From mboxrd@z Thu Jan  1 00:00:00 1970
From: Michael Hayes <m.hayes@elec.canterbury.ac.nz>
To: law@cygnus.com
Cc: Michael Hayes <m.hayes@elec.canterbury.ac.nz>, egcs@cygnus.com
Subject: Re: Re-run of loop pass
Date: Sun, 18 Oct 1998 03:11:00 -0000
Message-id: <13865.44885.953349.460182@ongaonga.elec.canterbury.ac.nz>
References: <13865.9178.455954.70561"@ongaonga.elec.canterbury.ac.nz> <18798.908694891@hurl.cygnus.com>
X-SW-Source: 1998-10/msg00680.html

Jeffrey A Law writes:
 > You should investigate why the second loop pass is unable to determine
 > loop iteration counts.

Consider a simple dot product function like the following:

float foo(float *a, float *b)
{
  int i;
  float result = 0.0;
  
  for (i = 0; i < 4; i++)
    result += a[i] * b[i];
  
  return result;
}

The first pass often eliminates the BIV associated with i and replaces
the end of loop test i < 4 with a comparison of the incremented
pointer associated with a + i with a + 4 (I think the elimination is
only performed for targets with autoincrement addressing modes).  The
second pass then finds that the initial value of BIV associated with a
is not a constant and thus cannot determine the iteration count.

I imagine that the information regarding iteration counts should be
gathered on the first pass and then used on the second pass if needed.

Michael.