LAPACK Archives

[Lapack] Different answers on second run


when you read DGESV header, the matrix A and the vector B are flagged as 
inpout/output. In input A is your matrix, B is your set of right-hand 
sides. In output, A is overwritren by the L and U factors and B is 
overwritten by the solution.

If you call it twice DGESV without changing anything., this going to do 
something weird and for sure the second output vector as nothing to do 
with the one from the first run.

Did you miss this?


On Fri, 2 Mar 2007, Abe Krebs wrote:


I am trying to use the dgesv routine from version 3.0 of the CLAPACK.  I
have a case where upon a second call to the routine, passing in the same
inputs I get different output vector.  The memory space has not changed
between runs, and this is important.  If I call the routine with a copy of
the data, the results are the same as if I ran it for the first time.  I
believe the results are correct for the second run.  Is there some global
memory that needs to be initialized for the routine to work properly?  Am I
missing something obvious?  I have been through the memory allocations and
the inputs with a fine comb and have come across no issues.

Best Regards,
Abe Krebs

<Prev in Thread] Current Thread [Next in Thread>

For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or