|
From: Arseny K. <ars...@gm...> - 2010-10-16 10:17:51
|
Hello, list!
I'm working on application for numerical computing.
It sets FPU exceptions mask, that every invalid floating operation or
operation with NaN or Inf should raise an exception. And handler
consider this as a fatal error.
So, with this settings, any dealing with uninitialised floating point
numbers is a error. It also includes just loading them to floating point
registers. To be more concrete, the example of incorrect code:
int main () {
double *a = new double [2];
double b = a[0] + a[1];
delete[] a;
return 0;
}
I found in old manuals, that such behaviour of valgrind was the default
in old versions:
> However the case with floating-point data is different.
> During a floating-point read instruction the 'V' bits corresponding to
the data are checked.
> Thus copying of uninitialized value will produce error in case of
floating-point numbers.
Is it possible now to enable this feature by some configuration?
Is it difficult to implement?
--
Klimovsky Arseny
|