ECP 5: Deploy production sliding mesh capability with linear solver benchmarking #59

spdomin · 2016-09-26T21:29:25Z

Activities:

Improve baseline sliding mesh capability at curved surfaces.
Evaluate ATDM-based parallel search methods.
Establish matrix set-up cost timings.
Evaluate possible lagging of matrix update.
Evaluate reduction of matrix system by omitting moving block column entries in favor of multiple matrix assembly/solve iterations.

This refactoring supports our design experimentation related to element-algorithms. * Adds two classes that mimic element-algorithms: TestElemAlgorithmWithVectors, TestElemAlgorithmWithViews * Illustrates the scratch-array resizing strategy which finds max nodes-per-elem and num-integ-points prior to the bucket loop and performs max resize, making subsequent resize operations fast. * Illustrates Kokkos views with shared-memory-views as prototyped by Christian in a separate repo. * Encapsulates looping mechanisms as described in recent meeting. * Some 'kokkos-helper' stuff (typedefs, small functions etc) are at the top of the file, essentially similar to the stuff that Christian had in the KokkosInterface.h header in his repo. * The test 'main' functions are at the bottom of the file, this is where you should start reading if you are interested in exploring.

spdomin added the enhancement label Sep 26, 2016

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ECP 5: Deploy production sliding mesh capability with linear solver benchmarking #59

ECP 5: Deploy production sliding mesh capability with linear solver benchmarking #59

spdomin commented Sep 26, 2016

ECP 5: Deploy production sliding mesh capability with linear solver benchmarking #59

ECP 5: Deploy production sliding mesh capability with linear solver benchmarking #59

Comments

spdomin commented Sep 26, 2016