1 | /* a-memset.s -- memset, optimised for fido asm |
---|
2 | * |
---|
3 | * Copyright (c) 2007 mocom software GmbH & Co KG) |
---|
4 | * |
---|
5 | * The authors hereby grant permission to use, copy, modify, distribute, |
---|
6 | * and license this software and its documentation for any purpose, provided |
---|
7 | * that existing copyright notices are retained in all copies and that this |
---|
8 | * notice is included verbatim in any distributions. No written agreement, |
---|
9 | * license, or royalty fee is required for any of the authorized uses. |
---|
10 | * Modifications to this software may be copyrighted by their authors |
---|
11 | * and need not follow the licensing terms described here, provided that |
---|
12 | * the new terms are clearly indicated on the first page of each file where |
---|
13 | * they apply. |
---|
14 | */ |
---|
15 | |
---|
16 | #include "m68kasm.h" |
---|
17 | |
---|
18 | .text |
---|
19 | .align 4 |
---|
20 | |
---|
21 | .globl SYM(memset) |
---|
22 | .type SYM(memset), @function |
---|
23 | |
---|
24 | | memset, optimised |
---|
25 | | |
---|
26 | | strategy: |
---|
27 | | - no argument testing (the original memcpy from the GNU lib does |
---|
28 | | no checking either) |
---|
29 | | - make sure the destination pointer (the write pointer) is long word |
---|
30 | | aligned. This is the best you can do, because writing to unaligned |
---|
31 | | addresses can be the most costfull thing one could do. |
---|
32 | | - we fill long word wise if possible |
---|
33 | | |
---|
34 | | VG, 2006 |
---|
35 | | |
---|
36 | | bugfixes: |
---|
37 | | - distribution of byte value improved - in cases someone gives |
---|
38 | | non-byte value |
---|
39 | | - residue byte transfer was not working |
---|
40 | | |
---|
41 | | VG, April 2007 |
---|
42 | | |
---|
43 | SYM(memset): |
---|
44 | move.l 4(sp),a0 | dest ptr |
---|
45 | move.l 8(sp),d0 | value |
---|
46 | move.l 12(sp),d1 | len |
---|
47 | cmp.l #16,d1 |
---|
48 | blo .Lbset | below, byte fills |
---|
49 | | |
---|
50 | move.l d2,-(sp) | need a register |
---|
51 | move.b d0,d2 | distribute low byte to all byte in word |
---|
52 | lsl.l #8,d0 |
---|
53 | move.b d2,d0 |
---|
54 | move.w d0,d2 |
---|
55 | swap d0 | rotate 16 |
---|
56 | move.w d2,d0 |
---|
57 | | |
---|
58 | move.l a0,d2 | copy of src |
---|
59 | neg.l d2 | 1 2 3 ==> 3 2 1 |
---|
60 | and.l #3,d2 |
---|
61 | beq 2f | is aligned |
---|
62 | | |
---|
63 | sub.l d2,d1 | fix length |
---|
64 | lsr.l #1,d2 | word align needed? |
---|
65 | bcc 1f |
---|
66 | move.b d0,(a0)+ | fill byte |
---|
67 | 1: |
---|
68 | lsr.l #1,d2 | long align needed? |
---|
69 | bcc 2f |
---|
70 | move.w d0,(a0)+ | fill word |
---|
71 | 2: |
---|
72 | move.l d1,d2 | number of long transfers (at least 3) |
---|
73 | lsr.l #2,d2 |
---|
74 | subq.l #1,d2 |
---|
75 | |
---|
76 | 1: |
---|
77 | move.l d0,(a0)+ | fill long words |
---|
78 | .Llset: |
---|
79 | #if !defined (__mcoldfire__) |
---|
80 | dbra d2,1b | loop until done |
---|
81 | sub.l #0x10000,d2 |
---|
82 | #else |
---|
83 | subq.l #1,d2 |
---|
84 | #endif |
---|
85 | bpl 1b |
---|
86 | and.l #3,d1 | residue byte transfers, fixed |
---|
87 | move.l (sp)+,d2 | restore d2 |
---|
88 | bra .Lbset |
---|
89 | |
---|
90 | 1: |
---|
91 | move.b d0,(a0)+ | fill residue bytes |
---|
92 | .Lbset: |
---|
93 | #if !defined (__mcoldfire__) |
---|
94 | dbra d1,1b | loop until done |
---|
95 | #else |
---|
96 | subq.l #1,d1 |
---|
97 | bpl 1b |
---|
98 | #endif |
---|
99 | move.l 4(sp),d0 | return value |
---|
100 | rts |
---|