1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _ASM_HASH_H
3#define _ASM_HASH_H
4
5/*
6 * The later H8SX models have a 32x32-bit multiply, but the H8/300H
7 * and H8S have only 16x16->32.  Since it's tolerably compact, this is
8 * basically an inlined version of the __mulsi3 code.  Since the inputs
9 * are not expected to be small, it's also simplfied by skipping the
10 * early-out checks.
11 *
12 * (Since neither CPU has any multi-bit shift instructions, a
13 * shift-and-add version is a non-starter.)
14 *
15 * TODO: come up with an arch-specific version of the hashing in fs/namei.c,
16 * since that is heavily dependent on rotates.  Which, as mentioned, suck
17 * horribly on H8.
18 */
19
20#if defined(CONFIG_CPU_H300H) || defined(CONFIG_CPU_H8S)
21
22#define HAVE_ARCH__HASH_32 1
23
24/*
25 * Multiply by k = 0x61C88647.  Fitting this into three registers requires
26 * one extra instruction, but reducing register pressure will probably
27 * make that back and then some.
28 *
29 * GCC asm note: %e1 is the high half of operand %1, while %f1 is the
30 * low half.  So if %1 is er4, then %e1 is e4 and %f1 is r4.
31 *
32 * This has been designed to modify x in place, since that's the most
33 * common usage, but preserve k, since hash_64() makes two calls in
34 * quick succession.
35 */
36static inline u32 __attribute_const__ __hash_32(u32 x)
37{
38	u32 temp;
39
40	asm(   "mov.w	%e1,%f0"
41	"\n	mulxu.w	%f2,%0"		/* klow * xhigh */
42	"\n	mov.w	%f0,%e1"	/* The extra instruction */
43	"\n	mov.w	%f1,%f0"
44	"\n	mulxu.w	%e2,%0"		/* khigh * xlow */
45	"\n	add.w	%e1,%f0"
46	"\n	mulxu.w	%f2,%1"		/* klow * xlow */
47	"\n	add.w	%f0,%e1"
48	: "=&r" (temp), "=r" (x)
49	: "%r" (GOLDEN_RATIO_32), "1" (x));
50	return x;
51}
52
53#endif
54#endif /* _ASM_HASH_H */
55