Im working on Cuda C++ right now. But I\'m having problems about reading and writing into files with CUDA.
How can I implement file input output processes in Cuda C+
Read the file using ordinary host (C++) file operations. Then transfer the data to the device if you need it there, using ordinary cudaMalloc
and cudaMemcpy
operations.
You won't be able to implement file I/O directly in CUDA C++, as there is no API for this and the GPU does not connect directly to the file system. You have to go through the OS for file system services.
For your file to get into CUDA memory space you will have to use the CUDA memcpy command after allocating memory for the information (so do a cudaMalloc(your, malloc, params, here) and a cudaMemcpy(your, memcpy, params,here) after you have loaded parsed and stored in memory the file as you would normally in C++.
You could try making a cuda system to access the file directly, but that is a process I would not envy trying to write, and the gain to effort ratio would render it a non-viable venture (apart from a proof of concept excercise).
Hope that helps!:)