hcho3 / eigenmap_gpu Goto Github PK

When I make the MakeFile and it reaches the objective line lanczos.o it reports a problem.
Namely, in line 123 of lanczos.cu, the function magma_dstedx() has the first input set as a character as shown here:
magma_dstedx('I', LANCZOS_ITR, 0, 1, 1, LANCZOS_ITR, &alpha[1], &beta[1], eigvec, LANCZOS_ITR, work, lwork, iwork, liwork, dwork, &info);
and this gives the following error:

lanczos.cu(123): error: argument of type "char" is incompatible with parameter of type "magma_range_t"

1 error detected in the compilation of "/tmp/tmpxft_00005bbf_00000000-6_lanczos.cpp1.ii".
Makefile:50: recipe for target 'lanczos.o' failed.

Any idea?

Thank you,

Ahmad

Hello, and thank you for sharing.

I am facing a problem with the Makefile compilation. When I run Makefile in eigenmap_gpu directory, specifically when it reaches target book.o, I get an error telling me that the function atomicAdd is already defined as seen below.

**_nvcc -arch=sm_35 -rdc=true -c book.cu -o book.o -Xcompiler -fPIC -I./matio/include
book.cu(161): error: function "atomicAdd(double *, double)" has already been defined

1 error detected in the compilation of "/tmp/tmpxft_00002825_00000000-4_book.cpp4.ii".
Makefile:47: recipe for target 'book.o' failed
make: *** [book.o] Error 2_**

I looked around and apparently it may be a problem that has to do with the cuda version, but I am still uncertain of what to do.

I am using Ubuntu 18.04, with nvidia-driver 390, and its compatible version of Cuda, which is 9.1.
I also made sure that matio, Atlas and Magma are in their correct locations, and made sure that said locations are defined in the Makefile.

Thank you for your help