]>
Commit | Line | Data |
---|---|---|
fe2f79db LM |
1 | /* Optimized memchr 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. | |
5 | ||
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. | |
10 | ||
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. | |
15 | ||
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 | |
19 | 02110-1301 USA. */ | |
20 | ||
21 | #include <sysdep.h> | |
22 | #include <bp-sym.h> | |
23 | #include <bp-asm.h> | |
24 | ||
25 | /* int [r3] memchr (char *s [r3], int byte [r4], int size [r5]) */ | |
26 | .machine power7 | |
27 | ENTRY (BP_SYM (__memchr)) | |
28 | CALL_MCOUNT 2 | |
29 | dcbt 0,r3 | |
30 | clrrdi r8,r3,3 | |
31 | rlwimi r4,r4,8,16,23 | |
32 | rlwimi r4,r4,16,0,15 | |
33 | insrdi r4,r4,32,0 | |
34 | add r7,r3,r5 /* Calculate the last acceptable address. */ | |
35 | cmpldi r5,32 | |
36 | ble L(small_range) | |
37 | ||
38 | cmpld cr7,r3,r7 /* Is the address equal or less than r3? If | |
39 | it's equal or less, it means size is either 0 | |
40 | or a negative number. */ | |
41 | ble cr7,L(proceed) | |
42 | ||
43 | li r7,-1 /* Make r11 the biggest if r4 <= 0. */ | |
44 | L(proceed): | |
45 | rlwinm r6,r3,3,26,28 /* Calculate padding. */ | |
46 | ld r12,0(r8) /* Load doubleword from memory. */ | |
47 | cmpb r10,r12,r4 /* Check for BYTE's in DWORD1. */ | |
48 | sld r10,r10,r6 | |
49 | srd r10,r10,r6 | |
50 | cmpldi cr7,r10,0 /* If r10 == 0, no BYTE's's have been found. */ | |
51 | bne cr7,L(done) | |
52 | ||
53 | /* Are we done already? */ | |
54 | addi r9,r8,8 | |
55 | cmpld cr6,r9,r7 | |
56 | bge cr6,L(null) | |
57 | ||
58 | mtcrf 0x01,r8 | |
59 | /* Are we now aligned to a quadword boundary? If so, skip to | |
60 | the main loop. Otherwise, go through the alignment code. */ | |
61 | ||
62 | bt 28,L(loop_setup) | |
63 | ||
64 | /* Handle DWORD2 of pair. */ | |
65 | ldu r12,8(r8) | |
66 | cmpb r10,r12,r4 | |
67 | cmpldi cr7,r10,0 | |
68 | bne cr7,L(done) | |
69 | ||
70 | /* Are we done already? */ | |
71 | addi r9,r8,8 | |
72 | cmpld cr6,r9,r7 | |
73 | bge cr6,L(null) | |
74 | ||
75 | L(loop_setup): | |
76 | sub r5,r7,r9 | |
77 | srdi r6,r5,4 /* Number of loop iterations. */ | |
78 | mtctr r6 /* Setup the counter. */ | |
79 | b L(loop) | |
80 | /* Main loop to look for BYTE backwards in the string. Since | |
81 | it's a small loop (< 8 instructions), align it to 32-bytes. */ | |
82 | .p2align 5 | |
83 | L(loop): | |
84 | /* Load two doublewords, compare and merge in a | |
85 | single register for speed. This is an attempt | |
86 | to speed up the byte-checking process for bigger strings. */ | |
87 | ||
88 | ld r12,8(r8) | |
89 | ldu r11,16(r8) | |
90 | cmpb r10,r12,r4 | |
91 | cmpb r9,r11,r4 | |
92 | or r5,r9,r10 /* Merge everything in one doubleword. */ | |
93 | cmpldi cr7,r5,0 | |
94 | bne cr7,L(found) | |
95 | bdnz L(loop) | |
96 | /* We're here because the counter reached 0, and that means we | |
97 | didn't have any matches for BYTE in the whole range. Just return | |
98 | the original range. */ | |
99 | addi r9,r8,8 | |
100 | cmpld cr6,r9,r7 | |
101 | blt cr6,L(loop_small) | |
102 | b L(null) | |
103 | ||
104 | /* OK, one (or both) of the doublewords contains BYTE. Check | |
105 | the first doubleword and decrement the address in case the first | |
106 | doubleword really contains BYTE. */ | |
107 | .align 4 | |
108 | L(found): | |
109 | cmpldi cr6,r10,0 | |
110 | addi r8,r8,-8 | |
111 | bne cr6,L(done) | |
112 | ||
113 | /* BYTE must be in the second doubleword. Adjust the address | |
114 | again and move the result of cmpb to r10 so we can calculate the | |
115 | pointer. */ | |
116 | ||
117 | mr r10,r9 | |
118 | addi r8,r8,8 | |
119 | ||
120 | /* r10 has the output of the cmpb instruction, that is, it contains | |
121 | 0xff in the same position as BYTE in the original | |
122 | doubleword from the string. Use that to calculate the pointer. | |
123 | We need to make sure BYTE is *before* the end of the | |
124 | range. */ | |
125 | L(done): | |
126 | cntlzd r0,r10 /* Count leading zeroes before the match. */ | |
127 | srdi r0,r0,3 /* Convert leading zeroes to bytes. */ | |
128 | add r3,r8,r0 | |
129 | cmpld r3,r7 | |
130 | bge L(null) | |
131 | blr | |
132 | ||
133 | .align 4 | |
134 | L(null): | |
135 | li r3,0 | |
136 | blr | |
137 | ||
138 | /* Deals with size <= 32. */ | |
139 | .align 4 | |
140 | L(small_range): | |
141 | cmpldi r5,0 | |
142 | beq L(null) | |
143 | ||
144 | rlwinm r6,r3,3,26,28 /* Calculate padding. */ | |
145 | ld r12,0(r8) /* Load word from memory. */ | |
146 | cmpb r10,r12,r4 /* Check for BYTE in DWORD1. */ | |
147 | sld r10,r10,r6 | |
148 | srd r10,r10,r6 | |
149 | cmpldi cr7,r10,0 | |
150 | bne cr7,L(done) | |
151 | ||
152 | /* Are we done already? */ | |
153 | addi r9,r8,8 | |
154 | cmpld r9,r7 | |
155 | bge L(null) | |
156 | b L(loop_small) | |
157 | ||
158 | .p2align 5 | |
159 | L(loop_small): | |
160 | ldu r12,8(r8) | |
161 | cmpb r10,r12,r4 | |
162 | addi r9,r8,8 | |
163 | cmpldi cr6,r10,0 | |
164 | bne cr6,L(done) | |
165 | cmpld r9,r7 | |
166 | bge L(null) | |
167 | b L(loop_small) | |
168 | ||
169 | END (BP_SYM (__memchr)) | |
170 | weak_alias (BP_SYM (__memchr), BP_SYM(memchr)) | |
171 | libc_hidden_builtin_def (memchr) |