#34599 [Opn]: Accuracy of round() function affected by compiler optimizations

From: Date: Thu, 22 Sep 2005 19:43:42 +0000
Subject: #34599 [Opn]: Accuracy of round() function affected by compiler optimizations
References: 1  Groups: php.bugs 
Request: Send a blank email to php-bugs+get-85464@lists.php.net to get a copy of this message 
 ID:               34599
 Updated by:       sniper@php.net
 Reported By:      osolo at wndtabs dot com
 Status:           Open
 Bug Type:         Math related
 Operating System: Linux
 PHP Version:      4.4.0
 New Comment:

See also bug #17772 which has some talk about optimizations and some
other options.



Previous Comments:
------------------------------------------------------------------------

[2005-09-22 21:16:05] osolo at wndtabs dot com

The bug is **not** about limited precision of floating point numbers.

The bug report is about the optimizer causing the function to return
different values.  The function should return the same value whether or
not its compiled with -O2.  The proposed fix does in fact solve this
problem.

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

[2005-09-22 16:18:52] tony2001@php.net

Floating point values have a limited precision. Hence a value might 
not have the same string representation after any processing. That also
includes writing a floating point value in your script and directly 
printing it without any mathematical operations.

If you would like to know more about "floats" and what IEEE
754 is read this:
http://docs.sun.com/source/806-3568/ncg_goldberg.html
 
Thank you for your interest in PHP.



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

[2005-09-22 16:04:20] osolo at wndtabs dot com

Description:
------------
The round() function will return inconsistent results when PHP is
compiled under certain Linux distro/CPU/compiler combinations.

For example, if when compiling PHP on Gentoo 1.12.0_pre6/cc
3.4.4/Pentium 4, we get:
  round(0.245,2) => 0.24
Instead of the expected 0.25.  

I've traced the problem to the PHP_ROUND_WITH_FUZZ macro in
ext/standard/math.c.  This macro starts out as:

#define PHP_ROUND_WITH_FUZZ(val, places) { \ 
  double tmp_val=val, f = pow(10.0, (double) places); \
  ...

Changing the first line to:
  volatile double tmp_val=val; double f = pow(10.0, (double) places);
\

I suspect the problem is that the optimizer is keeping tmp_val in a
floating point register that has a difference precision than a C
double.  Declaring it volatile keeps it in the variable and makes
things more consisntant.





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


-- 
Edit this bug report at http://bugs.php.net/?id=34599&edit=1

Thread (9 messages)

« previous php.bugs (#85464) next »