https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104721
--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> --- (In reply to Michael Matz from comment #2) > Is there a testcase where you noticed this, or was it just reading code? I am working on a patch to check currently_expanding_gimple_stmt != NULL in i386 backend. At -O2, currently_expanding_gimple_stmt isn't NULL outside of gimple to RTL expansion when compiling libgomp.fortran/reduction1.f90. I tracked it down to expand_gimple_basic_block.