Constrain HE TP depth sums to use only 11 bits #23117
Conversation
|
The code-checks are being triggered in jenkins. |
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-23117/4534 |
|
A new Pull Request was created by @christopheralanwest for master. It involves the following packages: SimCalorimetry/HcalTrigPrimAlgos @nsmith-, @rekovic, @cmsbuild, @thomreis can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
|
please test |
|
The tests are being triggered in jenkins. |
|
Comparison job queued. |
|
Comparison is ready @slava77 comparisons for the following workflows were not done due to missing matrix map:
Comparison Summary:
|
|
I don't understand what went wrong in the comparison tests. The discrepancies are entirely in tracking histograms, and appear to be virtually identical to those seen in the comparison tests for PR #23115, which uses the same comparison baseline. No such discrepancies are seen in the comparison tests for the backport of this PR. |
|
The unexpected discrepancies in tracking/vertexing/PV histograms of the comparison test are presumably a consequence of the problem described in #23105, and are unrelated to this PR. |
|
@thomreis Could you, please, take a look and eventually approve this important fix? |
|
+1 So the FW sets the sum value to 0x7ff if the sum is larger than what fits in 11 bits if I understand this correctly? It does not just cut away the bits above the 11th? |
|
This pull request is fully signed and it will be integrated in one of the next master IBs (tests are also fine). This pull request will now be reviewed by the release team before it's merged. @davidlange6, @slava77, @smuzaffar, @fabiocos (and backports should be raised in the release meeting by the corresponding L2) |
|
Comparison job queued. |
|
Comparison is ready Comparison Summary:
|
|
@fabiocos All of the changes in the comparison tests are in tracking/vertexing histograms. These differences appear to be completely spurious, as this PR only makes a simple change to the HCAL trigger primitive emulation. |
|
please test |
cmsbuild
added
comparison-pending
tests-pending
and removed
comparison-available
tests-approved
labels
|
The tests are being triggered in jenkins. |
|
Comparison job queued. |
|
Comparison is ready Comparison Summary:
|
|
+1 |
Quick summary:
This PR fixes an edge case in the emulation of HCAL TPs in the upgraded HE. Potentially these changes could have an effect on 2018 MC workflows in the PR comparison tests, but likely statistics are too low in these tests to see an effect.
Detailed summary:
TP ET mismatches have been seen in the TP correlation plot of the online DQM:
http://cmsonline.cern.ch/cms-elog/1038548
The mismatches only occur for TPs from the 10-degree segmentation region of HE. In this region, the physical segmentation of HE is 10 degrees but, for the convenience of the trigger, the uHTR divides the energy into two TPs to mimic 5-degree segmentation. The computation of TPs proceeds as follows:
input (linearization) LUT -> summing of depths -> (divide-by-two) -> summing of time slices -> compression LUT
where the divide-by-two is only relevant in the 10-degree segmentation region.
The firmware has the following widths:
The emulation uses the same LSB as the uHTR in the summing of the depths, but currently does not constrain the sum to be less than the firmware's 11-bit maximum of 0x7FF. This results in the emulated TP ET being larger than the data TP ET. In the 5-degree segmentation region, this would automatically result in a saturated TP. In contrast, in the 10-degree segmentation region, the divide-by-two reduces the ET so that it may fall in the region 64 < ET < 128 GeV, which does not necessarily result in a saturated TP after the sum of the two samples.
Constraining the depth sum to only 11 bits in the emulation resolves the discrepancies, as shown in the table below. These plots use run 315420 of the JetHT dataset. Though the HcalNZS dataset is normally used for TP emulation tests to avoid spurious mismatches between data and emulation as a result of zero suppression, the statistics are too low for this comparison because of the additional prescale of 4096 for NZS events. In these plots, the variable "soi" indicates the compressed ET in the sample of interest, and is equal to 2*ET with the current compression scheme. The features in abs(ieta)=21-28 disappear. There are additional mismatches at abs(ieta)=15-16 for reasons associated with zero-suppression, which are purely an artifact of using zero-suppressed data in the TP emulation, and are not relevant for this PR.