[PATCH v2 0/9] x86: operand size handling improvements

[Jan Beulich]

Not the least in order to make sure the main change here is fine
for both Linux and gcc, but also because of the need / desire to
have sufficient testsuite coverage (which by itself has pointed
out further issues addressed by this series), it has taken me
several years to finally get this set of changes ready. The main
goal is better consistency in the handling of insn operands, i.e.
in particular less unexpected behavior when deducing how things
would behave from observations with one (set of) insn(s) or
operand(s) towards other constructs.

1: x86: drop stray W
2: x86: slightly rearrange struct insn_template
3: x86: re-do "shorthand" handling
4: x86: add tests to cover defaulting of operand sizes for ambiguous insns
5: x86: improve handling of insns with ambiguous operand sizes
6: x86: replace adhoc ambiguous operand checking for CRC32
7: x86: move certain MOVSX/MOVZX tests
8: x86: replace adhoc ambiguous operand checking for MOVSX/MOVZX
9: x86: correct VFPCLASSP{S,D} operand size handling

On top of this I've also previously sent additions to the newly
added test cases that with the series applied still don't work.
The question there continues to be whether (and if so how) we'd
want operand size defaulting to work there.

v2: Two new patches. "x86: produce suffix in suffix-always mode for
    {,V}CVT{,T}S{S,D}2SI" dropped. See also individual patches.

Jan