fortran 3d tests failing: pivotL() encounters zero diagonal l.737 in symbfact.c

[drew-parsons]

fortran tests are building, running and passing for the debian build of 9.2.0

except for the 3d tests

test 20
      Start 20: f_pddrive3d

20: Test command: /usr/bin/mpiexec "-n" "8" "-oversubscribe" "./f_pddrive3d"
20: Working Directory: /projects/mathlibs/build/superlu-dist/obj-x86_64-linux-gnu/FORTRAN
20: Test timeout computed to be: 1500
20:   Process grid:            2  X           2  X           2
20: Time to read and distribute matrix 0.00
20:   Matrix A was set up: m          400  nnz         1920
20: At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
20: At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
20: At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
20: At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
20: --------------------------------------------------------------------------
20: prterun detected that one or more processes exited with non-zero status,
20: thus causing the job to be terminated. The first process to do so was:
20:
20:    Process name: [prterun-sandy-619727@1,1]
20:    Exit code:    255
20: --------------------------------------------------------------------------
20/22 Test #20: f_pddrive3d ......................***Failed    0.65 sec


test 22
      Start 22: f_pzdrive3d

22: Test command: /usr/bin/mpiexec "-n" "8" "-oversubscribe" "./f_pzdrive3d"
22: Working Directory: /projects/mathlibs/build/superlu-dist/obj-x86_64-linux-gnu/FORTRAN
22: Test timeout computed to be: 1500
22:   Process grid:            2  X           2  X           2
22: Time to read and distribute matrix 0.00
22:   Matrix A was set up: m          400  nnz         1920
22: At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
22: At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
22: At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
22: At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
22: --------------------------------------------------------------------------
22: prterun detected that one or more processes exited with non-zero status,
22: thus causing the job to be terminated. The first process to do so was:
22:
22:    Process name: [prterun-sandy-619769@1,0]
22:    Exit code:    255
22: --------------------------------------------------------------------------
22/22 Test #22: f_pzdrive3d ......................***Failed    0.44 sec

(I patched CMakeFile.txt to add the fortran tests to the cmake build)

[xiaoyeli]

I am not sure the problem. I just had a commit in master to make sure the input matrix g20/cg20 are copied to the build directory.
Also, can you check whether you are using 32-bit integer build (default) or 64-bit?

[drew-parsons]

This is with the default 32-bit integer build.

[xiaoyeli]

Can you try yesterday's commit? There was some improvement, but I couldn't reproduce your error.

[drew-parsons]

It confirms 32-bit,
e.g.

$ OMPI_MCA_btl_tcp_if_include=lo /usr/bin/mpiexec -n 8 -oversubscribe ./f_pddrive3d
  Process grid:            2  X           2  X           2
  use 32-bit integer for A matrix
Time to read and distribute matrix 0.00
  Matrix A was set up: m          400  nnz         1920
At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
At column 0 pivotL() encounters zero diagonal at line 737 in file ./SRC/prec-independent/symbfact.c
--------------------------------------------------------------------------
prterun detected that one or more processes exited with non-zero status,
thus causing the job to be terminated. The first process to do so was:

   Process name: [prterun-sandy-1047244@1,1]
   Exit code:    255
--------------------------------------------------------------------------

[xiaoyeli]

My feeling is the matrix is not read in correctly from the file g20.rua.

[drew-parsons]

The logic at l.737 in symbfact.c is

    diagind = jcol;
    diag = SLU_EMPTY;
    for (isub = nsupc; isub < nsupr; ++isub) {
        if ( lsub_ptr[isub] == diagind ) {
            diag = isub;
            break;
        }
    }
    /* Diagonal pivot exists? */
    if ( diag == SLU_EMPTY ) {
        printf("At column %d ", jcol);
        ABORT("pivotL() encounters zero diagonal");
    }

Adding some print statements to debug, it tells me that with diagind=0 (since jcol=0), it never gets lsub_ptr[isub]=0 for all values of isub (0,1,2,3 and 4). It finds 205, 60, 216, 396, 353 isntead. So diag remains set to SLU_EMPTY (which is -1).

$ LD_LIBRARY_PATH=.:../SRC/ OMPI_MCA_btl_tcp_if_include=lo /usr/bin/mpiexec -n 8 -oversubscribe ./f_pddrive3d
  Process grid:            2  X           2  X           2
  use 32-bit integer for A matrix
Time to read and distribute matrix 0.00
  Matrix A was set up: m          400  nnz         1920
with nsupc=0 nsupr=5  diagind=0  SLU_EMPTY=-1
isub=0 lsub_ptr[isub]=205
isub=1 lsub_ptr[isub]=60
isub=2 lsub_ptr[isub]=216
isub=3 lsub_ptr[isub]=396
isub=4 lsub_ptr[isub]=353
check diag -1 is not SLU_EMPTY -1
At column 0 pivotL() encounters zero diagonal at line 740 in file ./SRC/prec-independent/symbfact.c