1 /* Optimized memrchr implementation for PowerPC64/POWER7 using cmpb insn.
2 Copyright (C) 2010-2019 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 <https://www.gnu.org/licenses/>. */
22 /* int [r3] memrchr (char *s [r3], int byte [r4], int size [r5]) */
25 # define MEMRCHR __memrchr
28 ENTRY_TOCLESS (MEMRCHR)
30 add r7,r3,r5 /* Calculate the last acceptable address. */
36 dcbt r9,r6,8 /* Stream hint, decreasing addresses. */
38 /* Replicate BYTE to doubleword. */
44 rlwinm r0,r0,3,26,28 /* Calculate padding. */
51 #ifdef __LITTLE_ENDIAN__
54 ldbrx r12,0,r8 /* Load reversed doubleword from memory. */
56 cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */
58 cmpldi cr7,r3,0 /* If r3 == 0, no BYTEs have been found. */
62 /* Are we now aligned to a quadword boundary? If so, skip to
63 the main loop. Otherwise, go through the alignment code. */
66 /* Handle DWORD2 of pair. */
67 #ifdef __LITTLE_ENDIAN__
78 /* The last dword we want to read in the loop below is the one
79 containing the first byte of the string, ie. the dword at
80 s & ~7, or r0. The first dword read is at r8 - 8, we
81 read 2 * cnt dwords, so the last dword read will be at
82 r8 - 8 - 16 * cnt + 8. Solving for cnt gives
83 cnt = (r8 - r0) / 16 */
86 srdi r9,r5,4 /* Number of loop iterations. */
87 mtctr r9 /* Setup the counter. */
89 /* Main loop to look for BYTE backwards in the string.
90 FIXME: Investigate whether 32 byte align helps with this
91 9 instruction loop. */
94 /* Load two doublewords, compare and merge in a
95 single register for speed. This is an attempt
96 to speed up the byte-checking process for bigger strings. */
98 #ifdef __LITTLE_ENDIAN__
107 or r5,r9,r3 /* Merge everything in one doubleword. */
113 /* We may have one more word to read. */
117 #ifdef __LITTLE_ENDIAN__
129 /* OK, one (or both) of the dwords contains BYTE. Check
134 /* BYTE must be in the second word. Adjust the address
135 again and move the result of cmpb to r3 so we can calculate the
141 /* r3 has the output of the cmpb instruction, that is, it contains
142 0xff in the same position as BYTE in the original
143 word from the string. Use that to calculate the pointer.
144 We need to make sure BYTE is *before* the end of the
147 cntlzd r9,r3 /* Count leading zeros before the match. */
148 cmpld r8,r0 /* Are we on the last word? */
149 srdi r6,r9,3 /* Convert leading zeros to bytes. */
163 /* Deals with size <= 32. */
169 #ifdef __LITTLE_ENDIAN__
172 ldbrx r12,0,r8 /* Load reversed doubleword from memory. */
174 cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */
179 /* Are we done already? */
186 #ifdef __LITTLE_ENDIAN__
200 weak_alias (__memrchr, memrchr)
201 libc_hidden_builtin_def (memrchr)