Why do we require passing -msimd128 when passing -msse and related flags?

[ngzhian]

Right now, if we pass -msse without passing -msimd128 we get a compile error.

-msse looks like a clear signal that we should enable SIMD, maybe the presence of -msse should enable the -msimd128 flag?

Seems like @juj added this when implementing the SSE compatibility headers.

[sbc100]

I believe this has been discusses at length in the past.. I can't remember exactly why this path was chosen.. but I do think this was the decided on best approach.

@RReverser and @surma I believe were also involved with the discussion.

[RReverser]

Yeah I'm tempted to close this as duplicate of #12714 unless @ngzhian thinks this is a separate issue?

[ngzhian]

Ah, I should have searched! Thanks!

From that issue:

While this works fine for Wasm-specific projects, it adds friction to building existing projects relying on SSE intrinsics. For example, CMake checks that try to use those flags, error out by default, and the only ways to fix it are either 1) modify CFLAGS / CXXFLAGS / etc. on the caller side to include -msimd128, which will now affect all targets in CMake and not only those that used -msse2 or 2) selectively modify CMakeFiles to include this extra flag only in places where -msse2 is used too and only on Emscripten target.

This was exactly what I faced :)

Closing as duplicate.