https://sourceware.org/bugzilla/show_bug.cgi?id=29998
--- Comment #21 from Jan Janssen <medhefgo at web dot de> --- Of course the reproducer is nonsensical. It's just a minified version of systemd-boot. > These files appear to be redundant - their code is never called. When I minified the code, this appeared to be necessary for the miscompile to happen. For example, dropping test2.o from the final link will have popcount be compiled and called correctly. > Shouldn't this function be called main() rather than run() ? Since I am trying to build an EFI binary, I cannot use main(). The miscompile also happens if -Wl,--entry=run is passed when linking. > If you are linking without the standard libraries, can you really expect the program to run ? I don't expect the reproducer to run at all. I only expect it to build correctly. When building with stdlib+main produces correct output, but that's not what I need. So yeah, change the link command to this if you want: x86_64-w64-mingw32-gcc -o test.exe -flto -nostdlib -Wl,--entry=run test2.o test3.o libtest.a -lgcc > True - but as far as I can see this code will never be called. So having LTO eliminate the call to __popcountdi2 is understandable. But run does get called (by the EFI runtime). Or for userspace nostdlib stuff: whichever code loads the stuff is also responsible for calling into the entry point. I don't mind opening a bug on gcc if you think it's on their side. I would not be able to tell. -- You are receiving this mail because: You are on the CC list for the bug.
