Fix bad conversion from llvm_asm! to asm!

[tomaka]

When updating x86_64 from 0.11.x to 0.12.x, I've gotten the following compilation error:

error: invalid instruction mnemonic 'xchgw'
  --> /home/pierre/.cargo/registry/src/github.aaakk.us.kg-1ecc6299db9ec823/x86_64-0.12.3/src/instructions/mod.rs:51:15
   |
51 |         asm!("xchgw bx, bx", options(nomem, nostack));
   |               ^
   |
note: instantiated into assembly here
  --> <inline asm>:2:2
   |
2  |     xchgw bx, bx
   |     ^^^^^

According to this Bochs-related page, the instruction is indeed just xchg, while xchgw is the weird AT&T syntax equivalent.

My code compiles with this change, but I haven't tested if the Bochs breakpoint actually works.

[tomaka]

Note that just running cargo build on the source code of x86_64 doesn't seem to actually verify the assembly. You can try this by writing asm!("xchgfoobarbaz bx, bx") and it will still compile.
I don't know exactly how it works, but it's likely that it's only the linker that checks the correctness of the assembly.

[phil-opp]

Thanks a lot for reporting this! I wasn't aware that asm! is only checked when it's used, but it seems like you're right: https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=74fb87fe9c686b6e4beffb1b1e1fd564

I opened a Rust issue about this at rust-lang/rust#80440 to see if this is a bug or the intended behavior.