https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96239
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2020-07-20
Blocks| |53947
Status|UNCONFIRMED |NEW
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
I see at -O2
Coalescing successful!
Merged into 1 stores
16 bit bswap implementation found at: _9
New sequence of 1 stores to replace old one of 2 stores
_9 = from_2(D) r>> 8;
MEM[(union BytesOverlay *)&overlay] = _9;
and
movl %edi, %eax
rolw $8, %ax
ret
with -O3 the vectorizer gets in the way and elides overlay:
_7 = (unsigned char) from_2(D);
_8 = BIT_FIELD_REF <from_2(D), 8, 8>;
_9 = {_8, _7};
_3 = VIEW_CONVERT_EXPR<short unsigned int>(_9);
overlay ={v} {CLOBBER};
return _3;
now we could fold the V_C_E of the vector ctor to a bswap on from_2 though
that would be quite a big special pattern. Maybe vector-ctor folding can
consider this case to convert it to a V_C_E to a vector type from the
bswap result. OTOH "fixing" the vectorizer to emit a bswap would be
even nicer.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53947
[Bug 53947] [meta-bug] vectorizer missed-optimizations
