Matti Kortelainen

Results 1229 comments of Matti Kortelainen

> Unless I'm missing something, the pile-up input at RECO step is useless in any case. If the RECO (or any non-DIGI) step includes VALIDATION, the pileup mixing is played...

Both `BPHMonitor` and `pat::PATTriggerProducer::beginRun()` are stream modules, and have the `HLTPrescaleProvider` as data members https://github.com/cms-sw/cmssw/blob/8340b52ad5b3272e98a300259ff0c67959520800/DQMOffline/Trigger/plugins/BPHMonitor.cc#L142 https://github.com/cms-sw/cmssw/blob/8340b52ad5b3272e98a300259ff0c67959520800/PhysicsTools/PatAlgos/plugins/PATTriggerProducer.cc#L118 that then lead to the memory usage to increase with more streams. Is `HLTPrescaleProvider`...

In addition, the `l1t::L1TGlobalUtil::retrieveL1Setup()` does 50 million memory allocations within its body, and nearly all of these come via the call from `HLTPrescaleProvider::init()`.

> I tried to address this at [#48026](https://github.com/cms-sw/cmssw/pull/48026). Thanks! > (and also if there is a way to detect the allocations directly from the PR tests). The "max memory used"...