tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
sh: Optimised memset for SH4
Optimised version of memset for the SH4 which uses movca.l.
Signed-off-by: Stuart Menefy <stuart.menefy@st.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
authored by
Stuart Menefy
and committed by
Paul Mundt
dfc34940
a0458b07
+112
-2
+5
-2
arch/sh/lib/Makefile
+5
-2
arch/sh/lib/Makefile
···
2
2
# Makefile for SuperH-specific library files..
3
3
#
4
4
5
-
lib-y = delay.o memset.o memmove.o memchr.o \
5
+
lib-y = delay.o memmove.o memchr.o \
6
6
checksum.o strlen.o div64.o div64-generic.o
7
7
8
8
# Extracted from libgcc
···
23
23
memcpy-y := memcpy.o
24
24
memcpy-$(CONFIG_CPU_SH4) := memcpy-sh4.o
25
25
26
+
memset-y := memset.o
27
+
memset-$(CONFIG_CPU_SH4) := memset-sh4.o
28
+
26
29
lib-$(CONFIG_MMU) += copy_page.o __clear_user.o
27
30
lib-$(CONFIG_MCOUNT) += mcount.o
28
-
lib-y += $(memcpy-y) $(udivsi3-y)
31
+
lib-y += $(memcpy-y) $(memset-y) $(udivsi3-y)
29
32
30
33
EXTRA_CFLAGS += -Werror
+107
arch/sh/lib/memset-sh4.S
+107
arch/sh/lib/memset-sh4.S
···
1
+
/*
2
+
* "memset" implementation for SH4
3
+
*
4
+
* Copyright (C) 1999 Niibe Yutaka
5
+
* Copyright (c) 2009 STMicroelectronics Limited
6
+
* Author: Stuart Menefy <stuart.menefy:st.com>
7
+
*/
8
+
9
+
/*
10
+
* void *memset(void *s, int c, size_t n);
11
+
*/
12
+
13
+
#include <linux/linkage.h>
14
+
15
+
ENTRY(memset)
16
+
mov #12,r0
17
+
add r6,r4
18
+
cmp/gt r6,r0
19
+
bt/s 40f ! if it's too small, set a byte at once
20
+
mov r4,r0
21
+
and #3,r0
22
+
cmp/eq #0,r0
23
+
bt/s 2f ! It's aligned
24
+
sub r0,r6
25
+
1:
26
+
dt r0
27
+
bf/s 1b
28
+
mov.b r5,@-r4
29
+
2: ! make VVVV
30
+
extu.b r5,r5
31
+
swap.b r5,r0 ! V0
32
+
or r0,r5 ! VV
33
+
swap.w r5,r0 ! VV00
34
+
or r0,r5 ! VVVV
35
+
36
+
! Check if enough bytes need to be copied to be worth the big loop
37
+
mov #0x40, r0 ! (MT)
38
+
cmp/gt r6,r0 ! (MT) 64 > len => slow loop
39
+
40
+
bt/s 22f
41
+
mov r6,r0
42
+
43
+
! align the dst to the cache block size if necessary
44
+
mov r4, r3
45
+
mov #~(0x1f), r1
46
+
47
+
and r3, r1
48
+
cmp/eq r3, r1
49
+
50
+
bt/s 11f ! dst is already aligned
51
+
sub r1, r3 ! r3-r1 -> r3
52
+
shlr2 r3 ! number of loops
53
+
54
+
10: mov.l r5,@-r4
55
+
dt r3
56
+
bf/s 10b
57
+
add #-4, r6
58
+
59
+
11: ! dst is 32byte aligned
60
+
mov r6,r2
61
+
mov #-5,r0
62
+
shld r0,r2 ! number of loops
63
+
64
+
add #-32, r4
65
+
mov r5, r0
66
+
12:
67
+
movca.l r0,@r4
68
+
mov.l r5,@(4, r4)
69
+
mov.l r5,@(8, r4)
70
+
mov.l r5,@(12,r4)
71
+
mov.l r5,@(16,r4)
72
+
mov.l r5,@(20,r4)
73
+
add #-0x20, r6
74
+
mov.l r5,@(24,r4)
75
+
dt r2
76
+
mov.l r5,@(28,r4)
77
+
bf/s 12b
78
+
add #-32, r4
79
+
80
+
add #32, r4
81
+
mov #8, r0
82
+
cmp/ge r0, r6
83
+
bf 40f
84
+
85
+
mov r6,r0
86
+
22:
87
+
shlr2 r0
88
+
shlr r0 ! r0 = r6 >> 3
89
+
3:
90
+
dt r0
91
+
mov.l r5,@-r4 ! set 8-byte at once
92
+
bf/s 3b
93
+
mov.l r5,@-r4
94
+
!
95
+
mov #7,r0
96
+
and r0,r6
97
+
98
+
! fill bytes (length may be zero)
99
+
40: tst r6,r6
100
+
bt 5f
101
+
4:
102
+
dt r6
103
+
bf/s 4b
104
+
mov.b r5,@-r4
105
+
5:
106
+
rts
107
+
mov r4,r0