]>
Commit | Line | Data |
---|---|---|
04277e02 | 1 | /* Copyright (C) 1996-2019 Free Software Foundation, Inc. |
db31c863 UD |
2 | This file is part of the GNU C Library. |
3 | ||
4 | The GNU C Library is free software; you can redistribute it and/or | |
3214b89b AJ |
5 | modify it under the terms of the GNU Lesser General Public |
6 | License as published by the Free Software Foundation; either | |
7 | version 2.1 of the License, or (at your option) any later version. | |
db31c863 UD |
8 | |
9 | The GNU C Library is distributed in the hope that it will be useful, | |
10 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
3214b89b | 12 | Lesser General Public License for more details. |
db31c863 | 13 | |
3214b89b | 14 | You should have received a copy of the GNU Lesser General Public |
ab84e3ff PE |
15 | License along with the GNU C Library. If not, see |
16 | <http://www.gnu.org/licenses/>. */ | |
cf182b58 | 17 | |
910e2e14 | 18 | /* Return the address of the last occurrence of a given character |
cf182b58 UD |
19 | within a null-terminated string, or null if it is not found. |
20 | ||
21 | This is generally scheduled for the EV5 (got to look out for my own | |
22 | interests :-), but with EV4 needs in mind. There are, in fact, fewer | |
23 | stalls on the EV4 than there are on the EV5. | |
24 | */ | |
25 | ||
26 | #include <sysdep.h> | |
27 | ||
28 | .set noreorder | |
29 | .set noat | |
30 | ||
31 | ENTRY(strrchr) | |
60c74cf0 UD |
32 | #ifdef PROF |
33 | ldgp gp, 0(pv) | |
34 | lda AT, _mcount | |
35 | jsr AT, (AT), _mcount | |
36 | .prologue 1 | |
37 | #else | |
cf182b58 | 38 | .prologue 0 |
60c74cf0 | 39 | #endif |
cf182b58 | 40 | |
6a0148d1 | 41 | and a1, 0xff, a1 # e0 : zero extend our test character |
cf182b58 UD |
42 | mov zero, t6 # .. e1 : t6 is last match aligned addr |
43 | sll a1, 8, t5 # e0 : replicate our test character | |
44 | mov zero, t7 # .. e1 : t7 is last match byte compare mask | |
45 | or t5, a1, a1 # e0 : | |
46 | ldq_u t0, 0(a0) # .. e1 : load first quadword | |
47 | sll a1, 16, t5 # e0 : | |
48 | andnot a0, 7, v0 # .. e1 : align source addr | |
49 | or t5, a1, a1 # e0 : | |
50 | lda t4, -1 # .. e1 : build garbage mask | |
51 | sll a1, 32, t5 # e0 : | |
52 | cmpbge zero, t0, t1 # .. e1 : bits set iff byte == zero | |
53 | mskqh t4, a0, t4 # e0 : | |
54 | or t5, a1, a1 # .. e1 : character replication complete | |
55 | xor t0, a1, t2 # e0 : make bytes == c zero | |
56 | cmpbge zero, t4, t4 # .. e1 : bits set iff byte is garbage | |
57 | cmpbge zero, t2, t3 # e0 : bits set iff byte == c | |
58 | andnot t1, t4, t1 # .. e1 : clear garbage from null test | |
59 | andnot t3, t4, t3 # e0 : clear garbage from char test | |
60 | bne t1, $eos # .. e1 : did we already hit the terminator? | |
61 | ||
62 | /* Character search main loop */ | |
63 | $loop: | |
64 | ldq t0, 8(v0) # e0 : load next quadword | |
65 | cmovne t3, v0, t6 # .. e1 : save previous comparisons match | |
66 | cmovne t3, t3, t7 # e0 : | |
67 | addq v0, 8, v0 # .. e1 : | |
68 | xor t0, a1, t2 # e0 : | |
69 | cmpbge zero, t0, t1 # .. e1 : bits set iff byte == zero | |
70 | cmpbge zero, t2, t3 # e0 : bits set iff byte == c | |
71 | beq t1, $loop # .. e1 : if we havnt seen a null, loop | |
72 | ||
73 | /* Mask out character matches after terminator */ | |
74 | $eos: | |
75 | negq t1, t4 # e0 : isolate first null byte match | |
76 | and t1, t4, t4 # e1 : | |
77 | subq t4, 1, t5 # e0 : build a mask of the bytes upto... | |
78 | or t4, t5, t4 # e1 : ... and including the null | |
79 | ||
80 | and t3, t4, t3 # e0 : mask out char matches after null | |
81 | cmovne t3, t3, t7 # .. e1 : save it, if match found | |
82 | cmovne t3, v0, t6 # e0 : | |
83 | ||
84 | /* Locate the address of the last matched character */ | |
85 | ||
86 | /* Retain the early exit for the ev4 -- the ev5 mispredict penalty | |
87 | is 5 cycles -- the same as just falling through. */ | |
88 | beq t7, $retnull # .. e1 : | |
89 | ||
90 | and t7, 0xf0, t2 # e0 : binary search for the high bit set | |
91 | cmovne t2, t2, t7 # .. e1 (zdb) | |
92 | cmovne t2, 4, t2 # e0 : | |
93 | and t7, 0xcc, t1 # .. e1 : | |
94 | cmovne t1, t1, t7 # e0 : | |
95 | cmovne t1, 2, t1 # .. e1 : | |
96 | and t7, 0xaa, t0 # e0 : | |
97 | cmovne t0, 1, t0 # .. e1 (zdb) | |
98 | addq t2, t1, t1 # e0 : | |
99 | addq t6, t0, v0 # .. e1 : add our aligned base ptr to the mix | |
100 | addq v0, t1, v0 # e0 : | |
101 | ret # .. e1 : | |
102 | ||
103 | $retnull: | |
104 | mov zero, v0 # e0 : | |
105 | ret # .. e1 : | |
106 | ||
107 | END(strrchr) | |
108 | ||
109 | weak_alias (strrchr, rindex) | |
79b7c863 | 110 | libc_hidden_builtin_def (strrchr) |