Creation and implementation of T4 (quadruplet) object #189

jchismar · 2025-07-28T17:53:04Z

This PR introduces the T4 (quadruplet) object to increase efficiency at large displacement. T4s are built from two T3s that share a common LS. Only T3s that are not used in pT5s, pT3s or T5s are considered, and the T3s must start in either the endcap or on barrel layer 3 to reduce duplicates with the other objects. T4s are added as an additional track candidate so that the final TC collection is in the following order: pT5, pT3, T5, T4, and pLS. Motivation for the introduction of T4s was presented at the Tracking POG meeting on 30 Sep 2025.

5 selections are applied to T4s: requiring that the charges of the two T3s must be the same, a DNN classifier, a beta cut for consistency with a circle in r-phi (identical to that used for T5s), a chi-squared cut for consistency with a helix in r-z, and a cut on the impact parameter (dxy). The T4 DNN uses a multi-class architecture (similar to the T3 DNN) which classifies T4s as fake, real prompt, and real displaced. The DNN and dxy cut values are set to keep real displaced tracks while rejecting fake and real prompt tracks.

The displaced efficiency gain on a 50cm cube muon gun sample is shown below, where the current configuration is shown in blue and the addition of T4s is shown in red.

Full plots can be found here for ttbar PU200 and 50cm cube muon gun.

jchismar · 2025-07-28T17:54:17Z

/run all

github-actions · 2025-07-28T18:13:55Z

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     28.6    374.4    184.7    115.0     48.3    714.2    126.9    134.8    178.5      1.6    1907.0    1164.1+/- 279.4     555.8   explicit[s=4] (target branch)
   avg     28.3    376.9    184.4    111.4     54.4    717.4     25.8    128.7    136.0    198.5      2.1    1964.0    1218.3+/- 295.8     576.0   explicit[s=4] (this PR)

github-actions · 2025-07-28T18:53:47Z

There was a problem while building and running with CMSSW. The logs can be found here.

jchismar · 2025-07-30T18:27:57Z

/run cmssw

github-actions · 2025-07-30T20:50:57Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

jchismar · 2025-08-01T18:47:36Z

/run all

github-actions · 2025-08-01T19:05:37Z

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     28.2    372.6    185.0    108.9     48.5    717.1    127.1    134.3    176.7      1.7    1900.0    1154.7+/- 274.6     556.1   explicit[s=4] (target branch)
   avg     28.1    375.1    188.5    111.9     53.7    705.4     24.9    126.7    134.4    192.4      2.1    1943.1    1209.5+/- 287.6     572.0   explicit[s=4] (this PR)

github-actions · 2025-08-01T20:30:45Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

jchismar · 2025-08-04T16:14:29Z

/run gpu-all

github-actions · 2025-08-04T16:31:35Z

The PR was built and ran successfully in standalone mode on GPU. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
[target branch]
   avg     15.2      0.4      0.3      0.6      0.8      0.3      0.7      0.4      0.8      0.0      19.4       3.9+/-  0.7      19.4   explicit[s=1]
   avg      0.9      0.6      0.4      0.8      1.0      0.3      1.0      0.4      1.1      0.0       6.5       5.3+/-  1.0       3.3   explicit[s=2]
   avg      1.6      0.9      0.7      1.2      1.5      0.5      1.6      0.7      1.9      0.0      10.5       8.5+/-  1.7       2.7   explicit[s=4]
   avg      2.2      1.3      1.1      1.8      2.1      0.6      2.2      0.9      2.5      0.0      14.8      11.9+/-  2.8       2.5   explicit[s=6]
   avg      2.8      1.7      1.6      2.3      2.7      0.8      2.8      1.2      3.2      0.0      19.2      15.6+/-  3.8       2.5   explicit[s=8]
[this PR]
   avg     18.4      0.4      0.3      0.6      0.8      0.3      0.4      0.7      0.4      1.2      0.0      23.4       4.7+/-  1.3      23.4   explicit[s=1]
   avg      0.9      0.6      0.4      0.8      0.9      0.3      0.6      0.9      0.4      1.7      0.0       7.6       6.4+/-  1.6       3.8   explicit[s=2]
   avg      1.5      0.8      0.7      1.2      1.4      0.4      1.1      1.4      0.6      2.7      0.0      11.9      10.0+/-  2.4       3.0   explicit[s=4]
   avg      2.2      1.2      1.0      1.6      2.0      0.6      1.7      2.1      1.0      3.4      0.0      16.7      14.0+/-  3.6       2.9   explicit[s=6]
   avg      3.0      1.8      1.4      2.2      2.5      0.7      2.2      2.7      1.2      4.3      0.0      22.1      18.4+/-  4.7       2.8   explicit[s=8]

github-actions · 2025-08-04T17:34:08Z

The PR was built and ran successfully with CMSSW on GPU. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

jchismar · 2025-08-06T01:08:11Z

/run all

github-actions · 2025-08-06T01:27:04Z

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     28.3    373.6    185.5    110.1     49.3    720.0    126.6    135.4    178.2      1.7    1908.8    1160.4+/- 274.8     556.9   explicit[s=4] (target branch)
   avg     28.6    381.3    184.9    111.8     55.1    710.0     21.4    126.8    138.8    188.2      2.1    1949.0    1210.4+/- 290.9     575.1   explicit[s=4] (this PR)

github-actions · 2025-08-06T03:00:33Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

slava77 · 2025-08-06T15:46:15Z

/run gpu-all

github-actions · 2025-08-06T16:03:25Z

The PR was built and ran successfully in standalone mode on GPU. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
[target branch]
   avg     15.0      0.4      0.3      0.6      0.8      0.3      0.6      0.4      0.9      0.0      19.2       3.9+/-  0.7      19.3   explicit[s=1]
   avg      0.8      0.6      0.4      0.8      1.0      0.3      1.0      0.5      1.3      0.0       6.7       5.5+/-  1.0       3.4   explicit[s=2]
   avg      1.5      0.9      0.8      1.2      1.5      0.4      1.7      0.7      2.0      0.0      10.8       8.8+/-  1.7       2.7   explicit[s=4]
   avg      2.2      1.2      1.1      1.9      2.2      0.7      2.4      1.0      2.7      0.0      15.5      12.6+/-  2.7       2.6   explicit[s=6]
   avg      2.9      1.8      1.5      2.4      2.9      0.8      3.0      1.4      3.5      0.0      20.3      16.5+/-  3.9       2.6   explicit[s=8]
[this PR]
   avg     18.4      0.4      0.3      0.6      0.8      0.3      0.4      0.6      0.4      1.2      0.0      23.3       4.6+/-  0.9      23.3   explicit[s=1]
   avg      0.9      0.6      0.4      0.8      1.0      0.3      0.6      1.0      0.4      1.6      0.0       7.6       6.4+/-  1.2       3.8   explicit[s=2]
   avg      1.7      0.9      0.7      1.2      1.5      0.4      1.2      1.5      0.6      2.5      0.0      12.2      10.1+/-  2.1       3.1   explicit[s=4]
   avg      2.4      1.3      1.1      1.7      2.0      0.6      1.7      2.2      1.0      3.5      0.0      17.4      14.4+/-  3.3       3.0   explicit[s=6]
   avg      2.9      1.8      1.5      2.3      2.6      0.8      2.3      2.8      1.3      4.3      0.0      22.5      18.9+/-  4.4       2.9   explicit[s=8]

github-actions · 2025-08-06T16:58:22Z

The PR was built and ran successfully with CMSSW on GPU. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

jchismar · 2025-08-07T18:05:21Z

/run cmssw

slava77 · 2025-08-07T19:21:24Z

instead of adding T4s in parallel in CMSSW, I'd rather we keep them together with T5s.
Maintenance of the LST candidate split is already quite tedious and I don't see a strong reason to have a split.
Renaming corresponding T5 cmssw collections to T45 or T4T5 then would be enough.

github-actions · 2025-08-07T20:00:00Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

jchismar · 2025-08-07T20:01:48Z

instead of adding T4s in parallel in CMSSW, I'd rather we keep them together with T5s. Maintenance of the LST candidate split is already quite tedious and I don't see a strong reason to have a split. Renaming corresponding T5 cmssw collections to T45 or T4T5 then would be enough.

Got it, I'll do that instead.

jchismar · 2025-08-08T02:08:15Z

/run cmssw

github-actions · 2025-08-08T03:51:57Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

GNiendorf · 2025-08-22T20:51:14Z

Adding the charge cut to the T5's causes a big drop in efficiency? Any ideas why it works for T4s? @slava77
To give context to Jade, currently looking at quick cuts that could be added to the T5's counting kernel.

Commit here, added to counting + creation kernels: d9b77ef

I get the same plots only adding the cut to the creation kernel.

slava77 · 2025-08-22T21:44:27Z

Adding the charge cut to the T5's causes a big drop in efficiency? Any ideas why it works for T4s? @slava77

How large is the inefficiency?
One plot posted is not enough to see a change (I'm not sure which setup that is).

I can speculate that at T4 level some inefficiency was acceptable; also, having two MDs in common probably reduces the inefficiency of this cut

GNiendorf · 2025-08-22T21:56:31Z

Adding the charge cut to the T5's causes a big drop in efficiency? Any ideas why it works for T4s? @slava77

How large is the inefficiency? One plot posted is not enough to see a change (I'm not sure which setup that is).

This is just the normal PU200 setup, you can see the pT5 efficiency drops drastically. pT5 markers are at like ~20% above.

Here is the same plot without that charge cut for comparison:

slava77 · 2025-08-22T22:01:25Z

This is just the normal PU200 setup, you can see the pT5 efficiency drops drastically. pT5 markers are at like ~20% above.

ah, indeed, this looks broken

slava77 · 2025-08-22T22:13:10Z

Commit here, added to counting + creation kernels: d9b77ef

I didn't notice a bug.
You may want to recompile from scratch and add some printouts (debug)

github-actions · 2025-10-17T15:48:58Z

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     30.8    365.9    271.4    122.1     46.1    849.8    125.5    131.1    192.0      1.6    2136.2    1255.6+/- 295.7     643.9   explicit[s=4] (target branch)
   avg     28.6    363.9    284.8    128.7     49.2    591.2     12.1    125.0    133.4    170.9      1.5    1889.2    1269.4+/- 301.8     588.4   explicit[s=4] (this PR)

github-actions · 2025-10-17T17:41:15Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

slava77

many comments related to formatting are not repeated; please check for similar patterns in the added code

slava77 · 2025-08-06T15:50:20Z

RecoTracker/LSTCore/interface/QuadrupletsHostCollection.h

+namespace lst {
+  using QuadrupletsHostCollection = PortableHostMultiCollection<QuadrupletsSoA, QuadrupletsOccupancySoA>;
+}  // namespace lst
+#endif


end line is missing

slava77 · 2025-08-06T15:54:05Z

RecoTracker/LSTCore/interface/Common.h

+  struct Params_T4 {
+    static constexpr int kLayers = 4, kHits = 8;
+    using ArrayU8xLayers = edm::StdArray<uint8_t, kLayers>;
+    using ArrayU16xLayers = edm::StdArray<uint16_t, kLayers>;
+    using ArrayUxHits = edm::StdArray<unsigned int, kHits>;
+  };


I'd move this up, between T3 and T5 to be in more logical order

slava77 · 2025-08-28T00:02:29Z

RecoTracker/LST/plugins/LSTOutputConverter.cc

      // only when required by a flag for other pT objects.
-      if (includeNonpLSTSs_ || lstOutput_view.trackCandidateType()[i] == lst::LSTObjType::T5) {
+      if (includeNonpLSTSs_ || (lstOutput_view.trackCandidateType()[i] == lst::LSTObjType::T5) ||
+          (lstOutput_view.trackCandidateType()[i] == lst::LSTObjType::T4)) {


6 lines below is if (lstOutput_view.trackCandidateType()[i] == lst::LSTObjType::T5) {
I think that this check should stay for T4s that still have 2 first PS layers, if that logic is still required.
The barrel-3 T4s should be OK to require just the first P-side hit and pick up the second only on the next layer.

OTOH, perhaps it's worth to revisit that P-side selection for the first two hits, if it is still required.
I recall that it was needed due to the features of the initial state estimator (done using the first two hits) in the seed generator from hits (used below).
But maybe things changed since the initial introduction of this code.

This is an old comment but I think it might be still relevant. Currently the T4s aren't included in the check for having the first 2 hits be PS (in L218-222 in the current LSTOutputConverter.cc). Should I include them? From my understanding if it's just checking the first 2 hits then it should have no impact on any of the T4s.

this comment was indeed from the time when I started the review in August

(since the very early version), now that all barrel T4s start from B3, this requirement would kill all T4s.
The fact that we see something (visible in the fake+dr plots though) looks like using P+S for the first two hits still works.

But it could be that the lack of any change in efficiency is still indicative of a problem.

The logic should be less that "the first two should be P-side", but rather that if one of the two first hits is "S", it has to be on another layer and not in the same doublet module stack.

@jchismar
what do you see in the cube sample vs eta in CMSSW? (just to use a sample where T4s are expected to contribute)

I haven't checked on the cube sample in CMSSW, only ttbar with and without PU. I don't think I know which input file to use for the cube sample, do you know of one?

it may need to be remade.
@VourMa @GNiendorf did you make a cube sample relatively recently (to perhaps still have the step1 config around) ?

since the goal for the test in my question/request is not so much the cube but rather to see T4s vs eta, perhaps running a local test on ttbar with T5s disabled (during building) would be enough ... or even also pT3s as well.
This way the potential issue with the hit selection in the output converter would be more clear.

ttbar with T5s disabled (during building) would be enough ... or even also pT3s as well.

uhm, no, don't disable pT3s: the reference should have something to compare with

slava77 · 2025-10-28T22:13:34Z

RecoTracker/LSTCore/src/alpaka/Quadruplet.h

+      const int flatThreadExtent = blockSize;  // total threads per block
+
+      for (int iter : cms::alpakatools::uniform_groups_z(acc, nEligibleT4Modules)) {
+        uint16_t lowerModule1 = ranges.indicesOfEligibleT4Modules()[iter];


Suggested change

uint16_t lowerModule1 = ranges.indicesOfEligibleT4Modules()[iter];

const uint16_t lowerModule1 = ranges.indicesOfEligibleT4Modules()[iter];

check the rest of the code to more systematically keep things const if they are supposed to be

slava77 · 2025-10-28T22:23:11Z

RecoTracker/LSTCore/src/alpaka/Quadruplet.h

+            unsigned int innerOuterInnerMiniDoubletIndex =
+                segments.mdIndices()[secondSegmentIndex][0];  //inner triplet outer segment inner MD index


please try to keep the names more compact or at least more readable: innerOuterInner requires the comment to be understood.

if the name was more clear the comment is not necessary

Suggested change

unsigned int innerOuterInnerMiniDoubletIndex =

segments.mdIndices()[secondSegmentIndex][0]; //inner triplet outer segment inner MD index

unsigned int innerT3OuterLSInnerMDIndex = segments.mdIndices()[secondSegmentIndex][0];

or

Suggested change

unsigned int innerOuterInnerMiniDoubletIndex =

segments.mdIndices()[secondSegmentIndex][0]; //inner triplet outer segment inner MD index

unsigned int innerT3MD2Index = segments.mdIndices()[secondSegmentIndex][0];

slava77 · 2025-10-28T23:35:50Z

RecoTracker/LSTCore/standalone/code/core/trkCore.cc

 }

+//___________________________________________________________________________________________________________________________________________________________________________________________
+float runQuadruplet(LSTEvent* event) {


keep this close to T5s

slava77 · 2025-10-28T23:38:46Z

RecoTracker/LSTCore/standalone/code/core/trkCore.cc

-    timing_total_short += timing[7] * 1000;  // pT3
-    timing_total_short += timing[8] * 1000;  // TC
-    timing_total_short += timing[9] * 1000;  // Reset
+    timing_total += timing[0] * 1000;         // Hits


why is this not a loop

slava77 · 2025-10-28T23:40:55Z

RecoTracker/LSTCore/standalone/code/core/write_lst_ntuple.cc

 }

+//________________________________________________________________________________________________________________________________
+void createQuadrupletBranches() {


keep closer to T5 (similar code to be closer together)

slava77 · 2025-10-28T23:41:36Z

RecoTracker/LSTCore/standalone/code/core/write_lst_ntuple.cc

 }

+//________________________________________________________________________________________________________________________________
+std::map<unsigned int, unsigned int> setQuadrupletBranches(LSTEvent* event,


keep closer to T5 code

slava77 · 2025-10-28T23:42:35Z

RecoTracker/LSTCore/standalone/code/core/write_lst_ntuple.cc

+    std::vector<float> const& trk_ph2_z) {
+  auto const trackCandidatesExtended = event->getTrackCandidatesExtended();
+  auto const quadruplets = event->getQuadruplets<QuadrupletsSoA>();
+  unsigned int T4 = trackCandidatesExtended.directObjectIndices()[idx];


Suggested change

unsigned int T4 = trackCandidatesExtended.directObjectIndices()[idx];

unsigned int t4 = trackCandidatesExtended.directObjectIndices()[idx];

here and elsewhere to match the CMSSW code style (only types are capitalized first letter)

slava77 · 2025-11-03T18:42:44Z

RecoTracker/LST/plugins/LSTOutputConverter.cc

            auto hType = tracker.getDetectorType(hit.geographicalId());
            if (hType != TrackerGeometry::ModuleType::Ph2PSP && n < 2)
              continue;  // the first two should be P


Do I recall correctly that all T4s start on a PS layer?
In this case the logic may be just if (hType == TrackerGeometry::ModuleType::Ph2PSS && n < 1) continue;
and change the comment to assume the first layer is a PS; skip the first S-side hit

ah, no, if (hType == TrackerGeometry::ModuleType::Ph2PSS && n < 1) is not enough to skip the first S; need to count the total hits so far.

Manos pointed me to here to get the layer info. I'll take a look at it tomorrow, and also double check about T4s only starting on a PS layer.

During the meeting I was talking about layer index that we can get given a detId; it needs a tracker topology though.
Need to add const edm::ESGetToken<TrackerTopology, TrackerTopologyRcd> tTopoToken_; to the member data and call esConsumes on it in the constructor (similar to the geometry token). Then
const TrackerTopology& tTopo = iSetup.getData(tTopoToken_); at the start of ::produce and then tTopo.layer(hit.geographicalId()) will give you the layer.

I'd then get the first hit layer index and if (hType == TrackerGeometry::ModuleType::Ph2PSS && hitLayer == firstLayer) continue;

jchismar · 2025-11-10T11:00:52Z

/run all

github-actions · 2025-11-10T11:14:28Z

There was a problem while building and running in standalone mode. The logs can be found here.

jchismar · 2025-11-10T14:46:41Z

/run all

jchismar · 2025-11-10T14:56:15Z

/run all

github-actions · 2025-11-10T15:17:02Z

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     27.8    366.1    294.0    118.7     45.2    665.9    119.2    125.2    162.5      1.5    1926.1    1232.4+/- 294.1     584.8   explicit[s=4] (target branch)
   avg     30.1    369.3    266.7    114.8     51.6    684.6     11.7    119.8    130.3    184.9      1.5    1965.5    1250.7+/- 297.0     603.6   explicit[s=4] (this PR)

github-actions · 2025-11-10T16:56:50Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

jchismar · 2025-11-26T13:31:20Z

/run all

github-actions · 2025-11-26T13:52:36Z

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     28.5    378.5    302.0    124.3     48.0    686.6    123.1    131.0    166.7      1.6    1990.2    1275.1+/- 298.9     609.6   explicit[s=4] (target branch)
   avg     30.7    376.9    274.2    119.4     53.5    702.1     11.9    123.4    134.7    189.8      1.5    2018.1    1285.4+/- 304.4     626.7   explicit[s=4] (this PR)

github-actions · 2025-11-26T15:15:55Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

slava77 · 2025-12-04T14:50:35Z

/run all
/run gpu-all

github-actions · 2025-12-04T15:11:59Z

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     28.7    375.7    298.7    121.6     46.1    680.8    122.0    127.9    167.4      2.1    1970.9    1261.5+/- 300.0     602.4   explicit[s=4] (target branch)
   avg     30.4    379.4    271.9    117.7     52.0    700.5     11.7    121.7    131.7    189.5      1.5    2008.0    1277.0+/- 305.8     620.0   explicit[s=4] (this PR)

github-actions · 2025-12-04T15:12:06Z

The PR was built and ran successfully in standalone mode on GPU. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
[target branch]
   avg     24.5      0.4      0.4      0.5      0.9      0.3      0.6      0.4      0.9      0.0      28.9       4.1+/-  1.7      28.9   explicit[s=1]
   avg      1.1      0.6      0.6      0.7      1.1      0.3      1.0      0.5      1.3      0.0       7.1       5.7+/-  1.9       3.6   explicit[s=2]
   avg      2.2      0.9      1.0      1.2      1.6      0.4      1.5      0.6      2.0      0.0      11.5       8.9+/-  2.6       3.0   explicit[s=4]
   avg      3.0      1.3      1.5      1.8      2.3      0.6      2.1      0.9      2.7      0.0      16.1      12.6+/-  3.4       2.8   explicit[s=6]
   avg      4.3      1.8      2.1      2.5      3.0      0.7      2.7      1.2      3.3      0.0      21.7      16.7+/-  3.6       2.8   explicit[s=8]
[this PR]
   avg     33.9      0.4      0.4      0.5      0.9      0.3      0.6      0.6      0.4      1.2      0.0      39.2       5.0+/-  2.6      39.2   explicit[s=1]
   avg      1.3      0.6      0.6      0.7      1.1      0.3      0.9      0.9      0.4      1.6      0.0       8.3       6.7+/-  3.0       4.3   explicit[s=2]
   avg      2.5      0.8      0.9      1.2      1.6      0.4      1.4      1.5      0.7      2.5      0.0      13.6      10.6+/-  3.8       3.5   explicit[s=4]
   avg      3.2      1.3      1.5      1.8      2.2      0.6      2.0      1.9      0.8      3.3      0.0      18.6      14.8+/-  4.4       3.2   explicit[s=6]
   avg      4.1      1.7      2.0      2.5      2.8      0.7      2.6      2.6      1.2      4.4      0.0      24.5      19.7+/-  5.6       3.1   explicit[s=8]

slava77 · 2025-12-04T15:42:08Z

Here is a timing comparison:

it looks like s=8 is slower by about 10%.
@jchismar do you have results from a local run?

github-actions · 2025-12-04T16:22:18Z

The PR was built and ran successfully with CMSSW on GPU. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

jchismar · 2025-12-04T16:22:27Z

Here is a timing comparison:

it looks like s=8 is slower by about 10%. @jchismar do you have results from a local run?

Here is the timing from a local run:
Baseline

This PR

github-actions · 2025-12-04T17:12:06Z

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

jchismar force-pushed the t4_creation branch from f439084 to bd481ef Compare August 1, 2025 18:44

jchismar force-pushed the t4_creation branch from aec07f0 to add9c04 Compare August 8, 2025 02:06

jchismar force-pushed the t4_creation branch from 059fd46 to 8b51a41 Compare October 28, 2025 14:04

slava77 reviewed Oct 28, 2025

View reviewed changes

slava77 reviewed Nov 3, 2025

View reviewed changes

jchismar mentioned this pull request Nov 4, 2025

Move T3 charge calculation #211

Closed

jchismar force-pushed the t4_creation branch from 8b51a41 to 2358fff Compare November 10, 2025 10:59

jchismar force-pushed the t4_creation branch from 2358fff to 7840d07 Compare November 10, 2025 14:45

jchismar force-pushed the t4_creation branch from 7840d07 to 3ec1f23 Compare November 10, 2025 14:55

jchismar force-pushed the t4_creation branch from 3ec1f23 to 45e831b Compare November 26, 2025 13:30

Creation and implementation of T4 (quadruplet) object

acef004

jchismar force-pushed the t4_creation branch from 45e831b to acef004 Compare December 3, 2025 13:31

github-actions bot merged commit f759879 into master Dec 9, 2025
5 checks passed

slava77 mentioned this pull request Dec 9, 2025

LST T5-T5 Duplicate Merging #212

Merged

	uint16_t lowerModule1 = ranges.indicesOfEligibleT4Modules()[iter];
	const uint16_t lowerModule1 = ranges.indicesOfEligibleT4Modules()[iter];

		unsigned int innerOuterInnerMiniDoubletIndex =
		segments.mdIndices()[secondSegmentIndex][0]; //inner triplet outer segment inner MD index

	unsigned int innerOuterInnerMiniDoubletIndex =
	segments.mdIndices()[secondSegmentIndex][0]; //inner triplet outer segment inner MD index
	unsigned int innerT3OuterLSInnerMDIndex = segments.mdIndices()[secondSegmentIndex][0];

	unsigned int innerOuterInnerMiniDoubletIndex =
	segments.mdIndices()[secondSegmentIndex][0]; //inner triplet outer segment inner MD index
	unsigned int innerT3MD2Index = segments.mdIndices()[secondSegmentIndex][0];

	unsigned int T4 = trackCandidatesExtended.directObjectIndices()[idx];
	unsigned int t4 = trackCandidatesExtended.directObjectIndices()[idx];

Creation and implementation of T4 (quadruplet) object #189

Creation and implementation of T4 (quadruplet) object #189

Uh oh!

Conversation

jchismar commented Jul 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jchismar commented Jul 28, 2025

Uh oh!

github-actions bot commented Jul 28, 2025

Uh oh!

github-actions bot commented Jul 28, 2025

Uh oh!

jchismar commented Jul 30, 2025

Uh oh!

github-actions bot commented Jul 30, 2025

Uh oh!

jchismar commented Aug 1, 2025

Uh oh!

github-actions bot commented Aug 1, 2025

Uh oh!

github-actions bot commented Aug 1, 2025

Uh oh!

jchismar commented Aug 4, 2025

Uh oh!

github-actions bot commented Aug 4, 2025

Uh oh!

github-actions bot commented Aug 4, 2025

Uh oh!

jchismar commented Aug 6, 2025

Uh oh!

github-actions bot commented Aug 6, 2025

Uh oh!

github-actions bot commented Aug 6, 2025

Uh oh!

slava77 commented Aug 6, 2025

Uh oh!

github-actions bot commented Aug 6, 2025

Uh oh!

github-actions bot commented Aug 6, 2025

Uh oh!

jchismar commented Aug 7, 2025

Uh oh!

slava77 commented Aug 7, 2025

Uh oh!

github-actions bot commented Aug 7, 2025

Uh oh!

jchismar commented Aug 7, 2025

Uh oh!

jchismar commented Aug 8, 2025

Uh oh!

github-actions bot commented Aug 8, 2025

Uh oh!

GNiendorf commented Aug 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

slava77 commented Aug 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

GNiendorf commented Aug 22, 2025

Uh oh!

slava77 commented Aug 22, 2025

Uh oh!

slava77 commented Aug 22, 2025

Uh oh!

github-actions bot commented Oct 17, 2025

Uh oh!

github-actions bot commented Oct 17, 2025

Uh oh!

slava77 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jchismar commented Jul 28, 2025 •

edited

Loading

GNiendorf commented Aug 22, 2025 •

edited

Loading

slava77 commented Aug 22, 2025 •

edited

Loading