Skip to content

AllStepsBlocksInfo + Fortran: dimension order mismatch

Created by: pnorbert

The blocks returned have the dimensions in the order of the source data (e.g. Fortran), not in the order the reader expects (e.g. bpls2 as C++). Therefore bpls -D is incorrect for Fortran app output and bpls -Dd results in out of boundary read requests.

adiosvm/Tutorial/heat2d/fortran$ mpirun -n 12 simulation/heatSimulation_adios2   heat   4 3   4 5   1  1
Process decomposition  : 4 x 3
Array size per process : 4 x 5
Number of output steps : 1
Iterations per step    : 1

The Fortran global array is (44) by (35), which is listed in C/C++ readers as 15,16. The blocks, however, keep their original Fortran order.

adiosvm/Tutorial/heat2d/fortran$ /opt/adios2/bin/bpls2 heat.bp -D
  double   T     {15, 16}
        step 0: 
          block  0: [ 0: 3,  0: 4]
          block  1: [ 4: 7,  0: 4]
          block  2: [ 8:11,  0: 4]
          block  3: [12:15,  0: 4]
          block  4: [ 0: 3,  5: 9]
          block  5: [ 4: 7,  5: 9]
          block  6: [ 8:11,  5: 9]
          block  7: [12:15,  5: 9]
          block  8: [ 0: 3, 10:14]
          block  9: [ 4: 7, 10:14]
          block 10: [ 8:11, 10:14]
          block 11: [12:15, 10:14]
  double   dT    {15, 16}
        step 0: 
          block  0: [ 0: 3,  0: 4]
          block  1: [ 4: 7,  0: 4]
          block  2: [ 8:11,  0: 4]
          block  3: [12:15,  0: 4]
          block  4: [ 0: 3,  5: 9]
          block  5: [ 4: 7,  5: 9]
          block  6: [ 8:11,  5: 9]
          block  7: [12:15,  5: 9]
          block  8: [ 0: 3, 10:14]
          block  9: [ 4: 7, 10:14]
          block 10: [ 8:11, 10:14]
          block 11: [12:15, 10:14]

Dumping a block fails because of the start,count calculation is wrong in bpls2 (due to the wrong offset,size order in the blocks' info).

adiosvm/Tutorial/heat2d/fortran$ /opt/adios2/bin/bpls2 heat.bp -Dd
  double   T     {15, 16}
        step 0: 
          block  0: [ 0: 3,  0: 4]
    (0,0)    109.191 55.4442 114.575 65.3925 54.045 106.662
    (1,1)    52.9152 112.046 62.8634 51.5159 125.707 71.9605
    (2,2)    131.091 81.9088 70.5613 70.0993 16.3525 75.4833
    (3,3)    26.3008 14.9533 
          block  1: [ 4: 7,  0: 4]
    (0,0)    163.236 109.489 168.62 119.437 108.09 141.03
    (1,1)    87.2828 146.414 97.2311 85.8835 133.351 79.6042
    (2,2)    138.735 89.5525 78.205 106.03 52.2832 111.414
    (3,3)    62.2315 50.884 
          block  2: [ 8:11,  0: 4]
    (0,0)    112.352 58.6052 117.736 68.5535 57.206 85.031
    (1,1)    31.2842 90.415 41.2325 29.885 77.3524 23.6056
    (2,2)    82.7364 33.5539 22.2064 55.146 1.39925 60.53
    (3,3)    11.3475 0 
          block  3: [12:15,  0: 4]
terminate called after throwing an instance of 'std::invalid_argument'
  what():  ERROR: selection Start Dims(2):[12, 0] and Count Dims(2):[4, 5] (requested) is out of bounds of available Shape Dims(2):[16, 15] , when reading variable T, in call to Get

Note that the error message is also mixing orders. The selection is printed as coming from the C++ reader but the Shape is printed as coming from the Fortran data. Hence one could wonder what is wrong here when the selection looks good (in the error message)...