]>
Commit | Line | Data |
---|---|---|
734aad71 A |
1 | /* |
2 | * Copyright (c) 2002 Apple Computer, Inc. All rights reserved. | |
3 | * | |
4 | * @APPLE_LICENSE_HEADER_START@ | |
5 | * | |
6 | * Copyright (c) 1999-2003 Apple Computer, Inc. All Rights Reserved. | |
7 | * | |
8 | * This file contains Original Code and/or Modifications of Original Code | |
9 | * as defined in and that are subject to the Apple Public Source License | |
10 | * Version 2.0 (the 'License'). You may not use this file except in | |
11 | * compliance with the License. Please obtain a copy of the License at | |
12 | * http://www.opensource.apple.com/apsl/ and read it before using this | |
13 | * file. | |
14 | * | |
15 | * The Original Code and all software distributed under the License are | |
16 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER | |
17 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, | |
18 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, | |
19 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. | |
20 | * Please see the License for the specific language governing rights and | |
21 | * limitations under the License. | |
22 | * | |
23 | * @APPLE_LICENSE_HEADER_END@ | |
24 | */ | |
25 | #define ASSEMBLER | |
26 | #include <mach/ppc/asm.h> | |
27 | #undef ASSEMBLER | |
28 | ||
29 | // *************** | |
30 | // * S T R C P Y * | |
31 | // *************** | |
32 | // | |
33 | // char* strcpy(const char *dst, const char *src); | |
34 | // | |
35 | // We optimize the move by doing it word parallel. This introduces | |
36 | // a complication: if we blindly did word load/stores until finding | |
37 | // a 0, we might get a spurious page fault by touching bytes past it. | |
38 | // To avoid this, we never do a "lwz" that crosses a page boundary, | |
39 | // and never store a byte we don't have to. | |
40 | // | |
41 | // The test for 0s relies on the following inobvious but very efficient | |
42 | // word-parallel test: | |
43 | // x = dataWord + 0xFEFEFEFF | |
44 | // y = ~dataWord & 0x80808080 | |
45 | // if (x & y) == 0 then no zero found | |
46 | // The test maps any non-zero byte to zero, and any zero byte to 0x80, | |
47 | // with one exception: 0x01 bytes preceeding the first zero are also | |
48 | // mapped to 0x80. | |
49 | // | |
50 | // We align the _source_, which allows us to avoid all worries about | |
51 | // spurious page faults. Doing so is faster than aligning the dest. | |
52 | ||
53 | .text | |
54 | .globl EXT(strcpy) | |
55 | ||
56 | .align 5 | |
57 | LEXT(strcpy) // char* strcpy(const char *dst, const char *src); | |
58 | andi. r0,r4,3 // is source aligned? | |
59 | dcbt 0,r4 // touch in source | |
60 | lis r6,hi16(0xFEFEFEFF) // start to load magic constants | |
61 | lis r7,hi16(0x80808080) | |
62 | dcbtst 0,r3 // touch in dst | |
63 | ori r6,r6,lo16(0xFEFEFEFF) | |
64 | ori r7,r7,lo16(0x80808080) | |
65 | mr r9,r3 // use r9 for dest ptr (must return r3 intact) | |
66 | beq LwordloopEnter // source is aligned | |
67 | subfic r0,r0,4 // r0 <- #bytes to word align source | |
68 | mtctr r0 | |
69 | ||
70 | // Loop over bytes. | |
71 | // r4 = source ptr (unaligned) | |
72 | // r6 = 0xFEFEFEFF | |
73 | // r7 = 0x80808080 | |
74 | // r9 = dest ptr (unaligned) | |
75 | // ctr = byte count | |
76 | ||
77 | Lbyteloop: | |
78 | lbz r8,0(r4) // r8 <- next source byte | |
79 | addi r4,r4,1 | |
80 | cmpwi r8,0 // 0 ? | |
81 | stb r8,0(r9) // pack into dest | |
82 | addi r9,r9,1 | |
83 | bdnzf eq,Lbyteloop // loop until (ctr==0) | (r8==0) | |
84 | ||
85 | bne LwordloopEnter // 0-byte not found, so enter word loop | |
86 | blr // 0-byte found, done | |
87 | ||
88 | // Word loop: move a word at a time until 0-byte found. | |
89 | // r4 = source ptr (word aligned) | |
90 | // r6 = 0xFEFEFEFF | |
91 | // r7 = 0x80808080 | |
92 | // r9 = dest ptr (unaligned) | |
93 | ||
94 | .align 5 // align inner loop, which is 8 words ling | |
95 | Lwordloop: | |
96 | stw r8,0(r9) // pack word into destination | |
97 | addi r9,r9,4 | |
98 | LwordloopEnter: | |
99 | lwz r8,0(r4) // r8 <- next 4 source bytes | |
100 | addi r4,r4,4 | |
101 | add r10,r8,r6 // r10 <- word + 0xFEFEFEFF | |
102 | andc r12,r7,r8 // r12 <- ~word & 0x80808080 | |
103 | and. r0,r10,r12 // r0 <- nonzero iff word has a 0-byte | |
104 | beq Lwordloop // loop if ctr!=0 and cr0_eq | |
105 | ||
106 | // Found a 0-byte. Store last word up to and including the 0, a byte at a time. | |
107 | // r8 = last word, known to have a 0-byte | |
108 | // r9 = dest ptr | |
109 | ||
110 | Lstorelastbytes: | |
111 | srwi. r0,r8,24 // right justify next byte and test for 0 | |
112 | slwi r8,r8,8 // shift next byte into position | |
113 | stb r0,0(r9) // pack into dest | |
114 | addi r9,r9,1 | |
115 | bne Lstorelastbytes // loop until 0 stored | |
116 | ||
117 | blr | |
118 |