On 02/16/2010 12:29 AM, Christoph Groth wrote:
> Hello,
>
> using g++ 4.3.4 (on Debian testing) I discovered the following problem.
>
> I compile the program
>
> ################################################################
> #include <iostream>
> #include <iomanip>
> #include <cmath>
>
> using namespace std;
>
> double length = 0.00106722283403500495;
> double angle = 0.60066412215710574;
>
> struct Vec {
> double x[2];
> };
>
> int main()
> {
> cout << setprecision(18);
> Vec vec = { cos(angle) * length, sin(angle) * length };
> cout << vec.x[0] << ' ' << vec.x[1] << endl;
> cout << cos(angle) * length << endl;
> cout << vec.x[0] - cos(angle) * length << endl;
> }
> ################################################################
>
> with
>
> g++ -O0 -static test.cc
>
> and it produces
>
> 0.000880416620183556821 0.000603184177834822054
> 0.000880416620183556821
> 0
>
> However, with
>
> g++ -O1 -static test.cc
>
> it produces
>
> 0.00088041662018355693 0.000603184177834822054
> 0.000880416620183556821
> 1.08420217248550443e-19
>
> This is on a computer with an Intel Core 2 CPU (I tried several
> different variants).
>
> If I take the second (-O1) binary and execute it on an Opteron it
> produces the _first_ result. Thus, the same static binary produces
> different results on different processors!
That is strange. A bit of debugging would be required to understand
why that happens.
> I observe this problem also with g++ 4.4.2. It seems only a tiny
> difference, but this inconsistency makes my Montecarlo simulation
> produce different results on different machines.
>
> Any help with this would be greatly appreciated.
gcc applies an optimization to code where the same angle is passed to
sin and cos. This is turned into a single call to sincos() aka
__builtin_cexpi(). Clearly the lsb is different in the optimized
case.
Andrew.
