This project benchmarks the following libraries:

• hmatrix
• dense-linear-algebra (refered to as DLA henceforth)
• numhask
• massiv ("Massiv (Par)" refers to parallel computation
• matrix

To run:

(allocation) stack build :bench-alloc && stack exec bench-alloc

(runtime) stack build :bench-runtime && stack exec bench-runtime

Relevant details:

• the implementations of the "naive C" parts can be found in /naive. They were compiled with -O3
• the massiv benchmarks use the Primitive representation, which seems to be the fastest among what massiv offers
• the benchmarked functions from DLA are taken from the Fast module when available
• the norm function is called on n*n vectors
• instead of relying on hackage, the project's dependencies fetch the libraries directly from github (see stack.yaml).

Formely included:

• bed-and-breakfast (abandoned because too slow)
• matrices (abandoned because too slow, also see kaizhang/matrices#8)

TODO:

• have a cleaner/more abstract interface for the benches