master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp #121

valassi · 2024-07-19T12:36:08Z

Hi @oliviermattelaer this is a WIP MR into gpucpp of my nb_warp_used and of everything else I am using against master_june24 in my (madgraph5/madgraph4gpu#882).

The nb_warp_used fixes are described here madgraph5/madgraph4gpu#885 (comment)

NB: as discussed in madgraph5/madgraph4gpu#886, gpucpp_june24, with or without my changes, does NOT work for me with master_june24 (my modified version, but I am pretty sure that the problem was pre-existing). This is why here I am NOT merging gpucpp_june24, I am only merging a SUBSET of that (what was originally used in master_june24, plus my changes, see https://github.com/valassi/mg5amcnlo/commits/valassi_gpucpp_june24/

I keep this in WIP for the moment. I would only merge this when we agree to merge my madgraph5/madgraph4gpu#882 into master (as master is meant to use the head of gpucpp... I do not want to 'pollute' gpucpp before!).

I also filed PR #120 against gpucpp_june24, but as mentioned in madgraph5/madgraph4gpu#886, I am unable to use gpucpp_june24 and this is NOT what was used in master_june24 anyway.

Thanks Andrea

…n fortran)

…section (and wrap not correctly ported everywhere

…uto_dsigX.f

…F call and MLM weight)

…ake--

…fortran exporter

…pucpp_june24 (eef200f) This is needed to allow the merge of the following two branches of madgraph4gpu: - master 6ba6bae30 (which uses commit 1e2aa4b i.e. the head of origin/gpucpp) - master_june24 309a48654 (which uses eef200f that is currently a standalone commit)

…or icolamp mg5amcnlo#118) into valassi_gpucpp_june24

…mcnlo#107 and coloramps mg5amcnlo#118) into valassi_gpucpp_june24

…ph5/madgraph4gpu#765

mg5amcnlo#125) into valassi_gpucpp_june24

valassi · 2024-08-21T17:53:52Z

Hi @oliviermattelaer this is a preliminary step for merging the master_june24 work of madgraph5/madgraph4gpu#882

Can you please review and approve?

Note, this is related to madgraph5/madgraph4gpu#886: in practice, what is here my valassi_gpucpp_june24 gets merged into gpucpp, while we forget about gpucpp_june24.

Thanks!
Andrea

valassi · 2024-08-21T18:51:59Z

PS @oliviermattelaer PLEASE DO NOT SQUASH in case you merge this, thanks :-)

…g5amcnlo#132) into valassi_gpucpp_june24 Fix conflicts: madgraph/various/banner.py (keep a debug printout in get_value_from_include)

valassi · 2024-09-01T20:21:15Z

Hi @oliviermattelaer I just merged th elatest gpucpp (including your gpucpp_fixci #132) into my valassi_gpucpp_june24.

I will update also my madgraph5/madgraph4gpu#882 in madgraph4gpu to use this one.

(Then I will probably comment on your madgraph5/madgraph4gpu#981 that 981 becomes irrelevant)

valassi · 2024-09-01T21:12:28Z

Hi @oliviermattelaer this now includes everything we need for june24. Can you have a look and approve? Thanks
Andrea
PS Again please do not squash this...

oliviermattelaer · 2024-09-02T11:23:04Z

Looks like they are conflict to fix here.
Additionally don't you want to use nb_warp_used? in the file dsample.f
Or do you prefere anoter PR for that?

Olivier

valassi · 2024-09-02T11:39:32Z

madgraph/iolibs/export_v4.py

@@ -1738,7 +1738,25 @@ def get_pdf_lines(self, matrix_element, ninitial, subproc_group = False, vector=
        if vector:
            pdf_definition_lines_vec = ""
            pdf_data_lines_vec = ""
-            pdf_lines = " DO iVEC=1,VECSIZE_USED\n"
+            pdf_lines = """ NB_WARP_USED = VECSIZE_USED / WARP_SIZE


Hi @oliviermattelaer about your question: I am doing nb_warp_used in various places. This is one example here.

But you are right that there may be more. (See madgraph5/madgraph4gpu#983)

Ok I will check dsample.f more in detail...

valassi · 2024-09-02T11:40:55Z

Hi Olivier thanks

Looks like they are conflict to fix here.

Hm? No I do not see conflicts, which conflicts are you talking about?

Additionally don't you want to use nb_warp_used? in the file dsample.f
Or do you prefere anoter PR for that?

No I wanted to fix this here.

I am fixing it in various places.
See https://github.com/mg5amcnlo/mg5amcnlo/pull/121/files#r1740773122

I thought that this was enough because tests were passing.
But you are right that I may be missing some and that these go undetected.
Ok then lets put this back to WIP :-(

Andrea

oliviermattelaer · 2024-09-02T11:58:12Z

Ok looks like this is another subtelty of github, that I had to understand... But this is only because the button was on "rebase" if I move it to "merge" then github does not complain about conflict anymore ...
Sorry for the confusion

valassi · 2024-09-02T15:06:31Z

Additionally don't you want to use nb_warp_used? in the file dsample.f

Hi @oliviermattelaer again. I had a look at this, see the comments in madgraph5/madgraph4gpu#983. In my opinion there is nothing else to be done in dsample.f, but I may be missing something. Can you have a look yourself and comment in 983 please?

Then if we agree that there is nothing else to do , I think that we can start merging this.

Thanks Andrea

valassi · 2024-09-03T08:55:58Z

Hi @oliviermattelaer again. I had a look at this, see the comments in madgraph5/madgraph4gpu#983. In my opinion there is nothing else to be done in dsample.f, but I may be missing something. Can you have a look yourself and comment in 983 please?

Hi @oliviermattelaer thanks for having a look! You confirmed here madgraph5/madgraph4gpu#983 (comment) that there is nothing to do for 983, so I consider that this is good to go.

So, I AM MERGING THIS! Starting the full june24 merge into master+gpucpp...

… into ghav_gpucpp_june24 NB: at this point, the remaining changes in this branch w.r.t. gpucpp are only a few commits by Olivier (which may be elsewhere like360) git log --oneline gpucpp..HEAD 718f59a0e (HEAD -> ghav_gpucpp_june24) Merge branch 'gpucpp' (including june24 nb_warp_used via mg5amcnlo#121) into ghav_gpucpp_june24 268cee3 (ghav/ghav_gpucpp_june24) Merge branch 'valassi_gpucpp_june24' into ghav_gpucpp_june24 def7fa6 sync gpuccp_june24 with gpucpp 37d1871 fix change in cpp template within the test (spaces) 0e5e483 fixing pythia8 iotest 849194f fix some tests (template changed) 1a9e6dd fixing an issue in the fortran side if no vector size e2e7900 add a possibility to have multiple type of factorization (no actual code change at this stage) 942dab7 secure the code 45f8410 (origin/gpucpp_fix_scan_vector) Merge branch 'gpucpp_june24' into gpucpp_fix_scan_vector 0959edf Merge branch 'gpucpp_wrap' into gpucpp_june24 63d2d39 (origin/gpucpp_wrap) remove useless file + debug statement ecb0a03 add gpu in the autocompletion for group_subproceses 82f078b change makefile to recompile code as needed if vector size is changed

oliviermattelaer and others added 28 commits September 14, 2023 10:18

change interface to support the nb_wrap method (not implemented withi…

1d20bfd

…n fortran)

select grouping updated --and dsample-- but imirror/... still scalar

825dc1d

dirty commit implementing wrap control: code compile but wrong cross-…

abec832

…section (and wrap not correctly ported everywhere

fix issue that wrong channel was picked

b5ebd1b

comment line that should not be used

6c75cf3

allow infrastructure to avoid the plugin to need to use a patch for a…

488fc48

…uto_dsigX.f

set IB identity such that this is correctly set for each wrap (for PD…

45d14ed

…F call and MLM weight)

set a list of channel

7d45dda

merge with gpucpp

e14c9df

set IB identity such that this is correctly set for each wrap (for PD…

663473a

…F call and MLM weight)

fix conflicts

d4e3ee9

Merge branch 'gpucpp' into gpucpp_wrap

f0ad8ca

wrap -> warp

115cf97

move to 3.5.3

4a59e13

move back the symlink to status for Stefan/andrea --committed by mist…

9fd8122

…ake--

Merge branch 'gpucpp' into gpucpp_wrap

602a888

rename plugin directory

514895b

fixing some unittest issue

4355881

update py8 test

8a8622b

allow for second exporter to have access to all variable used in the …

10378b3

…fortran exporter

Merge branch 'gpucpp' into gpucpp_wrap

57e617c

Merge remote-tracking branch 'upstream/gpucpp_826' into gpucpp_wrap

eef200f

remove one useless? call to coloramp

493fe45

refactoring part of the writing of color (is_LC)

5a241ef

Merge remote-tracking branch 'origin/datastructure_coloramp' (fixes f…

434b4a7

…or icolamp mg5amcnlo#118) into valassi_gpucpp_june24

Merge remote-tracking branch 'origin/gpucpp' (including gridpack mg5a…

6c7fda8

…mcnlo#107 and coloramps mg5amcnlo#118) into valassi_gpucpp_june24

[june24] implement nb_warp_used madgraph5/madgraph4gpu#885 and madgra…

2f4c467

…ph5/madgraph4gpu#765

valassi requested a review from oliviermattelaer July 19, 2024 12:36

valassi self-assigned this Jul 19, 2024

valassi marked this pull request as draft July 19, 2024 12:36

This was referenced Jul 19, 2024

URGENT - master_june24 does not use gpucpp_june24 (and upgrading fails codegen) madgraph5/madgraph4gpu#886

Closed

master_june24 tmad tests crashes in dsig1_vec (CUDACPP_RUNTIME_VECSIZEUSED not correctly propagated?) madgraph5/madgraph4gpu#885

Closed

Merge remote-tracking branch 'origin/gpucpp' (including generate_events

f0b4299

mg5amcnlo#125) into valassi_gpucpp_june24

valassi marked this pull request as ready for review August 21, 2024 17:49

valassi changed the title ~~WIP: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp~~ master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp Aug 21, 2024

valassi mentioned this pull request Aug 21, 2024

Merge of master into master_june24 and channelid fixes/reimplementation madgraph5/madgraph4gpu#882

Merged

valassi changed the title ~~master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp~~ URGENT master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp Aug 21, 2024

Merge remote-tracking branch 'origin/gpucpp' (including gpucpp_fixci m…

4ef15ca

…g5amcnlo#132) into valassi_gpucpp_june24 Fix conflicts: madgraph/various/banner.py (keep a debug printout in get_value_from_include)

valassi commented Sep 2, 2024

View reviewed changes

valassi changed the title ~~URGENT master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp~~ WIP master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp Sep 2, 2024

valassi changed the title ~~WIP master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp~~ master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp Sep 2, 2024

valassi merged commit a696c93 into mg5amcnlo:gpucpp Sep 3, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp #121

master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp #121

valassi commented Jul 19, 2024

valassi commented Aug 21, 2024

valassi commented Aug 21, 2024

valassi commented Sep 1, 2024

valassi commented Sep 1, 2024 •

edited

Loading

oliviermattelaer commented Sep 2, 2024

valassi Sep 2, 2024

valassi commented Sep 2, 2024

oliviermattelaer commented Sep 2, 2024

valassi commented Sep 2, 2024 •

edited

Loading

valassi commented Sep 3, 2024

master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp #121

master_june24 completion: nb_warp_used (and a SUBset of gpucpp_june24) into gpucpp #121

Conversation

valassi commented Jul 19, 2024

valassi commented Aug 21, 2024

valassi commented Aug 21, 2024

valassi commented Sep 1, 2024

valassi commented Sep 1, 2024 • edited Loading

oliviermattelaer commented Sep 2, 2024

valassi Sep 2, 2024

Choose a reason for hiding this comment

valassi commented Sep 2, 2024

oliviermattelaer commented Sep 2, 2024

valassi commented Sep 2, 2024 • edited Loading

valassi commented Sep 3, 2024

valassi commented Sep 1, 2024 •

edited

Loading

valassi commented Sep 2, 2024 •

edited

Loading