flâneur

Writing CUDA Kernels — Numba 0.57.0+0.g4fd4e39c6.dirty documentation

numba.readthedocs.io · 1,403 words · saved by 1 readers

CUDA has an execution model unlike the traditional sequential model used for programming CPUs. In CUDA, the code you write will be executed by multiple threads at once (often hundreds or thousands). Your solution will be modeled by defining a thread hierarchy of grid, blocks and threads. Numba’s CUDA support exposes facilities to declare and manage this hierarchy of threads. The facilities are largely similar to those exposed by NVidia’s CUDA C language. Numba also exposes three kinds of GPU memory: global device memory (the large, relatively slow off-chip memory that’s connected to the GPU itself), on-chip shared memory and local memory. For all but the simplest algorithms, it is important that you carefully consider how to use and access memory in order to minimize bandwidth requirements and contention. A kernel function is a GPU function that is meant to be called from CPU code (*). It gives it two fundamental characteristics: kernels cannot explicitly return a value; all result dat

Writing CUDA Kernels - Numba 0+untagged.622.gaacf444.dirty documentation >> Numba for CUDA GPUs >> Writing CUDA Kernels View page source Writing CUDA Kernels ¶ CUDA Built-in Target deprecation notice The CUDA target built-in to Numba is deprecated, with further development moved to the NVIDIA numba-cuda package . Please see Built-in CUDA target deprecation and maintenance status . Introduction ¶ CUDA has an execution model unlike the traditional sequential model used for programming CPUs. In CUDA, the code you write will be executed by multiple threads at once (often hundreds or thousands). Yo

related reading