cuda_ops.hpp File Reference

CUDA kernel wrappers. More...

#include "core/types.hpp"

Go to the source code of this file.

Namespaces
namespace	num
namespace	num::cuda

Functions
real *	num::cuda::alloc (idx n)
	Allocate device memory.
void	num::cuda::free (real *ptr)
	Free device memory.
void	num::cuda::to_device (real *dst, const real *src, idx n)
	Copy host to device.
void	num::cuda::to_host (real *dst, const real *src, idx n)
	Copy device to host.
void	num::cuda::scale (real *v, idx n, real alpha)
	v = alpha * v
void	num::cuda::add (const real *x, const real *y, real *z, idx n)
	z = x + y
void	num::cuda::axpy (real alpha, const real *x, real *y, idx n)
	y = alpha*x + y
real	num::cuda::dot (const real *x, const real *y, idx n)
	dot product
void	num::cuda::matvec (const real *A, const real *x, real *y, idx rows, idx cols)
	y = A * x (row-major A)
void	num::cuda::matmul (const real *A, const real *B, real *C, idx m, idx k, idx n)
	C = A * B.
void	num::cuda::thomas_batched (const real *a, const real *b, const real *c, const real *d, real *x, idx n, idx batch_size)
	Batched Thomas algorithm for tridiagonal systems.

Detailed Description

CUDA kernel wrappers.

Definition in file cuda_ops.hpp.