cholesky decomposition ScaLapack error

Question

I\'m getting the following error and i\'m not sure why.

{    1,    1}:  On entry to PDPOTRF parameter number    2 had an illegal value
{    1,    0}:  On ent         


        

Answer 1

Right - I solved this. Here's what you have to do (I checked the result of the modified MPI program against a Cholesky decomp of your matrix in Octave -- it works.).

I found the following LAPACK reference by IBM to be more helpful than the one in your link: http://publib.boulder.ibm.com/infocenter/clresctr/vxrx/index.jsp?topic=%2Fcom.ibm.cluster.pessl.v4r2.pssl100.doc%2Fam6gr_lpotrf.htm

PDPOTRF( UPLO, N, A, IA, JA, DESCA, INFO )

You are passing Mb and Nb as IA and JA. However, those parameters are meant to provide the starting row and column of your global matrix inside a larger matrix. They are only relevant if you have a big matrix and only want the Cholesky decomp of a submatrix. In your case, IA and JA both have to be 1!

So all you need to do is:

int IA = 1;
int JA = 1;
pdpotrf_("L",&ord,A_loc,&IA,&JA,descA,&info);

You may also want to change your hardcoded int ord = 8; so that it depends on the value read from the command line, otherwise you'll run into problems later.

Output of your program modified as described above:

Cholesky MPI Run Time0.000659943
SUCCESS
Matrix A test:
13.4907 147 140 125 132  76 126 157 
10.8964 9.70923 185 150 209 114 166 188 
10.3775 7.4077 8.33269 129 194 142 199 205 
9.26562 5.0507 -0.548194 5.59806 148  81 104 150 
9.78449 10.5451 1.72175 0.897537 1.81524 122 172 189 
5.63349 5.41911 5.20784 0.765767 0.0442447 3.63139 129 117 
9.33974 6.61543 6.36911 -2.22569 1.03941 2.48498 1.79738 181 
11.6376 6.30249 4.50561 2.28799 -0.627688 -2.17633 7.27182 0.547228 

Octave output for comparison:

octave:1> A=dlmread("matrixfile4")
A =

   182   147   140   125   132    76   126   157
   147   213   185   150   209   114   166   188
   140   185   232   129   194   142   199   205
   125   150   129   143   148    81   104   150
   132   209   194   148   214   122   172   189
    76   114   142    81   122   102   129   117
   126   166   199   104   172   129   187   181
   157   188   205   150   189   117   181   259

octave:2> C=chol(A)
C =

   13.49074   10.89636   10.37749    9.26562    9.78449    5.63349    9.33974   11.63761
    0.00000    9.70923    7.40770    5.05070   10.54508    5.41911    6.61543    6.30249
    0.00000    0.00000    8.33269   -0.54819    1.72175    5.20784    6.36911    4.50561
    0.00000    0.00000    0.00000    5.59806    0.89754    0.76577   -2.22569    2.28799
    0.00000    0.00000    0.00000    0.00000    1.81524    0.04424    1.03941   -0.62769
    0.00000    0.00000    0.00000    0.00000    0.00000    3.63139    2.48498   -2.17633
    0.00000    0.00000    0.00000    0.00000    0.00000    0.00000    1.79738    7.27182
    0.00000    0.00000    0.00000    0.00000    0.00000    0.00000    0.00000    0.54723

(My earlier comment -now deleted- about matrices being wrongly copied to the nodes does not apply, the rest of your program seems fine to me.)