Re: How to detect whether we're inside of a loop??

["Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>]

 > From: Geoff Keating <geoffk at geoffk dot org>
 > 
 > "Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu> writes:
 > 
 > > I was interested in experimenting with heuristics which might expand a
 > > builtin depending on whether at the point of expansion we're in a loop
 > > or not.  E.g. It might make sense to expand something which might
 > > create slightly larger code only if we think it's likely that it'll be
 > > executed a number of times.  I'm guessing that being inside a loop is
 > > one way to increase that probability.  Thus I was curious if there's a
 > > way to detect this.
 > 
 > Why don't you just use the support that GCC already has for detecting
 > the probability that a block will be executed?

Well first, I don't know how to do that. :-)

But secondly, my limited understanding of probability detection leads
me to believe that using it would answer a different question than the
one I'm posing.  E.g. given this code fragment:

{
  <code1>
  if (foo)
    <code2>
  else
    {
      <code3>
      exit(1);
    }
 for (i=0; i<x; i++)
    <code4>
}

we know that <code1> is executed once.  And we can say that <code2>
has a higher probability of being executed that <code3>, but again it
would be once.  It's <code4> that is likely to be executed many times
(on average) for each pass through the fragment and therefore it is
most suitable for an expensive optimization.

I think your approach would be suited for finding which of the if-else
branches to choose.  Finding the looping spots is what I'm looking
for because I think it has more impact.

Do you agree with this line of reasoning?

		--Kaveh
--
Kaveh R. Ghazi			ghazi at caip dot rutgers dot edu