A simple, parallel, efficient neural network evaluator.
Source files are nn_omp.c and nn_cuda.cu.
Compile with: gcc -fopenmp nn_omp.c -o nn_omp
Please note that some GCC versions might require the additional -lm flag to link the math library.
Run with: ./nn_omp [N K]
Values default to N=500000, K=150
Compile with: nvcc nn_cuda.cu -o nn_cuda
Run with: ./nn_cuda [N K]
Values default to N=500000, K=150
Both programs print the execution time on the console.
In addition, the output activations are written to a text file, respectively omp.txt and cuda.txt.
To check that computed results of both programs are the same, the source file test.c is provided.
Compile with: gcc test.c -o test
Run with: ./test [file1 file2]
Values default to file1="cuda.txt", file2="omp.txt"
Bash scripts and the ipython notebook used to assess the performance for both programs are provided inside the evaluation folder. Instructions to run the files are provided within them.
A brief report describing the work is provided inside the report folder. It includes the source LaTeX files as well as the pdf, along with the plots used within the document.