darrnshn
commented
8 years ago Benchmarking can be done two ways: one would be to write benchmarking code in C++ or we can treat the whole thing as a blackbox (launch binaries using python and time). The former is good for microbenchmarks and profiling, the latter is better for benchmarking the whole system. What would be good here?
This will be helpful in the long run. Currently, Google Benchmark (https://github.com/google/benchmark) looks like a good candidate for performing the benchmarking. We just need a way to record benchmarking results and spot regressions.