Floating point optimization?

[Sam Lantinga <slouken at devolution dot com>]

There appears to be an inconsistency in the way floating point
values are handled under Intel egcs with and without optimization:

j.c --------
float new_past_cost = 899.999878, new_entry_cost = 33.3333359, m_future_cost = 1140;
float m_total_cost = 2073.33325;
------------

t.c --------
extern float new_past_cost, new_entry_cost, m_future_cost;
extern float m_total_cost;

main()
{
    float new_total_cost;

    new_total_cost = new_past_cost + new_entry_cost + m_future_cost;
    /* new_total_cost = 2073.333214, m_total_cost = 2073.333252 */
    if ( new_total_cost < m_total_cost ) {
        printf("New total cost is less than total cost\n");
    }
    printf("Done!\n");
}
------------

Output with egcs 1.1.1 and 1.1.2 on Linux i686 with -O0:
--
Done!

Output with egcs 1.1.1 and 1.1.2 on Linux i686 with -O1 or -O2:
--
New total cost is less than total cost
Done!

Why are they different?

An analysis of the assembly output shows that the -O0 version has
    fstps -4(%ebp)
    flds -4(%ebp)
between the floating point adds and the floating point compare.
I don't know assembly well enough to know what these are doing.

The result, however, is that non-optimized code results a different
branch of the logic being taken from the optimized code.

Comments?

	-Sam Lantinga				(slouken@devolution.com)

Lead Programmer, Loki Entertainment Software
--
Author of Simple DirectMedia Layer -
	http://www.devolution.com/~slouken/SDL/
--