]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
14c44b95 GS |
2 | #ifndef _ASM_HASH_H |
3 | #define _ASM_HASH_H | |
4 | ||
5 | /* | |
6 | * If CONFIG_M68000=y (original mc68000/010), this file is #included | |
7 | * to work around the lack of a MULU.L instruction. | |
8 | */ | |
9 | ||
10 | #define HAVE_ARCH__HASH_32 1 | |
11 | /* | |
12 | * While it would be legal to substitute a different hash operation | |
13 | * entirely, let's keep it simple and just use an optimized multiply | |
14 | * by GOLDEN_RATIO_32 = 0x61C88647. | |
15 | * | |
16 | * The best way to do that appears to be to multiply by 0x8647 with | |
17 | * shifts and adds, and use mulu.w to multiply the high half by 0x61C8. | |
18 | * | |
19 | * Because the 68000 has multi-cycle shifts, this addition chain is | |
20 | * chosen to minimise the shift distances. | |
21 | * | |
22 | * Despite every attempt to spoon-feed it simple operations, GCC | |
23 | * 6.1.1 doggedly insists on doing annoying things like converting | |
24 | * "lsl.l #2,<reg>" (12 cycles) to two adds (8+8 cycles). | |
25 | * | |
26 | * It also likes to notice two shifts in a row, like "a = x << 2" and | |
27 | * "a <<= 7", and convert that to "a = x << 9". But shifts longer | |
28 | * than 8 bits are extra-slow on m68k, so that's a lose. | |
29 | * | |
30 | * Since the 68000 is a very simple in-order processor with no | |
31 | * instruction scheduling effects on execution time, we can safely | |
32 | * take it out of GCC's hands and write one big asm() block. | |
33 | * | |
34 | * Without calling overhead, this operation is 30 bytes (14 instructions | |
35 | * plus one immediate constant) and 166 cycles. | |
36 | * | |
37 | * (Because %2 is fetched twice, it can't be postincrement, and thus it | |
38 | * can't be a fully general "g" or "m". Register is preferred, but | |
39 | * offsettable memory or immediate will work.) | |
40 | */ | |
41 | static inline u32 __attribute_const__ __hash_32(u32 x) | |
42 | { | |
43 | u32 a, b; | |
44 | ||
45 | asm( "move.l %2,%0" /* a = x * 0x0001 */ | |
46 | "\n lsl.l #2,%0" /* a = x * 0x0004 */ | |
47 | "\n move.l %0,%1" | |
48 | "\n lsl.l #7,%0" /* a = x * 0x0200 */ | |
49 | "\n add.l %2,%0" /* a = x * 0x0201 */ | |
50 | "\n add.l %0,%1" /* b = x * 0x0205 */ | |
51 | "\n add.l %0,%0" /* a = x * 0x0402 */ | |
52 | "\n add.l %0,%1" /* b = x * 0x0607 */ | |
53 | "\n lsl.l #5,%0" /* a = x * 0x8040 */ | |
54 | : "=&d,d" (a), "=&r,r" (b) | |
55 | : "r,roi?" (x)); /* a+b = x*0x8647 */ | |
56 | ||
57 | return ((u16)(x*0x61c8) << 16) + a + b; | |
58 | } | |
59 | ||
60 | #endif /* _ASM_HASH_H */ |