Skip to content

PADRE-5: Add GPU implementation of DDECal IterativeDiagonalSolver

Bram Veenboer requested to merge padre-5-ddecal into master

The AARTFAAC pipeline uses the IterativeDiagonalSolver of the DDECal step. In this MR, an (initial) GPU implementation is added. This implementation achieves better performance than the CPU implementation.

A new unit test (based on the test for the equivalent CPU implementation) is added. It can be executed (in the build directory) as follows:

./unittests -t solvers/iterative_diagonal_cuda

And passes.

This MR may eventually be extended to include the other DDECal kernels as well.

Edited by Bram Veenboer

Merge request reports