1 /* Optimized memchr implementation for PowerPC64/POWER7 using cmpb insn.
2 Copyright (C) 2010-2015 Free Software Foundation, Inc.
3 Contributed by Luis Machado <luisgpm@br.ibm.com>.
4 This file is part of the GNU C Library.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, see
18 <http://www.gnu.org/licenses/>. */
22 /* int [r3] memchr (char *s [r3], int byte [r4], int size [r5]) */
29 add r7,r3,r5 /* Calculate the last acceptable address. */
33 rlwinm r6,r3,3,26,28 /* Calculate padding. */
36 #ifdef __LITTLE_ENDIAN__
43 ld r12,0(r8) /* Load doubleword from memory. */
44 cmpb r3,r12,r4 /* Check for BYTEs in DWORD1. */
46 clrldi r5,r7,61 /* Byte count - 1 in last dword. */
47 clrrdi r7,r7,3 /* Address of last doubleword. */
48 cmpldi cr7,r3,0 /* Does r3 indicate we got a hit? */
52 /* Are we now aligned to a quadword boundary? If so, skip to
53 the main loop. Otherwise, go through the alignment code. */
56 /* Handle DWORD2 of pair. */
63 /* The last dword we want to read in the loop below is the one
64 containing the last byte of the string, ie. the dword at
65 (s + size - 1) & ~7, or r7. The first dword read is at
66 r8 + 8, we read 2 * cnt dwords, so the last dword read will
67 be at r8 + 8 + 16 * cnt - 8. Solving for cnt gives
68 cnt = (r7 - r8) / 16 */
70 srdi r6,r6,4 /* Number of loop iterations. */
71 mtctr r6 /* Setup the counter. */
73 /* Main loop to look for BYTE in the string. Since
74 it's a small loop (8 instructions), align it to 32-bytes. */
77 /* Load two doublewords, compare and merge in a
78 single register for speed. This is an attempt
79 to speed up the byte-checking process for bigger strings. */
84 or r6,r9,r3 /* Merge everything in one doubleword. */
89 /* We may have one more dword to read. */
101 /* OK, one (or both) of the doublewords contains BYTE. Check
102 the first doubleword and decrement the address in case the first
103 doubleword really contains BYTE. */
108 /* BYTE must be in the second doubleword. Adjust the address
109 again and move the result of cmpb to r3 so we can calculate the
115 /* r3 has the output of the cmpb instruction, that is, it contains
116 0xff in the same position as BYTE in the original
117 doubleword from the string. Use that to calculate the pointer.
118 We need to make sure BYTE is *before* the end of the range. */
120 #ifdef __LITTLE_ENDIAN__
123 popcntd r0,r0 /* Count trailing zeros. */
125 cntlzd r0,r3 /* Count leading zeros before the match. */
127 cmpld r8,r7 /* Are we on the last dword? */
128 srdi r0,r0,3 /* Convert leading/trailing zeros to bytes. */
130 cmpld cr7,r0,r5 /* If on the last dword, check byte offset. */
141 /* Deals with size <= 32. */
146 ld r12,0(r8) /* Load word from memory. */
147 cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */
150 clrldi r5,r7,61 /* Byte count - 1 in last dword. */
151 clrrdi r7,r7,3 /* Address of last doubleword. */
152 cmpld r8,r7 /* Are we done already? */
160 bne cr6,L(done) /* Found something. */
161 beqlr /* Hit end of string (length). */
184 weak_alias (__memchr, memchr)
185 libc_hidden_builtin_def (memchr)