On Fri, 24 Apr 2009, DJ Delorie wrote: > The fp-int-convert-long-double test does this: > > static volatile signed long long ivin, ivout; > static volatile long double fv1, fv2; > ivin = ((signed long long) (((unsigned long long) ~(unsigned long long) 0) > >> 1)); > fv1 = ((signed long long) (((unsigned long long) ~(unsigned long long) 0) > >> 1)); > fv2 = ivin; > ivout = fv2; > > When sizeof(long double) <= sizeof(long long), it cannot store that > ~0 value in a long double.
But it doesn't need to store it *exactly*; it only tests that the conversion reverses if PREC_OK (argument to TEST_I_F_VAL) is true, and TEST_I_F sets PREC_OK to what should be an appropriate value (based on the types involved, LDBL_MANT_DIG, etc.) in each case. The other tests are applicable even when the conversion is inexact (and even if it overflows to infinity): the compile-time conversions should agree with the runtime ones since the results of converting integers to floating-point are well-defined by IEEE. -- Joseph S. Myers jos...@codesourcery.com
