This is the code that my compiler generates for a 64/64 divide when it doesn't know that the divisor is max 32 bits. It works by using the x86 64/32 div instruction. The same trick would work on x86-64 by using the 128/64 divisions.
I am not sure how good the machine code you get out of your C++ is in the end, but it's hard to beat this handwritten assembly. Of course, you can do better on x86-64 because there are more registers and you don't have to spill on to the stack.
I faced a similar problem implemented 64/64 division on 32-bit x86. I cribbed some code off the internet and came up with this:
https://github.com/titzer/virgil/blob/2c674dbe3512da23c4a644...
This is the code that my compiler generates for a 64/64 divide when it doesn't know that the divisor is max 32 bits. It works by using the x86 64/32 div instruction. The same trick would work on x86-64 by using the 128/64 divisions.
I am not sure how good the machine code you get out of your C++ is in the end, but it's hard to beat this handwritten assembly. Of course, you can do better on x86-64 because there are more registers and you don't have to spill on to the stack.