mycuda_scan Namespace Reference

Module for computing prefix scans. More...

Functions
template<typename T >
__device__ void	scan_dev (T *r, T x, int i, int n)
	Device code for prefix scan of a single thread block. More...
template<typename T >
__global__ void	scan_blocks (T *w, T *x, T *blocksum, int inclusive)
	Kernel for doing prefix scan. More...

Detailed Description

Module for computing prefix scans.

Notes

• See scan example in Nvidia CUDA toolkit.

Function Documentation

template<typename T >

__global__ void mycuda_scan::scan_blocks	(	T *	w,
		T *	x,
		T *	blocksum,
		int	inclusive
	)

Kernel for doing prefix scan.

Notes

• Inclusive scan: y(i) = x(0) + x(1) + y(i-1),

  Exclusive scan: y(i) = x(0) + x(1) + y(i),

• block j scans elements x[j*blocksize], ... x[(j+1)*blocksize-1]

Requires

• sharedMemorySize = 2*n*sizeof(float)
• n <= blocksize

Parameters

w	(out) scanned vector
x	(in) vector to be scanned
blocksum	(out) Block sums
inclusive	(in) flag indicating if scan is to be inclusive or exclusive

Definition at line 63 of file mycuda_scan.h.

template<typename T >

__device__ void mycuda_scan::scan_dev	(	T *	r,
		T	x,
		int	i,
		int	n
	)

Device code for prefix scan of a single thread block.

Notes

• Uses shared memory to perform scan.

Parameters

r	(out) shared memory
x	(in) ith element of vector to be scanned
i	(in) index of element to be scanned
n	(in) size of vector to be scanned

Definition at line 22 of file mycuda_scan.h.