X-Authentication-Warning: delorie.com: mail set sender to djgpp-bounces using -f
From: "Bob W" <dontsend AT operamail DOT com>
Newsgroups: comp.os.msdos.djgpp
Subject: Sequence points, any?
Date: 5 Apr 2006 05:00:53 -0700
Organization: http://groups.google.com
Lines: 109
Message-ID: <1144238453.674596.302360@z34g2000cwc.googlegroups.com>
NNTP-Posting-Host: 84.102.38.22
Mime-Version: 1.0
Content-Type: text/plain; charset="iso-8859-1"
X-Trace: posting.google.com 1144238460 4629 127.0.0.1 (5 Apr 2006 12:01:00 GMT)
X-Complaints-To: groups-abuse AT google DOT com
NNTP-Posting-Date: Wed, 5 Apr 2006 12:01:00 +0000 (UTC)
User-Agent: G2/0.2
X-HTTP-UserAgent: Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1; SV1; .NET CLR 1.1.4322; .NET CLR 2.0.50727),gzip(gfe),gzip(gfe)
Complaints-To: groups-abuse AT google DOT com
Injection-Info: z34g2000cwc.googlegroups.com; posting-host=84.102.38.22;
   posting-account=CXf2IQ0AAADhHwR4LIBYSPHMQKV3cPd3
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com


Before reporting a gcc bug I would like to consult
the experienced public in this newsgroup for their
opinions:

As mentioned already in a previous post, gcc 4.10
tends to optimize code in a way so that it gets out
of sequence.

This happens already in a simple program which is
meant to do the following:
- call clock() to start timing
- call a recursive function 'Ack(..)'
- call clock() again to get the time elapsed
- display timing

After compilation the code is rearranged by gcc
the following way:
- Ack(..) gets modified by gcc
- gcc calls clock() to start timing
- Ack() is called the 1st time
- gcc calls clock() again to end the timing
- now Ack() is called again and this obsoletes timing
- just timing of the 1st Ack() call gets displayed


I have not made up my mind whether I'd consider this
a genuine bug or not. Theoretically the compiler needs
the results of Ack() only when the printf() statement
is being executed. So gcc seemingly does not care
whether the Ack() result is finally available before
clock() is called the second time.

Unfortunately this does not match the programmer's
expectations and after trying several methods of
getting gcc to compile in sequence with Ack() and
clock() calls I have found the following solutions:

- assign the return value of Ack() to a static or
  global variable instead of using a local var

- 'hide' the second clock() call after the first
  printf() statement


The following methods did not work in finding a
way to prevent gcc rearranging code sequence:

- declaring variables for clock() timing values
  as global, static or extern

- calling clock() a third time in order to get
  gcc to interleave its second call to Ack()
  between the 2nd and 3rd call to clock() as
  it is obviously done if a printf() statement
  is used instead of clock()

- using a do {..} while loop to call Ack()
  e.g.: do { ackret=Ack(..); } while(0);


One further observation:

If gcc's out-of-sequence optimisation is prevented
by using one of the methods mentioned previously,
the program actually executes a touch faster. So
from the programmer's point of view gcc's behaviour
of breaking the sequence cannot be considered
beneficial.


Finaly, here is a sample program which will
demonstrate the 'maybe-bug' - unless the
/* static */ is uncommented (use parameters
of 11 or 12 if you actually want to try it
out):

-------------------------------

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

extern clock_t t0,t1,t2;
extern int n;

int Ack(int m, int n) {
  return(m ? (Ack(m-1,n ? Ack(m,(n-1)) : 1)) : n+1);
}

int main(int ac, char **av) {
  /* static */ int ackret;
  n = (ac == 2) ? atoi(av[1]) : 1;

  t0=clock();

  do { ackret=Ack(3, n); } while(0);

  t1=clock();

  printf("Ack(3,%d): %d\n", n, ackret);
  t2=clock();
  printf("Time1: %g secs\n", 1.0*(t1-t0)/CLOCKS_PER_SEC);
  printf("Time2: %g secs\n", 1.0*(t2-t0)/CLOCKS_PER_SEC);
  printf("t2:%d, t1:%d, t0:%d, cps:%d\n",t2,t1,t0,CLOCKS_PER_SEC);

  return 0;
}