Search code examples
rustmicro-optimization

u8 as i32 vs i32::from_be_bytes([u8;4])

Total bike-shedding here, but I'm wondering about the relative efficiency of

let mut buf = [0u8];
input.read(&mut buf)?;
Ok(buf[0] as i32)

vs

let mut buf = [0u8; 4];
input.read(&mut buf[3..4])?;
Ok(i32::from_be_bytes(buf))

Even on aesthetic grounds its hard to choose between these since in isolation I prefer the first set of code, but I have parallel methods for reading 16-, 24- and 32-bit integer data that follow the latter pattern which makes it more appealing in context.

Solution

  • It looks like the first one is going to be slightly more efficient. It uses a movzx instead of a bswap.

    You can see it here: https://rust.godbolt.org/z/WxW1bj4Eh

    It's pretty difficult for the compiler to be able to convert the bswap into movzx because it will need to see through the read to understand that the rest of the buffer is 0.