[Bug tree-optimization/110557] [13/14 Regression] Wrong code for x86_64-linux-gnu with -O3 -mavx2: vectorized loop mishandles signed bit-fields

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110557

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
           Keywords|needs-bisection             |
   Last reconfirmed|                            |2023-07-05
     Ever confirmed|0                           |1
          Component|target                      |tree-optimization

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
ifcvt produces correctly the sign extend:

  <unnamed-signed:56> _ifc__18;


  _3 = a_14(D) + _2;
  _ifc__22 = _3->D.3439;
  _ifc__18 = BIT_FIELD_REF <_ifc__22, 56, 8>;
  _5 = (long long int) _ifc__18;
  _6 = _5 * -4;
  _7 = (long unsigned int) _6;
  _15 = MAX_EXPR <_7, size_19>;

But then the vectorizer misses that:
```
  _ifc__18 = BIT_FIELD_REF <_ifc__22, 56, 8>;
  vect_patt_13.20_76 = vect__ifc__22.19_74 & { 18446744073709551360,
18446744073709551360, 18446744073709551360, 18446744073709551360 };
  vect_patt_10.21_77 = vect_patt_13.20_76 >> 8;
  vect_patt_9.22_78 = VIEW_CONVERT_EXPR<vector(4) long long
int>(vect_patt_10.21_77);
  _5 = (long long int) _ifc__18;
  vect__6.23_80 = vect_patt_9.22_78 * { -4, -4, -4, -4 };
  _6 = _5 * -4;
```

vect_patt_9.22_78 should have been sign extended ...



With the C front-end, we get:
```
  <unnamed-signed:56> _24;
  <unnamed-signed:56> _23;

  _24 = _25->y;
  _23 = _24 * -4;
  _22 = (long unsigned int) _23;
```

Which the vectorizer does not understand could be prompted to 64bit and then
truncated so it does not vectorize it.


r13-3219-g25413fdb2ac24933214123e24ba16502 .