Re: PATCH: __nodebug__ attribute for use on SSE intrinsic wrappers

[Ian Ollmann <iano at apple dot com>]

On Jul 29, 2005, at 2:54 PM, Stuart Hastings wrote:
> > You can claim this information is never valuable to anyone till  
> > you are
> > blue in the face,
>
> (inhaling deeply and holding breath... :-)
>
> I'm not claiming this information is "never valuable," I'm claiming  
> it has "negative value."  Right now, with our current debug stack  
> (STABS + GDB), this information is /interfering/ with Developers  
> debugging their SSE codes.
>
>
> > and i will tell you there are users who care. Simply
> > because you do not see them or here from them right now, does not  
> > make
> > this true.

As one the people most strongly impacted by this decision, may I  
suggest that you just do this:

#if defined( GCC_SHOW_INTRINSIC_DEBUG_INFO )
        #define HIDE_DEBUG_INFO
#else
       #define HIDE_DEBUG_INFO        __attribute__ ((nodebug))  / 
*Pick yer favorite name here */
#endif

and then just use the HIDE_DEBUG_INFO where you would have used  
__attribute__ ((nodebug)) in  the header?

There are probably three people in the world who want to see this  
info. All three work on GCC. That's fine. We don't need to shut you  
out. However, the intrinsic header debug information is causing  
serious headaches for me and mine, and I would like to see it fixed.  
It has a profound impact on my productivity and my ability to detect  
the causes of performance problems.

I note that if the __builtin's were stable (didn't go away between  
major revisions of GCC) I'd just use those instead.

Ian Ollmann