A few comments on PIC32MZ math exceptions
At the risk of sounding like an advertisement, the PIC32MZ really is an amazing microcontroller! I'm using its floating point unit, and it's blazingly fast. I'm developing a C (non-Harmony) application, and was initially concerned about getting possible divide-by-zero errors. But the PIC32MZ automatically recognizes and propagates NaNs and +/-Infinity floating point results, flawlessly and transparently.
I haven't seen this documented anywhere, but with the PIC32MZ you don't need to modify the startup code, you don't need to write a _general_exception_handler() function to examine the CPU's (exception) "Cause" register, and you don't have to worry about how to recover from math errors. If you do get a math error, the IEEE 754-2008's "quiet NaN" (or [+/-]Infinity) simply propagates through all subsequent operations that use that value. This makes for a robust design, and it's saving me a lot of time and trouble. Thanks Microchip!