Refactor `PerfConfig`
arthurpaulino opened this issue · 1 comments
arthurpaulino commented
Our PerfConfig
can better represent the parallelizations that we have:
- Parallel witness generation for the slots
- Parallel witness generation for the frames within a
MultiFrame
- Folding in parallel with witness generation
Therefore, our PerfConfig
would better capture what actually happens with more meaningful attributes. For example:
struct PerfConfig {
slots: Flow,
frames: Flow,
witgen_vs_folding: Flow,
}
Then, our canned configs would be:
FullySequential
ParallelSlots
ParallelFrames
ParallelWitGenVsFolding
FullyParallel
samuelburnham commented
Sounds great, please ping me for review on the eventual PR so I can check it integrates with the user config and env vars