search

egraphs-good / extraction-gym

benchmarking e-graph extraction
MIT License
24 stars 15 forks source link

Initial Rover testcases - stressing common sub-expression aware extraction #13

Closed cowardsa closed 9 months ago

cowardsa commented 9 months ago

We include multiple e-graphs for each test case based on e-graphs grown until increasing iteration limits - good extraction algorithms will ensure monotonicity - namely as the e-graph grows the cost should reduce - this is not true for naive ILP implementations.

Three testcases added:

  1. MCM 3,7,21 - from O. Gustafsson, “A difference based adder graph heuristic for multiple constant multiplication problems,”
  2. FIR Filter - from C. Lee, M. Potkonjak, and W. H. Mangione-Smith, “MediaBench: A tool for evaluating and synthesizing multimedia and communications systems,”
  3. Box Filter - randomly generated testcase for constant factorization
mwillsey commented 9 months ago

Thanks Sam! This will be super valuable.