[ITS] Remove custom boilerplate in favour of MCCompLabel - #2235
Conversation
f160ae2 to
74fbf4e
Compare
| for (int iNextLayerClusterIndex{ firstRowClusterIndex }; iNextLayerClusterIndex < maxRowClusterIndex && iNextLayerClusterIndex < (int)clustersNextLayer.size(); ++iNextLayerClusterIndex) { | ||
| const Cluster& nextCluster{ clustersNextLayer[iNextLayerClusterIndex] }; | ||
| const char testMC{ !isMc || (nextLayerMClabels[iNextLayerClusterIndex] == currentLayerMClabels[iCurrentLayerClusterIndex] && nextLayerMClabels[iNextLayerClusterIndex] != -1) }; | ||
| const auto& lblNext = evt->getClusterLabels(layerIndex, nextCluster.clusterId); |
There was a problem hiding this comment.
Without checking in detail, I would where possible try to keep the reconstruction code and the MC code separate. Don't know if this is possible here, but e.g. for the TPC I first create the tracks, then if MC is present I loop over the tracks and assign the labels. In addition I have some debug code inside the tracking that might access the labels, but that is usually disabled at compile time.
There was a problem hiding this comment.
Ok, I see what you mean.
For the moment the upstream code is giving wrong results as a consequence of the changes to labels.
I'd rather merge this in order to make code behave correctly.
Later I'll factorise the MC part out from the reco code, maybe changing it in a way it will work also with the GPU version.
Cheers
| for (auto& trklet01 : mComb01) { | ||
| for (auto& trklet12 : mComb12) { | ||
| if (trklet01.secondClusterIndex == trklet12.firstClusterIndex) { | ||
| const float deltaTanLambda{ gpu::GPUCommonMath::Abs(trklet01.tanLambda - trklet12.tanLambda) }; |
There was a problem hiding this comment.
@mconcas : this gives me a build error with -Werror (apparently the CI didn't catch it, probably the cmake migration broke this, investigating). Anyway, can you move the definition inside the #if defined protection?
* Use MCompLabels in place of custom boilerplate * Fix unspotted warning
…p#2235) * Use MCompLabels in place of custom boilerplate * Fix unspotted warning
…p#2235) * Use MCompLabels in place of custom boilerplate * Fix unspotted warning
* propagated tracks * propagated tracks used in all histos * typo * clang * add more control histos * after review edit and new histogram added
Hi @shahor02, Anne,
this will remove my custom labels in favour of the
MCComplLabelmainstream ones.This will improve the consistency of code.
GPU code for the moment is still using my custom ones.
I'll sort it out in the upcoming reiteration on the GPU code I'm going to perform next.
@a-kieffer for the record, as soon as we get new simulated data the check on noise will be consistent.
Cheers
Matteo