https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123137
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |ASSIGNED
Ever confirmed|0 |1
Last reconfirmed| |2025-12-16
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
/var/gcc/regression/master/11.4-gcc-gas/build/i386-pc-solaris2.11/libstdc++-v3/include/complex:376:20:
missed: not vectorized: more than one data ref in stmt: __r = MEM[(const
struct complex &)_23];
/vol/gcc/src/hg/master/local/gcc/testsuite/g++.dg/vect/pr64410.cc:46:29:
missed: not vectorized: loop contains function calls or data references that
cannot be analyzed
the loop body is
<bb 19> [local count: 955630224]:
# i_196 = PHI <i_62(44), 0(42)>
_21 = i_196 * 16;
_22 = _124 + _21;
_23 = _213 + _21;
_24 = _111 + _21;
__r = MEM[(const struct complex &)_23];
_140 = REALPART_EXPR <MEM[(const struct complex *)_22]._M_value>;
_201 = IMAGPART_EXPR <MEM[(const struct complex *)_22]._M_value>;
_31 = REALPART_EXPR <__r._M_value>;
_26 = IMAGPART_EXPR <__r._M_value>;
_30 = _31 + _140;
_27 = _26 + _201;
REALPART_EXPR <__r._M_value> = _30;
IMAGPART_EXPR <__r._M_value> = _27;
*_24 = __r;
__r ={v} {CLOBBER(eos)};
i_62 = i_196 + 1;
if (size_40 > i_62)
complex lowering is not expanding the aggregate copy because it's of
struct complex type, not COMPLEX_TYPE. But I'd have expected SRA
to scalarize __r here.
As indicated we're vectorizing two unrelated functions in this testcase
(so the previous dump scan [failure] was spurious).
I'll note that x86_64 with -m32 suffers from the same issue and there SRA
is not performed because of x87 math, with -mfpmath=sse it works fine.
The reason is because aggregate copying is bit-for-bit while ld/st pairs
are not bit pattern preserving.
I guess a dg-skip for x32 but not -mfpmath=sse -msse is difficult so
I'll add -msse -mfpmath=sse instead.
