https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128587
--- Comment #6 from David C. Rankin <drankinatty at gmail dot com> --- You are much better at looking at those intricate details that I am. I'm just a old engineer that tinker and fights with gdb and valgrind on occasion to get things right. The l->nneur is set from the literal value 2 passed to layers init, and long before the total_error code is ever reached. Removing -march=native eliminates the warning with -O3, so it looks like you nailed the area for the cause. Leaves me scratching my head on why the processor specific intrinsics would lead to a warning -- but that's well beyond my understanding of what the compiler is doing with that option. Let me know if you need anything else from me. This covers the extent of what I'm seeing on my end.
