Re: missing eh regions?

Re: missing eh regions?

[Mike Stump]

Just a guess, but...  There is an optimization pass that removed
unneeded EH regions.  Since you didn't compile with -fasync-exceptions
(sp?), they were eliminated as there were no calls...  If people
really want it, we could smarten the compiler about math exception
stuff as it relates to EH.  It already has most of the smarts (in two
different places) but they don't know about eachother.

Also, we need to add support for EH sync points (to make the chip
flush pending insns that might trap) to make this really work as one
might expect.  That requires a new md insn type, and support for it in
the alpha.md file.

> Date: Mon, 6 Oct 1997 00:00:49 -0700
> From: Richard Henderson <rth@cygnus.com>
> To: egcs@cygnus.com

> I was looking to teach the alpha trap shadow processing about 
> exception regions, so that FP exceptions could be properly 
> caught, etc.

> But it appears that, in the appended program, the exception 
> region notes were eliminated.  What gives?  


> r~


> double foo(double x, double y)
> {
>   double result = 0.0;
>   if (x < y) {
>     try {
>       result = x / y;
>     }
>     catch (...) {
>     }
>   }
>   else {
>     try {
>       result = x * y;
>     }
>     catch(...) {
>     }
>   }
>   return result;
> }

Re: missing eh regions?

[Richard Henderson]

On Mon, Oct 06, 1997 at 12:59:04PM -0700, Mike Stump wrote:
>
> If people really want it, we could smarten the compiler about math
> exception stuff as it relates to EH.

I have no idea whether people really want it, but it certainly seems
like something that someone might want to reasonably do.

> Also, we need to add support for EH sync points (to make the chip
> flush pending insns that might trap) to make this really work as one
> might expect.  That requires a new md insn type, and support for it in
> the alpha.md file.

That is essentially what I was doing when I complained.

Incidentally, is there a quick way one can tell if a function has
exception regions?  I'd like to be able to abort the pass early
when I can.


r~

Re: missing eh regions?

[Mike Stump]

> Date: Mon, 6 Oct 1997 16:17:26 -0700
> From: Richard Henderson <rth@dot.cygnus.com>
> To: Mike Stump <mrs@wrs.com>

> Incidentally, is there a quick way one can tell if a function has
> exception regions?

If we want a coarse gained, this has, or doesn't have, we could add
it, but no, there isn't a preexisting one.

Re: missing eh regions?

[Jason Merrill]

>>>>> Richard Henderson <rth@cygnus.com> writes:

> I was looking to teach the alpha trap shadow processing about 
> exception regions, so that FP exceptions could be properly 
> caught, etc.

> But it appears that, in the appended program, the exception 
> region notes were eliminated.  What gives?  

Try -fasynchronous-exceptions.

Jason

missing eh regions?

[Richard Henderson]

I was looking to teach the alpha trap shadow processing about 
exception regions, so that FP exceptions could be properly 
caught, etc.

But it appears that, in the appended program, the exception 
region notes were eliminated.  What gives?  


r~


double foo(double x, double y)
{
  double result = 0.0;
  if (x < y) {
    try {
      result = x / y;
    }
    catch (...) {
    }
  }
  else {
    try {
      result = x * y;
    }
    catch(...) {
    }
  }
  return result;
}