Open mtaillefumier opened 7 months ago
we had long discussion on the 29th about the reduction algorithm that extended to the two versions of determinism, the strong vs weak implementation.
refs : https://groups.csail.mit.edu/commit/papers/09/asplos073-olszewski.pdf https://onlinelibrary.wiley.com/doi/full/10.1002/cpe.7763 https://www.physicsforums.com/threads/parallel-kahan-summation.716446/ (Kahan sum in parallel) https://indico.fnal.gov/event/57249/contributions/270632/attachments/169825/228082/reproduciblity_lattice2023.pdf (kate talk) https://developer.download.nvidia.com/assets/cuda/files/reduction.pdf (Mark harris talk about reduction, perf oriented)
we must distinguish between reproducibility and error minimization.
we had long discussion on the 29th about the reduction algorithm that extended to the two versions of determinism, the strong vs weak implementation.
refs : https://groups.csail.mit.edu/commit/papers/09/asplos073-olszewski.pdf https://onlinelibrary.wiley.com/doi/full/10.1002/cpe.7763 https://www.physicsforums.com/threads/parallel-kahan-summation.716446/ (Kahan sum in parallel) https://indico.fnal.gov/event/57249/contributions/270632/attachments/169825/228082/reproduciblity_lattice2023.pdf (kate talk) https://developer.download.nvidia.com/assets/cuda/files/reduction.pdf (Mark harris talk about reduction, perf oriented)
we must distinguish between reproducibility and error minimization.