Full HTML for
Given by Bryan Carpenter, Geoffrey Fox, Xinying Li, Yuhong wen, Guansong Zhang at Alliance 98 NCSA Illinois on April 27-29 98. Foils prepared 3 May 98
Outside Index
Summary of Material
| Current Status of HPJava and motivation in context of HPspmd |
| HPF like but SPMD not High Level Data parallel |
Denote Foils where Image Critical
Denote Foils where HTML is sufficient
HPJava: Structured SPMD programming in Java
Motivations
HPJava
Libraries
Distributed Array Syntax
Distributed Control
Example: Red-black iteration
Preliminary Performance
Other HPspmd Languages
Outside Index
Summary of Material
HTML version of Basic Foils prepared 3 May 98 
Full HTML Index
| Java binding of the High Level HPspmd model |
| Bryan Carpenter, Geoffrey Fox, Xinying Li, Yuhong Wen, Guansong Zhang |
HTML version of Basic Foils prepared 3 May 98
Full HTML Index
| Data parallelism vital in High Performance scientific computing. |
| High level HPF programming model attractive, but implementations problematic. |
| HPspmd model: Distributed array syntax, plus high-level class library bindings for communication and arithmetic on arrays. |
| Explicitly MIMD control flow. |
HTML version of Basic Foils prepared 3 May 98
Full HTML Index
| An instance of the HPspmd language model. |
| Adds limited new syntax to Java for describing and accessing distributed arrays. |
| Distributed array model is a superset of HPF 1.0 model. |
| 3 distributed control constructs replace ON HOME and FORALL constructs of HPF. |
HTML version of Basic Foils prepared 3 May 98
Full HTML Index
| In HPspmd model, all communications go through explicit calls to user-level libraries, initially: |
| Adlib: regular collective operations |
| MPI: low-level message passing |
| Later, add interfaces to other libraries, eg |
| Global Arrays: 1-sided access to remote data |
| CHAOS: irregular collective operations |
HTML version of Basic Foils prepared 3 May 98 
Full HTML Index
| HPJava |
| Procs p = new Procs1(4) ; |
| Range x = new BlockRange(100,p.dim(0)); |
| float [[,*]] a = new float [[x, 100]] on p ; |
| float [[]] b = new float [[x]] on p ; |
| HPF |
| !HPF$ PROCESSOR P(4) |
| !HPF$ DISTRIBUTE T(BLOCK) ONTO P |
| REAL A(100,100) |
| !HPF$ ALIGN A(:,*) WITH T(:) |
| REAL B(100) |
| !HPF$ ALIGN B(:) WITH T(:) |
HTML version of Basic Foils prepared 3 May 98
Full HTML Index
| Explicitly coordinate groups of processes. |
| Groups cooperate in loosely synchronous manner. |
| Control constructs on, at and overall dispatch control. |
| on(p) { |
| overall(i = x|:) |
| b [i] = 0 ; |
| at(j = x [3]) |
| b [j] = 23 ; |
| overall(i = x|2:99) { |
| for(int k = 0 ; k < 50) |
| a [i, k] = b [i] + k ; |
| } |
| } |
HTML version of Basic Foils prepared 3 May 98
Full HTML Index
| Procs p = new Procs2(NP, NP) ; |
| on(p) { |
| Range x = new BlockRange(N, p.dim(0), 1) ; // ghost width 1 |
| Range y = new BlockRange(N, p.dim(1), 1) ; // ... |
| float [[,]] u = new float [[x, y]] ; |
| for(int parity = 0 ; parity < 2 ; parity++) { |
| Adlib.writeHalo(u, widths) ; |
| overall(i = x | 1 : N - 2) |
| overall(j = y | 1 + (x.idx(i) + parity) % 2 : N - 2 : 2) |
| u [i, j] = 0.25 * (u [i - 1, j] + u [i + 1, j] + u [i, j - 1] + u [i, j + 1]) ; |
| } |
| } |
HTML version of Basic Foils prepared 3 May 98
Full HTML Index
| Hand translated Jacobi iteration (HPJava to Java). |
| Compared with sequential Java, C++ and Fortran (-O optimization level). |
| JDK 1.2Beta, JNI to Adlib, MPICH, Ultrasparc cluster |
HTML version of Basic Foils prepared 3 May 98
Full HTML Index
| We have outlined a dialect of Fortran following the same HPspmd programming model |
| C++ version of the model can be implemented through class libraries and macros alone (but inefficient). |
| Collaborators at University of Peking to implement compilers for these languages. |