Hi

I'm trying to parallelize some lapack based code by replacing the dgbsv call with spike. The matrix is 10000x107 with kl=ku=53 and one rhs. I set

mat%format = 'D' ! dense banded matrix

mat%astru = 'G' ! general non symmetric

mat%diagdo = 'N' ! no diagonal dominance

and

mat%A => ab(kl+1:3*kl+1,:)

to skip the pivoting indices.

For some reason Spike fails to solve the system (summary below) while lapack does it in no time.

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

>>>>>>>>>>>>>>>>>>> SPIKE SUMMARY >>>>>>>>>>>>>>>>>>>>>

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

Spike Strategy RP3

TIME FOR PARTITIONING 1.879501342773438e-02

TIME FOR SPIKE BANDED FACT 4.340887069702148e-02

TIME FOR SPIKE BANDED SOLV 5.887868404388428e-01

TIME FOR SPIKE (FACT+SOLV) 6.321957111358643e-01

RESIDUAL 2.132312888050483e+18

# Outside iterations: 51

SPIKE has failed (to reach the accuracy pspike%eps_out)

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

I've also tried autoadapt=true btw. which chooses EA3 but gives me "zero pivots" in addition to the above.

I must say that I'm fairly new to fortran and that I don't understand the math behind this code. (But I do have some experience withparallel programming, so the MPI part should be alright.)

I'm really running out of ideas here. If anybody has a hint of what might be wrong or where I should start looking, I'd appreciate it.

lukas