This is a semantics of Nvidia's CUDA-C implemented using the K framework term
rewriting system.
It's a fork of Chucky Ellison's C semantics with changes and additions to
support various features of Nvidia's "nvcc" compiler. In addition to the
special kernel launch syntax ("<<< >>>") and various attributes ("__device__",
"__shared__", etc.), many functions from the CUDA Runtime API are at least
partially supported. See semantics/includes/cuda.h for some clues about the
current state of things.
Installation and build instructions can be found in the README from Chucky's C
semantics ("README-C"), although we've renamed the frontend to the generated
interpreter to "cudak" (from "kcc") and made some other minor changes to the
build system to avoid conflict or confusion with the vanilla-C semantics. As
mentioned in README-C, make sure to download the K Framework version 2.7 and
not the latest version.
Race checking and other properties of the semantics can be configured in
semantics/cuda-properties.k.
Chris Hathhorn, chathhorn@gmail.com