1. Reference problem

This test is based on the first calculation CALC_MODES + OPTION =” BANDE “from the fdlv112b test case (\(N\mathrm{=}5229\) degrees of freedom). We are looking for 50 modes in 2, 3 or 4 sub-bands depending on the case. Each modal search is pre-calibrated by calling INFO_MODE parallel.

This test is dedicated to the computer and functional validation of the parallel calculation schemes of CALC_MODES + OPTION =” BANDE “with division into sub-bands, and of INFO_MODE.

The validity of a good number of possible parallel configurations is tested by playing on:

• The parallelism level (s) activated (if modeling has),

• The distribution of MPI processes by sub-bands (if modeling a),

• The Sturm post-verification criterion (if CALC_MODES),

• The combinatorics of numerical parameters (modal solvers and/or direct linear solvers).

The calculation is started on 4 processors (if modeling a) and 1 processor (if modeling b).

Thus, the proposed calculations are:

With potentially 2 levels of parallelism [1] _ (NIVEAU_PARALLELISME =” COMPLET “) :

if*number of processors < number of non-empty sub-bands: case treated in test case error015a; Stop in ERREUR_F MODAL_10 regardless of the values of the other parameters.

if*number of processors = number of non-empty sub-bands: here 4 processors and 4 sub-bands. Parallel calculation called « 4x1 [2] _ « . A INFO_MODE calculation and 3 CALC_MODES + MUMPS calculations, with the modal solver by default, by varying the values of the Sturm test (“GLOBAL”/”LOCAL”/”NON”). User concepts: nbmd41 and mode41, mode41l and mode41n.

Same by combining the linear solver (“MULT_FRONT” and “LDLT”), the Sturm criterion and the modal solver (“JACOBI” and “TRI_DIAG”). NBMDMF/NBMD LDLT and Modesm/Modelm/ModeJM User Concepts

if*number of processors = alpha*number of non-empty sub-bands (integer alpha>1): here 4 processors and 2 sub-bands (alpha=2).

If the direct solver is not MUMPS: case treated in test case error015a; Stop in ERREUR_F MODAL_9 /10. This is true regardless of the values of the other parameters.

If the direct solver is MUMPS: parallelism says « 2x2 [3] _ « . A INFO_MODE calculation and 3 CALC_MODES calculations, with the modal solver by default, by varying the values of the Sturm test (“GLOBAL”/”LOCAL”/”NON”). nbmd22 and mode22, mode22let and mode22n user concepts.

if*number of processors > alpha*number of non-empty sub-bands+1 (integer alpha>1): here 4 processors and 3 sub-bands (alpha=1).

If the direct solver is not MUMPS: case treated in test case error015a; Stop in ERREUR_F MODAL_9 /10. This is true regardless of the values of the other parameters.

If the direct solver is MUMPS: parallelism says « 1x2 + 2 (1x1) [4] _ « . A INFO_MODE calculation and 3 CALC_MODES calculations by varying the values of the Sturm test (“GLOBAL”/”LOCAL”/”NON”) with the modal solver by default. Nbmd211 and Mode211l, Mode211L, Mode211L and Mode211N User Concepts.

• With potentially 1 level of parallelism [5] _

(NIVEAU_PARALLELISME =” PARTIEL “) :

if*number of processors < number of non-empty sub-bands: here 4 processors and 5 sub-bands.

If the direct solver is not MUMPS: case treated in test case error015a; Stop in ERREUR_F MODAL_14. This is true regardless of the values of the other parameters.

If the direct solver is MUMPS: parallelism says « 5 (1x4) [6] _ « . A INFO_MODE calculation and 3 CALC_MODES calculations, with the modal solver by default, by varying the values of the Sturm test (“GLOBAL”/”LOCAL”/”NON”). nbmd15 and mode 15, mode 15 and mode 15 n user concepts.

if*number of processors = number of non-empty sub-bands: here 4 processors and 4 sub-bands.

If the direct solver is not MUMPS: case treated in test case error015a; Stop in ERREUR_F MODAL_14. This is true regardless of the values of the other parameters.

If the direct solver is MUMPS: parallel calculation says « 4 (1x4) [7] _ « . A INFO_MODE calculation and 3 CALC_MODES + MUMPS calculations, with the modal solver by default, by varying the values of the Sturm test (“GLOBAL”/”LOCAL”/”NON”). User concepts: nbmd14 and mode14, mode14l and mode14n.

if*number of processors > alpha*number of non-empty sub-bands (integer alpha>1):

• Not applicable.

if*number of processors > alpha*number of non-empty sub-bands+1 (integer alpha>1): here 4 processors and 3 sub-bands (alpha=1).

If the direct solver is not MUMPS: case treated in test case error015a; Stop in ERREUR_F MODAL_14. This is true regardless of the values of the other parameters.

If the direct solver is MUMPS: parallelism says « 3 (1x4) [8] _ « . A INFO_MODE calculation and 3 CALC_MODES calculations by varying the values of the Sturm test (“GLOBAL”/”LOCAL”/”NON”) with the modal solver by default. nbmd13 and mode 13, mode 13 and mode 13 n user concepts.

For each of the calculations CALC_MODES, we test 8 non-regression values of the mode_meca:

• 4 eigenvalues taken at the terminals of the sub-intervals 2,

• 4 values of the effective unit modal mass taken arbitrarily, by permuting the components.

For each of the parallel INFO_MODE calculations, we test 3 non-regression values of the “NB_MODE” components of the sd_table:

• the minimum value,

• the maximum value,

• the sum.

[1]

One at the sub-band level, completed, if there are enough MPI processes, by a second at the level of the linear solver (if this one is MUMPS).

[2]

Each sub-band has had its dedicated processor reserved.

[3]

Each sub-band was reserved for its dedicated group of 2 processors.

[4]

The first sub-band is processed on 2 processors, then each of the following sub-bands is processed by a different processor.

[5]

Only the second level of the linear solver (if this one is MUMPS).

[6]

The 5 sub-bands are processed one after the other (in ascending order) and, each time, 4 processors are devoted to them.

[7]

The 4 sub-bands are processed one after the other (in ascending order) and, each time, 4 processors are devoted to them.

[8]

The 3 sub-bands are processed one after the other (in ascending order) and, each time, 4 processors are devoted to them.