执行块LU分解

Question

我正在尝试执行图块LU分解，但是，当我选择图块siz（nbtiles）与矩阵大小（nbtiles = 1）相同时，它将产生正确的答案（这意味着函数Bllu中使用的是正确的），但是当选择不同的图块大小（例如nbtiles = 2）时，它将生成矩阵A但值错误

% A is a symmetric matrix
A = [0.9 0.06 -0.39 -0.24; 0.06 1.604 0.134 0.464; -0.39 0.134 2.685 0.802; -0.24 0.464 0.802 1.977];
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
nbtiles = 1;
[nbrow nbcol]=size(A);

   mink = zeros(size(nbtiles,1),size(acc,1));

for indtiles=1:size(nbtiles,1)
        % Init mink array
        mink(indtiles,:)=nbtiles(indtiles);

        % Number of tiles
        p=floor(nbrow/nbtiles(indtiles))
        q=floor(nbcol/nbtiles(indtiles))

        rows=nbtiles(indtiles)*ones(1,p);
        if mod(nbrow,nbtiles(indtiles)) ~= 0
           rows=[rows mod(nbrow,nbtiles(indtiles))];
           p=p+1;
        end 
        rows;
        cols=nbtiles(indtiles)*ones(1,q);
        if mod(nbcol,nbtiles(indtiles)) ~= 0
           cols=[cols mod(nbcol,nbtiles(indtiles))];
           q=q+1;
        end 
        cols;

        % Generate the tile format
        B = cell(p,q);
        B = mat2cell(A,rows,cols);

end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

[A]= Bllu(B, p)

function [ AA] = Bllu(A, p )
%for each panel
   %*************************  dpotrf  *************************
   for k = 1 : p,
        [L U] = lu(sparse(A{k,k}),0); %dpotrf(Akk,oAkk)
   %*************************  TRSM Lower *************************
    for i = (k+1) : p, 
        A{i,k} = dtrsm(A{i,k}, U);      
    end
     %*************************  TRSM Upper *************************
   for j = (k+1) : p, 
          A{k,j} = dtrsm( A{k,j},L);  
    end    
   %*************************  dia  *************************
    for j = (k+1) : p,
         A{j,j}  = dgemm(A{j,j}, A{j,k}, A{k,j}, -1);
   %*************************  gemm  *************************
        for i = (j+1):p,  
            A{i,j} = dgemm(A{i,j}, A{i,k}, A{k,j}, -1);
            A{j,i} = dgemm(A{j,i}, A{k,i}, A{j,k}, -1);
        end
    end
end
AA  = A;

function [ X ] = dtrsm( A, B )
[m,n] = size(B); 
X = A;
for i = 1 : m,
    for k = 1 : m,
        X(k,i) = A(k,i) / B(i,i);
    end
    for j = (i+1) : m,
        for k = 1 : m,
            A(k,j) = A(k,j) - X(k,i) * B(j,i);
        end
    end
end

function [oC] = dgemm(iC, iA, iB, ialpha)
    oC = iC + ialpha * iA * iB';

end