|
|
1.1 root 1: /*
2: * Copyright (c) 1999 Apple Computer, Inc. All rights reserved.
3: *
4: * @APPLE_LICENSE_HEADER_START@
5: *
6: * Portions Copyright (c) 1999 Apple Computer, Inc. All Rights
7: * Reserved. This file contains Original Code and/or Modifications of
8: * Original Code as defined in and that are subject to the Apple Public
9: * Source License Version 1.1 (the "License"). You may not use this file
10: * except in compliance with the License. Please obtain a copy of the
11: * License at http://www.apple.com/publicsource and read it before using
12: * this file.
13: *
14: * The Original Code and all software distributed under the License are
15: * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
16: * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
17: * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
18: * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
19: * License for the specific language governing rights and limitations
20: * under the License.
21: *
22: * @APPLE_LICENSE_HEADER_END@
23: */
24:
25: /* Copyright (c) 1991 NeXT Computer, Inc. All rights reserved.
26: *
27: * File: libc/i386/ansi/memcpy.c
28: * Author: Bruce Martin, NeXT Computer, Inc.
29: *
30: * This file contains machine dependent code for block copies
31: * on NeXT i386-based products. Currently tuned for the i486.
32: *
33: * HISTORY
34: * 12-Aug-92 Bruce Martin ([email protected])
35: * Created.
36: */
37:
38:
39: /*
40: * 486 Factoids:
41: * - if (move <= 16 bytes) use unrolled mem-mem mov instructions
42: * - if (move > 16 bytes) use "rep movs" [this avoids prefetch stalls]
43: * - Aligned read is better than aligned write. Both aligned is even better.
44: */
45:
46:
47: #define MIN_TO_ALIGN 16
48:
49: /*
50: * Convert a void * into a unsigned int so arithmetic can be done
51: * (Technically, the gnu C compiler is tolerant of arithmetic
52: * on void *, but ansi isn't; so we do this.)
53: */
54: #define UNS(voidp) ((unsigned int)(voidp))
55:
56: /*
57: * Number of bytes addr is past an integer alignment
58: */
59: #define BYTES_PAST_INT_ALIGNMENT(voidp) (UNS(voidp) & (sizeof(int) - 1))
60:
61:
62: #define CHAR(voidp, offset) (*(char *)(UNS(voidp) + (offset)))
63:
64: #if defined(NX_CURRENT_COMPILER_RELEASE) && (NX_CURRENT_COMPILER_RELEASE < 320)
65: #define SIREG "e"
66: #else
67: #define SIREG "S"
68: #endif
69:
70:
71: static __inline__ void
72: simple_fwd_char_copy(void *dst, const void *src, int len)
73: {
74: asm("rep; movsb"
75: : /* no outputs */
76: : "&c" (len), "D" (dst), SIREG (src)
77: : "ecx", "esi", "edi");
78: }
79:
80: static __inline__ void
81: simple_fwd_short_copy(void *dst, const void *src, int len)
82: {
83: asm("rep; movsw"
84: : /* no outputs */
85: : "&c" (len>>1), "D" (dst), SIREG (src)
86: : "ecx", "esi", "edi");
87: }
88:
89: static __inline__ void
90: simple_fwd_int_copy(void *dst, const void *src, int len)
91: {
92: asm("rep; movsl"
93: : /* no outputs */
94: : "&c" (len>>2), "D" (dst), SIREG (src)
95: : "ecx", "esi", "edi");
96: }
97:
98:
99: void *memcpy(void *, const void *, unsigned long);
100:
101: void *memcpy(void *dst, const void *src, unsigned long ulen)
102: {
103: int len = ulen;
104: int need_to_move;
105: void *orig_dst = dst;
106:
107: /* Always do a forward copy */
108:
109: if (len < MIN_TO_ALIGN) {
110: simple_fwd_char_copy(dst, src, len);
111: return orig_dst;
112: }
113:
114: if (need_to_move = BYTES_PAST_INT_ALIGNMENT(src)) {
115: need_to_move = sizeof(int) - need_to_move;
116: simple_fwd_char_copy(dst, src, need_to_move);
117: len -= need_to_move;
118: UNS(dst) += need_to_move;
119: UNS(src) += need_to_move;
120: }
121:
122: simple_fwd_int_copy(dst, src, len);
123:
124: if (need_to_move = (len & 3)) {
125: UNS(src) += (len & ~3);
126: UNS(dst) += (len & ~3);
127: switch(need_to_move) {
128: case 3: CHAR(dst, 2) = CHAR(src, 2);
129: case 2: CHAR(dst, 1) = CHAR(src, 1);
130: case 1: CHAR(dst, 0) = CHAR(src, 0);
131: }
132: }
133: return orig_dst;
134: }
135:
136:
137: /*
138: * Exactly the same as memcpy, except for the use of a 16-bit word
139: * move in the inner loop.
140: */
141: void bcopy16(const void *src, void *dst, unsigned long ulen)
142: {
143: int len = ulen;
144: int need_to_move;
145:
146: /* Always do a forward copy */
147:
148: if (len < MIN_TO_ALIGN) {
149: simple_fwd_char_copy(dst, src, len);
150: return;
151: }
152:
153: if (need_to_move = BYTES_PAST_INT_ALIGNMENT(src)) {
154: need_to_move = sizeof(int) - need_to_move;
155: simple_fwd_char_copy(dst, src, need_to_move);
156: len -= need_to_move;
157: UNS(dst) += need_to_move;
158: UNS(src) += need_to_move;
159: }
160:
161: simple_fwd_short_copy(dst, src, len);
162:
163: if (need_to_move = (len & 1)) {
164: UNS(src) += (len & ~1);
165: UNS(dst) += (len & ~1);
166: if (need_to_move) {
167: CHAR(dst, 0) = CHAR(src, 0);
168: }
169: }
170: return;
171: }
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.