RTX-CUDA-template

A simple RTX-CUDA interoperability test program based on the SAXPY example from NVIDIA's RTX Compute Samples https://github.com/NVIDIA/rtx_compute_samples

The sample code simulates n randomly generated numbers, by shifting their value (as if was a 1D coordinate) by a random translation, and computes the minimum value at each time step. The user can choose one of the following approaches for computing such minimum value in GPU:

• CUDA Kernel
• CUB
• Thrust
• RTX OptiX

Dependencies

• CUDA 12 or later
• OptiX 7.7 or later
• Nvidia driver 530.41 or later
• CUB (for comparison)
• THrust (for comparison)

Compile and run

mkdir build && cd build
cmake ../ -DOPTIX_HOME=<PATH-TO-OPTIX-MAIN-DIR> -DThrust_DIR=<PATH-TO-THRUST-CMAKE-CONFIG> -DCUB_DIR=<PATH-TO-CUB-CMAKE-CONFIG>
make
./rtxcuda <n> <s> <alg>
dev = device ID
n   = problem size
s   = number of simulation steps
alg = algorithm
   1 -> WARP_SHUFFLE
   2 -> CUB
   3 -> THRUST
   4 -> RTX_CLOSEST_HIT

Example compilation

>> cmake ../ -DOPTIX_HOME=~/NVIDIA-OptiX-SDK-7.7.0-linux64-x86_64 -DThrust_DIR=/opt/cuda/targets/x86_64-linux/lib/cmake/thrust -DCUB_DIR=/opt/cuda/targets/x86_64-linux/lib/cmake/cub
>> make

Example execution

./rtxcuda 0 $((2**20)) 1 4