fusedcg_cpld_kernel.h File Reference

#include <math/bcknd/device/cuda/math_kernel.h>

Include dependency graph for fusedcg_cpld_kernel.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Functions
template<typename T >
__global__ void	fusedcg_cpld_part1_kernel (T *__restrict__ a1, T *__restrict__ a2, T *__restrict__ a3, T *__restrict__ b1, T *__restrict__ b2, T *__restrict__ b3, T *__restrict__ tmp, const int n)
template<typename T >
__global__ void	fusedcg_cpld_update_p_kernel (T *__restrict__ p1, T *__restrict__ p2, T *__restrict__ p3, const T *__restrict__ z1, const T *__restrict__ z2, const T *__restrict__ z3, const T *__restrict__ po1, const T *__restrict__ po2, const T *__restrict__ po3, const T beta, const int n)
template<typename T >
__global__ void	fusedcg_cpld_update_x_kernel (T *__restrict__ x1, T *__restrict__ x2, T *__restrict__ x3, const T **p1, const T **p2, const T **p3, const T *__restrict__ alpha, const int p_cur, const int n)
template<typename T >
__global__ void	fusedcg_cpld_part2_kernel (T *__restrict__ a1, T *__restrict__ a2, T *__restrict__ a3, const T *__restrict__ b, const T *__restrict__ c1, const T *__restrict__ c2, const T *__restrict__ c3, const T alpha, T *buf_h, const int n)

Function Documentation

fusedcg_cpld_part1_kernel()

template<typename T >

__global__ void fusedcg_cpld_part1_kernel	(	T *__restrict__	a1,
		T *__restrict__	a2,
		T *__restrict__	a3,
		T *__restrict__	b1,
		T *__restrict__	b2,
		T *__restrict__	b3,
		T *__restrict__	tmp,
		const int	n
	)

Kernel for first coupled fusedcg part

Definition at line 43 of file fusedcg_cpld_kernel.h.

fusedcg_cpld_part2_kernel()

template<typename T >

__global__ void fusedcg_cpld_part2_kernel	(	T *__restrict__	a1,
		T *__restrict__	a2,
		T *__restrict__	a3,
		const T *__restrict__	b,
		const T *__restrict__	c1,
		const T *__restrict__	c2,
		const T *__restrict__	c3,
		const T	alpha,
		T *	buf_h,
		const int	n
	)

Device kernel for fusedcg_cpld_part2

Definition at line 125 of file fusedcg_cpld_kernel.h.

Here is the call graph for this function:

fusedcg_cpld_update_p_kernel()

template<typename T >

__global__ void fusedcg_cpld_update_p_kernel	(	T *__restrict__	p1,
		T *__restrict__	p2,
		T *__restrict__	p3,
		const T *__restrict__	z1,
		const T *__restrict__	z2,
		const T *__restrict__	z3,
		const T *__restrict__	po1,
		const T *__restrict__	po2,
		const T *__restrict__	po3,
		const T	beta,
		const int	n
	)

Kernel for update of px

Definition at line 65 of file fusedcg_cpld_kernel.h.

fusedcg_cpld_update_x_kernel()

template<typename T >

__global__ void fusedcg_cpld_update_x_kernel	(	T *__restrict__	x1,
		T *__restrict__	x2,
		T *__restrict__	x3,
		const T **	p1,
		const T **	p2,
		const T **	p3,
		const T *__restrict__	alpha,
		const int	p_cur,
		const int	n
	)

Kernel for update of xx

Definition at line 92 of file fusedcg_cpld_kernel.h.