1 | /* Optimized rawmemchr implementation for PowerPC64/POWER7 using cmpb insn. |
2 | Copyright (C) 2010-2024 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 | #include <sysdep.h> |
20 | |
21 | /* int [r3] rawmemchr (void *s [r3], int c [r4]) */ |
22 | |
23 | #ifndef RAWMEMCHR |
24 | # define RAWMEMCHR __rawmemchr |
25 | #endif |
26 | .machine power7 |
27 | ENTRY_TOCLESS (RAWMEMCHR) |
28 | CALL_MCOUNT 2 |
29 | dcbt 0,r3 |
30 | clrrdi r8,r3,3 /* Align the address to doubleword boundary. */ |
31 | |
32 | /* Replicate byte to doubleword. */ |
33 | insrdi r4,r4,8,48 |
34 | insrdi r4,r4,16,32 |
35 | insrdi r4,r4,32,0 |
36 | |
37 | /* Now r4 has a doubleword of c bytes. */ |
38 | |
39 | rlwinm r6,r3,3,26,28 /* Calculate padding. */ |
40 | ld r12,0(r8) /* Load doubleword from memory. */ |
41 | cmpb r5,r12,r4 /* Compare each byte against c byte. */ |
42 | #ifdef __LITTLE_ENDIAN__ |
43 | srd r5,r5,r6 |
44 | sld r5,r5,r6 |
45 | #else |
46 | sld r5,r5,r6 /* Move left to discard ignored bits. */ |
47 | srd r5,r5,r6 /* Bring the bits back as zeros. */ |
48 | #endif |
49 | cmpdi cr7,r5,0 /* If r5 == 0, no c bytes have been found. */ |
50 | bne cr7,L(done) |
51 | |
52 | mtcrf 0x01,r8 |
53 | |
54 | /* Are we now aligned to a quadword boundary? If so, skip to |
55 | the main loop. Otherwise, go through the alignment code. */ |
56 | |
57 | bt 28,L(loop) |
58 | |
59 | /* Handle DWORD2 of pair. */ |
60 | ldu r12,8(r8) |
61 | cmpb r5,r12,r4 |
62 | cmpdi cr7,r5,0 |
63 | bne cr7,L(done) |
64 | b L(loop) /* We branch here (rather than falling through) |
65 | to skip the nops due to heavy alignment |
66 | of the loop below. */ |
67 | |
68 | /* Main loop to look for the end of the string. Since it's a |
69 | small loop (< 8 instructions), align it to 32-bytes. */ |
70 | .p2align 5 |
71 | L(loop): |
72 | /* Load two doublewords, compare and merge in a |
73 | single register for speed. This is an attempt |
74 | to speed up the byte-checking process for bigger strings. */ |
75 | ld r12,8(r8) |
76 | ldu r11,16(r8) |
77 | cmpb r5,r12,r4 |
78 | cmpb r6,r11,r4 |
79 | or r7,r5,r6 |
80 | cmpdi cr7,r7,0 |
81 | beq cr7,L(loop) |
82 | |
83 | /* OK, one (or both) of the doublewords contains a 'c' byte. Check |
84 | the first doubleword and decrement the address in case the first |
85 | doubleword really contains a c byte. */ |
86 | |
87 | cmpdi cr6,r5,0 |
88 | addi r8,r8,-8 |
89 | bne cr6,L(done) |
90 | |
91 | /* The 'c' byte must be in the second doubleword. Adjust the address |
92 | again and move the result of cmpb to r10 so we can calculate the |
93 | pointer. */ |
94 | mr r5,r6 |
95 | addi r8,r8,8 |
96 | |
97 | /* r5 has the output of the cmpb instruction, that is, it contains |
98 | 0xff in the same position as the 'c' byte in the original |
99 | doubleword from the string. Use that fact to find out what is |
100 | the position of the byte inside the string. */ |
101 | L(done): |
102 | #ifdef __LITTLE_ENDIAN__ |
103 | addi r0,r5,-1 |
104 | andc r0,r0,r5 |
105 | popcntd r0,r0 /* Count trailing zeros. */ |
106 | #else |
107 | cntlzd r0,r5 /* Count leading zeros before the match. */ |
108 | #endif |
109 | srdi r0,r0,3 /* Convert leading zeros to bytes. */ |
110 | add r3,r8,r0 /* Return address of the matching char. */ |
111 | blr |
112 | END (RAWMEMCHR) |
113 | weak_alias (__rawmemchr,rawmemchr) |
114 | libc_hidden_builtin_def (__rawmemchr) |
115 | |