GPU implementations of common functions. More...

#include "alignment.h"
#include "chain.h"
#include "scorer.h"
#include "thread.h"

Include dependency graph for gpu_module.h:

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

Typedefs
typedef struct ChainDatabaseGpu	ChainDatabaseGpu
	GPU database scoring object. More...

Functions
void	hwEndDataGpu (int queryEnd, int targetEnd, int outScore, Chain query, Chain target, Scorer scorer, int score, int card, Thread *thread)
	GPU implementation of the semiglobal scoring function. More...

void	nwFindScoreGpu (int queryStart, int targetStart, Chain query, Chain target, Scorer scorer, int score, int card, Thread thread)
	GPU implementation of score finding function. More...

void	nwLinearDataGpu (int scores, int affines, Chain query, int queryFrontGap, Chain target, int targetFrontGap, Scorer scorer, int pLeft, int pRight, int card, Thread thread)
	GPU implementation of Needleman-Wunsch scoring function. More...

void	ovEndDataGpu (int queryEnd, int targetEnd, int outScore, Chain query, Chain target, Scorer scorer, int score, int card, Thread *thread)
	GPU implementation of the overlap scoring function. More...

void	ovFindScoreGpu (int queryStart, int targetStart, Chain query, Chain target, Scorer scorer, int score, int card, Thread thread)
	GPU implementation of score finding function. More...

void	swEndDataGpu (int queryEnd, int targetEnd, int outScore, int scores, int affines, Chain query, Chain target, Scorer scorer, int score, int card, Thread *thread)
	GPU implementation of Smith-Waterman scoring function. More...

ChainDatabaseGpu *	chainDatabaseGpuCreate (Chain *database, int databaseLen, int cards, int cardsLen)
	ChainDatabaseGpu constructor. More...

void	chainDatabaseGpuDelete (ChainDatabaseGpu *chainDatabaseGpu)
	ChainDatabaseGpu destructor. More...

void	scoreDatabaseGpu (int *scores, int type, Chain query, ChainDatabaseGpu chainDatabaseGpu, Scorer scorer, int indexes, int indexesLen, int cards, int cardsLen, Thread *thread)
	GPU database aligning function. More...

void	scoreDatabasesGpu (int scores, int type, Chain queries, int queriesLen, ChainDatabaseGpu chainDatabaseGpu, Scorer scorer, int indexes, int indexesLen, int cards, int cardsLen, Thread *thread)
	GPU shotgun database aligning function. More...

Detailed Description

GPU implementations of common functions.

Typedef Documentation

typedef struct ChainDatabaseGpu ChainDatabaseGpu

GPU database scoring object.

In the database aligning, queries are often changed and the database is fairly static. ChainDatabaseGpu is the chain database prepared for GPU usage to reduce the preperation time in repetitive aligning.

Function Documentation

ChainDatabaseGpu* chainDatabaseGpuCreate	(	Chain **	database,
		int	databaseLen,
		int *	cards,
		int	cardsLen
	)

ChainDatabaseGpu constructor.

Parameters

database	chain array
databaseLen	chain array length
cards	cuda cards index array which the database will be available on
cardsLen	cuda cards index array length, greater or equal to 1

Returns: chainDatabaseGpu object

void chainDatabaseGpuDelete ( ChainDatabaseGpu * chainDatabaseGpu )

ChainDatabaseGpu destructor.

Parameters

chainDatabaseGpu chainDatabaseGpu object

void hwEndDataGpu	(	int *	queryEnd,
		int *	targetEnd,
		int *	outScore,
		Chain *	query,
		Chain *	target,
		Scorer *	scorer,
		int	score,
		int	card,
		Thread *	thread
	)

GPU implementation of the semiglobal scoring function.

Function provides the semiglobal end data, the position of the maximum score on the query and target sequences as well as the maximum score. QueryEnd must be equal to the length of the query minus one.

Parameters

queryEnd	output, position of the maximum score on the query sequences
targetEnd	output, position of the maximum score on the target sequences
outScore	output, maximum score
query	query chain
target	target chain
scorer	scorer object used for alignment
score	input alignment score if known, otherwise NO_SCORE
card	CUDA card on which the function will be executed
thread	thread on which the function will be executed, if NULL function is executed on the current thread

void nwFindScoreGpu	(	int *	queryStart,
		int *	targetStart,
		Chain *	query,
		Chain *	target,
		Scorer *	scorer,
		int	score,
		int	card,
		Thread *	thread
	)

GPU implementation of score finding function.

Method uses Needleman-Wunsch algorithm with all of the start conditions set to infinity. This assures path contains the first cell and does not start with gaps. If the score is found it returns the coordinates of the cell with the provided score, (-1, -1) otherwise.

Parameters

queryStart	output, if found query index of found cell, -1 otherwise
targetStart	output, if found target index of found cell, -1 otherwise
query	query chain
target	target chain
scorer	scorer object used for alignment
score	input alignment score
card	CUDA card on which the function will be executed
thread	thread on which the function will be executed, if NULL function is executed on the current thread

void nwLinearDataGpu	(	int **	scores,
		int **	affines,
		Chain *	query,
		int	queryFrontGap,
		Chain *	target,
		int	targetFrontGap,
		Scorer *	scorer,
		int	pLeft,
		int	pRight,
		int	card,
		Thread *	thread
	)

GPU implementation of Needleman-Wunsch scoring function.

If scores and/or affines pointers are not equal to NULL method provides the last row of the scoring matrix and the affine deletion matrix, respectively. Method uses Ukkonen's banded optimization with the pLeft and pRight margins. PLeft margin is defined as diagonal left of the diagonal from the (0, 0) cell, and pRight as the right diagonal, respectively. Only the cells lying between those diagonals are calculated.

Parameters

scores	output, if not NULL the last row of the scoring matrix, new array is created
affines	output, if not NULL the last row of the affine deletion matrix, new array is created
query	query chain
queryFrontGap	if not 0, force that alignments start with a query gap
target	target chain
targetFrontGap	if not 0, force that alignments start with a target gap
scorer	scorer object used for alignment
pLeft	left Ukkonen's margin
pRight	right Ukkonen's margin
card	CUDA card on which the function will be executed
thread	thread on which the function will be executed, if NULL function is executed on the current thread

void ovEndDataGpu	(	int *	queryEnd,
		int *	targetEnd,
		int *	outScore,
		Chain *	query,
		Chain *	target,
		Scorer *	scorer,
		int	score,
		int	card,
		Thread *	thread
	)

GPU implementation of the overlap scoring function.

Function provides the overlap end data, the position of the maximum score on the query and target sequences as well as the maximum score. QueryEnd must be equal to the length of the query minus one or targetEnd must be equal to the length of the target minus one.

Parameters

queryEnd	output, position of the maximum score on the query sequences
targetEnd	output, position of the maximum score on the target sequences
outScore	output, maximum score
query	query chain
target	target chain
scorer	scorer object used for alignment
score	input alignment score if known, otherwise NO_SCORE
card	CUDA card on which the function will be executed
thread	thread on which the function will be executed, if NULL function is executed on the current thread

void ovFindScoreGpu	(	int *	queryStart,
		int *	targetStart,
		Chain *	query,
		Chain *	target,
		Scorer *	scorer,
		int	score,
		int	card,
		Thread *	thread
	)

GPU implementation of score finding function.

Method uses Needleman-Wunsch algorithm. If the score is found and the indicies of the coresponding cell are on the border of the solving matrix, functions returns the coordinates of the cell with the provided score, (-1, -1) otherwise.

Parameters

queryStart	output, if found query index of found cell, -1 otherwise
targetStart	output, if found target index of found cell, -1 otherwise
query	query chain
target	target chain
scorer	scorer object used for alignment
score	input alignment score
card	CUDA card on which the function will be executed
thread	thread on which the function will be executed, if NULL function is executed on the current thread

void scoreDatabaseGpu	(	int **	scores,
		int	type,
		Chain *	query,
		ChainDatabaseGpu *	chainDatabaseGpu,
		Scorer *	scorer,
		int *	indexes,
		int	indexesLen,
		int *	cards,
		int	cardsLen,
		Thread *	thread
	)

GPU database aligning function.

Function scores the query with every target in the chainDatabaseGpu, in other words with every chain in the database array with witch the chainDatabaseGpu was created. The new score array has legnth of databaseLen, where databaseLen is the argument with which the chainDatabaseGpu was created. If the indexes array is given only the targets with given indexes will be scored, other targets will have the NO_SCORE score. CUDA cards are necessary for this function to work.

Parameters

scores	output, array of scores coresponding to every chain in the database array with which the chainDatabaseGpu was created, new array is created
type	aligning type, can be SW_ALIGN, NW_ALIGN, HW_ALIGN or OV_ALIGN
query	query chain
chainDatabaseGpu	gpu chain database object
scorer	scorer object used for alignment
indexes	array of indexes of which chains from the database to score, if NULL all are solved
indexesLen	indexes array length
cards	cuda cards index array
cardsLen	cuda cards index array length, greater or equal to 1
thread	thread on which the function will be executed, if NULL function is executed on the current thread

void scoreDatabasesGpu	(	int **	scores,
		int	type,
		Chain **	queries,
		int	queriesLen,
		ChainDatabaseGpu *	chainDatabaseGpu,
		Scorer *	scorer,
		int *	indexes,
		int	indexesLen,
		int *	cards,
		int	cardsLen,
		Thread *	thread
	)

GPU shotgun database aligning function.

Function scores every query with every target in the chainDatabaseGpu, in other words with every chain in the database array with witch the chainDatabaseGpu was created. The new score array has legnth of databaseLen * queriesLen, where databaseLen is the argument with which the chainDatabaseGpu was created. Array is organized as a table of databaseLen columns and queriesLen rows, where rows correspond to queries and columns to targets in the database. If the indexes array is given only the targets with given indexes will be scored, other targets will have the NO_SCORE score. CUDA cards are necessary for this function to work. This function is faster than calling scoreDatabaseGpu() for every query separately.

Parameters

scores	output, array of scores coresponding to every query scored with every chain in the database array with which the chainDatabaseGpu was created, new array is created
type	aligning type, can be SW_ALIGN, NW_ALIGN, HW_ALIGN or OV_ALIGN
queries	query chains array
queriesLen	query chains array length
chainDatabaseGpu	gpu chain database object
scorer	scorer object used for alignment
indexes	array of indexes of which chains from the database to score, if NULL all are solved
indexesLen	indexes array length
cards	cuda cards index array
cardsLen	cuda cards index array length, greater or equal to 1
thread	thread on which the function will be executed, if NULL function is executed on the current thread

void swEndDataGpu	(	int *	queryEnd,
		int *	targetEnd,
		int *	outScore,
		int **	scores,
		int **	affines,
		Chain *	query,
		Chain *	target,
		Scorer *	scorer,
		int	score,
		int	card,
		Thread *	thread
	)

GPU implementation of Smith-Waterman scoring function.

Function provides the Smith-Waterman end data, the position of the maximum score on the query and target sequences as well as the maximum score. Additionally if scores and/or affines pointers are not equal to NULL method provides the last row of the scoring matrix and the affine deletion matrix, respectively.

Parameters

queryEnd	output, position of the maximum score on the query sequences
targetEnd	output, position of the maximum score on the target sequences
scores	output, if not NULL the last row of the scoring matrix, new array is created
affines	output, if not NULL the last row of the affine deletion matrix, new array is created
outScore	output, maximum score
query	query chain
target	target chain
scorer	scorer object used for alignment
score	input alignment score if known, otherwise NO_SCORE
card	CUDA card on which the function will be executed
thread	thread on which the function will be executed, if NULL function is executed on the current thread

Typedefs

Functions

Detailed Description

Typedef Documentation

Function Documentation