1
/* Optimized strchr implementation for PowerPC64/POWER7 using cmpb insn.
2
Copyright (C) 2010 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, write to the Free
18
Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA
25
/* int [r3] strchr (char *s [r3], int c [r4]) */
27
ENTRY (BP_SYM(strchr))
30
clrrdi r8,r3,3 /* Align the address to doubleword boundary. */
32
ld r12,0(r8) /* Load doubleword from memory. */
33
li r0,0 /* Doubleword with null chars to use
36
rlwinm r6,r3,3,26,28 /* Calculate padding. */
40
/* Replicate byte to doubleword. */
45
/* Now r4 has a doubleword of c bytes and r0 has
46
a doubleword of null bytes. */
48
cmpb r10,r12,r4 /* Compare each byte against c byte. */
49
cmpb r11,r12,r0 /* Compare each byte against null byte. */
51
/* Move the doublewords left and right to discard the bits that are
52
not part of the string and bring them back as zeros. */
58
or r5,r10,r11 /* OR the results to speed things up. */
59
cmpdi cr7,r5,0 /* If r5 == 0, no c or null bytes
65
/* Are we now aligned to a doubleword boundary? If so, skip to
66
the main loop. Otherwise, go through the alignment code. */
70
/* Handle WORD2 of pair. */
77
b L(loop) /* We branch here (rather than falling through)
78
to skip the nops due to heavy alignment
83
/* Load two doublewords, compare and merge in a
84
single register for speed. This is an attempt
85
to speed up the null-checking process for bigger strings. */
98
/* OK, one (or both) of the doublewords contains a c/null byte. Check
99
the first doubleword and decrement the address in case the first
100
doubleword really contains a c/null byte. */
106
/* The c/null byte must be in the second doubleword. Adjust the
107
address again and move the result of cmpb to r10 so we can calculate
114
/* r5 has the output of the cmpb instruction, that is, it contains
115
0xff in the same position as the c/null byte in the original
116
doubleword from the string. Use that to calculate the pointer. */
118
cntlzd r4,r10 /* Count leading zeroes before c matches. */
119
cntlzd r0,r11 /* Count leading zeroes before null matches. */
122
srdi r0,r4,3 /* Convert leading zeroes to bytes. */
123
add r3,r8,r0 /* Return address of the matching c byte
124
or null in case c was not found. */
132
/* We are here because strchr was called with a null byte. */
135
/* r0 has a doubleword of null bytes. */
137
cmpb r5,r12,r0 /* Compare each byte against null bytes. */
139
/* Move the doublewords left and right to discard the bits that are
140
not part of the string and bring them back as zeros. */
144
cmpdi cr7,r5,0 /* If r10 == 0, no c or null bytes
150
/* Are we now aligned to a quadword boundary? If so, skip to
151
the main loop. Otherwise, go through the alignment code. */
155
/* Handle WORD2 of pair. */
160
b L(loop_null) /* We branch here (rather than falling through)
161
to skip the nops due to heavy alignment
162
of the loop below. */
164
/* Main loop to look for the end of the string. Since it's a
165
small loop (< 8 instructions), align it to 32-bytes. */
168
/* Load two doublewords, compare and merge in a
169
single register for speed. This is an attempt
170
to speed up the null-checking process for bigger strings. */
179
/* OK, one (or both) of the doublewords contains a null byte. Check
180
the first doubleword and decrement the address in case the first
181
doubleword really contains a null byte. */
187
/* The null byte must be in the second doubleword. Adjust the address
188
again and move the result of cmpb to r10 so we can calculate the
194
/* r5 has the output of the cmpb instruction, that is, it contains
195
0xff in the same position as the null byte in the original
196
doubleword from the string. Use that to calculate the pointer. */
198
cntlzd r0,r5 /* Count leading zeros before the match. */
199
srdi r0,r0,3 /* Convert leading zeros to bytes. */
200
add r3,r8,r0 /* Return address of the matching null byte. */
202
END (BP_SYM (strchr))
203
weak_alias (BP_SYM (strchr), BP_SYM (index))
204
libc_hidden_builtin_def (strchr)
added
removed
sysdeps/powerpc/powerpc64/power7/strchr.S
