Help: Using PDGESVD with different row/col. blocking factors

[David Amsallem]

I have installed scalapack version 1.8.

I have set up a simple 6x5 matrix distributed across 2x2 processor grid.

In the first case, I used a row/col blocking factor of 2x2 to distributed the matrix and I get the correct results.

However, when I change the blocking factor to 2x1, I get the following error message:

{ 0, 1}: On entry to PDGEBRD parameter number 606 had an illegal value
{ 1, 1}: On entry to PDGEBRD parameter number 606 had an illegal value
{ 0, 0}: On entry to PDGEBRD parameter number 606 had an illegal value
{ 1, 0}: On entry to PDGEBRD parameter number 606 had an illegal value

The nice thing about using the 6x5 matrix is that the structure of the submatrices is the same whether the column blocking factor is 1 or 2. So, I am a bit baffled as to why scalapack reports an error.

Does anyone have any ideas or can offer a way to figure this out? Or is this indeed a bug?

Thanks in advance

[Klystron]

The row and column blocking factorsMB_A and NB_A have to be the same when using PDGESVD. In the comments to this routine, you find

* The distributed submatrix sub( A ) must verify some alignment proper-
* ties, namely the following expressions should be true:
* ( MB_A.EQ.NB_A .AND. IROFFA.EQ.ICOFFA ) ...

Best wishes
Klystron