Remove GPU-CA dependence on CPU doublets #126
Conversation
|
Successfully tested at #127 (comment) . |
| produces<RegionsSeedingHitSets>(); | ||
| } | ||
|
|
||
| void CAHitNtupletHeterogeneousEDProducer::fillDescriptions( | ||
| edm::ConfigurationDescriptions &descriptions) { | ||
| edm::ParameterSetDescription desc; | ||
|
|
||
| desc.add<edm::InputTag>("doublets", edm::InputTag("hitPairEDProducer")); | ||
| desc.add<edm::InputTag>("doublets", edm::InputTag("dummy"))->setComment("Not really used, kept to keep the python parameters"); |
There was a problem hiding this comment.
Can we set it to be empty by default, instead of using a dummy value ?
One of the checks we do at HLT is to avoid InputTag referencing non-existing modules, and these "dummy" labels end up giving false positives.
|
Everything is indeed hardcoded (even region params) as Quadruplets are Quadruplets. |
|
A new Pull Request was created by @makortel (Matti Kortelainen) for CMSSW_10_2_X_Patatrack. It involves the following packages: RecoPixelVertexing/PixelTrackFitting @cmsbot, @fwyzard can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
For simplicity I also removed the CPU-CA from the
CAHitNtupletHeterogeneousEDProducer(in fact it must be removed, otherwise it would be impossible to remove the dependence with the current framework prefetching logic).@felicepantaleo I was a bit surprised that the "seeding layer" information was not used in the end (implying that the "navigation logic" must be hardcoded at the moment).