When using CUDA 10, there are also the warnings, but the error below occurs on CUDA 9.2.88 and CUDA 10.
It is the same bug as in llama.
I try to fix it by updating it to the latest alpaka version, where it should be fixed.
Maybe it will take some time, but our students need cupla soon.
/opt/cuda/include/crt/math_functions.hpp:54:2: Warnung: #warning "crt/math_functions.hpp is an internal header file and must not be used directly. Please use cuda_runtime_api.h or c uda_runtime.h instead." [-Wcpp]
#warning "crt/math_functions.hpp is an internal header file and must not be used directly. Please use cuda_runtime_api.h or cuda_runtime.h instead."
^~~~~~~
In file included from /home/gcoe04/cuda-workspace/cupla/alpaka/include/alpaka/math/trunc/TruncCudaBuiltIn.hpp:38:0,
from /home/gcoe04/cuda-workspace/cupla/alpaka/include/alpaka/math/MathCudaBuiltIn.hpp:54,
from /home/gcoe04/cuda-workspace/cupla/alpaka/include/alpaka/acc/AccGpuCudaRt.hpp:38,
from /home/gcoe04/cuda-workspace/cupla/alpaka/include/alpaka/alpaka.hpp:40,
from /home/gcoe04/cuda-workspace/cupla/include/cupla_runtime.hpp:24,
from /home/gcoe04/cuda-workspace/cupla/src/stream.cpp:22:
/opt/cuda/include/crt/math_functions.hpp:54:2: Warnung: #warning "crt/math_functions.hpp is an internal header file and must not be used directly. Please use cuda_runtime_api.h or c uda_runtime.h instead." [-Wcpp]
#warning "crt/math_functions.hpp is an internal header file and must not be used directly. Please use cuda_runtime_api.h or cuda_runtime.h instead."
^~~~~~~
/home/gcoe04/cuda-workspace/cupla/alpaka/include/alpaka/core/Assert.hpp(102): error: this pragma must immediately precede a declaration
1 error detected in the compilation of "/tmp/tmpxft_0000275b_00000000-6_stream.cpp1.ii".
CMake Error at matrixMul_generated_stream.cpp.o.cmake:279 (message):
Error generating file
/home/gcoe04/cuda-workspace/cupla/build/CMakeFiles/matrixMul.dir/__/__/__/src/./matrixMul_generated_stream.cpp.o
make[2]: *** [CMakeFiles/matrixMul.dir/build.make:107: CMakeFiles/matrixMul.dir/__/__/__/src/matrixMul_generated_stream.cpp.o] Fehler 1
make[1]: *** [CMakeFiles/Makefile2:72: CMakeFiles/matrixMul.dir/all] Fehler 2
make: *** [Makefile:84: all] Fehler 2
When using CUDA 10, there are also the warnings, but the error below occurs on CUDA 9.2.88 and CUDA 10. It is the same bug as in llama. I try to fix it by updating it to the latest alpaka version, where it should be fixed. Maybe it will take some time, but our students need cupla soon.