EIG/TIMING (Error in csvdtim.out)

[MattR]

Hello everyone,

I am in the process of building LAPACK across multiple platforms,
i686/Linux with the Intel fortran compiler (ifort)9.0,
i686/Linux with the Lahey-Fujitsu fortran compiler (lf95)L6.20c, and on
alpha/OSF1 with the HP fortran compiler (f90)T5.6a.

So far, I have only had success building, testing, and timing using the Intel fortran compiler (and g77 for trivial testing purposes).

My current difficulties arise when trying to compile with Lahey. I have tried using the BLAS source that comes with LAPACK as well as the libblasmt.a that comes with the Lahey compiler suite with no success. Specifically, the problem arises during the timing phase, so maybe things aren't so bad, I hope. When the build process gets to the timing phase and runs the command

Code: Select all

./xeigtimc < csvdtim.in > csvdtim.out 2>&1

Most of that output file is populated correctly, but the process terminates with make giving an "Error 8".

The last few dozen lines of TIMING/csvdtim.out are as follows:

Code: Select all

      3       3.92E+02   5.16E+02   4.84E+02   5.18E+02   6.74E+02   6.36E+02   6.32E+02
      4       3.97E+02   4.99E+02   4.88E+02   5.44E+02   6.60E+02   6.23E+02   6.53E+02



 ****** Results for LINSVD(B) ******
     with LDA =   81

 *** Time in seconds ***
       M,N (  10,  10)(  10,  20)(  20,  10)(  20,  20)(  20,  40)(  40,  20)(  40,  40)
   Type
      1        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      2        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      3        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      4        0.0        0.0        0.0        0.0        0.0        0.0        0.0

 *** Number of floating-point operations ***
       M,N (  10,  10)(  10,  20)(  20,  10)(  20,  20)(  20,  40)(  40,  20)(  40,  40)
   Type
      1        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      2        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      3        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      4        0.0        0.0        0.0        0.0        0.0        0.0        0.0

 *** Speed in megaflops ***
       M,N (  10,  10)(  10,  20)(  20,  10)(  20,  20)(  20,  40)(  40,  20)(  40,  40)
   Type
      1        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      2        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      3        0.0        0.0        0.0        0.0        0.0        0.0        0.0
      4        0.0        0.0        0.0        0.0        0.0        0.0        0.0


 End o^A^@^@^@^B^@^@^@^C^@^@^@
The format specification has an invalid edit descriptor (unit= 6).
 Error occurs at or near line 648 of MAIN__


The exact same copy of LAPACK was built successfully using ifort and g77 on the same hardware, so perhaps there are some compiler or linker flags I should be using with this other compiler?

Does anyone have some ideas on what I can do to get the timing procedures to complete correctly?

Thank you,
--Matt

[Julie]

Hi Matt,

I don't access to any Fujitsu machine, so it will be very hard to help you on that one.
But it is not a big deal that the timing fails, what is important are the testing.

I looked at http://www.lahey.com/linux.htm and see that they ship LAPACK with their Fortran compiler. It is recommended to use the vendor LAPACK if it exists. Usually they tuned the LAPACK library for their machines and thus it should give you better performance.

Julie

[MattR]

Thank you for the quick reply Julie,

That was part of my hope too, that the timing phase of installation wasn't nearly as important as the others.

Since the build and testing phases seem to have completed correctly, I can assume the library itself is ready to use, yes?
If that's the case, I won't worry too much about the timing programs failing.

Thanks again,
--Matt

[Julie]

Yes Matt, the library is ready to use.

Have fun
Julie