floating point exception flags and compiler optimization

floating point exception flags and compiler optimization

[Richard B. Kreckel]

[-- Attachment #1: Type: text/plain, Size: 1056 bytes --]

Hi,

I just tried playing with the floating point exceptions using the
<fenv.h> functionality defined in C99. However, it seems to interfere
somehow with optimization. Can anyone give a hint what's going on?
Here is my test program:

$ cat fenv.c
#include <fenv.h>
#include <stdio.h>

int main()
{
    // disguise constants one and zero, in order to fool the compiler
    double x = (double)printf("") + 1.0; // one
    double y = (double)printf(""); // zero
    if (fetestexcept(FE_ALL_EXCEPT)) {
        printf("flag set before call.\n");
    }
    double z = x / y;
    if (fetestexcept(FE_ALL_EXCEPT)) {
        printf("flag set after call.\n");
    }
    printf("%f/%f==%f\n",x,y,z);

}
$ gcc -O0 fenv.c -lm && ./a.out
flag set after call.
1.000000/0.000000==inf
$ gcc -O1 fenv.c -lm && ./a.out
1.000000/0.000000==inf

I also found that the behavior appears to depend on the compiler
version used. gcc 3.x and 4.0 print "flat set after call." even with
optimization, but the above is gcc 4.1.

TIA
  -richy.
-- 
Richard B. Kreckel
<kreckel@ginac.de>


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/x-pkcs7-signature, Size: 3204 bytes --]