partial fix for EFT processes after merging alphas (only workd with HRDCOD=1 for now) by valassi · Pull Request #446 · madgraph5/madgraph4gpu

valassi · 2022-05-06T17:57:08Z

This is a WIP PR for #439.

Now it is possible to build heftggh after the alphas merge, but ONLY in the hardcoded HRDCOD=1 mode.

The problem with the non hardcoded mode is that several EFT parameters like mdl_MH__exp__6 which are needed in the calculation of couplings from alphas are not available in the G2COUP kernel. Various options

force the use of HRDCOD=1 (this works, because the parameters are constexpr in a namespace)
compute the Gs on the host rather than on the device (just get mdl_MH__exp__6 from the singleton Parameters class on the host)
(probably the best, but most complex) copy frm host to device the parameters like mdl_MH__exp__6 from the singleton Parameters class

Note that the last one has some overlap with what we do with cIPC/cIPD. Maybe easiest to generalize the cIPC/cIPD to all parameters? Overkill for the other cases? In any case the amount of global memory (or even constant memory) these would occupy is very little.

…ted with c0c2768 (BEFORE THE ALPHAS PR) This is the bginning of a patch to analyse and fix madgraph5#439

…is will be manually-developed code

…, already there in latest codegen)

…oat/hrd0 or double/hrd1 both fail...

…smpl<double> operator)

This fixes the cuda build in 'make HRDCOD=1' ../../src/Parameters_heft.h(116): error: expression must have a constant value ../../src/Parameters_heft.h(116): note #2703-D: cannot call non-constexpr function "pow(double, double) throw()" /usr/local/cuda-11.6/include/crt/math_functions.h(4990): here Note that g++ was succeeding (pow being constexpr is a gcc extension)

…r_pow

…_sqrt

…exponent

…as all ok) Revert "[heft] TMP move over codegen c0c2768 the three recent backports" This reverts commit ed65cda. Revert "[heft] TMP move codegen back to c0c2768" This reverts commit 4bb80a2.

…to to manu

…end of the Parameters.h file

…ixes the build (ONLY for HRDCOD=1) madgraph5#439 The build without HRDCOD=1 however i sstill failing

…aph5#439

valassi · 2022-05-06T18:01:50Z

In a way the question is, what determines what is in self.parms2order and self.coup2order? These are the only pieces of the Parameters.h file that are presently copied to constant memory...

… build was failing)

WARNING! eemumu runtest.exe fails... mdl_complexi is wrong

…but is is only 8 events!)

…complexi)

valassi · 2022-05-09T13:34:17Z

I have hopefully completed a PARTIAL patch, which indeed only works for EFTs with hardcoded parameters. The other options remain open.

It took quite some time. There are many corners that needed secial fixes. Examples

parameters are double but ME precision can be float, there is a psecial handling if using SIMD because cannot multiply a scalar double and a float vector
in hardcoded mode I use cxsmpl but the ME calculations need cxtype and some casts are needed
a constexpr cxsmpl cannot be used in device mode
etc etc

Now all issues should be ok both for heftggh and the standard 5 processes.

I am running a ful set of tests, and I will merge this. Then the other options I will file as a separate non urgent followup issue.

…ll ok STARTED AT Mon May 9 15:35:15 CEST 2022 ENDED(1) AT Mon May 9 18:07:17 CEST 2022 ENDED(2) AT Mon May 9 18:42:28 CEST 2022 ENDED(3) AT Mon May 9 18:46:55 CEST 2022 ENDED(4) AT Mon May 9 18:50:03 CEST 2022 ENDED(5) AT Mon May 9 18:53:06 CEST 2022

…ogs that are not redone in allTees.sh

…heftggh only builds ok with HRDCOD=1) OK: all logs are now from the latest code, check using grep -h DATE tput/logs_*/*txt | sort

valassi · 2022-05-09T17:57:53Z

All checks have completed. I am merging this.

This is only a partial patch, but it is fully functional. I have stripped off a few more things to do in #448, specifically
#448 (comment)

valassi added 22 commits May 6, 2022 15:32

[heft] add heft_gg_h to repo, copied as-is from heft_gg_h.auto genera…

4ed001c

…ted with c0c2768 (BEFORE THE ALPHAS PR) This is the bginning of a patch to analyse and fix madgraph5#439

[heft] remove CODEGEN_cudacpp_heft_gg_h_log.txt to make it clearer th…

9e928a8

…is will be manually-developed code

[heft] fix src/Makefile in older generated code (replace _sm by _heft…

ddd4e1a

…, already there in latest codegen)

[heft] add -heftggh option to the tput scripts

6c8b955

[heft] add double/hrd0 lohs for heftggh, all looks ok - NB however fl…

a35ebf1

…oat/hrd0 or double/hrd1 both fail...

[heft] fix "make FPTYPE=f AVX=none" build for heftggh (add float * cx…

dafe2af

…smpl<double> operator)

[heft] add log for heftggh testst with float/hrd0, now all ok

048918f

[heft] rename sqrtNR as constexpr_sqrt to be consistent with constexp…

f0cf4a7

…r_pow

[heft] backport to codegen the changes in constexpr_pow and constexpr…

b853749

…_sqrt

[heft] more useful version of constexpr_pow, taking a floating point …

6a2df46

…exponent

[heft] backport improved constexpr_pow(double,double) to codegen

5e42556

[heft] backport to codegen also the missing operator*

26f7884

[heft] TMP move codegen back to c0c2768

4bb80a2

[heft] TMP move over codegen c0c2768 the three recent backports

ed65cda

[heft] revert the two TMP changes (I was just checking that codegen w…

605dcc6

…as all ok) Revert "[heft] TMP move over codegen c0c2768 the three recent backports" This reverts commit ed65cda. Revert "[heft] TMP move codegen back to c0c2768" This reverts commit 4bb80a2.

[heft] regenerate heftggh with the latest codegen and move it from au…

86f0716

…to to manu

[heft] move Parameters_heft_[in]dependentCouplings and G2COUP to the …

54b8e27

…end of the Parameters.h file

[heft] add 'using namespace Parameters_heft;' for HRDCOD=1: finally f…

3b0b288

…ixes the build (ONLY for HRDCOD=1) madgraph5#439 The build without HRDCOD=1 however i sstill failing

[heft] backport to codegen the (partial, HRDCOD=1 only) fix for madgr…

4bec566

…aph5#439

[heft] regenerate all 5 processes auto and resync manual

8d59000

[heft] regenerate also eemumu.mad ggtt.mad and ggtt.madonly

d6319bb

valassi self-assigned this May 6, 2022

valassi marked this pull request as draft May 6, 2022 17:57

valassi added 5 commits May 6, 2022 20:10

[heft] fix the operator* for mixing double and float (eemumu FPTYPE=f…

814cbf3

… build was failing)

[heft] backport to codegen the fix in operator*

df54cc3

[heft] regenerate eemumu.mad ggtt.mad and ggtt.madonly

e8c79ce

[heft] port fix in operator* to heftggh

caf3e29

[heft] regenerate 5 processes auto and resync manual

1440bf8

valassi added 15 commits May 9, 2022 12:38

[heft] minor imprvements in tput allTees script

d073d7f

[heft] add an extra error for non-SM processes

4dac791

[heft] in codegen add an error in EFT unless HRDCOD=1 is used

2d507ef

[heft] remove 'FIXME' comment about mdl_complexi and clean up the code

c57ec48

[heft] backport to codegen of Parameters.cc

2ddc2d0

[heft] cosmetics in codegen

6c732c7

[heft] backport to codegen of Parameters.h with mdl_complexi changes

3f337ec

[heft] regenerate 5 processes auto and resync manu (fix mdl_complexi)

09b7f53

WARNING! eemumu runtest.exe fails... mdl_complexi is wrong

[heft] fix eemumu manual (go back to old mdl_complexi in many respects)

bd55f84

[heft] fix codegen and eemumu auto/manu for Parameters_sm.h

28cc805

[heft] fix codegen and eemumu auto for Parameters_sm.cc

d38ca9e

[heft] regenerate heftggh auto and move it to manual

579c6a9

[heft] rerun heftggh flat/hrd1, all looks ok (performance fluctuates …

f752450

…but is is only 8 events!)

[heft] regenerate 5 processes auto and resync manu (finally? fix mdl_…

205c0c7

…complexi)

[heft] regenerate also eemumu.mad ggtt.mad and ggtt.madonly

5efffda

valassi force-pushed the heft branch from 95a8f0c to 5efffda Compare May 9, 2022 13:28

[heft] rerun ggtt d/f hrd0/1, all seems ok

03b593b

This was referenced May 9, 2022

Fix heft generation after alphas PR: more fundamentally, remove reading of param_card.dat? #439

Closed

Improve handling of model parameters #448

Open

valassi added 3 commits May 9, 2022 19:24

[heft] cleanup: remove tput/logs_ggttg_manu/log_ggttg_manu_*_inl1_* l…

8a44ede

…ogs that are not redone in allTees.sh

[heft] ** COMPLETE HEFT ** cleanup: remove hrd0 logs for heftggh (as …

c112e7d

…heftggh only builds ok with HRDCOD=1) OK: all logs are now from the latest code, check using grep -h DATE tput/logs_*/*txt | sort

valassi marked this pull request as ready for review May 9, 2022 17:56

valassi changed the title ~~WIP: partial fix for EFT processes after merging alphas (only workd with HRDCOD=1 for now)~~ partial fix for EFT processes after merging alphas (only workd with HRDCOD=1 for now) May 9, 2022

valassi merged commit e4daf96 into madgraph5:master May 9, 2022

This was referenced May 10, 2022

Add clang and icx logs + FIX SUMMARY TABLES #449

Merged

Additional fixes for EFT gg->h in clang #450

Merged

valassi mentioned this pull request Mar 29, 2023

Support for non-SM UFO models: HRDCOD=1 build fails in model SMEFTsim_topU3l_MwScheme_UFO #614

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

partial fix for EFT processes after merging alphas (only workd with HRDCOD=1 for now)#446

partial fix for EFT processes after merging alphas (only workd with HRDCOD=1 for now)#446
valassi merged 74 commits intomadgraph5:masterfrom
valassi:heft

valassi commented May 6, 2022

Uh oh!

valassi commented May 6, 2022

Uh oh!

valassi commented May 9, 2022

Uh oh!

valassi commented May 9, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

valassi commented May 6, 2022

Uh oh!

valassi commented May 6, 2022

Uh oh!

valassi commented May 9, 2022

Uh oh!

valassi commented May 9, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant