WASM SIMD - Single instruction shuffle

[ppenzin]

Current implementation of wasm SIMD shuffle produces a scalar loop over the bytes of the vector. That is the leftover of the situation when there were multiple shuffles with different lane shuffles - byte shuffle was considered lower priority than other kinds.

To solve this Chakra would need to emit byteshuffle instructions (pshufb on x86).

[ppenzin]

Extra credit - detect shuffle masks that would be able to lower to hardware instructions other than shuffle bytes, but getting rid of the scalar loop would be already significant.

[ppenzin]

There is one case when detecting mask which would pretty straight-forward to implement: cases when opcodes normally resulting in Simd128LowerShuffle by Simd128LowerShuffle_4.