Test of mathematical libraries

This page gives some results obtained with MPCHECK 1.1.0. Contributions are welcome: contact information.

See also the page Test of Mathematical Functions of the Standard C Library from Vincent Lefèvre, which gives a percentage of incorrect roundings (to nearest) for different machines and functions on a set of hard-to-round cases.

LIBM, double type (53 bits)

Pentium 4, Linux 2.6.11-1.35_FC3, gcc 3.4.3, libc-2.3.5
Opteron, Linux 2.6.12-1.1398_FC4smp, gcc 4.0.1, libc-2.3.5 (note: the pow was disabled, since it produced a Seg Fault).
Sparc, SunOS 5.7, cc Sun WorkShop 6
Itanium 1, Debian Linux 2.4.20, gcc 3.2.3, libc 2.2.5
MIPS R16000, IRIX64 6.5, gcc 3.3
PowerPC 7400, Darwin 8.8.0, gcc 4.0.1
PowerPC 7410, Debian Linux, gcc 3.3.5 (note: the pow was disabled, since it produced a Seg Fault).
PowerPC 970, Darwin 8.6.0, gcc 4.0.1
Opteron, 32-bit i386 FreeBSD 4.9, gcc 3.3.3
Opteron, Solaris 10 1/06 s10x_u1wos_19a
Opteron, Solaris 10 1/06 s10x_u1wos_19a
Opteron, Solaris Nevada snv_29
Xeon, Solaris 10 3/05 s10_74L2a
Fujitsu PC with Transmeta Crusoe under Solaris 10 Update 1 (pre-release)
Opteron, Solaris 10 3/05 s10_74L2a
Xeon, Solaris 10 3/05 s10_74L2a
Pentium 3, Solaris 10 3/05 s10_74L2a
Itanium 2, HP-UX B.11.23

Important note: on x86 processors, mpcheck sets the rounding precision to 53 bits for double-precision. Some libraries give better results when the rounding precision is set to 64 bits (default):
X86, Solaris 10, RP64