Hi @oliviermattelaer ,

as discussed in https://github.com/madgraph5/madgraph4gpu/pull/935#issuecomment-2256328718

This is a reminder to check if the second helicity filtering in fortran for mirror processes can be removed.

For the moment, my proposal to solve #872 via PR #935 is to

keep computing twice helicities in fortran
keep computing only once helicities in cudacpp
add a second reset_cumulative_variable call in cudacpp for mirri processes (to fix 872 and ensure that frtran and cudacpp give the same results)
add a sanity check in fortran that the two lists of good helicities are the same

If, as you suggest, we eventually find that helicities can be computed only once in fortran, then

the second reset_cumative_variable call can be removed in cudacpp
there is no need for a sanity chec in fortran any more, as helicities are computed only once

I assign this to you... Low priority, in my opinion Thanks Andrea

The fact to have Ntry/goodhel to have a mirror index is ... as old as the matrix_madevent_group_v4.inc file (this is even the original reason of such file): https://github.com/mg5amcnlo/mg5amcnlo/tree/382e71efd61ef2c2f19be5664b1228220e095ebe (version number is 1.0.0 -> 2011)

Now I have added a test comparison that crash at fortran stage if this is not the case (in LTS) and in crashed for p p > t t~ j j

[x] P1_gq_ttxgq
[x] P1_qq_ttxgg
[x] P1_gg_ttxgg
[x] P1_gg_ttxqq
[x] P1_qq_ttxqq directory (see details report below):

REPORT for P1_qq_ttxqq :

     Added good helicity            4   3.1889029540868283       in event            1 local:           1 mirror           1
     Added good helicity            8  0.63182311673402225       in event            1 local:           1 mirror           1
     Added good helicity           12   2.3539048543925931       in event            1 local:           1 mirror           1
     Added good helicity           16   2.8011885049441809       in event            1 local:           1 mirror           1
     Added good helicity           18   1.3875388603326930       in event            1 local:           1 mirror           1
     Added good helicity           19   8.1889062609589534       in event            1 local:           1 mirror           1
     Added good helicity           22  0.44425339515692869       in event            1 local:           1 mirror           1
     Added good helicity           23   1.6084314717877597       in event            1 local:           1 mirror           1
     Added good helicity           26   1.0515001769765651       in event            1 local:           1 mirror           1
     Added good helicity           27   3.2861119324407699       in event            1 local:           1 mirror           1
     Added good helicity           30   2.2457791222478369       in event            1 local:           1 mirror           1
     Added good helicity           31   4.8116593499408635       in event            1 local:           1 mirror           1
     Added good helicity           34   4.8116593499408635       in event            1 local:           1 mirror           1
     Added good helicity           35   2.2457791222478369       in event            1 local:           1 mirror           1
     Added good helicity           38   3.2861119324407699       in event            1 local:           1 mirror           1
     Added good helicity           39   1.0515001769765657       in event            1 local:           1 mirror           1
     Added good helicity           42   1.6084314717877601       in event            1 local:           1 mirror           1
     Added good helicity           43  0.44425339515692841       in event            1 local:           1 mirror           1
     Added good helicity           46   8.1889062609589551       in event            1 local:           1 mirror           1
     Added good helicity           47   1.3875388603326930       in event            1 local:           1 mirror           1
     Added good helicity           49   2.8011885049441791       in event            1 local:           1 mirror           1
     Added good helicity           53   2.3539048543925931       in event            1 local:           1 mirror           1
     Added good helicity           57  0.63182311673402225       in event            1 local:           1 mirror           1
     Added good helicity           61   3.1889029540868288       in event            1 local:           1 mirror           1
     Added good helicity            1   1.2599838176532971       in event            2 local:           1 mirror           1
     Added good helicity            5  0.41908560994560851       in event            2 local:           1 mirror           1
     Added good helicity            9  0.93354495955340910       in event            2 local:           1 mirror           1
     Added good helicity           13   2.3615765200130308       in event            2 local:           1 mirror           1
     Added good helicity           52   2.3588857801434031       in event            2 local:           1 mirror           1
     Added good helicity           56  0.93295325385162231       in event            2 local:           1 mirror           1
     Added good helicity           60  0.41863175273867276       in event            2 local:           1 mirror           1
     Added good helicity           64   1.2586012261521977       in event            2 local:           1 mirror           1
     Added good helicity            1   1.2599838176532958       in event            1 local:           1 mirror           2
     Added good helicity            4   10.209795896012128       in event            1 local:           1 mirror           2
     Added good helicity            5  0.41908560994560839       in event            1 local:           1 mirror           2
     Added good helicity            8   1.9285010395071618       in event            1 local:           1 mirror           2
     Added good helicity            9  0.93354495955340966       in event            1 local:           1 mirror           2
     Added good helicity           12   3.5860261579071175       in event            1 local:           1 mirror           2
     Added good helicity           13   2.3615765200130303       in event            1 local:           1 mirror           2
     Added good helicity           16   4.8956255377021494       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            1 local:           1 mirror           2
     Added good helicity           19   2.3943788533168218       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            1 local:           1 mirror           2
     Added good helicity           23  0.52860032119394884       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            1 local:           1 mirror           2
     Added good helicity           27   1.9102840720006069       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            1 local:           1 mirror           2
     Added good helicity           31   1.5725397269893320       in event            1 local:           1 mirror           2
     Added good helicity           34   1.5725397269893304       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            1 local:           1 mirror           2
     Added good helicity           38   1.9102840720006067       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            1 local:           1 mirror           2
     Added good helicity           42  0.52860032119394840       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            1 local:           1 mirror           2
     Added good helicity           46   2.3943788533168200       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            1 local:           1 mirror           2
     Added good helicity           49   4.8967684628107904       in event            1 local:           1 mirror           2
     Added good helicity           52   2.3588857801434022       in event            1 local:           1 mirror           2
     Added good helicity           53   3.5867866708050791       in event            1 local:           1 mirror           2
     Added good helicity           56  0.93295325385162242       in event            1 local:           1 mirror           2
     Added good helicity           57   1.9288943118773216       in event            1 local:           1 mirror           2
     Added good helicity           60  0.41863175273867259       in event            1 local:           1 mirror           2
     Added good helicity           61   10.212733056325586       in event            1 local:           1 mirror           2
     Added good helicity           64   1.2586012261521971       in event            1 local:           1 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            2 local:           1 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            2 local:           1 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            2 local:           1 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            2 local:           1 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            2 local:           1 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            2 local:           1 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            2 local:           1 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            2 local:           1 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            3 local:           1 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            3 local:           1 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            3 local:           1 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            3 local:           1 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            3 local:           1 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            3 local:           1 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            3 local:           1 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            3 local:           1 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            4 local:           1 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            4 local:           1 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            4 local:           1 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            4 local:           1 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            4 local:           1 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            4 local:           1 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            4 local:           1 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            4 local:           1 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            5 local:           2 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            5 local:           2 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            5 local:           2 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            5 local:           2 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            5 local:           2 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            5 local:           2 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            5 local:           2 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            5 local:           2 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            6 local:           2 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            6 local:           2 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            6 local:           2 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            6 local:           2 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            6 local:           2 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            6 local:           2 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            6 local:           2 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            6 local:           2 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            7 local:           2 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            7 local:           2 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            7 local:           2 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            7 local:           2 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            7 local:           2 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            7 local:           2 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            7 local:           2 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            7 local:           2 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            8 local:           2 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            8 local:           2 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            8 local:           2 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            8 local:           2 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            8 local:           2 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            8 local:           2 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            8 local:           2 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            8 local:           2 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event            9 local:           3 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event            9 local:           3 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event            9 local:           3 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event            9 local:           3 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event            9 local:           3 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event            9 local:           3 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event            9 local:           3 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event            9 local:           3 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           10 local:           3 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           10 local:           3 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           10 local:           3 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           10 local:           3 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           10 local:           3 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           10 local:           3 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           10 local:           3 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           10 local:           3 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           11 local:           3 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           11 local:           3 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           11 local:           3 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           11 local:           3 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           11 local:           3 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           11 local:           3 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           11 local:           3 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           11 local:           3 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           12 local:           3 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           12 local:           3 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           12 local:           3 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           12 local:           3 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           12 local:           3 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           12 local:           3 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           12 local:           3 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           12 local:           3 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           13 local:           4 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           13 local:           4 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           13 local:           4 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           13 local:           4 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           13 local:           4 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           13 local:           4 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           13 local:           4 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           13 local:           4 mirror           2
     RESET CUMULATIVE VARIABLE
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           14 local:           4 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           14 local:           4 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           14 local:           4 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           14 local:           4 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           14 local:           4 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           14 local:           4 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           14 local:           4 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           14 local:           4 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           15 local:           4 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           15 local:           4 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           15 local:           4 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           15 local:           4 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           15 local:           4 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           15 local:           4 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           15 local:           4 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           15 local:           4 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           16 local:           4 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           16 local:           4 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           16 local:           4 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           16 local:           4 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           16 local:           4 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           16 local:           4 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           16 local:           4 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           16 local:           4 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           17 local:           5 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           17 local:           5 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           17 local:           5 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           17 local:           5 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           17 local:           5 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           17 local:           5 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           17 local:           5 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           17 local:           5 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           18 local:           5 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           18 local:           5 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           18 local:           5 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           18 local:           5 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           18 local:           5 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           18 local:           5 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           18 local:           5 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           18 local:           5 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           19 local:           5 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           19 local:           5 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           19 local:           5 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           19 local:           5 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           19 local:           5 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           19 local:           5 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           19 local:           5 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           19 local:           5 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           20 local:           5 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           20 local:           5 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           20 local:           5 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           20 local:           5 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           20 local:           5 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           20 local:           5 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           20 local:           5 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           20 local:           5 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           21 local:           6 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           21 local:           6 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           21 local:           6 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           21 local:           6 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           21 local:           6 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           21 local:           6 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           21 local:           6 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           21 local:           6 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           22 local:           6 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           22 local:           6 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           22 local:           6 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           22 local:           6 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           22 local:           6 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           22 local:           6 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           22 local:           6 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           22 local:           6 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           23 local:           6 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           23 local:           6 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           23 local:           6 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           23 local:           6 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           23 local:           6 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           23 local:           6 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           23 local:           6 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           23 local:           6 mirror           2
     not adding good helicity (good for other mirror)          18   0.0000000000000000       in event           24 local:           6 mirror           2
     not adding good helicity (good for other mirror)          22   0.0000000000000000       in event           24 local:           6 mirror           2
     not adding good helicity (good for other mirror)          26   0.0000000000000000       in event           24 local:           6 mirror           2
     not adding good helicity (good for other mirror)          30   0.0000000000000000       in event           24 local:           6 mirror           2
     not adding good helicity (good for other mirror)          35   0.0000000000000000       in event           24 local:           6 mirror           2
     not adding good helicity (good for other mirror)          39   0.0000000000000000       in event           24 local:           6 mirror           2
     not adding good helicity (good for other mirror)          43   0.0000000000000000       in event           24 local:           6 mirror           2
     not adding good helicity (good for other mirror)          47   0.0000000000000000       in event           24 local:           6 mirror           2

Adding only those helicities to the fortran code does not change (at all) the cross-section (so bit to bit compatibility and really 0 helicity

     Cross-section :   11.35 +- 0.03021 pb
     Nb of events :  10000

versus

     Cross-section :   11.35 +- 0.03021 pb
     Nb of events :  10000

So the issue is occuring due to uu>uu that has some helicity that contribute only for that process while in itself uu~ and u~u does have symmetric goodhel as expected but goodhel is here a global variable for all processes... Making it non global can create issue for the grid handling (and therefore for loop-induced/gridpack?)

But the good news is that for "gpu grouping" (so what we have now, this is indeed symmetric). This still needs more investigation to answer the question to with if we can handle this without the need of imirror or not.

Hi @oliviermattelaer thanks for looking at this :-)

I was wondering this morning, does Madgraph include the option of polarised beams? Because one physics case where I imagine different helicity lists is that of polarised beams. Well, typically that would be e+ against e- which is not symmetric anyway, but in principle if you have something like e- against e- and the left and right beams have different polarisations, then the list of helicities are different. Far fetched, but maybe something similar is the reason why this was introduced at the time?

(Reminder from yesterday's meeting: you said that the computation of helicities in fortran separately for each mirror was introduced by JohannA in one of the first bug fix releases 1.1.0 or similar, "to fix a bug" or similar, so maybe there was a use case - PS ops sorry I see you mentioned it it the post above, giving details , thanks!).

Anyway, in the meantime as discussed I will look at introducing a sanity check that if two lists of helicities are computed in fortran, then they are the same.

So the commit message of Johan is: Fixed problem when different helicities are zero for main process and mirror process for grouped subprocess output related information in the UpdateNotes:

                   JA: Improved helicity selection and automatic full helicity
                       sum if needed. Optimization of run parameters.

So the issue is occuring due to uu>uu that has some helicity that contribute only for that process while in itself uu~ and u~u does have symmetric goodhel as expected but goodhel is here a global variable for all processes... Making it non global can create issue for the grid handling (and therefore for loop-induced/gridpack?)

But the good news is that for "gpu grouping" (so what we have now, this is indeed symmetric). This still needs more investigation to answer the question to with if we can handle this without the need of imirror or not.

Hi Olivier, thanks again for looking at this :-)

I implemented my sanity checks, I see no issues in pp->ttjj (well, actually in pp_tt012j).

Now, there is one important difference: in the cudacpp generated code, remember that we split the processes (maybe this is what you mean by gpu grouping?). In particular pp -> tt 0,1,2 jets has

ls -d1 pp_tt012j.mad/SubProcesses/P*
pp_tt012j.mad/SubProcesses/P0_gg_ttx/
pp_tt012j.mad/SubProcesses/P0_uux_ttx/
pp_tt012j.mad/SubProcesses/P1_gg_ttxg/
pp_tt012j.mad/SubProcesses/P1_gu_ttxu/
pp_tt012j.mad/SubProcesses/P1_gux_ttxux/
pp_tt012j.mad/SubProcesses/P1_uux_ttxg/
pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/
pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/
pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/
pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/
pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/
pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/
pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/
pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/
pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/
pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/
pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/
pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/

\grep MIRRORPROCS pp_tt012j.mad/SubProcesses/P* -r | \grep DATA
pp_tt012j.mad/SubProcesses/P0_gg_ttx/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.FALSE./
pp_tt012j.mad/SubProcesses/P0_uux_ttx/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P1_gg_ttxg/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.FALSE./
pp_tt012j.mad/SubProcesses/P1_gu_ttxu/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P1_gux_ttxux/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P1_uux_ttxg/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_gg_ttxgg/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.FALSE./
pp_tt012j.mad/SubProcesses/P2_gg_ttxuux/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.FALSE./
pp_tt012j.mad/SubProcesses/P2_gu_ttxgu/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_gux_ttxgux/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_uc_ttxuc/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_ucx_ttxucx/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_uu_ttxuu/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.FALSE./
pp_tt012j.mad/SubProcesses/P2_uux_ttxccx/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_uux_ttxgg/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_uux_ttxuux/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_uxcx_ttxuxcx/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.TRUE./
pp_tt012j.mad/SubProcesses/P2_uxux_ttxuxux/mirrorprocs.inc:      DATA (MIRRORPROCS(I),I=1,1)/.FALSE./

In particular:

there is no qq because quark-quark and quark-antiquark collisions are separate (so uu_XXXX and uux_XXXX are separate)
there are a few directories with MIRRORPROCS=TRUE, I checked only P2_uux_ttxuux and P2_uc_ttxuc and in both cases the helicity lists from the two mirrors match

So, bottomline: remember that in cudacpp generated code (including generated fortran) we do NOT mix uu_XXXX and uux_XXXX together, so if these two have separate helicities, this is handled because they are separate directories. Changing this would require a much more general handling of nprocesses>2, which so far we do not use.

Rephrasing: I think that adding this sanity check should be enough, and in cudacpp we can safely keep two helicity computations in fortran and only one in cudacpp. Do you agree?

Yes, this is what I meant by GPU grouping.

Now this makes me realize that they are steps that I do not like in the way those helicities are handle and I did create a new branch(https://github.com/mg5amcnlo/mg5amcnlo/tree/goodhel) to implement some change and to allow to experiment more on this without being affected by the issue spotted for tt~. I do agree that the above tt~ case is not relevant for this plugin.

Now I still need to investigate this since

the commit message of Johan and the status of the code at the time, clearly linked that this is possible for the mirror
I LIKE to solve puzzle and this is a quite intriguing/funny one.
This a good opportunity to clean the mess that goodhel is at madevent level anyway

The obviously side effect of this fortran refactoring is that your patching method will likely crash since the base file will be different.

Hi Olivier, thanks. Yes good to investigate puzzles :-)

Just one question, are you trying to completely remove the second helicity calculation? Or to understand whty it is needed and keep it there? I mean I'd like to understand if eventually we need to implement two helicities in cudacpp

First step is to understand if /when it is needed. And if the answer is "it is not needed" -> then we can remove it.

Cheers,

Olivier

On 31 Jul 2024, at 11:39, Andrea Valassi @.***> wrote:

Hi Olivier, thanks. Yes good to investigate puzzles :-)

— Reply to this email directly, view it on GitHubhttps://github.com/madgraph5/madgraph4gpu/issues/941#issuecomment-2260097021, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AH6535S2OM56FTHTBZ27NWDZPCWGRAVCNFSM6AAAAABLUTRRYCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRQGA4TOMBSGE. You are receiving this because you were mentioned.Message ID: @.***>

With the new goodhel branch, performing additional tests (first in the SM):

[x] p p > t t~ j j
[x] p p > t t~ j j j
[x] p p > w+ w+ j j , w+ > l+ vl
[x] p p > w+ w+ j j QCD=0 , w+ > l+ vl
[x] p p > w+ w+ j j j QCD=1 , w+ > l+ vl
[x] p p > w+ w- j j QCD<=4 , w+ > l+ vl
[x] p p > w+{T} w-{0} j j QCD<=4 , w+ > l+ vl, w- > j j
[x] p p > w+{T} w-{0} j j QCD=0 , w+ > l+ vl, w- > j j
[x] p p > w+{T} w-{0} j j QCD<=0

Running test with sm-ckm model (like CMS often use) then the crash was on the first process (that I did test):

[x] p p > w+{T} w-{0} j j QCD<=0

So I'm going to investigate that process now:

     Added good helicity           18 for process 63   4.0343177724261636E-008  in event           12 local:          12 mirror           2
detected in  G345:
     fail for process 63
              18 F T

So given that the limit to be "accepted is LIMHEL=1e-8" and that here we are (after 12 events) hitting the first event with only 4 times bigger than the limit... and that many other channel (like G1, does not include 18 for process 63 (in none of the imirror), I think that this is safe to say that this is a threshold effect. So yes here non identical list are occuring and are "fine" so here a simple check of identical list on the fortran side will just lead to an un-necessary crash. (and the result will be equivalent (up to single precision calculation).

To avoid such threshold check, I have updated the test to (the quite heavy now):

diff --git a/madgraph/iolibs/template_files/matrix_madevent_group_v4.inc b/madgraph/iolibs/template_files/matrix_madevent_group_v4.inc
index f3d0ade09..92ad6a419 100644
--- a/madgraph/iolibs/template_files/matrix_madevent_group_v4.inc
+++ b/madgraph/iolibs/template_files/matrix_madevent_group_v4.inc
@@ -35,6 +35,8 @@ c
 c global (due to reading writting)
 c
         LOGICAL GOODHEL(NCOMB,2, MAXSPROC)
+        LOGICAL GOODHEL_STRONG(NCOMB,2)
+        DATA GOODHEL_STRONG /THEL*.false./
            INTEGER NTRY(2, MAXSPROC)
         common/BLOCK_GOODHEL/NTRY,GOODHEL

@@ -137,6 +139,20 @@ C ----------
     IF ((ISHEL(IMIRROR).EQ.0.and.ISUM_HEL.eq.0).or.(DS_get_dim_status('Helicity').eq.0).or.(HEL_PICKED.eq.-1)) THEN
       DO I=1,NCOMB
          IF (GOODHEL(I,IMIRROR,%(proc_id)s) .OR. NTRY(IMIRROR,%(proc_id)s).LE.MAXTRIES.or.(ISUM_HEL.NE.0).or.THIS_NTRY(IMIRROR).le.10) THEN
+          IF (IMIRROR.eq.2.and.NTRY(1,%(proc_id)s).ge.MAXTRIES.and.NTRY(2,%(proc_id)s).ge.MAXTRIES) then
+             IF (GOODHEL(I,1,%(proc_id)s).neqv.GOODHEL(I,2,%(proc_id)s))THEN
+                IF(GOODHEL_STRONG(I,1).or.GOODHEL_STRONG(I,2)) then
+
+                 write(*,*) 'fail for process %(proc_id)s'
+                 DO J=1,NCOMB
+                    IF (GOODHEL(J,1,%(proc_id)s).neqv.GOODHEL(J,2,%(proc_id)s))THEN
+                        write(*,*) J, GOODHEL(J,1,%(proc_id)s), GOODHEL(J,2,%(proc_id)s)
+                    ENDIF
+                 ENDDO
+                 STOP 34
+                 ENDIF
+             ENDIF
+            ENDIF
             T=MATRIX%(proc_id)s(P ,NHEL(1,I),JC(1),I)
             %(beam_polarization)s
                        IF (ISUM_HEL.NE.0.and.DS_get_dim_status('Helicity').eq.0.and.ALLOW_HELICITY_GRID_ENTRIES) then
@@ -166,6 +182,9 @@ C ----------
         JHEL(IMIRROR) = 1
         IF(NTRY(IMIRROR,%(proc_id)s).LE.MAXTRIES.or.THIS_NTRY(IMIRROR).le.10)THEN
            DO I=1,NCOMB
+              IF (.NOT.GOODHEL_STRONG(I,IMIRROR) .AND. (DABS(TS(I)).GT.100d0*ANS*LIMHEL/NCOMB)) THEN
+                  GOODHEL_STRONG(I,IMIRROR)=.TRUE.
+              ENDIF
              IF(init_mode) THEN
                   IF (DABS(TS(I)).GT.ANS*LIMHEL/NCOMB) THEN
                      PRINT *, 'Matrix Element/Good Helicity: %(proc_id)s ', i, 'IMIRROR', IMIRROR

With this factor of "100", the above process still crashes for 4 G directory out of 183. With the factor of "1000", all of them pass the check at survey time ... but one crash at refine stage :-(

 fail for process 60
          18 T F
          22 T F
STOP 34

helicity 18: -1 -1 1 0 -1 -1 helicity 22: -1 -1 -1 0 -1 -1

cross-section of helicity 18+12 : 0.6 (contribution from process 60 only:5.718838519992133e-12) cross-section reported for sum over helicity: 0.29 (Careful not same normalization a factor of 4 is possible here between the two cross-section)

I will ignore the crash at the refine stage, since this might be an issue on how the grid is read/... While the spread in returned helicity seems quite large here (more than a factor 100), the "problematic" matrix-element/helicity are clearly not physically relevant so accepting such variance is here ok.

So conclusion for this crashing test are

one can not just put a sanity check that the list are the same
this case is working fine and we do not care if that helicity is included or not (efficiency is impacted obviously).

Continue checking for more processes (MSSM):

[x] p p > go go
[x] p p > go go j j
[x] p p > sq sq j j
[x] (sm) p p > l+ l- j j j
[x] (sm) p p > l+ vl j j j

So given that the limit to be "accepted is LIMHEL=1e-8" and that here we are (after 12 events) hitting the first event with only 4 times bigger than the limit... and that many other channel (like G1, does not include 18 for process 63 (in none of the imirror), I think that this is safe to say that this is a threshold effect. So yes here non identical list are occuring and are "fine" so here a simple check of identical list on the fortran side will just lead to an un-necessary crash. (and the result will be equivalent (up to single precision calculation).

Hi Olivier, thanks, nice finds :-)

One point: in the fortran generated through cudacpp, we use LIMHEL=0, since more than two years, see #419. So there is no sensistivity to threshold effect. We had discussed this extensively at the time.

For the release, I would keep this as it is (LIMHEL=0) and add the check that the two helicity lists are the same, to ensure that cudacpp and fortran tests give bit by bit the same results. From a software testing point of view, this is essential.

Later on, this can be rediscussed. One can even think of computing helicities in fortran and passing them to cudacpp.

That said, I still kind of favour having LIMHEL=0 in production full stop. I wonder how much performance we gain and in which specific processes by using LIMHEL>0. But definitely I do not have the full picture here.

For sm-ckm model, using limhel=0 can giveperformance hit of a factor of 4 or more ... So this is not negligeable. Obviously, I do not want to have a "normal" Fortran and a "cudacpp" fortran. You can set limhel to 0 for test if you want but in production this should be the same fortran code/design.

Cheers,

Olivier

On 2 Aug 2024, at 09:34, Andrea Valassi @.***> wrote:

Hi Olivier, thanks, nice finds :-)

One point: in the fortran generated through cudacpp, we use LIMHEL=0, since more than two years, see #419https://github.com/madgraph5/madgraph4gpu/issues/419. So there is no sensistivity to threshold effect. We had discussed this extensively at the time.

Later on, this can be rediscussed. One can even think of computing helicities in fortran and passing them to cudacpp.

— Reply to this email directly, view it on GitHubhttps://github.com/madgraph5/madgraph4gpu/issues/941#issuecomment-2264759455, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AH6535Q5DZVUBYPTD5VRFULZPMZCBAVCNFSM6AAAAABLUTRRYCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRUG42TSNBVGU. You are receiving this because you were mentioned.Message ID: @.***>

So I guess that we can close this issue with the conclusion: Yes it can be removed. This will be done independently of the cudacpp plugin, I will follow the normal procedure for including such fix in 3.6.0. Which means that this associated issue will automatically be fixed when the merging with 3.6.0 will be complete. (which is our next big task anyway).

So I guess that we can close this issue with the conclusion: Yes it can be removed. This will be done independently of the cudacpp plugin, I will follow the normal procedure for including such fix in 3.6.0

Thanks a lot Olivier :-)

In the meantime while waiting for 360 etc, can we then merge #935 which adds the second reset cumulative call? Then of course this will need to be removed in 360 (I can do that at that moment, no problem), but at least in the meantime while waiting for 360 we have some internally consistent code where fortran and cudacpp are expected to give the same xsec bit by bit. I prefer to add a fix now that we need to remove later, rather than do nothing and have tests that keep failing. Esepecially because I want to test the CMS xsec issues and I need that bit by bit equality.

You can set limhel to 0 for test if you want but in production this should be the same fortran code/design.

Note, LIMHEL=0 is now in cudacpp production (NOT just in the madgrah4gpu repo). I am ok to remove that and move it ONLY to the madgraph4gpu repo, but can we at least add a runcard that sets limhel=0? (Or maybe it exists?) I think that would be very useful for debugging.

madgraph5 / madgraph4gpu

Check if the second helicity filtering in fortran for mirror processes can be removed #941

REPORT for P1_qq_ttxqq :