Detailed Description

This module provides a set of convenience wrappers to invoke device-wide CUB's parallel primitives without worrying about the memory management. All temporary storage is in fact allocated within a single thrust::device_vector passed by the user, which can be safely reused across function calls.

Functions
template<typename VectorType >
void	cugar::cuda::alloc_temp_storage (VectorType &vec, const uint64 size)

template<typename PredicateIterator >
bool	cugar::cuda::any (const uint32 n, const PredicateIterator pred)

template<typename PredicateIterator >
bool	cugar::cuda::all (const uint32 n, const PredicateIterator pred)

template<typename Iterator >
bool	cugar::cuda::is_sorted (const uint32 n, const Iterator values)

template<typename Iterator , typename Headflags >
bool	cugar::cuda::is_segment_sorted (const uint32 n, const Iterator values, const Headflags flags)

template<typename InputIterator , typename BinaryOp >
std::iterator_traits< InputIterator >::value_type	cugar::cuda::reduce (const uint32 n, InputIterator d_in, BinaryOp op, thrust::device_vector< uint8 > &d_temp_storage)

template<typename InputIterator , typename OutputIterator , typename BinaryOp >
void	cugar::cuda::inclusive_scan (const uint32 n, InputIterator d_in, OutputIterator d_out, BinaryOp op, thrust::device_vector< uint8 > &d_temp_storage)

template<typename InputIterator , typename OutputIterator , typename BinaryOp , typename Identity >
void	cugar::cuda::exclusive_scan (const uint32 n, InputIterator d_in, OutputIterator d_out, BinaryOp op, Identity identity, thrust::device_vector< uint8 > &d_temp_storage)

template<typename InputIterator , typename FlagsIterator , typename OutputIterator >
uint32	cugar::cuda::copy_flagged (const uint32 n, InputIterator d_in, FlagsIterator d_flags, OutputIterator d_out, thrust::device_vector< uint8 > &d_temp_storage)

template<typename InputIterator , typename OutputIterator , typename Predicate >
uint32	cugar::cuda::copy_if (const uint32 n, InputIterator d_in, OutputIterator d_out, const Predicate pred, thrust::device_vector< uint8 > &d_temp_storage)

template<typename InputIterator , typename OutputIterator , typename CountIterator >
uint32	cugar::cuda::runlength_encode (const uint32 n, InputIterator d_in, OutputIterator d_out, CountIterator d_counts, thrust::device_vector< uint8 > &d_temp_storage)

template<typename KeyIterator , typename ValueIterator , typename OutputKeyIterator , typename OutputValueIterator , typename ReductionOp >
uint32	cugar::cuda::reduce_by_key (const uint32 n, KeyIterator d_keys_in, ValueIterator d_values_in, OutputKeyIterator d_keys_out, OutputValueIterator d_values_out, ReductionOp reduction_op, thrust::device_vector< uint8 > &d_temp_storage)

Function Documentation

◆ all()

template<typename PredicateIterator >

bool cugar::cuda::all	(	const uint32	n,
		const PredicateIterator	pred
	)

return true if all items in the range [0,n) evaluate to true

◆ alloc_temp_storage()

template<typename VectorType >

void cugar::cuda::alloc_temp_storage	(	VectorType &	vec,
		const uint64	size
	)

make sure a given buffer is as big as size; note: upon reallocations, the contents of the buffer are invalidated

◆ any()

template<typename PredicateIterator >

bool cugar::cuda::any	(	const uint32	n,
		const PredicateIterator	pred
	)

return true if any item in the range [0,n) evaluates to true

◆ copy_flagged()

template<typename InputIterator , typename FlagsIterator , typename OutputIterator >

uint32 cugar::cuda::copy_flagged	(	const uint32	n,
		InputIterator	d_in,
		FlagsIterator	d_flags,
		OutputIterator	d_out,
		thrust::device_vector< uint8 > &	d_temp_storage
	)

device-wide copy of flagged items

Parameters

n	number of input items
d_in	a device input iterator
d_flags	a device flags iterator
d_out	a device output iterator
d_temp_storage	some temporary storage

Returns: the number of copied items

◆ copy_if()

template<typename InputIterator , typename OutputIterator , typename Predicate >

uint32 cugar::cuda::copy_if	(	const uint32	n,
		InputIterator	d_in,
		OutputIterator	d_out,
		const Predicate	pred,
		thrust::device_vector< uint8 > &	d_temp_storage
	)

device-wide copy of predicated items

Parameters

n	number of input items
d_in	a device input iterator
d_out	a device output iterator
pred	a unary predicate functor
d_temp_storage	some temporary storage

Returns: the number of copied items

◆ exclusive_scan()

template<typename InputIterator , typename OutputIterator , typename BinaryOp , typename Identity >

void cugar::cuda::exclusive_scan	(	const uint32	n,
		InputIterator	d_in,
		OutputIterator	d_out,
		BinaryOp	op,
		Identity	identity,
		thrust::device_vector< uint8 > &	d_temp_storage
	)

device-wide exclusive scan

Parameters

n	number of items to reduce
d_in	a device input iterator
d_out	a device output iterator
op	the binary reduction operator
identity	the identity element
d_temp_storage	some temporary storage

◆ inclusive_scan()

template<typename InputIterator , typename OutputIterator , typename BinaryOp >

void cugar::cuda::inclusive_scan	(	const uint32	n,
		InputIterator	d_in,
		OutputIterator	d_out,
		BinaryOp	op,
		thrust::device_vector< uint8 > &	d_temp_storage
	)

device-wide inclusive scan

Parameters

n	number of items to reduce
d_in	a device input iterator
d_out	a device output iterator
op	the binary reduction operator
d_temp_storage	some temporary storage

◆ is_segment_sorted()

template<typename Iterator , typename Headflags >

bool cugar::cuda::is_segment_sorted	(	const uint32	n,
		const Iterator	values,
		const Headflags	flags
	)

return true if the items in the range [0,n) are sorted by segment, where the beginning of each segment is identified by a set head flag

◆ is_sorted()

template<typename Iterator >

bool cugar::cuda::is_sorted	(	const uint32	n,
		const Iterator	values
	)

return true if the items in the range [0,n) are sorted

◆ reduce()

template<typename InputIterator , typename BinaryOp >

std::iterator_traits< InputIterator >::value_type cugar::cuda::reduce	(	const uint32	n,
		InputIterator	d_in,
		BinaryOp	op,
		thrust::device_vector< uint8 > &	d_temp_storage
	)

device-wide reduce

Parameters

n	number of items to reduce
d_in	a device iterator
op	the binary reduction operator
d_temp_storage	some temporary storage

◆ reduce_by_key()

template<typename KeyIterator , typename ValueIterator , typename OutputKeyIterator , typename OutputValueIterator , typename ReductionOp >

uint32 cugar::cuda::reduce_by_key	(	const uint32	n,
		KeyIterator	d_keys_in,
		ValueIterator	d_values_in,
		OutputKeyIterator	d_keys_out,
		OutputValueIterator	d_values_out,
		ReductionOp	reduction_op,
		thrust::device_vector< uint8 > &	d_temp_storage
	)

device-wide run-length encode

Parameters

n	number of input items
d_keys_in	a device input iterator
d_values_in	a device input iterator
d_keys_out	a device output iterator
d_values_out	a device output iterator
reduction_op	a reduction operator
d_temp_storage	some temporary storage

Returns: the number of copied items

◆ runlength_encode()

template<typename InputIterator , typename OutputIterator , typename CountIterator >

uint32 cugar::cuda::runlength_encode	(	const uint32	n,
		InputIterator	d_in,
		OutputIterator	d_out,
		CountIterator	d_counts,
		thrust::device_vector< uint8 > &	d_temp_storage
	)

device-wide run-length encode

Parameters

n	number of input items
d_in	a device input iterator
d_out	a device output iterator
d_counts	a device output count iterator
d_temp_storage	some temporary storage

Returns: the number of copied items

Detailed Description

Functions

Function Documentation

◆ all()

◆ alloc_temp_storage()

◆ any()

◆ copy_flagged()

◆ copy_if()

◆ exclusive_scan()

◆ inclusive_scan()

◆ is_segment_sorted()

◆ is_sorted()

◆ reduce()

◆ reduce_by_key()

◆ runlength_encode()