Re: [OMPI users] Use of __float128 with openmpi

[Patrick Boehl]

Hello Jeff,

thank you a lot for your reply!

On 01.02.2014, at 23:07, Jeff Hammond wrote:

> See Section 5.9.5 of MPI-3 or the section named "User-Defined
> Reduction Operations" but presumably numbered differently in older
> copies of the MPI standard.
> 
> An older but still relevant online reference is
> http://www.mpi-forum.org/docs/mpi-2.2/mpi22-report/node107.htm
> 

In this example they construct this "datatype"

- 
typedef struct {
double real,imag;
} Complex
-

and later

-
MPI_Datatype ctype;
/* explain to MPI how type Complex is defined
*/
MPI_Type_contiguous(2, MPI_DOUBLE, &ctype);
-

Do I understand correctly that I have to find out how __float128 is constructed 
internally and 
convert it to a form which is compatible with the standard MPI Datatypes?
In an analogue way as they do in the example. Up to now, I only found out that 
__float128 should 
be somehow the sum of two doubles.

Again, I am grateful for any help!

Best regards,
Patrick




> On Sat, Feb 1, 2014 at 2:28 PM, Tim Prince  wrote:
>> 
>> On 02/01/2014 12:42 PM, Patrick Boehl wrote:
>>> 
>>> Hi all,
>>> 
>>> I have a question on datatypes in openmpi:
>>> 
>>> Is there an (easy?) way to use __float128 variables with openmpi?
>>> 
>>> Specifically, functions like
>>> 
>>> MPI_Allreduce
>>> 
>>> seem to give weird results with __float128.
>>> 
>>> Essentially all I found was
>>> 
>>> http://beige.ucs.indiana.edu/I590/node100.html
>>> 
>>> where they state
>>> 
>>> MPI_LONG_DOUBLE
>>>   This is a quadruple precision, 128-bit long floating point number.
>>> 
>>> 
>>> But as far as I have seen, MPI_LONG_DOUBLE is only used for long doubles.
>>> 
>>> The Open MPI Version is 1.6.3 and gcc is 4.7.3 on a x86_64 machine.
>>> 
>> It seems unlikely that 10 year old course notes on an unspecified MPI
>> implementation (hinted to be IBM power3) would deal with specific details of
>> openmpi on a different architecture.
>> Where openmpi refers to "portable C types" I would take long double to be
>> the 80-bit hardware format you would have in a standard build of gcc for
>> x86_64.  You should be able to gain some insight by examining your openmpi
>> build logs to see if it builds for both __float80 and __float128 (or
>> neither).  gfortran has a 128-bit data type (software floating point
>> real(16), corresponding to __float128); you should be able to see in the
>> build logs whether that data type was used.
>> 
>> 
>> ___
>> users mailing list
>> us...@open-mpi.org
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
> 
> 
> 
> -- 
> Jeff Hammond
> jeff.scie...@gmail.com
> ___
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users

Re: [OMPI users] Use of __float128 with openmpi

[Patrick Boehl]

Hello Tim,

thank you for your reply!

On 01.02.2014, at 21:28, Tim Prince wrote:

> Where openmpi refers to "portable C types" I would take long double to be the 
> 80-bit hardware format you would have in a standard build of gcc for x86_64.  
> You should be able to gain some insight by examining your openmpi build logs 
> to see if it builds for both __float80 and __float128 (or neither).  

Unfortunately, these are not my machines, so I do not have access to any build 
logs. But I will talk to the admins and/or 
I try to compile openmpi myself. 


Best regards,
Patrick

Re: [OMPI users] Use of __float128 with openmpi

[George Bosilca]

Just go for the most trivial:

MPI_Type_contiguous(sizeof(__float128), MPI_BYTE, &my__float128);

A little bit more info about the optional quad-precision floating-point format 
is available on Wikipedia 
(https://en.wikipedia.org/wiki/Double-double_%28arithmetic%29#Double-double_arithmetic).

  George.


On Feb 2, 2014, at 13:41 , Patrick Boehl  
wrote:

> Hello Jeff,
> 
> thank you a lot for your reply!
> 
> On 01.02.2014, at 23:07, Jeff Hammond wrote:
> 
>> See Section 5.9.5 of MPI-3 or the section named "User-Defined
>> Reduction Operations" but presumably numbered differently in older
>> copies of the MPI standard.
>> 
>> An older but still relevant online reference is
>> http://www.mpi-forum.org/docs/mpi-2.2/mpi22-report/node107.htm
>> 
> 
> In this example they construct this "datatype"
> 
> - 
> typedef struct {
> double real,imag;
> } Complex
> -
> 
> and later
> 
> -
> MPI_Datatype ctype;
> /* explain to MPI how type Complex is defined
> */
> MPI_Type_contiguous(2, MPI_DOUBLE, &ctype);
> -
> 
> Do I understand correctly that I have to find out how __float128 is 
> constructed internally and 
> convert it to a form which is compatible with the standard MPI Datatypes?
> In an analogue way as they do in the example. Up to now, I only found out 
> that __float128 should 
> be somehow the sum of two doubles.
> 
> Again, I am grateful for any help!
> 
> Best regards,
> Patrick
> 
> 
> 
> 
>> On Sat, Feb 1, 2014 at 2:28 PM, Tim Prince  wrote:
>>> 
>>> On 02/01/2014 12:42 PM, Patrick Boehl wrote:
 
 Hi all,
 
 I have a question on datatypes in openmpi:
 
 Is there an (easy?) way to use __float128 variables with openmpi?
 
 Specifically, functions like
 
 MPI_Allreduce
 
 seem to give weird results with __float128.
 
 Essentially all I found was
 
 http://beige.ucs.indiana.edu/I590/node100.html
 
 where they state
 
 MPI_LONG_DOUBLE
  This is a quadruple precision, 128-bit long floating point number.
 
 
 But as far as I have seen, MPI_LONG_DOUBLE is only used for long doubles.
 
 The Open MPI Version is 1.6.3 and gcc is 4.7.3 on a x86_64 machine.
 
>>> It seems unlikely that 10 year old course notes on an unspecified MPI
>>> implementation (hinted to be IBM power3) would deal with specific details of
>>> openmpi on a different architecture.
>>> Where openmpi refers to "portable C types" I would take long double to be
>>> the 80-bit hardware format you would have in a standard build of gcc for
>>> x86_64.  You should be able to gain some insight by examining your openmpi
>>> build logs to see if it builds for both __float80 and __float128 (or
>>> neither).  gfortran has a 128-bit data type (software floating point
>>> real(16), corresponding to __float128); you should be able to see in the
>>> build logs whether that data type was used.
>>> 
>>> 
>>> ___
>>> users mailing list
>>> us...@open-mpi.org
>>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>> 
>> 
>> 
>> -- 
>> Jeff Hammond
>> jeff.scie...@gmail.com
>> ___
>> users mailing list
>> us...@open-mpi.org
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
> 
> ___
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users

[OMPI users] Planned support for Intel Phis

[Michael Thomadakis]

Hello OpenMPI,

I was wondering what is the support that is being implemented for the Intel
Phi platforms. That is would we be able to run MPI code in "symmetric"
fashion, where some ranks run on the cores of the multicore hostst and some
on the cores of the Phis in a multinode cluster environment.

Also is it based on OFED 1.5.4.1 or on which OFED?

Best regards
Michael

Re: [OMPI users] Planned support for Intel Phis

[Ralph Castain]

Support for the Phi is in the upcoming 1.7.4 release. It doesn't require any 
version of OFED as it uses the Phi's scif interface for communication to ranks 
on the local host. For communication off-host, OMPI will use whatever NICs are 
available


On Feb 2, 2014, at 6:44 AM, Michael Thomadakis  wrote:

> Hello OpenMPI,
> 
> I was wondering what is the support that is being implemented for the Intel 
> Phi platforms. That is would we be able to run MPI code in "symmetric" 
> fashion, where some ranks run on the cores of the multicore hostst and some 
> on the cores of the Phis in a multinode cluster environment.
> 
> Also is it based on OFED 1.5.4.1 or on which OFED?  
> 
> Best regards
> Michael
> 
> ___
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users

Re: [OMPI users] [EXTERNAL] Re: Planned support for Intel Phis

[Hammond, Simon David (-EXP)]

Will this support native execution? I.e. MIC only, no host involvement?

S



--
Si Hammond
Sandia National Laboratories
Remote Connection


-Original Message-
From: Ralph Castain [r...@open-mpi.org]
Sent: Sunday, February 02, 2014 09:02 AM Mountain Standard Time
To: Open MPI Users
Subject: [EXTERNAL] Re: [OMPI users] Planned support for Intel Phis


Support for the Phi is in the upcoming 1.7.4 release. It doesn't require any 
version of OFED as it uses the Phi's scif interface for communication to ranks 
on the local host. For communication off-host, OMPI will use whatever NICs are 
available


On Feb 2, 2014, at 6:44 AM, Michael Thomadakis  wrote:

> Hello OpenMPI,
>
> I was wondering what is the support that is being implemented for the Intel 
> Phi platforms. That is would we be able to run MPI code in "symmetric" 
> fashion, where some ranks run on the cores of the multicore hostst and some 
> on the cores of the Phis in a multinode cluster environment.
>
> Also is it based on OFED 1.5.4.1 or on which OFED?
>
> Best regards
> Michael
>
> ___
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users

___
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

Re: [OMPI users] [EXTERNAL] Re: Planned support for Intel Phis

[Ralph Castain]

It should - our intent is to support the Phi in both modes, so we would 
appreciate feedback. I'm going to post on the website FAQ some notes assembled 
by Tim Carlson and Michael Thomadakis about how to do it, and Nathan Hjelmn did 
the scif implementation, so hopefully others will chime in with suggestions as 
people explore this new capability.


On Feb 2, 2014, at 8:31 AM, Hammond, Simon David (-EXP)  
wrote:

> Will this support native execution? I.e. MIC only, no host involvement?
> 
> S
> 
> 
> 
> --
> Si Hammond
> Sandia National Laboratories
> Remote Connection
> 
> 
> -Original Message-
> From: Ralph Castain [r...@open-mpi.org]
> Sent: Sunday, February 02, 2014 09:02 AM Mountain Standard Time
> To: Open MPI Users
> Subject: [EXTERNAL] Re: [OMPI users] Planned support for Intel Phis
> 
> Support for the Phi is in the upcoming 1.7.4 release. It doesn't require any 
> version of OFED as it uses the Phi's scif interface for communication to 
> ranks on the local host. For communication off-host, OMPI will use whatever 
> NICs are available
> 
> 
> On Feb 2, 2014, at 6:44 AM, Michael Thomadakis  
> wrote:
> 
> > Hello OpenMPI,
> >
> > I was wondering what is the support that is being implemented for the Intel 
> > Phi platforms. That is would we be able to run MPI code in "symmetric" 
> > fashion, where some ranks run on the cores of the multicore hostst and some 
> > on the cores of the Phis in a multinode cluster environment.
> >
> > Also is it based on OFED 1.5.4.1 or on which OFED? 
> >
> > Best regards
> > Michael
> >
> > ___
> > users mailing list
> > us...@open-mpi.org
> > http://www.open-mpi.org/mailman/listinfo.cgi/users
> 
> ___
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users
> 
> ___
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users