Andrey Novoseltsev wrote:
On Sunday, 27 April 2014 09:31:13 UTC-6, leif wrote:
Tracking this further down...
Would you mind retrying whether
env CFLAGS="-O3 -march=native -mtune=native" ./sage -f -c ecm
also works for you?
It does not:
Ah, thanks, but I just noticed that's a different failure than I get
(more precisely, apparently in the same code fragment, but on different
data); for me it fails later in the P-1 tests*.
While tracking what I thought was your previous one down, my testcase
suddenly did no longer fail, until I noticed the wrong code had [also]
been triggered by an implicit, unintended '-mtune=generic', which
doesn't get overridden by just a subsequent '-march=bdver{1,2}'.
If you have the time, you may check whether
env CFLAGS="-O3 -march=bdver1 -mtune=bdver1" ./sage -f -c ecm
gives the same (on an Opteron 62xx; 'bdver1' or 'bdver2' on an Opteron
63xx).
By the way, different versions of GCC fail in different ways on
Bulldozer/Piledriver CPUs (with XOP extensions enabled); GMP-ECM's test
suite is just one example where bugs show up. With Ubuntu's GCC
4.6.3-1ubuntu5 from Precise, I for example don't get the test suite
failure in GMP-ECM, but a lot of doctests fail unless I compile Sage [or
at least parts of it, i.e., some spkgs] with '-mno-xop' (and
'-march=bdver1 -mtune=bdver1 -O3').
-leif
P.S.: Although MPIR 2.6.0's /test suite/ also hangs indefinitely for me
(with '-O3' and XOP enabled), the failures in GMP-ECM are unrelated to
GMP/MPIR. And MPIR 2.7.0 (alpha)'s test suite passes without problems...
_____________
*
GMP-ECM 6.4.4 [configured with GMP 6.0.0, --enable-asm-redc] [P-1]
Input number is 474476178924594486566271953891 (30 digits)
Using B1=9594209, B2=670420198, polynomial x^1, x0=2067103269
Step 1 took 416ms
lt-ecm: ../ecm-6.4.4/pm1fs2.c:1863: poly_from_sets_V: Assertion
`set->elem[i] == -set->elem[c - 1L - i]' failed.
Aborted
############### ERROR ###############
Expected return code 8 but got 134
(With vanilla FSF GCC 4.8.1, configured with '--with-arch-64=bdver2'.
Fails with e.g. '-march=bdver2 -mtune=generic -O3', works with e.g.
'-march=bdver2 -mtune=bdver2 -O3'.)
Now running GMP-ECM's test suite...
Making check in x86_64
make[1]: Entering directory
`/home/novoselt/sage/local/var/tmp/sage/build/ecm-6.4.4/src/x86_64'
make[1]: warning: -jN forced in submake: disabling jobserver mode.
make[1]: Nothing to be done for `check'.
make[1]: Leaving directory
`/home/novoselt/sage/local/var/tmp/sage/build/ecm-6.4.4/src/x86_64'
make[1]: Entering directory
`/home/novoselt/sage/local/var/tmp/sage/build/ecm-6.4.4/src'
make[1]: warning: -jN forced in submake: disabling jobserver mode.
make[1]: Leaving directory
`/home/novoselt/sage/local/var/tmp/sage/build/ecm-6.4.4/src'
./test.pp1 ./ecm
GMP-ECM 6.4.4 [configured with MPIR 2.6.0, --enable-asm-redc] [P+1]
Input number is 328006342451 (12 digits)
Using B1=120, B2=8008, polynomial x^1, x0=5
Step 1 took 0ms
Step 2 took 0ms
********** Factor found in step 2: 328006342451
Found input number N
GMP-ECM 6.4.4 [configured with MPIR 2.6.0, --enable-asm-redc] [P+1]
Input number is 328006342451 (12 digits)
Using B1=120, B2=8008, polynomial x^1, x0=262405073961
Step 1 took 0ms
Step 2 took 4ms
********** Factor found in step 2: 328006342451
Found input number N
GMP-ECM 6.4.4 [configured with MPIR 2.6.0, --enable-asm-redc] [P+1]
Input number is 2050449218179969792522461197 (28 digits)
Using B1=20, B2=-1578-1248978, polynomial x^1, x0=6
Step 1 took 0ms
ecm: pm1fs2.c:1863: poly_from_sets_V: Assertion `set->elem[i] ==
-set->elem[c - 1L - i]' failed.
Aborted
############### ERROR ###############
Expected return code 14 but got 134
make: *** [check] Error 1
Error: GMP-ECM's test suite failed.
--
() The ASCII Ribbon Campaign
/\ Help Cure HTML E-Mail
--
You received this message because you are subscribed to the Google Groups
"sage-release" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/sage-release.
For more options, visit https://groups.google.com/d/optout.
