Ley Foon Tan | eea9507 | 2014-11-06 15:20:00 +0800 | [diff] [blame] | 1 | /* Extracted from GLIBC memcpy.c and memcopy.h, which is: |
| 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, see |
| 18 | <http://www.gnu.org/licenses/>. */ |
| 19 | |
| 20 | #include <linux/types.h> |
| 21 | |
| 22 | /* Type to use for aligned memory operations. |
| 23 | This should normally be the biggest type supported by a single load |
| 24 | and store. */ |
| 25 | #define op_t unsigned long int |
| 26 | #define OPSIZ (sizeof(op_t)) |
| 27 | |
| 28 | /* Optimal type for storing bytes in registers. */ |
| 29 | #define reg_char char |
| 30 | |
| 31 | #define MERGE(w0, sh_1, w1, sh_2) (((w0) >> (sh_1)) | ((w1) << (sh_2))) |
| 32 | |
| 33 | /* Copy exactly NBYTES bytes from SRC_BP to DST_BP, |
| 34 | without any assumptions about alignment of the pointers. */ |
| 35 | #define BYTE_COPY_FWD(dst_bp, src_bp, nbytes) \ |
| 36 | do { \ |
| 37 | size_t __nbytes = (nbytes); \ |
| 38 | while (__nbytes > 0) { \ |
| 39 | unsigned char __x = ((unsigned char *) src_bp)[0]; \ |
| 40 | src_bp += 1; \ |
| 41 | __nbytes -= 1; \ |
| 42 | ((unsigned char *) dst_bp)[0] = __x; \ |
| 43 | dst_bp += 1; \ |
| 44 | } \ |
| 45 | } while (0) |
| 46 | |
| 47 | /* Copy *up to* NBYTES bytes from SRC_BP to DST_BP, with |
| 48 | the assumption that DST_BP is aligned on an OPSIZ multiple. If |
| 49 | not all bytes could be easily copied, store remaining number of bytes |
| 50 | in NBYTES_LEFT, otherwise store 0. */ |
| 51 | /* extern void _wordcopy_fwd_aligned __P ((long int, long int, size_t)); */ |
| 52 | /* extern void _wordcopy_fwd_dest_aligned __P ((long int, long int, size_t)); */ |
| 53 | #define WORD_COPY_FWD(dst_bp, src_bp, nbytes_left, nbytes) \ |
| 54 | do { \ |
| 55 | if (src_bp % OPSIZ == 0) \ |
| 56 | _wordcopy_fwd_aligned(dst_bp, src_bp, (nbytes) / OPSIZ);\ |
| 57 | else \ |
| 58 | _wordcopy_fwd_dest_aligned(dst_bp, src_bp, (nbytes) / OPSIZ);\ |
| 59 | src_bp += (nbytes) & -OPSIZ; \ |
| 60 | dst_bp += (nbytes) & -OPSIZ; \ |
| 61 | (nbytes_left) = (nbytes) % OPSIZ; \ |
| 62 | } while (0) |
| 63 | |
| 64 | |
| 65 | /* Threshold value for when to enter the unrolled loops. */ |
| 66 | #define OP_T_THRES 16 |
| 67 | |
| 68 | /* _wordcopy_fwd_aligned -- Copy block beginning at SRCP to |
| 69 | block beginning at DSTP with LEN `op_t' words (not LEN bytes!). |
| 70 | Both SRCP and DSTP should be aligned for memory operations on `op_t's. */ |
| 71 | /* stream-lined (read x8 + write x8) */ |
| 72 | static void _wordcopy_fwd_aligned(long int dstp, long int srcp, size_t len) |
| 73 | { |
| 74 | while (len > 7) { |
| 75 | register op_t a0, a1, a2, a3, a4, a5, a6, a7; |
| 76 | |
| 77 | a0 = ((op_t *) srcp)[0]; |
| 78 | a1 = ((op_t *) srcp)[1]; |
| 79 | a2 = ((op_t *) srcp)[2]; |
| 80 | a3 = ((op_t *) srcp)[3]; |
| 81 | a4 = ((op_t *) srcp)[4]; |
| 82 | a5 = ((op_t *) srcp)[5]; |
| 83 | a6 = ((op_t *) srcp)[6]; |
| 84 | a7 = ((op_t *) srcp)[7]; |
| 85 | ((op_t *) dstp)[0] = a0; |
| 86 | ((op_t *) dstp)[1] = a1; |
| 87 | ((op_t *) dstp)[2] = a2; |
| 88 | ((op_t *) dstp)[3] = a3; |
| 89 | ((op_t *) dstp)[4] = a4; |
| 90 | ((op_t *) dstp)[5] = a5; |
| 91 | ((op_t *) dstp)[6] = a6; |
| 92 | ((op_t *) dstp)[7] = a7; |
| 93 | |
| 94 | srcp += 8 * OPSIZ; |
| 95 | dstp += 8 * OPSIZ; |
| 96 | len -= 8; |
| 97 | } |
| 98 | while (len > 0) { |
| 99 | *(op_t *)dstp = *(op_t *)srcp; |
| 100 | |
| 101 | srcp += OPSIZ; |
| 102 | dstp += OPSIZ; |
| 103 | len -= 1; |
| 104 | } |
| 105 | } |
| 106 | |
| 107 | /* _wordcopy_fwd_dest_aligned -- Copy block beginning at SRCP to |
| 108 | block beginning at DSTP with LEN `op_t' words (not LEN bytes!). |
| 109 | DSTP should be aligned for memory operations on `op_t's, but SRCP must |
| 110 | *not* be aligned. */ |
| 111 | /* stream-lined (read x4 + write x4) */ |
| 112 | static void _wordcopy_fwd_dest_aligned(long int dstp, long int srcp, |
| 113 | size_t len) |
| 114 | { |
| 115 | op_t ap; |
| 116 | int sh_1, sh_2; |
| 117 | |
| 118 | /* Calculate how to shift a word read at the memory operation |
| 119 | aligned srcp to make it aligned for copy. */ |
| 120 | |
| 121 | sh_1 = 8 * (srcp % OPSIZ); |
| 122 | sh_2 = 8 * OPSIZ - sh_1; |
| 123 | |
| 124 | /* Make SRCP aligned by rounding it down to the beginning of the `op_t' |
| 125 | it points in the middle of. */ |
| 126 | srcp &= -OPSIZ; |
| 127 | ap = ((op_t *) srcp)[0]; |
| 128 | srcp += OPSIZ; |
| 129 | |
| 130 | while (len > 3) { |
| 131 | op_t a0, a1, a2, a3; |
| 132 | |
| 133 | a0 = ((op_t *) srcp)[0]; |
| 134 | a1 = ((op_t *) srcp)[1]; |
| 135 | a2 = ((op_t *) srcp)[2]; |
| 136 | a3 = ((op_t *) srcp)[3]; |
| 137 | ((op_t *) dstp)[0] = MERGE(ap, sh_1, a0, sh_2); |
| 138 | ((op_t *) dstp)[1] = MERGE(a0, sh_1, a1, sh_2); |
| 139 | ((op_t *) dstp)[2] = MERGE(a1, sh_1, a2, sh_2); |
| 140 | ((op_t *) dstp)[3] = MERGE(a2, sh_1, a3, sh_2); |
| 141 | |
| 142 | ap = a3; |
| 143 | srcp += 4 * OPSIZ; |
| 144 | dstp += 4 * OPSIZ; |
| 145 | len -= 4; |
| 146 | } |
| 147 | while (len > 0) { |
| 148 | register op_t a0; |
| 149 | |
| 150 | a0 = ((op_t *) srcp)[0]; |
| 151 | ((op_t *) dstp)[0] = MERGE(ap, sh_1, a0, sh_2); |
| 152 | |
| 153 | ap = a0; |
| 154 | srcp += OPSIZ; |
| 155 | dstp += OPSIZ; |
| 156 | len -= 1; |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | void *memcpy(void *dstpp, const void *srcpp, size_t len) |
| 161 | { |
| 162 | unsigned long int dstp = (long int) dstpp; |
| 163 | unsigned long int srcp = (long int) srcpp; |
| 164 | |
| 165 | /* Copy from the beginning to the end. */ |
| 166 | |
| 167 | /* If there not too few bytes to copy, use word copy. */ |
| 168 | if (len >= OP_T_THRES) { |
| 169 | /* Copy just a few bytes to make DSTP aligned. */ |
| 170 | len -= (-dstp) % OPSIZ; |
| 171 | BYTE_COPY_FWD(dstp, srcp, (-dstp) % OPSIZ); |
| 172 | |
| 173 | /* Copy whole pages from SRCP to DSTP by virtual address |
| 174 | manipulation, as much as possible. */ |
| 175 | |
| 176 | /* PAGE_COPY_FWD_MAYBE (dstp, srcp, len, len); */ |
| 177 | |
| 178 | /* Copy from SRCP to DSTP taking advantage of the known |
| 179 | alignment of DSTP. Number of bytes remaining is put in the |
| 180 | third argument, i.e. in LEN. This number may vary from |
| 181 | machine to machine. */ |
| 182 | |
| 183 | WORD_COPY_FWD(dstp, srcp, len, len); |
| 184 | |
| 185 | /* Fall out and copy the tail. */ |
| 186 | } |
| 187 | |
| 188 | /* There are just a few bytes to copy. Use byte memory operations. */ |
| 189 | BYTE_COPY_FWD(dstp, srcp, len); |
| 190 | |
| 191 | return dstpp; |
| 192 | } |
| 193 | |
| 194 | void *memcpyb(void *dstpp, const void *srcpp, unsigned len) |
| 195 | { |
| 196 | unsigned long int dstp = (long int) dstpp; |
| 197 | unsigned long int srcp = (long int) srcpp; |
| 198 | |
| 199 | BYTE_COPY_FWD(dstp, srcp, len); |
| 200 | |
| 201 | return dstpp; |
| 202 | } |