elementwise.cuh File Reference

#include "pangolin/namespace.hpp"
#include "search.cuh"

Include dependency graph for elementwise.cuh:

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

Functions
template<size_t BLOCK_DIM_X, typename IndexType , typename ValueType >
__device__ void	inner_product_inplace_block (const IndexType *indA, ValueType *valA, const IndexType nA, const IndexType *indB, const ValueType *valB, const IndexType nB)
template<size_t BLOCK_DIM_X, typename IndexType , typename ValueType >
__global__ void	csr_elementwise_inplace (const IndexType *csrRowPtrA, const IndexType *csrColIndA, ValueType *csrValA, const IndexType *csrRowPtrB, const IndexType *csrColIndB, const ValueType *csrValB, const IndexType numRows)
	CSR elementwise matrix multiplication in-place. More...
template<typename IndexType , typename ValueType >
__global__ void	csr_compress (const IndexType *csrRowPtrA, const IndexType *csrColIndA, const ValueType *csrValA, const IndexType *csrRowPtrB, const IndexType *csrColIndB, const ValueType *csrValB, const IndexType numRows, void *tmp)
	Compress CSR. More...

Function Documentation

csr_compress()

template<typename IndexType , typename ValueType >

__global__ void csr_compress	(	const IndexType *	csrRowPtrA,
		const IndexType *	csrColIndA,
		const ValueType *	csrValA,
		const IndexType *	csrRowPtrB,
		const IndexType *	csrColIndB,
		const ValueType *	csrValB,
		const IndexType	numRows,
		void *	tmp
	)

Compress CSR.

if tmp == nullptr, figures out how much tmp storage to allocate

else, compress the CSR

Parameters

numRows

number of rows in A and B

csr_elementwise_inplace()

template<size_t BLOCK_DIM_X, typename IndexType , typename ValueType >

__global__ void csr_elementwise_inplace	(	const IndexType *	csrRowPtrA,
		const IndexType *	csrColIndA,
		ValueType *	csrValA,
		const IndexType *	csrRowPtrB,
		const IndexType *	csrColIndB,
		const ValueType *	csrValB,
		const IndexType	numRows
	)

CSR elementwise matrix multiplication in-place.

A = A .* B

may put zeros into A's rows

Parameters

numRows

number of rows in A and B

inner_product_inplace_block()

template<size_t BLOCK_DIM_X, typename IndexType , typename ValueType >

__device__ void inner_product_inplace_block	(	const IndexType *	indA,
		ValueType *	valA,
		const IndexType	nA,
		const IndexType *	indB,
		const ValueType *	valB,
		const IndexType	nB
	)

non-zero elements in outer product of two sparse vectors

inner product of sparse A and B in-place in A

may put zeros into A