1 /* SPDX-License-Identifier: GPL-2.0 */ <<
2 /* 1 /*
3 * strlen.S (c) 1995 David Mosberger (davidm@c 2 * strlen.S (c) 1995 David Mosberger (davidm@cs.arizona.edu)
4 * 3 *
5 * Finds length of a 0-terminated string. Opt 4 * Finds length of a 0-terminated string. Optimized for the
6 * Alpha architecture: 5 * Alpha architecture:
7 * 6 *
8 * - memory accessed as aligned quadwords 7 * - memory accessed as aligned quadwords only
9 * - uses bcmpge to compare 8 bytes in pa 8 * - uses bcmpge to compare 8 bytes in parallel
10 * - does binary search to find 0 byte in 9 * - does binary search to find 0 byte in last
11 * quadword (HAKMEM needed 12 instructi 10 * quadword (HAKMEM needed 12 instructions to
12 * do this instead of the 9 instruction 11 * do this instead of the 9 instructions that
13 * binary search needs). 12 * binary search needs).
14 */ 13 */
15 #include <linux/export.h> !! 14 #include <asm/export.h>
16 .set noreorder 15 .set noreorder
17 .set noat 16 .set noat
18 17
19 .align 3 18 .align 3
20 19
21 .globl strlen 20 .globl strlen
22 .ent strlen 21 .ent strlen
23 22
24 strlen: 23 strlen:
25 ldq_u $1, 0($16) # load first q 24 ldq_u $1, 0($16) # load first quadword ($16 may be misaligned)
26 lda $2, -1($31) 25 lda $2, -1($31)
27 insqh $2, $16, $2 26 insqh $2, $16, $2
28 andnot $16, 7, $0 27 andnot $16, 7, $0
29 or $2, $1, $1 28 or $2, $1, $1
30 cmpbge $31, $1, $2 # $2 <- bitma 29 cmpbge $31, $1, $2 # $2 <- bitmask: bit i == 1 <==> i-th byte == 0
31 bne $2, found 30 bne $2, found
32 31
33 loop: ldq $1, 8($0) 32 loop: ldq $1, 8($0)
34 addq $0, 8, $0 # addr += 8 33 addq $0, 8, $0 # addr += 8
35 nop # helps dual i 34 nop # helps dual issue last two insns
36 cmpbge $31, $1, $2 35 cmpbge $31, $1, $2
37 beq $2, loop 36 beq $2, loop
38 37
39 found: blbs $2, done # make aligned 38 found: blbs $2, done # make aligned case fast
40 negq $2, $3 39 negq $2, $3
41 and $2, $3, $2 40 and $2, $3, $2
42 41
43 and $2, 0x0f, $1 42 and $2, 0x0f, $1
44 addq $0, 4, $3 43 addq $0, 4, $3
45 cmoveq $1, $3, $0 44 cmoveq $1, $3, $0
46 45
47 and $2, 0x33, $1 46 and $2, 0x33, $1
48 addq $0, 2, $3 47 addq $0, 2, $3
49 cmoveq $1, $3, $0 48 cmoveq $1, $3, $0
50 49
51 and $2, 0x55, $1 50 and $2, 0x55, $1
52 addq $0, 1, $3 51 addq $0, 1, $3
53 cmoveq $1, $3, $0 52 cmoveq $1, $3, $0
54 53
55 done: subq $0, $16, $0 54 done: subq $0, $16, $0
56 ret $31, ($26) 55 ret $31, ($26)
57 56
58 .end strlen 57 .end strlen
59 EXPORT_SYMBOL(strlen) 58 EXPORT_SYMBOL(strlen)
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.