Statically define an array whose size is known at runtime. Expected/unexpected behavior?

Accelerated Computing CUDA CUDA Programming and Performance
1

This is related to C/C++ and not directly related to CUDA. Well, it’s part of a CUDA project.

In CS 101 I was taught that statically defined arrays could not depend on something like:

int N=atoi(argv[1]);

float arrayOfFloats[N];

However, both NVCC, for compiling, and GCC 4.6, linking, ate this up and had no complains. Does the compiler cleverly replaces it by a heap allocation?

What is the behavior for this sort of thing, other than the segmentation fault I got. Shouldn’t the compiler complain a-priori?

[EDIT:]

I had a similar thing on an OpenCL project (my bad) that complained neither at compile nor at runtime on one system, but for other systems crashed. Is unexpected, isn’t it?

2

I’m not sure what the crashing is related to, but a number of compilers (including GCC) allow static arrays of variable size as an extension to the language. NVCC calls out to GCC for host-side compilation, so I’m not surprised you don’t see any complaints.

(GCC can be run with warning flags turned on that will complain about this kind of non-standard usage.)