Re: [Bug lto/47222] New: [4.6 Regression] Revision 168593 failed many tests

[Jan Hubicka <hubicka at ucw dot cz>]

> One failure is
> 
> /tmp/ccJSP9PV.lto.o: In function `main':^M
> ccWUIi0z.o:(.text.startup+0x1d): undefined reference to `g_ebp_save'^M
> ccWUIi0z.o:(.text.startup+0x23): undefined reference to `g_esp_save'^M
> ccWUIi0z.o:(.text.startup+0x3b): undefined reference to `g_edi'^M
> ccWUIi0z.o:(.text.startup+0x41): undefined reference to `g_esi'^M
> ccWUIi0z.o:(.text.startup+0x47): undefined reference to `g_ebx'^M
> ccWUIi0z.o:(.text.startup+0x4d): undefined reference to `g_ebp'^M
> ccWUIi0z.o:(.text.startup+0x53): undefined reference to `g_esp'^M
> collect2: ld returned 1 exit status^M

Hmm, this does not seem to be dependent on plugin implementation, but it seems like
bug in 32bit only testcase.  The testcase seems to use:

int g_edi=INIT_EDI, g_esi=INIT_ESI, g_ebx=INIT_EBX;
int g_ebp, g_ebp_save, g_esp, g_esp_save;
int n_error;

and

        __asm__ __volatile__ (
        "movl %ebp," ASMNAME("g_ebp_save")"\n\t"
        "movl %esp," ASMNAME("g_esp_save")"\n\t"
        );

I guess either g_ebp_save needs externally_visible attribute or the asm
statement needs to be modified to use g_ebp_save as an destination.

I think this testcase is "yours", so perhaps you can fix it?
Was there some reason to use the ASMNAME hack instead of memory output constraint?

Thanks,
Honza