@dpaoliello by the way, is f16/f128 support for arm64ec on the near future radar? As far as I know, since LLVM21 it is our only remaining target to still have an LLVM crash trying to use the f16 type. (f128 has a bit further to go)

@dpaoliello by the way, is f16/f128 support for arm64ec on the near future radar? As far as I know, since LLVM21 it is our only remaining target to still have an LLVM crash trying to use the f16 type. (f128 has a bit further to go)

I'm afraid that I don't know, my team doesn't own defining the ABI/thunks.

@pmsjt @wesleywiser do either of you know if Arm64EC will have a defined way to define thunks for 16- and 128-bit floating point types?

I assumed you may be the one to know since you self-assigned llvm/llvm-project#94434 :)

I would expect at least f16 should be reasonably easy, given aarch64 hardware support and a de facto x86 ABI of passing and returning in xmm (of course that is unfortunately not MSVC-blessed though). That's the type we are much closer to stabilizing anyway.

Unfortunately, I don't have any information on that