How is Float16 implemented?

Found a Float16 bug, the rounding of subnormals seems incorrect, however I can’t find the implementation of Float16. The help that’s built into the REPL and @which both just point me to “intrinsic functions”.

Is the Float16 implementation in Julia at all, or is it all-LLVM?

1 Like

All the basic Float16 arithmetic functions are implemented by LLVM, I believe.


It’s mostly LLVM, together with libgcc functions and whatnot.
Could you open an issue on julia with an MWE so that we can track it?

1 Like