Hello,
I’m switching a code I wrote from using PUGH to using Carpet but now I get lots of NAN’s when I run it. Are there any special parameters needed to run Carpet as unigrid and behave like PUGH? I am also using MoL and get this issue even when I use Euler methods and trivial initial data.
-DG
Hello David,
you should be able to run basic PUGH parfiles with Carpet.
The one difference that comes to mind is that Carpet by default will poison (fill with NaN) unitialized memory while PUGH leaves it the way the OS provided it (which usually means either 0 or whatever value it had before.
You an set:
Carpet::poison_new_timelevels = no CarpetLib::poison_new_memory = no
and if that "fixes" the problem then in fact you have a bug in your code and are accessing unitialized memory.
Can you provide a working example parfile to reproduce the problem?
Yours, Roland
Hello,
I’m switching a code I wrote from using PUGH to using Carpet but now I get lots of NAN’s when I run it. Are there any special parameters needed to run Carpet as unigrid and behave like PUGH? I am also using MoL and get this issue even when I use Euler methods and trivial initial data.
-DG
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
Thank you,
That was driving me nuts. It works now.
-DG
On Oct 30, 2020, at 2:06 PM, Roland Haas rhaas@illinois.edu wrote:
Hello David,
you should be able to run basic PUGH parfiles with Carpet.
The one difference that comes to mind is that Carpet by default will poison (fill with NaN) unitialized memory while PUGH leaves it the way the OS provided it (which usually means either 0 or whatever value it had before.
You an set:
Carpet::poison_new_timelevels = no CarpetLib::poison_new_memory = no
and if that "fixes" the problem then in fact you have a bug in your code and are accessing unitialized memory.
Can you provide a working example parfile to reproduce the problem?
Yours, Roland
Hello,
I’m switching a code I wrote from using PUGH to using Carpet but now I get lots of NAN’s when I run it. Are there any special parameters needed to run Carpet as unigrid and behave like PUGH? I am also using MoL and get this issue even when I use Euler methods and trivial initial data.
-DG
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
-- My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://pgp.mit.edu .
Hello David,
ok, Please note that I made sure to mention that if setting
Carpet::poison_new_timelevels = no CarpetLib::poison_new_memory = no
makes the NaN go away *then* your code has a bug, since it reads from unitialized memory. You saw the NaN when Carpet ensured that the undefined value stored in newly allocated memory is a NaN. Nothing says that memory allocated via malloc() has any particular value, the OS is free to set it all to a value of say 42.
Ie setting
Carpet::poison_new_timelevels = no CarpetLib::poison_new_memory = no
does not actually fix the bug. It just hides its effect.
Yours, Roland
Thank you,
That was driving me nuts. It works now.
-DG
On Oct 30, 2020, at 2:06 PM, Roland Haas rhaas@illinois.edu wrote:
Hello David,
you should be able to run basic PUGH parfiles with Carpet.
The one difference that comes to mind is that Carpet by default will poison (fill with NaN) unitialized memory while PUGH leaves it the way the OS provided it (which usually means either 0 or whatever value it had before.
You an set:
Carpet::poison_new_timelevels = no CarpetLib::poison_new_memory = no
and if that "fixes" the problem then in fact you have a bug in your code and are accessing unitialized memory.
Can you provide a working example parfile to reproduce the problem?
Yours, Roland
Hello,
I’m switching a code I wrote from using PUGH to using Carpet but now I get lots of NAN’s when I run it. Are there any special parameters needed to run Carpet as unigrid and behave like PUGH? I am also using MoL and get this issue even when I use Euler methods and trivial initial data.
-DG
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
-- My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://pgp.mit.edu .
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
users@lists.einsteintoolkit.org
-
Garrison, David -
Roland Haas