Learning Forum

[Amrith Mariappan]

Hi everyone,

 

I am encountering a Signal 11 (Segmentation Violation) during MPP initialization that occurs only with my custom-built LS-DYNA R16.2 executable, while the standard Ansys-distributed binary runs the identical keyword file to normal termination without any errors.

 

Environment & Executables

- Custom build: mppdyna.exe compiled from the R16.2 source makefile (Intel Fortran 19.0 + Intel MPI 2019.6.166, double precision I8R8, WINX64). I did not modify any source files or build flags — I simply ran the default makefile to verify that the custom library would execute without errors.

- Standard binary: lsdyna_mpp_sp_msmpi.exe (Ansys-distributed) — same keyword file, same machine, same MPI run command completes successfully with normal termination.

- Version: LS-DYNA R16.2-151-g276cb33db2

 

Problem Summary

When I run the custom mppdyna.exe with thermal properties included, the solver crashes immediately after initialization:

 

initialization completed

>>>>>  Process 0  <<<<<

>>>>> Signal 11 : Segmentation Violation <<<<<

 

The crash happens at Problem cycle = 0, right after MPP decomposition and initialization finish. No d3plot or binout data is written beyond the initial state.

 

Critical Observation: Thermal Dependency

- With thermal cards active: The custom MPP executable segfaults every time (tested with -n 1 and -n 2).

- Without thermal cards: The custom MPP executable runs perfectly fine to normal termination.

- With thermal cards + standard Ansys binary: Runs perfectly fine to normal termination.

 

This strongly suggests the issue is tied to how the custom-built executable handles thermal initialization in an MPP + S-ALE environment, rather than a keyword syntax error.

 

Keyword Cards Used in the Model

The simulation is a 2D axisymmetric S-ALE hypervelocity impact problem using the following cards:

 

- *KEYWORD, *TITLE

- *MAT_ELASTIC_PLASTIC_HYDRO_TITLE (HSLA 100 Steel, Tungsten Carbide)

- *MAT_ALE_VACUUM_TITLE

- *MAT_THERMAL_ISOTROPIC_PHASE_CHANGE_TITLE (Steel)

- *MAT_THERMAL_ISOTROPIC_TITLE (Tungsten Carbide, Vacuum)

- *EOS_GRUNEISEN_TITLE

- *NODE

- *ALE_STRUCTURED_MESH, *ALE_STRUCTURED_MESH_CONTROL_POINTS

- *ALE_MULTI-MATERIAL_GROUP

- *ALE_STRUCTURED_MESH_VOLUME_FILLING (ALL, BOXCOR, ELLIPSOID)

- *PART, *SECTION_ALE2D_TITLE

- *INITIAL_TEMPERATURE_SET

- *DEFINE_BOX_TITLE, *DEFINE_VECTOR_TITLE

- *BOUNDARY_SALE_MESH_FACE (SYM, NONREFL)

- *SET_PART_LIST_TITLE

- *CONTROL_ALE, *CONTROL_BULK_VISCOSITY, *CONTROL_ENERGY, *CONTROL_HOURGLASS

- *CONTROL_MPP_DECOMPOSITION_DISTRIBUTE_ALE_ELEMENTS

- *CONTROL_SOLUTION, *CONTROL_TERMINATION

- *CONTROL_THERMAL_NONLINEAR, *CONTROL_THERMAL_SOLVER, *CONTROL_THERMAL_TIMESTEP

- *CONTROL_TIMESTEP, *CONTROL_UNITS

- *DATABASE_ family (D3PLOT, D3THDT, GLSTAT, MATSUM, NODOUT, TPRINT, etc.)

- *DATABASE_EXTENT_BINARY

- *END

 

Run Command (Custom — Fails)

bash

mpiexec -n 2 "...\mppdyna.exe" i="sale_shot10_umat44v_T09-fine.k" memory=200m

 

Run Command (Standard — Succeeds)

bash

mpiexec -n 2 "...\lsdyna_mpp_sp_msmpi.exe" i="sale_shot10_umat44v_T09-fine.k" memory=200m

 

Question

Since the standard Ansys MPP binary handles the identical thermal + S-ALE setup without issue, but the default, unmodified custom build crashes during thermal initialization, what could be causing this? Is there a known compiler/MPI flag, preprocessor directive, or library linkage issue in the R16.2 makefile that affects thermal solver initialization in MPP mode? Any guidance on resolving this would be greatly appreciated.

 

Thanks,

Amrith