Closed danielstumpp closed 3 years ago
I think I have a fix for this. I'm going to do a test run overnight to make sure and will push in the morning if successful
For documentation sake, the bug occurred because when creating the normal distribution on the pynq, sometimes the value was so small it got rounded to zero. This was fixed using a long double type
a floating point exception occurs when running on the pynq with the following error message:
I haven't seen it happen on my local machine or during the benchmarks. This leads me to believe we are dividing by zero or performing some other illegal operation during fault injection.