Sometimes the CUDA runtime would not allocate sufficient stack
space for the particle advection code to run. This issue was exposed by
!1737 -- for some reason, once those changes to unrelated filters/worklets
are added to VTK, CUDA allocates less stack and the following tests would
fail:
UnitTestLagrangianFilterCUDA
UnitTestLagrangianStructuresFilterCUDA
UnitTestStreamlineFilterCUDA
UnitTestStreamSurfaceFilterCUDA
These were fixed by increasing the stack size in the particle advection
worklet Run(...) methods.
An RAII helper has been added that will restore the previous stack size
in case an exception is thrown, and the KDTree code has been updated
to use this helper when it adjusts the CUDA stack allocation.
We previously had to mark the make_FunctionInterface function as
host/device as it could be used in either place. The ramifications
of this is that each time we launched a worklet any input parameter
had to either suppress cuda exceptions, or had to rely on the
DispatcherBase to suppress the warnings.
By making make_FunctionInterface only host callable ( as it is ),
we can remove all of our unneccesary suppression logic and better
expose real issues with code that is marked host/device but can't
be due to calling things such as std::abort
Now that the dispatcher does its own TryExecute, filters do not need to
do that. This change requires all worklets called by filters to be able
to execute without knowing the device a priori.
By making these helper classes ExecObject, they no longer need to have a
device to be constructed. This is important for future plans to make the
filters not determine a device adapter now that dispatchers no longer
require a device adapter.
This will make it easier to support integrators as an ExecObject.
One side effect is that the integrators and partical advection
are not templated by the type of the field.
Regardless of the type of the field, there is probably little reason to
compute particle advection with less than 64 bit floats to account for
accumulated errors. This will make it easier to use these classes.
Rather than force all dispatchers to be templated on a device adapter,
instead use a TryExecute internally within the invoke to select a device
adapter.
Because this removes the need to declare a device when invoking a
worklet, this commit also removes the need to declare a device in
several other areas of the code.
Adding Utkarsh's changes from 86bd81de to Rectilinear grid as well
- To make rectilinear grids work with any storage type
Adding Rob's changes from 7d6fe568 to the RK4Integrator
- Previously the RK4Integrator could compute velocity with uninitialized inputs.
Adding Rob's changes from a6bf830f to ParticleAdvection
- Previous ParticleAdvection would dispatch some of the worklets on the default device adapter
and not the explicitly requested device adapter
Adding Ken's change from 0753131a
- For supporting all the previous merge conflicts with the new ExecutionObjectBase change.
- PushOutOfDomain would loop over without exiting indefinately utilizing all max steps.
It's supposed to execute only once the particle goes out of spatio-temporal boundary.
- Removing the check for the steamlines polyline points to be same as the maxSteps from
the unit test, as if the particle has already taken some steps, then the polyline
points written out will always be lower than the total number of steps taken by the
current node.
- Adding code to support taking short steps (minimize boundary errors)
- Adding time of advection to the particle from the Integrator
- API changes for TakeStep
- TakeStep should only care about the maximum steps as it's a field on a Particle
- The check for out of temporal and spatial bounds has been moved to the Worklet
- Changing some of the internal APIs for supporting these new features/fixes
- Changing inheritance in the Integrators class to be more obvious than something convoluted
Bug Fixes in the UniformGridEvaluator
Added Temporal Grid Evaluator for Uniform Grid
Updated Integrator to add temporal integration
Changing the TakeStep method in Particles to only deal with max steps
ideally, particles should be unaware of spatio-temporal constrains
Updating the ParticleAdvectionWorklet file to handle out of spatio-temporal
issues.
Missing: Code to take small steps (AT_SPATIAL_BOUNDARY and AT_TEMPORAL_BOUNDARY).
Sandia National Laboratories recently changed management from the
Sandia Corporation to the National Technology & Engineering Solutions
of Sandia, LLC (NTESS). The copyright statements need to be updated
accordingly.
