GMP failing make check on Sun Ultrasparc
Torbjorn Granlund
tg-this-will-bounce-but-I-am-subscribed-to-the-list-honest at swox.com
Thu Apr 19 19:24:58 CEST 2007
"James Avery" <avery at diku.dk> writes:
I've just compiled the latest release of libgmp on Solaris. The
"t-get_d" test fails, as shown below:
PASS: t-asmtype
PASS: t-aors_1
PASS: t-divrem_1
PASS: t-fat
mpn_get_d wrong on random data
sign 0
n =0x1FF800FE007FC0
nsize 1
exp -37
want =[43 3F F8 00 FE 00 7F C0] 8998407423164352
got =[40 EF F8 00 FE 00 7F C0] 65472.031006097327918
/bin/bash: line 4: 21355 Abort ${dir}$tst
FAIL: t-get_d
PASS: t-instrument
PASS: t-iord_u
PASS: t-mp_bases
PASS: t-perfsqr
PASS: t-scan
==================================
1 of 10 tests failed
Please report to gmp-bugs at swox.com
==================================
I compiled for 64 bit using the following C/CXXFLAGS:
"-fast -xtarget=ultra3 -xarch=v9b"
This is most likely a bug/feature in the Slowlaris compiler that you
trigger by overriding GMP's option choice. We pass a couple of extra
options to make -fast more well-behaved for floating-point operations.
If you play with different compilation options than the default ones,
you should read the compiler documentation really well.
(It might be a mistake of us to pass -fast for the Sun compilers, as
it triggers non-conformant behaviour. We now work around problems in
one release, but Sun might get the brilliant idea to let it enable
some new non-conformant behaviour in a future compiler release. There
are problems with -fast for more compilers, like Apple's gcc. I
suspect these -fast options exist for inflating SPEC numbers, and
should be avoided if one wants general code to work properly.)
--
Torbjörn
More information about the gmp-bugs
mailing list