mirror of
https://github.com/gigablast/open-source-search-engine.git
synced 2024-10-04 20:27:43 +03:00
249 lines
7.1 KiB
C++
249 lines
7.1 KiB
C++
#include "gb-include.h"
|
|
|
|
#include <sys/time.h> // gettimeofday()
|
|
#include "sort.h"
|
|
|
|
static int cmp (const void *p1, const void *p2) ;
|
|
|
|
|
|
class fslot1 {
|
|
public:
|
|
long m_score;
|
|
char *m_docIdBitsPtr;
|
|
unsigned short m_termBits;
|
|
//unsigned short align;
|
|
};
|
|
|
|
class fslot2 {
|
|
public:
|
|
long m_score;
|
|
long long m_docIdBits;
|
|
unsigned short m_termBits;
|
|
//unsigned short align;
|
|
};
|
|
|
|
static long long gettimeofdayInMilliseconds() ;
|
|
|
|
long long gettimeofdayInMilliseconds() {
|
|
struct timeval tv;
|
|
gettimeofday ( &tv , NULL );
|
|
long long now=(long long)(tv.tv_usec/1000)+((long long)tv.tv_sec)*1000;
|
|
return now;
|
|
}
|
|
|
|
|
|
main ( ) {
|
|
// fill our tbl
|
|
unsigned long g_hashtab[256];
|
|
static bool s_initialized = false;
|
|
// bail if we already called this
|
|
if ( s_initialized ) return true;
|
|
// show RAND_MAX
|
|
//printf("RAND_MAX = %lu\n", RAND_MAX ); it's 0x7fffffff
|
|
// seed with same value so we get same rand sequence for all
|
|
srand ( 1945687 );
|
|
for ( long i = 0 ; i < 256 ; i++ )
|
|
g_hashtab [i] = (unsigned long)rand();
|
|
|
|
// . # of docIds to hash
|
|
long nd = 2000000; //8192*4*4;
|
|
// make a list of compressed (6 byte) docIds
|
|
char *docIds1 = (char *) malloc ( 6 * nd );
|
|
char *docIds2 = (char *) malloc ( 6 * nd );
|
|
// print start time
|
|
//fprintf (stderr,"hashtest:: randomizing begin."
|
|
// " %li 6-byte docIds.\n",nd);
|
|
// randomize docIds
|
|
unsigned char *p = (unsigned char *)docIds1;
|
|
for ( long i = 0 ; i < nd ; i++ ) {
|
|
*(unsigned long *)p = rand() | 0x01 ; p += 4;
|
|
*(unsigned short *)p = rand() % 0xffff ; p += 2;
|
|
}
|
|
p = (unsigned char *)docIds2;
|
|
for ( long i = 0 ; i < nd ; i++ ) {
|
|
*(unsigned long *)p = rand() | 0x01 ; p += 4;
|
|
*(unsigned short *)p = rand() % 0xffff ; p += 2;
|
|
}
|
|
// . make a hash table, small...
|
|
// . numSlots MUST be power of 2
|
|
// . ENSURE this fits into the L1 or L2 cache!!!!!!!!!!!!!!
|
|
// . That is the whole POINT of this!!!!!!!
|
|
// . This seems to run fastest at about 2k, going much less than
|
|
// that doesn't seem to help much...
|
|
#define numSlots 1024*2
|
|
#define mask (numSlots-1)
|
|
//long numSlots = 1024*4;
|
|
fprintf(stderr,"numslots = %lik\n", numSlots/1024);
|
|
fslot1 *slots = (fslot1 *) calloc (sizeof(fslot1), numSlots );
|
|
// point to 6 byte docIds
|
|
p = (unsigned char *)docIds1;
|
|
// set our total end
|
|
unsigned char *max = p + nd*6;
|
|
// only hash enough to fill up our small hash table
|
|
long score;
|
|
long collisions = 0;
|
|
unsigned long n;
|
|
unsigned short termBitMask = 1;
|
|
//unsigned long mask = numSlots - 1;
|
|
long scoreWeight = 13;
|
|
// debug msg
|
|
fprintf (stderr,"hashtest:: starting loop (nd=%li)\n",nd);
|
|
// time stamp
|
|
long long t = gettimeofdayInMilliseconds();
|
|
again:
|
|
// advance the ending point before clearing the hash table
|
|
unsigned char *pend = p + (numSlots>>1)*6;
|
|
if ( pend >= max ) pend = max;
|
|
if ( p >= max ) goto finalDone;
|
|
// skip clear of table if we don't need to do it
|
|
if ( p == (unsigned char *)docIds1 ) goto top;
|
|
// otherwise msg
|
|
//fprintf(stderr,"looping\n");
|
|
// empty hash table
|
|
for ( long i = 0 ; i < numSlots ; i++ )
|
|
slots[i].m_score = 0;
|
|
top:
|
|
if ( p >= pend ) goto done;
|
|
//*(((unsigned char *)(&docIdBits))+0) = p[0];
|
|
//*(((unsigned char *)(&docIdBits))+1) = p[1];
|
|
//*(((unsigned char *)(&docIdBits))+2) = p[2];
|
|
//*(((unsigned char *)(&docIdBits))+3) = p[3];
|
|
//*(((unsigned char *)(&docIdBits))+4) = p[4];
|
|
n = ( (*(unsigned long *)p) ^ g_hashtab[p[0]] ) & mask;
|
|
chain:
|
|
if ( slots[n].m_score == 0 ) {
|
|
slots[n].m_score = ~p[5];
|
|
slots[n].m_docIdBitsPtr = (char *)p;
|
|
slots[n].m_termBits = termBitMask;
|
|
p += 6;
|
|
goto top;
|
|
}
|
|
// if equal, add
|
|
if ( *(long *)p == *slots[n].m_docIdBitsPtr &&
|
|
p[5] == slots[n].m_docIdBitsPtr[5] ) {
|
|
slots[n].m_score += ~p[5];
|
|
slots[n].m_termBits |= termBitMask;
|
|
p += 6;
|
|
goto top;
|
|
}
|
|
// otherwise, chain
|
|
//collisions++;
|
|
if ( ++n >= (unsigned long)numSlots ) n = 0;
|
|
goto chain;
|
|
|
|
done:
|
|
// dec loopcount
|
|
goto again;
|
|
|
|
finalDone:
|
|
// completed
|
|
long long now = gettimeofdayInMilliseconds();
|
|
fprintf (stderr,"hashtest:: addList took %llu ms\n" , now - t );
|
|
// how many did we hash
|
|
long hashed = (p - (unsigned char *)docIds1) / 6;
|
|
fprintf(stderr,"hashtest:: hashed %li docids\n", hashed);
|
|
// stats
|
|
double d = (1000.0*(double)hashed) / ((double)(now - t));
|
|
fprintf (stderr,"hashtest:: each add took %f cycles\n" ,
|
|
400000000.0 / d );
|
|
fprintf (stderr,"hashtest:: we can do %li adds per second\n" ,(long)d);
|
|
fprintf (stderr,"hashtest:: collisions = %li\n", collisions);
|
|
|
|
//////////////////////////////////////////////
|
|
//////////////////////////////////////////////
|
|
// now do merge test of the lists
|
|
//////////////////////////////////////////////
|
|
//////////////////////////////////////////////
|
|
|
|
// sort docIds1 and docIds2
|
|
fprintf(stderr,"hashtest:: sorting docIds1 and docIds2 for merge\n");
|
|
gbsort ( docIds1 , nd , 6 , cmp );
|
|
gbsort ( docIds2 , nd , 6 , cmp );
|
|
|
|
// current ptrs
|
|
unsigned char *p1 = (unsigned char *)docIds1;
|
|
unsigned char *p2 = (unsigned char *)docIds2;
|
|
|
|
unsigned char *pend1 = (unsigned char *)(docIds1 + nd * 6);
|
|
unsigned char *pend2 = (unsigned char *)(docIds2 + nd * 6);
|
|
|
|
// list to hold them
|
|
//char *results = (char *) malloc ( nd * 6 * 2 );
|
|
//char *pr = results ;
|
|
|
|
t = gettimeofdayInMilliseconds();
|
|
|
|
// pt to slots
|
|
fslot2 *pr = (fslot2 *) calloc (sizeof(fslot2), nd * 2 );
|
|
loop:
|
|
// now merge the two lists sorted by docId
|
|
if ( p1[5] < p2[5] ) {
|
|
*(long *)(&pr->m_docIdBits) = *(long *)p1;
|
|
*(((short *)(&pr->m_docIdBits))+2) = *(short *)(p1+4);
|
|
pr->m_score = ~p1[5];
|
|
pr->m_termBits = 0x01;
|
|
pr++;;
|
|
p1+=6;
|
|
if ( p1 < pend1 ) goto loop;
|
|
goto done2;
|
|
}
|
|
if ( p1[5] > p2[5] ) {
|
|
*(long *)(&pr->m_docIdBits) = *(long *)p2;
|
|
*(((short *)(&pr->m_docIdBits))+2) = *(short *)(p2+4);
|
|
pr->m_score = ~p2[5];
|
|
pr->m_termBits = 0x02;
|
|
pr++;
|
|
p2+=6;
|
|
if ( p2 < pend2 ) goto loop;
|
|
goto done2;
|
|
}
|
|
// add together
|
|
*(long *)(&pr->m_docIdBits) = *(long *)p2;
|
|
*(((short *)(&pr->m_docIdBits))+2) = *(short *)(p2+4);
|
|
pr->m_score = ((long)~p1[5]) + ((long)~p2[5]) ;
|
|
pr->m_termBits = 0x01 | 0x02;
|
|
pr++;
|
|
p2 += 6;
|
|
p1 += 6;
|
|
if ( p2 >= pend2 ) goto done2;
|
|
if ( p1 >= pend1 ) goto done2;
|
|
goto loop;
|
|
|
|
done2:
|
|
|
|
// completed
|
|
now = gettimeofdayInMilliseconds();
|
|
fprintf (stderr,"hashtest:: 2 list MERGE took %llu ms\n" , now - t );
|
|
// how many did we hash
|
|
fprintf(stderr,"hashtest:: merged %li docids\n", nd*2);
|
|
// stats
|
|
d = (1000.0*(double)nd*2.0) / ((double)(now - t));
|
|
fprintf (stderr,"hashtest:: each add took %f cycles\n" ,
|
|
400000000.0 / d );
|
|
fprintf (stderr,"hashtest:: we can do %li adds per second\n" ,(long)d);
|
|
|
|
// exit gracefully
|
|
exit ( 0 );
|
|
}
|
|
|
|
int cmp (const void *p1, const void *p2) {
|
|
|
|
|
|
long long n1 = 0;
|
|
long long n2 = 0;
|
|
|
|
*(((unsigned char *)(&n1))+0) = *(((char *)p1)+0);
|
|
*(((unsigned char *)(&n1))+1) = *(((char *)p1)+1);
|
|
*(((unsigned char *)(&n1))+2) = *(((char *)p1)+2);
|
|
*(((unsigned char *)(&n1))+3) = *(((char *)p1)+3);
|
|
*(((unsigned char *)(&n1))+4) = *(((char *)p1)+4);
|
|
|
|
*(((unsigned char *)(&n2))+0) = *(((char *)p2)+0);
|
|
*(((unsigned char *)(&n2))+1) = *(((char *)p2)+1);
|
|
*(((unsigned char *)(&n2))+2) = *(((char *)p2)+2);
|
|
*(((unsigned char *)(&n2))+3) = *(((char *)p2)+3);
|
|
*(((unsigned char *)(&n2))+4) = *(((char *)p2)+4);
|
|
|
|
return n1 - n2;
|
|
}
|