The Exelixis Lab

Enabling Research in Evolutionary Biology

Microbenchmark for Denormalized Floating Point Numbers

Denormalized floating point values can have a dramatic impact on program performance, i.e., operations of O(n) theoretical run-time can have significantly different execution times, depending on the input data.

The microbenchmark has been extracted from RAxML that exhibited some unexplicable run time variations due to this problem.

The benchmark is available under GNU GPL from Alexi's github repository.

Please read:

Björndalen J., Anshus O.: "Trusting floating point benchmarks-are your benchmarks really data-independent?" Applied Parallel Computing. State of the art in Scientific Computing 2010; pp 178-188, Springer.

for some background info.