nvcc error : 'ptxas' died due to signal 24
This can arise when the ptxas part of the CUDA C/C++ compiler
fails, eg due to exceeding CPU limits.
Also seen ptxas failure reported with signal 11 (Invalid memory reference)
Work Around
Possibly associated with buggy CUDA code.
E.g. missing i++ in for loop header.
Work around is then to fix the buggy code before recompiling.
Possible alternatives:
- Compile for less advanced nVidia GPU, eg sm_13.
- Try different version of nvcc, eg 4.0 or 5.5
(It appears
the signal 24 bug is fixed in CUDA 5.5)
W.B.Langdon
1 September 2013
(Last update 2nd Sep)