A Python package

Driver API bindings

CUDA made easy

PyCu provides ctypes bindings to the CUDA Driver API along with a few other CUDA related libraries, each of which can be imported separately.

To access the direct API's of the Driver, NVRTC, or NVVM simply import any of the following

from pycu import driver
from pycu import nvrtc
from pycu import nvvm

To access a higher level API for each you can import "core" from the above libraries

from pycu.driver import core as driver
from pycu.nvrtc import core as nvrtc
from pycu.nvvm import core as nvvm

If you want to import everything import pycu

import pycu

pycu also has its own "core" which contains Jitify, a compiler that uses NVRTC and NVVM to jit compile kernel functions.

from pycu import core as pycu

Currently import pycu will auto import all of the above packages including pycu.core but this can be turned off (and may be turned of by defualt in the future) if only a select few of the functionalities are needed. This is a project I started developing in order to work on something else. Almost the entirity of the Driver API currently has a corrisponding python binding. The missing bindings are primarily related to the lesser used features (at least for me) like the graph API. There are a number of incomplete features as of right now, almost all related to the higher level Python API. They are mostly implemented as I need them.

There is currently one example located here https://github.com/uchytilc/Recursive-Grid-Based-Integration demonstrating how to compile and call kernels written in C, create device arrays, and query GPU properties.