Is _Float16 supported by Emscripten?

[vittorioromeo]

I tried to compile code from my simulation that uses _Float16 under the latest version of Emscripten, but I got this compiler error:

Particles.cpp:495:9: error: _Float16 is not supported on this target
  495 |         _Float16 f16test = 100.f16;
      |         ^
Particles.cpp:495:31: error: invalid suffix 'f16' on floating constant
  495 |         _Float16 f16test = 100.f16;
      |                                ^

I can compile the same code under Windows x64 and Linux x64 with both GCC and Clang. Is _Float16 not supported at all by Emscripten, or is there a flag to enable support? (hardware support, ideally)

Thanks!

[hoodmane]

WebAssembly has no float16 type but there is some discussion about it:

WebAssembly/design#1497
https://github.com/tc39/proposal-float16array

It seems unlikely that Emscripten will do anything to support this before webassembly adds native support for float16.

[brendandahl]

There's a proposal for simd instructions for fp16 (no scalar support). There's experimental support for this in LLVM/emscripten (flag -mfp16) and V8 (flag --experimental-wasm-fp16). _Float16 is not supported, but __fp16 is.