Float-explorer, a tool for generating very precise assembly programs to explore the darkest recesses of floating-point behavior on your processor without having to bully the compiler into generating the code for you.

And when I say darkest recesses, I'm not referring to "0.1 + 0.2 != 0.3" (which is fairly well-known) but things like "so when you turn on denormal flushing, how exactly are you defining it because there's at least three different definitions..." Or also "does my emulator actually emulate floating-point behavior correctly, or is it delegating to the current hardware which might have a slightly different definition?"