https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103053
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
In this particular case it is cause by my
r7-6851-gb607e75e25261ac2c81d71fe1a4dcf5fcd166603
where we try to emit the division as both signed and unsigned if value range
shows both will work the same, and pick the cheaper one.
The underlying problem is different, we call assemble_external_libcall
prematurely from emit_library_call_value_1, to avoid such spurious .globl
directives we'd need to somehow mark the CALL_INSN and emit those only late
during RTL optimizations, either during final if those directives can be
intermixed with instructions, or close before final otherwise (in some late
always done pass that walks the whole IL).
Because otherwise we can emit it even for library calls we just tried and chose
not to emit, or for code that we even expand, but later during RTL
optimizations optimize away.
