[KaranG]

Respected members

While writing a fortran subroutine I am using this statement

write(UNIT=10000+myrank) tamp_mat(1:endx,1:endy, nn_loc)

and on running the code the error comes

Segmentation fault - invalid memory reference

this nn_loc represents the time step and this subroutine is called at each and every time step.

and the same variable is mentioned elsewhere in a subroutine from which it writes the values

there it is given as

tamp_mat(i,j,nn_loc)=tmp_mat(i,j)

in this nn_loc varies from 3: endt-1

Please help me to find out the cause of error.

your sincerely
Karan Gupta

[flotus1]

Use debugging flags to compile your code. At least -traceback -check bounds
edit: or -fbounds-check -fbacktrace for the gnu compiler

[piu58]

Originally Posted by KaranG

[FONT=Arial]

write(UNIT=10000+myrank) tamp_mat(1:endx,1:endy, nn_loc)
Karan Gupta

May it be that myrank points to anywhere and you address a unit which doesn't exist?

[KaranG]

This is the make file I used for running the open source code for which the above program is a part.


What do I do to debug the code using this makefile. I am new to FORTRAN programming so please help me.


# make clean : remove .o .mod .M -files and PROG
# make mod : compile modules
# make : build coulwave_MPI.out
# make db : compile files in $(DEBUG) with -g flag

PROG = pCOULWAVE.out

MODULES = mainvar_module.f FV_var_module.f

OBJS = coulwave_MPI.o var_grp.o \
add_bottom_friction.o add_breaking.o add_breaking_ktrans.o \
add_sponge.o allocate_matrices.o bl_define.o \
boundary_condition.o calc_corrector_error.o \
calc_Courant_number.o calc_mass.o calc_maxs_means.o \
calc_overflow.o calc_vel_DA.o calc_vel_dz.o calc_vel_z.o \
create_depth.o create_file.o create_rand_matrix.o \
create_write_xyth.o decomp2d.o dhdt_calc.o exchange2d.o \
exchange2d_int.o find_dxdydt.o find_max_hxx.o find_max_zeta.o \
find_ts_indices.o find_wavelengths.o global_dims.o \
initial_condition.o init_variables.o internal_source_type2.o \
load_eta_in.o load_inputs.o move_shoreline.o \
set_dispersion_coefs.o set_internal_source_coefs.o \
set_loop_limits.o shift_matrices_back.o solit_cnoidal_ic.o \
spectral_analysis.o store_previous_iter.o \
tridiag.o tridiagp.o user_interface.o write_disp_info.o \
write_inputs.o write_maxs_means.o write_surfaces.o \
write_timeseries.o \
FV_var_grp.o FV_allocate_matrices.o FV_interp_xei.o \
FV_interp_xoi.o FV_interp_yei.o FV_interp_yoi.o FV_ix_iy_eval.o

F90OBJS = FV_efg_calc.o FV_efg_calc_1D.o FV_estime.o \
FV_interpolations.o FV_ldu.o FV_limiters.o FV_Riemanns4.o

DEBUG_FLAGS = -C -g

# mpif90 with gfortran compiler
COMPILER=mpif90
F90FLAGS= -O3 -ffixed-line-length-0 -ffree-line-length-0
LFLAGS= -O3 -ffixed-line-length-0 -ffree-line-length-0

MOD_OBJ=$(MODULES:.f=.o)

main : $(OBJS) $(F90OBJS)
$(COMPILER) -o $(PROG) -p $(LFLAGS) $(OBJS) $(F90OBJS) $(MOD_OBJ)

%.o : %.f
$(COMPILER) -p -c $(F90FLAGS) $< -o $@
%.o : %.f90
$(COMPILER) -p -c $(F90FLAGS) $< -o $@

clean:
rm *.o *.mod *~ $(PROG)
db:
$(COMPILER) $(DEBUG_FLAGS) -o $(PROG) $(OBJS) $(F90OBJS) $(MOD_OBJ)
mod:
$(COMPILER) -c $(F90FLAGS) $(MODULES)

[juliom]

Segmentation fault means that you are reading a section in the memory that does not exist.
I can suggest you different things, since you are using writing instruction, make sure the argument of the function are characters, it seems that you are providing scalars instead.
Change tamp_mat(1:endx,1:endy, nn_loc), by
tamp_mat(:,:, nn_loc). And see if you can solve the issue. My guess is that you call this instruction at the end of each time step. If you call it at the end of the computation change nn_loc by any integer to see if you can access that part of the memory.
Segmentation faults are sometimes hard to spot but the best way is to do that.

[KaranG]

Thanks for the answer. Segmentation fault has gone by choosing tamp_mat(:,:, nn_loc).

[juliom]

Excellent, then check your nnumerical model, because it means that you introduced an error when you performed divided the domain, for example if you have 21 nodes, it means you have 20 cells or volumes. If you have the information stored at each node then i= 1, imax, but if you have the information stored at the center of each node, then i = 1, imax-1.
The solution of changing Change tamp_mat(1:endx,1:endy, nn_loc), by tamp_mat(:,:, nn_loc) basically reads the whole chunk from the memory, but that procedure will mask any error you have in your algebraic equation.

Happy Computations!!!

[FMDenaro]

This error is generally due to the fact that the type of the index used in the matrix is not an integer or if is an integer it violates the bounds of the matrix...

[FMDenaro]

Originally Posted by KaranG

This is the make file I used for running the open source code for which the above program is a part.


What do I do to debug the code using this makefile. I am new to FORTRAN programming so please help me.


# make clean : remove .o .mod .M -files and PROG
# make mod : compile modules
# make : build coulwave_MPI.out
# make db : compile files in $(DEBUG) with -g flag

PROG = pCOULWAVE.out

MODULES = mainvar_module.f FV_var_module.f

OBJS = coulwave_MPI.o var_grp.o \
add_bottom_friction.o add_breaking.o add_breaking_ktrans.o \
add_sponge.o allocate_matrices.o bl_define.o \
boundary_condition.o calc_corrector_error.o \
calc_Courant_number.o calc_mass.o calc_maxs_means.o \
calc_overflow.o calc_vel_DA.o calc_vel_dz.o calc_vel_z.o \
create_depth.o create_file.o create_rand_matrix.o \
create_write_xyth.o decomp2d.o dhdt_calc.o exchange2d.o \
exchange2d_int.o find_dxdydt.o find_max_hxx.o find_max_zeta.o \
find_ts_indices.o find_wavelengths.o global_dims.o \
initial_condition.o init_variables.o internal_source_type2.o \
load_eta_in.o load_inputs.o move_shoreline.o \
set_dispersion_coefs.o set_internal_source_coefs.o \
set_loop_limits.o shift_matrices_back.o solit_cnoidal_ic.o \
spectral_analysis.o store_previous_iter.o \
tridiag.o tridiagp.o user_interface.o write_disp_info.o \
write_inputs.o write_maxs_means.o write_surfaces.o \
write_timeseries.o \
FV_var_grp.o FV_allocate_matrices.o FV_interp_xei.o \
FV_interp_xoi.o FV_interp_yei.o FV_interp_yoi.o FV_ix_iy_eval.o

F90OBJS = FV_efg_calc.o FV_efg_calc_1D.o FV_estime.o \
FV_interpolations.o FV_ldu.o FV_limiters.o FV_Riemanns4.o

DEBUG_FLAGS = -C -g

# mpif90 with gfortran compiler
COMPILER=mpif90
F90FLAGS= -O3 -ffixed-line-length-0 -ffree-line-length-0
LFLAGS= -O3 -ffixed-line-length-0 -ffree-line-length-0

MOD_OBJ=$(MODULES:.f=.o)

main : $(OBJS) $(F90OBJS)
$(COMPILER) -o $(PROG) -p $(LFLAGS) $(OBJS) $(F90OBJS) $(MOD_OBJ)

%.o : %.f
$(COMPILER) -p -c $(F90FLAGS) $< -o $@
%.o : %.f90
$(COMPILER) -p -c $(F90FLAGS) $< -o $@

clean:
rm *.o *.mod *~ $(PROG)
db:
$(COMPILER) $(DEBUG_FLAGS) -o $(PROG) $(OBJS) $(F90OBJS) $(MOD_OBJ)
mod:
$(COMPILER) -c $(F90FLAGS) $(MODULES)

Note that you have DEBUG_FLAGS = -C -g but also some otpimization options like -O3. For debugging purpose, you must not use optimization.