CUDA specific functions

Accessing ArrayFire's stream, and native device id with other CUDA code. More...

Functions
AFAPI af_err	afcu_get_stream (cudaStream_t *stream, int id)
	This file contain functions that apply only to the CUDA backend.
AFAPI af_err	afcu_get_native_id (int *nativeid, int id)
	Get the native device id of the CUDA device with id in ArrayFire context.
AFAPI af_err	afcu_set_native_id (int nativeid)
	Set the CUDA device with given native id as the active device for ArrayFire.
AFAPI af_err	afcu_cublasSetMathMode (cublasMath_t mode)
	Sets the cuBLAS math mode for the internal handle.
static cudaStream_t	getStream (int id)
	Get the stream for the CUDA device with id in ArrayFire context.
static int	getNativeId (int id)
	Get the native device id of the CUDA device with id in ArrayFire context.
static void	setNativeId (int nativeId)
	Set the CUDA device with given native id as the active device for ArrayFire.

Detailed Description

Accessing ArrayFire's stream, and native device id with other CUDA code.

If your software is using ArrayFire's CUDA backend, you can also write custom kernels and do custom memory operations using native CUDA commands. The functions contained in the afcu namespace provide methods to get the stream and native device id that ArrayFire is using.

Function Documentation

afcu_cublasSetMathMode()

AFAPI af_err afcu_cublasSetMathMode

(

cublasMath_t

mode

)

Sets the cuBLAS math mode for the internal handle.

See the cuBLAS documentation for additional details

Parameters

[in]

mode

The cublasMath_t type to set

Returns

af_err error code

afcu_get_native_id()

AFAPI af_err afcu_get_native_id	(	int *	nativeid,
		int	id
	)

Get the native device id of the CUDA device with id in ArrayFire context.

Parameters

[out]	nativeid	native device id of the CUDA device with id in ArrayFire context
[in]	id	ArrayFire device id

Returns

af_err error code

afcu_get_stream()

AFAPI af_err afcu_get_stream	(	cudaStream_t *	stream,
		int	id
	)

This file contain functions that apply only to the CUDA backend.

It will include cuda headers when it is built with NVCC. Otherwise the you can define the AF_DEFINE_CUDA_TYPES before including this file and it will define the cuda types used in this header. Get the stream for the CUDA device with id in ArrayFire context

Parameters

[out]	stream	CUDA Stream of device with id in ArrayFire context
[in]	id	ArrayFire device id

Returns

af_err error code

afcu_set_native_id()

AFAPI af_err afcu_set_native_id

(

int

nativeid

)

Set the CUDA device with given native id as the active device for ArrayFire.

Parameters

[in]

nativeid

native device id of the CUDA device

Returns

af_err error code

getNativeId()

static int getNativeId ( int  id )

inlinestatic

Get the native device id of the CUDA device with id in ArrayFire context.

Parameters

[in]

id

ArrayFire device id

Returns

cuda native id of device

Definition at line 129 of file cuda.h.

{
    int retVal;
    af_err err = afcu_get_native_id(&retVal, id);
    if (err!=AF_SUCCESS)
        throw af::exception("Failed to get CUDA device native id from ArrayFire");
    return retVal;
}

getStream()

static cudaStream_t getStream ( int  id )

inlinestatic

Get the stream for the CUDA device with id in ArrayFire context.

Parameters

[in]

id

ArrayFire device id

Returns

cuda stream used by CUDA device

Definition at line 110 of file cuda.h.

{
    cudaStream_t retVal;
    af_err err = afcu_get_stream(&retVal, id);
    if (err!=AF_SUCCESS)
        throw af::exception("Failed to get CUDA stream from ArrayFire");
    return retVal;
}

setNativeId()

static void setNativeId ( int  nativeId )

inlinestatic

Set the CUDA device with given native id as the active device for ArrayFire.

Parameters

[in]

nativeId

native device id of the CUDA device

Definition at line 147 of file cuda.h.

{
    af_err err = afcu_set_native_id(nativeId);
    if (err!=AF_SUCCESS)
        throw af::exception("Failed to change active CUDA device to the device with given native id");
}