CUDAKernel not recognizing ptx filename.

Hello,

I am attempting to use a CUDA function written in a file called 'cluster.cu' and compiled to 'cluster.ptx' in matlab. It worked fine the first time I tried it, but ever since then I get the following error:

"Error using parallel.internal.gpu.handleKernelArgs

The first input to parallel.gpu.CUDAKernel must be the name of a file that contains PTX code or an array containing PTX code."

Here is a snippet of the code involved:

cudaname = 'cluster.cu';
ptxname = 'cluster.ptx';

kernel = parallel.gpu.CUDAKernel(ptxname,cudaname);

I am not sure what I am doing wrong here. I am passing it the name of a ptx file. Any ideas?

[Joss Knight]

How did you compile your PTX file? What commands did you use?

[Guilherme]

I'm getting the same error using the following command to generate the PTX file

nvcc pctdemo_processMandelbrotElement.cu -ccbin "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.25.28610\bin\Hostx64\x64" -ptx

[Pierre E.]

The same for me... it seems that the generated .ptx file is empty. It seems to be a nvcc problem

Answer 1

[weui zhang]

Have you solved this problem?

I meet the same questions to you.But do not solve this.

[Linda Koletsou Soulti]

Hello Weui,

Has the PTX file been created properly with the expected name? A way to inspect the contents of the file is by using the command type followed by the name of the file inside the Command Window:

type filename.ptx

Could you also please let me know what are the commands you use to compile the file to PTX?

Cheers,

Linda

[weui zhang]

I found that the .ptx file has not been generated.

cudaFilename = [specific_filename, '.cu'];
ptxFilename = [specific_filename, '.ptx'];
system(['nvcc -ptx ', cudaFilename, ' --output-file ', ptxFilename]);
kernel = parallel.gpu.CUDAKernel( ptxFilename, cudaFilename );

[Linda Koletsou Soulti]

Hello Weui,

are you sure that the name of the file in specific_filename variable is properly spelled and resides in the folder you are when the system command is executed?

Cheers,

Linda

[weui zhang]

I am sure the questions you mentioned are all correct

[Linda Koletsou Soulti]

Hello Weui,

I will need some additional information to understand the issue you are facing. Does the system command throw any errors? Could you please add the verbose option, too:

system(['nvcc -v -ptx ', cudaFilename, ' --output-file ', ptxFilename]);

Have you tried to compile the file to ptx outside MATLAB? Could you please share some additional information like the .cu file you are trying to compile?

Can you successfully follow the steps in the example in the following link:

https://www.mathworks.com/help/parallel-computing/parallel.gpu.cudakernel.html

Cheers,

Linda

[weui zhang]

if single_yes

specific_filename = 'cuda/calculate_tensors_single';

else

specific_filename = 'cuda/calculate_tensors_double';

end

cudaFilename = [specific_filename, '.cu'];

ptxFilename = [specific_filename, '.ptx'];

system(['nvcc -ptx ', cudaFilename, ' --output-file ', ptxFilename]);

kernel = parallel.gpu.CUDAKernel( ptxFilename, cudaFilename );

The above is my code, it still mentioned "Error using parallel.internal.gpu.handleKernelArgs

The first input to parallel.gpu.CUDAKernel must be the name of a file that contains PTX code or an array containing PTX code."

[凯仁 解]

The same for me,and my ptx file has the content:

//

// Generated by NVIDIA NVVM Compiler

//

// Compiler Build ID: CL-24330188

// Cuda compilation tools, release 9.2, V9.2.148

// Based on LLVM 3.4svn

//

.version 6.2

.target sm_30

.address_size 64