Independent C++ engineer. High-performance parallel code, written close to the metal. Optimisation and scaling problems, mostly.
I write high-performance C++: parallel code, memory and cache behaviour, MPI across nodes, SIMD where it pays. Close to the metal, because that is where the difference between a good and a bad implementation stops being a few percent and becomes an order of magnitude.
That covers a lot of ground. Numerical and simulation code, scheduling, anything where an optimisation has to finish inside a budget, and anything that refuses to parallelise cleanly. The common thread is that the obvious implementation is too slow and buying more hardware has stopped helping.
Combinatorial search is one of those areas, and the one I also do research in. The papers are below.
/Preprint, Zenodo
with the GridSAT Stiftung
10.5281/zenodo.22062716
Karim Daghbouche, Deniz Duman
Deterministic models treat two configurations as different states whenever their encodings differ. This paper makes that choice an explicit parameter instead of a tacit one.
/Preprint, Zenodo
with the GridSAT Stiftung
10.5281/zenodo.22062774
Karim Daghbouche, Deniz Duman
The state-counting framework carried onto a concrete formula family: bounded-width layered 3-CNF from Tseytin encodings of bounded-fanin layered circuits.
/Preprint, Zenodo
with the GridSAT Stiftung
10.5281/zenodo.22063161
Karim Daghbouche, Deniz Duman
An audit schema for exact dynamic-programming arguments, worked through on an Ising instance with a fixed defect budget and on a fixed-cross-section polymer tube.
