#include <utility>
#include <type_traits>
#include <cstring>
#include <string>
#include <cstdio>
#include <iostream>
#include <thread>
#include <mutex>
#include <shared_mutex>
#include <memory>
#include <vector>
#include <climits>
#include <new>
#include <cmath>

Classes
struct	CaloRecGPU::CUDA_Helpers::CUDAStreamPtrHolder
struct	CaloRecGPU::Helpers::MemoryContext::CPU
struct	CaloRecGPU::Helpers::MemoryContext::CUDAGPU
struct	CaloRecGPU::Helpers::MemoryContext::CUDAPinnedCPU
class	CaloRecGPU::Helpers::MemoryManagement< T, indexer >
	! Handles allocation of a type `T`, using `indexer` as the integer type to indicate sizes. More...
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::unary_helper< MemoryContext::CPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::unary_helper< MemoryContext::CUDAGPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::unary_helper< MemoryContext::CUDAPinnedCPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CPU, MemoryContext::CPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CPU, MemoryContext::CUDAGPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CUDAGPU, MemoryContext::CUDAGPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CUDAGPU, MemoryContext::CPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CUDAPinnedCPU, MemoryContext::CPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CPU, MemoryContext::CUDAPinnedCPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CUDAPinnedCPU, MemoryContext::CUDAPinnedCPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CUDAPinnedCPU, MemoryContext::CUDAGPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::copy_helper< MemoryContext::CUDAGPU, MemoryContext::CUDAPinnedCPU, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::move_helper< C1, C2, dummy >
struct	CaloRecGPU::Helpers::MemoryManagement< T, indexer >::move_helper< C, C, dummy >
class	CaloRecGPU::Helpers::SimpleContainer< T, indexer, Context, true >
class	CaloRecGPU::Helpers::SimpleContainer< T, indexer, Context, false >
class	CaloRecGPU::Helpers::SimpleHolder< T, Context, true >
class	CaloRecGPU::Helpers::SimpleHolder< T, Context, false >
class	CaloRecGPU::Helpers::separate_thread_holder< T >
	Manages objects of type `T` in a thread-safe way, ensuring that there's an object available for each separate thread while minimizing the number of allocations. More...
struct	CaloRecGPU::Helpers::separate_thread_accessor< T >
	! More...
struct	CaloRecGPU::Helpers::maybe_allocate< T >
	Possibly holds an object in its internal buffer. More...

Namespaces
namespace	CaloRecGPU
	Copyright (C) 2002-2025 CERN for the benefit of the ATLAS collaboration.
namespace	CaloRecGPU::CUDA_Helpers
namespace	CaloRecGPU::Helpers
namespace	CaloRecGPU::Helpers::Constants
	!
namespace	CaloRecGPU::Helpers::MemoryContext
	! Holds dummy classes just to identify the place in which memory lives.

Macros
#define	CUDA_AVAILABLE 0
#define	CUDA_HOS_DEV
#define	CUDA_ERRCHECK(...)

Typedefs
template<class T, class indexer = unsigned int>
using	CaloRecGPU::Helpers::CPU_array = SimpleContainer<T, indexer, MemoryContext::CPU, true>
	Holds a run-time specified amount of objects of type `T` in CPU memory.
template<class T, class indexer = unsigned int>
using	CaloRecGPU::Helpers::CUDA_array = SimpleContainer<T, indexer, MemoryContext::CUDAGPU, true>
	Holds a run-time specified amount of objects of type `T` in CUDA GPU memory.
template<class T, class indexer = unsigned int>
using	CaloRecGPU::Helpers::CUDA_kernel_array = SimpleContainer<T, indexer, MemoryContext::CUDAGPU, false>
	Non-owning pointer to an array of `T` in CUDA GPU memory.
template<class T>
using	CaloRecGPU::Helpers::CPU_object = SimpleHolder<T, MemoryContext::CPU, true>
	Holds an object of type `T` in CPU memory.
template<class T>
using	CaloRecGPU::Helpers::CUDA_object = SimpleHolder<T, MemoryContext::CUDAGPU, true>
	Holds an object of type `T` in CUDA GPU memory.
template<class T>
using	CaloRecGPU::Helpers::CUDA_kernel_object = SimpleHolder<T, MemoryContext::CUDAGPU, false>
	Non-owning pointer to an object of type `T` in CUDA GPU memory.
template<class T>
using	CaloRecGPU::Helpers::CUDA_pinned_CPU_object = SimpleHolder<T, MemoryContext::CUDAPinnedCPU, true>
	Holds an object of type `T` in CUDA GPU memory.

Functions
void *	CaloRecGPU::CUDA_Helpers::allocate (const size_t num)
	Allocates and returns the address of `num` bytes from GPU memory.
void	CaloRecGPU::CUDA_Helpers::deallocate (void *address)
	Deallocates `address` in GPU memory.
void *	CaloRecGPU::CUDA_Helpers::allocate_pinned (const size_t num)
	Allocates and returns the address of `num` bytes from CPU pinned memory.
void	CaloRecGPU::CUDA_Helpers::deallocate_pinned (void *address)
	Deallocates `address` in CPU pinned memory.
void	CaloRecGPU::CUDA_Helpers::GPU_to_CPU (void dest, const void const source, const size_t num)
	Copies `num` bytse from `source` in GPU memory to `dest` in CPU memory.
void	CaloRecGPU::CUDA_Helpers::CPU_to_GPU (void dest, const void const source, const size_t num)
	Copies `num` bytes from `source` in CPU memory to `dest` in GPU memory.
void	CaloRecGPU::CUDA_Helpers::GPU_to_GPU (void dest, const void const source, const size_t num)
	Copies `num` bytes from `source` to `dest`, both in GPU memory.
void	CaloRecGPU::CUDA_Helpers::GPU_to_CPU_async (void dest, const void const source, const size_t num, CUDAStreamPtrHolder stream={})
	Copies `num` bytes from `source` in GPU memory to `dest` in CPU memory, asynchronously.
void	CaloRecGPU::CUDA_Helpers::CPU_to_GPU_async (void dest, const void const source, const size_t num, CUDAStreamPtrHolder stream={})
	Copies `num` bytes from `source` in CPU memory to `dest` in GPU memory, asynchronously.
void	CaloRecGPU::CUDA_Helpers::GPU_to_GPU_async (void dest, const void const source, const size_t num, CUDAStreamPtrHolder stream={})
	Copies `num` bytes from `source` to `dest`, both in GPU memory, asynchronously.
void	CaloRecGPU::CUDA_Helpers::GPU_synchronize (CUDAStreamPtrHolder stream={})
	Synchronizes the `stream`.
void	CaloRecGPU::CUDA_Helpers::optimize_block_and_grid_size (void *func, int &block_size, int &grid_size, const int dynamic_memory=0, const int block_size_limit=0)
	Optimizes block and grid size according to `cudaOccupancyMaxPotentialBlockSize`.
void	CaloRecGPU::CUDA_Helpers::optimize_block_and_grid_size_for_cooperative_launch (void *func, int &block_size, int &grid_size, const int dynamic_memory=0, const int block_size_limit=0)
	Optimizes block and grid size for a cooperative launch.
bool	CaloRecGPU::CUDA_Helpers::supports_cooperative_launches ()
bool	CaloRecGPU::CUDA_Helpers::supports_dynamic_parallelism ()
std::string	CaloRecGPU::CUDA_Helpers::GPU_name ()
template<class T1, class T2>
constexpr auto	CaloRecGPU::Helpers::int_ceil_div (const T1 num, const T2 denom)
	Returns the ceiling of num/denom, with proper rounding.
template<class T1, class T2>
constexpr auto	CaloRecGPU::Helpers::int_floor_div (const T1 num, const T2 denom)
	Returns the floor of num/denom, with proper rounding.
template<class Base = float, class Exp = int>
constexpr Base	CaloRecGPU::Helpers::compile_time_pow2 (const Exp exp)
	Returns 2 to the power of `exp`.
template<class T>
constexpr unsigned int	CaloRecGPU::Helpers::int_ceil_log_2 (T num)
	Returns the ceiling of the base-2 logarithm of a number (i.
template<class T>
constexpr unsigned char	CaloRecGPU::Helpers::Pearson_hash (const T number)
	Calculates a Pearson hash from @ number.
template<class T>
constexpr unsigned short	CaloRecGPU::Helpers::Pearson_hash_16_bit (const T number)
	Calculates a 16-bit Pearson hash from @ number.
static CUDA_HOS_DEV float	CaloRecGPU::Helpers::erf_inv_wrapper (const float x)
static CUDA_HOS_DEV float	CaloRecGPU::Helpers::regularize_angle (const float b, const float a=0.f)
static CUDA_HOS_DEV double	CaloRecGPU::Helpers::regularize_angle (const double b, const double a=0.)
template<class T>
static CUDA_HOS_DEV T	CaloRecGPU::Helpers::angular_difference (const T x, const T y)
static CUDA_HOS_DEV float	CaloRecGPU::Helpers::eta_from_coordinates (const float x, const float y, const float z)
static CUDA_HOS_DEV double	CaloRecGPU::Helpers::eta_from_coordinates (const double x, const double y, const double z)
static CUDA_HOS_DEV void	CaloRecGPU::Helpers::partial_kahan_babushka_neumaier_sum (const float &to_add, float &sum, float &corr)
	Implements one step of a Kahan-Babushka-Neumaier sum by adding `to_add` to `sum` with the correction term `corr`.
template<class ... Floats, class disabler = std::enable_if_t < (std::is_same_v<std::decay_t<Floats>, float> && ...) >>
CUDA_HOS_DEV float	CaloRecGPU::Helpers::sum_kahan_babushka_neumaier (const Floats &... fs)
	Adds a list of floats using the Kahan-Babushka-Neumaier algorithm for greater precision (at the cost of additional operations).
static CUDA_HOS_DEV float	CaloRecGPU::Helpers::product_sum_cornea_harrison_tang (const float a, const float b, const float c, const float d)
static CUDA_HOS_DEV float	CaloRecGPU::Helpers::corrected_dot_product (const float a_1, const float a_2, const float a_3, const float b_1, const float b_2, const float b_3)
CUDA_HOS_DEV float	CaloRecGPU::Helpers::corrected_dot_product (const float(&a)[3], const float(&b)[3])
static CUDA_HOS_DEV void	CaloRecGPU::Helpers::corrected_cross_product (float(&res)[3], const float a1, const float a2, const float a3, const float b1, const float b2, const float b3)
static CUDA_HOS_DEV void	CaloRecGPU::Helpers::corrected_cross_product (float(&res)[3], const float(&x)[3], const float(&y)[3])
static CUDA_HOS_DEV float	CaloRecGPU::Helpers::corrected_magn_cross_product (const float a1, const float a2, const float a3, const float b1, const float b2, const float b3)
static CUDA_HOS_DEV float	CaloRecGPU::Helpers::corrected_magn_cross_product (const float(&x)[3], const float(&y)[3])

Variables
template<class T>
constexpr T	CaloRecGPU::Helpers::Constants::pi = T(3.1415926535897932384626433832795028841971693993751058209749445923078164062862089986280348253421170679821480865132823066470938446095505822317253594081284811174502841027019385211055596446229489549303819644288109756659334461284756482337867831652712019091456485669234603486104543266482133936072602491412737245870066063155881748815209209628292540917153643678925903600113305305488204665213841469519415116094330572703657595919530921861173819326117931051185480744623799627495673518857527248912279381830119491298336733624L)
template<class T>
constexpr T	CaloRecGPU::Helpers::Constants::sqrt2 = T(1.4142135623730950488016887242096980785696718753769480731766797379907324784621070388503875343276415727350138462309122970249248360558507372126441214970999358314132226659275055927557999505011527820605714701095599716059702745345968620147285174186408891986095523292304843087143214508397626036279952514079896872533965463318088296406206152583523950547457502877599617298355752203375318570113543746034084988471603868999706990048150305440277903164542478230684929369186215805784631115966687130130156185689872372352885092649L)
template<class T>
constexpr T	CaloRecGPU::Helpers::Constants::inv_sqrt2 = T(0.70710678118654752440084436210484903928483593768847403658833986899536623923105351942519376716382078636750692311545614851246241802792536860632206074854996791570661133296375279637789997525057639103028573505477998580298513726729843100736425870932044459930477616461524215435716072541988130181399762570399484362669827316590441482031030762917619752737287514387998086491778761016876592850567718730170424942358019344998534950240751527201389515822712391153424646845931079028923155579833435650650780928449361861764425463243L)