14 February 2020
Futhark hedgehog logo

Our extended abstract Efficient Translation of Certain Irregular Data-Parallel Array Comprehensions, with Ken Friis Larsen, was accepted for presentation at TFP 2020 in Krakow, held alongside Lambda Days. The talk was scheduled for February 14.

Array comprehensions offer a compact way to describe queries and calculations, but irregular nesting makes efficient GPU execution difficult. We introduce a comprehension language supporting nesting, zipping, filtering, and sorting, and show how to translate it into Futhark using segmented operations. Examples demonstrate how these familiar abstractions can express useful computations while still producing efficient parallel code.

Read the paper (PDF).

Conference programme and abstract.