cudafit_analytic.cu File Reference

Provide trace fitting using the libucdafit library. More...

#include "cudafit_analytic.cuh"
#include <limits>
#include <ctime>
#include <iostream>
#include <stdexcept>
#include <string>
#include <float.h>
#include <DE_Optimizer.h>
#include <PSO_Optimizer.h>
#include "fit_extensions.h"
#include "functions_analytic.h"
#include "reductions.cu"

Include dependency graph for cudafit_analytic.cu:

Functions
__host__ __device__ float	logistic (float A, float k, float x1, float x)
	Evaluate a logistic function for the specified parameters and point. More...
__host__ __device__ float	decay (float A, float k, float x1, float x)
	Exponential decay function. More...
__host__ __device__ float	singlePulse (float A1, float k1, float k2, float x1, float C, float x)
	Evaluate the value of a single pulse in accordance with our canonical functional form. More...
__host__ __device__ float	doublePulse (float A1, float k1, float k2, float x1, float A2, float k3, float k4, float x2, float C, float x)
	Evaluate the canonical form of a double pulse. More...
__host__ __device__ float	chiFitness1 (const float *pParams, float x, float y, float wt)
	Computes chi-square fitness for one point in one solution given that d_fitness has pulled out what we need. This fitness is for a single-pulse fit. More...
__host__ __device__ float	chiFitness2 (const float *pParams, float x, float y, float wt)
	Computes chi-square fitness contribution for one point in one solution given that our caller has pulled out what we need. More...
__global__ void	d_fitness1 (const float *pSolutions, float *pFitnesses, int nParams, int nSol, unsigned short *pXcoords, unsigned short *pYcoords, float *pWeights, int nPoints)
	This function lives in the GPU and: More...
__global__ void	d_fitness2 (const float *pSolutions, float *pFitnesses, int nParams, int nSol, unsigned short *pXcoords, unsigned short *pYcoords, float *pWeights, int nPoints)
	This function lives in the GPU and: More...
void	h_fitSingle (const CudaOptimize::SolutionSet *solutions, CudaOptimize::FitnessSet *fitnesses, dim3 grid, dim3 block)
	Invokes the kernel that produces the fitness measure. The fitness is computed in the GPU and is the chi square. More...
void	h_fitDouble (const CudaOptimize::SolutionSet *solutions, CudaOptimize::FitnessSet *fitnesses, dim3 grid, dim3 block)
	Host part to setup computation of the fitnesses across the swarm for our fits for a double pulse. Really this just sets up the kernel call for fitness2 which does the rest. More...

Detailed Description

Provide trace fitting using the libucdafit library.

Author

Ron Foxfox@n.nosp@m.scl..nosp@m.msu.e.nosp@m.du

Note

We provide call compatible interfaces with lmfit1 and lmfit2.

This fit will not thread due to libcudaoptimize's need for us to have global data for the device pointers to the trace.

Function Documentation

chiFitness1()

__host__ __device__ float chiFitness1	(	const float *	pParams,
		float	x,
		float	y,
		float	wt
	)

Computes chi-square fitness for one point in one solution given that d_fitness has pulled out what we need. This fitness is for a single-pulse fit.

The choice of weight determines whether the returned value is Neyman's (weight by variance estimated from data wt = 1/y) or Pearson's (weight by variance estimated from fit wt = 1/fit).

Parameters

pParams	Pointer to this solutions parameters.
x	x-coordinate.
y	y-coordinate.
wt	Weight for this coordinate.

Returns

Chi-square between fit function and trace data.

chiFitness2()

__host__ __device__ float chiFitness2	(	const float *	pParams,
		float	x,
		float	y,
		float	wt
	)

Computes chi-square fitness contribution for one point in one solution given that our caller has pulled out what we need.

The choice of weight determines whether the returned value is Neyman's (weight by variance estimated from data wt = 1/y) or Pearson's (weight by variance estimated from fit wt = 1/fit).

Parameters

pParams	Pointer to this solutions parameters.
x	x-coordinate.
y	y-coordinate.
wt	Weight for this coordinate.

Returns

Chi-square between fit function and trace data.

d_fitness1()

__global__ void d_fitness1	(	const float *	pSolutions,
		float *	pFitnesses,
		int	nParams,
		int	nSol,
		unsigned short *	pXcoords,
		unsigned short *	pYcoords,
		float *	pWeights,
		int	nPoints
	)

This function lives in the GPU and:

• Computes the chi-square contribution for a single point for a single solution in the swarm for a single pulse with an offset.
• Uses reduceToSum to sum the chisquare contributions over the entire trace. The result is put into the fitness value for our solution.

Parameters

pSolutions	Pointer to solutions array in the GPU.
pFitnesses	Pointer to the array of fitnesse for all solutions in the swarm.
nParams	Number of parameters in the fit (should be 5).
nSol	Number of solutions in the swarm.
pXcoords	Trace x-coordinates array.
pYcoords	Trace y-coordinates array.
pWeights	y weights to apply.
nPoints	Number of points in the trace.

d_fitness2()

__global__ void d_fitness2	(	const float *	pSolutions,
		float *	pFitnesses,
		int	nParams,
		int	nSol,
		unsigned short *	pXcoords,
		unsigned short *	pYcoords,
		float *	pWeights,
		int	nPoints
	)

This function lives in the GPU and:

• Computes the chi-square contribution for a single point for a single solution in the swarm for a single pulse with an offset.
• Uses reduceToSum to sum the chisquare contributions over the entire trace. The result is put into the fitness value for our solution.

Parameters

pSolutions	Pointer to solutions array in the GPU.
pFitnesses	Pointer to the array of fitnesse for all solutions in the swarm.
nParams	Number of parameters in the fit (should be 5).
nSol	Number of solutions in the swarm.
pXcoords	Trace x-coordinates array.
pYcoords	Trace y-coordinates array.
pWeights	y weights to apply.
nPoints	Number of points in the trace.

decay()

__host__ __device__ float decay	(	float	A,
		float	k,
		float	x1,
		float	x
	)

Exponential decay function.

Signals from detectors usually have a falling shape that approximates an exponential. This function evaluates this decay at some point.

Parameters

A	Amplitude of the signal
k	Decay time factor f the signal.
x1	Position of the pulse.
x	Where to evaluate the signal.

Returns

Exponential decay evaluated at x.

doublePulse()

__host__ __device__ float doublePulse	(	float	A1,
		float	k1,
		float	k2,
		float	x1,
		float	A2,
		float	k3,
		float	k4,
		float	x2,
		float	C,
		float	x
	)

Evaluate the canonical form of a double pulse.

This is done by summing two single pulses. The constant term is thrown into the first pulse. The second pulse gets a constant term of 0.

Parameters

A1	Amplitude of the first pulse.
k1	Steepness of first pulse rise.
k2	Decay time of the first pulse.
x1	Position of the first pulse.
A2	Amplitude of the second pulse.
k3	Steepness of second pulse rise.
k4	Decay time of second pulse.
x2	Position of second pulse.
C	Constant offset the pulses sit on.
x	Position at which to evaluate the pulse.

Returns

Value of the double-pulse function evaluated at x.

h_fitDouble()

void h_fitDouble	(	const CudaOptimize::SolutionSet *	solutions,
		CudaOptimize::FitnessSet *	fitnesses,
		dim3	grid,
		dim3	block
	)

Host part to setup computation of the fitnesses across the swarm for our fits for a double pulse. Really this just sets up the kernel call for fitness2 which does the rest.

Parameters

solutions	Pointer to the current Solution set.
fitnesses	Pointer to the current fitness set
grid	Computaional grid geometry.
block	Shapes of the blocks within the grid.

h_fitSingle()

void h_fitSingle	(	const CudaOptimize::SolutionSet *	solutions,
		CudaOptimize::FitnessSet *	fitnesses,
		dim3	grid,
		dim3	block
	)

Invokes the kernel that produces the fitness measure. The fitness is computed in the GPU and is the chi square.

Parameters

solutions	Pointer to the Cuda solution set.
fitnesses	Pointer to the current fitness set
grid	Computational grid being used.
block	Shapes of blocks within the grid.

logistic()

__host__ __device__ float logistic	(	float	A,
		float	k,
		float	x1,
		float	x
	)

Evaluate a logistic function for the specified parameters and point.

A logistic function is a function with a sigmoidal shape. We use it to fit the rising edge of signals DDAS digitizes from detectors. See e.g. https://en.wikipedia.org/wiki/Logistic_function for a discussion of this function.

Parameters

A	Amplitude of the signal.
k	Steepness of the signal (related to the rise time).
x1	Mid point of the rise of the sigmoid.
x	Location at which to evaluate the function.

Returns

Logistic function evaluated at x.

singlePulse()

__host__ __device__ float singlePulse	(	float	A1,
		float	k1,
		float	k2,
		float	x1,
		float	C,
		float	x
	)

Evaluate the value of a single pulse in accordance with our canonical functional form.

The form is a sigmoid rise with an exponential decay that sits on top of a constant offset. The exponential decay is turned on with switchOn() above when x > the rise point of the sigmoid.

Parameters

A1	Pulse amplitiude
k1	Sigmoid rise steepness.
k2	Exponential decay time constant.
x1	Sigmoid position.
C	Constant offset.
x	Position at which to evaluate this function.

Returns

Single pulse evaluated at x.