Atlas on MacPro6,1 (Quad-Core Intel Xeon E5) fails

[fultz@…]

port install atlas +nofortran

fails on MacPro6,1. Here is the output:

sudo port install atlas +nofortran
--->  Computing dependencies for atlas
--->  Fetching archive for atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33+nofortran.darwin_13.x86_64.tbz2 from http://packages.macports.org/atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33+nofortran.darwin_13.x86_64.tbz2 from http://lil.fr.packages.macports.org/atlas
--->  Attempting to fetch atlas-3.10.1_5+mpclang33+nofortran.darwin_13.x86_64.tbz2 from http://mse.uk.packages.macports.org/sites/packages.macports.org/atlas
--->  Fetching distfiles for atlas
--->  Verifying checksums for atlas
--->  Extracting atlas
--->  Applying patches to atlas
--->  Configuring atlas
Selected C compiler: /opt/local/bin/clang-mp-3.3
--->  Building atlas
Error: org.macports.build for port atlas returned: command execution failed
Please see the log file for port atlas for details:
    /opt/local/var/macports/logs/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_math_atlas/atlas/main.log
To report a bug, follow the instructions in the guide:
    http://guide.macports.org/#project.tickets
Error: Processing of port atlas failed

The CPU is not identified properly by the atlas configure script: "Architecture configured as UNKNOWNx86 (35)"

[Veence (Vincent)]

That's the new MacPro, isn't it? If yes, then I can handle you a private Portfile for the latest unstable release. I think it'll work, but, up to that point, I've failed to get any advice from upstream as of the usability and reliability of the code. At your own risks.

[fultz@…]

Yessir, it is the new MacPro (trash can).

[Veence (Vincent)]

Well, well, well, nice machine, kudos… Regarding your difficulty, either you can try with the makeshift/workaround I suggested, the other way being to rely on the OS X provided Accelerate framework for whatever other port you want to install (e.g. octave or scipy). Tell me. I don't hide from you I'd be very interested if you could compile the experimental version…

[fultz@…]

What makeshift/workaround?

[Veence (Vincent)]

Well, as I said, I can post here an experimental Portfile that would boost you to 3.11.28, latest unstable release, that works with my Haswell-based MacBook Air.

Air > port installed | grep atlas
  atlas @3.11.28_0+perf (active)

[fultz@…]

Please post, thanks.

[Veence (Vincent)]

Portfile for unstable 3.11 release

[Veence (Vincent)]

Copy the attached Portfile in /opt/local/var/macports/sources/rsync.macports.org/release/tarballs/ports/math/atlas/ Then try again and tell me.

[Veence (Vincent)]

Since I have no news, I surmise it worked, didn't it?

[fultz@…]

main.log for atlas-3.11.28

[fultz@…]

Sadly, no. It didn't work. It still seems to be an issue with CPU identification.

[Veence (Vincent)]

Ooops sorry about that. I'll be away for three days without computer, but I'll tackle this snag next week. Thanks for your patience.

[Veence (Vincent)]

Portfile + patch

[Veence (Vincent)]

Please can you retry with the new files I uploaded? There is a slightly modified Portfile and a patch that should go in the "files" sub-directory of …/…/math/atlas. Sorry for being a little sluggish and thanks for your patience.

[fultz@…]

New atlas-3.11.28_0+clang+nofortran.darwin_13.x86_64.main.log

[fultz@…]

Essentially the same error.

[Veence (Vincent)]

Not quite. Granted, there is something wrong with FMA, that I'll try to fix. But note that the CPU architecture is now correctly detected, which is a step forward.

[Veence (Vincent)]

Can you post me the result of:

sysctl hw.optional

in a Terminal window? Thanks.

[fultz@…]

Replying to vince@…:

Not quite. Granted, there is something wrong with FMA, that I'll try to fix. But note that the CPU architecture is now correctly detected, which is a step forward.

If this machine really has a Corei3 CPU, Apple has a LOT of explaining to do.

Can you post me the result of:

sysctl hw.optional

in a Terminal window? Thanks.

% sysctl hw.optional
hw.optional.floatingpoint: 1
hw.optional.mmx: 1
hw.optional.sse: 1
hw.optional.sse2: 1
hw.optional.sse3: 1
hw.optional.supplementalsse3: 1
hw.optional.sse4_1: 1
hw.optional.sse4_2: 1
hw.optional.x86_64: 1
hw.optional.aes: 1
hw.optional.avx1_0: 1
hw.optional.rdrand: 1
hw.optional.f16c: 1
hw.optional.enfstrg: 1
hw.optional.fma: 0
hw.optional.avx2_0: 0
hw.optional.bmi1: 0
hw.optional.bmi2: 0
hw.optional.rtm: 0
hw.optional.hle: 0

[Veence (Vincent)]

It's obviously NOT a Core i3, since it has neither AVX2 nor (consequently) FMA. It's a Core-i2 (Ivy bridge) and my assumption was wrong. I can't believe this machine has no Haswell CPU. But, eh.

Okay, I'm going to fix it in a new version of the patch tonight. If you can wait a couple of hours more… Thanks for your patience!

[fultz@…]

Replying to vince@…:

Okay, I'm going to fix it in a new version of the patch tonight. If you can wait a couple of hours more… Thanks for your patience!

I'm at your mercy :) No rush, and thanks for your work!

[Veence (Vincent)]

New version of the patch

[Veence (Vincent)]

Can you retry with the new version of the patch I just uploaded? Thanks.

[fultz@…]

--->  No broken files found.

Success! Thanks a lot. Consider this resolved.

[Veence (Vincent)]

Great! I'll hand the patch over upstream. Thanks for your patience again!