1 /* Optimized strchr implementation for PowerPC64/POWER7 using cmpb insn.
2 Copyright (C) 2010-2014 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] strchr (char *s [r3], int c [r4]) */
27 clrrdi r8,r3,3 /* Align the address to doubleword boundary. */
29 ld r12,0(r8) /* Load doubleword from memory. */
30 li r0,0 /* Doubleword with null chars to use
33 rlwinm r6,r3,3,26,28 /* Calculate padding. */
37 /* Replicate byte to doubleword. */
42 /* Now r4 has a doubleword of c bytes and r0 has
43 a doubleword of null bytes. */
45 cmpb r10,r12,r4 /* Compare each byte against c byte. */
46 cmpb r11,r12,r0 /* Compare each byte against null byte. */
48 /* Move the doublewords left and right to discard the bits that are
49 not part of the string and bring them back as zeros. */
50 #ifdef __LITTLE_ENDIAN__
61 or r5,r10,r11 /* OR the results to speed things up. */
62 cmpdi cr7,r5,0 /* If r5 == 0, no c or null bytes
68 /* Are we now aligned to a doubleword boundary? If so, skip to
69 the main loop. Otherwise, go through the alignment code. */
73 /* Handle WORD2 of pair. */
80 b L(loop) /* We branch here (rather than falling through)
81 to skip the nops due to heavy alignment
86 /* Load two doublewords, compare and merge in a
87 single register for speed. This is an attempt
88 to speed up the null-checking process for bigger strings. */
101 /* OK, one (or both) of the doublewords contains a c/null byte. Check
102 the first doubleword and decrement the address in case the first
103 doubleword really contains a c/null byte. */
109 /* The c/null byte must be in the second doubleword. Adjust the
110 address again and move the result of cmpb to r10 so we can calculate
117 /* r10/r11 have the output of the cmpb instructions, that is,
118 0xff in the same position as the c/null byte in the original
119 doubleword from the string. Use that to calculate the pointer. */
121 #ifdef __LITTLE_ENDIAN__
130 cntlzd r0,r10 /* Count leading zeros before c matches. */
134 srdi r0,r0,3 /* Convert leading zeros to bytes. */
135 add r3,r8,r0 /* Return address of the matching c byte
136 or null in case c was not found. */
144 /* We are here because strchr was called with a null byte. */
147 /* r0 has a doubleword of null bytes. */
149 cmpb r5,r12,r0 /* Compare each byte against null bytes. */
151 /* Move the doublewords left and right to discard the bits that are
152 not part of the string and bring them back as zeros. */
153 #ifdef __LITTLE_ENDIAN__
160 cmpdi cr7,r5,0 /* If r10 == 0, no c or null bytes
166 /* Are we now aligned to a quadword boundary? If so, skip to
167 the main loop. Otherwise, go through the alignment code. */
171 /* Handle WORD2 of pair. */
176 b L(loop_null) /* We branch here (rather than falling through)
177 to skip the nops due to heavy alignment
178 of the loop below. */
180 /* Main loop to look for the end of the string. Since it's a
181 small loop (< 8 instructions), align it to 32-bytes. */
184 /* Load two doublewords, compare and merge in a
185 single register for speed. This is an attempt
186 to speed up the null-checking process for bigger strings. */
195 /* OK, one (or both) of the doublewords contains a null byte. Check
196 the first doubleword and decrement the address in case the first
197 doubleword really contains a null byte. */
203 /* The null byte must be in the second doubleword. Adjust the address
204 again and move the result of cmpb to r10 so we can calculate the
210 /* r5 has the output of the cmpb instruction, that is, it contains
211 0xff in the same position as the null byte in the original
212 doubleword from the string. Use that to calculate the pointer. */
214 #ifdef __LITTLE_ENDIAN__
219 cntlzd r0,r5 /* Count leading zeros before the match. */
221 srdi r0,r0,3 /* Convert leading zeros to bytes. */
222 add r3,r8,r0 /* Return address of the matching null byte. */
225 weak_alias (strchr, index)
226 libc_hidden_builtin_def (strchr)