Go to the documentation of this file.
15 namespace Autoscheduler {
73 double *cost_ptr) = 0;
79 virtual void reset() = 0;
84 #endif // COST_MODEL_H
int parallelism
Maximum level of parallelism available.
int beam_size
Beam size to use in the beam search.
int64_t memory_limit
If >= 0, only consider schedules that allocate at most this much memory (measured in bytes).
int random_dropout
percent chance of accepting each state in the beam.
virtual void evaluate_costs()=0
int disable_subtiling
If set to nonzero value: limits the search space to that of Mullapudi et al.
This file defines the class FunctionDAG, which is our representation of a Halide pipeline,...
@ Internal
Not visible externally, similar to 'static' linkage in C.
int random_dropout_seed
Random seed used by the random dropout.
PerfectHashMap< FunctionDAG::Node::Stage, ScheduleFeatures > StageMapOfScheduleFeatures
signed __INT64_TYPE__ int64_t
virtual void enqueue(const Internal::Autoscheduler::FunctionDAG &dag, const Halide::Internal::Autoscheduler::StageMapOfScheduleFeatures &schedule_feats, double *cost_ptr)=0
virtual ~CostModel()=default
virtual void set_pipeline_features(const Internal::Autoscheduler::FunctionDAG &dag, const Internal::Autoscheduler::Adams2019Params ¶ms)=0
std::string weights_path
When training or schedule, read weights from this directory or file.
int disable_memoized_features
If set to nonzero value: features of possible schedules are always recalculated, and are not cached a...
int disable_memoized_blocks
If set to nonzero value: tiling sizes are not cached across passes.