Autotuner

The genetic-algorithm-based autotuner tries to optimize a TC by tuning the available mapping options.

Each autotuning session starts with a set (population) of candidate options which can be initialized randomly and/or from known starting points. Each candidate is benchmarked and the best ones have a higher chance of surviving and breeding to produce the next generation of candidates. This procedure is repeated for a pre-defined number of generations. In the end, the best candidate is returned.

At the end of each generation new candidates must be selected. Each candidate is either a combination of parent candidates (crossover) or one that survives from the previous generation. Both types are potentially randomly changed (mutation). The top candidates (elites) survive intact (without mutations) between generations.

Parameters for Autotuning

The parameters that control the autotuner’s behavior are the following:

  • Number of generations: The number of tuning generation to be run.
  • Population size: The number of candidates in each generation.
  • Number of elites: The number of best candidates that are preserved intact between generations (without any mutations).
  • Crossover rate: The rate at which new candidates are bred instead of just surviving across generations.
  • Mutation rate: The rate at which candidate options are randomly changed (mutated).
  • Number of threads: The number of threads that are used to compile different candidates in parallel.
  • GPUs: A comma separated list of GPUs (ids) to use for evaluating candidates (e.g., “0,1,2,3”).
  • RNG state: The state used to seed the tuner’s RNG.
  • min_launch_total_threads: Prune out kernels mapped to fewer than this many threads and block. Set this to 1 to avoid pruning.

Caching

After each autotuning session the best candidates’ profiling information and compilation results are stored in a cache. They can be subsequently retrieved to seed a new autotuning session.