See the README one level up about where to find the reference manual for the stm32f4 and schematics for the board. Just having some fun. Curious to know how the hardware fpu compared to TestFloat. Extracted only the add function which is easy/boring and it looks good. A NaN is not a fixed bit pattern so not surprising they dont match. I guess I could have said if this is a nan and that is a nan then pass, instead I said if TestFloat says nan, continue; llvm likes to throw memcpy()'s into code with structures where gcc doesnt. Very ugly memcpy() implementation, which probably hurts llvm's overall slowfloat performance.