Quickest way to change endianness

Question

What is the quickest way to reverse the endianness of a 16 bit and 32 bit integer. I usually do something like (this coding was done in Visual Studio in C++):

un         


        

Answer 1

Why aren't you using the built-in swab function, which is likely optimized better than your code?

Beyond that, the usual bit-shift operations should be fast to begin with, and are so widely used they may be recognized by the optimizer and replaced by even better code.

---

Because other answers have serious bugs, I'll post a better implementation:

int16_t changeEndianness16(int16_t val)
{
    return (val << 8) |          // left-shift always fills with zeros
          ((val >> 8) & 0x00ff); // right-shift sign-extends, so force to zero
}

None of the compilers I tested generate rolw for this code, I think a slightly longer sequence (in terms of instruction count) is actually faster. Benchmarks would be interesting.

For 32-bit, there are a few possible orders for the operations:

//version 1
int32_t changeEndianness32(int32_t val)
{
    return (val << 24) |
          ((val <<  8) & 0x00ff0000) |
          ((val >>  8) & 0x0000ff00) |
          ((val >> 24) & 0x000000ff);
}

//version 2, one less OR, but has data dependencies
int32_t changeEndianness32(int32_t val)
{
    int32_t tmp = (val << 16) |
                 ((val >> 16) & 0x00ffff);
    return ((tmp >> 8) & 0x00ff00ff) | ((tmp & 0x00ff00ff) << 8);
}