Implement io for thermodynamic force #41
Conversation
|
Sorry for spamming this pull request with commits, but there is problems when using the newly added function on the cluster that I'm working on. Hence, I'm trying a lot to get rid of these. |
|
The current state of the branch can be compiled and passes all tests on a GPU node of the cluster. |
That's not an issue. We can fix this when the PR is finished. Committing often is better than not committing. :) |
Very good. The last time I checked Github did not provided GPU runners. So, unfortunately, we cannot automatically test on GPUs. |
mrmd/io/DumpThermoForce.hpp
Outdated
| ScalarView::HostMirror forceView("forceView", numBins); | ||
| auto thermoForce = Kokkos::create_mirror_view_and_copy(Kokkos::HostSpace(), | ||
| thermodynamicForce.getForce(typeId)); | ||
| for (idx_t idx = 0; idx < numBins; ++idx) | ||
| { | ||
| forceView(idx) = thermoForce(idx); | ||
| } | ||
|
|
||
| dumpThermoForce.dump(filename, thermodynamicForce.getForce().createGrid(), forceView); |
There was a problem hiding this comment.
Why the copying? Directly use thermoForce?
There was a problem hiding this comment.
If I have a thermodynamicForce with multiple species, getForce(typeId) does, apparently, return a View with an incompatible Layout to ScalarView and the ThermoForce.test.cpp fails. I haven't figured out how to solve this issue without the naive for loop, but I would be happy for further suggestions.
mrmd/io/DumpThermoForce.hpp
Outdated
| ScalarView::HostMirror forceView("forceTest", numBins); | ||
| auto thermoForce = Kokkos::create_mirror_view_and_copy(Kokkos::HostSpace(), | ||
| thermodynamicForce.getForce(typeId)); | ||
| for (idx_t idx = 0; idx < numBins; ++idx) | ||
| { | ||
| forceView(idx) = thermoForce(idx); | ||
| } | ||
| dumpThermoForce.dumpStep(forceView); |
There was a problem hiding this comment.
If I have a
thermodynamicForcewith multiple species,getForce(typeId)does, apparently, return aViewwith an incompatible Layout toScalarViewand theThermoForce.test.cppfails. I haven't figured out how to solve this issue without the naive for loop, but I would be happy for further suggestions.
Here, this problem can not be ignored, since we necessarily have multiple species.
There are multiple ways to clean up the commit. One way would be.
|
Solves #40
I'm not at all sure about the use of the Kokkos structures and functions, but it seems to work fine.