This repository has been archived on 2023-08-20. You can view files and clone it, but cannot push or open issues or pull requests.
yap-6.3/packages/cuda/joincpu.cpp

501 lines
11 KiB
C++
Raw Normal View History

2016-04-22 18:21:05 +01:00
#include "CC_CSSTree.h"
#include <vector>
#include <thrust/sort.h>
#include <thrust/system/omp/execution_policy.h>
#include "pred.h"
void partInlj(Record *R, int rLen, CC_CSSTree *tree, Record *S, int startS, int endS, int of1, int of2, vector<int> *res, int *p1, int *p2, int *perm, int *proj, int wj, int halfrul, int lenrul)
{
//set_thread_affinity(cpuid,NUM_T);
int i=0;
int k=0;
int curIndex=0;
int keyForSearch;
int y, posS, posR;
for(k=startS; k<endS; k++)
{
if(S == NULL)
posS = k * of2;
else
posS = S[k] * of2;
keyForSearch=p2[posS + wj];
curIndex=tree->search(keyForSearch);
for(i=curIndex-1;i>0;i--)
{
if(keyForSearch == R[i])
{
//cout << keyForSearch << endl;
posR = perm[i] * of1;
for(y = 0; y < halfrul; y++)
res->push_back(p1[posR + proj[y]]);
for(; y < lenrul; y++)
res->push_back(p2[posS + proj[y]]);
}
else
if(R[i]<keyForSearch)
break;
}
for(i=curIndex;i<rLen;i++)
{
if(keyForSearch == R[i])
{
//cout << -i << " " << keyForSearch << endl;
posR = perm[i] * of1;
for(y = 0; y < halfrul; y++)
res->push_back(p1[posR + proj[y]]);
for(; y < lenrul; y++)
res->push_back(p2[posS + proj[y]]);
}
else
if(R[i]>keyForSearch)
break;
}
}
}
void partInlj2(Record *R, int rLen, CC_CSSTree *tree, Record *S, int startS, int endS, int of1, int of2, vector<int> *res, int *p1, int *p2, int *perm, int *proj, int cols, int wj)
{
//set_thread_affinity(cpuid,NUM_T);
int i=0;
int k=0;
int curIndex=0;
int keyForSearch;
int y, cond, posS, posR;
for(k=startS; k<endS; k++)
{
if(S == NULL)
posS = k * of2;
else
posS = S[k] * of2;
keyForSearch=p2[posS + wj];
curIndex=tree->search(keyForSearch);
for(i=curIndex-1;i>0;i--)
{
if(keyForSearch == R[i])
{
//cout << keyForSearch << endl;
posR = perm[i] * of1 - 1;
for(y = 0; y < cols; y++)
{
cond = proj[y];
if(cond > 0)
res->push_back(p1[posR + cond]);
else
res->push_back(p2[posS - cond - 1]);
}
}
else
if(R[i]<keyForSearch)
break;
}
for(i=curIndex;i<rLen;i++)
{
if(keyForSearch == R[i])
{
//cout << -i << " " << keyForSearch << endl;
posR = perm[i] * of1 - 1;
for(y = 0; y < cols; y++)
{
cond = proj[y];
if(cond > 0)
res->push_back(p1[posR + cond]);
else
res->push_back(p2[posS - cond - 1]);
}
}
else
if(R[i]>keyForSearch)
break;
}
}
}
void multipartInlj(Record *R, int rLen, CC_CSSTree *tree, Record *S, int startS, int endS, int of1, int of2, vector<int> *res, int *p1, int *p2, int *perm, int *proj, int *wj, int numj, int halfrul, int lenrul)
{
//set_thread_affinity(cpuid,NUM_T);
int i=0;
int k=0;
int curIndex=0;
int keyForSearch;
int y, posS, posR;
for(k=startS; k<endS; k++)
{
if(S == NULL)
posS = k * of2;
else
posS = S[k] * of2;
keyForSearch=p2[posS + wj[1]];
curIndex=tree->search(keyForSearch);
for(i=curIndex-1;i>0;i--)
{
if(keyForSearch == R[i])
{
posR = perm[i] * of1;
for(y = 2; y < numj; y += 2)
{
if(p1[posR + wj[y]] != p2[posS + wj[y+1]])
break;
}
if(y < numj)
continue;
for(y = 0; y < halfrul; y++)
res->push_back(p1[posR + proj[y]]);
for(; y < lenrul; y++)
res->push_back(p2[posS + proj[y]]);
}
else
if(R[i]<keyForSearch)
break;
}
for(i=curIndex;i<rLen;i++)
{
if(keyForSearch == R[i])
{
posR = perm[i] * of1;
for(y = 2; y < numj; y += 2)
{
if(p1[posR + wj[y]] != p2[posS + wj[y+1]])
break;
}
if(y < numj)
continue;
for(y = 0; y < halfrul; y++)
res->push_back(p1[posR + proj[y]]);
for(; y < lenrul; y++)
res->push_back(p2[posS + proj[y]]);
}
else
if(R[i]>keyForSearch)
break;
}
}
}
void multipartInlj2(Record *R, int rLen, CC_CSSTree *tree, Record *S, int startS, int endS, int of1, int of2, vector<int> *res, int *p1, int *p2, int *perm, int *proj, int cols, int *wj, int numj)
{
//set_thread_affinity(cpuid,NUM_T);
int i=0;
int k=0;
int curIndex=0;
int keyForSearch;
int y, cond, posS, posR;
for(k=startS; k<endS; k++)
{
if(S == NULL)
posS = k * of2;
else
posS = S[k] * of2;
keyForSearch=p2[posS + wj[1]];
curIndex=tree->search(keyForSearch);
for(i=curIndex-1;i>0;i--)
{
if(keyForSearch == R[i])
{
posR = perm[i] * of1;
for(y = 2; y < numj; y += 2)
{
if(p1[posR + wj[y]] != p2[posS + wj[y+1]])
break;
}
if(y < numj)
continue;
for(y = 0; y < cols; y++)
{
cond = proj[y];
if(cond > 0)
res->push_back(p1[posR + cond - 1]);
else
res->push_back(p2[posS - cond - 1]);
}
}
else
if(R[i]<keyForSearch)
break;
}
for(i=curIndex;i<rLen;i++)
{
if(keyForSearch == R[i])
{
posR = perm[i] * of1;
for(y = 2; y < numj; y += 2)
{
if(p1[posR + wj[y]] != p2[posS + wj[y+1]])
break;
}
if(y < numj)
continue;
for(y = 0; y < cols; y++)
{
cond = proj[y];
if(cond > 0)
res->push_back(p1[posR + cond - 1]);
else
res->push_back(p2[posS - cond - 1]);
}
}
else
if(R[i]>keyForSearch)
break;
}
}
}
void inlj_omp(Record *R, int rLen, CC_CSSTree *tree, Record *S, int sLen, int of1, int of2, vector<int> *res, int *p1, int *p2, int *perm, int *proj, int2 projp, int cols, int* wj, int numj, int tipo)
{
int i=0;
int j=0;
int *startS=new int[NUM_T];
int *endS=new int[NUM_T];
int chunkSize=sLen/NUM_T;
for(i=0;i<NUM_T;i++)
{
startS[i]=i*chunkSize;
if(i==(NUM_T-1))
endS[i]=sLen;
else
endS[i]=(i+1)*chunkSize;
//cout<<"T"<<i<<", "<<endS[i]-startS[i]<<"; ";
}
//cout<<endl;
//omp_set_num_threads(NUM_T);
//cout << "inicio" << endl;
#pragma omp parallel for
for(j=0;j<NUM_T;j++)
{
if(tipo)
{
if(numj > 2)
multipartInlj2(R, rLen, tree, S, startS[j], endS[j], of1, of2, &res[j], p1, p2, perm, proj, cols, wj, numj);
else
partInlj2(R, rLen, tree, S, startS[j], endS[j], of1, of2, &res[j], p1, p2, perm, proj, cols, wj[1]);
}
else
{
if(numj > 2)
multipartInlj(R, rLen, tree, S, startS[j], endS[j], of1, of2, &res[j], p1, p2, perm, proj, wj, numj, projp.x, projp.y);
else
partInlj(R, rLen, tree, S, startS[j], endS[j], of1, of2, &res[j], p1, p2, perm, proj, wj[1], projp.x, projp.y);
}
}
//cout << "fin" << endl;
delete startS;
delete endS;
}
int joincpu(int *p1, int *p2, int rLen, int sLen, int of1, int of2, list<rulenode>::iterator rule, int pos, int bothops, int **ret)
{
int pos2 = pos + 1;
int *sel1, nsel1 = 0;
int *sel2 = rule->select[pos2];
int nsel2 = rule->numsel[pos2];
int *proj = rule->project[pos];
int2 projp = rule->projpos[pos];
int *sjoin1, nsj1 = 0;
int *sjoin2 = rule->selfjoin[pos2];
int nsj2 = rule->numselfj[pos2];
int *wherej = rule->wherejoin[pos];
int numj = rule->numjoin[pos];
int size, *fres, ini[NUM_T], *temp;
int x, tipo = 0;
int *Sres = NULL, *Rres, Snl, Rnl, *permutation;
if(bothops)
{
sel1 = rule->select[pos];
nsel1 = rule->numsel[pos];
sjoin1 = rule->selfjoin[pos];
nsj1 = rule->numselfj[pos];
}
#ifdef TIMER
2016-07-31 16:14:02 +01:00
hipEvent_t start, stop;
2016-04-22 18:21:05 +01:00
float time;
2016-07-31 16:14:02 +01:00
hipEventCreate(&start);
hipEventCreate(&stop);
hipEventRecord(start, 0);
2016-04-22 18:21:05 +01:00
#endif
if(nsel1 > 0 || nsj1 > 0)
Rnl = selectproyectcpu2(p1, rLen, of1, sel1, nsel1, sjoin1, nsj1, wherej[0], &Rres, &permutation);
else
{
/*cout << "sin sel" << endl;
cout << "valores = " << rLen << " " << of1 << " " << wherej[0] << endl;
for(x = 0; x < 100; x++)
cout << p1[x] << " ";
cout << endl;
cout << "ultimo = " << p1[of1 * rLen - 1] << endl;*/
Rnl = rLen;
size = Rnl * sizeof(int);
permutation = (int *)malloc(size);
Rres = (int *)malloc(size);
#pragma omp parallel for firstprivate(of1)
for(x = 0; x < Rnl; x++)
{
permutation[x] = x;
Rres[x] = p1[of1 * x + wherej[0]];
}
//cout << "sin sel fin" << endl;
}
#ifdef TIMER
2016-07-31 16:14:02 +01:00
hipEventRecord(stop, 0);
hipEventSynchronize(stop);
hipEventElapsedTime(&time, start, stop);
2016-04-22 18:21:05 +01:00
cuda_stats.select1_time += time;
2016-07-31 16:14:02 +01:00
hipEventDestroy(start);
hipEventDestroy(stop);
hipEventCreate(&start);
hipEventCreate(&stop);
hipEventRecord(start, 0);
2016-04-22 18:21:05 +01:00
#endif
if(nsel2 > 0 || nsj2 > 0)
{
//cout << "con sel S" << endl;
Snl = selectproyectcpu2(p2, sLen, of2, sel2, nsel2, sjoin2, nsj2, wherej[1], &Sres, NULL);
}
else
Snl = sLen;
#ifdef TIMER
2016-07-31 16:14:02 +01:00
hipEventRecord(stop, 0);
hipEventSynchronize(stop);
hipEventElapsedTime(&time, start, stop);
2016-04-22 18:21:05 +01:00
cuda_stats.select2_time += time;
2016-07-31 16:14:02 +01:00
hipEventDestroy(start);
hipEventDestroy(stop);
hipEventCreate(&start);
hipEventCreate(&stop);
hipEventRecord(start, 0);
2016-04-22 18:21:05 +01:00
#endif
//cout << "antes" << endl;
/*cout << "antes" << endl;
for(x = 0; x < Rnl; x++)
cout << permutation[x] << " ";
cout << endl;
for(x = 0; x < 100; x++)
cout << Rres[x] << " ";
cout << endl;*/
thrust::stable_sort_by_key(thrust::omp::par, Rres, Rres + Rnl, permutation);
#ifdef TIMER
2016-07-31 16:14:02 +01:00
hipEventRecord(stop, 0);
hipEventSynchronize(stop);
hipEventElapsedTime(&time, start, stop);
2016-04-22 18:21:05 +01:00
cuda_stats.sort_time += time;
2016-07-31 16:14:02 +01:00
hipEventDestroy(start);
hipEventDestroy(stop);
hipEventCreate(&start);
hipEventCreate(&stop);
hipEventRecord(start, 0);
2016-04-22 18:21:05 +01:00
#endif
/*cout << "despues" << endl;
for(x = 0; x < Rnl; x++)
cout << permutation[x] << " ";
cout << endl;
for(x = 0; x < Rnl; x++)
cout << Rres[x] << " ";
cout << endl;*/
//cout << "despues sort" << endl;
vector<int> *res = new vector<int>[NUM_T];
for(x = 0; x < NUM_T; x++)
res[x].reserve(INISIZE);
CC_CSSTree *tree = new CC_CSSTree(Rres, Rnl, CSS_TREE_FANOUT);
if(pos == (rule->num_rows - 3)) // && rule->num_bpreds.x == 0)
tipo = 1;
inlj_omp(Rres, Rnl, tree, Sres, Snl, of1, of2, res, p1, p2, permutation, proj, projp, rule->num_columns, wherej, numj, tipo);
/*cout << "proj = ";
for(x = 0; x < rule->num_columns; x++)
cout << proj[x] << " ";
cout << endl;
int y,z;
for(x = 0; x < NUM_T; x++)
{
cout << "Thread " << x << endl;
for(y = 0; y < res[x].size() / projp.y; y++)
{
for(z = 0; z < projp.y; z++)
cout << res[x][y * projp.y + z] << " ";
cout << endl;
}
}
cout << "Tamanios" << endl;*/
size = 0;
for(x = 0; x < NUM_T; x++)
{
ini[x] = res[x].size();
size += ini[x];
//cout << ini[x] << " " << size << endl;
}
fres = (int *)malloc(size * sizeof(int));
temp = fres;
for(x = 0; x < NUM_T; x++)
{
2018-06-30 14:33:32 +01:00
memmove(temp, res[x].data(), ini[x] * sizeof(int));
2016-04-22 18:21:05 +01:00
temp += ini[x];
}
if(*ret != NULL)
free(*ret);
free(Rres);
free(permutation);
if(Sres != NULL)
free(Sres);
delete tree;
delete [] res;
*ret = fres;
#ifdef TIMER
2016-07-31 16:14:02 +01:00
hipEventRecord(stop, 0);
hipEventSynchronize(stop);
hipEventElapsedTime(&time, start, stop);
2016-04-22 18:21:05 +01:00
cuda_stats.join_time += time;
#endif
//cout << "Projp.x = " << projp.x << " projp.y = " << projp.y << endl;
/*if(numj > 2)
{
cout << "total = " << rLen << " " << size / projp.y << " " << projp.y << " " << rule->num_columns << endl;
exit(1);
}*/
return size / projp.y;
}