Re: PATCH: PR c++/16276: Group jump table with function

["H. J. Lu" <hjl at lucon dot org>]

On Fri, Jul 09, 2004 at 01:54:29PM -0700, Richard Henderson wrote:
> On Fri, Jul 09, 2004 at 11:31:28AM -0700, H. J. Lu wrote:
> > -  if (!(flags & SECTION_DEBUG))
> > -    *f++ = 'a';
> > -  if (flags & SECTION_WRITE)
> > -    *f++ = 'w';
> > -  if (flags & SECTION_CODE)
> > -    *f++ = 'x';
> > -  if (flags & SECTION_SMALL)
> > -    *f++ = 's';
> > -  if (flags & SECTION_MERGE)
> > -    *f++ = 'M';
> > -  if (flags & SECTION_STRINGS)
> > -    *f++ = 'S';
> > -  if (flags & SECTION_TLS)
> > -    *f++ = 'T';
> > -  *f = '\0';
> 
> I think you most certainly do not want to be pulling out the flags
> as a string into the caller.  They do not line up 100% between the
> targets.

I changed readonly_data_section_in_function_group to call
named_section_flags instead.

> 
> > +#ifdef HAVE_GAS_COMDAT_GROUP
> > +#define PUT_JUMP_TABLES_IN_TEXT_SECTION JUMP_TABLES_IN_TEXT_SECTION
> > +#else
> > +#define PUT_JUMP_TABLES_IN_TEXT_SECTION \
> > +  (JUMP_TABLES_IN_TEXT_SECTION || DECL_ONE_ONLY (current_function_decl))
> > +#endif
> 
> I think the !COMDAT case is wrong.
> 
> Suppose SUPPORTS_ONE_ONLY is false, but SUPPORTS_WEAK is true.  In that
> case we don't actually have to worry about what section jump tables go in.
> 
> Suppose it's a strict harvard architecture machine.  In which case you 
> can't go putting *any* data in the text section.
> 
> I think the only workable solution is to put all the private read-only
> data for a function in an rodata linkonce section for the function.

OK.

> 
> > +#ifdef READONLY_DATA_SECTION_IN_FUNCTION_GROUP 
> > +    READONLY_DATA_SECTION_IN_FUNCTION_GROUP (decl);
> > +#else
> 
> Now I think you're inventing macros that no one will ever use.

Removed. Here is the new patch.


H.J.

Attachment: gcc-3.4-comdat-9.patch
Description: Text document