RFR(S): 8178811: Minimize the AVX <-> SSE transition penalty on x86

[Andrew Haley]

I'm sorry: my reply sounded terribly mean spirited but I didn't intend
it to be.

3% on an important benchmark is certainly worth having, and I'd
support the patch for that reason.  I'm just confused about what is
allowed: for example, there are a couple of missing vector patterns in
AArch64.ad, but I haven't posted the fix because we're in rampdown.

Andrew.