Build failure on ARM

[H-Plus-Time]

This seems (well, is) down to the inclusion of the SSE flags in CMakeLists.txt, line 98-99. Removing those two lines resolves the problem.

As far as I'm aware, cmake does not provide an architecture flag (though WIN32, MSVC evidently work), though executing uname -m should be sufficient.

How central are SIMD operations to binaryen's performance? Ideally I'd like to keep SIMD and support ARM, but given the...weirdness of gcc/g++ flags (see Stackoverflow|Detect neon availability), it would be considerably more expedient to provide degraded performance (no neon instructions) on ARM, followed by a more complete solution at a later date.

[kripken]

I'd guess it doesn't matter much. In a few tests I ran now I see no difference. So we can just remove those.

[binji]

IIRC, using sse2 is required on x86 32-bit so it performs the correct floating point arithmetic. This is why I turned it on in wabt, anyway.

[kripken]

Oh, right! This was a correctness fix for the interpreter, thanks @binji.

So not so easy to remove...

[dschuff]

Surely it shouldn't be too hard to include the SSE flags conditionally on the target architecture though.

[binji]

Perhaps there's a clever way to do this, but I ended up compiling an executable and using file to do this in wabt: https://github.com/WebAssembly/wabt/blob/master/CMakeLists.txt#L116

[H-Plus-Time]

Compiling a dummy executable probably is the clever way of doing it, since file is a fair bit more stable than uname. PR 978 is pretty much identical to the linked wabt implementation.

[froydnj]

Looks like #978 got fixed, so this can be closed now?