|
96
by Marcus Shawcroft
Add AArch64 optimized strlen. |
1 |
/* Copyright (c) 2013, Linaro Limited |
2 |
All rights reserved. |
|
3 |
||
4 |
Redistribution and use in source and binary forms, with or without |
|
5 |
modification, are permitted provided that the following conditions are met: |
|
6 |
* Redistributions of source code must retain the above copyright |
|
7 |
notice, this list of conditions and the following disclaimer. |
|
8 |
* Redistributions in binary form must reproduce the above copyright |
|
9 |
notice, this list of conditions and the following disclaimer in the |
|
10 |
documentation and/or other materials provided with the distribution. |
|
11 |
* Neither the name of the Linaro nor the |
|
12 |
names of its contributors may be used to endorse or promote products |
|
13 |
derived from this software without specific prior written permission. |
|
14 |
||
15 |
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
|
16 |
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
|
17 |
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
|
18 |
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
|
19 |
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
|
20 |
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
|
21 |
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
|
22 |
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
|
23 |
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
|
24 |
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
|
25 |
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ |
|
26 |
||
27 |
/* Assumptions: |
|
28 |
*
|
|
29 |
* ARMv8-a, AArch64 |
|
30 |
*/
|
|
31 |
||
32 |
/* Arguments and results. */ |
|
33 |
#define srcin x0
|
|
34 |
#define len x0
|
|
35 |
||
36 |
/* Locals and temporaries. */ |
|
37 |
#define src x1
|
|
38 |
#define data1 x2
|
|
39 |
#define data2 x3
|
|
40 |
#define data2a x4
|
|
41 |
#define has_nul1 x5
|
|
42 |
#define has_nul2 x6
|
|
43 |
#define tmp1 x7
|
|
44 |
#define tmp2 x8
|
|
45 |
#define tmp3 x9
|
|
46 |
#define tmp4 x10
|
|
47 |
#define zeroones x11
|
|
48 |
#define pos x12
|
|
49 |
||
50 |
.macro def_fn f p2align=0 |
|
51 |
.text
|
|
52 |
.p2align \p2align |
|
53 |
.global \f |
|
54 |
.type \f, %function |
|
55 |
\f: |
|
56 |
.endm
|
|
57 |
||
58 |
#define REP8_01 0x0101010101010101
|
|
59 |
#define REP8_7f 0x7f7f7f7f7f7f7f7f
|
|
60 |
#define REP8_80 0x8080808080808080
|
|
61 |
||
62 |
/* Start of critial section -- keep to one 64Byte cache line. */ |
|
63 |
def_fn strlen p2align=6 |
|
64 |
mov zeroones, #REP8_01 |
|
65 |
bic src, srcin, #15 |
|
66 |
ands tmp1, srcin, #15 |
|
67 |
b.ne .Lmisaligned |
|
68 |
/* NUL detection works on the principle that (X - 1) & (~X) & 0x80 |
|
69 |
(=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and |
|
70 |
can be done in parallel across the entire word. */ |
|
71 |
/* The inner loop deals with two Dwords at a time. This has a |
|
72 |
slightly higher start-up cost, but we should win quite quickly, |
|
73 |
especially on cores with a high number of issue slots per |
|
74 |
cycle, as we get much better parallelism out of the operations. */ |
|
75 |
.Lloop: |
|
76 |
ldp data1, data2, [src], #16 |
|
77 |
.Lrealigned: |
|
78 |
sub tmp1, data1, zeroones |
|
79 |
orr tmp2, data1, #REP8_7f |
|
80 |
sub tmp3, data2, zeroones |
|
81 |
orr tmp4, data2, #REP8_7f |
|
82 |
bic has_nul1, tmp1, tmp2 |
|
83 |
bics has_nul2, tmp3, tmp4 |
|
84 |
ccmp has_nul1, #0, #0, eq /* NZCV = 0000 */ |
|
85 |
b.eq .Lloop |
|
86 |
/* End of critical section -- keep to one 64Byte cache line. */ |
|
87 |
||
88 |
sub len, src, srcin |
|
89 |
cbz has_nul1, .Lnul_in_data2 |
|
90 |
#ifdef __AARCH64EB__
|
|
91 |
mov data2, data1 |
|
92 |
#endif
|
|
93 |
sub len, len, #8 |
|
94 |
mov has_nul2, has_nul1 |
|
95 |
.Lnul_in_data2: |
|
96 |
#ifdef __AARCH64EB__
|
|
97 |
/* For big-endian, carry propagation (if the final byte in the |
|
98 |
string is 0x01) means we cannot use has_nul directly. The |
|
99 |
easiest way to get the correct byte is to byte-swap the data |
|
100 |
and calculate the syndrome a second time. */ |
|
101 |
rev data2, data2 |
|
102 |
sub tmp1, data2, zeroones |
|
103 |
orr tmp2, data2, #REP8_7f |
|
104 |
bic has_nul2, tmp1, tmp2 |
|
105 |
#endif
|
|
106 |
sub len, len, #8 |
|
107 |
rev has_nul2, has_nul2 |
|
108 |
clz pos, has_nul2 |
|
109 |
add len, len, pos, lsr #3 /* Bits to bytes. */ |
|
110 |
ret
|
|
111 |
||
112 |
.Lmisaligned: |
|
113 |
cmp tmp1, #8 |
|
114 |
neg tmp1, tmp1 |
|
115 |
ldp data1, data2, [src], #16 |
|
116 |
lsl tmp1, tmp1, #3 /* Bytes beyond alignment -> bits. */ |
|
117 |
mov tmp2, #~0 |
|
118 |
#ifdef __AARCH64EB__
|
|
119 |
/* Big-endian. Early bytes are at MSB. */ |
|
120 |
lsl tmp2, tmp2, tmp1 /* Shift (tmp1 & 63). */ |
|
121 |
#else
|
|
122 |
/* Little-endian. Early bytes are at LSB. */ |
|
123 |
lsr tmp2, tmp2, tmp1 /* Shift (tmp1 & 63). */ |
|
124 |
#endif
|
|
125 |
orr data1, data1, tmp2 |
|
126 |
orr data2a, data2, tmp2 |
|
127 |
csinv data1, data1, xzr, le |
|
128 |
csel data2, data2, data2a, le |
|
129 |
b .Lrealigned |
|
130 |
||
131 |
.size strlen, . - strlen |