This directory contains scripts that can be used to automatically generate wrappers for C headers by NVIDIA, such as CUBLAS or CUDNN. This is done using Clang.jl, with some CSTParser.jl-based scripts to clean-up the result.
In CUDA.jl, the wrappers need to know whether pointers passed into the
library point to CPU or GPU memory (i.e. Ptr or CuPtr). This information is
not available from the headers, and the headers will need to be reviewed up manually.
Either run wrap.jl directly, or include it using Revise.jl and call the main() function.
Be sure to activate the project environment in this folder, which will download CUDA from
artifacts (if you want to upgrade the headers, be sure to update the relevant JLLs in the
project environment).
For each library, the script performs the following steps:
- generate wrappers with Clang.jl
- rewrite the headers: wrap functions that result status codes with
@check, add calls to API initializers, etc. - apply manual patches: these are read from the
patchesfolder, and can be used to incompatible code
Clang.jl generates headers with two files: a common file with type definitions, aliases,
etc, and a main wrapper that contains function definitions. The former will be copied over
the existing files automatically, while for the latter we scan for changes: Removed
functions are put in the libXXX_deprecated.jl file, new ones are concatenated to the
libXXX.jl file.
You should always review any changes to the headers! Specifically, to correct Ptr
signature and possibly change them to:
CuPtr: if this pointer is a device pointerPtrOrCuPtr: if this pointer can be either a device or host pointerRef: if the pointer represents a scalar or single-value argument on the hostCuRef: idem, but on the deviceRefOrCuRef: idem, but either on the host or device
Finally, it might also be useful to diff the generated wrapper (generated from scratch) in
the res/wrap folder with the one in the lib folder (which is incrementally changed) to
see if no function signatures have changed.