RooFitPhotopeakKernels.hpp File Reference

Host-callable launch wrappers for the photopeak PDF CUDA kernels. More...

#include <cstddef>

Include dependency graph for RooFitPhotopeakKernels.hpp:

Go to the source code of this file.

Functions
void	RooLowExpTail_launchKernel (double *output, const double *x_vals, double mu, double inv_tau, double inv_sqrt2_sigma, size_t n)
	Launch the low-energy exponential tail kernel.
void	RooStepShelf_launchKernel (double *output, const double *x_vals, double mu, double inv_sigma, size_t n)
	Launch the resolution-smeared step-shelf kernel.
void	RooLowLinTail_launchKernel (double *output, const double *x_vals, double mu, double slope, double inv_sqrt2_sigma, size_t n)
	Launch the low-energy linear tail kernel.
void	RooHighExpTail_launchKernel (double *output, const double *x_vals, double mu, double inv_tau, double inv_sqrt2_sigma, size_t n)
	Launch the high-energy exponential tail kernel.

Detailed Description

Host-callable launch wrappers for the photopeak PDF CUDA kernels.

Each launcher takes device pointers plus the constants hoisted once per Minuit evaluation — reciprocals are passed in already inverted so the kernel multiplies rather than divides per element — dispatches the kernel, and returns as soon as the stream has accepted the launch.

These are declared unconditionally, but only defined when the library is built with -DAU_USE_CUDA=ON. The CPU build never compiles gpu/*.cu, so calling one of these in a CPU build is a link error. The custom PDFs reach them only from a doEval path already guarded by canComputeBatchWithCuda().

Warning

Every pointer argument must be device-resident. The callers obtain them from RooFit's evaluation context and check with cudaPointerGetAttributes before dispatching.

Note

No implicit synchronisation: the launch is asynchronous and the caller is responsible for any synchronisation the surrounding evaluation needs.

Definition in file RooFitPhotopeakKernels.hpp.

Function Documentation

RooHighExpTail_launchKernel()

void RooHighExpTail_launchKernel	(	double *	output,
		const double *	x_vals,
		double	mu,
		double	inv_tau,
		double	inv_sqrt2_sigma,
		size_t	n )

Launch the high-energy exponential tail kernel.

Parameters

output	Device buffer for n PDF values.
x_vals	Device buffer of n observable values.
mu	Peak centroid, in the observable's units.
inv_tau	Reciprocal of the tail decay constant.
inv_sqrt2_sigma	1 / (sqrt(2) * sigma), the resolution factor.
n	Number of elements.

Definition at line 37 of file RooHighExpTail.cu.

Referenced by RooHighExpTail::doEval().

RooLowExpTail_launchKernel()

void RooLowExpTail_launchKernel	(	double *	output,
		const double *	x_vals,
		double	mu,
		double	inv_tau,
		double	inv_sqrt2_sigma,
		size_t	n )

Launch the low-energy exponential tail kernel.

Parameters

output	Device buffer for n PDF values.
x_vals	Device buffer of n observable values.
mu	Peak centroid, in the observable's units.
inv_tau	Reciprocal of the tail decay constant.
inv_sqrt2_sigma	1 / (sqrt(2) * sigma), the resolution factor.
n	Number of elements.

Definition at line 37 of file RooLowExpTail.cu.

Referenced by RooLowExpTail::doEval().

RooLowLinTail_launchKernel()

void RooLowLinTail_launchKernel	(	double *	output,
		const double *	x_vals,
		double	mu,
		double	slope,
		double	inv_sqrt2_sigma,
		size_t	n )

Launch the low-energy linear tail kernel.

Parameters

output	Device buffer for n PDF values.
x_vals	Device buffer of n observable values.
mu	Peak centroid, in the observable's units.
slope	Linear tail slope s in the (1 + s*(x - mu)) factor, which is floored at zero to keep the PDF non-negative.
inv_sqrt2_sigma	1 / (sqrt(2) * sigma), the resolution factor.
n	Number of elements.

Definition at line 24 of file RooLowLinTail.cu.

Referenced by RooLowLinTail::doEval().

RooStepShelf_launchKernel()

void RooStepShelf_launchKernel	(	double *	output,
		const double *	x_vals,
		double	mu,
		double	inv_sigma,
		size_t	n )

Launch the resolution-smeared step-shelf kernel.

Parameters

output	Device buffer for n PDF values.
x_vals	Device buffer of n observable values.
mu	Peak centroid, in the observable's units.
inv_sigma	Reciprocal of the Gaussian resolution.
n	Number of elements.

Definition at line 27 of file RooStepShelf.cu.

Referenced by RooStepShelf::doEval().