MULX on haswell triggers ILLOP

Torbjorn Granlund tg at
Fri Oct 25 09:37:08 UTC 2013

Tijl Coosemans <tijl at> writes:

  A comment on that PR says that it has been fixed in
  I don't know if it's feasible to merge this to stable/9.  Committer CCed.
Thanks for your reply.

You should notice that the PR is in the "open" state.  That usually
means that it has not been analysed, and that no action has been taken.

If the PR is taken care of, one would assume that either,

(1) the bug would be fixed.  I'd say that'd need 10 minutes for a
    reasonably talented hacker, or

(2) that the documentation is updated to reflect that argument 2 and 3
    are ignored.

That >= 4 FreeBSD releases have been made since this became known makes
me worry about how FreeBSD is maintained these days.  I cannot believe
that they left the bug there without at least fixing the documentation.

I have decided to not implement further workarounds for GMP; FreeBSD
needs to address its own bugs.  GMP 5.2 and later running on an Intel
Haswell will need NetBSD, OpenBSD, Solaris, GNU/Linux, or even Windoze
to operate.  On FreeBSD a 'make check' looks like this:

PASS: t-asmtype
Illegal instruction
FAIL: t-aors_1
Illegal instruction
FAIL: t-divrem_1
Illegal instruction
FAIL: t-mod_1
Illegal instruction
FAIL: t-fat
Illegal instruction
FAIL: t-get_d
PASS: t-instrument
Illegal instruction
FAIL: t-iord_u
PASS: t-mp_bases
Illegal instruction
FAIL: t-perfsqr
Illegal instruction
FAIL: t-scan
Illegal instruction
FAIL: logic


More information about the freebsd-questions mailing list