"esparanoia" tests fail
DARIN JOHNSON
darin@usa.net
Fri Oct 28 21:31:00 GMT 2005
I ran the "esparanoia" floating point test suite on an MPC5200
using newlib-1.13.0 and GCC 4.0.2, and got an "unacceptable" rating.
The test is available at http://www.leshatton.org/ESP_903.html.
Has anyone gotten good results with PowerPC and newlib? These may
be problems with my configuration or porting.
Using double-precision, I got these anomalies:
COMMENT: sqrt( X * X ) - X = 7.10542735760100186e-15 rather than 0.
COMMENT: sqrt(X) is non-monotonic for X near 2.00000000000000000e+00
FLAW : Square root is neither chopped nor correctly rounded.
COMMENT: Observed errors run from
COMMENT: -5.00000000000000111e-01 to 1.00000000000000000e+00 ulps.
COMMENT: Comparison alleges that what prints as Z
=2.22507385850720188e-308
COMMENT: is too far from 2.22507385850720237e-308
COMMENT: Comparison alleges that what prints as Z
=4.94065645841246544e-324
COMMENT: is too far from 1.11253692925360119e-308
DEFECT : Badly
COMMENT: unbalanced range: UfThold * V = 4.00000000000000000e+00
is too far from 1.
COMMENT: Comparison alleges that Z = Inf
COMMENT: is too far from sqrt(Z) ^ 2 = 0.00000000000000000e+00
COMMENT: Test, exp(log(X)) - X =
COMMENT: 5.00000000000000000e-01 -> -6.97999210823864928e+02
--
Darin Johnson
darin@usa.net
More information about the Newlib
mailing list