Re: [atlas-devel] my status

[José L. G. P. <jgp...@gm...>]

2013/9/8 R. Clint Whaley <rcw...@ls...>:
>  > Related to the threads rewrite, could be possible to stole the code
>  > from the openBLAS project (former GotoBLAS). It's now under the BSD
>  > license...
>
> That is interesting, didn't know they had changed the license.  How did
> that happen?  I would have thought that UT-Austin & Goto and a few other
> folks would have to agree?

Hi:

I've read (I don't remember where) that Kazushige Goto is now at
Microsoft (according to wikipedia he is now at Intel since July 2012).
The copyright was owned by the TACC and I suppose when Goto leaves the
group nobody could maintain the library. This is, of course, only my
opinion. The fact is that GotoBLAS is now under the BSD terms:
http://www.tacc.utexas.edu/tacc-projects/gotoblas2/

The active development was then continued by Zhang Xianyi, from the
Institute of Software in the Chinese Academy of Sciences, who renamed
the project to OpenBLAS (https://github.com/xianyi/OpenBLAS).


>
> What features of the OpenBLAS threading model are you interested in?  I
> doubt we could share much of the threading code due to our differing
> frameworks, but we might get some ideas for things to try if they have
> some particularly interesting ideas in this area.

I don't know the internals, but could be possible to copy the way for
the selection of the number of threads to use at execution time?

Of course I agree with you that the principal task is the GEMM rewrite
in ATLAS. How about the develop? Is near completed yet? I understand
also the dynamic selection of threads can not be an  urgent task (I
think lot of people uses ATLAS as single threaded inside other hihger
lever parallelized routines). Another important feature I think is the
optimized support for the near future ARM chips. AMD annouces new ARM
based chips next year:
http://arstechnica.com/information-technology/2013/06/amd-announces-its-first-64-bit-8-and-16-core-arm-based-server-socs/

Cheers

>
> ?
> Clint
>
> On 09/08/2013 04:12 AM, José Luis García Pallero wrote:
>> 2013/9/8 R. Clint Whaley <rcw...@ls...>:
>>> GEMM rewrite is all about performance.  I think ATLAS gets > 90% of peak
>>> performance on the all the newer Intel archs now (from the first Corei
>>> on, AFAIK) due to the rewrite.  Theoretically, I think the new framework
>>> should scale better, but that has not been proven.
>>>
>>> Haswell has at least decent support (I haven't tuned kernel crazy-like,
>>> but I recall it doing quite well nonetheless).  It may not be >90%, but
>>> I think its close if not (can't remember).
>>>
>>> As for the threads rewrite, that is strictly after the GEMM rewrite is
>>> finished.  Whether I'll be able to encompass all requested features in a
>>> threading rewrite will remain to be seen.  I have ideas on how it'll
>>> work, but often they cannot withstand the harsh contact with reality . . .
>>
>>
>> Hello:
>>
>> Related to the threads rewrite, could be possible to stole the code
>> from the openBLAS project (former GotoBLAS). It's now under the BSD
>> license...
>>
>> Cheers
>>
>>>
>>> Cheers,
>>> Clint
>>>
>>> On 09/07/2013 05:07 PM, José Luis García Pallero wrote:
>>>> I wish you good luck in your new job, Clint.
>>>>
>>>> About ATLAS, I suppose that the long term requested feature of
>>>> selection of threads at runtime is now far :(. And the GEMM rewrite,
>>>> how will improve the performance? I suppose also that an important
>>>> feature is the AVX2 instructions present in Haswell.
>>>>
>>>> Cheers
>>>>
>>>> 2013/9/7 R. Clint Whaley <rcw...@ls...>:
>>>>> Guys,
>>>>>
>>>>> You will have noticed no ATLAS release in a while, and almost no help on
>>>>> the support list for even longer.  This is primarily due to stuff going
>>>>> on with me, which I thought I might ought to share for those of you
>>>>> frustrated with the pace.
>>>>>
>>>>> I wound up having to leave UTSA due to administration hamstringing of
>>>>> the department, and had to spend a lot of my personal time I normally
>>>>> spend on ATLAS applying for positions.  Fortunately for me, the academic
>>>>> season just completed was a hot one for HPC, and I was able to find
>>>>> several very attractive offers for the kinds of research I like.
>>>>>
>>>>> It was a tough decision where to go, but I eventually accepted a
>>>>> position at LSU with a joint appointment in their computer science
>>>>> department and a lab-style place called CCT (Center for Computation &
>>>>> Technology):
>>>>>        http://www.csc.lsu.edu/
>>>>>        http://www.cct.lsu.edu/
>>>>>
>>>>> Through CCT, I have the opportunity to work with a lot of computational
>>>>> users that I have never had before (CCT has physicists, chemists,
>>>>> engineers, mathematicians and so forth; pretty much every discipline
>>>>> that uses ATLAS), in addition to my normal collaborations within
>>>>> computer science (and here at LSU, with folks in computer engineering as
>>>>> well).
>>>>>
>>>>> I've just finished the move, and still have an office not completely set
>>>>> up, and am getting my feet under me at LSU, which has kept my ATLAS
>>>>> productivity low.
>>>>>
>>>>> It is even worse news on the support front.  Due to the move, I lost the
>>>>> DoD contract that I used to justify the time spent doing pure software
>>>>> development and support (as opposed to research).  I'm pretty much
>>>>> supporting ATLAS on my CAREER grant & LSU startup right now.  However,
>>>>> at least for the short-term, Tony Castaldo (Tony got his PhD with me
>>>>> back in 2010 working on ATLAS-related topics) has assumed that contract.
>>>>>     So, we are going to try to have Tony help with support (since he can
>>>>> justify it funding-wise in ways I cannot), but ATLAS is a deep & wide
>>>>> package, so that will take time to get up and going.
>>>>>
>>>>> What we are trying to do is for Tony to take initial cracks at support
>>>>> questions, and call me in when things really get stuck.  Because of the
>>>>> large backlog, it will probably be a while before we get back to the
>>>>> responsiveness we had before all this happened (and yes, I realize that
>>>>> that wasn't all that great either :)
>>>>>
>>>>> I was on very low responsiveness even before all this, just due to the
>>>>> volume of work required by the GEMM rewrite that I'm presently pursuing;
>>>>> that is very important work for the future of ATLAS and must get done,
>>>>> so I was using all my spare time to push it forward, and this means we
>>>>> are even more behind on support than even all the above would indicate.
>>>>>     I was only responding to questions that I was clear were likely ATLAS
>>>>> bugs that the user couldn't get around by the themselves, and even there
>>>>> my lack of time means I'm sure we have some of those.
>>>>>
>>>>> Anyway, that's what's been going on with me, and hopefully the pace of
>>>>> ATLAS development will get back on track, and we'll slowly start getting
>>>>> through some support stuff as times go on.
>>>>>
>>>>> Cheers,
>>>>> Clint
>>>>>
>>>>> --
>>>>> **********************************************************************
>>>>> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
>>>>> **********************************************************************
>>>>>
>>>>>
>>>>> ------------------------------------------------------------------------------
>>>>> Learn the latest--Visual Studio 2012, SharePoint 2013, SQL 2012, more!
>>>>> Discover the easy way to master current and previous Microsoft technologies
>>>>> and advance your career. Get an incredible 1,500+ hours of step-by-step
>>>>> tutorial videos with LearnDevNow. Subscribe today and save!
>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=58041391&iu=/4140/ostg.clktrk
>>>>> _______________________________________________
>>>>> Math-atlas-devel mailing list
>>>>> Mat...@li...
>>>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>>>
>>>>
>>>>
>>>
>>> --
>>> **********************************************************************
>>> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
>>> **********************************************************************
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Learn the latest--Visual Studio 2012, SharePoint 2013, SQL 2012, more!
>>> Discover the easy way to master current and previous Microsoft technologies
>>> and advance your career. Get an incredible 1,500+ hours of step-by-step
>>> tutorial videos with LearnDevNow. Subscribe today and save!
>>> http://pubads.g.doubleclick.net/gampad/clk?id=58041391&iu=/4140/ostg.clktrk
>>> _______________________________________________
>>> Math-atlas-devel mailing list
>>> Mat...@li...
>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>
>>
>>
>
> --
> **********************************************************************
> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
> **********************************************************************
>
>
> ------------------------------------------------------------------------------
> Learn the latest--Visual Studio 2012, SharePoint 2013, SQL 2012, more!
> Discover the easy way to master current and previous Microsoft technologies
> and advance your career. Get an incredible 1,500+ hours of step-by-step
> tutorial videos with LearnDevNow. Subscribe today and save!
> http://pubads.g.doubleclick.net/gampad/clk?id=58041391&iu=/4140/ostg.clktrk
> _______________________________________________
> Math-atlas-devel mailing list
> Mat...@li...
> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel



-- 
*****************************************
José Luis García Pallero
jgp...@gm...
(o<
/ / \
V_/_
Use Debian GNU/Linux and enjoy!
*****************************************

2013/9/8 R. Clint Whaley <rcw...@ls...>:
>  > Related to the threads rewrite, could be possible to stole the code
>  > from the openBLAS project (former GotoBLAS). It's now under the BSD
>  > license...
>
> That is interesting, didn't know they had changed the license.  How did
> that happen?  I would have thought that UT-Austin & Goto and a few other
> folks would have to agree?
>
> What features of the OpenBLAS threading model are you interested in?  I
> doubt we could share much of the threading code due to our differing
> frameworks, but we might get some ideas for things to try if they have
> some particularly interesting ideas in this area.
>
> ?
> Clint
>
> On 09/08/2013 04:12 AM, José Luis García Pallero wrote:
>> 2013/9/8 R. Clint Whaley <rcw...@ls...>:
>>> GEMM rewrite is all about performance.  I think ATLAS gets > 90% of peak
>>> performance on the all the newer Intel archs now (from the first Corei
>>> on, AFAIK) due to the rewrite.  Theoretically, I think the new framework
>>> should scale better, but that has not been proven.
>>>
>>> Haswell has at least decent support (I haven't tuned kernel crazy-like,
>>> but I recall it doing quite well nonetheless).  It may not be >90%, but
>>> I think its close if not (can't remember).
>>>
>>> As for the threads rewrite, that is strictly after the GEMM rewrite is
>>> finished.  Whether I'll be able to encompass all requested features in a
>>> threading rewrite will remain to be seen.  I have ideas on how it'll
>>> work, but often they cannot withstand the harsh contact with reality . . .
>>
>>
>> Hello:
>>
>> Related to the threads rewrite, could be possible to stole the code
>> from the openBLAS project (former GotoBLAS). It's now under the BSD
>> license...
>>
>> Cheers
>>
>>>
>>> Cheers,
>>> Clint
>>>
>>> On 09/07/2013 05:07 PM, José Luis García Pallero wrote:
>>>> I wish you good luck in your new job, Clint.
>>>>
>>>> About ATLAS, I suppose that the long term requested feature of
>>>> selection of threads at runtime is now far :(. And the GEMM rewrite,
>>>> how will improve the performance? I suppose also that an important
>>>> feature is the AVX2 instructions present in Haswell.
>>>>
>>>> Cheers
>>>>
>>>> 2013/9/7 R. Clint Whaley <rcw...@ls...>:
>>>>> Guys,
>>>>>
>>>>> You will have noticed no ATLAS release in a while, and almost no help on
>>>>> the support list for even longer.  This is primarily due to stuff going
>>>>> on with me, which I thought I might ought to share for those of you
>>>>> frustrated with the pace.
>>>>>
>>>>> I wound up having to leave UTSA due to administration hamstringing of
>>>>> the department, and had to spend a lot of my personal time I normally
>>>>> spend on ATLAS applying for positions.  Fortunately for me, the academic
>>>>> season just completed was a hot one for HPC, and I was able to find
>>>>> several very attractive offers for the kinds of research I like.
>>>>>
>>>>> It was a tough decision where to go, but I eventually accepted a
>>>>> position at LSU with a joint appointment in their computer science
>>>>> department and a lab-style place called CCT (Center for Computation &
>>>>> Technology):
>>>>>        http://www.csc.lsu.edu/
>>>>>        http://www.cct.lsu.edu/
>>>>>
>>>>> Through CCT, I have the opportunity to work with a lot of computational
>>>>> users that I have never had before (CCT has physicists, chemists,
>>>>> engineers, mathematicians and so forth; pretty much every discipline
>>>>> that uses ATLAS), in addition to my normal collaborations within
>>>>> computer science (and here at LSU, with folks in computer engineering as
>>>>> well).
>>>>>
>>>>> I've just finished the move, and still have an office not completely set
>>>>> up, and am getting my feet under me at LSU, which has kept my ATLAS
>>>>> productivity low.
>>>>>
>>>>> It is even worse news on the support front.  Due to the move, I lost the
>>>>> DoD contract that I used to justify the time spent doing pure software
>>>>> development and support (as opposed to research).  I'm pretty much
>>>>> supporting ATLAS on my CAREER grant & LSU startup right now.  However,
>>>>> at least for the short-term, Tony Castaldo (Tony got his PhD with me
>>>>> back in 2010 working on ATLAS-related topics) has assumed that contract.
>>>>>     So, we are going to try to have Tony help with support (since he can
>>>>> justify it funding-wise in ways I cannot), but ATLAS is a deep & wide
>>>>> package, so that will take time to get up and going.
>>>>>
>>>>> What we are trying to do is for Tony to take initial cracks at support
>>>>> questions, and call me in when things really get stuck.  Because of the
>>>>> large backlog, it will probably be a while before we get back to the
>>>>> responsiveness we had before all this happened (and yes, I realize that
>>>>> that wasn't all that great either :)
>>>>>
>>>>> I was on very low responsiveness even before all this, just due to the
>>>>> volume of work required by the GEMM rewrite that I'm presently pursuing;
>>>>> that is very important work for the future of ATLAS and must get done,
>>>>> so I was using all my spare time to push it forward, and this means we
>>>>> are even more behind on support than even all the above would indicate.
>>>>>     I was only responding to questions that I was clear were likely ATLAS
>>>>> bugs that the user couldn't get around by the themselves, and even there
>>>>> my lack of time means I'm sure we have some of those.
>>>>>
>>>>> Anyway, that's what's been going on with me, and hopefully the pace of
>>>>> ATLAS development will get back on track, and we'll slowly start getting
>>>>> through some support stuff as times go on.
>>>>>
>>>>> Cheers,
>>>>> Clint
>>>>>
>>>>> --
>>>>> **********************************************************************
>>>>> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
>>>>> **********************************************************************
>>>>>
>>>>>
>>>>> ------------------------------------------------------------------------------
>>>>> Learn the latest--Visual Studio 2012, SharePoint 2013, SQL 2012, more!
>>>>> Discover the easy way to master current and previous Microsoft technologies
>>>>> and advance your career. Get an incredible 1,500+ hours of step-by-step
>>>>> tutorial videos with LearnDevNow. Subscribe today and save!
>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=58041391&iu=/4140/ostg.clktrk
>>>>> _______________________________________________
>>>>> Math-atlas-devel mailing list
>>>>> Mat...@li...
>>>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>>>
>>>>
>>>>
>>>
>>> --
>>> **********************************************************************
>>> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
>>> **********************************************************************
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Learn the latest--Visual Studio 2012, SharePoint 2013, SQL 2012, more!
>>> Discover the easy way to master current and previous Microsoft technologies
>>> and advance your career. Get an incredible 1,500+ hours of step-by-step
>>> tutorial videos with LearnDevNow. Subscribe today and save!
>>> http://pubads.g.doubleclick.net/gampad/clk?id=58041391&iu=/4140/ostg.clktrk
>>> _______________________________________________
>>> Math-atlas-devel mailing list
>>> Mat...@li...
>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>
>>
>>
>
> --
> **********************************************************************
> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
> **********************************************************************
>
>
> ------------------------------------------------------------------------------
> Learn the latest--Visual Studio 2012, SharePoint 2013, SQL 2012, more!
> Discover the easy way to master current and previous Microsoft technologies
> and advance your career. Get an incredible 1,500+ hours of step-by-step
> tutorial videos with LearnDevNow. Subscribe today and save!
> http://pubads.g.doubleclick.net/gampad/clk?id=58041391&iu=/4140/ostg.clktrk
> _______________________________________________
> Math-atlas-devel mailing list
> Mat...@li...
> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel



-- 
*****************************************
José Luis García Pallero
jgp...@gm...
(o<
/ / \
V_/_
Use Debian GNU/Linux and enjoy!
*****************************************