cudaModelMatrix.cu File Reference

CUDA accelerated extension of Markov::API::ModelMatrix. More...

#include "cudaModelMatrix.h"
#include "cudarandom.h"
#include <curand_kernel.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <device_launch_parameters.h>

Include dependency graph for cudaModelMatrix.cu:

Go to the source code of this file.

Namespaces
	Markov
	Namespace for the markov-model related classes. Contains Model, Node and Edge classes.
	Markov::API
	Namespace for the MarkovPasswords API.
	Markov::API::CUDA
	Namespace for objects requiring CUDA libraries.

Functions
__global__ void	Markov::API::CUDA::FastRandomWalkCUDAKernel (unsigned long int n, int minLen, int maxLen, char *outputBuffer, char *matrixIndex, long int *totalEdgeWeights, long int *valueMatrix, char *edgeMatrix, int matrixSize, int memoryPerKernelGrid, unsigned long *seed)
	CUDA kernel for the FastRandomWalk operation. More...
__device__ char *	Markov::API::CUDA::strchr (char *p, char c, int s_len)
	srtchr implementation on device space More...

Detailed Description

CUDA accelerated extension of Markov::API::ModelMatrix.

Authors

Ata Hakçıl

Extension of Markov::API::ModelMatrix which is modified to run on GPU devices. This implementation only supports Nvidia devices.

Class to flatten and reduce Markov::Model to a Matrix. Matrix level operations can be used for Generation events, with a significant performance optimization at the cost of O(N) memory complexity (O(1) memory space for slow mode)

To limit the maximum memory usage, each generation operation is partitioned into 50M chunks for allocation. Threads are sychronized and files are flushed every 50M operations.

Definition in file cudaModelMatrix.cu.