Move LIMHEL=0 to madgraph4gpu only and add a runcard allowing LIMHEL=0

[valassi]

Hi @oliviermattelaer this is a followup to #941 and PR #935. Two points

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

(1) In #941 (comment) you suggest that LIMHEL=0 should not be the default in production cudacpp. Now it is. So it should be changed eventually. I still want to use LIMHEL=0 for all my tests in madgraph4gpu, so I can move that to generateAndCompare.sh for the code generation in the madgraph4gpu repo.

(2) However, I still think that LIMHEL=0 is a useful setting for tests, also for users, when debugging issues (eg the CMS xsec difference). So I suggest adding it as a hidden parameter in the runcards. Objections or other suggestions?

Thanks
Andrea

[oliviermattelaer]

Yes make sense to allow to control such parameter via the run_card. Will look at that Sent from Outlook for iOS<https://aka.ms/o0ukef>

…

________________________________ From: Andrea Valassi ***@***.***> Sent: Friday, August 2, 2024 6:34:45 PM To: madgraph5/madgraph4gpu ***@***.***> Cc: Olivier Mattelaer ***@***.***>; Mention ***@***.***> Subject: [madgraph5/madgraph4gpu] Move LIMHEL=0 to madgraph4gpu only and add a runcard allowing LIMHEL=0 (Issue #950) Hi @oliviermattelaer<https://github.com/oliviermattelaer> this is a followup to #941<#941> and PR #935<#935>. Two points You can set limhel to 0 for test if you want but in production this should be the same fortran code/design. (1) In #941 (comment)<#941 (comment)> you suggest that LIMHEL=0 should not be the default in production cudacpp. Now it is. So it should be changed eventually. I still want to use LIMHEL=0 for all my tests in madgraph4gpu, so I can move that to generateAndCompare.sh for the code generation in the madgraph4gpu repo. (2) However, I still think that LIMHEL=0 is a useful setting for tests, also for users, when debugging issues (eg the CMS xsec difference). So I suggest adding it as a hidden parameter in the runcards. Objections or other suggestions? Thanks Andrea — Reply to this email directly, view it on GitHub<#950>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AH6535WGQDWC2HJLS5GQVLDZPOYKLAVCNFSM6AAAAABL43ICDCVHI2DSMVQWIX3LMV43ASLTON2WKOZSGQ2DKMZWG4ZTEMA>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[oliviermattelaer]

I'm confused: limhel is not set to zero for madevent (at least for the gpucpp branch).
https://github.com/mg5amcnlo/mg5amcnlo/blob/8c70b1ba4c0f01257d8aa2d337fff716f1ef7e55/Template/LO/Source/genps.inc#L33

So either you are confusing with the standalone mode or we are not understanding each other.
But anyway can make sense to control limhel from the run_card

Olivier

[valassi]

I'm confused: limhel is not set to zero for madevent (at least for the gpucpp branch).

Hi Olivier, LIMHEL=0 is in the patch

madgraph4gpu/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/MG5aMC_patches/PROD/patch.common

Line 11 in 2f76535

+ PARAMETER(LIMHEL=0) ! ME threshold for helicity filtering (force Fortran to mimic cudacpp, see #419)

The fact that LIMHEL=0 for madeven in cudacpp was discussed many times long ago, Examples

• Allow LIMHEL>0 in cudacpp (using the exact same algorithm as in fortran) #564 Allow LIMHEL>0 in cudacpp (using the exact same algorithm as in fortran)
• Also related xsec from fortran and cpp differ in gg_uu tmad tests (bug in getGoodHel implementation) #630, PR Fix cudacpp getGoodHel to define a good helicity as ME>0 (fix mismatch in cross sections for fortran and cpp in gg_uu) #705, Physics results are different in fortran and cudacpp MEs (by factors or orders of magnitude) #417, Debug fortran vs cpp differences in multichannel (using SD1 to bypass "get_channel_cut" issues and LIMHEL=0 to bypass helicity filtering issues) #419

But anyway can make sense to control limhel from the run_card

Very good :-)

So lets keep this on the todo list

Andrea

[oliviermattelaer]

Ok, I will include this control on limhel on the goodhel branch targetting 3.6.0. Olivier On 5 Aug 2024, at 09:28, Andrea Valassi ***@***.***> wrote: I'm confused: limhel is not set to zero for madevent (at least for the gpucpp branch). <https://github.com/mg5amcnlo/mg5amcnlo/blob/8c70b1ba4c0f01257d8aa2d337fff716f1ef7e55/Template/LO/Source/genps.inc#L33> Hi Olivier, LIMHEL=0 is in the patch https://github.com/madgraph5/madgraph4gpu/blob/2f76535c41e3efcd065d31d36e356b90fdb9b554/epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/MG5aMC_patches/PROD/patch.common#L11 The fact that LIMHEL=0 for madeven in cudacpp was discussed many times long ago, Examples * #564<#564> Allow LIMHEL>0 in cudacpp (using the exact same algorithm as in fortran) * Also related #630<#630>, PR #705<#705>, #417<#417>, #419<#419> But anyway can make sense to control limhel from the run_card Very good :-) So lets keep this on the todo list Andrea — Reply to this email directly, view it on GitHub<#950 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AH6535TZGCXM5TM4OG3U3B3ZP4SSZAVCNFSM6AAAAABL43ICDCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRYGM3DKNRZGI>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[oliviermattelaer]

limhel is now controllable within 3.6.0 branch.
Now we need to put back some support on the cudacpp side.

[valassi]

Hi Olivier, thanks a lot

limhel is now controllable within 3.6.0 branch.

Yes this will be merged in #955 (also #986)

Now we need to put back some support on the cudacpp side.

I have moved this to #564 (no need to add again #988, I have now closed that).

What you have done already complets this part ie #950

• we can set limhel in the runcard for fortran
• for cudacpp, tyhe setting of the runcard hardcodes limhel=0 (see df130b0)

Closing this, thanks to Olivier

[valassi]

Reopen and mark that this is actually closed by #992 (which includes #986 which includes #955)