Add three new lines to select charged kaon decays in HLT1 for Run 3

This MR includes a 3-body SV maker (as compared to the current 2+1 combinations) and a set of lines to select three-body charged kaon decays in HLT1. For now only \pi e e, \pi\mu\mu and 3\pi are implemented. The lines' cuts have been aligned with their analogous HLT2 lines Hlt2RD_KpToPiMuMu, Hlt2RD_KpToPiPiPi and Hlt2RD_KpToPiEE. In addition, the K \rightarrow 3\pi has been pre-scaled to 0.01.

All lines have been tested locally on expected 2024 minimum bias bandwidth division samples with the sequence hlt1_pp_forward_then_matching_1200KHz yielding the following rates:

Hlt1Kplus2PiEE:                                     9/200000, (    1.35 +/-     0.45) kHz
Hlt1Kplus2PiMuMu:                                   6/200000, (    0.90 +/-     0.37) kHz
Hlt1Kplus2PiPiPi:                                   2/200000, (    0.30 +/-     0.21) kHz
Inclusive:                                       9909/200000, ( 1486.35 +/-    14.56) kHz

Without the new lines for comparison:

Inclusive:                                       9895/200000, ( 1484.25 +/-    14.55) kHz

So the lines add about 2 kHz.

The throughput check yields

Device                        Throughput (kHz)    Reference Throughput (kHz)  Speedup    % change    Status
--------------------------  ------------------  ----------------------------  ---------  ----------  --------
NVIDIA RTX A5000                         75.13                         76.14  0.99x      -1.34%      OK
AMD EPYC 7502 32-Core                     8.03                          8.42  0.95x      -4.61%      OK
NVIDIA GeForce RTX 2080 Ti               53.72                         56.09  0.96x      -4.23%      OK
NVIDIA GeForce RTX 3090                  92.44                         93.64  0.99x      -1.28%      OK
Device-averaged speedup: 0.97x
               % change: -2.87%
                 status: DECREASED

with a few % slowdown on the CPU and 2080 NVIDIA GeForce RTX 2080 Ti card and ~ 1% slowdown in the rest. This decrease is mainly due to the 3-body SVs combiner which this MR adds.

Todo:

• Test using 2024 expected minbias and signal MC
• Add LFV decay combinations (this could be postponed to a second stage)
• Adapt to the new monitoring machinery from @ahennequ once it is merged (see !1483 (merged))

Uses the new 2+1 body combiner machinery from @mstahl and @thboettc

fyi @fdettori @acontu @fdordei

Reviewer assignment

• @thboettc for cuda changes
• @mvesteri for overall RTA WP3
• .... TBD

configuration/python/AllenConf/HLT1.py

@@ -22,6 +22,7 @@ from AllenConf.hlt1_monitoring_lines import *
 from AllenConf.hlt1_smog2_lines import *
 from AllenConf.hlt1_downstream_lines import *
 from AllenConf.filters import *
+from AllenConf.hlt1_charged_kaon_lines import *
 
 from AllenConf.hlt1_photon_lines import make_diphotonhighmass_line
 from AllenConf.persistency import make_persistency
@@ -49,6 +50,7 @@ def default_physics_lines(reconstructed_objects, with_calo, with_muon,
     lambda_track_from_c = reconstructed_objects["lambda_track_from_c"]
     v0_twotrack_pairs = reconstructed_objects["v0_sv_twotrack_pairs"]
     dstars = reconstructed_objects["dstars"]
+    three_body_svs = reconstructed_objects["three_body_svs"]
     v0_pairs = reconstructed_objects["v0_pairs"]
     v0_hh_pairs = reconstructed_objects["v0_hh_pairs"]
     muon_stubs = reconstructed_objects["muon_stubs"]
@@ -95,6 +97,12 @@ def default_physics_lines(reconstructed_objects, with_calo, with_muon,
             name="Hlt1DiProtonHighMass",
             pre_scaler=1.,
             enable_tupling=enable_tupling),
+        make_kplus_to_piee_line(
+            three_body_svs, name="Hlt1Kplus2PiEE", enable_monitoring=False),
+        make_kplus_to_pimumu_line(
+            three_body_svs, name="Hlt1Kplus2PiMuMu", enable_monitoring=False),
+        make_kplus_to_3pi_line(
+            three_body_svs, name="Hlt1Kplus2PiPiPi", enable_monitoring=False)
     ]
 
     if 'downstream_tracks' in reconstructed_objects and 'downstream_secondary_vertices' in reconstructed_objects: