https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122219
--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #3)
> Actually it is because in this case it is a partial write and not a full
> write of the v256_t .
>
> So my patch does not fix it.
But it might help the second example.
```
_11 = VIEW_CONVERT_EXPR<uint128_t>(a_4(D));
MEM[(union simde__m128_private *)&D.62853] = _11;
a_ = D.62853;
r_.m128_private[0] = D.62853;
v = r_;
_7 = MEM <simde__m256> [(char * {ref-all})&v];
v ={v} {CLOBBER(eos)};
<retval> = _7;
```
Should be converted into:
```
_11 = VIEW_CONVERT_EXPR<uint128_t>(a_4(D));
MEM[(union simde__m128_private *)&D.62853] = _11;
a_ = VIEW_CONVERT_EXPR<union simde__m128_private>(_11);
VIEW_CONVERT_EXPR<union simde__m128_private>(r_.m128_private[0]) = _11;
// v = r_; // removed with the dse
_7 = VIEW_CONVER_EXPR<simde__m256>(r_);
v ={v} {CLOBBER(eos)};
<retval> = _7;
```
Which should provide SRA with better input.
On the insert:
```
a_ ={v} {CLOBBER(bob)};
MEM <simde__m256> [(char * {ref-all})&a_] = a_4(D);
b_ ={v} {CLOBBER(bob)};
_15 = VIEW_CONVERT_EXPR<uint128_t>(b_7(D));
MEM[(union simde__m128_private *)&D.62863] = _15;
b_ = D.62863;
a_.m128_private[imm8_9(D)] = D.62863;
v = a_;
_11 = MEM <simde__m256> [(char * {ref-all})&v];
v ={v} {CLOBBER(eos)};
<retval> = _11;
a_ ={v} {CLOBBER(eos)};
b_ ={v} {CLOBBER(eos)};
```
Converted to:
```
a_ ={v} {CLOBBER(bob)};
MEM <simde__m256> [(char * {ref-all})&a_] = a_4(D);
b_ ={v} {CLOBBER(bob)};
_15 = VIEW_CONVERT_EXPR<uint128_t>(b_7(D));
MEM[(union simde__m128_private *)&D.62863] = _15;
//b_ = D.62863;
a_.m128_private[imm8_9(D)] = D.62863;
//v = a_;
_11 = VIEW_CONVERT_EXPR <simde__m256>(a_);
v ={v} {CLOBBER(eos)};
<retval> = _11;
a_ ={v} {CLOBBER(eos)};
b_ ={v} {CLOBBER(eos)};
```
Which also should help ESRA.
