Add PCR Thomas Laplacian inversion solver #2340
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Currently just a copy of pcr
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
| } | ||
|
|
||
| if (dst) { | ||
| BOUT_OMP(parallel) { |
There was a problem hiding this comment.
warning: if with identical then and else branches [bugprone-branch-clone]
if (dst) {
^
src/invert/laplace/impls/pcr_thomas/pcr_thomas.cxx:230:5: note: else branch starts here
} else {
^
dschwoerer
reviewed
Jun 3, 2021
Comment on lines
192
to
194
| BoutReal zlen = coords->dz * (localmesh->LocalNz - 3); | ||
| BoutReal kwave = | ||
| kz * 2.0 * PI / (2. * zlen); // wave number is 1/[rad]; DST has extra 2. |
There was a problem hiding this comment.
Similar to what I asked for the pcr - why is there a 3 but the comment says 2?
Besides that, would that also work if dz is a Field2D?
I assume variation of dz along z wouldn't work due to FFTs require being uniform ...
|
This would work best when transformed into a node-aware solver. The best way to do that might be running one MPI rank per node, and then filling up the node with OpenMP threads. So, we might want to wait for the outer-loop stuff in order to get the best performance. |
* next: (1519 commits) Remove some useless forward declarations Revert "Remove some unused headers from vecops" Remove some unused headers from vecops Fix bad merge: missed removing extraneous `const` on return value Remove `Up/DownXSplitIndex` Remove last remaining uses of `deprecated.hxx` Apply clang-format Apply clang-format changes CI: Don't update submodules (already updated) CI: Workaround for git ownership checks in container CI: Bump actions/checkout version CI: Use latest version of cmake in clang-tidy-review Use MPI_COMM_NULL rather than nullptr Add 'u' format flag to BOUT.settings output Adds a 'u' format, commenting unused options Apply clang-format changes Apply clang-format changes Don't compile LaplaceXY2 with 3D metrics Apply clang-format changes Apply clang-format changes ...
- implicit narrowing conversion - C-style array - loop variable name too short - multiple declarations on same line - dead stores
ZedThree
approved these changes
Jul 28, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This adds another direct parallel solver,
pcr_thomas, based on Laszlo, Gilles and Appleyard, Manycore Algorithms for Batch Scalar and Block Tridiagonal Solvers, ACM TOMS, 42, 31 (2016) and adapted from the parallel_tdma_cpp library.It is very similar to the parallel cyclic reduction solver (#2330), except instead of performing PCR directly on the distributed tridiagonal matrix, it uses the Thomas algorithm on each core to get a reduced tridiagonal system with 2 equations per processor. It then calls PCR on this reduced system. This reduces the number of communications from
O(log2(nx))toO(log2(nxpe)), without changing theO(nx)complexity.This solver passes the Laplace integrated tests, but I haven't done any performance analysis on it. This is really intended as a stepping stone to getting a node-aware solver, which would further reduce the communication count to
O(log2(nnodes))inter-node communications. To do this, we need to replace the Thomas solver with a parallel direct solver and work out how to extract the coefficients to feed into a node-level PCR. This PR gives the code structure to do this, but since it also adds a complete solver, I thought I'd add it as an option.