Quadruple precision float

Greetings,

my current projects requires high-precision numbers. My calculations indicate that double is insufficient, but quadruple, with its 112+1 bit mantissa, will do fine. My current solution is to use the mpfr library, however, I'm not so sure regarding portability (which is required among POSIX platforms, including Linux, AIX, Solaris and HP-UX). Furthermore, it is rather slow, and serialization is a PITA (for IPC, I currently use the string I/O routines, which impedes performance even further).

A system-dependent solution, preferably for Linux on amd64, would be acceptable, as long as I can switch back to the software emulation if not on the given platform. Any suggestions regarding a mpfr-replacement would be much appreciated. Oh, and yes, I *do* require IEEE 754 compliance, which is also why I use mpfr with its option to operate on subnormal numbers in just the way IEEE does, and not the higher-level wrappers for C++. The one thing I can't afford is a different result on different platforms...
I can't see any reason why MPFR would not be portable. This list of client projects suggest it's portable.

Like all cross platform tools, you need to test on your target platforms. Think of testing this way; if you haven't tested, it doesn't work.
Topic archived. No new replies allowed.