Re: [PATCH] Fix gcc.dg/torture/pr26565.c at -O0 on SPARC/SH

[Roger Sayle <roger at eyesopen dot com>]

Hi Eric,

On Wed, 3 May 2006, Eric Botcazou wrote:
> Regtesting on SPARC/Solaris and AMD64/Linux indeed reveals a few problems:
>
> FAIL: gcc.c-torture/execute/builtins/memops-asm.c execution,  -O0
> eric@linux:~/build/gcc/native> gcc/xgcc -Bgcc -c memops-asm.c
> memops-asm.c:12: warning: asm declaration ignored due to conflict with
> previousrename
>
> extern void *memcpy (void *, const void *, size_t)
>   __asm (ASMNAME ("my_memcpy"));
>
> doesn't cause a call to my_memcpy to be emitted anymore.

Hmm.  I think that this is still the best way to go, but that we'll
also need to tweak c-pragma:maybe_apply_renaming_pragma to avoid the
above fallout; complaining about DECL_ASSEMBLER_SET_P.

Something like

	if (DECL_ASSEMBLER_NAME_SET_P (decl)
	    && !DECL_BUILT_IN (decl))

Ideally, we'd continue to issue the renaming warning if we could
determine whether the built-in had been explicitly prototyped or
defined earlier.  Unfortunately, this information is front-end
specific.

So for C it looks like:
	  || C_DECL_DECLARED_BUILTIN (decl)))
	  ...

and for C++ it looks like:
	|| !DECL_ANTICIPATED (decl))

which is a real pain for maybe_apply_renaming_pragma which is
shared between the C family front-ends.  It might be easiest to
just allow builtins to be renamed without warnings, unless someone
has a better idea as how to solve this.  I thought we used to
use DECL_ARTIFICIAL for this, but it doesn't look like there's
any language-independent way to check whether a buitlin has been
explicitly prototyped.  Perhaps we could strcmp the assembler
name with the function's name, i.e. test "memcmp" == "memcmp"?

Thoughts?

Roger
--