tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
sh: Optimise memcpy_to/fromio for SH4
Optimise memcpy_to/fromio. This is used extensivly by MTD, so is a
worthwhile performance gain. The main savings come from not repeatedly
calling readl/writel, and doing word instead of byte at a time
transfers. Also using "movca.l" on SH4 gives a small performance win.
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
5e9377ec
8af57f8b
+72
-21
+72
-21
arch/sh/kernel/io.c
+72
-21
arch/sh/kernel/io.c
···
1
1
/*
2
-
* linux/arch/sh/kernel/io.c
2
+
* arch/sh/kernel/io.c - Machine independent I/O functions.
3
3
*
4
-
* Copyright (C) 2000 Stuart Menefy
4
+
* Copyright (C) 2000 - 2009 Stuart Menefy
5
5
* Copyright (C) 2005 Paul Mundt
6
-
*
7
-
* Provide real functions which expand to whatever the header file defined.
8
-
* Also definitions of machine independent IO functions.
9
6
*
10
7
* This file is subject to the terms and conditions of the GNU General Public
11
8
* License. See the file "COPYING" in the main directory of this archive
···
15
18
16
19
/*
17
20
* Copy data from IO memory space to "real" memory space.
18
-
* This needs to be optimized.
19
21
*/
20
22
void memcpy_fromio(void *to, const volatile void __iomem *from, unsigned long count)
21
23
{
22
-
unsigned char *p = to;
23
-
while (count) {
24
-
count--;
25
-
*p = readb(from);
26
-
p++;
27
-
from++;
28
-
}
24
+
/*
25
+
* Would it be worthwhile doing byte and long transfers first
26
+
* to try and get aligned?
27
+
*/
28
+
#ifdef CONFIG_CPU_SH4
29
+
if ((count >= 0x20) &&
30
+
(((u32)to & 0x1f) == 0) && (((u32)from & 0x3) == 0)) {
31
+
int tmp2, tmp3, tmp4, tmp5, tmp6;
32
+
33
+
__asm__ __volatile__(
34
+
"1: \n\t"
35
+
"mov.l @%7+, r0 \n\t"
36
+
"mov.l @%7+, %2 \n\t"
37
+
"movca.l r0, @%0 \n\t"
38
+
"mov.l @%7+, %3 \n\t"
39
+
"mov.l @%7+, %4 \n\t"
40
+
"mov.l @%7+, %5 \n\t"
41
+
"mov.l @%7+, %6 \n\t"
42
+
"mov.l @%7+, r7 \n\t"
43
+
"mov.l @%7+, r0 \n\t"
44
+
"mov.l %2, @(0x04,%0) \n\t"
45
+
"mov #0x20, %2 \n\t"
46
+
"mov.l %3, @(0x08,%0) \n\t"
47
+
"sub %2, %1 \n\t"
48
+
"mov.l %4, @(0x0c,%0) \n\t"
49
+
"cmp/hi %1, %2 ! T if 32 > count \n\t"
50
+
"mov.l %5, @(0x10,%0) \n\t"
51
+
"mov.l %6, @(0x14,%0) \n\t"
52
+
"mov.l r7, @(0x18,%0) \n\t"
53
+
"mov.l r0, @(0x1c,%0) \n\t"
54
+
"bf.s 1b \n\t"
55
+
" add #0x20, %0 \n\t"
56
+
: "=&r" (to), "=&r" (count),
57
+
"=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4),
58
+
"=&r" (tmp5), "=&r" (tmp6), "=&r" (from)
59
+
: "7"(from), "0" (to), "1" (count)
60
+
: "r0", "r7", "t", "memory");
61
+
}
62
+
#endif
63
+
64
+
if ((((u32)to | (u32)from) & 0x3) == 0) {
65
+
for (; count > 3; count -= 4) {
66
+
*(u32 *)to = *(volatile u32 *)from;
67
+
to += 4;
68
+
from += 4;
69
+
}
70
+
}
71
+
72
+
for (; count > 0; count--) {
73
+
*(u8 *)to = *(volatile u8 *)from;
74
+
to++;
75
+
from++;
76
+
}
77
+
78
+
mb();
29
79
}
30
80
EXPORT_SYMBOL(memcpy_fromio);
31
81
32
82
/*
33
83
* Copy data from "real" memory space to IO memory space.
34
-
* This needs to be optimized.
35
84
*/
36
85
void memcpy_toio(volatile void __iomem *to, const void *from, unsigned long count)
37
86
{
38
-
const unsigned char *p = from;
39
-
while (count) {
40
-
count--;
41
-
writeb(*p, to);
42
-
p++;
43
-
to++;
44
-
}
87
+
if ((((u32)to | (u32)from) & 0x3) == 0) {
88
+
for ( ; count > 3; count -= 4) {
89
+
*(volatile u32 *)to = *(u32 *)from;
90
+
to += 4;
91
+
from += 4;
92
+
}
93
+
}
94
+
95
+
for (; count > 0; count--) {
96
+
*(volatile u8 *)to = *(u8 *)from;
97
+
to++;
98
+
from++;
99
+
}
100
+
101
+
mb();
45
102
}
46
103
EXPORT_SYMBOL(memcpy_toio);
47
104