Re: [POC] verifying UTF-8 using SIMD instructions

The cfbot reported a build failure on Windows because of the use of binary
literals. I've turned those into hex for v6, so let's see how far it gets
now.

I also decided to leave out the patch that adds an ascii fast path to
non-UTF-8 encodings. That would really require more testing than I have
time for.

As before, 0001 is v4 of Heikk's noError conversion patch, whose
regressions tests I build upon.

0002 has no ascii fast path in the fallback implementation. 0003 and 0004
add it back in using 8- and 16-byte strides, respectively. That will make
it easier to test on non-Intel platforms, so we can decide which way to go
here. Also did a round of editing the comments in the SSE4.2 file.

I ran the multibyte conversion regression test found in the message below,
and it passed. That doesn't test UTF-8 explicitly, but all conversions
round-trip through UTF-8, so it does get some coverage.

https://www.postgresql.org/message-id/b9e3167f-f84b-7aa4-5738-be578a4db924%40iki.fi
--
John Naylor
EDB: http://www.enterprisedb.com