ClusterMomentsCalculator Namespace Reference

Classes
struct	ClusterMomentCalculationOptions
struct	CMCOptionsHolder
struct	RealSymmetricMatrixSolver
struct	RealSymmetricMatrixSolverIterative

Functions
CUDA_HOS_DEV void	partial_kahan_babushka_neumaier_sum (const float &to_add, float &sum, float &corr)
template<class ... Floats, class disabler = std::enable_if_t < (std::is_same_v<std::decay_t<Floats>, float> && ...) >>
CUDA_HOS_DEV float	sum_kahan_babushka_neumaier (const Floats &... fs)
CUDA_HOS_DEV float	product_sum_cornea_harrison_tang (const float a, const float b, const float c, const float d)
CUDA_HOS_DEV float	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	corrected_dot_product (const float(&a)[3], const float(&b)[3])
CUDA_HOS_DEV void	corrected_cross_product (float(&res)[3], const float a1, const float a2, const float a3, const float b1, const float b2, const float b3)
CUDA_HOS_DEV void	corrected_cross_product (float(&res)[3], const float(&x)[3], const float(&y)[3])
CUDA_HOS_DEV float	corrected_magn_cross_product (const float a1, const float a2, const float a3, const float b1, const float b2, const float b3)
CUDA_HOS_DEV float	corrected_magn_cross_product (const float(&x)[3], const float(&y)[3])
void	register_kernels (IGPUKernelSizeOptimizer &optimizer)
void	calculateClusterPropertiesAndMoments (CaloRecGPU::EventDataHolder &holder, const CaloRecGPU::ConstantDataHolder &instance_data, const CMCOptionsHolder &options, const IGPUKernelSizeOptimizer &optimizer, size_t(&times)[num_time_measurements], const bool synchronize=false, CaloRecGPU::CUDA_Helpers::CUDAStreamPtrHolder stream={}, const bool defer_instead_of_oversize=false)

Variables
constexpr unsigned int	num_time_measurements = 11

Function Documentation

calculateClusterPropertiesAndMoments()

void ClusterMomentsCalculator::calculateClusterPropertiesAndMoments	(	CaloRecGPU::EventDataHolder &	holder,
		const CaloRecGPU::ConstantDataHolder &	instance_data,
		const CMCOptionsHolder &	options,
		const IGPUKernelSizeOptimizer &	optimizer,
		size_t(&)	times[num_time_measurements],
		const bool	synchronize = false,
		CaloRecGPU::CUDA_Helpers::CUDAStreamPtrHolder	stream = {},
		const bool	defer_instead_of_oversize = false
	)

corrected_cross_product() [1/2]

CUDA_HOS_DEV void ClusterMomentsCalculator::corrected_cross_product ( float(&)  res[3], const float  a1, const float  a2, const float  a3, const float  b1, const float  b2, const float  b3  )

inline

Definition at line 93 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    res[0] = product_sum_cornea_harrison_tang(a2, b3, -a3, b2);
    res[1] = product_sum_cornea_harrison_tang(a3, b1, -a1, b3);
    res[2] = product_sum_cornea_harrison_tang(a1, b2, -a2, b1);
  }

corrected_cross_product() [2/2]

CUDA_HOS_DEV void ClusterMomentsCalculator::corrected_cross_product ( float(&)  res[3], const float(&)  x[3], const float(&)  y[3]  )

inline

Definition at line 101 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    corrected_cross_product(res, x[0], x[1], x[2], y[0], y[1], y[2]);
  }

corrected_dot_product() [1/2]

CUDA_HOS_DEV float ClusterMomentsCalculator::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  )

inline

Definition at line 69 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    using namespace std;
 
    const float w_1 = a_1 * b_1;
    const float w_2 = a_2 * b_2;
    const float w_3 = a_3 * b_3;
 
    const float e_1 = fmaf(a_1, b_1, -w_1);
    const float e_2 = fmaf(a_2, b_2, -w_2);
    const float e_3 = fmaf(a_3, b_3, -w_3);
 
    return sum_kahan_babushka_neumaier(w_1, w_2, w_3, e_1, e_2, e_3);
  }

corrected_dot_product() [2/2]

CUDA_HOS_DEV float ClusterMomentsCalculator::corrected_dot_product ( const float(&)  a[3], const float(&)  b[3]  )

inline

Definition at line 86 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    return corrected_dot_product(a[0], a[1], a[2], b[0], b[1], b[2]);
  }

corrected_magn_cross_product() [1/2]

CUDA_HOS_DEV float ClusterMomentsCalculator::corrected_magn_cross_product ( const float  a1, const float  a2, const float  a3, const float  b1, const float  b2, const float  b3  )

inline

Definition at line 108 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    using namespace std;
 
    const float r_1 = product_sum_cornea_harrison_tang(a2, b3, -a3, b2);
    const float r_2 = product_sum_cornea_harrison_tang(a3, b1, -a1, b3);
    const float r_3 = product_sum_cornea_harrison_tang(a1, b2, -a2, b1);
 
#ifdef __CUDA_ARCH__
    return norm3df(r_1, r_2, r_3);
#else
    return hypot(r_1, r_2, r_3);
#endif
 
  }

corrected_magn_cross_product() [2/2]

CUDA_HOS_DEV float ClusterMomentsCalculator::corrected_magn_cross_product ( const float(&)  x[3], const float(&)  y[3]  )

inline

Definition at line 125 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    return corrected_magn_cross_product(x[0], x[1], x[2], y[0], y[1], y[2]);
  }

partial_kahan_babushka_neumaier_sum()

CUDA_HOS_DEV void ClusterMomentsCalculator::partial_kahan_babushka_neumaier_sum ( const float &  to_add, float &  sum, float &  corr  )

inline

Definition at line 21 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    const float t = sum + to_add;
 
    const bool test = fabsf(sum) >= fabsf(to_add);
 
    const float opt_1 = (sum - t) + to_add;
    const float opt_2 = (to_add - t) + sum;
 
    corr += (test) * opt_1 + (!test) * opt_2;
 
    sum = t;
  }

product_sum_cornea_harrison_tang()

CUDA_HOS_DEV float ClusterMomentsCalculator::product_sum_cornea_harrison_tang ( const float  a, const float  b, const float  c, const float  d  )

inline

Definition at line 54 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    using namespace std;
 
    const float w_1 = a * b;
    const float w_2 = c * d;
 
    const float e_1 = fmaf(a, b, -w_1);
    const float e_2 = fmaf(c, d, -w_2);
 
    return sum_kahan_babushka_neumaier(w_1, w_2, e_1, e_2);
  }

register_kernels()

void ClusterMomentsCalculator::register_kernels

(

IGPUKernelSizeOptimizer &

optimizer

)

sum_kahan_babushka_neumaier()

template<class ... Floats, class disabler = std::enable_if_t < (std::is_same_v<std::decay_t<Floats>, float> && ...) >>

CUDA_HOS_DEV float ClusterMomentsCalculator::sum_kahan_babushka_neumaier

(

const Floats &...

fs

)

Definition at line 40 of file GPUClusterInfoAndMomentsCalculatorImpl.h.

  {
    float ret = 0.f;
    float corr = 0.f;
 
    (partial_kahan_babushka_neumaier_sum(fs, ret, corr), ...);
 
    return ret + corr;
  }

Variable Documentation

num_time_measurements

constexpr unsigned int ClusterMomentsCalculator::num_time_measurements = 11

constexpr

Definition at line 801 of file GPUClusterInfoAndMomentsCalculatorImpl.h.