Cannot compile properly on tiger about athena

[c-white]

tigressdata is currently configured to load icc version 15.0.2 with module load intel. On the other hand, tiger is still configured to load version 13.0.1 by default. However, you can force version 15 with module load intel/15 or module load intel/15.0 on tiger, and it will load the latest compiler.

Does this get rid of compilation issues? It seems to work for me.

Note that until recently we only had icc 14 at best, and that the code is significantly faster on icc 15.

[zhuzh1983]

Today, Kengo and I have tried intel/15.0.
There are no compiler errors but dt still goes to 0 very quickly.
This does not happen when the code is compiled in tigressdata.

Z

---

Zhaohuan Zhu, PhD
Hubble Fellow
Princeton University
Email: [email protected]
Homepage: http://www.astro.princeton.edu/~zhzhu/

On Apr 6, 2015, at 7:43 PM, Chris White wrote:

tigressdata is currently configured to load icc version 15.0.2 with module load intel. On the other hand, tiger is still configured to load version 13.0.1 by default. However, you can force version 15 with module load intel/15 or module load intel/15.0 on tiger, and it will load the latest compiler.

Does this get rid of compilation issues? It seems to work for me.

Note that until recently we only had icc 14 at best, and that the code is significantly faster on icc 15.

—
Reply to this email directly or view it on GitHub.

[tomidakn]

Hi Zhaohuan and Chris,

This is a part of my .bashrc file. I do not have any problem with this on tiger.
Please try these configurations.

---

module load intel/15.0
module load python
module load openmpi
module load hdf5

module switch intel/13.0 intel/15.0

Cheers,
Kengo

On 2015/04/06 21:46, zhuzh1983 wrote:

Today, Kengo and I have tried intel/15.0.
There are no compiler errors but dt still goes to 0 very quickly.
This does not happen when the code is compiled in tigressdata.

Z

---

Zhaohuan Zhu, PhD
Hubble Fellow
Princeton University
Email: [email protected]
Homepage: http://www.astro.princeton.edu/~zhzhu/

On Apr 6, 2015, at 7:43 PM, Chris White wrote:

tigressdata is currently configured to load icc version 15.0.2 with module
load intel. On the other hand, tiger is still configured to load version 13.0.1
by default. However, you can force version 15 with module load intel/15 or
module load intel/15.0 on tiger, and it will load the latest compiler.

Does this get rid of compilation issues? It seems to work for me.

Note that until recently we only had icc 14 at best, and that the code is
significantly faster on icc 15.

�$B!=�(B
Reply to this email directly or view it on GitHub.

�$B!=�(B
Reply to this email directly or view it on GitHub
#8 (comment).

Kengo TOMIDA [email protected]
Department of Astrophysical Sciences, Princeton University

[gustavogrds]

This is related with the previous messages, that's why I am posting it here.
There is a compilation problem with tiger. The problem is that you can't compile with the -d option for debugging. Here is the error message:

mpicxx -g -O0 -openmp -o bin/athena obj/main.o obj/meshblocktree.o obj/mesh.o obj/parameter_input.o obj/tasklist.o obj/utils.o obj/wrapio.o obj/bvals.o obj/outflow_bfield.o obj/outflow_fluid.o obj/reflect_bfield.o obj/reflect_fluid.o obj/cartesian.o obj/field.o obj/corner_emf.o obj/ct.o obj/field_integrator.o obj/fluid.o obj/srcterms.o obj/adiabatic_mhd.o obj/fluid_integrator.o obj/van_leer2.o obj/dc.o obj/plm.o obj/ppm.o obj/hlld.o obj/athdf5.o obj/formatted_table.o obj/history.o obj/outputs.o obj/restart.o obj/vtk.o obj/shock_tube.o -lhdf5
obj/fluid.o: In function Fluid::Fluid(MeshBlock*, ParameterInput*)': /home/grsilva/new_athena/athena/src/fluid/fluid.cpp:93: undefined reference toViscosity::Viscosity(Fluid_, ParameterInput_)'
obj/fluid.o: In function Fluid::~Fluid()': /home/grsilva/new_athena/athena/src/fluid/fluid.cpp:114: undefined reference toViscosity::~Viscosity()'
obj/fluid.o: In function L__ZN5Fluid16NewBlockTimeStepEP9MeshBlock_139__par_region0_2_6': /home/grsilva/new_athena/athena/src/fluid/fluid.cpp:197: undefined reference toViscosity::VisDt(double, int, int, int)'
/home/grsilva/new_athena/athena/src/fluid/fluid.cpp:198: undefined reference to Viscosity::VisDt(double, int, int, int)' /home/grsilva/new_athena/athena/src/fluid/fluid.cpp:199: undefined reference toViscosity::VisDt(double, int, int, int)'
obj/van_leer2.o: In function L__ZN15FluidIntegrator7OneStepEP9MeshBlockR11AthenaArrayIdES4_R14InterfaceFieldS4_i_76__par_region0_2_8': /home/grsilva/new_athena/athena/src/fluid/integrators/van_leer2.cpp:92: undefined reference toViscosity::ViscosityTerms(double, AthenaArray const&, AthenaArray&)'
make: *** [bin/athena] Error 1

Additional info that can be relevant:

• I am using: python configure.py --flux hlld --prob shock_tube -b -m -omp --cxx icc -hdf5 -d (in order to be sure that is not my problem generator)
• in this folder new_athena I cloned the last version of the repository again
• I am using module load intel/15.0 module load python module load openmpi module load hdf5 module switch intel/13.0 intel/15.0 in my .bashrc as recommended for tiger.

Maybe I am missing something...Please let me know if this is the case.

[tomidakn]

Hi,

This is due to the improper implementation of viscosity; when the VISCOSITY
option is off some functions are called but not defined because the viscosity
files are not compiled. The optimizer of the Intel Compiler removes those
function calls and suppresses these errors, but grammatically they are still
wrong. I was aware of this but did not fix it expecting someone (i.e. Zhaohuan)
will fix it, so please ask him to fix this. It can be fixed very easily; simply
change the configuration file so that the viscosity files are compiled no matter
whether the VISCOSITY switch is on or off. Then the functions are defined
regardless of whether they are actually used or not. This will slightly slow
down compilation but should not be a problem.

Cheers,
Kengo

On 2015/10/03 0:38, gustavogrds wrote:

This is related with the previous messages, that's why I am posting it here.
There is a compilation problem with tiger. The problem is that you can't compile
with the -d option for debugging. Here is the error message:

mpicxx -g -O0 -openmp -o bin/athena obj/main.o obj/meshblocktree.o obj/mesh.o
obj/parameter_input.o obj/tasklist.o obj/utils.o obj/wrapio.o obj/bvals.o
obj/outflow_bfield.o obj/outflow_fluid.o obj/reflect_bfield.o
obj/reflect_fluid.o obj/cartesian.o obj/field.o obj/corner_emf.o obj/ct.o
obj/field_integrator.o obj/fluid.o obj/srcterms.o obj/adiabatic_mhd.o
obj/fluid_integrator.o obj/van_leer2.o obj/dc.o obj/plm.o obj/ppm.o obj/hlld.o
obj/athdf5.o obj/formatted_table.o obj/history.o obj/outputs.o obj/restart.o
obj/vtk.o obj/shock_tube.o -lhdf5
obj/fluid.o: In function |Fluid::Fluid(MeshBlock_, ParameterInput_)':
/home/grsilva/new_athena/athena/src/fluid/fluid.cpp:93: undefined reference
to|Viscosity::Viscosity(Fluid/, ParameterInput/)'
obj/fluid.o: In function |Fluid::~Fluid()':
/home/grsilva/new_athena/athena/src/fluid/fluid.cpp:114: undefined reference
to|Viscosity::~Viscosity()'
obj/fluid.o: In function
|L__ZN5Fluid16NewBlockTimeStepEP9MeshBlock_139__par_region0_2_6':
/home/grsilva/new_athena/athena/src/fluid/fluid.cpp:197: undefined reference
to|Viscosity::VisDt(double, int, int, int)'
/home/grsilva/new_athena/athena/src/fluid/fluid.cpp:198: undefined reference to
|Viscosity::VisDt(double, int, int, int)'
/home/grsilva/new_athena/athena/src/fluid/fluid.cpp:199: undefined reference
to|Viscosity::VisDt(double, int, int, int)'
obj/van_leer2.o: In function
|L__ZN15FluidIntegrator7OneStepEP9MeshBlockR11AthenaArrayIdES4_R14InterfaceFieldS4_i_76__par_region0_2_8':
/home/grsilva/new_athena/athena/src/fluid/integrators/van_leer2.cpp:92:
undefined reference to|Viscosity::ViscosityTerms(double, AthenaArray const&,
AthenaArray&)'
make: *** [bin/athena] Error 1

Additional info that can be relevant:

• I am using: python configure.py --flux hlld --prob shock_tube -b -m -omp
  --cxx icc -hdf5 -d (in order to be sure that is not my problem generator)
• in this folder new_athena I cloned the last version of the repository again
• I am using module load intel/15.0 module load python module load openmpi
  module load hdf5 module switch intel/13.0 intel/15.0 in my .bashrc as
  recommended for tiger.

Maybe I am missing something...Please let me know if this is the case.

�$B!=�(B
Reply to this email directly or view it on GitHub
#8 (comment).

Kengo TOMIDA [email protected]
Department of Astrophysical Sciences, Princeton University

[zhuzh1983]

Problem Fixed. Using ICC with -d option can detect the uncompiled functions. Now I force the code to compile viscosity functions.