Replied: Wed, 25 Jun 1997 23:27:05 -0400
Replied: Tomasz Haupt <haupt@npac.syr.edu>
Received: from aga.npac.syr.edu (aga.npac.syr.edu [128.230.117.22]) by postoffice.npac.syr.edu (8.7.5/8.7.1) with SMTP id LAA25723; Mon, 23 Jun 1997 11:52:13 -0400 (EDT)
Sender: haupt
Message-ID: <33AE9BA2.1CFB@npac.syr.edu>
Date: Mon, 23 Jun 1997 11:52:02 -0400
From: Tomasz Haupt <haupt@npac.syr.edu>
X-Mailer: Mozilla 3.01SC-SGI (X11; I; IRIX 6.3 IP32)
MIME-Version: 1.0
To: gcf
Subject: ACTS: my response to Jarek
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=us-ascii
Content-Length: 3697

This is what I intend to send
I replaced sect 3 of Jarek's draft. I would appreciate your comments.


Scope of proposed work:

We propose to extend and integrate Global Arrays NUMA programming
toolkit with SCiTL and other important facilities including HPF. 


1. GA component

   - interoperability with SciTL and other important to DOE tools
      MPI, Scalapack, Plapack, PETSc, P++ 
      (most already implemented)

   - add C++ interface (object oriented design already in place)

   - emphasize NUMA model, SMP optimizations (important for ASCI)


2. Functionality extensions (needed for CFD and relativistic 
   quantum chemistry)
   
   - higher-dimensional (>2) arrays

   - support for cyclic distribution

   - support for sparsity


3. Interoperability with HPF and language support

   - Integration of interpreted and compiled HPF using Web Technologies

     . compiled mode HPF : high performance 
     . dynamic on-and-off switching for the interpreter mode
        - steering
        - interactive visualizations, and data analysis
        - debugging
        - fast prototyping 
        - dynamical software integration (interpreted interfaces to
          non-HPF runtime libraries)

   - System components:

      . compiler front-end for HPF (done), Java (done), C++ (under
development,
        funded independently) - to be adopted from PCRC
         - used for automatic instrumentation of the source code
(working
           prototype at this time) 
         - used to process user's extensions to the language

     . common runtime support
         - support for HPF compiler requests (done; PCRC)
         - support for HPF-2.0 extensions  (under development; PCRC)
         - extensions beyond HPF standard to support other programming
           paradigms (e.g. GA, DAGH, and other SciTL components)- this
proposal

     . HPF extrinsic interfaces to SciTL components (we will start with
GA) 
         - this proposal

     . HPF server that accepts and executes user commands (sent by a
client 
       implemented as an Java applet), such as:
         - suspend/resume execution of the HPF code
         - send data (visualizations/dubugging)
         - interprete source code (in HPF, Fortran 77, Java, C++ or
user's
           extensions to these languages)
         - execute precompiled (e.g., library) code on the current data 
       (HPF server: working, limited prototype available at this time)  
     
     . GUI implemeted in Java (under development now)


references:

1. <a href=http://www.npac.syr.edu/projects/pcrc/demo1997/> PCRC
front-end, and runtime</a>
2a. <a href=http://www.npac.syr.edu/users/haupt/HPFI/> Integrated HPF
compiler and Interpreter </a>
2b. <a
href=http://www.npac.syr.edu/users/akarsu/SC97/ps/correct_draftvis.html>
Integration of Interpreted and Compiled HPF using Web Technologies</a>
3. <a
href=http://www.npac.syr.edu/users/akarsu/HPFInterpreter/HPFInterpreter.html>HPF
interpreter</a>
4. <a
href=http://www.emsl.pnl.gov:2080/docs/parsoft/mattworks/compiler.html>
Language extensions</a>


====================================================================

original Jarek's text:

3. Interoperability with HPF and language support

   - language support needed (preferably F90 compilation target)

   - propose interface with HPF through EXTRINSIC interface and
     HPF_LOCAL_LIBRARY:

      .  HPF handles data-parallel and GA task-parallel code sections
      .  provides get/put/reduction/scatter/gather on HPF arrays
         through library interface (GAlite)

   - next step is to use language constructs (F90 array notation) 
     and employ preprocessor technology to translate them into 
     the library calls