Serial code implementation serial/serial_code.cu
To run the code
Without Valgrind
cd serial
make
./serial <grid_points> <num_rays>
With Valgrind
cd serial
make RayTracingSerialWithValgrind
valgrind --leak-check=yes ./serial <grid_points> <num_rays>
The image is made for grid size = 1000 and num of rays = 100000000
Parallel Code implementation parallel/parallel_code.cu
cd parallel
make
./parallel <grid_points> <num_rays> <threads_per_block>
Here grid size = 1000 and no of rays = 10000000 (both are kept constant)
| No of Threads Per Block | 32 | 64 | 128 | 256 |
|---|---|---|---|---|
| time elapsed device: (sec) | 0.428597 | 0.430513 | 0.419821 | 0.414252 |
Optimal settings = 256 threads per block
The optimal structure used for GPU here is 256 threads per block and no of blocks = num of rays /(256)
grid_size = 1000 (constant)
| No of rays | 10000 | 100000 | 1000000 | 10000000 | 100000000 |
|---|---|---|---|---|---|
| CPU time: (sec) | 0.163046 | 0.200292 | 0.571290 | 4.267502 | 41.105978 |
| GPU time: (sec) | 0.001255 | 0.005879 | 0.049273 | 0.434843 | 3.752551 |
The image is made for grid size = 1000 and num of rays = 1000000000
