potential speedup in HLT: AXOL1TLCondition::evaluateCondition consider caching the model? #46740

slava77 · 2024-11-19T20:12:30Z

Looking at a profile of HLT in 14_1_X with callgrind, on MC running only MC_ReducedIterativeTracking_v22, I see that 78% of L1TGlobalProducer::produce is spent in l1t::AXOL1TLCondition::evaluateCondition
https://github.com/cms-sw/cmssw/blob/CMSSW_14_1_0_pre5/L1Trigger/L1TGlobal/src/AXOL1TLCondition.cc#L100

In my test L1TGlobalProducer::produce takes 9.7% of the time; in the full menu it's apparently around ~~2.5%~~ 0.9% (updated to 0.9%, see notes below)

Of all the time spent in l1t::AXOL1TLCondition::evaluateCondition

hls4mlEmulator::ModelLoader::load_model() is 54%
hls4mlEmulator::ModelLoader::~ModelLoader() is 30%
GTADModel_emulator_v4::predict() is 15%
the rest is around 1%

IIUC, the load and destruction of the model happens 10 times per event; I don't see any dependence on the current event variables.
Some kind of caching may be useful to get HLT to run a bit faster (seems like ~~1.5%~~ (updated) 0.6% or so).

The text was updated successfully, but these errors were encountered:

cmsbuild · 2024-11-19T20:12:52Z

cms-bot internal usage

cmsbuild · 2024-11-19T20:12:53Z

A new Issue was created by @slava77.

@Dr15Jones, @antoniovilela, @makortel, @mandrenguyen, @rappoccio, @sextonkennedy, @smuzaffar can you please review it and eventually sign/assign? Thanks.

cms-bot commands are listed here

mmusich · 2024-11-19T21:18:42Z

@aloeliger @artlbv FYI

aloeliger · 2024-11-19T21:20:19Z

@quinnanm

aloeliger · 2024-11-19T21:20:38Z

I can also take a look at this when I get some time, frankly I've been meaning to for a while.

mmusich · 2024-11-19T21:27:30Z

In my test L1TGlobalProducer::produce takes 9.7% of the time; in the full menu it's apparently around 2.5%

in a more "realistic" setup it's more like sub-percent

(details from #45631 (comment))

makortel · 2024-11-19T21:33:54Z

assign l1

makortel · 2024-11-19T21:34:14Z

type performance-improvements

cmsbuild · 2024-11-19T21:34:16Z

New categories assigned: l1

@aloeliger,@epalencia you have been requested to review this Pull request/Issue and eventually sign? Thanks

slava77 · 2024-11-19T21:49:41Z

in a more "realistic" setup it's more like sub-percent

My 2.5% comes from a more recent test (but on GRun menu) by @bdanzi
The odd part there is that e.g. in the first iteration in the first 6 jobs L1TGlobalProducer takes under 1%, but in the last two it's around 6%

mmusich · 2024-11-19T21:52:49Z

My 2.5% comes from a more recent test (but on GRun menu) by @bdanzi

I'd rather trust more the manual measurement than what comes from the timing server, tbh.

slava77 · 2024-11-20T23:04:29Z

essentially all of the cost is in dlopen and dlclose calls.
https://github.com/cms-hls4ml/hls4mlEmulatorExtras/blob/v1.1.3/src/hls4ml/emulator.cc#L12-L22
create_model has orders of magnitude lower cost

perhaps just having the .so file loaded in the constructor or begin job is enough

artlbv · 2024-11-21T08:51:55Z

IIUC, the load and destruction of the model happens 10 times per event; I don't see any dependence on the current event variables.

The "problem" is that the model is evaluated (and loaded) for every threshold and BX separately, even though it is the same model. If caching is possible that would help of course and in principle the model(s) are known at the beginning of the job as they are fixed in the L1 menu.

IMO it would be good to have some common approach to this loading of HLS4ML models within L1/CMSSW, as e.g. here it seems to be done differently:
https://github.com/cms-sw/cmssw/blob/CMSSW_14_2_X/L1Trigger/Phase2L1ParticleFlow/interface/JetId.h#L50-L53

makortel · 2024-11-21T17:41:32Z

The "problem" is that the model is evaluated (and loaded) for every threshold and BX separately, even though it is the same model. If caching is possible that would help of course and in principle the model(s) are known at the beginning of the job as they are fixed in the L1 menu.

Given that the interface of hls4mlEmulator::Model is not const, the best that one can do is to call ModelLoader::load_model() once per module per stream (in constructor of edm::stream module, or using StreamCache with edm::global module, or in constructor of edm::one module).

This is how

cmssw/L1Trigger/Phase2L1ParticleFlow/interface/JetId.h

Lines 50 to 53 in 92333e3

    
             unique_ptr<float[]> fDY_; 
        
             tensorflow::Session *sessionRef_; 
        
             std::shared_ptr<hls4mlEmulator::Model> modelRef_; 
        
           };

is used in

cmssw/L1Trigger/Phase2L1ParticleFlow/plugins/L1BJetProducer.cc

Lines 51 to 52 in 92333e3

    
           fBJetId_ = std::make_unique<JetId>( 
        
               cfg.getParameter<std::string>("NNInput"), cfg.getParameter<std::string>("NNOutput"), cache, fNParticles_);

(now if hls4mlEmulator::Model interface would be const and thread-efficient, we could avoid the per-stream copies of the hls4mlEmulator::Model)

mmusich · 2024-11-22T13:27:12Z

assign hlt

to keep it on the radar

cmsbuild · 2024-11-22T13:27:35Z

New categories assigned: hlt

@Martin-Grunewald,@mmusich you have been requested to review this Pull request/Issue and eventually sign? Thanks

jpearkes · 2024-12-10T11:16:48Z

Hi @slava77,

For these tests is the --timing flag being used in hltGetConfiguration? From tests we were running last year, we saw a big slowdown without using the --timing flag, e.g.
hltGetConfiguration /dev/CMSSW_13_3_0/GRun --data --output none --globaltag 132X_dataRun3_HLT_for2024TSGStudies_v1 --process jannicke_test1 --timing > hlt.py

I am attaching screenshots from a presentation here . The "Dec" results are when the --timing flag was not being used, the "New" results are when the --timing was being used. The difference in throughput was on the order of 50%.

mmusich · 2024-12-10T11:21:52Z

@jpearkes

For these tests is the --timing flag being used in hltGetConfiguration?

As far as I understand these measurement were derived using the timing server, so the flag should have been used by default.
Irrespective of how the timing was run, there is an opportunity to make the code more efficient, see suggestion at #46740 (comment). Is it something that is going to picked up by the AXO team?

jpearkes · 2024-12-10T11:37:57Z

@mmusich We were using the timing server for the tests too, and at least around this time last year it was not enabled by default.

Completely agreed that it is an opportunity to make the code more efficient. As after running the --timing flag the timing was no longer an issue the AXO deprioritized it. We could bump it up in priority if needed though. @quinnanm and @aloeliger are our main emulator contacts from both AXO and CICADA.

missirol · 2025-01-05T18:06:39Z

I gave it a try in https://github.com/missirol/cmssw/commits/devel_cmssw46740_v1, which includes two commits on top of CMSSW_15_0_0_pre1.

missirol@9dc135c moves several operations from L1TGlobalProducer::produce to L1TGlobalProducer::beginRun (these are operations that depend only on the L1T trigger menu, which (afaik) may only change during run transitions). In particular, this leads to initialising the map of "trigger conditions" (instances of classes inhering from ConditionEvaluation, one of them being AXOL1TLCondition) only once per run, as opposed to once per event. (this kind of update is unrelated to this issue per se)
missirol@4be82dd moves the initialisation of the hls4ml model in AXOL1TLCondition from its evaluateCondition method (used multiple times per event) to its constructor (called only a few times per run if missirol@9dc135c is used). missirol@4be82dd requires missirol/hls4mlEmulatorExtras@69f7d1e (an update of the hls4mlEmulatorExtras external), because the current methods of ModelLoader were not enough for me to get things to work properly, and since (as far as I understand) the Model and ModelLoader must have the same "lifetime" in order for things to work, it made sense to package them in a single class (called ModelWrapper in missirol/hls4mlEmulatorExtras@69f7d1e).

I tested this update with the script in [1] (running the L1-uGT emulator on roughly 50k events of 2024 pp data), and I got the outputs in [2] from the FastTimerService: it is not an accurate timing measurement, but it seems that these updates bring some speedup.

Did others have something else in mind ? If you think it's useful, I can put this in a PR (incl. missirol/hls4mlEmulatorExtras@69f7d1e, assuming that's an acceptable way to update that external).

[1]

import FWCore.ParameterSet.Config as cms

import FWCore.ParameterSet.VarParsing as vpo
opts = vpo.VarParsing('standard')
opts.setDefault('maxEvents', 1000)
opts.parseArguments()

process = cms.Process('TEST')

process.options.numberOfThreads = 1
process.options.numberOfStreams = 0
process.options.wantSummary = True
process.maxEvents.input = opts.maxEvents

# Global Tag
from Configuration.AlCa.GlobalTag import GlobalTag
process.load('Configuration.StandardSequences.FrontierConditions_GlobalTag_cff')
process.GlobalTag = GlobalTag(process.GlobalTag, '141X_dataRun3_HLT_v2', '')

# Input source
process.source = cms.Source('PoolSource',
    fileNames = cms.untracked.vstring(
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/593/00000/00aa2e37-b442-4fa6-b087-73c54699784d.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/593/00000/018e0570-6f5f-48db-ab22-3c156607f248.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/593/00000/036cb164-49e7-490f-8a3b-c1f37f713deb.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/593/00000/074e8d3f-d60f-46a4-8fbe-e16e86d2b693.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/593/00000/078f24ee-6776-45b1-a437-5b58de809fb6.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/807/00000/136403f7-a7fe-40c2-8af6-280f91f82976.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/807/00000/247d49c3-5efe-4de1-8046-303e510b6a00.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/807/00000/357f3289-56d5-413c-958f-ecc5844bd4ed.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/807/00000/42371903-8e35-487a-a646-2ba330cf8cf3.root',
        'root://eoscms.cern.ch//eos/cms/store/data/Run2024I/EphemeralHLTPhysics7/RAW/v1/000/386/807/00000/46de66dd-f64c-4ee7-8553-1ebb2e08cc68.root',
    )
)

# EventSetup modules
process.GlobalParametersRcdSource = cms.ESSource('EmptyESSource',
    recordName = cms.string('L1TGlobalParametersRcd'),
    iovIsRunNotTime = cms.bool(True),
    firstValid = cms.vuint32(1)
)

process.GlobalParameters = cms.ESProducer('StableParametersTrivialProducer',
    # trigger decision
    NumberPhysTriggers = cms.uint32(512), # number of physics trigger algorithms
    # trigger objects
    NumberL1Muon = cms.uint32(8),    # muons
    NumberL1EGamma = cms.uint32(12), # e/gamma and isolated e/gamma objects
    NumberL1Jet = cms.uint32(12),    # jets
    NumberL1Tau = cms.uint32(12),    # taus
    # hardware
    NumberChips = cms.uint32(1),  # number of maximum chips defined in the xml file
    PinsOnChip = cms.uint32(512), # number of pins on the GTL condition chips
    # correspondence 'condition chip - GTL algorithm word' in the hardware
    # e.g.: chip 2: 0 - 95;  chip 1: 96 - 128 (191)
    OrderOfChip = cms.vint32(1),
)

# EventData modules
process.simGtExtFakeStage2Digis = cms.EDProducer('L1TExtCondProducer',
    bxFirst = cms.int32(-2),
    bxLast = cms.int32(2),
    setBptxAND = cms.bool(True),
    setBptxMinus = cms.bool(True),
    setBptxOR = cms.bool(True),
    setBptxPlus = cms.bool(True),
    tcdsRecordLabel = cms.InputTag('')
)

process.simGtStage2Digis = cms.EDProducer('L1TGlobalProducer',
    AlgoBlkInputTag = cms.InputTag(''),
    AlgorithmTriggersUnmasked = cms.bool(True),
    AlgorithmTriggersUnprescaled = cms.bool(True),
    EGammaInputTag = cms.InputTag('simCaloStage2Digis'),
    EtSumInputTag = cms.InputTag('simCaloStage2Digis'),
    ExtInputTag = cms.InputTag('simGtExtFakeStage2Digis'),
    JetInputTag = cms.InputTag('simCaloStage2Digis'),
    MuonInputTag = cms.InputTag('simGmtStage2Digis'),
    MuonShowerInputTag = cms.InputTag('simGmtShowerDigis'),
    TauInputTag = cms.InputTag('simCaloStage2Digis'),
    useMuonShowers = cms.bool(True),
    RequireMenuToMatchAlgoBlkInput = cms.bool(False),
)

# Sequence definition
process.l1tSequence = cms.Sequence( process.simGtExtFakeStage2Digis + process.simGtStage2Digis )

# Path definition
process.l1tPath = cms.Path( process.l1tSequence )

## Analyser of L1T-menu results
#process.l1tGlobalSummary = cms.EDAnalyzer( 'L1TGlobalSummary',
#    AlgInputTag = cms.InputTag( 'simGtStage2Digis' ),
#    ExtInputTag = cms.InputTag( 'simGtStage2Digis' ),
#    MinBx = cms.int32( 0 ),
#    MaxBx = cms.int32( 0 ),
#    DumpTrigResults = cms.bool( False ),
#    DumpRecord = cms.bool( False ),
#    DumpTrigSummary = cms.bool( True ),
#    ReadPrescalesFromFile = cms.bool( False ),
#    psFileName = cms.string( '' ),
#    psColumn = cms.int32( 0 )
#)
#
## EndPath definition
#process.l1tEndPath = cms.EndPath( process.l1tGlobalSummary )

# MessageLogger
process.load('FWCore.MessageService.MessageLogger_cfi')
process.MessageLogger.cerr.FwkReport.reportEvery = 1000
process.MessageLogger.L1TGlobalSummary = cms.untracked.PSet()
process.MessageLogger.FastReport = cms.untracked.PSet()

# FastTimerService
process.FastTimerService = cms.Service( "FastTimerService",
    printEventSummary = cms.untracked.bool( False ),
    printRunSummary = cms.untracked.bool( False ),
    printJobSummary = cms.untracked.bool( True ),
    writeJSONSummary = cms.untracked.bool( False ),
    jsonFileName = cms.untracked.string( "resources.json" ),
    enableDQM = cms.untracked.bool( False ),
    enableDQMbyModule = cms.untracked.bool( False ),
    enableDQMbyPath = cms.untracked.bool( False ),
    enableDQMbyLumiSection = cms.untracked.bool( False ),
    enableDQMbyProcesses = cms.untracked.bool( False ),
    enableDQMTransitions = cms.untracked.bool( False ),
    dqmTimeRange = cms.untracked.double( 2000.0 ),
    dqmTimeResolution = cms.untracked.double( 5.0 ),
    dqmMemoryRange = cms.untracked.double( 1000000.0 ),
    dqmMemoryResolution = cms.untracked.double( 5000.0 ),
    dqmPathTimeRange = cms.untracked.double( 100.0 ),
    dqmPathTimeResolution = cms.untracked.double( 0.5 ),
    dqmPathMemoryRange = cms.untracked.double( 1000000.0 ),
    dqmPathMemoryResolution = cms.untracked.double( 5000.0 ),
    dqmModuleTimeRange = cms.untracked.double( 40.0 ),
    dqmModuleTimeResolution = cms.untracked.double( 0.2 ),
    dqmModuleMemoryRange = cms.untracked.double( 100000.0 ),
    dqmModuleMemoryResolution = cms.untracked.double( 500.0 ),
    dqmLumiSectionsRange = cms.untracked.uint32( 2500 ),
    dqmPath = cms.untracked.string( "HLT/TimerService" ),
)

[2] Tested on lxplus8 (without full control over the machine, but at least I ran the test a few times and got consistent results). The L1-uGT emulator corresponds to the module simGtStage2Digis.

CMSSW_15_0_0_pre1

FastReport ---------------------------- Job Summary ----------------------------
FastReport   CPU time avg.      when run  Real time avg.      when run     Alloc. avg.      when run   Dealloc. avg.      when run  Modules
FastReport         0.0 ms         0.0 ms         0.1 ms         0.1 ms          +3 kB          +3 kB          -3 kB          -3 kB  source
FastReport         2.5 ms                        2.9 ms                       +457 kB                       -322 kB                 process TEST
FastReport         0.0 ms         0.0 ms         0.0 ms         0.0 ms          +1 kB          +1 kB          -1 kB          -1 kB    simGtExtFakeStage2Digis
FastReport         2.5 ms         2.5 ms         2.9 ms         2.9 ms        +455 kB        +455 kB        -321 kB        -321 kB    simGtStage2Digis
FastReport         0.0 ms         0.0 ms         0.0 ms         0.0 ms          +0 kB          +0 kB          +0 kB          +0 kB    TriggerResults
FastReport         0.0 ms         0.0 ms         0.0 ms         0.0 ms          +0 kB          +0 kB          +0 kB          +0 kB    l1tPath
FastReport         2.5 ms                        3.0 ms                       +461 kB                       -326 kB                 total
FastReport         0.0 ms                        0.0 ms                         +0 kB                         +0 kB                 eventsetup
FastReport         0.1 ms                        0.6 ms                        +41 kB                       -174 kB                 other
FastReport         0.0 ms                        0.0 ms                         +0 kB                         +0 kB                 idle

CMSSW_15_0_0_pre1 + (missirol@9dc135c + missirol@4be82dd) + missirol/hls4mlEmulatorExtras@69f7d1e

FastReport ---------------------------- Job Summary ----------------------------
FastReport   CPU time avg.      when run  Real time avg.      when run     Alloc. avg.      when run   Dealloc. avg.      when run  Modules
FastReport         0.0 ms         0.0 ms         0.1 ms         0.1 ms          +3 kB          +3 kB          -3 kB          -3 kB  source
FastReport         0.4 ms                        0.4 ms                       +409 kB                       -273 kB                 process TEST
FastReport         0.0 ms         0.0 ms         0.0 ms         0.0 ms          +1 kB          +1 kB          -1 kB          -1 kB    simGtExtFakeStage2Digis
FastReport         0.4 ms         0.4 ms         0.4 ms         0.4 ms        +407 kB        +407 kB        -272 kB        -272 kB    simGtStage2Digis
FastReport         0.0 ms         0.0 ms         0.0 ms         0.0 ms          +0 kB          +0 kB          +0 kB          +0 kB    TriggerResults
FastReport         0.0 ms         0.0 ms         0.0 ms         0.0 ms          +0 kB          +0 kB          +0 kB          +0 kB    l1tPath
FastReport         0.4 ms                        0.5 ms                       +413 kB                       -277 kB                 total
FastReport         0.0 ms                        0.0 ms                         +0 kB                         +0 kB                 eventsetup
FastReport         0.1 ms                        0.5 ms                        +41 kB                       -174 kB                 other
FastReport         0.0 ms                        0.0 ms                         +0 kB                         +0 kB                 idle

missirol · 2025-01-05T20:50:14Z

After reading the discussion in #40277 (in particular, the part on not allowing copy/move operators for hls4mlEmulator::ModelLoader), I force-pushed a couple of updates and edited my previous comment accordingly (#46740 (comment)).

mmusich · 2025-01-08T20:54:26Z

Did others have something else in mind ? If you think it's useful, I can put this in a PR (incl. missirol/hls4mlEmulatorExtras@69f7d1e, assuming that's an acceptable way to update that external).

@cms-sw/l1-l2 can you please comment? From the HLT side, I think a PR would be useful.

missirol · 2025-01-09T01:23:41Z

Latest (and hopefully, final) evolution of #46740 (comment).

missirol:devel_cmssw46740_v2 == CMSSW_15_0_0_pre1 + missirol@9dc135c + missirol@fe94271.
missirol/hls4mlEmulatorExtras@4abd8ad

The latest changes are mainly in the hls4mlEmulatorExtras external, going for a minimal update of hls4mlEmulatorExtras::ModelLoader, rather than introducing a new class.

If there are no objections, I'll open PRs to CMSSW and hls4mlEmulatorExtras by the end of the week. (the CMSSW PR will initially conflict with #47030, then one of the two will have to be rebased eventually)

aloeliger · 2025-01-09T02:20:13Z

Did others have something else in mind ? If you think it's useful, I can put this in a PR (incl. missirol/hls4mlEmulatorExtras@69f7d1e, assuming that's an acceptable way to update that external).

@cms-sw/l1-l2 can you please comment? From the HLT side, I think a PR would be useful.

Sorry, catching back up after the holidays here. I'm a little confused on the intended purpose of the ModelWrapper. It's fine, and I can approve that PR to the hls4ml extras myself on a moment's notice, but I would have thought the primary change needed here was caching the model itself to the uGT GlobalBoard or whichever top level GT producer contains most of the module members for use. Did I miss plans or PRs from the AXO team to do that? Is this just a convenience?

mmusich · 2025-01-09T06:40:12Z

but I would have thought the primary change needed here was caching the model itself to the uGT GlobalBoard or whichever top level GT producer contains most of the module members for use. Did I miss plans or PRs from the AXO team to do that? Is this just a convenience?

Isn't it what Marino proposed as missirol:devel_cmssw46740_v2 == CMSSW_15_0_0_pre1 + missirol@9dc135c + missirol@fe94271. (see #46740 (comment) ) ?

missirol · 2025-01-09T08:37:20Z

Did others have something else in mind ? If you think it's useful, I can put this in a PR (incl. missirol/hls4mlEmulatorExtras@69f7d1e, assuming that's an acceptable way to update that external).

@cms-sw/l1-l2 can you please comment? From the HLT side, I think a PR would be useful.

Sorry, catching back up after the holidays here. I'm a little confused on the intended purpose of the ModelWrapper. It's fine, and I can approve that PR to the hls4ml extras myself on a moment's notice, but I would have thought the primary change needed here was caching the model itself to the uGT GlobalBoard or whichever top level GT producer contains most of the module members for use. Did I miss plans or PRs from the AXO team to do that? Is this just a convenience?

In short, ModelWrapper is just something I used for convenience while testing (I don't fully understand why one needs to carry around both a Model and ModelLoader in client code, having a single object would be more intuitive to me). In my latest comment (#46740 (comment)), I removed the ModelWrapper and simply updated minimally ModelLoader: that will be enough for this issue.

As for this issue itself, the point is simply that (a) the construction of the GT conditions is moved from produce to beginRun, and (b) the anomaly-detection-model loading is moved from the evaluateCondition to the constructor of the GT condition, so the model loading is done only once per condition as opposed to once per evaluation (I guess, in other words, the model is cached in the GT condition). I don't see how one could do it less often, since a job can in principle span multiple L1T menus (across different runs), and each menu can contain GT conditions using different models. That's as far as I got: if L1T experts can provide a better update before 15_0_X closes, I'm of course in favor of that (over my suggestion). In the meantime, I'll prepare the PRs.

missirol · 2025-01-09T18:09:26Z

#46740 (comment) is now implemented in #47070 + cms-hls4ml/hls4mlEmulatorExtras#6.

cmsbuild added the pending-assignment label Nov 19, 2024

cmsbuild added l1-pending pending-signatures performance-improvements and removed pending-assignment labels Nov 19, 2024

cmsbuild added the hlt-pending label Nov 22, 2024

This was referenced Jan 9, 2025

add default ctor and reset method to ModelLoader cms-hls4ml/hls4mlEmulatorExtras#6

Merged

move per-run changes of L1-uGT emulator to beginRun (incl. loading of AXOL1TL model) #47070

Open

missirol mentioned this issue Jan 10, 2025

update hls4mlEmulatorExtras to v1.1.4 cms-sw/cmsdist#9606

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

potential speedup in HLT: AXOL1TLCondition::evaluateCondition consider caching the model? #46740

potential speedup in HLT: AXOL1TLCondition::evaluateCondition consider caching the model? #46740

slava77 commented Nov 19, 2024 •

edited

Loading

cmsbuild commented Nov 19, 2024 •

edited

Loading

cmsbuild commented Nov 19, 2024

mmusich commented Nov 19, 2024 •

edited

Loading

aloeliger commented Nov 19, 2024

aloeliger commented Nov 19, 2024

mmusich commented Nov 19, 2024

makortel commented Nov 19, 2024

makortel commented Nov 19, 2024

cmsbuild commented Nov 19, 2024

slava77 commented Nov 19, 2024

mmusich commented Nov 19, 2024

slava77 commented Nov 20, 2024

artlbv commented Nov 21, 2024

makortel commented Nov 21, 2024

mmusich commented Nov 22, 2024

cmsbuild commented Nov 22, 2024

jpearkes commented Dec 10, 2024 •

edited

Loading

mmusich commented Dec 10, 2024

jpearkes commented Dec 10, 2024

missirol commented Jan 5, 2025 •

edited

Loading

missirol commented Jan 5, 2025

mmusich commented Jan 8, 2025

missirol commented Jan 9, 2025 •

edited

Loading

aloeliger commented Jan 9, 2025

mmusich commented Jan 9, 2025

missirol commented Jan 9, 2025

missirol commented Jan 9, 2025

potential speedup in HLT: AXOL1TLCondition::evaluateCondition consider caching the model? #46740

potential speedup in HLT: AXOL1TLCondition::evaluateCondition consider caching the model? #46740

Comments

slava77 commented Nov 19, 2024 • edited Loading

cmsbuild commented Nov 19, 2024 • edited Loading

cmsbuild commented Nov 19, 2024

mmusich commented Nov 19, 2024 • edited Loading

aloeliger commented Nov 19, 2024

aloeliger commented Nov 19, 2024

mmusich commented Nov 19, 2024

makortel commented Nov 19, 2024

makortel commented Nov 19, 2024

cmsbuild commented Nov 19, 2024

slava77 commented Nov 19, 2024

mmusich commented Nov 19, 2024

slava77 commented Nov 20, 2024

artlbv commented Nov 21, 2024

makortel commented Nov 21, 2024

mmusich commented Nov 22, 2024

cmsbuild commented Nov 22, 2024

jpearkes commented Dec 10, 2024 • edited Loading

mmusich commented Dec 10, 2024

jpearkes commented Dec 10, 2024

missirol commented Jan 5, 2025 • edited Loading

missirol commented Jan 5, 2025

mmusich commented Jan 8, 2025

missirol commented Jan 9, 2025 • edited Loading

aloeliger commented Jan 9, 2025

mmusich commented Jan 9, 2025

missirol commented Jan 9, 2025

missirol commented Jan 9, 2025

slava77 commented Nov 19, 2024 •

edited

Loading

cmsbuild commented Nov 19, 2024 •

edited

Loading

mmusich commented Nov 19, 2024 •

edited

Loading

jpearkes commented Dec 10, 2024 •

edited

Loading

missirol commented Jan 5, 2025 •

edited

Loading

missirol commented Jan 9, 2025 •

edited

Loading