https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121488
Bug ID: 121488
Summary: A missing BIT_FIELD_REF mem optimization due to type
differences
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: pinskia at gcc dot gnu.org
Target Milestone: ---
Created attachment 62091
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62091&action=edit
Reduced testcase
This shows up in the testcase of PR 92645:
```
vector(2) long long int _18;
vector(2) long long int _22;
vector(32) unsigned char _23;
vector(4) float _34;
vector(4) float _35;
D.215655 ={v} {CLOBBER(bob)};
D.215655.fLo.fVec = _18;
D.215655.fHi.fVec = _22;
_23 = MEM <vector(32) unsigned char> [(char * {ref-all})&D.215655];
D.215655 ={v} {CLOBBER(eos)};
...
_34 = BIT_FIELD_REF <_23, 128, 0>;
_35 = BIT_FIELD_REF <_23, 128, 128>;
```
Attached is the reduced testcase.
In the reduced testcase if I change all of the vector types to int, FRE1 is
able to optimize this.
