port-i386/46108: some math tests works on qemu but fail on real hardware

[bouyer%antioche.eu.org@localhost]

>Number:         46108
>Category:       port-i386
>Synopsis:       some math tests works on qemu but fail on real hardware
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    port-i386-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon Feb 27 23:05:00 +0000 2012
>Originator:     bouyer%antioche.eu.org@localhost
>Release:        NetBSD 6.0_BETA
>Organization:
>Environment:
System: NetBSD pop.soc.lip6.fr 6.0_BETA NetBSD 6.0_BETA (XEN3PAE_DOMU) #1: Mon 
Feb 27 23:05:19 MET 2012 
bouyer%pop.soc.lip6.fr@localhost:/dsk/l1/misc/bouyer/tmp/i386/obj/dsk/l1/misc/bouyer/netbsd-6/src/sys/arch/i386/compile/XEN3PAE_DOMU
 i386
Architecture: i386
Machine: i386
>Description:
        While running tests on Xen, I noticed some failed math tests 
        which don't seem to fail on a anita/qemu run, nor amd64 (xen or qemu).
        Theses are:
        lib/libm/t_atan:atan_inf_neg
        lib/libm/t_atan:atan_inf_pos
        lib/libm/t_atan:atan_tan

        Just to make sure, I tried on a plain NetBSD/i386 host too (no
        Xen involved), they fails in the same way. The i386 binary also
        fails on amd64, when with the amd64 binary the test passes.

        It could be a compiler bug: when the test is recompiled by
        hang without -O2 (e.g. gcc -o t_atan t_atan.c -lm -latf-c),
        it passes. Adding -O2 to the compilation line makes it fails again.

>How-To-Repeat:
        cd /usr/tests/lib/libm; atf-run |atf-report
>Fix: