CUDA error messages fail to write

[Checkmate50]

Messages generated by the CUDA renderer fail to write to standard output. The writer fails on line 58 with a read access violation (this was 0x10). Calling StdWriters::initDefaultSingleton() manually beforehand changes the error message to a different access violation.

Reproducing this error is a bit complicated, since a minimal CUDA example involves many lines of code (and a failing example may be fixed), but an illustration can be viewed by running the cuda-hello-world example on the current version of this experimental branch.

[tangent-vector]

Are you able to debug this issue more and work toward a fix? If so, that is probably the easiest path forward.

If you don't think you have line-of-sight to a fix, then we will need to figure out some way to provide a repro to the other devs. Can you point me at a branch that has your CUDA example, so that I can use it to reproduce the issue?

[Checkmate50]

I'm working on debugging this issue, so I think I can push out a fix shortly. I'll let you know if I get stuck for some reason though; this branch should allow reproduction of the issue if needed.