稀疏对称矩阵的LDLT分解在GPU上的高效实现

陈鑫峰,王武

An Effective Implementation of LDLT Decomposition of Sparse Symmetric Matrix on GPU

Chen Xinfeng,Wang Wu

伪代码2. 动态并行分解核函数

1: function dynamic(Lp, Li, Lx, level_p, level_i, tmpMem, tmpMem1, n, level, offset) 2: k = level_i[level_p[level]+offset+blockIdx.x]; 3: d = Lx[Lp[k]]; 4: if abs(d)<1e-5 then 5: Lx[Lp[k]]=1e-5; 6: d=1e-5; 7: end if 8: subColSize=Lp[k+1]-Lp[k]-1; 9: factorize<<<(subColSize+1023)/1024,1024>>> (Lp,Li,Lx,tmpMem,tmpMem1,d,n,k,blockIdx.x); 10: update<<<subColSize,1024>>>(Lp,Li,Lx, tmpMem,tmpMem1,n,k,blockIdx.x); 11: cleartmpMem<<<(subColSize+1023)/1024, 1024>>>(Lp,Li,tmpMem,tmpMem1,n,k,blockIdx.x); 12: end function