Expand description
Execution backends, dataset queries, and reduction support for compiled laddu models.
Structs§
- BinSpec
- Validated, monotonically increasing bin edges.
- Binning
Axis - A finite, strictly increasing one-dimensional bin axis.
- CpuBackend
- Prepares compiled models for CPU execution.
- CpuBatch
Cache - Materialized event-dependent values for one batch.
- CpuCached
Batch - A cached event batch and its associated weights.
- CpuCached
Dataset - A dataset whose event-dependent model values are fully cached in memory.
- CpuOptions
- CPU-specific execution options.
- CpuPlan
- A compiled model prepared for CPU evaluation.
- Dataset
Bin - A lazily filtered dataset corresponding to one bin.
- Device
Identity - Stable information used to match a runtime accelerator to platform telemetry.
- Execution
- Resolved resources and policies used to execute models.
- Execution
Options - Options used to construct an
Executioncontext. - GpuOptions
- GPU-specific execution options.
- Joint
Histogram - An N-dimensional empirical weighted histogram stored in row-major order.
- Joint
Histogram Diagnostics - Aggregate diagnostics for events that could not be placed in a joint histogram.
- Memory
Decision - One memory-derived execution decision.
- Memory
Lease - A live memory reservation. Clones share one reservation.
- Memory
Plan - Host and optional accelerator budgets for one execution.
- Memory
Pool - A resolved, reservable limit within one physical resource.
- Memory
Pool Report - Report for one resolved pool.
- Memory
Report - Report covering all resources in a
crate::MemoryState. - Memory
Resource - Snapshot of one physical memory resource.
- Memory
State - Live resource discovery and process-wide
laddureservation state. - Prepared
Dataset Stats - Fixed statistics collected when a dataset is prepared for repeated evaluation.
- Prepared
Normalization - Prepared sufficient statistics and their parameter-only contraction.
- Prepared
Normalization Diagnostics - Runtime diagnostics for one compiler-native accepted normalization.
- Prepared
Query - Scalar expressions compiled and prepared once for repeated batch evaluation. This object retains no event data and owns no cross-request compilation cache.
- Process
Memory Report - Sampled operating-system memory counters for the current process.
- Reduction
Evaluation - The scalar value and free-parameter gradient produced by a reduction.
- Value
Gradient - A complex model value and its derivatives with respect to free parameters.
Enums§
- Capacity
Source - The basis used to obtain a resource’s capacity information.
- Comparison
- Comparison operation used by a dataset predicate.
- CpuExecution
Mode - CPU scalar-kernel execution strategy.
- CpuPrepared
Dataset - A dataset prepared according to its
CacheStoragepolicy. - Device
- Device on which a model should execute.
- Execution
Error - Error produced while resolving execution options.
- Final
Upper Edge - Policy for a value exactly equal to an axis’s final upper edge.
- GpuBackend
- GPU implementation to use.
- GpuDevice
Selector - Rule used to select a GPU device.
- Interval
Closure - Determines which endpoints are included by an interval predicate.
- JitPolicy
- Policy controlling CPU just-in-time compilation.
- Memory
Budget - A requested memory limit.
- Memory
Error - Errors produced while discovering or reserving memory.
- Memory
Resource Kind - Kind of physical memory resource.
- Normalization
Mode - Policy controlling compiler-native accepted normalization.
- Precision
- Numeric precision used to execute a model.
- Predicate
- Boolean expression used to select events from a dataset.
- Prepared
Dataset - A dataset prepared for a concrete execution backend.
- Prepared
Model - A compiled model prepared for a concrete execution backend.
- Runtime
Error - Error produced while preparing or evaluating a model.
- Thread
Policy - Policy controlling CPU worker-thread use.
Traits§
- Dataset
Expr Ext - Expression-based query operations for datasets.
- Event
Lookup - Supplies event-dependent scalar values for direct CPU evaluation.
Functions§
- bin_
shape - Return the row-major shape for ordered axes.
- checked_
bin_ count - Return the flattened bin count, or
Nonewhen the ordered shape overflowsusize. - flat_
bin_ index - Assign ordered coordinates to a row-major flattened bin.
- flat_
bin_ index_ for_ event - Assign one event from axis-major coordinate columns to a flattened bin.
Type Aliases§
- Runtime
Result - Result type returned by runtime operations.