diff options
| author |  Andrew Myers <atmyers@lbl.gov>
| 2021-10-18 14:06:08 -0700 |
|---|---|---|
| committer |  GitHub <noreply@github.com>
| 2021-10-18 14:06:08 -0700 |
| commit | 2bc1322aa76f7026e5b42639bb3d12125da2407c (patch) | |
| tree | 7b8dbd4285fea5d47bf2b2b18a24af767342f34a /Source/Diagnostics/ComputeDiagFunctors | |
| parent | 3e9903c07561c24bb8309a7bfa8feeb76ba2eebd (diff) | |
| download | WarpX-2bc1322aa76f7026e5b42639bb3d12125da2407c.tar.gz WarpX-2bc1322aa76f7026e5b42639bb3d12125da2407c.tar.zst WarpX-2bc1322aa76f7026e5b42639bb3d12125da2407c.zip | |
Option to do single precision mesh communication. (#2294)
* option to use single precision guard cell exhanges
* add missing files
* fix namespace issue
* change precision of comms to float
* ParmParse the single_precision_comms flag
* set back to real
* test
* make sure dst is filled
* nGrow -> nGrowVect
* restore float
* don't override valid cells
* single precision mesh
* whitespace
* wrap SumBoundary
* Wrap additional uses of FillBoundary.
* catch missing copies of ParallelCopy
* missing OverrideSyncs
* add wrapper for iMultifab
* fix typo
* moar typos
* typo
* strip single_precision_mesh option
* fix original copy
* update fusible syntax
Co-authored-by: Weiqun Zhang <WeiqunZhang@lbl.gov>
* Fix: Single Precision Builds
Should not copy around data for `do_single_precision_comms`
* Docs: warpx.do_single_precision_comms
* initialize this tmp multifab to 0.0
* fix tiny profile label
* remove orig copies, they are only correct for FillBoundary
* loosen tolerance for space charge tests for single precision
* missing some setVal
* another missing setVal
* missing setVal
* add wrapper for new version of sumboundary
* add explicit cast to silence compiler warning
* add a test for single precision comms
* revert change to test precision
* add benchmark for single precision comms test
* restore tolerance I removed by mistake
* tolerance
* copyright headers
* drop tolerance for single precision tests in default analysis script
* missing python module
* bump tol again
* fix bad merge
* Apply suggestions from code review
Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>
Co-authored-by: Remi Lehe <remi.lehe@normalesup.org>
Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja>
Co-authored-by: Weiqun Zhang <WeiqunZhang@lbl.gov>
Diffstat (limited to 'Source/Diagnostics/ComputeDiagFunctors')
| -rw-r--r-- | Source/Diagnostics/ComputeDiagFunctors/BackTransformFunctor.cpp | 7 |
1 files changed, 5 insertions, 2 deletions
diff --git a/Source/Diagnostics/ComputeDiagFunctors/BackTransformFunctor.cpp b/Source/Diagnostics/ComputeDiagFunctors/BackTransformFunctor.cpp index 78e070bfd..4af43b4b8 100644 --- a/Source/Diagnostics/ComputeDiagFunctors/BackTransformFunctor.cpp +++ b/Source/Diagnostics/ComputeDiagFunctors/BackTransformFunctor.cpp @@ -1,6 +1,7 @@ #include "BackTransformFunctor.H" #include "Diagnostics/ComputeDiagFunctors/ComputeDiagFunctor.H" +#include "Parallelization/WarpXCommUtil.H" #include "Utils/WarpXConst.H" #include "WarpX.H" @@ -70,11 +71,13 @@ BackTransformFunctor::operator ()(amrex::MultiFab& mf_dst, int /*dcomp*/, const // containing all ten components that were in the slice generated from m_mf_src. std::unique_ptr< amrex::MultiFab > tmp_slice_ptr = nullptr; tmp_slice_ptr = std::make_unique<MultiFab> ( slice_ba, mf_dst.DistributionMap(), - slice->nComp(), 0 ); + slice->nComp(), 0 ); + tmp_slice_ptr->setVal(0.0); // Parallel copy the lab-frame data from "slice" MultiFab with // ncomp=10 and boosted-frame dmap to "tmp_slice_ptr" MultiFab with // ncomp=10 and dmap of the destination Multifab, which will store the final data - tmp_slice_ptr->ParallelCopy( *slice, 0, 0, slice->nComp() ); + WarpXCommUtil::ParallelCopy(*tmp_slice_ptr, *slice, 0, 0, slice->nComp(), + IntVect(AMREX_D_DECL(0, 0, 0)), IntVect(AMREX_D_DECL(0, 0, 0))); // Now we will cherry pick only the user-defined fields from // tmp_slice_ptr to dst_mf const int k_lab = m_k_index_zlab[i_buffer]; |