GPUFunctions.h File Reference

Header file for using ReMAS with Nvidia GPUs. More...

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include <float.h>
#include <cuda_runtime.h>
#include <cufft.h>
#include <cublas_v2.h>
#include "../common/defines.h"

Go to the source code of this file.

Functions
void	BlocksAndThreads (int *, int *, int *, const int, const int)
	BlocksAndThreads calculates the suitable number of blocks and threads, and the needed shared memory. More...
int	HaveCompatibleGPU (int &)
	HaveCompatibleGPU checks if the system has an appropiate GPU for ReMAS. More...
bool	IsPow2 (unsigned int)
	IsPow2 decides if a number is power of 2. More...
unsigned int	NextPow2 (unsigned int)
	NextPow2 returns the next power of 2 of a given number. More...
int	FFTGPU (MyType *, MyType *, MyFFTGPUType *)
	FFTGPU computes FFT. More...
void	InitSxD (MyType *, MyType *, const MyType *__restrict__, const int *__restrict__, const int, const int)
int	AllocAuxiGPU (MyType **, short **, short **, MyType **, MyType **, const int, const int, const int)
	AllocAuxiGPU memory reservation for norms, frame, v_cfreq and v_dxState vectors. More...
int	AllocDataGPU (MyType **, int **, int **, int **, int **, int **, int *, const int, const int, DTWfiles)
	AllocDataGPU Allocates memory and initializes some structures reading info from files. More...
int	AllocDTWGPU (MyType **, MyType **, MyType **, const int, const int, const int)
	AllocDTWGPU Allocates memory for DTW vectors and auxiliar structures. More...
int	AllocFFTGPU (MyFFTGPUType *, MyType **, MyType **, MyType **, int *, int *, const int, DTWfiles)
	AllocFFTGPU Allocates "Unified" GPU memory for FFT vector and reads some fft information from files. More...
int	AllocS_fkGPU (MyType **, MyType **, MyType **, const MyType, const int, const int, DTWfiles)
	AllocS_fkGPU Allocates memory for S_fk vector, read its data from file and initializes other auxiliar vectors. More...
int	OneImin (MyType *, int *, MyType *, const int, const int)
	This function launches cuda kernels to find one minimun and its position. More...
int	FirstImin (MyType *, int *, MyType *, const int, const int)
	This function launches cuda kernels to find the first minimun and its position. More...
int	LastImin (MyType *, int *, MyType *, const int, const int)
	This function launches cuda kernels to find the last minimun and its position. More...
int	ReadWavGPU1st (short *, short *, FILE *)
	ReadWavGPU1st reads first audio (frame) from WAV file when NVIDIA GPU is used. More...
int	ReadWavGPU (short *, short *, FILE *)
	ReadFileGPU reads current audio (frame) from WAV file when NVIDIA GPU is used. More...
__global__ void	kernel_ApplyWindow (MyType *__restrict__, const short *__restrict__, const MyType *__restrict__, const int, const int)
__global__ void	kernel_InitDTW (MyType *__restrict__, const int, const int)
__global__ void	kernel_CompNorB0 (MyType *__restrict__, const MyType, const int)
__global__ void	kernel_CompNorB1 (MyType *__restrict__, const MyType *__restrict__, const int, const int)
__global__ void	kernel_CompNorBG (MyType *__restrict__, MyType *__restrict__, const MyType *__restrict__, const int, const MyType, const int)
__global__ void	kernel_PowToReal (MyType *__restrict__, const MyType *__restrict__, const MyType, const int)
__global__ void	kernel_Cfreq (MyType *__restrict__, const MyType *__restrict__)
__global__ void	kernel_Modul (MyType *__restrict__, const MyType *__restrict__, const int)
__global__ void	kernel_Reduction (MyType *__restrict__, const int)
__global__ void	kernel_InitSxD (MyType *__restrict__, MyType *__restrict__, const MyType *__restrict__, const int *__restrict__, const int, const bool, const int)
__global__ void	kernel_Sum (MyType *__restrict__, const MyType *__restrict__, const int, const bool, const int)
__global__ void	kernel_Vnorm (MyType *__restrict__)
__global__ void	kernel_UpdateSxD (MyType *__restrict__, const MyType, const MyType *__restrict__, const int)
__global__ void	kernel_DTW (const MyType *__restrict__, MyType *__restrict__, MyType *__restrict__, int *__restrict__, const int, const int, const int)
__global__ void	kernel_CompDisB0 (MyType *__restrict__, const MyType *__restrict__, const MyType *__restrict__, const MyType *__restrict__, const int, const int)
__global__ void	kernel_CompDisB1 (MyType *__restrict__, const MyType *__restrict__, const MyType *__restrict__, const MyType *__restrict__, const int, const int)
__global__ void	kernel_CompDisBG (MyType *__restrict__, const MyType *__restrict__, const MyType *__restrict__, const MyType *__restrict__, const MyType *__restrict__, const MyType *__restrict__, const MyType, const int, const int)
__global__ void	kernel_Shift (short *__restrict__, const int, const int)
__global__ void	kernel_OneImin (MyType *__restrict__, int *__restrict__, const MyType *__restrict__, const int, const bool, const int)
__global__ void	kernel_FirstImin (MyType *__restrict__, int *__restrict__, const MyType *__restrict__, const int, const bool, const int)
__global__ void	kernel_LastImin (MyType *__restrict__, int *__restrict__, const MyType *__restrict__, const int, const bool, const int)
__global__ void	kernel_OneIminLast (MyType *__restrict__, int *__restrict__, const MyType *__restrict__, const int *__restrict__, const int, const bool, const int)
__global__ void	kernel_FirstIminLast (MyType *__restrict__, int *__restrict__, const MyType *__restrict__, const int *__restrict__, const int, const bool, const int)
__global__ void	kernel_LastIminLast (MyType *__restrict__, int *__restrict__, const MyType *__restrict__, const int *__restrict__, const int, const bool, const int)

Detailed Description

Header file for using ReMAS with Nvidia GPUs.

Author

Information Retrieval and Parallel Computing Group, University of Oviedo, Spain

Interdisciplinary Computation and Communication Group, Universitat Politecnica de Valencia, Spain

Signal Processing and Telecommunication Systems Research Group, University of Jaen, Spain.

Contact: remas.nosp@m.pack.nosp@m.@gmai.nosp@m.l.co.nosp@m.m

Date

February 13, 2017

Definition in file GPUFunctions.h.

Function Documentation

AllocAuxiGPU()

int AllocAuxiGPU	(	MyType **	norms,
		short **	GPUframe,
		short **	CPUframe,
		MyType **	v_cfreq,
		MyType **	v_dxState,
		const int	nbases,
		const int	tamframe,
		const int	nmidi
	)

AllocAuxiGPU memory reservation for norms, frame, v_cfreq and v_dxState vectors.

Parameters

norms	(out) Norms vector
GPUframe	(out) Vector for frames in GPU memory
CPUframe	(out) Vector for frames in CPU memory
v_cfreq	(out) v_cfreq vector
v_dxState	(out) v_dxState vector
nbases	(in) Number of bases/combinations, sizeof norms and v_dxState
tamframe	(in) Size of frames in samples
nmidi	(in) Number of midi notes, v_cfreq size

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 259 of file GPUFunctions.cu.

AllocDataGPU()

int AllocDataGPU	(	MyType **	v_hanning,
		int **	states_time_i,
		int **	states_time_e,
		int **	states_seq,
		int **	states_corr,
		int **	I_SxD,
		int *	DTWSize,
		const int	tamtrama,
		const int	nstates,
		DTWfiles	NameFiles
	)

AllocDataGPU Allocates memory and initializes some structures reading info from files.

Parameters

v_hanning	(out) v_hanning vector
states_time_i	(out) states_time_i vector, contains the start-time of each state in frames
states_time_e	(out) states_time_e vector, contains the end-time of each state in frames
states_seq	(out) states_seq vector, contains the base/combination that is performed in each state
states_corr	(out) states_corr vector
I_SxD	(out) I_SxD vector
DTWSize	(in) Size of DTW vectors
tamtrama	(in) Size of frames in samples
nstates	(in) Number of states
NameFiles	(in) Struct with the file names

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 148 of file GPUFunctions.cu.

AllocDTWGPU()

int AllocDTWGPU	(	MyType **	pV,
		MyType **	v_SxD,
		MyType **	sdata,
		const int	maxGrid,
		const int	DTWSize,
		const int	DTWSizePlusPad
	)

AllocDTWGPU Allocates memory for DTW vectors and auxiliar structures.

Parameters

pV	(out) DTW pV vector
v_SxD	(out) v_SxD vector
sdata	(out) sdata vector, auxiliar
maxGrid	(in) maxGridSize supported by GPU
DTWSize	(in) Size of DTW vectors
DTWSizePlusPad	(in) Size of DTW vectors plus padding

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 232 of file GPUFunctions.cu.

AllocFFTGPU()

int AllocFFTGPU	(	MyFFTGPUType *	plan,
		MyType **	X_fft,
		MyType **	Out_fft,
		MyType **	Mod_fft,
		int *	kmin_fft,
		int *	kmax_fft,
		const int	nfft,
		DTWfiles	NameFiles
	)

AllocFFTGPU Allocates "Unified" GPU memory for FFT vector and reads some fft information from files.

Parameters

plan	(out) FFT scheduler
X_fft	(out) X_fft vector
Out_fft	(out) Out_fft vector
Mod_fft	(out) Mod_fft vector
kmin_fft	(out) Where kmin_fft is stored
kmax_fft	(out) Where kmax_fft is stored
nfft	(in) As is to be, Mod_fft and Out_fft size
NameFiles	(in) Struct with the file names

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 197 of file GPUFunctions.cu.

AllocS_fkGPU()

int AllocS_fkGPU	(	MyType **	s_fk,
		MyType **	tauxi,
		MyType **	ts_fk,
		const MyType	BETA,
		const int	nmidi,
		const int	nbases,
		DTWfiles	NameFiles
	)

AllocS_fkGPU Allocates memory for S_fk vector, read its data from file and initializes other auxiliar vectors.

Parameters

s_fk	(out) s_fk vector
tauxi	(out) Auxiliar vector tauxi
ts_fk	(out) Auxiliar vector ts_fk
BETA	(in) BETA value
nmidi	(in) Number of midi notes
nbases	(in) Number of bases/combinations
NameFiles	(in) Struct with the file names

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 117 of file GPUFunctions.cu.

BlocksAndThreads()

void BlocksAndThreads	(	int *	blocks,
		int *	threads,
		int *	sharedsize,
		const int	maxGrid,
		const int	size
	)

BlocksAndThreads calculates the suitable number of blocks and threads, and the needed shared memory.

Parameters

blocks	(out) Number of blocks
threads	(out) Number of threads
sharedsize	(out) Size of shared memory
maxGrid	(in) maxGridSize supported by GPU
size	(in) Size of the vectors

Returns

: Nothing, it is void

Definition at line 283 of file GPUFunctions.cu.

FFTGPU()

int FFTGPU	(	MyType *	X_fft,
		MyType *	Out_fft,
		MyFFTGPUType *	plan
	)

FFTGPU computes FFT.

Parameters

X_fft	(inout) X_fft vector
Out_fft	(inout) Out_fft vector
plan	(out) FFT scheduler

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 307 of file GPUFunctions.cu.

FirstImin()

int FirstImin	(	MyType *	odata,
		int *	opos,
		MyType *	idata,
		const int	maxGrid,
		const int	size
	)

This function launches cuda kernels to find the first minimun and its position.

Parameters

odata	(inout) Intermedial data vector
opos	(inout) Intermedial positions vector
idata	(in) Input data vector
maxGrid	(in) maxGridSize supported by GPU
size	(in) Size of the vector

Returns

: The pos of "a" minimum

Definition at line 394 of file GPUFunctions.cu.

HaveCompatibleGPU()

int HaveCompatibleGPU

(

int &

maxGrid

)

HaveCompatibleGPU checks if the system has an appropiate GPU for ReMAS.

Parameters

maxGrid

(out) MaxGrid stores the GPU maxGridSize property

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 75 of file GPUFunctions.cu.

IsPow2()

bool IsPow2 ( unsigned int  x )

inline

IsPow2 decides if a number is power of 2.

Parameters

x	(in) The number

Returns

: True is the number is power of 2, 0 otherwise false

Definition at line 66 of file GPUFunctions.cu.

LastImin()

int LastImin	(	MyType *	odata,
		int *	opos,
		MyType *	idata,
		const int	maxGrid,
		const int	size
	)

This function launches cuda kernels to find the last minimun and its position.

Parameters

odata	(inout) Intermedial data vector
opos	(inout) Intermedial positions vector
idata	(in) Input data vector
maxGrid	(in) maxGridSize supported by GPU
size	(in) Size of the vector

Returns

: The pos of "a" minimum

Definition at line 426 of file GPUFunctions.cu.

NextPow2()

unsigned int NextPow2

(

unsigned int

x

)

NextPow2 returns the next power of 2 of a given number.

Parameters

x	(in) The number

Returns

: Next power of 2 of a number

Definition at line 47 of file GPUFunctions.cu.

OneImin()

int OneImin	(	MyType *	odata,
		int *	opos,
		MyType *	idata,
		const int	maxGrid,
		const int	size
	)

This function launches cuda kernels to find one minimun and its position.

Parameters

odata	(inout) Intermedial data vector
opos	(inout) Intermedial positions vector
idata	(in) Input data vector
maxGrid	(in) maxGridSize supported by GPU
size	(in) Size of the vector

Returns

: The pos of "a" minimum

Definition at line 362 of file GPUFunctions.cu.

ReadWavGPU()

int ReadWavGPU	(	short *	GPUframe,
		short *	CPUframe,
		FILE *	fp
	)

ReadFileGPU reads current audio (frame) from WAV file when NVIDIA GPU is used.

Parameters

GPUframe	(out) Vector to store the current frame whitin NVIDIA GPU
CPUframe	(—) Vector to store the current frame whitin ARM
fp	(in) ID of file with the information

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 473 of file GPUFunctions.cu.

ReadWavGPU1st()

int ReadWavGPU1st	(	short *	GPUframe,
		short *	CPUframe,
		FILE *	fp
	)

ReadWavGPU1st reads first audio (frame) from WAV file when NVIDIA GPU is used.

Parameters

GPUframe	(out) Vector to store the first frame within NVIDIA GPU
CPUframe	(—) Vector to store the first frame within ARM
fp	(in) ID of file with the information

Returns

: 0 if all is OK, otherwise a code error (see defines.h)

Definition at line 456 of file GPUFunctions.cu.