Parallel Edge Detection with OpenMP and CUDA

Build

Requirements

• Nvidia GPU
• Up-to-date GPU driver
• gcc and nvcc

make

Run

# For sequential program
./seq_main assets/papagan.JPG papagan2.JPG

# For CUDA program
./cuda_main assets/papagan.JPG papagan2.JPG 4 8 16

# For CUDA and OpenMP program
./cuda_openMP_main assets/papagan.JPG papagan2.JPG 4 8 16

Explanation

There is 2 CUDA-OpenMP and 1 C file

• Sequential file works in a traditional way
• In cuda_main.cu, you can use either CUDA or OpenMP
  • For using CUDA, you need to put comment line in front of the 147th line
  • For using OpenMP, you need to put comment line in front of the 150th, 151st and 156th lines
• In cuda_openMP_main.cu, we used both CUDA and OpenMP libraries
  • It is basically use CUDA(GPU parallelism) and OpenMP(CPU parallelism, in main) at the same time.

There are 3 parameters after the image names

• First one indicates the number of OpenMP threads
• Second one indicates the number of blocks in CUDA
• Third one indicates the number of threads per block in CUDA

Discussion

You can find a discussion about how execution time changes when we change the parameters in Report.pdf