diff options
author | Jim Meyering <jim@meyering.net> | 1995-02-11 14:02:02 +0000 |
---|---|---|
committer | Jim Meyering <jim@meyering.net> | 1995-02-11 14:02:02 +0000 |
commit | b4994dc3feb70726e7fbafac59d6d95710a8971e (patch) | |
tree | 000f5c9bc0b0a8a6936611ef0ea4ffa7946c73d4 /lib | |
parent | 4ef39a33cb05527878b940b229e4b3583fdab31a (diff) | |
download | coreutils-b4994dc3feb70726e7fbafac59d6d95710a8971e.tar.xz |
.
Diffstat (limited to 'lib')
-rw-r--r-- | lib/memcmp.c | 366 |
1 files changed, 366 insertions, 0 deletions
diff --git a/lib/memcmp.c b/lib/memcmp.c new file mode 100644 index 000000000..dce6d3a8c --- /dev/null +++ b/lib/memcmp.c @@ -0,0 +1,366 @@ +/* Copyright (C) 1991, 1993 Free Software Foundation, Inc. + Contributed by Torbjorn Granlund (tege@sics.se). + +NOTE: The canonical source of this file is maintained with the GNU C Library. +Bugs can be reported to bug-glibc@prep.ai.mit.edu. + +This program is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +This program is distributed in the hope that it will be useful, +but WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +GNU General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; if not, write to the Free Software +Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#undef __ptr_t +#if defined (__cplusplus) || (defined (__STDC__) && __STDC__) +#define __ptr_t void * +#else /* Not C++ or ANSI C. */ +#undef const +#define const +#define __ptr_t char * +#endif /* C++ or ANSI C. */ + +#if defined (HAVE_STRING_H) || defined (_LIBC) +#include <string.h> +#endif + +#ifdef _LIBC + +#include <memcopy.h> + +#else /* Not in the GNU C library. */ + +#include <sys/types.h> + +/* Type to use for aligned memory operations. + This should normally be the biggest type supported by a single load + and store. Must be an unsigned type. */ +#define op_t unsigned long int +#define OPSIZ (sizeof(op_t)) + +/* Threshold value for when to enter the unrolled loops. */ +#define OP_T_THRES 16 + +/* Type to use for unaligned operations. */ +typedef unsigned char byte; + +#ifndef WORDS_BIGENDIAN +#define MERGE(w0, sh_1, w1, sh_2) (((w0) >> (sh_1)) | ((w1) << (sh_2))) +#else +#define MERGE(w0, sh_1, w1, sh_2) (((w0) << (sh_1)) | ((w1) >> (sh_2))) +#endif + +#endif /* In the GNU C library. */ + +#ifdef WORDS_BIGENDIAN +#define CMP_LT_OR_GT(a, b) ((a) > (b) ? 1 : -1) +#else +#define CMP_LT_OR_GT(a, b) memcmp_bytes ((a), (b)) +#endif + +/* BE VERY CAREFUL IF YOU CHANGE THIS CODE! */ + +/* The strategy of this memcmp is: + + 1. Compare bytes until one of the block pointers is aligned. + + 2. Compare using memcmp_common_alignment or + memcmp_not_common_alignment, regarding the alignment of the other + block after the initial byte operations. The maximum number of + full words (of type op_t) are compared in this way. + + 3. Compare the few remaining bytes. */ + +#ifndef WORDS_BIGENDIAN +/* memcmp_bytes -- Compare A and B bytewise in the byte order of the machine. + A and B are known to be different. + This is needed only on little-endian machines. */ +#ifdef __GNUC__ +__inline +#endif +static int +memcmp_bytes (a, b) + op_t a, b; +{ + long int srcp1 = (long int) &a; + long int srcp2 = (long int) &b; + op_t a0, b0; + + do + { + a0 = ((byte *) srcp1)[0]; + b0 = ((byte *) srcp2)[0]; + srcp1 += 1; + srcp2 += 1; + } + while (a0 == b0); + return a0 - b0; +} +#endif + +/* memcmp_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN `op_t' + objects (not LEN bytes!). Both SRCP1 and SRCP2 should be aligned for + memory operations on `op_t's. */ +#ifdef __GNUC__ +__inline +#endif +static int +memcmp_common_alignment (srcp1, srcp2, len) + long int srcp1; + long int srcp2; + size_t len; +{ + op_t a0, a1; + op_t b0, b1; + + switch (len % 4) + { + case 2: + a0 = ((op_t *) srcp1)[0]; + b0 = ((op_t *) srcp2)[0]; + srcp1 -= 2 * OPSIZ; + srcp2 -= 2 * OPSIZ; + len += 2; + goto do1; + case 3: + a1 = ((op_t *) srcp1)[0]; + b1 = ((op_t *) srcp2)[0]; + srcp1 -= OPSIZ; + srcp2 -= OPSIZ; + len += 1; + goto do2; + case 0: + if (OP_T_THRES <= 3 * OPSIZ && len == 0) + return 0; + a0 = ((op_t *) srcp1)[0]; + b0 = ((op_t *) srcp2)[0]; + goto do3; + case 1: + a1 = ((op_t *) srcp1)[0]; + b1 = ((op_t *) srcp2)[0]; + srcp1 += OPSIZ; + srcp2 += OPSIZ; + len -= 1; + if (OP_T_THRES <= 3 * OPSIZ && len == 0) + goto do0; + /* Fall through. */ + } + + do + { + a0 = ((op_t *) srcp1)[0]; + b0 = ((op_t *) srcp2)[0]; + if (a1 != b1) + return CMP_LT_OR_GT (a1, b1); + + do3: + a1 = ((op_t *) srcp1)[1]; + b1 = ((op_t *) srcp2)[1]; + if (a0 != b0) + return CMP_LT_OR_GT (a0, b0); + + do2: + a0 = ((op_t *) srcp1)[2]; + b0 = ((op_t *) srcp2)[2]; + if (a1 != b1) + return CMP_LT_OR_GT (a1, b1); + + do1: + a1 = ((op_t *) srcp1)[3]; + b1 = ((op_t *) srcp2)[3]; + if (a0 != b0) + return CMP_LT_OR_GT (a0, b0); + + srcp1 += 4 * OPSIZ; + srcp2 += 4 * OPSIZ; + len -= 4; + } + while (len != 0); + + /* This is the right position for do0. Please don't move + it into the loop. */ + do0: + if (a1 != b1) + return CMP_LT_OR_GT (a1, b1); + return 0; +} + +/* memcmp_not_common_alignment -- Compare blocks at SRCP1 and SRCP2 with LEN + `op_t' objects (not LEN bytes!). SRCP2 should be aligned for memory + operations on `op_t', but SRCP1 *should be unaligned*. */ +#ifdef __GNUC__ +__inline +#endif +static int +memcmp_not_common_alignment (srcp1, srcp2, len) + long int srcp1; + long int srcp2; + size_t len; +{ + op_t a0, a1, a2, a3; + op_t b0, b1, b2, b3; + op_t x; + int shl, shr; + + /* Calculate how to shift a word read at the memory operation + aligned srcp1 to make it aligned for comparison. */ + + shl = 8 * (srcp1 % OPSIZ); + shr = 8 * OPSIZ - shl; + + /* Make SRCP1 aligned by rounding it down to the beginning of the `op_t' + it points in the middle of. */ + srcp1 &= -OPSIZ; + + switch (len % 4) + { + case 2: + a1 = ((op_t *) srcp1)[0]; + a2 = ((op_t *) srcp1)[1]; + b2 = ((op_t *) srcp2)[0]; + srcp1 -= 1 * OPSIZ; + srcp2 -= 2 * OPSIZ; + len += 2; + goto do1; + case 3: + a0 = ((op_t *) srcp1)[0]; + a1 = ((op_t *) srcp1)[1]; + b1 = ((op_t *) srcp2)[0]; + srcp2 -= 1 * OPSIZ; + len += 1; + goto do2; + case 0: + if (OP_T_THRES <= 3 * OPSIZ && len == 0) + return 0; + a3 = ((op_t *) srcp1)[0]; + a0 = ((op_t *) srcp1)[1]; + b0 = ((op_t *) srcp2)[0]; + srcp1 += 1 * OPSIZ; + goto do3; + case 1: + a2 = ((op_t *) srcp1)[0]; + a3 = ((op_t *) srcp1)[1]; + b3 = ((op_t *) srcp2)[0]; + srcp1 += 2 * OPSIZ; + srcp2 += 1 * OPSIZ; + len -= 1; + if (OP_T_THRES <= 3 * OPSIZ && len == 0) + goto do0; + /* Fall through. */ + } + + do + { + a0 = ((op_t *) srcp1)[0]; + b0 = ((op_t *) srcp2)[0]; + x = MERGE(a2, shl, a3, shr); + if (x != b3) + return CMP_LT_OR_GT (x, b3); + + do3: + a1 = ((op_t *) srcp1)[1]; + b1 = ((op_t *) srcp2)[1]; + x = MERGE(a3, shl, a0, shr); + if (x != b0) + return CMP_LT_OR_GT (x, b0); + + do2: + a2 = ((op_t *) srcp1)[2]; + b2 = ((op_t *) srcp2)[2]; + x = MERGE(a0, shl, a1, shr); + if (x != b1) + return CMP_LT_OR_GT (x, b1); + + do1: + a3 = ((op_t *) srcp1)[3]; + b3 = ((op_t *) srcp2)[3]; + x = MERGE(a1, shl, a2, shr); + if (x != b2) + return CMP_LT_OR_GT (x, b2); + + srcp1 += 4 * OPSIZ; + srcp2 += 4 * OPSIZ; + len -= 4; + } + while (len != 0); + + /* This is the right position for do0. Please don't move + it into the loop. */ + do0: + x = MERGE(a2, shl, a3, shr); + if (x != b3) + return CMP_LT_OR_GT (x, b3); + return 0; +} + +int +memcmp (s1, s2, len) + const __ptr_t s1; + const __ptr_t s2; + size_t len; +{ + op_t a0; + op_t b0; + long int srcp1 = (long int) s1; + long int srcp2 = (long int) s2; + op_t res; + + if (len >= OP_T_THRES) + { + /* There are at least some bytes to compare. No need to test + for LEN == 0 in this alignment loop. */ + while (srcp2 % OPSIZ != 0) + { + a0 = ((byte *) srcp1)[0]; + b0 = ((byte *) srcp2)[0]; + srcp1 += 1; + srcp2 += 1; + res = a0 - b0; + if (res != 0) + return res; + len -= 1; + } + + /* SRCP2 is now aligned for memory operations on `op_t'. + SRCP1 alignment determines if we can do a simple, + aligned compare or need to shuffle bits. */ + + if (srcp1 % OPSIZ == 0) + res = memcmp_common_alignment (srcp1, srcp2, len / OPSIZ); + else + res = memcmp_not_common_alignment (srcp1, srcp2, len / OPSIZ); + if (res != 0) + return res; + + /* Number of bytes remaining in the interval [0..OPSIZ-1]. */ + srcp1 += len & -OPSIZ; + srcp2 += len & -OPSIZ; + len %= OPSIZ; + } + + /* There are just a few bytes to compare. Use byte memory operations. */ + while (len != 0) + { + a0 = ((byte *) srcp1)[0]; + b0 = ((byte *) srcp2)[0]; + srcp1 += 1; + srcp2 += 1; + res = a0 - b0; + if (res != 0) + return res; + len -= 1; + } + + return 0; +} |