https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123105
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2025-12-13
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed (with GCC 15.2). One reason for some expected slowdown is that we
emit much more code with AVX512, 126795 lines of assembly vs. 45348 with AVX2.
That results in 3x more time spent in late optimization.
I'm not sure to what extent this is caused by Eigen itself (maybe we are
comparing apples and oranges here) but for example
unrolling fully peeling vector epilogues with AVX512 could lead to 2x the
number
of statements. Likewise in GCC 16 znver5 with AVX512 will sometimes generate
two vectorized epilogue copies, one with AVX2 and one with SSE2.
-fopt-info is quite verbose and being C++ the locations reported are not
very easy to compare.
