Calling method in the GNU Scientific Library

I want to call a function in the GNU Scientific Library (GSL) from MATLAB. In particular, I want to call this method:

int gsl_linalg_SV_decomp(gsl_matrix * A, gsl_matrix * V, gsl_vector * S, gsl_vector * work)

I have the following folder structure

.

├── Makefile

├── call_gsl_svd.m

├── include

│ ├── calculate_svd_example.h

│ └── run_svd.h

├── project

└── src

├── calculate_svd_example.c

├── project.c

└── run_svd.c

I can call make and the project builds without error. I can then call ./project and a call is made to the gsl_linalg_SV_decomp function and runs fine. calculate_svd_example.c has one simple method, calculate_svd_example, that takes no input and returns no output.

I want to call this simple method from MATLAB (as a first step, anyway). My call_gsl_svd.m reads:

function [] = call_gsl_svd()

%UNTITLED Summary of this function goes here

% Detailed explanation goes here

%#codegen

% for code generation, preinitialize the output variable

% data type, size, and complexity

% generate an include in the C code

coder.cinclude('./include/calculate_svd_example.h');

% evaluate the C function

coder.ceval('./src/calculate_svd_example');

end

I then run this command:

>> codegen call_gsl_svd calculate_svd_example.c

I get this error:

[1/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c /panfs/roc/msisoft/matlab/R2019a/extern/version/c_mexapi_version.c -o build/glnxa64/c_mexapi_version.o

[2/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c call_gsl_svd.c -o build/glnxa64/call_gsl_svd.o

[3/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c call_gsl_svd_initialize.c -o build/glnxa64/call_gsl_svd_initialize.o

[4/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c interface/_coder_call_gsl_svd_api.c -o build/glnxa64/_coder_call_gsl_svd_api.o

[5/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c call_gsl_svd_data.c -o build/glnxa64/call_gsl_svd_data.o

[6/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c call_gsl_svd_terminate.c -o build/glnxa64/call_gsl_svd_terminate.o

[7/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c interface/_coder_call_gsl_svd_mex.c -o build/glnxa64/_coder_call_gsl_svd_mex.o

[8/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c /panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src/calculate_svd_example.c -o build/glnxa64/calculate_svd_example.o

[9/10] /bin/gcc -fexceptions -fPIC -fno-omit-frame-pointer -pthread -D_GNU_SOURCE -DMATLAB_MEX_FILE -O2 -fwrapv -DNDEBUG -std=c99 -I "." -I "/panfs/roc/groups/4/miran045/reine097/projects/connectotyping/src" -I "./interface" -I "/panfs/roc/msisoft/matlab/R2019a/extern/include" -I "." -c interface/_coder_call_gsl_svd_info.c -o build/glnxa64/_coder_call_gsl_svd_info.o

[10/10] /bin/gcc build/glnxa64/call_gsl_svd_data.o build/glnxa64/call_gsl_svd_initialize.o build/glnxa64/call_gsl_svd_terminate.o build/glnxa64/call_gsl_svd.o build/glnxa64/_coder_call_gsl_svd_info.o build/glnxa64/_coder_call_gsl_svd_api.o build/glnxa64/_coder_call_gsl_svd_mex.o build/glnxa64/calculate_svd_example.o build/glnxa64/c_mexapi_version.o -pthread -Wl,--no-undefined -Wl,-rpath-link,/panfs/roc/msisoft/matlab/R2019a/bin/glnxa64 -shared -L"/panfs/roc/msisoft/matlab/R2019a/bin/glnxa64" -lmx -lmex -lmat -lm -lstdc++ -o call_gsl_svd_mex.mexa64 -lemlrt -lcovrt -lut -lmwmathutil -Wl,--version-script,call_gsl_svd_mex.map

FAILED: call_gsl_svd_mex.mexa64

/bin/gcc build/glnxa64/call_gsl_svd_data.o build/glnxa64/call_gsl_svd_initialize.o build/glnxa64/call_gsl_svd_terminate.o build/glnxa64/call_gsl_svd.o build/glnxa64/_coder_call_gsl_svd_info.o build/glnxa64/_coder_call_gsl_svd_api.o build/glnxa64/_coder_call_gsl_svd_mex.o build/glnxa64/calculate_svd_example.o build/glnxa64/c_mexapi_version.o -pthread -Wl,--no-undefined -Wl,-rpath-link,/panfs/roc/msisoft/matlab/R2019a/bin/glnxa64 -shared -L"/panfs/roc/msisoft/matlab/R2019a/bin/glnxa64" -lmx -lmex -lmat -lm -lstdc++ -o call_gsl_svd_mex.mexa64 -lemlrt -lcovrt -lut -lmwmathutil -Wl,--version-script,call_gsl_svd_mex.map

build/glnxa64/calculate_svd_example.o: In function `calculate_svd_example':

calculate_svd_example.c:(.text+0x73): undefined reference to `gsl_matrix_view_array'

calculate_svd_example.c:(.text+0x82): undefined reference to `gsl_matrix_alloc'

calculate_svd_example.c:(.text+0x94): undefined reference to `gsl_matrix_alloc'

calculate_svd_example.c:(.text+0xa1): undefined reference to `gsl_vector_alloc'

calculate_svd_example.c:(.text+0xae): undefined reference to `gsl_vector_alloc'

calculate_svd_example.c:(.text+0xbc): undefined reference to `gsl_matrix_transpose_memcpy'

calculate_svd_example.c:(.text+0xcd): undefined reference to `gsl_linalg_SV_decomp'

calculate_svd_example.c:(.text+0xe1): undefined reference to `pretty_print'

calculate_svd_example.c:(.text+0xf5): undefined reference to `pretty_print_vector'

calculate_svd_example.c:(.text+0x109): undefined reference to `pretty_print'

calculate_svd_example.c:(.text+0x111): undefined reference to `gsl_matrix_free'

calculate_svd_example.c:(.text+0x119): undefined reference to `gsl_matrix_free'

calculate_svd_example.c:(.text+0x121): undefined reference to `gsl_vector_free'

calculate_svd_example.c:(.text+0x129): undefined reference to `gsl_vector_free'

collect2: error: ld returned 1 exit status

ninja: build stopped: subcommand failed.

------------------------------------------------------------------------

??? Build error: C compiler produced errors. See the Build Log for further details.

More information

Code generation failed: View Error Report

Error using codegen

How do I do this call this simple C method? Do the .c, .h. and .m files all have to be in the same directory? How do I link with the GSL libraries?

(I have attached the source files as a ZIP file.)

[Bjorn Gustavsson]

Why do you want to use that SVD-decomposition-function instead of the LAPACK-based one matlab uses?

[Paul Reiners]

Does the LAPACK=based one that MATLAB uses run as fast as the GSL version? If it does, I'd be happy to continue using it.

[Bjorn Gustavsson]

(I don't work in the field of numerical linear algebra) I've lived with the impression that LAPACK (et al.) are the state-of-the-art when it comes to linear algebra algorithms. That might be wrong. You might have some special matrices where other algorithms are faster. Perhaps you can time the SVD-algorithms on your type of matrices? Not by interfacing to matlab in the first place but to write a test-program in C...

[Paul Reiners]

The GSL C function is aboutI’m 200x faster than the MATLAB function.

[Bjorn Gustavsson]

That's not bad as such. But my idea was that you'd compare directly with the LAPACK routines also in C: LAPACK-routines for Singular Value Decomposition. If you're building an interface with an external library (which I've never done...) I'd expect you want the best bang for your effort...

[Jan]

this might be useful: https://www.mathworks.com/matlabcentral/fileexchange/16777-lapack

[Paul Reiners]

How do you install SVD_LAPACK? That seems to be what I want.

Answer 1

[Paul Reiners]

>> mex -v -L/home/reine097/c-libs/lib/ -R2017b run_svd.c calculate_svd_example.c -lgsl -lgslcblas -lm

[Matthew Faulkner]

Hi, where did you obtain/find the directory -L/home/reine097/c-libs/lib/ ?

Answer 2

[Jan]

You can create a C++-Mex function and call the library throught it. Unfortunately all inputs must be transposed, because Matlab uses the columnwise storing in opposite to GSL. But the rest of the conversion from Matlab arrays to GSL matrices is easy.

[Paul Reiners]

Hi Jan,

I added more details to my description, including attaching the source code files. If you could take another look, that would be great.