cuda

README

CUDA bindings for Go
--------------------

This package provides Go bindings for nVIDIA CUDA.
Access is provided to the low-level functionality,
together with high-level (more idiomatic Go) functions.

Low-level usage example:

	import "cuda"

	runtime.LockOSThread()              // cuda contexts are tied to OS threads
	pointer := cuda.Malloc(4 * 1024)    // low-level memory allocation
	cuda.Free(pointer)                  // low level free


High-level usage example:

	import "cuda"

	runtime.LockOSThread()                   // cuda contexts are tied to OS threads
	deviceArray := cuda.NewArray(1024, 4)    // space for 1024 float32's
	hostArray := make([]float32, 1024)
	cuda.Copy(deviceArray, hostArray)        // copy from host to device
	cuda.Copy(hostArray, deviceArray)        // copy from device to host
	deviceArray.Free()


CUDA error statuses are checked automatically and passed to panic() when not successful.
Also, return values are returned directly, and not via a pointer passed as function argument.
E.g.:

The C code:
	void *ptr;
	cudaError_t error = cudaMalloc(&ptr, 1024);
	if(error != cudaSuccess){
		fprintf(stderr, cudaErrorString(error))
		abort()
	}

Becomes in Go:

	ptr := cuda.Malloc(1024)


Installation
------------

* To use this package, the CUDA SDK must be installed.
  It has been tested with CUDA 4.0 on Linux. See:
	http://developer.nvidia.com/object/cuda_3_2_downloads.html

* Have a look at cuda/Makefile.
  The Makefile assumes that the header files (cuda.h and friends) are located in:
  	/usr/local/cuda/include.
  and that the libraries (cudart.so and friends) are locted in:
	/usr/local/cuda/lib  /usr/local/cuda/lib64
  Check that this applicable for your system and change these paths if neccesarey.

* Finally:
	cd cuda/
	make install