tjh.dev
+1
arch/h8300/Kconfig
+1
arch/h8300/Kconfig
+53
arch/h8300/include/asm/hash.h
+53
arch/h8300/include/asm/hash.h
···
1
+
#ifndef _ASM_HASH_H
2
+
#define _ASM_HASH_H
3
+
4
+
/*
5
+
* The later H8SX models have a 32x32-bit multiply, but the H8/300H
6
+
* and H8S have only 16x16->32. Since it's tolerably compact, this is
7
+
* basically an inlined version of the __mulsi3 code. Since the inputs
8
+
* are not expected to be small, it's also simplfied by skipping the
9
+
* early-out checks.
10
+
*
11
+
* (Since neither CPU has any multi-bit shift instructions, a
12
+
* shift-and-add version is a non-starter.)
13
+
*
14
+
* TODO: come up with an arch-specific version of the hashing in fs/namei.c,
15
+
* since that is heavily dependent on rotates. Which, as mentioned, suck
16
+
* horribly on H8.
17
+
*/
18
+
19
+
#if defined(CONFIG_CPU_H300H) || defined(CONFIG_CPU_H8S)
20
+
21
+
#define HAVE_ARCH__HASH_32 1
22
+
23
+
/*
24
+
* Multiply by k = 0x61C88647. Fitting this into three registers requires
25
+
* one extra instruction, but reducing register pressure will probably
26
+
* make that back and then some.
27
+
*
28
+
* GCC asm note: %e1 is the high half of operand %1, while %f1 is the
29
+
* low half. So if %1 is er4, then %e1 is e4 and %f1 is r4.
30
+
*
31
+
* This has been designed to modify x in place, since that's the most
32
+
* common usage, but preserve k, since hash_64() makes two calls in
33
+
* quick succession.
34
+
*/
35
+
static inline u32 __attribute_const__ __hash_32(u32 x)
36
+
{
37
+
u32 temp;
38
+
39
+
asm( "mov.w %e1,%f0"
40
+
"\n mulxu.w %f2,%0" /* klow * xhigh */
41
+
"\n mov.w %f0,%e1" /* The extra instruction */
42
+
"\n mov.w %f1,%f0"
43
+
"\n mulxu.w %e2,%0" /* khigh * xlow */
44
+
"\n add.w %e1,%f0"
45
+
"\n mulxu.w %f2,%1" /* klow * xlow */
46
+
"\n add.w %f0,%e1"
47
+
: "=&r" (temp), "=r" (x)
48
+
: "%r" (GOLDEN_RATIO_32), "1" (x));
49
+
return x;
50
+
}
51
+
52
+
#endif
53
+
#endif /* _ASM_HASH_H */