Skip to content

openPMD plugin: optimized restart from partial domain #5582

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
PrometheusPi merged 2 commits into from
Jan 6, 2026
Merged

openPMD plugin: optimized restart from partial domain #5582

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3690da4
Dont use remap buffer
franzpoeschel Dec 17, 2025
9ede729
Move comments to their own lines
franzpoeschel Jan 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 13 additions & 27 deletions include/picongpu/plugins/openPMD/restart/LoadSpecies.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,12 +51,13 @@ namespace picongpu
template<typename T_Identifier>
struct RedistributeFilteredParticles
{
template<typename FrameType, typename FilterType, typename RemapType>
template<typename FrameType, typename FilterType>
HINLINE void operator()(
FrameType& frame,
FilterType const& filter,
RemapType const& remap,
uint64_t const numParticlesCurrentBatch,
// becomes numParticlesAfterFiltering after this function call
MemIdxType& particleIndexTarget,
char const filterRemove)
{
using Identifier = T_Identifier;
Expand All @@ -65,13 +66,14 @@ namespace picongpu

ValueType* dataPtr = frame.getIdentifier(Identifier()).getPointer();

for(size_t particleIndex = 0; particleIndex < numParticlesCurrentBatch; ++particleIndex)
particleIndexTarget = 0;
for(MemIdxType particleIndexSrc = 0; particleIndexSrc < numParticlesCurrentBatch; ++particleIndexSrc)
{
if(filter[particleIndex] == filterRemove)
if(filter[particleIndexSrc] == filterRemove)
{
continue;
}
dataPtr[remap[particleIndex]] = dataPtr[particleIndex];
Copy link
Member

@PrometheusPi PrometheusPi Jan 5, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This looks like magic to me and a find the new version much clearer.

Copy link
Contributor Author

@franzpoeschel franzpoeschel Jan 5, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The major change is that the previous version (unnecessarily) precomputed the target indices. Could have been made a bit clearer by:

auto previouslyComputedNewIndex = remap[particleIndex];
dataPtr[previouslyComputedNewIndex] = dataPtr[particleIndex];

dataPtr[particleIndexTarget++] = dataPtr[particleIndexSrc];
}
}
};
Expand Down Expand Up @@ -291,17 +293,13 @@ namespace picongpu
manager::Device<HostDevice>::get().current(),
manager::Device<ComputeDevice>::get().getPlatform(),
MemSpace<DIM1>(maxChunkSize).toAlpakaMemVec());
auto remap = alpaka::allocMappedBuf<MemIdxType, MemIdxType>(
manager::Device<HostDevice>::get().current(),
manager::Device<ComputeDevice>::get().getPlatform(),
MemSpace<DIM1>(maxChunkSize).toAlpakaMemVec());

loadMatchesGeneric(
partialMatches,
threadParams,
totalNumParticles,
maxChunkSize, /* forEachPatch = */
[this, threadParams, &filter, &remap, &patchTotalOffset, &patchExtent, &patchUpperCorner](
[this, threadParams, &filter, &patchTotalOffset, &patchExtent, &patchUpperCorner](
uint64_t loadRound,
uint64_t numParticlesCurrentBatch,
FrameType& mappedFrame,
Expand Down Expand Up @@ -330,39 +328,27 @@ namespace picongpu
alpaka::getPtrNative(filter));
eventSystem::getTransactionEvent().waitForFinished();

// For simplicity, do the remapping on the CPU again
MemIdxType remapCurrent = 0;
for(size_t particleIndex = 0; particleIndex < numParticlesCurrentBatch; ++particleIndex)
{
if(filter[particleIndex] == filterKeep)
{
remap[particleIndex] = remapCurrent++;
}
else
{
remap[particleIndex] = std::numeric_limits<MemIdxType>::max();
}
}

MemIdxType numParticlesAfterFiltering = 0;
meta::ForEach<
typename NewParticleDescription::ValueTypeSeq,
RedistributeFilteredParticles<boost::mpl::_1>>
redistributeFilteredParticles;
redistributeFilteredParticles(
mappedFrame,
filter,
remap,
numParticlesCurrentBatch,
numParticlesAfterFiltering,
filterRemove);

log<picLog::INPUT_OUTPUT>(
"openPMD: Keeping %1% of the current batch's %2% particles after filtering.")
% remapCurrent % numParticlesCurrentBatch;
% numParticlesAfterFiltering % numParticlesCurrentBatch;

pmacc::particles::operations::splitIntoListOfFrames(
*speciesTmp,
mappedFrame,
remapCurrent, // !! not numParticlesCurrentBatch, filtered vs. unfiltered number
// @attention not numParticlesCurrentBatch, filtered vs. unfiltered number
numParticlesAfterFiltering,
cellOffsetToTotalDomain,
totalCellIdx_,
*threadParams->cellDescription,
Expand Down