1 /* Optimized memchr implementation for PowerPC64/POWER7 using cmpb insn.
2 Copyright (C) 2010-2018 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]) */
25 # define MEMCHR __memchr
28 ENTRY_TOCLESS (MEMCHR)
34 /* Calculate the last acceptable address and check for possible
35 addition overflow by using satured math:
47 rlwinm r6,r3,3,26,28 /* Calculate padding. */
50 #ifdef __LITTLE_ENDIAN__
57 ld r12,0(r8) /* Load doubleword from memory. */
58 cmpb r3,r12,r4 /* Check for BYTEs in DWORD1. */
60 clrldi r5,r7,61 /* Byte count - 1 in last dword. */
61 clrrdi r7,r7,3 /* Address of last doubleword. */
62 cmpldi cr7,r3,0 /* Does r3 indicate we got a hit? */
66 /* Are we now aligned to a quadword boundary? If so, skip to
67 the main loop. Otherwise, go through the alignment code. */
70 /* Handle DWORD2 of pair. */
77 /* The last dword we want to read in the loop below is the one
78 containing the last byte of the string, ie. the dword at
79 (s + size - 1) & ~7, or r7. The first dword read is at
80 r8 + 8, we read 2 * cnt dwords, so the last dword read will
81 be at r8 + 8 + 16 * cnt - 8. Solving for cnt gives
82 cnt = (r7 - r8) / 16 */
84 srdi r6,r6,4 /* Number of loop iterations. */
85 mtctr r6 /* Setup the counter. */
87 /* Main loop to look for BYTE in the string. Since
88 it's a small loop (8 instructions), align it to 32-bytes. */
91 /* Load two doublewords, compare and merge in a
92 single register for speed. This is an attempt
93 to speed up the byte-checking process for bigger strings. */
98 or r6,r9,r3 /* Merge everything in one doubleword. */
103 /* We may have one more dword to read. */
115 /* OK, one (or both) of the doublewords contains BYTE. Check
116 the first doubleword and decrement the address in case the first
117 doubleword really contains BYTE. */
122 /* BYTE must be in the second doubleword. Adjust the address
123 again and move the result of cmpb to r3 so we can calculate the
129 /* r3 has the output of the cmpb instruction, that is, it contains
130 0xff in the same position as BYTE in the original
131 doubleword from the string. Use that to calculate the pointer.
132 We need to make sure BYTE is *before* the end of the range. */
134 #ifdef __LITTLE_ENDIAN__
137 popcntd r0,r0 /* Count trailing zeros. */
139 cntlzd r0,r3 /* Count leading zeros before the match. */
141 cmpld r8,r7 /* Are we on the last dword? */
142 srdi r0,r0,3 /* Convert leading/trailing zeros to bytes. */
144 cmpld cr7,r0,r5 /* If on the last dword, check byte offset. */
155 /* Deals with size <= 32. */
160 ld r12,0(r8) /* Load word from memory. */
161 cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */
164 clrldi r5,r7,61 /* Byte count - 1 in last dword. */
165 clrrdi r7,r7,3 /* Address of last doubleword. */
166 cmpld r8,r7 /* Are we done already? */
174 bne cr6,L(done) /* Found something. */
175 beqlr /* Hit end of string (length). */
198 weak_alias (__memchr, memchr)
199 libc_hidden_builtin_def (memchr)