Class AAHD_AngleSet

Inheritance Relationships

Base Type

Class Documentation

class AAHD_AngleSet : public opensn::AngleSet

AAHD angle set.

Fused all-at-once batch interface

Used only by SweepScheduler::ScheduleAlgoAAOFused.

A batch run sweeps every level of every member angle set in one uninterrupted sequence of launches on one stream. No per-level progress is kept between calls. See AAHDSweepChunk::SubmitBatchRun.

void SetFLUDSStream(crb::Stream &stream)

Direct this angle set’s FLUDS host-device copies to a stream.

Parameters:

stream – Stream for the copies.

void PrepareForBatch(SweepChunk &sweep_chunk, crb::Stream &stream)

Prepare this angle set’s device buffers for a batch run.

This mirrors the first part of AngleSetAdvance. Call it once per sweep, before the run is submitted.

Parameters:

  • sweep_chunk – Sweep chunk that runs the batch.

  • stream – Stream of the lane whose batch run this angle set joins.

std::size_t NumLevels() const

Return the number of levels in this angle set’s levelized local subgrid.

void FillBatchSlots(std::size_t num_levels, std::size_t slot_stride, gpu_kernel::AAHD_BatchSlot *slots) const

Write this angle set’s batch slot for each level of a batch run.

Levels at or beyond NumLevels() get an empty slot because every batch member runs through the same level range.

Parameters:

  • num_levels – Number of levels in the batch run.

  • slot_stride – Distance between this angle set’s slots of consecutive levels.

  • slots – Slot array. The slot for level level is written to slots[level * slot_stride].

void SubmitCompletionCopies()

Enqueue this angle set’s post-sweep device-to-host copies without synchronizing.

The copies go on the angle set’s FLUDS stream, which is its run’s lane. The caller waits for that stream before FinishCompletion.

void FinishCompletion(SweepChunk &sweep_chunk)

Finish this angle set after its SubmitCompletionCopies copies have completed.

This decrements the followers’ dependency counters, sends the downstream angular fluxes, and copies the saved angular flux to its destination. It mirrors the second part of AngleSetAdvance.

Parameters:

sweep_chunk – Sweep chunk that ran the batch.

Public Functions

AAHD_AngleSet(size_t id, const LBSGroupset &groupset, const SPDS &spds, std::shared_ptr<FLUDS> &fluds, std::vector<size_t> &angle_indices, std::map<uint64_t, std::shared_ptr<SweepBoundary>> &boundaries, int maximum_message_size, const SweepCommunicator &sweep_communicator)
inline crb::Stream &GetStream()
inline std::uint32_t *GetDeviceAngleIndices()
virtual void InitializeDelayedUpstreamData() override

Initializes delayed upstream data.

This method gets called when a sweep scheduler is constructed.

inline virtual int GetMaxBufferMessages() const override

Returns the maximum buffer size from the sweepbuffer.

inline virtual void SetMaxBufferMessages(int count) override

Sets the maximum buffer size for the sweepbuffer.

void PrepostReceives()
bool IsReady()
void WaitForDownstreamAndDelayed()
virtual AngleSetStatus AngleSetAdvance(SweepChunk &sweep_chunk, AngleSetStatus permission) override

This function advances the work stages of an angleset.

inline virtual AngleSetStatus FlushSendBuffers() override
inline virtual void ResetSweepBuffers() override

Resets the sweep buffer.

inline virtual bool ReceiveDelayedData() override

Instructs the sweep buffer to receive delayed data.

virtual void SyncDeviceAngleIndices() override

Update the angle index data on the device to match the host.

inline void AddBoundaryOffset(std::uint64_t offset)
void CopyBoundaryOffsetToDevice()
inline std::uint64_t *GetDeviceBoudnaryOffset()

Public Static Attributes

static std::mutex m

Protected Attributes

crb::Stream stream_ = crb::Stream::get_null_stream()

Associated CUDA stream.

AAHD_ASynchronousCommunicator async_comm_

Asynchronous communicator.

crb::DeviceMemory<std::uint32_t> device_angle_indices_

Angle indices on GPU.

std::vector<std::uint64_t> host_boundary_offsets_

Vector of offsets into the boundary bank.

crb::DeviceMemory<std::uint64_t> device_boundary_offsets_

Boundary offset on GPU.