Is it ok to use 64bit integers in a 32bit application?

Question

I notice in C and C++, we can use int64_t, or simply a long long.

If I compile 32bit code using these types, will I suffer any perform

Answer 1

If I compile 32bit code using these types, will I suffer any performance issues on 64bit and/or 32bit machines?

64 bit integers on 32 bit architectures require two registers to store the value*. 32 bit values require only one register. This matters because:

• You may run out of registers sooner, requiring registers to be spilled to memory. This takes more time.
• Loading and storing 2 registers typically takes two instructions, not one.
• Doing addition or subtraction on operands in two registers takes two or more instructions versus just one for operands in just one register.

Bottom line is that if 32 bit performance is important, don't use 64 bit integers unless you need them.s

* This is true for x86, ARM, and PowerPC processors, which covers most of the processors people program for these days. It is probably true of most other processors as well.

Answer 2

If I compile 32bit code using these types, will I suffer any performance issues on 64bit and/or 32bit machines?

Your compiler may need to generate several machine code instructions to perform operations on the 64 bit values, slowing down those operations by several times. If that might be a concern, you'd want to do some benchmarking to assess the impact on a particular program with realistic data. That issue exists where you're executing the 32 bit executable on a 32 or 64 bit machine.

would I ever have a reason to just use int?

Aside from performance and memory usage, there's occasionally reason to use ints because other APIs/streams etc. that you work with use int. There's also subtle documentary value in using int if it's clearly adequate, otherwise other programmers may waste time wondering why you'd gone out of your way to use a long long.

After all, 64bit ints are far more useful in storing large numbers.

Far more useful in storing very large numbers - sure - but that's relatively rarely needed. If you're storing something like a year or someone's age, there's just no particular point in having 64 bits.