Parallel Primitives

Functions
void	componentSumReduction (FLOAT_ARRAY_TYPE *dInData, FLOAT_ARRAY_TYPE *dInterData, int numInterData, int numThreads, int N)
	Parallel component wise sum procedure.
__global__ void	gpuComponentSumReduce1 (FLOAT_ARRAY_TYPE *dInData, FLOAT_ARRAY_TYPE *dInterData, int numInterData, int numThreads, int N)
	Kernel for first component wise reduction step.
__global__ void	gpuComponentSumReduce2 (FLOAT_ARRAY_TYPE *dInterData, int numInterData, int numThreads)
	Kernel for second component wise reduction step.
__global__ void	gpuReduce1 (float *dInData, float *dInterData, int numInterData, int numThreads, int N)
	Kernel for first sum reduction step.
__global__ void	gpuReduce2 (float *dInterData, int numInterData, int numThreads)
	Kernel for second rsum eduction step.
__global__ void	gpuSqSumReduce1 (FLOAT_ARRAY_TYPE *dInData, float *dInterData, int numInterData, int numThreads, int N)
	Kernel for first squared reduction step.
void	SqVecSumReduction (FLOAT_ARRAY_TYPE *dInData, float *dInterData, int numInterData, int numThreads, int N)
	Parallel squared vector sum procedure.
void	sumReduction (float *dInData, float *dInterData, int numInterData, int numThreads, int N)
	Parallel summation procedure.

---

Detailed Description

This module contains parallel implementations for typical tasks

---

Function Documentation

void componentSumReduction	(	FLOAT_ARRAY_TYPE *	dInData,
		FLOAT_ARRAY_TYPE *	dInterData,
		int	numInterData,
		int	numThreads,
		int	N
	)

Parallel componentwise sum function, used as wrapper for gpuComponentSumReduce1 and gpuComponentSumReduce2 . The final result is stored in the first element of dInterData .

Parameters:

[in]	dInData	Input data
[in,out]	dInterData	Used for intermedite storage, content will be overwritten
[in]	numInterData	Elements in intermediate output data
[in]	numThreads	Number of used threads
[in]	N	Number of elements to be reduced

Returns:

Returns CUDA error information.

Definition at line 804 of file gpu_tools.cu.

References checkCudaError(), and FLOAT_ARRAY_TYPE.

Referenced by removeDrift().

__global__ void gpuComponentSumReduce1	(	FLOAT_ARRAY_TYPE *	dInData,
		FLOAT_ARRAY_TYPE *	dInterData,
		int	numInterData,
		int	numThreads,
		int	N
	)

This kernel function sums componentwise all N elements of the input data up into numInterData intermediate partial sums.

Parameters:

[in]	dInData	Input data
[out]	dInterData	Intermediate output data
[in]	numInterData	Elements in intermediate output data
[in]	numThreads	Number of used threads
[in]	N	Number of elements to be reduced

Returns:

Returns CUDA error information.

Definition at line 707 of file gpu_tools.cu.

References FLOAT_ARRAY_TYPE.

__global__ void gpuComponentSumReduce2	(	FLOAT_ARRAY_TYPE *	dInterData,
		int	numInterData,
		int	numThreads
	)

This kernel function takes the intermediate componentwise sums of gpuComponentSumReduce1 and calculates the final componentwise sum, which is stored in the first element of dInterData

Parameters:

[in,out]	dInterData	Intermediate output data
[in]	numInterData	Elements in intermediate output data
[in]	numThreads	Number of used threads

Returns:

Returns CUDA error information.

Definition at line 745 of file gpu_tools.cu.

References FLOAT_ARRAY_TYPE.

__global__ void gpuReduce1	(	float *	dInData,
		float *	dInterData,
		int	numInterData,
		int	numThreads,
		int	N
	)

This kernel function sums all N elements of the inputdata up into numInterData intermediate partial sums.

Parameters:

[in]	dInData	Input data
[out]	dInterData	Intermediate output data
[in]	numInterData	Elements in intermediate output data
[in]	numThreads	Number of used threads
[in]	N	Number of elements to be reduced

Returns:

Returns CUDA error information.

Definition at line 637 of file gpu_tools.cu.

__global__ void gpuReduce2	(	float *	dInterData,
		int	numInterData,
		int	numThreads
	)

This kernel function takes the intermediate sums of gpuReduce1 and calculates the final sum, which is stored in the first element of dInterData

Parameters:

[in,out]	dInterData	Intermediate output data
[in]	numInterData	Elements in intermediate output data
[in]	numThreads	Number of used threads

Returns:

Returns CUDA error information.

Definition at line 660 of file gpu_tools.cu.

__global__ void gpuSqSumReduce1	(	FLOAT_ARRAY_TYPE *	dInData,
		float *	dInterData,
		int	numInterData,
		int	numThreads,
		int	N
	)

This kernel function takes the squared sums of all N elements of the input data and stores numInterData intermediate results.

Parameters:

[in]	dInData	Input data
[out]	dInterData	Intermediate output data
[in]	numInterData	Elements in intermediate output data
[in]	numThreads	Number of used threads
[in]	N	Number of elements to be reduced

Returns:

Returns CUDA error information.

Definition at line 684 of file gpu_tools.cu.

void SqVecSumReduction	(	FLOAT_ARRAY_TYPE *	dInData,
		float *	dInterData,
		int	numInterData,
		int	numThreads,
		int	N
	)

Parallel squared vector sum function, used as wrapper for gpuSqSumReduce1 and gpuReduce1 . The final result is stored in the first element of dInterData .

Parameters:

[in]	dInData	Input data
[in,out]	dInterData	Used for intermedite storage, content will be overwritten
[in]	numInterData	Elements in intermediate output data
[in]	numThreads	Number of used threads
[in]	N	Number of elements to be reduced

Returns:

Returns CUDA error information.

Definition at line 794 of file gpu_tools.cu.

References checkCudaError().

Referenced by getKineticEnergy(), and rescaleT().

void sumReduction	(	float *	dInData,
		float *	dInterData,
		int	numInterData,
		int	numThreads,
		int	N
	)

Parallel sum function, used as wrapper for gpuReduce1 and gpuReduce1 . The final result is stored in the first element of dInterData .

Parameters:

[in]	dInData	Input data
[in,out]	dInterData	Used for intermedite storage, content will be overwritten
[in]	numInterData	Elements in intermediate output data
[in]	numThreads	Number of used threads
[in]	N	Number of elements to be reduced

Returns:

Returns CUDA error information.

Definition at line 783 of file gpu_tools.cu.

References checkCudaError().

Referenced by getHarmonicPotentialEnergy(), and getLJCutPotentialEnergy().