[PATCH] fix compiling gcc in SPEC CPU 2000 on darwin with -finline-functions and -funit-at-a-time

[Fariborz Jahanian]

This patch is not right. It eliminates the symptom of the bug by 
inserting an extra label. Bug is in the middle-end. In this
special case, function called in the inlined version of of LIST() has 
its 'static' flag set. I am looking at this bug now.

- Fariborz

On Wednesday, November 5, 2003, at 03:10 PM, Andrew Pinski wrote:

> When compiling the following testcase (derived from gcc in SPEC CPU 
> 2000) on powerpc-apple-darwin,
> the stub for simple_cst_equal is not defined at all.  To fix the stub 
> either has be to defined or the reference to the stub has to be 
> changed to the non-stub version, having the stub be defined is easier
> for now and works for this testcase and also the full gcc source in 
> SPEC.
>
> Ok?
>
> Thanks,
> Andrew Pinski
>
> ChangeLog:
>
> 	* config/darwin.c (machopic_output_possible_stub_label): Allow stub 
> symbol
> 	be not defined when outputting possible stub label.
>
> Patch:
>
> Index: darwin.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/config/darwin.c,v
> retrieving revision 1.53
> diff -u -p -r1.53 darwin.c
> --- darwin.c	16 Oct 2003 01:16:38 -0000	1.53
> +++ darwin.c	5 Nov 2003 22:53:21 -0000
> @@ -1115,7 +1115,7 @@ machopic_output_possible_stub_label (FIL
>        const char *sym_name;
>
>        sym_name = IDENTIFIER_POINTER (TREE_VALUE (temp));
> -      if (sym_name[0] == '!' && sym_name[1] == 'T'
> +      if (sym_name[0] == '!' && (sym_name[1] == 'T' || sym_name[1] == 
> 't')
>  	  && ! strcmp (name+2, sym_name+2))
>  	{
>  	  ASM_OUTPUT_LABEL (file, IDENTIFIER_POINTER (TREE_PURPOSE (temp)));
>
>
> <temp.fix.gcc.diff.txt>
>
>
> Testcase:
>
> int
> attribute_list_contained ()
> {
>   return (simple_cst_equal ());
> }
> int
> simple_cst_list_equal ()
> {
>   return (simple_cst_equal ());
> }
>
> int __attribute__((noinline))
> simple_cst_equal ()
> {
>   return simple_cst_list_equal ();
> }