xref: /linux/arch/m68k/include/asm/hash.h (revision 8c13415c8a4383447c21ec832b20b3b283f0e01a)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_HASH_H
3 #define _ASM_HASH_H
4 
5 #include <linux/wordpart.h>
6 
7 /*
8  * If CONFIG_M68000=y (original mc68000/010), this file is #included
9  * to work around the lack of a MULU.L instruction.
10  */
11 
12 #define HAVE_ARCH__HASH_32 1
13 /*
14  * While it would be legal to substitute a different hash operation
15  * entirely, let's keep it simple and just use an optimized multiply
16  * by GOLDEN_RATIO_32 = 0x61C88647.
17  *
18  * The best way to do that appears to be to multiply by 0x8647 with
19  * shifts and adds, and use mulu.w to multiply the high half by 0x61C8.
20  *
21  * Because the 68000 has multi-cycle shifts, this addition chain is
22  * chosen to minimise the shift distances.
23  *
24  * Despite every attempt to spoon-feed it simple operations, GCC
25  * 6.1.1 doggedly insists on doing annoying things like converting
26  * "lsl.l #2,<reg>" (12 cycles) to two adds (8+8 cycles).
27  *
28  * It also likes to notice two shifts in a row, like "a = x << 2" and
29  * "a <<= 7", and convert that to "a = x << 9".  But shifts longer
30  * than 8 bits are extra-slow on m68k, so that's a lose.
31  *
32  * Since the 68000 is a very simple in-order processor with no
33  * instruction scheduling effects on execution time, we can safely
34  * take it out of GCC's hands and write one big asm() block.
35  *
36  * Without calling overhead, this operation is 30 bytes (14 instructions
37  * plus one immediate constant) and 166 cycles.
38  *
39  * (Because %2 is fetched twice, it can't be postincrement, and thus it
40  * can't be a fully general "g" or "m".  Register is preferred, but
41  * offsettable memory or immediate will work.)
42  */
43 static inline u32 __attribute_const__ __hash_32(u32 x)
44 {
45 	u32 a, b;
46 
47 	asm(   "move.l %2,%0"	/* a = x * 0x0001 */
48 	"\n	lsl.l #2,%0"	/* a = x * 0x0004 */
49 	"\n	move.l %0,%1"
50 	"\n	lsl.l #7,%0"	/* a = x * 0x0200 */
51 	"\n	add.l %2,%0"	/* a = x * 0x0201 */
52 	"\n	add.l %0,%1"	/* b = x * 0x0205 */
53 	"\n	add.l %0,%0"	/* a = x * 0x0402 */
54 	"\n	add.l %0,%1"	/* b = x * 0x0607 */
55 	"\n	lsl.l #5,%0"	/* a = x * 0x8040 */
56 	: "=&d,d" (a), "=&r,r" (b)
57 	: "r,roi?" (x));	/* a+b = x*0x8647 */
58 
59 	return (lower_16_bits(x * 0x61c8) << 16) + a + b;
60 }
61 
62 #endif	/* _ASM_HASH_H */
63