RooBatchCompute.cu File Reference

#include "RooBatchCompute.h"
#include "RooNaNPacker.h"
#include "Batches.h"
#include "CudaInterface.h"
#include <algorithm>
#include <array>
#include <cassert>
#include <cstring>
#include <functional>
#include <map>
#include <queue>
#include <unordered_map>
#include <vector>

Include dependency graph for RooBatchCompute.cu:

Classes
class	RooBatchCompute::CUDA::RooBatchComputeClass
	This class overrides some RooBatchComputeInterface functions, for the purpose of providing a cuda specific implementation of the library. More...

Namespaces
namespace	RooBatchCompute
	Namespace for dispatching RooFit computations to various backends.
namespace	RooBatchCompute::CUDA

Functions
std::vector< void(*)(Batches &)>	RooBatchCompute::CUDA::getFunctions ()
	Returns a std::vector of pointers to the compute functions in this file.
__global__ void	RooBatchCompute::CUDA::kahanSum (const double *__restrict__ input, const double *__restrict__ carries, size_t n, double *__restrict__ result, bool nll)
__device__ void	RooBatchCompute::CUDA::kahanSumReduction (double *shared, size_t n, double *__restrict__ result, int carry_index)
__device__ void	RooBatchCompute::CUDA::kahanSumUpdate (double &sum, double &carry, double a, double otherCarry)
__global__ void	RooBatchCompute::CUDA::nllSumKernel (const double *__restrict__ probas, const double *__restrict__ weights, const double *__restrict__ offsetProbas, size_t nProbas, double scalarProba, size_t nWeights, double *__restrict__ result, double *__restrict__ stats)
	Computes the negative log likelihood sum with the same semantics as the CPU implementation of RooBatchComputeInterface::reduceNLL(): zero-weight events are skipped, and evaluation problems are counted and accumulated into a "badness" value that the host can pack into a NaN for the error recovery in the minimizer.

Variables
constexpr int	RooBatchCompute::CUDA::blockSize = 512
static RooBatchComputeClass	RooBatchCompute::CUDA::computeObj
	Static object to trigger the constructor which overwrites the dispatch pointer.

Variable Documentation

_arr

CudaInterface::PinnedHostArray<double> _arr

private

Definition at line 643 of file RooBatchCompute.cu.

_cudaStream

CudaInterface::CudaStream* _cudaStream = nullptr

private

Definition at line 707 of file RooBatchCompute.cu.

_deferredCursor

std::size_t _deferredCursor = 0

private

Definition at line 225 of file RooBatchCompute.cu.

_deferredSlots

std::vector<DeferredSlot> _deferredSlots

private

Definition at line 224 of file RooBatchCompute.cu.

_gpuBuffer

GPUBufferContainer _gpuBuffer

private

Definition at line 706 of file RooBatchCompute.cu.

_lastAccess

LastAccessType _lastAccess = LastAccessType::CPU_READ

mutableprivate

Definition at line 708 of file RooBatchCompute.cu.

_next

std::size_t _next = 0

private

Definition at line 223 of file RooBatchCompute.cu.

_queue

Queue& _queue

private

Definition at line 741 of file RooBatchCompute.cu.

_queuesMaps

std::unique_ptr<BufferQueuesMaps> _queuesMaps

private

Definition at line 781 of file RooBatchCompute.cu.

_slots

std::array<Slot, 64> _slots

private

Definition at line 222 of file RooBatchCompute.cu.

_val

double _val

private

Definition at line 584 of file RooBatchCompute.cu.

_vec

std::unique_ptr<Container> _vec

private

Definition at line 612 of file RooBatchCompute.cu.

capacity

std::size_t capacity = 0

Definition at line 105 of file RooBatchCompute.cu.

cpuBufferQueuesMap

std::map<std::size_t, CPUBuffer::Queue> cpuBufferQueuesMap

Definition at line 751 of file RooBatchCompute.cu.

device

char* device = nullptr

Definition at line 104 of file RooBatchCompute.cu.

dst

double* dst = nullptr

Definition at line 162 of file RooBatchCompute.cu.

event

cudaEvent_t event = nullptr

Definition at line 106 of file RooBatchCompute.cu.

gpuBufferQueuesMap

std::map<std::size_t, GPUBuffer::Queue> gpuBufferQueuesMap

Definition at line 752 of file RooBatchCompute.cu.

host

char* host = nullptr

Definition at line 103 of file RooBatchCompute.cu.

inFlight

bool inFlight = false

Definition at line 107 of file RooBatchCompute.cu.

nPending

std::size_t nPending = 0

Definition at line 163 of file RooBatchCompute.cu.

pinnedBufferQueuesMap

std::map<std::size_t, PinnedBuffer::Queue> pinnedBufferQueuesMap

Definition at line 753 of file RooBatchCompute.cu.

scalarBufferQueuesMap

std::map<std::size_t, ScalarBuffer::Queue> scalarBufferQueuesMap

Definition at line 750 of file RooBatchCompute.cu.