The(SSE1) P3/P4 implementation of ippsCartToPolar_32fc generates floating point infinite magnitudesfor from finite input. This occurs when the complex input is sufficiently close to {re,im} = {0,0}.
The following, whenlinked with the PIII library,generates an incorrect, floating-point infinity, magnitude:
const Ipp32fc complex[] = { { 0.0f, 1e-22f } };
float magnitude[] = { 0 };
float phase[] = { 0 };
::ippsCartToPolar_32fc( complex, magnitude, phase, 1 );
printf( "magnitude: %g
", magnitude[0] );
On my system (Visual C++ 6, SP5) this outputs "magnitude: -1.#INF".
The generic library outputs the correct result.
This occurs in IPP 3.0 and 4.0.
Other magnitude related functions are also flawed.
