Recent changes to feature-requests

RFE: migrate VCS to git

Tomasz Kłoczko — Wed, 07 Sep 2022 16:36:40 -0000

IMO it woild be good to migrate VCS from current VCS to git
Best to github or gitlab.
sf.net no longer provide web frontend to CVS and by this it is harder to track changes.

#85 Please consider adding extern "C" block to headers for C++ users

Ryan Livingston — Fri, 13 Apr 2018 14:41:36 -0000

Great to hear! Thanks Clint!

#85 Please consider adding extern "C" block to headers for C++ users

R. Clint Whaley — Fri, 13 Apr 2018 13:57:19 -0000

assigned_to: R. Clint Whaley
Group: Dreaming --> Next_developer_release
Priority: 5 --> 6

#85 Please consider adding extern "C" block to headers for C++ users

R. Clint Whaley — Fri, 13 Apr 2018 13:56:52 -0000

Thank you for careful writeup with historical links!

This seems reasonable to me, so I'm going to scope adding it to the next developer release. Assuming that doesn't trigger problems, it can then go to next stable.

Please consider adding extern "C" block to headers for C++ users

Ryan Livingston — Thu, 12 Apr 2018 20:09:24 -0000

Thanks for investing the time to develop ATLAS. I work on a product that generates C and C++ code from a higher level language that includes matrix operations. To help optimize our generated code, we're looking into generating calls to BLAS libraries (via the CBLAS interface).

When testing with various libraries and language configurations, we came across the fact that the ATLAS cblas.h header doesn't declare functions as extern "C" for C++ compilation. This will cause a difficult to diagnose linker failure for our C++ users who decide to try ATLAS when this feature ships. C++ is becoming more of a focus for our customers, and I'd love to be able to make integration of various BLAS libraries, including ATLAS, as simple as possible for those customers. We can certainly document the workaround you've shown in the errata to minimize user pain.

I came across:

http://math-atlas.sourceforge.net/errata.html#CPPheaders

and thought I'd put in a vote for adding and extern "C" block protected by #ifdef __cplusplus checks to the headers.

The last request for this I could find was updated in 2009:

https://sourceforge.net/p/math-atlas/support-requests/550/

My arguments for the addition of extern "C" are:
As that requestor noted, debugging the absence of extern "C" is tricky since it manifests as a link error because of name mangling differences between C and C++.
Many other CBLAS headers provide this affordance including OpenBLAS, the Netlib reference BLAS (http://www.netlib.org/blas/cblas.h), and Intel MKL. Adding it would improve interchangeability and allow code like the code that we generate to remain mostly agnostic of which BLAS library is being used.
* I'm not aware of any functional downsides to existing C users if the extern "C" is protected by #ifdef guards. For existing C++ users who have already added their own extern "C" wrappers, it looks like extern specifiers nest and prefer the innermost specifier:
http://en.cppreference.com/w/cpp/language/language_linkage
so this change shouldn't break them either.

Thanks for the consideration.

Please consider adding extern "C" block to headers for C++ users

Ryan Livingston — Thu, 12 Apr 2018 20:09:24 -0000

Ticket 85 has been modified: Please consider adding extern "C" block to headers for C++ users
Edited By: R. Clint Whaley (rwhaley)
_priority updated: u'5' => u'6'
_milestone updated: u'Dreaming' => u'Next_developer_release'
Owner updated: None => u'rwhaley'

Matrix Indexing

Steven Evans — Tue, 19 Sep 2017 01:09:27 -0000

Many years ago, I fell in love with the MATLAB programming language. In order to keep software capability up-to-date with my ever advancing hardware, I primarily switched to Octave about two years ago.

I've recently begun a project that requires me to work with large data sets. After learning that Octave is single-threaded software, I discovered that some of its basic operations could be allowed to use multiple cores if I compile it from source with ATLAS. Due to the increased size of my data sets, a multi-threaded process is what I need.

I've learned that ATLAS provides linear operations on matricies, such as addition and different forms of multiplication. However, my understanding is that matrix indexing isn't included in its set of operations. For example, if "A" and "B" are both matricies of the same type and size, and "b" is also the same size, but is a logical array of zeros and ones, then matrix indexing could be the following operation: B(b) = A(b). The concept could also extend to "b" being an array of numerical indicies.

I've recently learned that the backbone of MATLAB is based on many FORTRAN concepts, which are also linked to BLAS, and also ATLAS. Matrix manipulation is the underlying, unique power of all of these software packages. I think it would be great if ATLAS was able to provide multi-threaded matrix indexing along with its other matrix operations. This would vastly open the door for further development for all of us who use your useful program.

In the meantime, I've been looking for a program which does what I've described. The last thing I want to do is re-invent the wheel if someone's already done this. If matrix indexing is something that isn't likely to be included with ATLAS, are you aware of any other programs which may have this feature? My continued goal is to compile Octave, with such a package, so that I could use it with multi-threading capability.

I want to take the time to thank you for developing ATLAS and offering it to the world. Because of people like you, I'm able to use software packages like ATLAS and Octave in order to continue my work. This means a lot to me. Thank you!

#83 POWER8 VSX tuning

Fabricio — Thu, 05 Mar 2015 23:38:15 -0000

Hi Clint.

I don't know if you're aware of this patch:

sourceforge.net/p/math-atlas/patches/65/attachment/atlas.ppc64le-abiv2.patch

#83 POWER8 VSX tuning

R. Clint Whaley — Thu, 05 Mar 2015 00:33:27 -0000

Fabricio,

I've been talking with David in order to get access to POWER8. If everything goes according to plan, I should do some work here very soon, but nothing so far.

Regards,
Clint

#83 POWER8 VSX tuning

Fabricio — Wed, 04 Mar 2015 23:48:20 -0000

Hello!

is there any news about this issue?

I'm about to get hold of an IBM POWER8 machine and would like very much to use ATLAS in my tests.

TIA,
Fabricio