1 /* Optimized memrchr 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.
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.
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.
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/>. */
21 /* int [r3] memrchr (char *s [r3], int byte [r4], int size [r5]) */
24 # define MEMRCHR __memrchr
27 ENTRY_TOCLESS (MEMRCHR)
29 add r7,r3,r5 /* Calculate the last acceptable address. */
35 dcbt r9,r6,8 /* Stream hint, decreasing addresses. */
37 /* Replicate BYTE to doubleword. */
43 rlwinm r0,r0,3,26,28 /* Calculate padding. */
50 #ifdef __LITTLE_ENDIAN__
53 ldbrx r12,0,r8 /* Load reversed doubleword from memory. */
55 cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */
57 cmpldi cr7,r3,0 /* If r3 == 0, no BYTEs have been found. */
61 /* Are we now aligned to a quadword boundary? If so, skip to
62 the main loop. Otherwise, go through the alignment code. */
65 /* Handle DWORD2 of pair. */
66 #ifdef __LITTLE_ENDIAN__
77 /* The last dword we want to read in the loop below is the one
78 containing the first byte of the string, ie. the dword at
79 s & ~7, or r0. The first dword read is at r8 - 8, we
80 read 2 * cnt dwords, so the last dword read will be at
81 r8 - 8 - 16 * cnt + 8. Solving for cnt gives
82 cnt = (r8 - r0) / 16 */
85 srdi r9,r5,4 /* Number of loop iterations. */
86 mtctr r9 /* Setup the counter. */
88 /* Main loop to look for BYTE backwards in the string.
89 FIXME: Investigate whether 32 byte align helps with this
90 9 instruction loop. */
93 /* Load two doublewords, compare and merge in a
94 single register for speed. This is an attempt
95 to speed up the byte-checking process for bigger strings. */
97 #ifdef __LITTLE_ENDIAN__
106 or r5,r9,r3 /* Merge everything in one doubleword. */
112 /* We may have one more word to read. */
116 #ifdef __LITTLE_ENDIAN__
128 /* OK, one (or both) of the dwords contains BYTE. Check
133 /* BYTE must be in the second word. Adjust the address
134 again and move the result of cmpb to r3 so we can calculate the
140 /* r3 has the output of the cmpb instruction, that is, it contains
141 0xff in the same position as BYTE in the original
142 word from the string. Use that to calculate the pointer.
143 We need to make sure BYTE is *before* the end of the
146 cntlzd r9,r3 /* Count leading zeros before the match. */
147 cmpld r8,r0 /* Are we on the last word? */
148 srdi r6,r9,3 /* Convert leading zeros to bytes. */
162 /* Deals with size <= 32. */
168 #ifdef __LITTLE_ENDIAN__
171 ldbrx r12,0,r8 /* Load reversed doubleword from memory. */
173 cmpb r3,r12,r4 /* Check for BYTE in DWORD1. */
178 /* Are we done already? */
185 #ifdef __LITTLE_ENDIAN__
199 libc_hidden_def (__memrchr)
200 weak_alias (__memrchr, memrchr)
201 libc_hidden_builtin_def (memrchr)