Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 05 Mar 2016 21:22:21 +0100
From:      Raphael Kubo da Costa <rakuco@freebsd.org>
To:        Andrew Reilly <areilly@bigpond.net.au>
Cc:        ports@freebsd.org
Subject:   Re: openblas 0.2.15,1 port broke build for BARCELONA AMD core
Message-ID:  <86d1r87k02.fsf@FreeBSD.org>
In-Reply-To: <7251C627-A9CC-49F8-A12D-1412D77493EF@bigpond.net.au> (Andrew Reilly's message of "Sat, 27 Feb 2016 12:30:33 %2B1100")
References:  <7251C627-A9CC-49F8-A12D-1412D77493EF@bigpond.net.au>

next in thread | previous in thread | raw e-mail | index | archive | help
Andrew Reilly <areilly@bigpond.net.au> writes:
> Hi there,
>
> My portmsaster run just broke on the recent openblas upgrade, with version
> stamp in head/math/openblas/Makefile 409114 26-02-18 16:35:48Z rakuco. No=
w that
> I=E2=80=99ve hacked on this to make it build, I remember having done some=
thing like
> this the last time, so perhaps nothing has actually changed in this respe=
ct,
> there=E2=80=99s just been a version increase that still doesn=E2=80=99t k=
now about my CPU.
>
> My build/target machine is, according to sysctl:
> hw.machine: amd64
> hw.model: AMD A6-3500 APU with Radeon(tm) HD Graphics
> hw.ncpu: 3
>
> The auto-config system for openblas has determined that this is (in confi=
g.h):
> #define AMD_UNKNOWN
> but also
> #define CORE_BARCELONA
> #define CHAR_CORENAME =E2=80=9CBARCELONA=E2=80=9D
>
> So on the assumption that the AMD_UNKNOWN should be BARCELONA, I have twe=
aked
> cpuid_x86.c to add case 3 along with case 1 and case 10 in the switch(exf=
amily)
> to return CPUTYPE_BARCELONA.
>
> I=E2=80=99m not sure if it wouldn=E2=80=99t have been better to slide tha=
t in with case 6,
> because model=3D1 and support_avx()->0, so that would have worked too.
>
> the full set of get_cputype() values for my machine are:
> family=3D0xf, exfamily=3D0x3, model=3D0x1, exmodel=3D0x0, support_avx=3D0
>
> The code in cpuid_x86.c thinks that NUM_CORES is 1, despite the sysctl
> information above, so I guess that only one core is going to be used, but=
 since
> this is not a high performance machine and is generally thermally challen=
ged,
> I=E2=80=99m not too concerned about that. Just want it to build at all.
>
> With this couple of tweaks the math/openblas port seems to have built OK,=
 so I
> hope that this change can be included in the port or the up-stream so tha=
t I
> don=E2=80=99t have to remember to do this all again next time!

Hi Andrew,

I don't maintain the openblas port, I just committed a patch for it that
was sent to Bugzilla.

This looks like a problem that should be reported upstream
(https://github.com/xianyi/OpenBLAS/issues), it's even possible it's
already been fixed in their git repository.

You can also file a bug in our bug tracker so the port maintainer can
take a look at this, but I don't think there's much he can do besides
relaying that information upstream.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?86d1r87k02.fsf>