X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=isamc%2Fisamd.c;h=8e711aead10255a3638976d616ad3cc3032a0f8e;hb=b918342ded2c57bcd57bf7af6748c17bce4d1cf6;hp=6013abda048f44a1ed464b95041222465a077a7e;hpb=3d9123eca4abdaa6582bb4d77b68e7d18650507c;p=idzebra-moved-to-github.git diff --git a/isamc/isamd.c b/isamc/isamd.c index 6013abd..8e711ae 100644 --- a/isamc/isamd.c +++ b/isamc/isamd.c @@ -1,10 +1,14 @@ /* * Copyright (c) 1995-1998, Index Data. * See the file LICENSE for details. - * Heikki Levanto - * + * $Id: isamd.c,v 1.12 1999-09-13 13:28:28 heikki Exp $ + * * Isamd - isam with diffs + * Programmed by: Heikki Levanto * + * Todo + * - Statistics are missing and/or completely wrong + * - Lots of code stolen from isamc, not all needed any more */ @@ -23,23 +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 keys before switching size. Unused time being */ - { 20, 40 }, - { 32, 0 }, +/* blocksz, max. Unused time being */ + { 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... @@ -51,13 +55,9 @@ ISAMD_M isamd_getmethod (ISAMD_M me) { 32768, 0 }, */ -/* assume about 2 bytes per pointer, when compressed. The head uses */ -/* 16 bytes, and other blocks use 8 for header info... If you want 3 */ -/* blocks of 32 bytes, say max 16+24+24 = 64 keys */ - }; - ISAMD_M m = (ISAMD_M) xmalloc (sizeof(*m)); /*??? never released??? */ - m->filecat = def_cat; + ISAMD_M m = (ISAMD_M) xmalloc (sizeof(*m)); /* never released! */ + m->filecat = def_cat; /* ok, only alloc'd once */ m->code_start = NULL; m->code_item = NULL; @@ -66,7 +66,7 @@ ISAMD_M isamd_getmethod (ISAMD_M me) m->compare_item = NULL; - m->debug = 1; + m->debug = 0; /* default to no debug */ m->max_blocks_mem = 10; @@ -89,11 +89,11 @@ ISAMD isamd_open (BFiles bfs, const char *name, int writeflag, ISAMD_M method) assert (filecat); /* determine number of block categories */ - if (is->method->debug) + if (is->method->debug>0) logf (LOG_LOG, "isamd: bsize maxkeys"); do { - if (is->method->debug) + if (is->method->debug>0) logf (LOG_LOG, "isamd:%6d %6d", filecat[i].bsize, filecat[i].mblocks); } while (filecat[i++].mblocks); @@ -104,13 +104,6 @@ ISAMD isamd_open (BFiles bfs, const char *name, int writeflag, ISAMD_M method) assert (is->max_cat <=8 ); /* we have only 3 bits for it */ is->files = (ISAMD_file) xmalloc (sizeof(*is->files)*is->no_files); - if (writeflag) - { - /* TODO - what ever needs to be done here... */ - } - else - { - } for (i = 0; ino_files; i++) { @@ -143,6 +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); } @@ -168,11 +168,12 @@ int isamd_close (ISAMD is) { int i; - if (is->method->debug) + 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 ? @@ -184,8 +185,8 @@ int isamd_close (ISAMD is) (double) is->files[i].sum_backward/is->files[i].no_backward : 0.0); } - if (is->method->debug) - logf (LOG_LOG, "isamd: writes reads skipped alloc released remap"); + if (is->method->debug>0) + logf (LOG_LOG, "f writes reads skipped alloc released "); for (i = 0; ino_files; i++) { release_fc (is, i); @@ -193,18 +194,38 @@ int isamd_close (ISAMD is) if (is->files[i].head_is_dirty) bf_write (is->files[i].bf, 0, 0, sizeof(ISAMD_head), &is->files[i].head); - if (is->method->debug) - logf (LOG_LOG, "isamd:%8d%8d%8d%8d%8d%8d", + if (is->method->debug>0) + 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); @@ -214,14 +235,16 @@ int isamd_close (ISAMD is) 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); return bf_read (is->files[cat].bf, pos, 0, 0, dst); } int isamd_write_block (ISAMD is, int cat, int pos, char *src) { ++(is->files[cat].no_writes); - if (is->method->debug > 2) - logf (LOG_LOG, "isamd: write_block %d %d", cat, pos); + if (is->method->debug > 6) + logf (LOG_LOG, "isamd: write_block %d:%d", cat, pos); return bf_write (is->files[cat].bf, pos, 0, 0, src); } @@ -229,7 +252,7 @@ int isamd_write_dblock (ISAMD is, int cat, int pos, char *src, int nextpos, int offset) { ISAMD_BLOCK_SIZE size = offset + ISAMD_BLOCK_OFFSET_N; - if (is->method->debug > 2) + if (is->method->debug > 4) logf (LOG_LOG, "isamd: write_dblock. size=%d nextpos=%d", (int) size, nextpos); src -= ISAMD_BLOCK_OFFSET_N; @@ -388,14 +411,14 @@ int isamd_alloc_block (ISAMD is, int cat) } if (!block) block = alloc_block (is, cat); - if (is->method->debug > 3) + if (is->method->debug > 4) logf (LOG_LOG, "isamd: alloc_block in cat %d: %d", cat, block); return block; } void isamd_release_block (ISAMD is, int cat, int pos) { - if (is->method->debug > 3) + if (is->method->debug > 4) logf (LOG_LOG, "isamd: release_block in cat %d: %d", cat, pos); assert(pos!=0); @@ -443,7 +466,7 @@ void isamd_pp_close (ISAMD_PP pp) isamd_free_diffs(pp); /* see merge-d.h */ xfree (pp->buf); xfree (pp); - if (is->method->debug > 2) + if (is->method->debug > 5) logf (LOG_LOG, "isamd_pp_close %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) ); @@ -471,7 +494,6 @@ ISAMD_PP isamd_pp_open (ISAMD is, ISAMD_P ipos) pp->decodeClientData = (*is->method->code_start)(ISAMD_DECODE); pp->numKeys = 0; pp->diffs=0; - pp->diffbuf=0; pp->diffinfo=0; @@ -490,12 +512,17 @@ ISAMD_PP isamd_pp_open (ISAMD is, ISAMD_P ipos) assert (pp->next != pp->pos); pp->offset = src - pp->buf; assert (pp->offset == ISAMD_BLOCK_OFFSET_1); -// if (is->method->debug > 2) -// 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) ); + if (0==pp->diffs) + ++(is->files[pp->cat].no_op_nodiff); + else + if(pp->diffs&1) + ++(is->files[pp->cat].no_op_extdiff); + else + ++(is->files[pp->cat].no_op_intdiff); + // if (!pp->diffbuf) + // pp->diffbuf=pp->buf; } - if (is->method->debug > 2) + 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) ); @@ -519,7 +546,7 @@ void isamd_buildfirstblock(ISAMD_PP pp){ memcpy(dst, &pp->diffs, sizeof(pp->diffs)); dst += sizeof(pp->diffs); assert (dst - pp->buf == ISAMD_BLOCK_OFFSET_1); - if (pp->is->method->debug > 2) + 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, @@ -536,7 +563,7 @@ void isamd_buildlaterblock(ISAMD_PP pp){ memcpy(dst, &pp->size,sizeof(pp->size)); dst += sizeof(pp->size); assert (dst - pp->buf == ISAMD_BLOCK_OFFSET_N); - if (pp->is->method->debug > 2) + if (pp->is->method->debug > 5) logf (LOG_LOG, "isamd: l8r: sz=%d p=%d/%d>%d/%d", pp->size, pp->pos, pp->cat, @@ -564,6 +591,7 @@ 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) { @@ -605,20 +633,24 @@ 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)); - //if (pp->deleteFlag) - // isamd_release_block (is, pp->cat, pp->pos); (*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 > 2) - 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; } @@ -650,23 +682,33 @@ 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]; 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 d=%d", + pp->numKeys, pp->offset, pp->size, pp->diffs); + 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; @@ -693,8 +735,20 @@ void isamd_pp_dump (ISAMD is, ISAMD_P ipos) /* * $Log: isamd.c,v $ - * Revision 1.7 1999-08-18 10:45:27 heikki - * Another fix for the difflen problem. + * 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 * Fixed memory leaks