mexcuda in ubuntu error

Try to complie the Cuda file in ubuntu system but get error.

mexcuda GEMM_GPU.cu -lcublas
Building with 'nvcc'.
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu: In function ‘void mexFunction(int, mxArray**, int, const mxArray**)’:
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu:164:55: warning: narrowing conversion of ‘numCRows’ from ‘long long int’ to ‘mwSize {aka long unsigned int}’ inside { } [-Wnarrowing]
  const mwSize dims[] = {numCRows,numCCols,batch_count};  

                                                       ^
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu:164:55: warning: narrowing conversion of ‘numCCols’ from ‘long long int’ to ‘mwSize {aka long unsigned int}’ inside { } [-Wnarrowing]
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu:164:55: warning: narrowing conversion of ‘batch_count’ from ‘int’ to ‘mwSize {aka long unsigned int}’ inside { } [-Wnarrowing]

Error using mex
/usr/bin/x86_64-linux-gnu-ld: cannot find -lcublas
collect2: error: ld returned 1 exit status


Error in mexcuda (line 166)
    [varargout{1:nargout}] = mex(mexArguments{:});

As the cuda file uses cublas_v2 liburary, I add the path of the CUDA h files in the commend line but the error still exists:

    mexcuda GEMM_GPU.cu -lcublas -I'/usr/local/cuda-9.0/include'
Building with 'nvcc'.
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu(164): warning: invalid narrowing conversion from "long long" to "unsigned long"
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu: In function ‘void mexFunction(int, mxArray**, int, const mxArray**)’:
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu:164:55: warning: narrowing conversion of ‘numCRows’ from ‘long long int’ to ‘mwSize {aka long unsigned int}’ inside { } [-Wnarrowing]
  const mwSize dims[] = {numCRows,numCCols,batch_count};  

                                                       ^
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu:164:55: warning: narrowing conversion of ‘numCCols’ from ‘long long int’ to ‘mwSize {aka long unsigned int}’ inside { } [-Wnarrowing]
/home/ubuntu/New_folder/Local_version/GEMM_GPU.cu:164:55: warning: narrowing conversion of ‘batch_count’ from ‘int’ to ‘mwSize {aka long unsigned int}’ inside { } [-Wnarrowing]

Error using mex
/usr/bin/x86_64-linux-gnu-ld: cannot find -lcublas
collect2: error: ld returned 1 exit status


Error in mexcuda (line 166)
    [varargout{1:nargout}] = mex(mexArguments{:});

Here are my device information after calling gpuDevice:

gpuDevice(1)

ans = 

  CUDADevice with properties:

                      Name: 'GeForce GTX 1080 Ti'
                     Index: 1
         ComputeCapability: '6.1'
            SupportsDouble: 1
             DriverVersion: 10.2000
            ToolkitVersion: 10.2000
        MaxThreadsPerBlock: 1024
          MaxShmemPerBlock: 49152
        MaxThreadBlockSize: [1024 1024 64]
               MaxGridSize: [2.1475e+09 65535 65535]
                 SIMDWidth: 32
               TotalMemory: 1.1718e+10
           AvailableMemory: 1.1399e+10
       MultiprocessorCount: 28
              ClockRateKHz: 1582000
               ComputeMode: 'Default'
      GPUOverlapsTransfers: 1
    KernelExecutionTimeout: 0
          CanMapHostMemory: 1
           DeviceSupported: 1
            DeviceSelected: 1

Here are the information after running gpu environment checker

puEnvObj = coder.gpuEnvConfig;
gpuEnvObj.GpuId = 1;
gpuEnvObj.BasicCodegen = 1;
gpuEnvObj.BasicCodeexec = 1;
results = coder.checkGpuInstall(gpuEnvObj)
Compatible GPU           : PASSED 
CUDA Environment         : PASSED 
	Runtime   : PASSED 
	cuFFT     : PASSED 
	cuSOLVER  : PASSED 
	cuBLAS    : PASSED 
Basic Code Generation    : PASSED 
Basic Code Execution     : PASSED 

results = 

  struct with fields:

                 gpu: 1
                cuda: 1
               cudnn: 0
            tensorrt: 0
        basiccodegen: 1
       basiccodeexec: 1
         deepcodegen: 0
        deepcodeexec: 0
    tensorrtdatatype: 0
           profiling: 0

I try to find whether it is because of the problem of GCC, but it looks fine once I just try to use mexcuda to complie some CUDA files which don't need cublas lib:

           mexcuda cuda_file_without_cublas_lib.cu  

The cu file will be successfully completed without any warnning or errors.

[Joss Knight]

The path to cublas is automatically inserted into the build line. Since adding a path to an incompatible version of the CUDA toolkit was one of the things you tried, perhaps MATLAB is using an incompatible version of the toolkit? Please run your call to mexcuda with the -v option so we can see the complete command line output.

Answer 1

[Joss Knight]

Try

mexcuda -lcublas GEMM_GPU.cu

Answer 2

[Shadaab Siddiqie]

Such errors might be caused due to the following reasons:

1) The 'mexcuda' is using incorrect version of 'nvcc'. Set the 'MW_NVCC_PATH' to the bin directory of the correct cuda installation.

>>setenv('MW_NVCC_PATH','/usr/local/CUDA/bin')

2) The 'g++' version is not compatible with the 'nvcc' version.

Check the the NVIDIA Developers website to find the compatible versions of g++ and update gcc/g++ to the recommended version. This is important because updating the gcc version will update the standard c/c++ libraries.

Answer 3

[Joss Knight]

Your declaration

const mwSize dims[] = {numCRows,numCCols,batch_count};

Is casting away the signed-ness of your variable numCRows and numCCols, because they are signed int64s and mwSize is a size_t (unsigned long int). All you have to do is force it (e.g. use a static_cast) or disable this compiler warning. Or just declare numCRows and numCCols as type mwSize in the first place.

Edit: Oh, and the same goes for batch_count.