Optimizing mul function in diyp

[wjr-z]

I noticed that the mul function in diyfp still uses a simple algorithm. In fact, simple inline assembly can be used in clang and GCC, and in MSVC__ UMULH for optimization.