[Bug tree-optimization/99082] manual bit-field creation followed by manual extraction does not always produce good code

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99082

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
```
  _8 = (unsigned int) _7;
  _9 = _8 << 16;
  _10 = _6 | _9;
  r8_20 = (const uint8_t) _10;
```

Obviously r8_20 should be assigned from (const uint8_t)_6 since _9's lower
values are 0.

Having a simple pattern like:
(simplify
 (maybe_truncate (bit_or:c @0 @1))
 (if (known_zero_bits (@1) & wi::typemask(type, TREE_TYPE (@0)) == 0)
  (convert @0))

Will help fix this I think but I am not sure if this is a reasonable thing to
do though.

We also have:
  _9 = _8 << 16;
  _10 = _6 | _9;
  r8_20 = (const uint8_t) _10;
  _21 = _10 >> 8;
  b8_22 = (const uint8_t) _21;
  _23 = _10 >> 16;

Where we know that _6's upper 16 bits are 0 so _23 becomes _9 >> 16. and we
know _8 already has only possible 8 bits in lower so further down to just _8.