>From e56992c4bfdb2e02a114b14c34780672a9c8cee9 Mon Sep 17 00:00:00 2001 From: Norihiro Tanaka Date: Sun, 23 Mar 2014 20:14:33 +0900 Subject: [PATCH] grep: optimization of DFA by reuse of multi-byte buffers in non-UTF8 locales * src/dfa.c (struct dfa): New members `mblen_buf', `nmblen_buf', `inputwcs', `ninputwcs', `mb_follows' and `mb_match_lens'. (SKIP_REMAINS_MB_IF_INITIAL_STATE): Use them instead of global variables. (match_anychar): Use them instead of global variables. (match_mb_charset): Use them instead of global variables. (check_matching_with_multibyte_ops): Use it instead of new allocation. (transit_state_consume_1char): Use them instead of global variables. (transit_state): Use them instead of global variables. (prepare_wc_buf): Use them instead of global variables. (dfaexec): Initializations of new members. (free_mbdata): Memory Deallocations for new members. --- src/dfa.c | 124 +++++++++++++++++++++++++++++++------------------------------- 1 file changed, 63 insertions(+), 61 deletions(-) diff --git a/src/dfa.c b/src/dfa.c index 5e60cd5..73e2432 100644 --- a/src/dfa.c +++ b/src/dfa.c @@ -424,6 +424,30 @@ struct dfa struct dfamust *musts; /* List of strings, at least one of which is known to appear in any r.e. matching the dfa. */ + unsigned char *mblen_buf; /* Correspond to the input buffer in dfaexec. + Each element stores the number of remaining + bytes of the corresponding multibyte + character in the input string. A element's + value is 0 if the corresponding character is + single-byte. + e.g., input : 'a', , , + mblen_buf : 0, 3, 2, 1 + */ + size_t nmblen_buf; /* Length of the mblen buffer currently + allocated. */ + wchar_t *inputwcs; /* Wide character representation of the input + string in dfaexec. + The length of this array is the same as + the length of input string (char array). + inputstring[i] is a single-byte char, + or the first byte of a multibyte char; + inputwcs[i] is the codepoint. */ + size_t ninputwcs; /* Length of the input wide characters + currently allocated. */ + position_set *mb_follows; /* Follow set added by ANYCHAR and/or MBCSET + on demand. */ + int *mb_match_lens; /* Array of length reduced by ANYCHAR and/or + MBCSET. */ }; /* Some macros for user access to dfa internals. */ @@ -806,22 +830,6 @@ static int cur_mb_len = 1; /* Length of the multibyte representation of static mbstate_t mbs; /* Mbstate for mbrlen. */ static wchar_t wctok; /* Wide character representation of the current multibyte character. */ -static unsigned char *mblen_buf;/* Correspond to the input buffer in dfaexec. - Each element stores the number of remaining - bytes of the corresponding multibyte - character in the input string. A element's - value is 0 if the corresponding character is - single-byte. - e.g., input : 'a', , , - mblen_buf : 0, 3, 2, 1 - */ -static wchar_t *inputwcs; /* Wide character representation of the input - string in dfaexec. - The length of this array is the same as - the length of input string (char array). - inputstring[i] is a single-byte char, - or the first byte of a multibyte char; - inputwcs[i] is the codepoint. */ static unsigned char const *buf_begin; /* reference to begin in dfaexec. */ static unsigned char const *buf_end; /* reference to end in dfaexec. */ @@ -2907,14 +2915,12 @@ build_state_zero (struct dfa *d) #define SKIP_REMAINS_MB_IF_INITIAL_STATE(s, p) \ if (s == 0) \ { \ - while (inputwcs[p - buf_begin] == 0 \ - && mblen_buf[p - buf_begin] > 0 \ + while (d->inputwcs[p - buf_begin] == 0 \ + && d->mblen_buf[p - buf_begin] > 0 \ && (unsigned char const *) p < buf_end) \ ++p; \ if ((char *) p >= end) \ { \ - free (mblen_buf); \ - free (inputwcs); \ *end = saved_end; \ return NULL; \ } \ @@ -3008,8 +3014,8 @@ match_anychar (struct dfa *d, state_num s, position pos, size_t idx) wchar_t wc; int mbclen; - wc = inputwcs[idx]; - mbclen = (mblen_buf[idx] == 0) ? 1 : mblen_buf[idx]; + wc = d->inputwcs[idx]; + mbclen = (d->mblen_buf[idx] == 0) ? 1 : d->mblen_buf[idx]; /* Check syntax bits. */ if (wc == (wchar_t) eolbyte) @@ -3050,7 +3056,7 @@ match_mb_charset (struct dfa *d, state_num s, position pos, size_t idx) int context; wchar_t wc; /* Current referring character. */ - wc = inputwcs[idx]; + wc = d->inputwcs[idx]; /* Check syntax bits. */ if (wc == (wchar_t) eolbyte) @@ -3071,7 +3077,7 @@ match_mb_charset (struct dfa *d, state_num s, position pos, size_t idx) /* Assign the current referring operator to work_mbc. */ work_mbc = &(d->mbcsets[(d->multibyte_prop[pos.index]) >> 2]); match = !work_mbc->invert; - match_len = (mblen_buf[idx] == 0) ? 1 : mblen_buf[idx]; + match_len = (d->mblen_buf[idx] == 0) ? 1 : d->mblen_buf[idx]; /* Match in range 0-255? */ if (wc < NOTCHAR && work_mbc->cset != -1 @@ -3149,7 +3155,7 @@ check_matching_with_multibyte_ops (struct dfa *d, state_num s, size_t idx) size_t i; int *rarray; - MALLOC (rarray, d->states[s].mbps.nelem); + rarray = d->mb_match_lens; for (i = 0; i < d->states[s].mbps.nelem; ++i) { position pos = d->states[s].mbps.elems[i]; @@ -3189,7 +3195,7 @@ transit_state_consume_1char (struct dfa *d, state_num s, /* Calculate the length of the (single/multi byte) character to which p points. */ - *mbclen = (mblen_buf[*pp - buf_begin] == 0) ? 1 : mblen_buf[*pp - buf_begin]; + *mbclen = (d->mblen_buf[*pp - buf_begin] == 0) ? 1 : d->mblen_buf[*pp - buf_begin]; /* Calculate the state which can be reached from the state 's' by consuming '*mbclen' single bytes from the buffer. */ @@ -3218,9 +3224,6 @@ transit_state_consume_1char (struct dfa *d, state_num s, insert (d->follows[d->states[s].mbps.elems[i].index].elems[j], pps); } - if (match_lens == NULL && work_mbls != NULL) - free (work_mbls); - /* FIXME: this return value is always ignored. */ return rs; } @@ -3268,12 +3271,11 @@ transit_state (struct dfa *d, state_num s, unsigned char const **pp) if (rs == TRANSIT_STATE_DONE) ++*pp; - free (match_lens); return s1; } /* This state has some operators which can match a multibyte character. */ - alloc_position_set (&follows, d->nleaves); + follows = *d->mb_follows; /* 'maxlen' may be longer than the length of a character, because it may not be a character but a (multi character) collating element. @@ -3281,7 +3283,7 @@ transit_state (struct dfa *d, state_num s, unsigned char const **pp) 'maxlen' bytes. */ transit_state_consume_1char (d, s, pp, match_lens, &mbclen, &follows); - wc = inputwcs[*pp - mbclen - buf_begin]; + wc = d->inputwcs[*pp - mbclen - buf_begin]; s1 = state_index (d, &follows, wchar_context (wc)); realloc_trans_if_necessary (d, s1); @@ -3298,12 +3300,10 @@ transit_state (struct dfa *d, state_num s, unsigned char const **pp) &follows); } - wc = inputwcs[*pp - mbclen - buf_begin]; + wc = d->inputwcs[*pp - mbclen - buf_begin]; s1 = state_index (d, &follows, wchar_context (wc)); realloc_trans_if_necessary (d, s1); } - free (match_lens); - free (follows.elems); return s1; } @@ -3311,7 +3311,7 @@ transit_state (struct dfa *d, state_num s, unsigned char const **pp) /* Initialize mblen_buf and inputwcs with data from the next line. */ static void -prepare_wc_buf (const char *begin, const char *end) +prepare_wc_buf (struct dfa *d, const char *begin, const char *end) { #if MBS_SUPPORT unsigned char eol = eolbyte; @@ -3325,35 +3325,35 @@ prepare_wc_buf (const char *begin, const char *end) if (remain_bytes == 0) { remain_bytes - = mbrtowc (inputwcs + i, begin + i, end - begin - i + 1, &mbs); + = mbrtowc (d->inputwcs + i, begin + i, end - begin - i + 1, &mbs); if (remain_bytes < 1 || remain_bytes == (size_t) -1 || remain_bytes == (size_t) -2 - || (remain_bytes == 1 && inputwcs[i] == (wchar_t) begin[i])) + || (remain_bytes == 1 && d->inputwcs[i] == (wchar_t) begin[i])) { remain_bytes = 0; - inputwcs[i] = (wchar_t) begin[i]; - mblen_buf[i] = 0; + d->inputwcs[i] = (wchar_t) begin[i]; + d->mblen_buf[i] = 0; if (begin[i] == eol) break; } else { - mblen_buf[i] = remain_bytes; + d->mblen_buf[i] = remain_bytes; remain_bytes--; } } else { - mblen_buf[i] = remain_bytes; - inputwcs[i] = 0; + d->mblen_buf[i] = remain_bytes; + d->inputwcs[i] = 0; remain_bytes--; } } buf_end = (unsigned char *) (begin + i); - mblen_buf[i] = 0; - inputwcs[i] = 0; /* sentinel */ + d->mblen_buf[i] = 0; + d->inputwcs[i] = 0; /* sentinel */ #endif /* MBS_SUPPORT */ } @@ -3391,10 +3391,18 @@ dfaexec (struct dfa *d, char const *begin, char *end, if (d->mb_cur_max > 1) { - MALLOC (mblen_buf, end - begin + 2); - MALLOC (inputwcs, end - begin + 2); + static bool mb_alloc = false; + REALLOC_IF_NECESSARY (d->mblen_buf, d->nmblen_buf, end - begin + 2); + REALLOC_IF_NECESSARY (d->inputwcs, d->ninputwcs, end - begin + 2); memset (&mbs, 0, sizeof (mbstate_t)); - prepare_wc_buf ((const char *) p, end); + prepare_wc_buf (d, (const char *) p, end); + if (!mb_alloc) + { + MALLOC (d->mb_match_lens, d->nleaves); + MALLOC (d->mb_follows, 1); + alloc_position_set (d->mb_follows, d->nleaves); + mb_alloc = true; + } } for (;;) @@ -3421,8 +3429,6 @@ dfaexec (struct dfa *d, char const *begin, char *end, if (backref) { *backref = 1; - free (mblen_buf); - free (inputwcs); *end = saved_end; return (char *) p; } @@ -3455,11 +3461,6 @@ dfaexec (struct dfa *d, char const *begin, char *end, { if (backref) *backref = (d->states[s].backref != 0); - if (d->mb_cur_max > 1) - { - free (mblen_buf); - free (inputwcs); - } *end = saved_end; return (char *) p; } @@ -3484,17 +3485,12 @@ dfaexec (struct dfa *d, char const *begin, char *end, ++*count; if (d->mb_cur_max > 1) - prepare_wc_buf ((const char *) p, end); + prepare_wc_buf (d, (const char *) p, end); } /* Check if we've run off the end of the buffer. */ if ((char *) p > end) { - if (d->mb_cur_max > 1) - { - free (mblen_buf); - free (inputwcs); - } *end = saved_end; return NULL; } @@ -3545,6 +3541,12 @@ free_mbdata (struct dfa *d) free (d->mbcsets); d->mbcsets = NULL; d->nmbcsets = 0; + + free (d->inputwcs); + if (d->mb_follows != NULL) + free (d->mb_follows->elems); + free (d->mb_follows); + free (d->mb_match_lens); } /* Initialize the components of a dfa that the other routines don't -- 1.9.1