Thomas Quinn
Thomas Quinn
The checks for table overflow/underflow in clRateMetalTable() in cooling_metal.c and cooling_metal_H2.c have roundoff problems. E.g. if cl->z is greater than cl->MetalzMin by an amount that will be lost in roundoff...
This only effects the "nSteps = 0" trick of creating output files from a checkpoint. CoolingSetTime is not called, so any outputs that depend on cooling (e.g. Edot) will be...
This may be a bad interaction with the Luster file system, but I'm frequently seeing corruption of the Tipsy output files when the "-npio NN" flag is zero. The default...
Binary output (iBinaryOutput != 0) is always single precision. Double precision should be implemented.
A bad Nchilada file will just cause an " ("Unhandled C++ exception in user code.") Handling the XDRExceptions could make a more useful error message.
With recent versions (post 6.10.1) of charm++, changa can occasionally hang right at startup when run with an mpi-*-smp build. This is a bug in recent versions of charm (see...
The PUPable derivation is only needed for polymorphic classes.
The SFC domain decomposition attempts to get equal numbers of particles in each domain within a tolerance of 10% (a compile time constant ddTolerance in ParallelGravity.h). Currently ChaNGa aborts if...
This is obviously an issue for the future, but TreePiece::countType(), whose only current use is to determine the number of sink particles, uses 32 bit ints, which limits the number...
Charm++ has some new options that make it easier to specify core layouts for SMP programs. Furthermore, the core to socket mapping varies between architectures (AMD vs Intel). E.g., dual...