CUDA architecture has a limited amount of memory available for
constants. The CUDA compiler uses this space to hold constants for some
optimizations. However, for large kernels, the number of constants
needed might be larger than the constant space available. For these
conditions, you have to disable this form of optimization with the `-
Xptxas --disable-optimizer-constants` flags.
Currently, the only file that seems to have this issue is the test for
the Lagrangian filter. Someone should take a closer look to see if this
filter in particular is making unnecessarily large worklet/kernel. (In
particular, why does the Lagrangian filter have a larger kernel than the
streamline and stream surface filters?)
If this occurance happens more often, we might need to add some ways to
configure it in the build.
Consumers of VTK-m when enabling of dropping of unused functions
will see VTK-m functions dropped. Previously this didn't happen
as VTK-m didn't build object files with the correct flags for this.
By allowing the linker to remove unused symbols we see a significant
saving the file size of VTK-m tests, examples, and benchmarks.
An OpenMP build of the tests and benchmarks goes from 168MB to
141MB which is roughly a 16% filesize reduction.
Initially I had presumed that these changes would increase link times.
But in measurements the total wall time for compilation of VTK-m has
stayed about the same ( seeing a decrease of 1.5% ). Presumably the
increased computation is offset by the reduction in file writing.
This is a library that contains parts of worklets that can be
precompiled into a library.
Currently, this library contains the implementation of ScatterCounting.
The added files provide support for Lagrangian analysis of velocity fields of time-varying data. Examples show how to use the filter to generate data and a second example demonstrates consuming generated information to calculate new particle trajectories.