ScaLAPACK & ACML

[Reuti]

Although this is not a list forum from AMD, it would be interesting to know, whether anyone else experienced the same:

While compiling ScaLAPACK 1.7.5 I found something strange. I got three errors in its testsuite with PGI6.1/ACML3.5, two errors wth PGI6.2/ACML3.6. Then I turned to ATLAS 3.6.0 and an also self compiled LAPACK 3.1.0 with the result, that there was only one complain about an illegal parameter to a function call (always -fast -Kieee used for PGI). I'm still waiting for an answer from AMD.

Anyone used Intel's Compiler & MKL for a self-compiled ScaLAPACK? Or found something similar while using ACML with any program?

-- Reuti

[Julie]

Reuti
Could you tell us in which testing the problem happens and give us the error you get.
Thanks
Julie

[Reuti]

Sure, here it is (PGI6.2/ACML3.6):

mpirun -np 4 -machinefile machinefile xssep < SEP.dat

27 3 1 1 10 N 0.00 -1.00 0.35E-01 0.89 PASSED EVD
C
ISEED( 1 ) = 3198
ISEED( 2 ) = 2611
ISEED( 3 ) = 15
ISEED( 4 ) = 287
UPLO= 'U'
SUBTESTS= 'N'
N= 27
NPROW= 2
NPCOL= 1
NB= 1
MATTYPE= 10
ABSTOL= 0.235099D-37
THRESH= 0.500000D+02
C
27 1 2 1 10 N 0.01 -1.00 0.32 0.18E+03 FAILED
C
ISEED( 1 ) = 3198
ISEED( 2 ) = 2611
ISEED( 3 ) = 15
ISEED( 4 ) = 287
UPLO= 'U'
N= 27
NPROW= 2
NPCOL= 1
NB= 1
MATTYPE= 10
ABSTOL= 0.235099D-37
THRESH= 0.500000D+02
C
27 1 2 1 10 N 0.01 -1.00 0.32 0.18E+03 SKIPPED EV
C
ISEED( 1 ) = 3198
ISEED( 2 ) = 2611
ISEED( 3 ) = 15
ISEED( 4 ) = 287
UPLO= 'U'
N= 27
NPROW= 2
NPCOL= 1
NB= 1
MATTYPE= 10
ABSTOL= 0.235099D-37
THRESH= 0.500000D+02
C
27 1 2 1 10 N 0.01 -1.00 0.32 0.18E+03 SKIPPED EVD

and the other:

mpirun -np 4 -machinefile machinefile xzsep < SEP.dat

END OF TESTS.
{ 0, 1}: On entry to PZSEPCHK parameter number 2 had an illegal value
{ 0, 1}: On entry to PZSEPCHK parameter number 2 had an illegal value
{ 0, 1}: On entry to PZSEPCHK parameter number 2 had an illegal value
{ 0, 1}: On entry to PZSEPCHK parameter number 2 had an illegal value