1 /* Optimized strrchr implementation for PowerPC64/POWER7 using cmpb insn.
2 Copyright (C) 2014-2024 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] strrchr (char *s [r3], int c [r4]) */
24 # define STRRCHR strrchr
28 ENTRY_TOCLESS (STRRCHR)
31 clrrdi r8,r3,3 /* Align the address to doubleword boundary. */
33 ld r12,0(r8) /* Load doubleword from memory. */
34 li r9,0 /* used to store last occurrence */
35 li r0,0 /* Doubleword with null chars to use
38 rlwinm r6,r3,3,26,28 /* Calculate padding. */
42 /* Replicate byte to doubleword. */
47 /* r4 is changed now ,if its passed as more chars
48 check for null again */
51 /* Now r4 has a doubleword of c bytes and r0 has
52 a doubleword of null bytes. */
54 cmpb r10,r12,r4 /* Compare each byte against c byte. */
55 cmpb r11,r12,r0 /* Compare each byte against null byte. */
57 /* Move the doublewords left and right to discard the bits that are
58 not part of the string and bring them back as zeros. */
59 #ifdef __LITTLE_ENDIAN__
70 or r5,r10,r11 /* OR the results to speed things up. */
71 cmpdi cr7,r5,0 /* If r5 == 0, no c or null bytes
78 /* Are we now aligned to a doubleword boundary? If so, skip to
79 the main loop. Otherwise, go through the alignment code. */
83 /* Handle WORD2 of pair. */
90 b L(loop) /* We branch here (rather than falling through)
91 to skip the nops due to heavy alignment
95 /* Load two doublewords, compare and merge in a
96 single register for speed. This is an attempt
97 to speed up the null-checking process for bigger strings. */
110 /* OK, one (or both) of the doublewords contains a c/null byte. Check
111 the first doubleword and decrement the address in case the first
112 doubleword really contains a c/null byte. */
117 /* The c/null byte must be in the second doubleword. Adjust the
118 address again and move the result of cmpb to r10 so we can calculate
125 /* r10/r11 have the output of the cmpb instructions, that is,
126 0xff in the same position as the c/null byte in the original
127 doubleword from the string. Use that to calculate the pointer. */
130 /* if there are more than one 0xff in r11, find the first pos of ff
131 in r11 and fill r10 with 0 from that position */
134 #ifdef __LITTLE_ENDIAN__
143 #ifdef __LITTLE_ENDIAN__
150 #ifdef __LITTLE_ENDIAN__
151 cntlzd r0,r10 /* Count leading zeros before c matches. */
159 addi r3,r10,-1 /* Count trailing zeros before c matches. */
165 srdi r0,r0,3 /* Convert trailing zeros to bytes. */
167 add r9,r8,r0 /* Return address of the matching c byte
168 or null in case c was not found. */
170 cmpdi cr7,r11,0 /* If r11 == 0, no null's have been found. */
178 /* We are here because strrchr was called with a null byte. */
181 /* r0 has a doubleword of null bytes. */
183 cmpb r5,r12,r0 /* Compare each byte against null bytes. */
185 /* Move the doublewords left and right to discard the bits that are
186 not part of the string and bring them back as zeros. */
187 #ifdef __LITTLE_ENDIAN__
194 cmpdi cr7,r5,0 /* If r10 == 0, no c or null bytes
200 /* Are we now aligned to a quadword boundary? If so, skip to
201 the main loop. Otherwise, go through the alignment code. */
205 /* Handle WORD2 of pair. */
210 b L(loop_null) /* We branch here (rather than falling through)
211 to skip the nops due to heavy alignment
212 of the loop below. */
214 /* Main loop to look for the end of the string. Since it's a
215 small loop (< 8 instructions), align it to 32-bytes. */
218 /* Load two doublewords, compare and merge in a
219 single register for speed. This is an attempt
220 to speed up the null-checking process for bigger strings. */
229 /* OK, one (or both) of the doublewords contains a null byte. Check
230 the first doubleword and decrement the address in case the first
231 doubleword really contains a null byte. */
237 /* The null byte must be in the second doubleword. Adjust the address
238 again and move the result of cmpb to r10 so we can calculate the
244 /* r5 has the output of the cmpb instruction, that is, it contains
245 0xff in the same position as the null byte in the original
246 doubleword from the string. Use that to calculate the pointer. */
248 #ifdef __LITTLE_ENDIAN__
253 cntlzd r0,r5 /* Count leading zeros before the match. */
255 srdi r0,r0,3 /* Convert trailing zeros to bytes. */
256 add r3,r8,r0 /* Return address of the matching null byte. */
259 weak_alias (strrchr, rindex)
260 libc_hidden_builtin_def (strrchr)