divBy0 ieee754 compatibility

Is this behaviour the same in the CPU than CUDA hardware:

1.0f/-0.0f = -inf
1.0f/0.0f = +inf


And also… what happens if I do a fminf(+inf,-inf) ???

And… is ±inf considered a NaN?


Section A.2 of the programming guide explains all the FP behaviors. But your questions are still valid since the programming guide just lists the rare exceptions where GPU floats differ from IEEE. Neither of your two examples are mentioned as exceptions, so I’d expect they’d work just as you expect.

However, it’s probably a good idea to test your results, since they ARE edge cases. It’d only take a few lines of code just to do a CPU and GPU eval and find out if the guide is correct.