3.9.8

[Clint Whaley]

Guys,

I have just released 3.9.8.  The main thing is a huge bug fix for machines
where the number of processors isn't a power of 2 (eg., Athlon X3 or
SiCortex MIPS machine).  On such systems, I failed to return the proper
number of processors to use when the general GEMM threading algorithm is
used, which usually means no threading is done!  

I didn't announce it, but I released 3.9.7 just a couple of days ago, and
it fixed a bug in tgemm that could cause seg faults when K is split.

So, if you are using the developer release on a parallel machine, I
highly recommend you install 3.9.8.

Cheers,
Clint

ATLAS 3.9.8 released 02/23/09, changes from 3.9.7
   * Fixed analagous problem in ATL_tsyrk as the 3.9.6 did for ATL_tgemm;
     however, tsyrk bug could not have been exercised by current decomposition.
   * Fixed bug in ATL_Xtgemm where ATL_thrdecompMM failed to return the
     number of processors on non-power-of-2 processor systems
   * Introduced some fixes & workarounds for SiCortex/MIPSICE9:
     - Changed default MIPSICE9 compiler back to gcc, since pathcc produces
       bad ATL_tsyrk when optimization is above -O1 (confirmed compiler error)
   * Added dependence on atlas_ptalias3.h in cblas interface Makefile.
ATLAS 3.9.7 released 02/20/09, changes from 3.9.6:
   * Fixed bug in ATL_tgemm that caused seg faults for some small-M tGEMMs
   * Added architectural defaults for K7323DNow (Athlon "classic")


**************************************************************************
** R. Clint Whaley, PhD ** Assist Prof, UTSA ** www.cs.utsa.edu/~whaley **
**************************************************************************

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Math-atlas-devel mailing list
Math-atlas-devel@...
https://lists.sourceforge.net/lists/listinfo/math-atlas-devel

[Jessica Jones]

Apologies if this is a bit of a stupid question, but I assume that you
are including cores in with processors (I know many people do)?  The
Athlon X3 is I believe a single 3-core processor?

Thanks

Jess

Clint Whaley wrote:

> Guys,
>
> I have just released 3.9.8.  The main thing is a huge bug fix for machines
> where the number of processors isn't a power of 2 (eg., Athlon X3 or
> SiCortex MIPS machine).  On such systems, I failed to return the proper
> number of processors to use when the general GEMM threading algorithm is
> used, which usually means no threading is done!  
>
> I didn't announce it, but I released 3.9.7 just a couple of days ago, and
> it fixed a bug in tgemm that could cause seg faults when K is split.
>
> So, if you are using the developer release on a parallel machine, I
> highly recommend you install 3.9.8.
>
> Cheers,
> Clint
>
> ATLAS 3.9.8 released 02/23/09, changes from 3.9.7
>    * Fixed analagous problem in ATL_tsyrk as the 3.9.6 did for ATL_tgemm;
>      however, tsyrk bug could not have been exercised by current decomposition.
>    * Fixed bug in ATL_Xtgemm where ATL_thrdecompMM failed to return the
>      number of processors on non-power-of-2 processor systems
>    * Introduced some fixes & workarounds for SiCortex/MIPSICE9:
>      - Changed default MIPSICE9 compiler back to gcc, since pathcc produces
>        bad ATL_tsyrk when optimization is above -O1 (confirmed compiler error)
>    * Added dependence on atlas_ptalias3.h in cblas interface Makefile.
> ATLAS 3.9.7 released 02/20/09, changes from 3.9.6:
>    * Fixed bug in ATL_tgemm that caused seg faults for some small-M tGEMMs
>    * Added architectural defaults for K7323DNow (Athlon "classic")
>
>
> **************************************************************************
> ** R. Clint Whaley, PhD ** Assist Prof, UTSA ** www.cs.utsa.edu/~whaley **
> **************************************************************************
>
> ------------------------------------------------------------------------------
> Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
> -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
> -Strategies to boost innovation and cut costs with open source participation
> -Receive a $600 discount off the registration fee with the source code: SFAD
> http://p.sf.net/sfu/XcvMzF8H
> _______________________________________________
> Math-atlas-devel mailing list
> Math-atlas-devel@...
> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>  

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Math-atlas-devel mailing list
Math-atlas-devel@...
https://lists.sourceforge.net/lists/listinfo/math-atlas-devel

[Clint Whaley]

Jess,

>Apologies if this is a bit of a stupid question, but I assume that you
>are including cores in with processors (I know many people do)?  The
>Athlon X3 is I believe a single 3-core processor?

Yeah, some people have started using that terminology, which I'm not a big
fan of.  The X3 is a 3-processor CMP, in the original terminology of
dinosaurs like myself.  I know the current trend is redefine processor
to physical package, and then use core to mean processor (eg, central
*core* unit, I guess now), but this redefinition has never made sense to me.
Therefore, if there is some special distinction to be made, I usually
use the term "physical package", or "socket", and then I usually use
"processor" for each CPU, though I will occasionally call it "core".
So, yes, the discussion below says if you are using multiple "cores"
and/or "processors", in the terminology you prefer . . . .

Regards,
Clint

Clint Whaley wrote:

> Guys,
>
> I have just released 3.9.8.  The main thing is a huge bug fix for machines
> where the number of processors isn't a power of 2 (eg., Athlon X3 or
> SiCortex MIPS machine).  On such systems, I failed to return the proper
> number of processors to use when the general GEMM threading algorithm is
> used, which usually means no threading is done!  
>
> I didn't announce it, but I released 3.9.7 just a couple of days ago, and
> it fixed a bug in tgemm that could cause seg faults when K is split.
>
> So, if you are using the developer release on a parallel machine, I
> highly recommend you install 3.9.8.
>
> Cheers,
> Clint
>
> ATLAS 3.9.8 released 02/23/09, changes from 3.9.7
>    * Fixed analagous problem in ATL_tsyrk as the 3.9.6 did for ATL_tgemm;
>      however, tsyrk bug could not have been exercised by current decomposition.
>    * Fixed bug in ATL_Xtgemm where ATL_thrdecompMM failed to return the
>      number of processors on non-power-of-2 processor systems
>    * Introduced some fixes & workarounds for SiCortex/MIPSICE9:
>      - Changed default MIPSICE9 compiler back to gcc, since pathcc produces
>        bad ATL_tsyrk when optimization is above -O1 (confirmed compiler error)
>    * Added dependence on atlas_ptalias3.h in cblas interface Makefile.
> ATLAS 3.9.7 released 02/20/09, changes from 3.9.6:
>    * Fixed bug in ATL_tgemm that caused seg faults for some small-M tGEMMs
>    * Added architectural defaults for K7323DNow (Athlon "classic")
>
>
> **************************************************************************
> ** R. Clint Whaley, PhD ** Assist Prof, UTSA ** www.cs.utsa.edu/~whaley **
> **************************************************************************
>
> ------------------------------------------------------------------------------
> Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
> -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
> -Strategies to boost innovation and cut costs with open source participation
> -Receive a $600 discount off the registration fee with the source code: SFAD
> http://p.sf.net/sfu/XcvMzF8H
> _______________________________________________
> Math-atlas-devel mailing list
> Math-atlas-devel@...
> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>  

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Math-atlas-devel mailing list
Math-atlas-devel@...
https://lists.sourceforge.net/lists/listinfo/math-atlas-devel


**************************************************************************
** R. Clint Whaley, PhD ** Assist Prof, UTSA ** www.cs.utsa.edu/~whaley **
**************************************************************************

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Math-atlas-devel mailing list
Math-atlas-devel@...
https://lists.sourceforge.net/lists/listinfo/math-atlas-devel

[Jessica Jones]

Clint Whaley wrote:

> Jess,
>
>  
>> Apologies if this is a bit of a stupid question, but I assume that you
>> are including cores in with processors (I know many people do)?  The
>> Athlon X3 is I believe a single 3-core processor?
>>    
>
> Yeah, some people have started using that terminology, which I'm not a big
> fan of.  The X3 is a 3-processor CMP, in the original terminology of
> dinosaurs like myself.  I know the current trend is redefine processor
> to physical package, and then use core to mean processor (eg, central
> *core* unit, I guess now), but this redefinition has never made sense to me.
> Therefore, if there is some special distinction to be made, I usually
> use the term "physical package", or "socket", and then I usually use
> "processor" for each CPU, though I will occasionally call it "core".
> So, yes, the discussion below says if you are using multiple "cores"
> and/or "processors", in the terminology you prefer . . . .
>
> Regards,
> Clint
>
>  

Thanks, I had seen it described that way, but being a young and
inexperienced thing I had not realised that there was a different,
'original terminology'.  (I've heard 'socket' used for something else
entirely, so it is all very confusing!)

Regards

Jess

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Math-atlas-devel mailing list
Math-atlas-devel@...
https://lists.sourceforge.net/lists/listinfo/math-atlas-devel

[Clint Whaley]

>Thanks, I had seen it described that way, but being a young and
>inexperienced thing I had not realised that there was a different,
>'original terminology'.  (I've heard 'socket' used for something else
>entirely, so it is all very confusing!)

I find it so, but from the other end of course :).  

Originally processor was the term that is now being replaced
by "core".  SMP meant "symmetric multiprocessor", which essentially meant
non-NUMA.  Nowadays, SMP is often used to mean "shared-memory-processor",
which includes NUMA.  When multiple processors started appearing on the
same die, we added CMP, for chip multiprocessor.

Then, someone had the brainwave to redefine the word processor (because
the SMP/SMP confusion wasn't adequate, I suppose).  I believe Intel calling
their new chips "Core" was just the logical extension of the devolvement
of our terminology into complete incomprehensability (I eagerly await their
next several lines of architectures, which I suppose will be called
"processor", "architecture" and "chip").  

The first time I was aware of this nifty new terminology overload was when
reviewing a grant proposal, where I noticed that the authors had seemed to
redefine processor in a weird way, and I was fixing to tell them to please use
the standard terminology in their grant applications going forward, when a
friend said that they'd seen that terminology as the new "standard".

Frankly, I hate it.  I've spent 15 years talking about doing the fundamental
unit of computation occuring on the processor, and now I'm supposed to call
that "core", and change the meaning of processor to mean physical package
(one physical package goes into one socket on the motherboard, is where
the 'socket' usage comes from).

Anyway, I realize the boat has already sailed, but I have used "processor"
too long to able to make the change stick now :)

Cheers,
Clint

**************************************************************************
** R. Clint Whaley, PhD ** Assist Prof, UTSA ** www.cs.utsa.edu/~whaley **
**************************************************************************

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Math-atlas-devel mailing list
Math-atlas-devel@...
https://lists.sourceforge.net/lists/listinfo/math-atlas-devel

[Jessica Jones]

Clint Whaley wrote:

>> Thanks, I had seen it described that way, but being a young and
>> inexperienced thing I had not realised that there was a different,
>> 'original terminology'.  (I've heard 'socket' used for something else
>> entirely, so it is all very confusing!)
>
> I find it so, but from the other end of course :).  
>
> Originally processor was the term that is now being replaced
> by "core".  SMP meant "symmetric multiprocessor", which essentially meant
> non-NUMA.  Nowadays, SMP is often used to mean "shared-memory-processor",
> which includes NUMA.  When multiple processors started appearing on the
> same die, we added CMP, for chip multiprocessor.
>
> Then, someone had the brainwave to redefine the word processor (because
> the SMP/SMP confusion wasn't adequate, I suppose).  I believe Intel calling
> their new chips "Core" was just the logical extension of the devolvement
> of our terminology into complete incomprehensability (I eagerly await their
> next several lines of architectures, which I suppose will be called
> "processor", "architecture" and "chip").  
>
> The first time I was aware of this nifty new terminology overload was when
> reviewing a grant proposal, where I noticed that the authors had seemed to
> redefine processor in a weird way, and I was fixing to tell them to please use
> the standard terminology in their grant applications going forward, when a
> friend said that they'd seen that terminology as the new "standard".
>
> Frankly, I hate it.  I've spent 15 years talking about doing the fundamental
> unit of computation occuring on the processor, and now I'm supposed to call
> that "core", and change the meaning of processor to mean physical package
> (one physical package goes into one socket on the motherboard, is where
> the 'socket' usage comes from).
>
> Anyway, I realize the boat has already sailed, but I have used "processor"
> too long to able to make the change stick now :)
>
> Cheers,
> Clint
>
> **************************************************************************
> ** R. Clint Whaley, PhD ** Assist Prof, UTSA ** www.cs.utsa.edu/~whaley **
> **************************************************************************
>

Well, at the moment I have no real allegiance to any of the definitions
so I'm quite happy to use whichever terminology is preferred in a
particular context/forum - as long as I know which dialect we are all
meant to be speaking, it isn't too confusing.  (That is, if I remember ..)

Thanks for the clarification.  I would blame Intel too .. their site is
intensely confusing (eg
http://www.intel.com/technology/itj/2006/volume10issue02/art02_cmp_implementation/p02_intro.htm).

Jess

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Math-atlas-devel mailing list
Math-atlas-devel@...
https://lists.sourceforge.net/lists/listinfo/math-atlas-devel