Applying std::reduce to floats would be a similarly bad idea for having the compiler do it, except people will write it without thinking. Of course, people alwa

Applying std::reduce to floats would be a similarly bad idea for having the compiler do it, except people will write it without thinking. Of course, people always struggle with writing correct floating point so I doubt it makes things better or worse, but it’s conceptually the library equivalent of -fast-math (which coincidentally would also auto-vectorize the loop but can impact more than just that).