00:18:29 Ian Hinder: it's scalable because you're doing a lot more work than you need to :) 00:21:18 Roland Haas: It works out if you have cores to spare since you suddenly can actually usefully parallelize things. 00:23:06 Eloisa Bentivegna (IBM): Agree. The two alternatives are usually doing work you don’t need vs. leaving resources idle. 00:23:25 Ian Hinder: having an rhs means it is part of the state vector, probably? 00:27:24 Roland Haas: not quite. just that it is picked up by the ODESolver thorn 00:28:20 Roland Haas: state vector would be everything that is checkpointed which could be eg things you cannot recompute but aren't evolved uisng a time derivative (things like the Weyl scalars could be like that). 00:29:22 Roland Haas: checkpoint=no has some other side effects wrt. CarpetX detecting if the variable becomes undefined after a regrid or not. 00:34:18 Ian Hinder: are the "after" clauses in WaveToyCPU_EstimateError needed or can they be derived from READS/WRITES? 00:35:27 Ian Hinder: thanks 00:38:41 Ian Hinder: how are interior and boundary defined? this was always tricky to pin down... 00:38:57 Ian Hinder: (+1 for all the tracking and checking, btw!) 00:39:38 Ian Hinder: halo 00:40:11 Ian Hinder: hence my comment :) 00:53:14 Ian Hinder: this is due to the cache-locality of having small grids to loop over? 00:53:20 Ian Hinder: like tiling? 00:54:53 Ian Hinder: thanks 00:56:20 Zachariah Etienne: what is the highest finite differencing order attempted with z4c and carpetX? Also what is the prolongation/restriction order at that FD order? 00:58:35 Ian Hinder: minimum C++ version required? fortran support? 00:58:57 Zachariah Etienne: ^ 00:59:20 Roland Haas: C++17 right now 00:59:45 Roland Haas: no Fortran scheduled functins at all right now (due to multithreading and static variables). 00:59:52 Ian Hinder: C? 01:00:05 Roland Haas: c99 01:00:21 Roland Haas: CarpetX uses no C so you inherti the Cactus requirements on C 01:00:40 Roland Haas: the static variables for Fortran are in the FrotranWrappers in case you wonder. 01:07:08 Eloisa Bentivegna (IBM): Can HydroToy do multi-fluid? Or be easily generalised to multi-fluid? 01:08:01 Ian Hinder: what would be the main advantages of using CarpetX over using pure AMReX? 01:08:43 Gabriele Bozzola: A general comment: pretty much everything I know about the Einstein Toolkit is because there are examples I could learn/copy from. I think it would be extremely important to have a lot of documentation and examples 01:09:14 Zachariah Etienne: ^ 01:09:26 Ian Hinder: but Cactus provides higher level abstractions than AMReX? 01:10:37 Ian Hinder: WeylScal4 computes Psi4, would need to be updated. Multipole uses only the interpolator. 01:11:56 Roland Haas: Multipole works (we think) since it was the testing code Sam and I worked with to port the itnerpolator. 01:12:14 Zachariah Etienne: what is the highest finite differencing order attempted with z4c and carpetX? Also what is the prolongation/restriction order at that FD order? 01:13:37 W. Kastaun: does amrex have python bindings that could be used to postprocess 3D AMR data? 01:13:41 Ian Hinder: care to say something briefly about CUDA / accelerators? 01:14:20 Roland Haas: Steve has Python bindings 01:18:10 Ian Hinder: sycl? 01:19:49 William Gabella: What is your repo for CarpetX and WaveToy example? 01:20:28 Erik Schnetter: https://bitbucket.org/eschnett/cactusamrex 01:21:06 Gabriele Bozzola: Is there a getting-started tutorial? 01:21:46 Ian Hinder: i have to go now - thanks erik! 01:27:07 Eloisa Bentivegna (IBM): Thanks all! 01:27:09 Geraint Pratten: Thanks! 01:27:14 Bruno Giacomazzo: Thanks! 01:27:16 Zachariah Etienne: great talk!