Paraview PIO Reader wraparound
xRage, the LANL code, normally uses 6 digits to write the cycle number in its dump filename. If a simulation reaches past 1,000,000 cycles, the number will wraparound and start back at 000 000. @pkf
List of things to do:
- Deal with wrapround by checking first and last dump files, and seeing if the number of expected files are correct
- Add PIOFileIndex as a TimeValue
- Share ordering in CycleIndex, SimulationTime and dumpFileName via MPI after reading only on proc 0
- Change reading of metadata to be done only on proc 0 including variables and share via MPI
- Allow multiple pio dump directories
- More far fetched because of scheduling, but read variable data only on proc 0 and send the portions needed to the other processors This wouldn't be too bad for unstructured grid, but hyper tree grid with the dispersing of hypertrees is likely to be as confusing as the XRAGEAdaptor for hypertrees
-- Newer Update --
All issues have been resolved, MR is vtk/vtk!7275 (merged).
-- Update --
Following up on this with an update: xRAGE has been modified to handle dump file naming differently, in order to avoid re-writing dump files when cycle number > 1 million. The new default behavior is to use six numbers in the dump file name (-dmp######) and, if the simulations goes over 1 million, 10 million, 100 million cycles, the dump file name will be automatically adjusted to use 7, 8, 9 numbers in the dump file name.
-- Current Issues --
- Loading a 2D dataset in parallel will seg fault
- Slight counting error when only loading one dump file
- Loading variables as doubles looks wrong