cuda_launch_config | automatically selecting a CUDA kernel launch configuration | GPU library

cuda_launch_config is a C++ library typically used in Hardware, GPU applications. cuda_launch_config has no bugs, it has no vulnerabilities and it has low support. You can download it from GitHub.

It is often the case in CUDA programming that we wish to launch a kernel but don't know which block size to use. Perhaps the kernel is an essential piece of our algorithm, but its performance is not important enough to warrant a thorough tuning. Perhaps we aren't able to benchmark the kernel because it is a library function whose runtime behavior isn't known at authorship time. Perhaps we're simply in a hurry and have more interesting things to work on. Whatever the case, we need to choose some block size because it is necessary to launch the kernel. We'd like a block size which is both. One way to choose a block size is to use a heuristic which promotes utilization, or "occupancy". A kernel with high potential occupancy can often perform well, and isn't likely to perform pathologically slow. So, without further information beyond a kernel's resource requirements and the GPU of interest, high occupancy is a reasonable goal. The functions included in this library compute CUDA block sizes which are intended to promote high occupancy for a CUDA kernel.