1/* memcopy.h -- definitions for memory copy functions.  Generic C version.
2   Copyright (C) 1991, 1992, 1993, 1997, 2004 Free Software Foundation, Inc.
3   This file is part of the GNU C Library.
4   Contributed by Torbjorn Granlund (tege@sics.se).
5
6   The GNU C Library is free software; you can redistribute it and/or
7   modify it under the terms of the GNU Lesser General Public
8   License as published by the Free Software Foundation; either
9   version 2.1 of the License, or (at your option) any later version.
10
11   The GNU C Library is distributed in the hope that it will be useful,
12   but WITHOUT ANY WARRANTY; without even the implied warranty of
13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14   Lesser General Public License for more details.
15
16   You should have received a copy of the GNU Lesser General Public
17   License along with the GNU C Library; if not, write to the Free
18   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
19   02111-1307 USA.  */
20
21/* The strategy of the memory functions is:
22
23     1. Copy bytes until the destination pointer is aligned.
24
25     2. Copy words in unrolled loops.  If the source and destination
26     are not aligned in the same way, use word memory operations,
27     but shift and merge two read words before writing.
28
29     3. Copy the few remaining bytes.
30
31   This is fast on processors that have at least 10 registers for
32   allocation by GCC, and that can access memory at reg+const in one
33   instruction.
34
35   I made an "exhaustive" test of this memmove when I wrote it,
36   exhaustive in the sense that I tried all alignment and length
37   combinations, with and without overlap.  */
38
39#include <sys/cdefs.h>
40#include <endian.h>
41
42/* The macros defined in this file are:
43
44   BYTE_COPY_FWD(dst_beg_ptr, src_beg_ptr, nbytes_to_copy)
45
46   BYTE_COPY_BWD(dst_end_ptr, src_end_ptr, nbytes_to_copy)
47
48   WORD_COPY_FWD(dst_beg_ptr, src_beg_ptr, nbytes_remaining, nbytes_to_copy)
49
50   WORD_COPY_BWD(dst_end_ptr, src_end_ptr, nbytes_remaining, nbytes_to_copy)
51
52   MERGE(old_word, sh_1, new_word, sh_2)
53     [I fail to understand.  I feel stupid.  --roland]
54*/
55
56/* Type to use for aligned memory operations.
57   This should normally be the biggest type supported by a single load
58   and store.  */
59#define	op_t	unsigned long int
60#define OPSIZ	(sizeof(op_t))
61
62/* Type to use for unaligned operations.  */
63typedef unsigned char byte;
64
65/* Optimal type for storing bytes in registers.  */
66#define	reg_char	char
67
68#if __BYTE_ORDER == __LITTLE_ENDIAN
69#define MERGE(w0, sh_1, w1, sh_2) (((w0) >> (sh_1)) | ((w1) << (sh_2)))
70#endif
71#if __BYTE_ORDER == __BIG_ENDIAN
72#define MERGE(w0, sh_1, w1, sh_2) (((w0) << (sh_1)) | ((w1) >> (sh_2)))
73#endif
74
75/* Copy exactly NBYTES bytes from SRC_BP to DST_BP,
76   without any assumptions about alignment of the pointers.  */
77#define BYTE_COPY_FWD(dst_bp, src_bp, nbytes)				      \
78  do									      \
79    {									      \
80      size_t __nbytes = (nbytes);					      \
81      while (__nbytes > 0)						      \
82	{								      \
83	  byte __x = ((byte *) src_bp)[0];				      \
84	  src_bp += 1;							      \
85	  __nbytes -= 1;						      \
86	  ((byte *) dst_bp)[0] = __x;					      \
87	  dst_bp += 1;							      \
88	}								      \
89    } while (0)
90
91/* Copy exactly NBYTES_TO_COPY bytes from SRC_END_PTR to DST_END_PTR,
92   beginning at the bytes right before the pointers and continuing towards
93   smaller addresses.  Don't assume anything about alignment of the
94   pointers.  */
95#define BYTE_COPY_BWD(dst_ep, src_ep, nbytes)				      \
96  do									      \
97    {									      \
98      size_t __nbytes = (nbytes);					      \
99      while (__nbytes > 0)						      \
100	{								      \
101	  byte __x;							      \
102	  src_ep -= 1;							      \
103	  __x = ((byte *) src_ep)[0];					      \
104	  dst_ep -= 1;							      \
105	  __nbytes -= 1;						      \
106	  ((byte *) dst_ep)[0] = __x;					      \
107	}								      \
108    } while (0)
109
110/* Copy *up to* NBYTES bytes from SRC_BP to DST_BP, with
111   the assumption that DST_BP is aligned on an OPSIZ multiple.  If
112   not all bytes could be easily copied, store remaining number of bytes
113   in NBYTES_LEFT, otherwise store 0.  */
114extern void _wordcopy_fwd_aligned (long int, long int, size_t) __THROW;
115extern void _wordcopy_fwd_dest_aligned (long int, long int, size_t) __THROW;
116#define WORD_COPY_FWD(dst_bp, src_bp, nbytes_left, nbytes)		      \
117  do									      \
118    {									      \
119      if (src_bp % OPSIZ == 0)						      \
120	_wordcopy_fwd_aligned (dst_bp, src_bp, (nbytes) / OPSIZ);	      \
121      else								      \
122	_wordcopy_fwd_dest_aligned (dst_bp, src_bp, (nbytes) / OPSIZ);	      \
123      src_bp += (nbytes) & -OPSIZ;					      \
124      dst_bp += (nbytes) & -OPSIZ;					      \
125      (nbytes_left) = (nbytes) % OPSIZ;					      \
126    } while (0)
127
128/* Copy *up to* NBYTES_TO_COPY bytes from SRC_END_PTR to DST_END_PTR,
129   beginning at the words (of type op_t) right before the pointers and
130   continuing towards smaller addresses.  May take advantage of that
131   DST_END_PTR is aligned on an OPSIZ multiple.  If not all bytes could be
132   easily copied, store remaining number of bytes in NBYTES_REMAINING,
133   otherwise store 0.  */
134extern void _wordcopy_bwd_aligned (long int, long int, size_t) __THROW;
135extern void _wordcopy_bwd_dest_aligned (long int, long int, size_t) __THROW;
136#define WORD_COPY_BWD(dst_ep, src_ep, nbytes_left, nbytes)		      \
137  do									      \
138    {									      \
139      if (src_ep % OPSIZ == 0)						      \
140	_wordcopy_bwd_aligned (dst_ep, src_ep, (nbytes) / OPSIZ);	      \
141      else								      \
142	_wordcopy_bwd_dest_aligned (dst_ep, src_ep, (nbytes) / OPSIZ);	      \
143      src_ep -= (nbytes) & -OPSIZ;					      \
144      dst_ep -= (nbytes) & -OPSIZ;					      \
145      (nbytes_left) = (nbytes) % OPSIZ;					      \
146    } while (0)
147
148
149/* Threshold value for when to enter the unrolled loops.  */
150#define	OP_T_THRES	16
151