Compilation error with complex NLO processes in 2.2.2

[Josh Bendavid]

Trying W+0,1,2 jets in 2.2.2 leads to a compilation error.

Process card like:

import model loop_sm-no_b_mass
define p = p b b~
define j = j b b~
generate p p > w- [QCD] @0
add process p p > w- j [QCD] @1
add process p p > w- j j [QCD] @2

Leads to errors during compilation like below (full log attached as well):

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

Backtrace for this error:
#0 0x2B5D7ECCD3C7
#1 0x2B5D7ECCD9CE
#2 0x2B5D7F0F769F
#3 0x2B5D7E92804A
#4 0x4BDB35
#5 0x4BE33C
#6 0x4BF67B
#7 0x4BCCC1
#8 0x4B0852
#9 0x40F945
#10 0x412124
#11 0x2B5D7F0E3D5C
#12 0x40F598

This same process works fine with only the 0 and 1 jet contributions, and it also worked fine with 0,1,2 jets in 2.2.1.

[marco zaro (marco-zaro)]

Ciao Josh,
do I understand correctly that the only parameter that you change is ickkw in the run_card, which is set to 3 (apart to lhapdf which should not be relevant for this bug?
2.2.3 on a mac seems to work ok.
Cheers,

Marco

[marco zaro (marco-zaro)]

Ciao Josh,
I have also tried 2.2.2 on a mac and 2.2.3 on linux (Louvain cluster) and everything looks ok (no segfault)…
Can you try 2.2.3?
It is still a dev branch, you can get it from here
https://code.launchpad.net/~maddevelopers/mg5amcnlo/2.2.3

Cheers,

Marco

On 28 Dec 2014, at 11:47, marco zaro <email address hidden> wrote:

> Ciao Josh,
> do I understand correctly that the only parameter that you change is ickkw in the run_card, which is set to 3 (apart to lhapdf which should not be relevant for this bug?
> 2.2.3 on a mac seems to work ok.
> Cheers,
>
> Marco
>
> --
> You received this bug notification because you are a bug assignee.
> https://bugs.launchpad.net/bugs/1406000
>
> Title:
> Compilation error with complex NLO processes in 2.2.2
>
> Status in MadGraph5_aMC@NLO Generator:
> New
>
> Bug description:
> Trying W+0,1,2 jets in 2.2.2 leads to a compilation error.
>
> Process card like:
>
> import model loop_sm-no_b_mass
> define p = p b b~
> define j = j b b~
> generate p p > w- [QCD] @0
> add process p p > w- j [QCD] @1
> add process p p > w- j j [QCD] @2
>
> Leads to errors during compilation like below (full log attached as
> well):
>
> Program received signal SIGSEGV: Segmentation fault - invalid memory
> reference.
>
> Backtrace for this error:
> #0 0x2B5D7ECCD3C7
> #1 0x2B5D7ECCD9CE
> #2 0x2B5D7F0F769F
> #3 0x2B5D7E92804A
> #4 0x4BDB35
> #5 0x4BE33C
> #6 0x4BF67B
> #7 0x4BCCC1
> #8 0x4B0852
> #9 0x40F945
> #10 0x412124
> #11 0x2B5D7F0E3D5C
> #12 0x40F598
>
> This same process works fine with only the 0 and 1 jet contributions,
> and it also worked fine with 0,1,2 jets in 2.2.1.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/mg5amcnlo/+bug/1406000/+subscriptions

[Josh Bendavid (joshbendavid)]

Yes only ickkw was changed (to 3), plus LHAPDF.

I'm running with 2.2.2 on Linux. Will try 2.2.3

Sent from my Android.
On 28 Dec 2014 06:25, "marco zaro" <email address hidden> wrote:

> Ciao Josh,
> I have also tried 2.2.2 on a mac and 2.2.3 on linux (Louvain cluster) and
> everything looks ok (no segfault)…
> Can you try 2.2.3?
> It is still a dev branch, you can get it from here
> https://code.launchpad.net/~maddevelopers/mg5amcnlo/2.2.3
>
> Cheers,
>
> Marco
>
>
> On 28 Dec 2014, at 11:47, marco zaro <email address hidden> wrote:
>
> > Ciao Josh,
> > do I understand correctly that the only parameter that you change is
> ickkw in the run_card, which is set to 3 (apart to lhapdf which should not
> be relevant for this bug?
> > 2.2.3 on a mac seems to work ok.
> > Cheers,
> >
> > Marco
> >
> > --
> > You received this bug notification because you are a bug assignee.
> > https://bugs.launchpad.net/bugs/1406000
> >
> > Title:
> > Compilation error with complex NLO processes in 2.2.2
> >
> > Status in MadGraph5_aMC@NLO Generator:
> > New
> >
> > Bug description:
> > Trying W+0,1,2 jets in 2.2.2 leads to a compilation error.
> >
> > Process card like:
> >
> > import model loop_sm-no_b_mass
> > define p = p b b~
> > define j = j b b~
> > generate p p > w- [QCD] @0
> > add process p p > w- j [QCD] @1
> > add process p p > w- j j [QCD] @2
> >
> > Leads to errors during compilation like below (full log attached as
> > well):
> >
> > Program received signal SIGSEGV: Segmentation fault - invalid memory
> > reference.
> >
> > Backtrace for this error:
> > #0 0x2B5D7ECCD3C7
> > #1 0x2B5D7ECCD9CE
> > #2 0x2B5D7F0F769F
> > #3 0x2B5D7E92804A
> > #4 0x4BDB35
> > #5 0x4BE33C
> > #6 0x4BF67B
> > #7 0x4BCCC1
> > #8 0x4B0852
> > #9 0x40F945
> > #10 0x412124
> > #11 0x2B5D7F0E3D5C
> > #12 0x40F598
> >
> > This same process works fine with only the 0 and 1 jet contributions,
> > and it also worked fine with 0,1,2 jets in 2.2.1.
> >
> > To manage notifications about this bug go to:
> > https://bugs.launchpad.net/mg5amcnlo/+bug/1406000/+subscriptions
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1406000
>
> Title:
> Compilation error with complex NLO processes in 2.2.2
>
> Status in MadGraph5_aMC@NLO Generator:
> New
>
> Bug description:
> Trying W+0,1,2 jets in 2.2.2 leads to a compilation error.
>
> Process card like:
>
> import model loop_sm-no_b_mass
> define p = p b b~
> define j = j b b~
> generate p p > w- [QCD] @0
> add process p p > w- j [QCD] @1
> add process p p > w- j j [QCD] @2
>
> Leads to errors during compilation like below (full log attached as
> well):
>
> Program received signal SIGSEGV: Segmentation fault - invalid memory
> reference.
>
> Backtrace for this error:
> #0 0x2B5D7ECCD3C7
> #1 0x2B5D7ECCD9CE
> #2 0x2B5D7F0F769F
> #3 0x2B5D7E92804A
> #4 0x4BDB35
> #5 0x4BE33C
> #6 0x4BF67B
> #7 0x4BCCC1
> #8 0x4B0852
> #9 0x40F945
> #10 0x412124
> #11 0x2B5D7F0E3D5C
> #12 0x40F598
>
> This same process works fine with only the 0 and 1 jet contribu...

[Josh Bendavid (joshbendavid)]

Hi Marco,
I've done some further tests on linux with 2.2.2.

Everything is indeed ok as long as I don't use lhapdf. As soon as I run with lhapdf I get the errors as reported in the initial post. (I'm running with lhapdf 6.1.4)

[marco zaro (marco-zaro)]

Ciao Josh,
Thanks for looking into this. Indeed I was checking without lhapdf, so it
makes sense.
I will be with bad connection this week, as I will be in the mountains.
When I will be back I will try to fix the bug ASAP.
In the meantime, note that nnpdf 2.3 is available also without lhapdf.
Cheers,

Marco
On Dec 29, 2014 12:45 AM, "Josh Bendavid" <email address hidden> wrote:

> Hi Marco,
> I've done some further tests on linux with 2.2.2.
>
> Everything is indeed ok as long as I don't use lhapdf. As soon as I run
> with lhapdf I get the errors as reported in the initial post. (I'm
> running with lhapdf 6.1.4)
>
> --
> You received this bug notification because you are a bug assignee.
> https://bugs.launchpad.net/bugs/1406000
>
> Title:
> Compilation error with complex NLO processes in 2.2.2
>
> Status in MadGraph5_aMC@NLO Generator:
> New
>
> Bug description:
> Trying W+0,1,2 jets in 2.2.2 leads to a compilation error.
>
> Process card like:
>
> import model loop_sm-no_b_mass
> define p = p b b~
> define j = j b b~
> generate p p > w- [QCD] @0
> add process p p > w- j [QCD] @1
> add process p p > w- j j [QCD] @2
>
> Leads to errors during compilation like below (full log attached as
> well):
>
> Program received signal SIGSEGV: Segmentation fault - invalid memory
> reference.
>
> Backtrace for this error:
> #0 0x2B5D7ECCD3C7
> #1 0x2B5D7ECCD9CE
> #2 0x2B5D7F0F769F
> #3 0x2B5D7E92804A
> #4 0x4BDB35
> #5 0x4BE33C
> #6 0x4BF67B
> #7 0x4BCCC1
> #8 0x4B0852
> #9 0x40F945
> #10 0x412124
> #11 0x2B5D7F0E3D5C
> #12 0x40F598
>
> This same process works fine with only the 0 and 1 jet contributions,
> and it also worked fine with 0,1,2 jets in 2.2.1.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/mg5amcnlo/+bug/1406000/+subscriptions
>

[marco zaro (marco-zaro)]

Ciao Josh,
I am back to work.
I have done some tests this morning and I have reproduced your problem with lhapdf6.1.0 and 2.2.3
This said, the backtrace of the error is quite unhelpful even if i recompile everything (including lhapdf) with the -g flag.

I have re-checked that all the information that was hardcoded in 2.2.1 and now is read at run time since 2.2.2 is exactly the same.

It looks like it is a nasty memory-allocation related problem.

I have tried to remove the 'save' stataments in configs_and_props_inc_chooser.f and leshouche_inc_chooser.f (such that the memory allocation should be the quite the same in 2.2.1 and 2.2.2) and the segfault is still there.

Which version of lhapdf are you using?
If it is not 6.1.5, can you try it?

I am a bit out of ideas for the moment...

Let me know what you think.

Cheers,

Marco

[Josh Bendavid (joshbendavid)]

I was using LHAPDF 6.1.4. We don't have 6.1.5 set up on the cms software
environment yet (but will do it soon).

The changes between the two versions are pretty small from what I
understand.

Sent from my Android.
On 5 Jan 2015 07:45, "marco zaro" <email address hidden> wrote:

> Ciao Josh,
> I am back to work.
> I have done some tests this morning and I have reproduced your problem
> with lhapdf6.1.0 and 2.2.3
> This said, the backtrace of the error is quite unhelpful even if i
> recompile everything (including lhapdf) with the -g flag.
>
> I have re-checked that all the information that was hardcoded in 2.2.1
> and now is read at run time since 2.2.2 is exactly the same.
>
> It looks like it is a nasty memory-allocation related problem.
>
> I have tried to remove the 'save' stataments in
> configs_and_props_inc_chooser.f and leshouche_inc_chooser.f (such that
> the memory allocation should be the quite the same in 2.2.1 and 2.2.2)
> and the segfault is still there.
>
> Which version of lhapdf are you using?
> If it is not 6.1.5, can you try it?
>
> I am a bit out of ideas for the moment...
>
> Let me know what you think.
>
> Cheers,
>
> Marco
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1406000
>
> Title:
> Compilation error with complex NLO processes in 2.2.2
>
> Status in MadGraph5_aMC@NLO Generator:
> New
>
> Bug description:
> Trying W+0,1,2 jets in 2.2.2 leads to a compilation error.
>
> Process card like:
>
> import model loop_sm-no_b_mass
> define p = p b b~
> define j = j b b~
> generate p p > w- [QCD] @0
> add process p p > w- j [QCD] @1
> add process p p > w- j j [QCD] @2
>
> Leads to errors during compilation like below (full log attached as
> well):
>
> Program received signal SIGSEGV: Segmentation fault - invalid memory
> reference.
>
> Backtrace for this error:
> #0 0x2B5D7ECCD3C7
> #1 0x2B5D7ECCD9CE
> #2 0x2B5D7F0F769F
> #3 0x2B5D7E92804A
> #4 0x4BDB35
> #5 0x4BE33C
> #6 0x4BF67B
> #7 0x4BCCC1
> #8 0x4B0852
> #9 0x40F945
> #10 0x412124
> #11 0x2B5D7F0E3D5C
> #12 0x40F598
>
> This same process works fine with only the 0 and 1 jet contributions,
> and it also worked fine with 0,1,2 jets in 2.2.1.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/mg5amcnlo/+bug/1406000/+subscriptions
>

[marco zaro (marco-zaro)]

Ciao Josh,
so, the bug should be fixed now (rev324 of 2.2.3).
It was due to some wrong declaration of size of some arrays used in functions in the 'Source' dir.
In order to have it working, please do as follows inside the Source directory:
1) replace the setrun.f inside Source with the one in the patch attached to this thread.
2) remove the symbolic link to leshouche_decl.inc
3) create a symbolic link to ../SubProcesses/PXXXX/born_leshouche.inc, where PXXXX is the SAME dir to which the nexternal.inc link points to

This should normally work for the processes you have already generated.

To be able to generate new processes, please replace the madgraph/iolibs/export_fks.py file inside the madgraph installation with the one in the patch

Please let me know if I can consider the bug as solved.

Cheers,

Marco

[Josh Bendavid (joshbendavid)]

Hi Marco,
This indeed fixes the compilation for newly generated processes. (Did not try patching an existing generated process, and not too concerned about that)

Still running the phase space integration in my W+0,1,2j test, but since it successfully compiled everything now, hopefully it should be ok.

[marco zaro (marco-zaro)]

Ciao Josh,
thanks for checking. I consider the (at least this) bug as solved.
Please don't hesitate to write if you find anything odd.
Cheers,

Marco

Marco Zaro

On Wed, Jan 7, 2015 at 3:44 PM, Josh Bendavid <email address hidden>
wrote:

> Hi Marco,
> This indeed fixes the compilation for newly generated processes. (Did not
> try patching an existing generated process, and not too concerned about
> that)
>
> Still running the phase space integration in my W+0,1,2j test, but since
> it successfully compiled everything now, hopefully it should be ok.
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1406000
>
> Title:
> Compilation error with complex NLO processes in 2.2.2
>
> Status in MadGraph5_aMC@NLO Generator:
> In Progress
>
> Bug description:
> Trying W+0,1,2 jets in 2.2.2 leads to a compilation error.
>
> Process card like:
>
> import model loop_sm-no_b_mass
> define p = p b b~
> define j = j b b~
> generate p p > w- [QCD] @0
> add process p p > w- j [QCD] @1
> add process p p > w- j j [QCD] @2
>
> Leads to errors during compilation like below (full log attached as
> well):
>
> Program received signal SIGSEGV: Segmentation fault - invalid memory
> reference.
>
> Backtrace for this error:
> #0 0x2B5D7ECCD3C7
> #1 0x2B5D7ECCD9CE
> #2 0x2B5D7F0F769F
> #3 0x2B5D7E92804A
> #4 0x4BDB35
> #5 0x4BE33C
> #6 0x4BF67B
> #7 0x4BCCC1
> #8 0x4B0852
> #9 0x40F945
> #10 0x412124
> #11 0x2B5D7F0E3D5C
> #12 0x40F598
>
> This same process works fine with only the 0 and 1 jet contributions,
> and it also worked fine with 0,1,2 jets in 2.2.1.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/mg5amcnlo/+bug/1406000/+subscriptions
>

[Josh Bendavid (joshbendavid)]

For reference/notification:
The issue in
https://bugs.launchpad.net/mg5amcnlo/+bug/1425058
is possibly related to this one