)
68 {
69   const unsigned char *s = (const unsigned char *) str;
70   unsigned int h = 5381;
71   unsigned int c0, c1;
72   for (;;)
73     {
74       c0 = s[0];
75       /* Since hashed string is normally not empty, this is unlikely on the
76 	 first iteration of the loop.  */
77       if (__glibc_unlikely (c0 == 0))
78 	return h;
79 
80       c1 = s[1];
81       if (c1 == 0)
82 	{
83 	  /* Ideal computational order is:
84 	 c0 += h;
85 	 h *= 32;
86 	 h += c0;  */
87 	  c0 += h;
88 	  __asm_reassociation_barrier("" : "+r"(h) : "r"(c0));
89 	  h = h * 32 + c0;
90 	  return h;
91 	}
92 
93       /* Ideal computational order is:
94 	 c1 += c0;
95 	 h *= 33 * 33;
96 	 c0 *= 32;
97 	 c1 += c0;
98 	 h  += c1;  */
99       c1 += c0;
100       __asm_reassociation_barrier("" : "+r"(c1), "+r"(c0));
101       h *= 33 * 33;
102       c1 += c0 * 32;
103       __asm_reassociation_barrier("" : "+r"(c1));
104       h += c1;
105       s += 2;
106     }
107 }
108 
109 #endif /* dl-new-hash.h */
110