From: Nicholas Clark Date: Sat, 31 Mar 2007 20:52:53 +0000 (+0000) Subject: Reorder the members of various regexp structs to reduce their size on X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=commitdiff_plain;h=1b5c067c53616dae46b48a67460c08ec2c2ec4c4;p=p5sagit%2Fp5-mst-13.2.git Reorder the members of various regexp structs to reduce their size on LP64 platforms, by pairing up I32 and U32 members. Notably structs _reg_trie_data, reg_ac_data, regexp and regmatch_state down by 8 bytes, re_save_state by 16, and regmatch_slab up by 48 (ie one more state per slab) p4raw-id: //depot/perl@30815 --- diff --git a/regcomp.h b/regcomp.h index 4799b1f..a13d487 100644 --- a/regcomp.h +++ b/regcomp.h @@ -561,20 +561,20 @@ typedef struct _reg_trie_trans reg_trie_trans; optimisation in Perl_regdupe. */ struct _reg_trie_data { U32 refcount; /* number of times this trie is referenced */ - U16 uniquecharcount; /* unique chars in trie (width of trans table) */ U32 lasttrans; /* last valid transition element */ U16 *charmap; /* byte to charid lookup array */ reg_trie_state *states; /* state data */ reg_trie_trans *trans; /* array of transition elements */ char *bitmap; /* stclass bitmap */ - U32 startstate; /* initial state - used for common prefix optimisation */ - STRLEN minlen; /* minimum length of words in trie - build/opt only? */ - STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ U32 *wordlen; /* array of lengths of words */ U16 *jump; /* optional 1 indexed array of offsets before tail for the node following a given word. */ U16 *nextword; /* optional 1 indexed array to support linked list of duplicate wordnums */ + U16 uniquecharcount; /* unique chars in trie (width of trans table) */ + U32 startstate; /* initial state - used for common prefix optimisation */ + STRLEN minlen; /* minimum length of words in trie - build/opt only? */ + STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ U32 statecount; /* Build only - number of states in the states array (including the unused zero state) */ U32 wordcount; /* Build only */ @@ -600,9 +600,9 @@ typedef struct _reg_trie_data reg_trie_data; optimisation in Perl_regdupe. */ struct _reg_ac_data { U32 refcount; + U32 trie; U32 *fail; reg_trie_state *states; - U32 trie; }; typedef struct _reg_ac_data reg_ac_data; diff --git a/regexp.h b/regexp.h index e195f67..bb73dab 100644 --- a/regexp.h +++ b/regexp.h @@ -84,13 +84,13 @@ typedef struct regexp { char *subbeg; /* saved or original string so \digit works forever. */ - I32 sublen; /* Length of string pointed by subbeg */ SV_SAVED_COPY /* If non-NULL, SV which is COW from original */ + I32 sublen; /* Length of string pointed by subbeg */ /* Information about the match that isn't often used */ - const char *precomp; /* pre-compilation regular expression */ I32 prelen; /* length of precomp */ + const char *precomp; /* pre-compilation regular expression */ /* wrapped can't be const char*, as it is returned by sv_2pv_flags */ char *wrapped; /* wrapped version of the pattern */ I32 wraplen; /* length of wrapped */ @@ -373,8 +373,8 @@ typedef struct regmatch_state { inner and outer rexen */ CHECKPOINT cp; /* remember current savestack indexes */ CHECKPOINT lastcp; - regnode *B; /* the node following us */ U32 close_paren; /* which close bracket is our end */ + regnode *B; /* the node following us */ } eval; struct { @@ -403,12 +403,12 @@ typedef struct regmatch_state { /* this first element must match u.yes */ struct regmatch_state *prev_yes_state; struct regmatch_state *prev_curlyx; /* previous cur_curlyx */ + regnode *A, *B; /* the nodes corresponding to /A*B/ */ CHECKPOINT cp; /* remember current savestack index */ bool minmod; int parenfloor;/* how far back to strip paren data */ int min; /* the minimal number of A's to match */ int max; /* the maximal number of A's to match */ - regnode *A, *B; /* the nodes corresponding to /A*B/ */ /* these two are modified by WHILEM */ int count; /* how many instances of A we've matched */ @@ -491,6 +491,9 @@ typedef struct regmatch_slab { struct re_save_state { U32 re_state_reg_flags; /* from regexec.c */ + U32 re_state_reg_start_tmpl; /* from regexec.c */ + I32 re_state_reg_eval_set; /* from regexec.c */ + bool re_state_reg_match_utf8; /* from regexec.c */ char *re_state_bostr; char *re_state_reginput; /* String-input pointer. */ char *re_state_regeol; /* End of input, for $ check. */ @@ -498,20 +501,17 @@ struct re_save_state { U32 *re_state_reglastparen; /* Similarly for lastparen. */ U32 *re_state_reglastcloseparen; /* Similarly for lastcloseparen. */ char **re_state_reg_start_tmp; /* from regexec.c */ - U32 re_state_reg_start_tmpl; /* from regexec.c */ - I32 re_state_reg_eval_set; /* from regexec.c */ - bool re_state_reg_match_utf8; /* from regexec.c */ MAGIC *re_state_reg_magic; /* from regexec.c */ - I32 re_state_reg_oldpos; /* from regexec.c */ PMOP *re_state_reg_oldcurpm; /* from regexec.c */ PMOP *re_state_reg_curpm; /* from regexec.c */ char *re_state_reg_oldsaved; /* old saved substr during match */ STRLEN re_state_reg_oldsavedlen; /* old length of saved substr during match */ + STRLEN re_state_reg_poscache_size; /* size of pos cache of WHILEM */ + I32 re_state_reg_oldpos; /* from regexec.c */ I32 re_state_reg_maxiter; /* max wait until caching pos */ I32 re_state_reg_leftiter; /* wait until caching pos */ - char *re_state_reg_poscache; /* cache of pos of WHILEM */ - STRLEN re_state_reg_poscache_size; /* size of pos cache of WHILEM */ U32 re_state_regsize; /* from regexec.c */ + char *re_state_reg_poscache; /* cache of pos of WHILEM */ char *re_state_reg_starttry; /* from regexec.c */ #ifdef PERL_OLD_COPY_ON_WRITE SV *re_state_nrs; /* was placeholder: unused since 5.8.0 (5.7.2 patch #12027 for bug ID 20010815.012). Used to save rx->saved_copy */