scourge
October 25th, 2006, 08:50 PM
I just upgraded to Edgy and was pleased to see that GCC now produces considerably faster binaries. However, everything isn't working as expected.
In my project (a chess engine) I have a function like this:
double getPhase(int mat)
{
ASSERT(mat >= 0);
if (mat > max_mat)
mat = max_mat;
return (double)(max_mat - mat) / (double)max_mat;
}
If I compile the app with assertions (I use my own assert macro) the function works as expected. But if I comment out the ASSERT line or disable all assertions the result isn't always correct. Most of the time the return value is what I want, but if I run the program's benchmark where the function is executed millions of times I see some problems:
With ASSERT: 4992637 nodes (correct result)
Without ASSERT: 5009880 nodes (incorrect result)
GCC 3.3 compiles the function correctly with or without the ASSERT line, but I've had no such luck with GCC 4.1.2 20060928 (prerelease) (Ubuntu 4.1.1-13ubuntu5). So does anyone have any idea what could be causing the problem?
In my project (a chess engine) I have a function like this:
double getPhase(int mat)
{
ASSERT(mat >= 0);
if (mat > max_mat)
mat = max_mat;
return (double)(max_mat - mat) / (double)max_mat;
}
If I compile the app with assertions (I use my own assert macro) the function works as expected. But if I comment out the ASSERT line or disable all assertions the result isn't always correct. Most of the time the return value is what I want, but if I run the program's benchmark where the function is executed millions of times I see some problems:
With ASSERT: 4992637 nodes (correct result)
Without ASSERT: 5009880 nodes (incorrect result)
GCC 3.3 compiles the function correctly with or without the ASSERT line, but I've had no such luck with GCC 4.1.2 20060928 (prerelease) (Ubuntu 4.1.1-13ubuntu5). So does anyone have any idea what could be causing the problem?