On Wed, Jul 25, 2001 at 03:34:00PM -0700, David Schleef wrote: > On Wed, Jul 25, 2001 at 01:23:38AM -0400, Matt Zimmerman wrote: > > Hercules 2.12 doesn't seem to build on powerpc due to its use of the > > 'sqrtl' function, which computes the square root of a long double. This > > function is in libm on other architectures (though it doesn't appear to be > > documented), but is absent on powerpc. > > > > How should I work around this problem? > > 'long double' is the same as 'double' on PowerPC, so you can define sqrtl() > to be sqrt(). However, that brings up the question of whether extended > precision floating point is necessary for proper emulation.
It shouldn't make too much of a difference, judging from these comments in the source: /* * WARNING * This code is presently incomplete * * Instructions: * Only a subset of the instructions is actually implemented. * [...] * Precision: * This code assumes the following relations between C data types and * emulated IEEE formats: * - float can represent 32-bit short format * - double can represent 64-bit long format * - long double can represent 128-bit extended format * On some host systems (including Intel), the long double type is * actually only 80-bits, so the conversion from extended format to native * long double format will cause loss of precision and range. */ > > (Please CC, as I'm not on the list) > > Uh... then why did you use Mail-Followup-To: to specifically _not_ > do that? =) At the last minute, I decided to copy debian-devel to reach a larger audience. mutt must have noticed that I was subscribed to debian-devel, and tried to do the right thing. -- - mdz
