X-Git-Url: http://lists.indexdata.com/cgi-bin?a=blobdiff_plain;f=isamc%2Fisamd.c;h=b1a2122ed3000eef80f41965504a13e53d0a1e06;hb=3ecb370349f3ea43b0b34cf4d097918c349b1aac;hp=3052990259a3bcf2c2a1da9ee34cb090a862cc4f;hpb=391f50227af9b86500ad1ddee46f3bd9d461aa92;p=idzebra-moved-to-github.git diff --git a/isamc/isamd.c b/isamc/isamd.c index 3052990..b1a2122 100644 --- a/isamc/isamd.c +++ b/isamc/isamd.c @@ -1,7 +1,7 @@ /* * Copyright (c) 1995-1998, Index Data. * See the file LICENSE for details. - * $Id: isamd.c,v 1.8 1999-08-18 13:28:16 heikki Exp $ + * $Id: isamd.c,v 1.15 1999-09-27 14:36:36 heikki Exp $ * * Isamd - isam with diffs * Programmed by: Heikki Levanto @@ -27,23 +27,23 @@ static void init_fc (ISAMD is, int cat); #define ISAMD_FREELIST_CHUNK 1 -#define SMALL_TEST 1 +#define SMALL_TEST 0 ISAMD_M isamd_getmethod (ISAMD_M me) { static struct ISAMD_filecat_s def_cat[] = { #if SMALL_TEST /* blocksz, max. Unused time being */ - { 20, 40 }, - { 32, 0 }, + { 32, 40 }, /* 24 is the smallest unreasonable size! */ + { 64, 0 }, #else - { 24, 1 }, { 32, 1 }, - { 64, 1 }, { 128, 1 }, - { 256, 1 }, - { 1024, 1 }, - { 2048, 0 }, + { 512, 1 }, + { 2048, 1 }, + { 8192, 1 }, + { 32768, 1 }, + {131072, 0 }, #endif /* old values from isamc, long time ago... @@ -136,6 +136,13 @@ ISAMD isamd_open (BFiles bfs, const char *name, int writeflag, ISAMD_M method) is->files[i].sum_backward = 0; is->files[i].no_next = 0; is->files[i].no_prev = 0; + is->files[i].no_op_nodiff=0; + is->files[i].no_op_intdiff=0; + is->files[i].no_op_extdiff=0; + is->files[i].no_fbuilds=0; + is->files[i].no_appds=0; + is->files[i].no_merges=0; + is->files[i].no_remerges=0; init_fc (is, i); } @@ -163,9 +170,10 @@ int isamd_close (ISAMD is) if (is->method->debug>0) { - logf (LOG_LOG, "isamd: next forw mid-f prev backw mid-b"); + logf (LOG_LOG, "isamd statistics"); + logf (LOG_LOG, "f nxt forw mid-f prev backw mid-b"); for (i = 0; ino_files; i++) - logf (LOG_LOG, "isamd:%8d%8d%8.1f%8d%8d%8.1f", + logf (LOG_LOG, "%d%8d%8d%8.1f%8d%8d%8.1f",i, is->files[i].no_next, is->files[i].no_forward, is->files[i].no_forward ? @@ -178,7 +186,7 @@ int isamd_close (ISAMD is) : 0.0); } if (is->method->debug>0) - logf (LOG_LOG, "isamd: writes reads skipped alloc released remap"); + logf (LOG_LOG, "f writes reads skipped alloc released "); for (i = 0; ino_files; i++) { release_fc (is, i); @@ -187,17 +195,37 @@ int isamd_close (ISAMD is) bf_write (is->files[i].bf, 0, 0, sizeof(ISAMD_head), &is->files[i].head); if (is->method->debug>0) - logf (LOG_LOG, "isamd:%8d%8d%8d%8d%8d%8d", + logf (LOG_LOG, "%d%8d%8d%8d%8d%8d",i, is->files[i].no_writes, is->files[i].no_reads, is->files[i].no_skip_writes, is->files[i].no_allocated, - is->files[i].no_released, - is->files[i].no_remap); + is->files[i].no_released); xfree (is->files[i].fc_list); flush_block (is, i); bf_close (is->files[i].bf); } + + if (is->method->debug>0) + { + logf (LOG_LOG, "f opens simple int ext"); + for (i = 0; ino_files; i++) + { + logf (LOG_LOG, "%d%8d%8d%8d%8d",i, + is->files[i].no_op_nodiff+ + is->files[i].no_op_intdiff+ + is->files[i].no_op_extdiff, + is->files[i].no_op_nodiff, + is->files[i].no_op_intdiff, + is->files[i].no_op_extdiff); + } + logf (LOG_LOG, " build append merge remrg"); + logf (LOG_LOG, "=%8d%8d%8d%8d", + is->files[0].no_fbuilds, + is->files[0].no_appds, + is->files[0].no_merges, + is->files[0].no_remerges); + } xfree (is->files); xfree (is->method); xfree (is); @@ -208,7 +236,7 @@ int isamd_read_block (ISAMD is, int cat, int pos, char *dst) { ++(is->files[cat].no_reads); if (is->method->debug > 6) - logf (LOG_LOG, "isamd: read_block %d %d", cat, pos); + logf (LOG_LOG, "isamd: read_block %d:%d",cat, pos); return bf_read (is->files[cat].bf, pos, 0, 0, dst); } @@ -216,7 +244,7 @@ int isamd_write_block (ISAMD is, int cat, int pos, char *src) { ++(is->files[cat].no_writes); if (is->method->debug > 6) - logf (LOG_LOG, "isamd: write_block %d %d", cat, pos); + logf (LOG_LOG, "isamd: write_block %d:%d", cat, pos); return bf_write (is->files[cat].bf, pos, 0, 0, src); } @@ -452,10 +480,12 @@ ISAMD_PP isamd_pp_open (ISAMD is, ISAMD_P ipos) char *src; int sz = is->method->filecat[is->max_cat].bsize; /* always allocate for the largest blocks, saves trouble */ - - pp->cat = isamd_type(ipos); - pp->pos = isamd_block(ipos); - + struct it_key singlekey; + char *c_ptr; /* for fake encoding the singlekey */ + char *i_ptr; + int ofs; + + pp->numKeys = 0; src = pp->buf = (char *) xmalloc (sz); memset(src,'\0',sz); /* clear the buffer, for new blocks */ @@ -463,13 +493,44 @@ ISAMD_PP isamd_pp_open (ISAMD is, ISAMD_P ipos) pp->size = 0; pp->offset = 0; pp->is = is; - pp->decodeClientData = (*is->method->code_start)(ISAMD_DECODE); - pp->numKeys = 0; pp->diffs=0; - pp->diffbuf=0; pp->diffinfo=0; + pp->decodeClientData = (*is->method->code_start)(ISAMD_DECODE); + if ( is_singleton(ipos) ) + { + pp->cat=0; + pp->pos=0; + if (is->method->debug > 5) + logf (LOG_LOG, "isamd_pp_open %p %d=%d:%d sz=%d n=%d=%d:%d", + pp, isamd_addr(pp->pos, pp->cat), pp->cat, pp->pos, pp->size, + pp->next, isamd_type(pp->next), isamd_block(pp->next) ); + singleton_decode(ipos, &singlekey ); + pp->offset=ISAMD_BLOCK_OFFSET_1; + pp->numKeys = 1; + ofs=pp->offset+sizeof(int); /* reserve length of diffsegment */ + singlekey.seqno = singlekey.seqno * 2 + 1; /* make an insert diff */ + c_ptr=&(pp->buf[ofs]); + i_ptr=(char*)(&singlekey); + (*is->method->code_item)(ISAMD_ENCODE, pp->decodeClientData, + &c_ptr, &i_ptr); + (*is->method->code_reset)(pp->decodeClientData); + ofs += c_ptr-&(pp->buf[ofs]); + memcpy( &(pp->buf[pp->offset]), &ofs, sizeof(int) ); + /* since we memset buf earlier, we already have a zero endmark! */ + pp->size = ofs; + if (is->method->debug > 5) + logf (LOG_LOG, "isamd_pp_open single %d=%x: %d.%d sz=%d", + ipos,ipos, + singlekey.sysno, singlekey.seqno/2, + pp->size ); + return pp; + } /* singleton */ + + pp->cat = isamd_type(ipos); + pp->pos = isamd_block(ipos); + if (pp->pos) { src = pp->buf; @@ -480,18 +541,16 @@ ISAMD_PP isamd_pp_open (ISAMD is, ISAMD_P ipos) src += sizeof(pp->size); memcpy (&pp->numKeys, src, sizeof(pp->numKeys)); src += sizeof(pp->numKeys); - memcpy (&pp->diffs, src, sizeof(pp->diffs)); - src += sizeof(pp->diffs); assert (pp->next != pp->pos); pp->offset = src - pp->buf; assert (pp->offset == ISAMD_BLOCK_OFFSET_1); + assert(pp->size>=ISAMD_BLOCK_OFFSET_1); /*??*/ } if (is->method->debug > 5) logf (LOG_LOG, "isamd_pp_open %p %d=%d:%d sz=%d n=%d=%d:%d", pp, isamd_addr(pp->pos, pp->cat), pp->cat, pp->pos, pp->size, pp->next, isamd_type(pp->next), isamd_block(pp->next) ); - - + return pp; } @@ -507,15 +566,14 @@ void isamd_buildfirstblock(ISAMD_PP pp){ dst += sizeof(pp->size); memcpy(dst, &pp->numKeys, sizeof(pp->numKeys)); dst += sizeof(pp->numKeys); - memcpy(dst, &pp->diffs, sizeof(pp->diffs)); - dst += sizeof(pp->diffs); +// memcpy(dst, &pp->diffs, sizeof(pp->diffs)); +// dst += sizeof(pp->diffs); assert (dst - pp->buf == ISAMD_BLOCK_OFFSET_1); if (pp->is->method->debug > 5) - logf (LOG_LOG, "isamd: first: sz=%d p=%d/%d>%d/%d nk=%d d=%d", - pp->size, - pp->cat, pp->pos, - isamd_type(pp->next), isamd_block(pp->next), - pp->numKeys, pp->diffs); + logf (LOG_LOG, "isamd: bldfirst: p=%d=%d:%d n=%d:%d:%d sz=%d nk=%d ", + isamd_addr(pp->pos,pp->cat),pp->cat, pp->pos, + pp->next, isamd_type(pp->next), isamd_block(pp->next), + pp->size, pp->numKeys); } void isamd_buildlaterblock(ISAMD_PP pp){ @@ -539,9 +597,10 @@ void isamd_buildlaterblock(ISAMD_PP pp){ /* returns non-zero if item could be read; 0 otherwise */ int isamd_pp_read (ISAMD_PP pp, void *buf) { + return isamd_read_item (pp, (char **) &buf); - /* note: isamd_read_item is in merge-d.c, because it is so */ - /* convoluted with the merge process */ + /* note: isamd_read_item is in merge-d.c, because it is so */ + /* convoluted with the merge process */ } /* read one main item from file - decode and store it in *dst. @@ -555,6 +614,7 @@ int isamd_read_main_item (ISAMD_PP pp, char **dst) ISAMD is = pp->is; char *src = pp->buf + pp->offset; int newcat; + int oldoffs; if (pp->offset >= pp->size) { @@ -596,18 +656,24 @@ int isamd_read_main_item (ISAMD_PP pp, char **dst) memcpy (&pp->size, src, sizeof(pp->size)); src += sizeof(pp->size); /* assume block is non-empty */ - assert (src - pp->buf == ISAMD_BLOCK_OFFSET_N); + pp->offset = oldoffs = src - pp->buf; + assert (pp->offset == ISAMD_BLOCK_OFFSET_N); assert (pp->next != isamd_addr(pp->pos,pp->cat)); (*is->method->code_reset)(pp->decodeClientData); + /* finally, read the item */ (*is->method->code_item)(ISAMD_DECODE, pp->decodeClientData, dst, &src); pp->offset = src - pp->buf; - if (is->method->debug > 4) - logf (LOG_LOG, "isamd: read_block size=%d %d %d next=%d", - pp->size, pp->cat, pp->pos, pp->next); + if (is->method->debug > 8) + logf (LOG_LOG, "isamd: read_m: block %d:%d sz=%d ofs=%d-%d next=%d", + pp->cat, pp->pos, pp->size, oldoffs, pp->offset, pp->next); return 2; } + oldoffs=pp->offset; (*is->method->code_item)(ISAMD_DECODE, pp->decodeClientData, dst, &src); pp->offset = src - pp->buf; + if (is->method->debug > 8) + logf (LOG_LOG, "isamd: read_m: got %d:%d sz=%d ofs=%d-%d next=%d", + pp->cat, pp->pos, pp->size, oldoffs, pp->offset, pp->next); return 1; } @@ -639,23 +705,35 @@ void isamd_pp_dump (ISAMD is, ISAMD_P ipos) int i,n; int occur =0; int oldoffs; + int diffmax=1; + int diffidx; char hexbuff[64]; + int olddebug= is->method->debug; + is->method->debug=0; /* no debug logs while reading for dump */ logf(LOG_LOG,"dumping isamd block %d (%d:%d)", (int)ipos, isamd_type(ipos), isamd_block(ipos) ); pp=isamd_pp_open(is,ipos); - logf(LOG_LOG,"numKeys=%d, ofs=%d d=%d", - pp->numKeys, - pp->offset, pp->diffs); - oldoffs= pp->offset; + logf(LOG_LOG,"numKeys=%d, ofs=%d sz=%d", + pp->numKeys, pp->offset, pp->size ); + diffidx=oldoffs= pp->offset; + while ((diffidx < is->method->filecat[pp->cat].bsize) && (diffmax>0)) + { + memcpy(&diffmax,&(pp->buf[diffidx]),sizeof(int)); + logf (LOG_LOG,"diff set at %d-%d: %s", diffidx, diffmax, + hexdump(pp->buf+diffidx,8,0)); + /*! todo: dump the actual diffs as well !!! */ + diffidx=diffmax; + + } /* dump diffs */ while(isamd_pp_read(pp, &key)) { if (oldaddr != isamd_addr(pp->pos,pp->cat) ) { oldaddr = isamd_addr(pp->pos,pp->cat); - logf(LOG_LOG,"block %d (%d:%d) sz=%d nx=%d (%d:%d) ofs=%d", - isamd_addr(pp->pos,pp->cat), - pp->cat, pp->pos, pp->size, + logf(LOG_LOG,"block %d=%d:%d sz=%d nx=%d=%d:%d ofs=%d", + isamd_addr(pp->pos,pp->cat), pp->cat, pp->pos, + pp->size, pp->next, isamd_type(pp->next), isamd_block(pp->next), pp->offset); i=0; @@ -678,11 +756,33 @@ void isamd_pp_dump (ISAMD is, ISAMD_P ipos) } /*!*/ /*TODO: dump diffs too!!! */ isamd_pp_close(pp); + is->method->debug=olddebug; } /* dump */ /* * $Log: isamd.c,v $ - * Revision 1.8 1999-08-18 13:28:16 heikki + * Revision 1.15 1999-09-27 14:36:36 heikki + * singletons + * + * Revision 1.14 1999/09/23 18:01:18 heikki + * singleton optimising + * + * Revision 1.13 1999/09/20 15:48:06 heikki + * Small changes + * + * Revision 1.12 1999/09/13 13:28:28 heikki + * isam-d optimizing: merging input data in the same go + * + * Revision 1.11 1999/08/25 18:09:24 heikki + * Starting to optimize + * + * Revision 1.10 1999/08/24 13:17:42 heikki + * Block sizes, comments + * + * Revision 1.9 1999/08/20 12:25:58 heikki + * Statistics in isamd + * + * Revision 1.8 1999/08/18 13:28:16 heikki * Set log levels to decent values * * Revision 1.6 1999/08/17 19:44:25 heikki