Namespace Acts::Cuda::Details

namespace Acts::Cuda::Details

Namespace holding some implementation detail types that should not be used directly in client code.

Functions

DubletCounts countDublets(std::size_t maxBlockSize, std::size_t nMiddleSP, const device_array<unsigned int> &middleBottomCounts, const device_array<unsigned int> &middleTopCounts)

Calculate summary values for the dublet search.

After the dublet search is done, we need to know some information about how many duplets were found exactly. As this information is necessary for the scheduing of the subsequent steps of the execution on the GPU. This function is used to collect this information

Parameters
  • maxBlockSize – The maximum block size to use on the GPU

  • nMiddleSP – The number of middle spacepoints for which the dublet reconstruction was run

  • middleBottomCounts – The output from the Acts::Cuda::Details::findDublets(…) function with the same name

  • middleTopCounts – The output from the Acts::Cuda::Details::findDublets(…) function with the same name

Returns

An object holding all the summary statistics necessary for the subsequent steps of GPU execution

void findDublets(std::size_t maxBlockSize, std::size_t nBottomSPs, const device_array<SpacePoint> &bottomSPs, std::size_t nMiddleSPs, const device_array<SpacePoint> &middleSPs, std::size_t nTopSPs, const device_array<SpacePoint> &topSPs, float deltaRMin, float deltaRMax, float cotThetaMax, float collisionRegionMin, float collisionRegionMax, device_array<unsigned int> &middleBottomCounts, device_array<std::size_t> &middleBottomDublets, device_array<unsigned int> &middleTopCounts, device_array<std::size_t> &middleTopDublets)

Find all viable middle-bottom and middle-top dublets.

This function is run as the first step in the seed finding, looking for viable middle-bottom and middle-top spacepoint pairs for the subsequent steps of the code.

Note that middleBottomCounts and middleTopCounts have type “unsigned int” instead of “std::size_t”, because the GPU code needs to execute atomic operations on these arrays. And CUDA does not define such operations on std::size_t (i.e. unsigned long).

Parameters
  • maxBlockSize[in] The maximum block size to use on the GPU

  • nBottomSPs[in] The number of bottom spacepoints in bottomSPs

  • bottomSPs[in] Properties of all of the bottom spacepoints

  • nMiddleSPs[in] The number of middle spacepoints in middleSPs

  • middleSPs[in] Properties of all of the middle spacepoints

  • nTopSPs[in] The number of top spacepoints in topSPs

  • topSPs[in] Properties of all of the top spacepoints

  • deltaRMin[in] Configuration parameter from Acts::SeedfinderConfig

  • deltaRMax[in] Configuration parameter from Acts::SeedfinderConfig

  • cotThetaMax[in] Configuration parameter from Acts::SeedfinderConfig

  • collisionRegionMin[in] Configuration parameter from Acts::SeedfinderConfig

  • collisionRegionMax[in] Configuration parameter from Acts::SeedfinderConfig

  • middleBottomCounts[out] 1-D array of the number of middle-bottom dublets found for each middle spacepoint

  • middleBottomDublets[out] 2-D matrix of size nMiddleSPs x nBottomSPs, holding the bottom spacepoint indices for the identified middle-bottom dublets

  • middleTopCounts[out] 1-D array of the number of middle-top dublets found for each middle spacepoint

  • middleTopDublets[out] 2-D matrix of size nMiddleSPs x nTopSPs, holding the top spacepoint indices for the identified middle-top dublets

std::vector<std::vector<Triplet>> findTriplets(const Info::Device &device, std::size_t maxBlockSize, const DubletCounts &dubletCounts, const SeedFilterConfig &seedConfig, const TripletFilterConfig &filterConfig, std::size_t nBottomSPs, const device_array<SpacePoint> &bottomSPs, std::size_t nMiddleSPs, const device_array<SpacePoint> &middleSPs, std::size_t nTopSPs, const device_array<SpacePoint> &topSPs, const device_array<unsigned int> &middleBottomCounts, const device_array<std::size_t> &middleBottomDublets, const device_array<unsigned int> &middleTopCounts, const device_array<std::size_t> &middleTopDublets, float maxScatteringAngle2, float sigmaScattering, float minHelixDiameter2, float pT2perRadius, float impactMax)

Find all viable triplets from the provided spacepoint dublets.

This function is used to find a “loosely selected” set of seed candidates that still need to be filtered through Acts::SeedFilter::filterSeeds_1SpFixed before returning it to the user.

Parameters
  • device[in] Properties of the device that the code will be running on

  • maxBlockSize[in] The maximum block size to use on the GPU

  • dubletCounts[in] The output object from Acts::Cuda::Details::countDublets

  • seedConfig[in] Configuration parameters for the triplet finding/filtering

  • filterConfig[in] User provided settings (code…) for the triplet filtering

  • nBottomSPs[in] The number of bottom spacepoints in bottomSPs

  • bottomSPs[in] Properties of all of the bottom spacepoints

  • nMiddleSPs[in] The number of middle spacepoints in middleSPs

  • middleSPs[in] Properties of all of the middle spacepoints

  • nTopSPs[in] The number of top spacepoints in topSPs

  • topSPs[in] Properties of all of the top spacepoints

  • middleBottomCounts[in] 1-D array of the number of middle-bottom dublets found for each middle spacepoint

  • middleBottomDublets[in] 2-D matrix of size nMiddleSPs x nBottomSPs, holding the bottom spacepoint indices for the identified middle-bottom dublets

  • middleTopCounts[in] 1-D array of the number of middle-top dublets found for each middle spacepoint

  • middleTopDublets[in] 2-D matrix of size nMiddleSPs x nTopSPs, holding the top spacepoint indices for the identified middle-top dublets

  • maxScatteringAngle2[in] Configuration parameter from Acts::SeedfinderConfig

  • sigmaScattering[in] Configuration parameter from Acts::SeedfinderConfig

  • minHelixDiameter2[in] Configuration parameter from Acts::SeedfinderConfig

  • pT2perRadius[in] Configuration parameter from Acts::SeedfinderConfig

  • impactMax[in] Configuration parameter from Acts::SeedfinderConfig

Returns

A 2-D structure holding the parameters of the identified triplets for each middle spacepoint

class DeviceArrayDeleter
#include </home/docs/checkouts/readthedocs.org/user_builds/acts/checkouts/v13.0.0/Plugins/Cuda/include/Acts/Plugins/Cuda/Utilities/Arrays.hpp>

Class performing the deletion of a CUDA device memory array.

Public Functions

void operator()(void *ptr)

Operator performing the deletion of the memory.

struct DubletCounts
#include </home/docs/checkouts/readthedocs.org/user_builds/acts/checkouts/v13.0.0/Plugins/Cuda/include/Acts/Plugins/Cuda/Seeding2/Details/Types.hpp>

Helper struct summarising the results of the dublet search.

Public Members

unsigned int maxMBDublets = 0

The maximal number of middle-bottom dublets.

unsigned int maxMTDublets = 0

The maximal number of middle-top dublets.

unsigned int maxTriplets = 0

The maximal number of triplets for any middle SP.

unsigned int nDublets = 0

The total number of dublets (M-B and M-T) found.

unsigned int nTriplets = 0

The total number of triplet candidates found.

class HostArrayDeleter
#include </home/docs/checkouts/readthedocs.org/user_builds/acts/checkouts/v13.0.0/Plugins/Cuda/include/Acts/Plugins/Cuda/Utilities/Arrays.hpp>

Class performing the deletion of pinned host memory.

Public Functions

void operator()(void *ptr)

Operator performing the deletion of the memory.

struct LinCircle
#include </home/docs/checkouts/readthedocs.org/user_builds/acts/checkouts/v13.0.0/Plugins/Cuda/include/Acts/Plugins/Cuda/Seeding2/Details/Types.hpp>

Helper struct holding the linearly transformed coordinates of spacepoints.

Public Members

float cotTheta = 0.0f
float Er = 0.0f
float iDeltaR = 0.0f
float U = 0.0f
float V = 0.0f
float Zo = 0.0f
struct SpacePoint
#include </home/docs/checkouts/readthedocs.org/user_builds/acts/checkouts/v13.0.0/Plugins/Cuda/include/Acts/Plugins/Cuda/Seeding2/Details/Types.hpp>

Helper struct describing a spacepoint on the device.

Public Members

float radius = 0.0f

radius in beam system coordinates

float varianceR = 0.0f
float varianceZ = 0.0f
float x = 0.0f

x-coordinate in beam system coordinates

float y = 0.0f

y-coordinate in beam system coordinates

float z = 0.0f

z-coordinate in beam system coordinates

struct Triplet
#include </home/docs/checkouts/readthedocs.org/user_builds/acts/checkouts/v13.0.0/Plugins/Cuda/include/Acts/Plugins/Cuda/Seeding2/Details/Types.hpp>

Structure used in the CUDA-based triplet finding.

Public Members

unsigned int bottomIndex = static_cast<unsigned int>(-1)
float impactParameter = 0.0f
float invHelixDiameter = 0.0f
unsigned int middleIndex = static_cast<unsigned int>(-1)
unsigned int topIndex = static_cast<unsigned int>(-1)
float weight = 0.0f