1: function LDLT(Lp, Li, Lx, level_p, level_i)
2:分配GPU内存,将数据从CPU端复制到GPU端
3: cudaMemGetInfo(&free, &total);
4: free -= 1024ull*1024ull*1024ull; //预留1GB的空闲内存
5: free /= (2*n*sizeof(数据类型));
6: for level = 0 to nlevel - 1 do
7: //level 级别的并行
8: lev_size = level_p[level+1]-level_p[level];
9: tmp = min(lev_size,free);
10: //GPU上为tmpMem1和tmpMem分配全局内存
11: offset = 0;
12: while lev_size >0 do
13: restCol = min(lev_size tmp);
14: dynamic<<<restCol, 1>>>(Lp, Li, Lx, level_p,
level_i, tmpMem, tmpMem1, n,level,offset);
15: lev_size -= tmp;
16: offset += tmp;
17: end while
18: 释放tmpMem
19: end for
20:从GPU端到CPU端拷贝数据,释放GPU内存 |
|