How to Turn Off Floating-Point Exception Interruption

[Ray Lu]

If I have overflowed floating-point values in my C program, the program is interrupted and floating-point value exception message is returned by default. Is there any way to turn off this exception handling, like a compiler option or system function?

I'm running on a OpenVMS Alpha system. I mainly care about VAX floating-point types. I've tried cc option /PDSC_MASK=SILENT but saw no difference. I believe there's no such systerm functions as feholdexcept() on Unix to turn off exception complaint. My short test program is attached below.

Thanks for helping me.

Ray

------------------
#include
#include

int main(void)
{
int i, j;
double d;
unsigned char *dp = (unsigned char*)&d;
float f;

for (i=0; i<16; i++) {
for (j=0; j dp[j] = rand();
f = (float)d;
printf("f=%g\n", (float)f);
}

return 0;
}

[Craig A Berry · Accepted solution]

I don't think there's a way to do that with VAX doubles (D_FLOAT). Is there any reason you can't use IEEE floats? Try compiling with

$ cc/float=ieee/ieee=denorm

[John Gillings]

Ray,

It's very dangerous to ignore fatal exceptions. However, if you really want to you can catch any exception using an exception handler and then do what you like with it.

Here's an example based on your sample program using the standard handler LIB$SIG_TO_RET which converts any signalled exception into a return status and unwinds from the declaring routine.

#include
#include
#include

int sub(float *f){
int i, j;
double d;
unsigned char *dp = (unsigned char*)&d;

VAXC$ESTABLISH(LIB$SIG_TO_RET);
for (i=0; i<16; i++) {
for (j=0; j dp[j] = rand();
*f = (float)d;
}
return 1;
}

int main(void)
{
float f;

printf("stat=%d f=%g \n", sub(&f),(float)f);

return 0;
}

The result is

stat=1284 f=0

1284 is HPARITH, and f is 0 because the store operation that resulted in the exception did not complete. By the time you're back in the main program, all the useful information from the signal array and mechanism args has been lost.

If you want different behaviour you can write your own exception handler, but I'm not sure if you'll be able to continue from a fatal exception, so unwinding may be your only option.

[Jim_McKinney]

Perhaps sigaction()...


#include
#include
#include

int main(void)
{
int i, j;
double d;
unsigned char *dp = (unsigned char*)&d;
float f;

struct sigaction ignore;

ignore.sa_handler = SIG_IGN;
ignore.sa_flags = 0;
sigemptyset(&ignore.sa_mask);
sigaction(SIGFPE,&ignore,(struct sigaction *)NULL);

for (i=0; i<16; i++) {
for (j=0; jdp[j] = rand();
f = (float)d;
printf("f=%g\n", (float)f);
}

return 0;
}