The origin:

    http://www.eecg.toronto.edu/parallel/publications.html

    (archival copy: https://archive.is/rZUhA )

    ftp://ftp.cs.toronto.edu/pub/parallel

Many, may be NOT all, of the HTML-files that the wget downloaded, 
were modified so that they reference the files from the 

    ./manually_copied_ftp_colon_doubleslash_ftp_cs_toronto_eduwww.eecg.toronto.edu

in stead of containing URLs to the original FTP-site.

---------------------------------------------------------------------
File: Ravi_Stumm_ICPP95.ps.Z

Title: Hierarchical Ring Topologies and the effect of their Bisection
       Bandwidth Constraints

Authors: G. Ravindran and M. Stumm

Where  : Proc. Intl. Conf. on Parallel Processing, pp.I/51-55, 1995

Keywords: Multiprocessor architectures, Interconnection networks,
          Hierarchical rings, Bisection bandwidth

Abstract:

Ring-based hierarchical networks are interesting alternatives to
popular direct networks such as 2D meshes or tori. They allow for
simple router designs, wider communications paths, and faster networks
than their direct network counterparts. However, they have a constant
bisection bandwidth, regardless of system size. In this paper, we
present the results of a simulation study to determine how large
hierarchical ring networks can become before their performance
deteriorates due to their bisection bandwidth constraint. We show that
a system with a maximum of 128 processors can sustain most memory
access behaviors, but that larger systems can be sustained, only if
their bisection bandwidth is increased.

---------------------------------------------------------------------
File: Ravi_Stumm_JIEICE96.ps.Z

Title : A Comparison of Blocking and Non-blocking Packet Switching
        Techniques in Hierarchical Ring Networks

Authors: G. Ravindran and M. Stumm

Where : IEICE Trans. Inf. & Syst., vol. E79-D, No. 8, August 1996

keywords:

Networks, Switching, Wormhole, Virtual Cut-through, Hierarchical Ring Networks,
Slotted Rings

Abstract :

This paper presents the results of a simulation study of blocking and
non-blocking switching for hierarchical ring networks.
The switching techniques include wormhole, virtual cut-through, and
slotted ring.
We conclude that slotted ring network performs better than
the more popular wormhole and virtual cut-through networks.
We also show that the size of the node buffers is an important parameter and
that choosing them too large can hurt performance in some cases.
Slotted rings have the advantage that the choice of buffer size is easier in
that larger than necessary buffers do not hurt performance and hence a single
choice of buffer size performs well for all system configurations.
In contrast, the optimal buffer size for virtual cut-through and wormhole
switching nodes varies depending on the system configuration and the level
in the hierarchy in which the switching node lies.

---------------------------------------------------------------------
File: Zhou_Brecht_SM91.ps.Z

Title: Processor Pool-Based Scheduling for Large-Scale NUMA Multiprocessors

Where: Appears in: Proceedings of  the  1991  ACM  SIGMETRICS  Conference  on
Measurement and Modeling of Computer Systems, May (1991), pp. 133-142.

Authors: Songnian Zhou and Timothy Brecht

Keywords: NUMA, Schedulling, multiprocessor performance

Abstract:
Large-scale  Non-Uniform  Memory  Access  (NUMA)  multiprocessors  are
gaining  increased attention due to their potential for achieving high
performance through the replication of relatively  simple  components.
Because  of  the complexity of such systems, scheduling algorithms for
parallel  applications  are  crucial  in  realizing  the   performance
potential  of  these  systems.  In particular, scheduling methods must
consider the scale of the system, with  the  increased  likelihood  of
creating  bottlenecks,  along  with  the  NUMA  characteristics of the
system, and the benefits to be gained  by  placing  threads  close  to
their code and data.

We propose a class of scheduling algorithms based on processor  pools.
A processor pool is a software construct for organizing and managing a
large number of processors by dividing them into groups called  pools.
The  parallel  threads  of  a  job are run in a single processor pool,
unless there are performance advantages for a  job  to  span  multiple
pools.   Several  jobs may share one pool.  Our simulation experiments
show that processor pool-based scheduling may effectively  reduce  the
average  job  response time.  The performance improvements attained by
using processor pools increase with the  average  parallelism  of  the
jobs, the load level of the system, the differentials in memory access
costs, and the likelihood of having system bottlenecks.  As the system
size   increases,  while  maintaining  the  workload  composition  and
intensity, we observed that processor pools can  be  used  to  provide
significant  performance  improvements.   We  therefore  conclude that
processor pool-based scheduling may  be  an  effective  and  efficient
technique for scalable systems.


---------------------------------------------------------------------
File: Brecht_SEDMS93.ps.Z

Title: On the Importance of Parallel Application Placement
                       in NUMA Multiprocessors

Authors: Timothy Brecht

Where: Proceedings of the Fourth Symposium on Experiences  with
Distributed  and  Multiprocessor  Systems  (SEDMS  IV), San Diego, CA,
September, 1993.

Keywords: NUMA, multiprocessor scheduling, multiprocessor performance  

Abstract:
The thesis of this paper is that scheduling decisions in  large-scale,
shared-memory,  NUMA  (Non-Uniform Memory Access) multiprocessors must
consider not only how many processors, but also  which  processors  to
allocate  to  each  application.   We  call  the  problem of assigning
parallel  processes  of  an  application  to  processors   application
placement.

We explore the importance of  placement  decisions  by  measuring  the
execution  time  of  several  parallel  applications  using  different
placements on a shared-memory NUMA  multiprocessor.   The  results  of
these  experiments  lead  us  to conclude that, as expected, in small-
scale mildly NUMA multiprocessors, placement  decisions  have  only  a
minor affect on the execution time of parallel applications.  However,
the  results  also  show  that  placement  decisions  in   large-scale
multiprocessors  are  critical  and  localization  that  considers the
architectural clusters inherent in these systems  is  essential.   Our
experiments  also  show  that  the  importance  of placement decisions
increases substantially with the size and NUMAness of the  system  and
that  the  placement  of individual processes of an application within
the  subset  of   chosen   processors   also   significantly   impacts
performance.

---------------------------------------------------------------------

File: Kumar_Kulkarni_ICPP91.ps.Z
      (does not contain figures)
Title: Generalized Unimodular Loop Transformations for 
       Distributed Memory Multiprocessors

Authors: K G Kumar*, D Kulkarni+ and A Basu
        Center for Development of Advanced Computing
        2/1 Brunton Road, Bangalore 560 025, India
* Now at IBM TJ Watson, York Town Heights, NY 10598
+ Now at Dept of Computer Science, University of Toronto,
         Toronto, ON M5S 1A4

Where: International Conference of Parallel Processing -91

Keywords: Parallelizing Compilers, Restructuring Transformations, 
          Loop Partitioning, Iteration Spaces, Dependence Vectors.

                           Abstract
In this paper, we present a generalized unimodular loop
transformation as a simple, systematic and elegant method
for partitioning the iteration spaces of nested loops for
execution on distributed memory multiprocessors. We present
a methodology for deriving the transformations that internalize
multiple dependences in a multidimensional iteration space
without resulting in a deadlocking situation. We then
derive the general expression for the bounds of the
transformed loops in terms of the bounds of the original
space and the transformation matrix elements.

-------------------------------------------------------------------

File: Kumar_Kulkarni_ICS92.ps.Z

Title: Deriving Good Transformations for Mapping
       Nested Loops on Hierarchical Parallel Machines
       in Polynomial Time
Authors: K G Kumar*, D Kulkarni+ and A Basu
        Center for Development of Advanced Computing
        2/1 Brunton Road, Bangalore 560 025, India
* IBM TJ Watson, York Town Heights, NY 10598
+ Dept of Computer Science, University of Toronto,
         Toronto, ON M5S 1A4

Where: International Conference on Supercomputing 92

Keywords: Parallelizing Compilers, Restructuring Transformations,
          Loop Partitioning, Iteration Spaces, Dependence Vectors.

We present a computationally efficient method for deriving
the most appropriate transformation and mapping of a nested
loop for a given hierarchical parallel machine. This method
is in the context of our systematic and general theory of
unimodular loop transformations for the problem of iteration
space partitioning \cite{kandk6}. Finding an optimal mapping or
an optimal associated unimodular transformation is NP-complete.
We present a polynomial time method for obtaining a `good'
transformation using a simple parameterized model of the
hierarchical machine. We outline a systematic methodology
for obtaining the most appropriate mapping.

------------------------------------------------------------------- 

File: Li_Tandri_et_ICPP93.ps.Z

Title: LOCALITY AND LOOP SCHEDULING ON NUMA MULTIPROCESSORS

Authors: Hui Li, Sudarsan Tandri Michael Stumm, and Kenneth C. Sevcik

Where: International Conference on Parallel Processing 93

Keywords: NUMA multiprocessors, Locality, Scheduling 

Abstract:
An important issue in the parallel execution of loops is how to
partition and schedule the loops onto the available processors.
While most existing dynamic scheduling algorithms manage load
imbalances well, they fail to take locality into account and
therefore perform poorly on parallel systems with non-uniform
memory access times.  In this paper, we propose a new loop
scheduling algorithm, Locality-based Dynamic Scheduling (LDS),
that exploits locality, and dynamically balances the load.

--------------------------------------------------------------
File: Sandhu_et_al_PPOPP.ps.Z

Title: The shared  regions  approach to software cache coherence on 
       multiprocessors 

Where: Appears in: Proceedings of  the  1993  ACM  SIGPLAN Symposium on
Principles and Pranctice of Parallel Programming, May (1993).

Authors: Harjinder Sandhu, Benjamin Gamsa and Songnian Zhou 

Keywords: NUMA, cache coherence, multiprocessor performance

Abstract: The effective management of caches is critical to the performance of
applications on shared-memory multiprocessors. In this paper,
we discuss a technique for software cache coherence that is
based upon the integration of a program-level abstraction for shared data
with software cache management. The program-level abstraction,
called {\it Shared Regions}, explicitly relates synchronization objects
with the data they protect. Cache coherence algorithms are presented
which use the information provided by shared region primitives, and
ensure that shared regions are always cacheable by the processors accessing
them. Measurements and experiments of the Shared Region approach on a
shared-memory multiprocessor are shown. Comparisons with other software
based coherence strategies, including a user-controlled strategy and an
operating system-based strategy, show that this approach is able to
deliver better performance, with relatively low corresponding overhead
and only a small increase in the programming effort.  Compared to
a compiler-based coherence strategy, the Shared Regions
approach still performs better  than a compiler that can achieve 90\%
accuracy in allowing cacheing, as long as the regions are a few 
hundred bytes or larger, or they are re-used a few times in the cache.

-------------------------------------------------------------------
File: Wilton_Vranesic_SPDP.ps.Z
Title: 
Architectural Support for Block Transfers in a Shared-Memory Multiprocessor

Authors: Steven J.E. Wilton and Zvonko G. Vranesic

To appear in the Fifth IEEE Symposium on Parallel and Distributed Processing,
Irving, Texas, December 1993

Keywords:

Shared-memory multiprocessor, block transfer support

Abstract:

This paper examines how the performance of a shared-memory multiprocessor
can be improved by including hardware support for block transfers.  A system
similar to the Hector multiprocessor developed at the University of Toronto
is used as a base architecture.  It is shown that such hardware support can
improve the performance of initialization code by as much as 50%, but that
the amount of improvement depends on the memory access behavior of the
program and the way in which the operating system issues block transfer
requests.

----------------------------------------------------------------------

File: Sevcik_Zhou_PERF93.ps.Z

Title: Performance Benefits and Limitations of Large NUMA Multiprocessors

Authors: Kenneth C. Sevcik and Songnian Zhou

Where:  appeared in the Proceedings of Performance '93 , Rome, Italy,
  September 27 to October 1, 1993, pp. 183-204, Elsevier Science Publ.

Abstract: Please see the ps file.

----------------------------------------------------------------------
File: Harz_Sevcik_SC93.ps.Z

Title: Hot Spot Analysis in Large Scale Shared Memory Multiprocessors

Authors:  Karim Harzallah and Kenneth C. Sevcik

Where: will appear in the Proceedings of the Supercomputing '93
Conference, November, 1993, Portland, Oregon.

Abstract: Please see the ps file.

-----------------------------------------------------------------------
File: Sevcik_JPE.ps.Z

Title: Application Scheduling and Processor Allocation in 
Multiprogrammed Parallel Processing Systems

Authors:  Kenneth C. Sevcik

Where:  This paper will appear in a special issue of the journal
  "Performance Evaluation" on the performance evaluation of parallel
  systems in late 1993 or early 1994.

Abstract: Please see the ps file.

-----------------------------------------------------------------------
File :   Holliday_Stumm_IEEETC.ps.Z

Title:   Performance Evaluation of Hierarchical Ring-Based Shared 
         Memory Multiprocessors

Authors: Mark Holliday 
         Dept. of Computer Science, Duke University, Durham, NC 27706

         Michael Stumm
         Dept. of Electrical and Computer Engineering
         University of Toronto, Toronto, Canada M5S 1A4 

Date:    November 1992; revised April 1993

Published: Technical Report CS-1992-18, Duke University
         Accepted for publication in IEEE Transactions on Computers

Keywords: communication locality; hierarchical ring-based networks;
         hot spots; large scale parallel systems; memory banks;
         performance evaluation; prefetching; shared memory multiprocessors;
         simulation.

Abstract:

	This paper investigates the performance of word-packet, slotted
	unidirectional ring-based hierarchical direct networks in the
	context of large-scale shared memory multiprocessors.
	Slotted unidirectional rings are attractive because their
	electrical characteristics and simple interfaces allow for fast
	cycle times and large bandwidths.
	For large-scale systems, it is necessary to use multiple rings
	for increased aggregate bandwidth.
	Hierarchies are attractive because the topology ensures unique
	paths between nodes, simple node interfaces and simple inter-ring
	connections.

	To ensure that a realistic region of the design space is examined, 
	the architecture of the network used in the Hector prototype is 
	adopted as the initial design point.
	A simulator of that architecture has been developed and validated with
	measurements from the prototype.
	The system and workload parameterization reflects conditions
	expected in the near future.

	The results of our study show the importance of system balance on 
	performance. Large-scale systems inherently have large
	communication delays for distant accesses, so processor efficiency
	will be low, unless the processors can operate with multiple 
	outstanding transactions using techniques such as prefetching, 
	asynchronous writes and multiple hardware contexts.
	However with multiple outstanding transactions and only one memory
	bank per processing module, memory quickly saturates.
	Memory saturation can be alleviated by having multiple memory
	banks per processing module, but this shifts the bottleneck to the
	ring subsystem.
	While the topology of the ring hierarchy affects performance,
	the ring subsystem will inherently limit the throughput of the system.
	Hence increasing the number of
	outstanding transactions per processor beyond a certain point only
	has a limiting effect on performance, since it causes some of the
	rings to become congested.
	An adaptive maximum number of outstanding
	transactions appears necessary to adjust for the appropriate tradeoff
	between concurrency and contention as the communication locality 
	changes. We show the relationships between processor, ring and memory 
	speeds, and their effects on performance.
	Using block transfers for prefetching seems unlikely to be advantageous
	in that the improvement in the cache hit ratio needed to compensate for
	the increased network utilization is substantial.


-------------------------------------------------------------------------
File : Curran_Stumm_CS.ps.Z

Title:     A Comparison of basic CPU Scheduling Algorithms for 
           Multiprocessor Unix

Authors:   Stephen Curran  and  Michael Stumm
           Department of Electrical and Computer Engineering
           University of Toronto, Toronto, Canada M5S 1A4

Published: Computer Systems, 3(4), Oct., 1990, pp. 551--579.

Abstract:
	In this paper, we present the results of a simulation study
	comparing three basic algorithms that schedule independent tasks in 
	multiprocessor versions of Unix.
	Two of these algorithms, namely Central Queue and Initial Placement,
	are obvious extensions to the standard uniprocessor scheduling 
	algorithm and are in use in a number of multiprocessor systems.
	A third algorithm, Take, is a variation on Initial Placement,
	where processors are allowed to raid the task queues of the other
	processors.
	Our simulation results show the difference between the
	performance of the three algorithms to be small when
	scheduling a typical Unix workload running on a small, bus-based,
	shared memory multiprocessor.
	They also show that the Take algorithm performs best for those
	multiprocessors on which tasks incur overhead each time they migrate.
	In particular, the Take algorithm appears to be more stable
	than the other two algorithms under extreme conditions.


-----------------------------------------------------------------------

File: Stumm_Unrau_Krieger_USENIX92.ps.Z

Title:  HIERARCHICAL CLUSTERING: A STRUCTURE FOR SCALABLE MULTIPROCESSOR
          OPERATING SYSTEM DESIGN

Authors: Michael Stumm, Ron Unrau, and Orran Krieger 

Where:  Extended version of Clustering Micro-Kernels for Scalability,
        Proc.\ of the Usenix Workshop on Micro-Kernels and
        Other Kernel Architectures, April, 1992.

Abstract: Please see the ps file.

----------------------------------------------------------------------

File: Stumm_Vranesic_White_IPPS93.ps.Z

Title: EXPERIENCE WITH THE HECTOR MULTIPROCESSOR 

Authors: Michael Stumm, Zvonko Vranesic, Ron White

Where:  Extended version of paper with same title
        in Proc.\ Intl.\ Parallel Processing Symposium
        Parallel Systems Fair, 1993, pp.\ 9--16.

Abstract: Please see the ps file.

----------------------------------------------------------------------

File: Krieger_Stumm_Unrau_USENIX92.ps.Z

Title: The Alloc Stream Facility: A redesign of application-level
Stream I/O 

Authors: O. Krieger, M. Stumm, and R. Unrau

Where: Extended version of
``Exploiting the advantages of mapped files for stream I/O''
in Proc.\ of the Winter 1992 Usenix Conference, January, 1992.

Abstract:
This paper describes the design and implementation of a new
application level I/O facility, called the Alloc Stream Facility.
The Alloc Stream Facility has several key advantages.  First,
performance is substantially improved as a result of a)~the
structure of the facility that allows it to take advantage of
system specific features like mapped files, and b)~a reduction in
data copying and the number of I/O system calls.  Second, the
facility is designed for multi-threaded applications running on
multiprocessors and allows for a high degree of concurrency.
Finally, the facility can support a variety of I/O interfaces,
including stdio, emulated Unix I/O, ASI, and C++ streams, in a
way that allows applications to freely intermix calls to the
different interfaces, resulting in improved code reusability.

We show that on several Unix workstation platforms the
performance of Unix applications using the Alloc Stream Facility
can be substantially better that when the applications use the
original I/O facilities.


----------------------------------------------------------------------

File: Krieger_Stumm_DAGS93.ps.Z

Title: HFS: A Flexible File System for large-scale Multiprocessors 

Authors: Orran Krieger and Michael Stumm

Where: Proceedings of the 1993 DAGS/PC Symposium

Abstract:
The {H{\sc urricane}} File System (HFS) is a new file system
  being developed for large-scale shared memory multiprocessors with
  distributed disks. The main goal of this file system is scalability; that is,
  the file system is designed to handle demands that are expected to grow
  linearly with the number of processors in the system. To achieve this goal,
  HFS is designed using a new structuring technique called Hierarchical
  Clustering. HFS is also designed to be flexible in supporting a variety of
  policies for managing file data and for managing file system state. This
  flexibility is necessary to support in a scalable fashion the diverse
  workloads we expect for a multiprocessor file system.


----------------------------------------------------------------------

File: Krieger_etal_ICPP93.ps.Z

Title: A fair fast scalable reader-writer lock 

Authors: O. Krieger, M. Stumm, R. Unrau, and J. Hanna,

Where: Proc. Intl. Conf. on Parallel Processing, 1993.

Abstract:
A reader-writer lock allows either multiple readers to inspect
  shared data or a single writer exclusive access to that data.  On
  shared memory multiprocessors, the cost of acquiring and
  releasing these locks can have a large impact on the performance
  of parallel applications.  Other researchers have shown how to
  implement scalable locks, that is, locks that can become
  contended without resulting in memory or interconnection network
  contention.  This paper describes a new algorithm for a
  reader-writer lock that, while being scalable in the contended
  case, has a low overhead in the uncontended case.  This is
  important because most parallel applications are written so that
  locks are typically uncontended.  The only atomic operation
  required by this algorithm is fetch_and_store and hence
  it can be used on most current multiprocessor systems.
  Experimental results are provided.

----------------------------------------------------------------------

File: Kulkarni_Stumm_Tutorial.ps.Z

Title: Loop and Data Transformations: A tutorial

Authors: Dattatraya Kulkarni and Michael Stumm

Where: Internal document, a tutorial guide.

Abstract:
Hierarchically structured machines appear to be becoming the
dominant parallel computing structure. These systems have
non-uniform access times. We address the problem of restructuring
a possibly sequential program to execute efficiently on such
parallel machines. This restructuring involves transforming and
partitioning the loop structures and the data to so as to improve
{\it parallelism}, {\it static} and {\it dynamic locality}, and {\it
load balance}.
The objective of this paper is to present previous and ongoing work
on loop and data transformations and motivate a {\it unified} framework to
restructuring of a sequence of loops and data so as to execute efficiently on
parallel machines with several levels of hierarchy.

----------------------------------------------------------------------


File: Baru_Zilio_PADS93.ps.Z

Title: Data reorganization in parallel database systems

Author: Chaitanya Baru & Daniel C. Zilio

Where : Proc. of the IEEE Workshop on Advances in Parallel and
Distributed Systems}, Princeton, NJ, pp.102-107, Oct.  1993.

Abstract: Parallel database systems are suitable for use in
applications with high capacity and high performance and availability
requirements.  The trend in such systems is to provide efficient
on-line capability for performing various system administration
functions such as, index creation and maintenance, backup/restore,
reorganization, and gathering of statistics.  For some of these
functions, the on-line capability can be efficiently supported by the
use of ``incremental algorithms", i.e., algorithms that achieve the
function in several, relatively small (i.e., less time-consuming)
steps, rather than in a single, large step.  Incremental algorithms
ensure that only small parts of the database become inaccessible for
short durations as opposed to non-incremental algorithms which may
lock large portions of the database or the entire database for a
longer duration.  In this paper, we discuss issues in providing
concurrent data reorganization capability using incremental algorithms
in parallel database systems.


----------------------------------------------------------------------

File: Kulkarni_Stumm_292.ps.Z

Title: Computational Alignment:
A new, unified program transformation for local and global optimization

Authors: Dattatraya Kulkarni and Michael Stumm

Where: CSRI Tech report 292, ISSN 0834-1648

Abstract:
{\small {\em Computational Alignment} is a new class of program
transformations suitable for both local and global optimization.
Computational Alignment transforms all of the computations of a
{\em portion} of the loop body in order to align them to other
computations either in the same loop or in another loop. It
extends along a new dimension and is significantly more powerful
than linear transformations because $i)$
it can transform subsets of dependences and references;
$ii)$ it is sensitive to the location of data in that it can move
the computation relative to data; $iii)$ it applies to imperfect
loop nests; and $iv)$ it is the first loop transformation that
can change {\it access vectors}. Linear transformations are just a
special case of Computational Alignment.
Computational Alignment is highly suitable for global
optimization because it can transform given loops to access data
in similar ways. Two important subclasses of Computational
Alignment are presented as well, namely, {\em Freeing} and {\em
Isomerizing} Computational Alignment.}

-------------------------------------------------------------
File: Brecht_PhD_303.ps.Z

Title: Multiprogrammed Parallel Application Scheduling in NUMA Multiprocessors

Authors: Timothy B. Brecht

Where: Ph.D. Dissertation - CSRI Technical Report CSRI-303

Abstract:

          The   invention,   acceptance,    and    proliferation    of
     multiprocessors  are  primarily a result of the quest to increase
     computer system performance.   The  most  promising  features  of
     multiprocessors are their potential to solve problems faster than
     previously possible and to solve larger problems than  previously
     possible.    Large-scale  multiprocessors  offer  the  additional
     advantage of being able to execute multiple parallel applications
     simultaneously.

          The execution time of a  parallel  application  is  directly
     related  to  the  number  of  processors  it is allocated and, in
     shared-memory   non-uniform    memory    access    time    (NUMA)
     multiprocessors,  which processors it is allocated.  As a result,
     efficient and effective scheduling becomes  critical  to  overall
     system  performance.   In fact, it is likely to be a contributing
     factor in  ultimately  determining  the  success  or  failure  of
     shared-memory NUMA multiprocessors.

          The subjects  of  this  dissertation  are  the  problems  of
     processor  allocation  and  application placement.  The processor
     allocation problem involves determining the number of  processors
     to  allocate to each of several simultaneously executing parallel
     applications and possibly dynamically adjusting those allocations
     to  improve  overall  system performance.  The performance metric
     used is mean response time.   We  show  that  by  differentiating
     between  applications  based on the amount of remaining work they
     have to execute, performance can be improved significantly.  Then
     we  propose  techniques  for estimating an application's expected
     remaining work along with policies for using these  estimates  to
     make  improved  processor  allocation decisions.  An experimental
     evaluation demonstrates the promise of this approach.

          The placement problem involves determining which of the many
     processors  to  assign  to  each  application.  Using experiments
     conducted on a representative  system,  we  demonstrate  that  in
     large-scale  NUMA  multiprocessors the execution time of parallel
     applications is significantly affected by the  placement  of  the
     application.  This motivates the need for new techniques designed
     explicitly  for  NUMA  multiprocessors.   We  introduce  such   a
     technique,   called  processor  pool-based  scheduling,  that  is
     designed to  localize  the  execution  of  parallel  applications
     within  a  NUMA  architecture  and  to isolate different parallel
     applications from each other.  An experimental evaluation of this
     scheduling  method  shows  that  it  can be used to significantly
     reduce mean response time over methods that do not  consider  the
     placement of parallel applications.

-------------------------------------------------------------------
File: Gamsa_MASc.ps.Z

Title: Region-Oriented Main Memory Management
       in Shared-Memory NUMA Multiprocessors


Authors: Benjamin Gamsa

Where: M.Sc. Thesis

Abstract:

     In Non-Uniform Memory Access  time  (NUMA)  multiprocessors,
     distribution of the memory modules facilitates architectural
     scaling,  but creates complications for the programmers  who
     must  be  concerned  with the physical distribution of their
     data in order to obtain good performance.

     In order to reduce the impact of remote  accesses,  in  this
     thesis  we  propose  that  data  be  partitioned into Shared
     Regions that reflect the  granularity  of  data  sharing  in
     programs, and that special synchronization calls be added to
     enforce proper ordering of accesses to the  shared  data  as
     well  as to manage replication and consistency transparently
     to the programmer.

     Results   from   measurements   on   a   16-processor   NUMA
     multiprocessor  and from a model of the system indicate that
     the Shared Regions approach is successful in  obtaining  the
     necessary  locality critical to performance, while incurring
     only minimal overhead.  Data distribution methods  are  also
     observed  to have a significant impact on the performance of
     the  system,  especially  in  the   larger   multiprocessors
     modeled.

-------------------------------------------------------------------

File: Unrau_PhD.ps.Z

Title: Scalable Memory Management through Hierarchical 
       Symmetric Multiprocessing

Authors: Ronald C. Unrau

Where: Ph.D. Disseration

Abstract:

     This dissertation examines scalability issues in  the  design  of
     operating systems for large-scale, shared-memory multiprocessors.
     In particular, the thesis focuses on structuring issues  as  they
     relate to memory management.

     From a set of simple, well-known  queuing  network  formulas,  we
     derive  a  set  of properties that describe sufficient conditions
     for an operating system to scale. From these properties we  first
     develop  a  set of guidelines for designing scalable systems, and
     then develop  a  new  structuring  philosophy  for  shared-memory
     multiprocessor  operating  systems, called Hierarchical Symmetric
     Multiprocessing (HSM).

     HSM  manages  the  system  resources  in  clusters,  using  tight
     coupling  within  a  cluster, and loose coupling across clusters.
     Distributed systems principles are applied  by  distributing  and
     replicating   system   services  and  data  objects  to  increase
     locality,  increase  concurrency,  and   to   avoid   centralized
     bottlenecks,  thus  making  the  system scalable.  However, tight
     coupling is used within a cluster, so the  system  performs  well
     for  local  interactions.  HSM maximizes locality which is key to
     good performance in large systems, and systems based on  HSM  can
     easily  be  adapted  to  different  hardware  configurations  and
     architectures by changing the size of the clusters.  Finally, HSM
     leads  to a modular system composed from easy-to-design and hence
     efficient building blocks.

     Memory  management  is  a  particularly  challenging  service  to
     implement  within  the HSM framework, because it must provide the
     applications with an integrated and coherent  view  of  a  single
     system,  while  distributing and replicating services in order to
     fully exploit the hardware potential. We describe in  detail  the
     implementation  of an HSM structured memory management subsystem,
     and evaluate the performance of our implementation on  Hector,  a
     prototype scalable shared memory multiprocessor.

-------------------------------------------------------------------

File: Wu_MASc.ps.Z

Title: Processor Scheduling in Multiprogrammed Shared Memory 
       NUMA Multiprocessors

Authors: Chee-Shong Wu

Where: M.Sc. Thesis

Abstract:

     In a multiprogrammed multiprocessor, the scheduler  is  not  only
     responsible for deciding when to activate an application and when
     to suspend it, but is also responsible for determining  how  many
     processors  to  allocate  to each application. In a scalable Non-
     Uniform Memory Access  (NUMA)  multiprocessor,  it  must  further
     resolve  the  problem  of which processor(s) to allocate to which
     application since the memory reference times are not the same for
     all processor-memory pairs.

     In this thesis, we study  the  problem  of  how  to  characterize
     parallel   applications  and  how  to  apply  this  knowledge  in
     scheduling for NUMA systems.  We also study  the  performance  of
     several  scheduling  algorithms  in  an  NUMA  environment. These
     algorithms differ in the frequency of reallocations.  We  propose
     two  policies,  the  Static policy and the Immediate Start Static
     policy, that  utilize  application  characteristics  when  making
     scheduling  decisions.   The performance of these two policies is
     compared  with  that  of  the  Dynamic   policy,   on   an   NUMA
     multiprocessor, Hector.

---------------------------------------------------------------------
File: Parsons_Sevcik_IPPS95.ps.Z

Title: Multiprocessor Scheduling for High-Variability Service Time
       Distributions
 
Where: IPPS '95 Workshop on Job Scheduling Strategies for Parallel Processing
  reprinted in Springer-Verlag Lecture Notes in Computer Science, Vol 949,
  pages 127--145.

Authors: Eric W. Parsons and Kenneth C. Sevcik 

Keywords: Scheduling, multiprocessor performance

Abstract:
  Many disciplines have been proposed for scheduling and processor
  allocation in multiprogrammed multiprocessors for parallel processing.
  These have been, for the most part, designed and evaluated for workloads
  having relatively low variability in service demand.  But with reports
  that variability in service demands at high performance computing centers
  can actually be quite high, these disciplines must be reevaluated.  In
  this paper, we examine the performance of two well-known static
  scheduling disciplines, and propose preemptive versions of these that
  offer much better mean response times when the variability in service
  demand is high.  We argue that, in systems in which dynamic
  repartitioning in applications is expensive or impossible, these
  preemptive disciplines are well suited for handling high variability in
  service demand.

---------------------------------------------------------------------
File:  Okrieg_PhD.ps.Z

Title: HFS: A flexible file system for shared-memory multiprocessors 

Where: PhD Dissertation, Department of Electrical and Computer Engineering,
University of Toronto

Authors: Orran Krieger 

Keywords: File System, I/O, Hurricane, Hector NUMA multiprocessor

Abstract:

  The Hurricane File System (HFS) is designed for large-scale, shared-memory
  multiprocessors.  Its architecture is based on the principle that a
  file system must support a wide variety of file structures, file
  system policies and I/O interfaces to maximize performance
  for a wide variety of applications.  HFS uses a novel, object-oriented
  building-block approach to provide the flexibility needed to support
  this variety of file structures, policies, and I/O interfaces.  File
  structures can be defined in HFS that optimize for sequential or
  random access, read-only, write-only or read/write access, sparse or
  dense data, large or small file sizes, and different degrees of
  application concurrency.  Policies that can be defined on a per-file
  or per-open instance basis include locking policies, prefetching
  policies, compression/decompression policies and file cache management
  policies.  In contrast, most existing file systems have been designed
  to support a single file structure and a small set of policies.

  We have implemented large portions of HFS as part of the Hurricane operating
  system running on the \hec\ shared-memory multiprocessor.  We
  demonstrate that the flexibility of HFS comes with little processing
  or I/O overhead.  Also, we show that HFS is able to deliver the full
  I/O bandwidth of the disks on our system to the applications.

---------------------------------------------------------------------
File: Vranesic_etal_IEEEC.ps.Z 

Title: Hector -- A hierarchically structured shared memory multiprocessor

Where: IEEE Computer, 24(1): 72-80, January, 1991.

Authors: Z. Vranesic, M. Stumm, D. Lewis and R. White 

Keywords: Shared memory multiprocessors, slotted rings, NUMA, Scalability. 

Abstract: Please see the ps file.

---------------------------------------------------------------------
File: Kulkarni_etal_317.ps.Z 

Title: A Generalized Theory of Linear Loop Transformations

Where: CSRI Tech Report 317

Authors: D. Kulkarni, M. Stumm, R. Unrau 

Keywords: Computational Alignment, Computation Decomposition,
Linear loop transformations, SPMD code generation.

Abstract: In this paper we present a new theory of linear loop transformations
called {\em Computation Decomposition and Alignment\/} (CDA). A CDA
transformation has two components: {\em Computation Decomposition\/}
first decomposes the computations in the loop into computations
of finer granularity, from iterations to instances of subexpressions.
{\em Computation Alignment\/} subsequently, linearly transforms each
of these sets of computations, possibly by using a
different transformation for each set.  This framework subsumes all
existing linear transformation frameworks in that it reduces to a
conventional linear loop transformation when the smallest granularity
is an iteration, and it reduces to some of the more recently extended
frameworks when the smallest granularity is a statement instance.
The possibility  of being able to align computations at arbitrary
granularities adds a new dimensions to performance optimization
on high performance computing platforms.

We describe Computation Decomposition and Alignment and provide
examples of CDA transformations. We present some heuristics to
derive appropriate CDA transformations, given a desired optimization
objective. We present the results of experiments run on
the KSR1 multiprocessor and various RS6000 and
Sparc platforms that demonstrate that CDA can result in substantial
performance improvements.

---------------------------------------------------------------------
File: Kulkarni_Stumm_ACJ95.ps.Z 

Title: Linear Loop Transformations in Optimizing Compilers for Parallel Machines

Where: To appear in the Australian Computer Journal

Authors: D. Kulkarni, M. Stumm 

Keywords:  Linear loop transformations

Abstract: We present the linear loop transformation framework which is the formal
basis for state of the art optimization techniques in restructuring
compilers for parallel machines. The framework unifies most existing
transformations and provides a systematic set of code generation
techniques for arbitrary compound loop transformations. The algebraic
representation of the loop structure and its transformation give
way to quantitative techniques for optimizing performance on parallel
machines. We discuss in detail the techniques for generating
the transformed loop and deriving the desired linear transformation.

---------------------------------------------------------------------
File: Manjikian_Abdelrahman_315.ps.Z 

Title: Fusion of Loops for Parallelism and Locality

Where: CSRI Tech Report 315

Authors: N. Manjikian and T. Abdelrahaman

Keywords:  Loop fusion, cache performance, locality, NUMA

Abstract: Loop fusion improves data locality and reduces synchronization in
data-parallel applications. However, loop fusion is not always
legal. Even when legal, fusion may introduce loop-carried dependences
which reduce parallelism. In addition, performance losses result from
cache conflicts in fused loops. We present new, systematic techniques
which: (1) allow fusion of loop nests in the presence of
fusion-preventing dependences, (2) allow parallel execution of fused
loops with minimal synchronization, and (3) eliminate cache conflicts
in fused loops. We evaluate our techniques on a 56-processor KSR2
multiprocessor, and show performance improvements of up to 20% for
representative loop nest sequences. The results also indicate a
performance tradeoff as more processors are used, suggesting a careful
evaluation of the profitability of fusion.

 
<!--------------------------------------------------------------------->
<HR>
<A NAME="Kulkarni_Stumm_LCR95">.</A>
<HR>

<B>Title:</B> <A HREF="ftp://ftp.cs.toronto.edu/pub/parallel/Kulkarni_Stumm_LCR95.ps.Z">CDA Loop Transformations</A> 
 <P>
<B>Authors:</B> Dattatraya Kulkarni and Michael Stumm 
 <P>
<B>Where:</B> Proceedings of the Third workshop on languages, compilers and run-
time systems
  for scalable computers}, Troy, NY, May 1995, Kluwer Academic. 
 <P>
<B>Abstract:</B>
 <P>

In this paper we present a new loop transformation technique
called {\em Computation Decomposition and Alignment\/} (CDA).
{\em Computation Decomposition\/}
first decomposes the iteration space into finer computation spaces.
{\em Computation  Alignment\/} subsequently, linearly transforms each
computation space independently.
CDA is a general framework in that
linear transformations and its recent extensions are just special cases of CDA.
CDA's fine grained loop restructuring can incur considerable
computational effort, but can exploit optimization
opportunities that earlier frameworks cannot. We present four
optimization contexts in which CDA can be useful. Our
initial experiments demonstrate that CDA adds a new dimension to
performance optimization.

<!--------------------------------------------------------------------->
<HR>
<A NAME="Kulkarni_Stumm_EuroPar95">.</A>
<HR>

<B>Title:</B> <A HREF="ftp://ftp.cs.toronto.edu/pub/parallel/Kulkarni_Stumm_EuroPar95.ps.Z">Implementing Flexible Computation Rules with
Subexpression-level Loop Transformations</A> 
 <P>
<B>Authors:</B> Dattatraya Kulkarni, Michael Stumm  and Ronald C. Unrau
 <P>
<B>Where:</B>Proceedings of the Euro-Par95, Stockholm, Aug 28-31, 1995. 
 <P>
<B>Abstract:</B>
 <P>

  Computation Decomposition and Alignment (CDA) is a new loop transformation
  framework that extends the linear loop transformation framework and
the more recently proposed
  Computation Alignment frameworks by linearly transforming
  computations at the granularity of subexpressions.
  It can be applied to achieve a number of optimization objectives,
  including the removal of data alignment constraints, the elimination
  of ownership tests, the reduction of cache conflicts, and
improvements in data access locality.

  In this paper we show how CDA can be used to effectively implement
  flexible computation rules with the objective of minimizing
  communication and, whenever possible, eliminating intrinsics that
  test whether computations need to be executed or not.
  We describe CDA, show how it can be used to implement flexible
  computation rules, and present an algorithm for deriving appropriate
  CDA transformations.


<!--------------------------------------------------------------------->
<HR>
<A NAME="Unrau_etal_EuroPar95">.</A>
<HR>

<B>Title:</B> <A HREF="ftp://ftp.cs.toronto.edu/pub/parallel/Unrau_etal_EuroPar95.ps.Z">On the Scalability of Demand-Driven Parallel Systems </A> 
 <P>
<B>Authors:</B> Ronald C. Unrau and Michael Stumm and Orran Krieger 
 <P>
<B>Where:</B>Proceedings of the Euro-Par95, Stockholm, Aug 28-31, 1995. 
 <P>
<B>Abstract:</B>
 <P>

Demand-driven systems follow the model where customers enter the system,
request some service, and then depart.
Examples are databases, transaction processing systems and operating systems,
which form the system software layer between the applications and the hardware.
Achieving scalability at the system software layer is critical
for the scalability of the system as a whole, and yet this layer
has largely been ignored.

In this paper, we characterize the scalability of the system software layer
of demand-driven parallel systems 
based on fundamental metrics of quantitative system performance analysis.
We develop a set of sufficient conditions so that
if a system satisfies these conditions, then the system is scalable.
We further argue that in practice these conditions are also necessary.
In the remainder of the paper, we use the necessary and sufficient conditions 
to develop a set of practical design guidelines, to study the effect of
application workloads, and to examine the scalability behavior of a system with
only a limited number of processors.

<!--------------------------------------------------------------------->
<HR>
<A NAME="Parsons_etal_IWOOS95">.</A>
<HR>

<B>Title:</B> <A HREF="ftp://ftp.cs.toronto.edu/pub/parallel/Parsons_etal_IWOOS95.ps.Z">(De-)Clustering Objects for Multiprocessor System Software </A> 
 <P>
<B>Authors:</B> Eric Parsons, Ben Gamsa, Orran Krieger, Michael Stumm
<P>
<B>Where:</B> IWOOS95 (Fourth International Workshop on Object Orientation in Operating Systems 95)  
<P>
<B>Abstract:</B>
 <P>

  Designing system software for large-scale shared-memory multiprocessors
  is challenging because of the level of performance demanded by the
  application workload and the distributed nature of the system.  Adopting
  an object-oriented approach for our system, we have developed a framework
  for de-clustering objects, where each object may migrate, replicate, and
  distribute all or part of its data across the system memory using the
  policies that will best meet the locality requirements for that data.  The
  mechanism for object invocation hides the internal structure of an
  object, allowing a request to be made directly to the most suitable part
  of the object on a per-processor basis without any knowledge of how the
  object is de-clustered.  Method invocation is very efficient, both within
  and across address spaces, involving no remote memory accesses in the
  common case.  We describe the design and implementation of this framework
  in Tornado, our multiprocessor operating system.


<!--------------------------------------------------------------------->
<HR>
<A NAME="Ben_etal_OOPSLAW94">.</A>
<HR>

<B>Title:</B> <A HREF="ftp://ftp.cs.toronto.edu/pub/parallel/Ben_etal_OOPSLAW94.ps.Z">The Importance of Performance-Oriented Flexibility in
System Software for Large-Scale Shared-Memory Multiprocessors
 </A> 
 <P>
<B>Authors:</B> 
Orran Krieger, Ben Gamsa, Karen Reid, Paul Lu, Eric Parsons and Michael Stumm
<P>
<B>Where:</B> 
OOPSLA Workshop on Flexible System Software. October 1994.
<P>
<B>Abstract:</B>
 <P>
See paper for abstract.


<!--------------------------------------------------------------------->
<HR>
<A NAME="Orran_etal_SPDPW95">.</A>
<HR>

<B>Title:</B> <A HREF="ftp://ftp.cs.toronto.edu/pub/parallel/Orran_etal_SPDPW95.ps.Z">
 Exploiting Mapped Files for Parallel I/O
</A> 
 <P>
<B>Authors:</B> 
Orran Krieger, Karen Reid and Michael Stumm
<P>
<B>Where:</B> 
SPDP Workshop on Modeling and Specification of I/O (MSIO), October 1995

<P>
<B>Abstract:</B>
 <P>
  Harnessing the full I/O capabilities of a large-scale multiprocessor is
  difficult and requires a great deal of cooperation between the
  application programmer, the compiler and the operating (/file) system.
  Hence, the parallel I/O interface used by the application to communicate
  with the system is crucial in achieving good performance.  We present a
  set of properties we believe that a good I/O interface should have and
  consider current parallel I/O interfaces from the perspective of these
  properties.  We describe the advantages and disadvantages of mapped-file
  I/O and argue that if properly implemented it can be a good basis for a
  parallel I/O interface that can fulfill the suggested properties.  To
  demonstrate that such an implementation is feasible, we describe
  methodology used in our previous work on the Hurricane operating system
  and in our current work on the Tornado operating system to implement
  mapped files.

%!PS-Adobe-2.0
%%Creator: dvips 5.512 Copyright 1986, 1993 Radical Eye Software
%%Title: pdpta.dvi
%%CreationDate: Thu Nov 23 17:27:55 1995
%%Pages: 10
%%PageOrder: Ascend
%%BoundingBox: 0 0 612 792
%%DocumentFonts: Times-Bold Times-Roman Times-Italic Courier
%%EndComments
%DVIPSCommandLine: dvips -o pdpta.ps pdpta.dvi
%DVIPSSource:  TeX output 1995.08.11:1234
%%BeginProcSet: tex.pro
/TeXDict 250 dict def TeXDict begin /N{def}def /B{bind def}N /S{exch}N /X{S N}
B /TR{translate}N /isls false N /vsize 11 72 mul N /@rigin{isls{[0 -1 1 0 0 0]
concat}if 72 Resolution div 72 VResolution div neg scale isls{Resolution hsize
-72 div mul 0 TR}if Resolution VResolution vsize -72 div 1 add mul TR matrix
currentmatrix dup dup 4 get round 4 exch put dup dup 5 get round 5 exch put
setmatrix}N /@landscape{/isls true N}B /@manualfeed{statusdict /manualfeed
true put}B /@copies{/#copies X}B /FMat[1 0 0 -1 0 0]N /FBB[0 0 0 0]N /nn 0 N
/IE 0 N /ctr 0 N /df-tail{/nn 8 dict N nn begin /FontType 3 N /FontMatrix
fntrx N /FontBBox FBB N string /base X array /BitMaps X /BuildChar{
CharBuilder}N /Encoding IE N end dup{/foo setfont}2 array copy cvx N load 0 nn
put /ctr 0 N[}B /df{/sf 1 N /fntrx FMat N df-tail}B /dfs{div /sf X /fntrx[sf 0
0 sf neg 0 0]N df-tail}B /E{pop nn dup definefont setfont}B /ch-width{ch-data
dup length 5 sub get}B /ch-height{ch-data dup length 4 sub get}B /ch-xoff{128
ch-data dup length 3 sub get sub}B /ch-yoff{ch-data dup length 2 sub get 127
sub}B /ch-dx{ch-data dup length 1 sub get}B /ch-image{ch-data dup type
/stringtype ne{ctr get /ctr ctr 1 add N}if}B /id 0 N /rw 0 N /rc 0 N /gp 0 N
/cp 0 N /G 0 N /sf 0 N /CharBuilder{save 3 1 roll S dup /base get 2 index get
S /BitMaps get S get /ch-data X pop /ctr 0 N ch-dx 0 ch-xoff ch-yoff ch-height
sub ch-xoff ch-width add ch-yoff setcachedevice ch-width ch-height true[1 0 0
-1 -.1 ch-xoff sub ch-yoff .1 add]{ch-image}imagemask restore}B /D{/cc X dup
type /stringtype ne{]}if nn /base get cc ctr put nn /BitMaps get S ctr S sf 1
ne{dup dup length 1 sub dup 2 index S get sf div put}if put /ctr ctr 1 add N}
B /I{cc 1 add D}B /bop{userdict /bop-hook known{bop-hook}if /SI save N @rigin
0 0 moveto /V matrix currentmatrix dup 1 get dup mul exch 0 get dup mul add
.99 lt{/QV}{/RV}ifelse load def pop pop}N /eop{SI restore showpage userdict
/eop-hook known{eop-hook}if}N /@start{userdict /start-hook known{start-hook}
if pop /VResolution X /Resolution X 1000 div /DVImag X /IE 256 array N 0 1 255
{IE S 1 string dup 0 3 index put cvn put}for 65781.76 div /vsize X 65781.76
div /hsize X}N /p{show}N /RMat[1 0 0 -1 0 0]N /BDot 260 string N /rulex 0 N
/ruley 0 N /v{/ruley X /rulex X V}B /V{}B /RV statusdict begin /product where{
pop product dup length 7 ge{0 7 getinterval dup(Display)eq exch 0 4
getinterval(NeXT)eq or}{pop false}ifelse}{false}ifelse end{{gsave TR -.1 -.1
TR 1 1 scale rulex ruley false RMat{BDot}imagemask grestore}}{{gsave TR -.1
-.1 TR rulex ruley scale 1 1 false RMat{BDot}imagemask grestore}}ifelse B /QV{
gsave transform round exch round exch itransform moveto rulex 0 rlineto 0
ruley neg rlineto rulex neg 0 rlineto fill grestore}B /a{moveto}B /delta 0 N
/tail{dup /delta X 0 rmoveto}B /M{S p delta add tail}B /b{S p tail}B /c{-4 M}
B /d{-3 M}B /e{-2 M}B /f{-1 M}B /g{0 M}B /h{1 M}B /i{2 M}B /j{3 M}B /k{4 M}B
/w{0 rmoveto}B /l{p -4 w}B /m{p -3 w}B /n{p -2 w}B /o{p -1 w}B /q{p 1 w}B /r{
p 2 w}B /s{p 3 w}B /t{p 4 w}B /x{0 S rmoveto}B /y{3 2 roll p a}B /bos{/SS save
N}B /eos{SS restore}B end
%%EndProcSet
%%BeginProcSet: texps.pro
TeXDict begin /rf{findfont dup length 1 add dict begin{1 index /FID ne 2 index
/UniqueID ne and{def}{pop pop}ifelse}forall[1 index 0 6 -1 roll exec 0 exch 5
-1 roll VResolution Resolution div mul neg 0 0]/Metrics exch def dict begin
Encoding{exch dup type /integertype ne{pop pop 1 sub dup 0 le{pop}{[}ifelse}{
FontMatrix 0 get div Metrics 0 get div def}ifelse}forall Metrics /Metrics
currentdict end def[2 index currentdict end definefont 3 -1 roll makefont
/setfont load]cvx def}def /ObliqueSlant{dup sin S cos div neg}B /SlantFont{4
index mul add}def /ExtendFont{3 -1 roll mul exch}def /ReEncodeFont{/Encoding
exch def}def end
%%EndProcSet
%%BeginProcSet: special.pro
TeXDict begin /SDict 200 dict N SDict begin /@SpecialDefaults{/hs 612 N /vs
792 N /ho 0 N /vo 0 N /hsc 1 N /vsc 1 N /ang 0 N /CLIP 0 N /rwiSeen false N
/rhiSeen false N /letter{}N /note{}N /a4{}N /legal{}N}B /@scaleunit 100 N
/@hscale{@scaleunit div /hsc X}B /@vscale{@scaleunit div /vsc X}B /@hsize{/hs
X /CLIP 1 N}B /@vsize{/vs X /CLIP 1 N}B /@clip{/CLIP 2 N}B /@hoffset{/ho X}B
/@voffset{/vo X}B /@angle{/ang X}B /@rwi{10 div /rwi X /rwiSeen true N}B /@rhi
{10 div /rhi X /rhiSeen true N}B /@llx{/llx X}B /@lly{/lly X}B /@urx{/urx X}B
/@ury{/ury X}B /magscale true def end /@MacSetUp{userdict /md known{userdict
/md get type /dicttype eq{userdict begin md length 10 add md maxlength ge{/md
md dup length 20 add dict copy def}if end md begin /letter{}N /note{}N /legal{
}N /od{txpose 1 0 mtx defaultmatrix dtransform S atan/pa X newpath clippath
mark{transform{itransform moveto}}{transform{itransform lineto}}{6 -2 roll
transform 6 -2 roll transform 6 -2 roll transform{itransform 6 2 roll
itransform 6 2 roll itransform 6 2 roll curveto}}{{closepath}}pathforall
newpath counttomark array astore /gc xdf pop ct 39 0 put 10 fz 0 fs 2
F/|______Courier fnt invertflag{PaintBlack}if}N /txpose{pxs pys scale ppr
aload pop por{noflips{pop S neg S TR pop 1 -1 scale}if xflip yflip and{pop S
neg S TR 180 rotate 1 -1 scale ppr 3 get ppr 1 get neg sub neg ppr 2 get ppr 0
get neg sub neg TR}if xflip yflip not and{pop S neg S TR pop 180 rotate ppr 3
get ppr 1 get neg sub neg 0 TR}if yflip xflip not and{ppr 1 get neg ppr 0 get
neg TR}if}{noflips{TR pop pop 270 rotate 1 -1 scale}if xflip yflip and{TR pop
pop 90 rotate 1 -1 scale ppr 3 get ppr 1 get neg sub neg ppr 2 get ppr 0 get
neg sub neg TR}if xflip yflip not and{TR pop pop 90 rotate ppr 3 get ppr 1 get
neg sub neg 0 TR}if yflip xflip not and{TR pop pop 270 rotate ppr 2 get ppr 0
get neg sub neg 0 S TR}if}ifelse scaleby96{ppr aload pop 4 -1 roll add 2 div 3
1 roll add 2 div 2 copy TR .96 dup scale neg S neg S TR}if}N /cp{pop pop
showpage pm restore}N end}if}if}N /normalscale{Resolution 72 div VResolution
72 div neg scale magscale{DVImag dup scale}if 0 setgray}N /psfts{S 65781.76
div N}N /startTexFig{/psf$SavedState save N userdict maxlength dict begin
/magscale false def normalscale currentpoint TR /psf$ury psfts /psf$urx psfts
/psf$lly psfts /psf$llx psfts /psf$y psfts /psf$x psfts currentpoint /psf$cy X
/psf$cx X /psf$sx psf$x psf$urx psf$llx sub div N /psf$sy psf$y psf$ury
psf$lly sub div N psf$sx psf$sy scale psf$cx psf$sx div psf$llx sub psf$cy
psf$sy div psf$ury sub TR /showpage{}N /erasepage{}N /copypage{}N /p 3 def
@MacSetUp}N /doclip{psf$llx psf$lly psf$urx psf$ury currentpoint 6 2 roll
newpath 4 copy 4 2 roll moveto 6 -1 roll S lineto S lineto S lineto closepath
clip newpath moveto}N /endTexFig{end psf$SavedState restore}N /@beginspecial{
SDict begin /SpecialSave save N gsave normalscale currentpoint TR
@SpecialDefaults count /ocount X /dcount countdictstack N}N /@setspecial{CLIP
1 eq{newpath 0 0 moveto hs 0 rlineto 0 vs rlineto hs neg 0 rlineto closepath
clip}if ho vo TR hsc vsc scale ang rotate rwiSeen{rwi urx llx sub div rhiSeen{
rhi ury lly sub div}{dup}ifelse scale llx neg lly neg TR}{rhiSeen{rhi ury lly
sub div dup scale llx neg lly neg TR}if}ifelse CLIP 2 eq{newpath llx lly
moveto urx lly lineto urx ury lineto llx ury lineto closepath clip}if
/showpage{}N /erasepage{}N /copypage{}N newpath}N /@endspecial{count ocount
sub{pop}repeat countdictstack dcount sub{end}repeat grestore SpecialSave
restore end}N /@defspecial{SDict begin}N /@fedspecial{end}B /li{lineto}B /rl{
rlineto}B /rc{rcurveto}B /np{/SaveX currentpoint /SaveY X N 1 setlinecap
newpath}N /st{stroke SaveX SaveY moveto}N /fil{fill SaveX SaveY moveto}N
/ellipse{/endangle X /startangle X /yrad X /xrad X /savematrix matrix
currentmatrix N TR xrad yrad scale 0 0 1 startangle endangle arc savematrix
setmatrix}N end
%%EndProcSet
TeXDict begin 40258431 52099146 1000 300 300
(/stumm/a0/tandri/pdpta/pdpta.dvi) @start /Fa 175[27 7[27 1[27
70[{}3 45.833332 /Courier rf /Fb 80[25 25 51[20 23 23 33 23
23 13 18 15 23 23 23 23 36 13 23 1[13 23 23 15 20 23 20 23
20 3[15 1[15 2[33 2[33 28 25 30 1[25 33 33 41 28 33 1[15 33
1[25 28 33 30 30 33 5[13 3[23 23 4[23 2[11 15 11 1[23 15 15
3[23 2[15 33[{}60 45.833332 /Times-Roman rf /Fc 81[29 51[23
26 2[26 29 16 23 23 2[29 29 42 16 2[16 29 29 16 26 29 26 29
29 13[29 2[36 42 1[48 6[36 1[42 39 1[36 11[29 29 29 29 29 2[15
19 45[{}36 58.333336 /Times-Italic rf /Fd 134[30 2[30 30 30
30 30 1[30 30 30 30 30 30 1[30 30 30 30 30 30 30 30 30 12[30
6[30 3[30 2[30 30 30 30 30 30 14[30 4[30 30 1[30 30 30 40[{}36
50.000000 /Courier rf /Fe 134[22 22 33 1[25 14 19 19 25 25
25 25 36 14 22 1[14 25 25 14 22 25 22 25 25 9[41 2[28 25 30
1[30 36 1[41 28 33 22 17 36 2[30 36 33 1[30 7[25 4[25 25 25
25 2[12 17 5[17 39[{}47 50.000000 /Times-Italic rf /Ff 1 1
df<FFFFF0FFFFF014027D881B>0 D E /Fg 4 117 df<1F0006000600060006000C000C000C00
0C0018F01B181C08180838183018301830306030603160616062C022C03C10177E9614>104
D<0300038003000000000000000000000000001C002400460046008C000C001800180018003100
3100320032001C0009177F960C>I<383C0044C6004702004602008E06000C06000C06000C0C00
180C00180C40181840181880300880300F00120E7F8D15>110 D<030003000600060006000600
FFC00C000C000C001800180018001800300030803080310031001E000A147F930D>116
D E /Fh 3 3 df<FFFFFFFCFFFFFFFC1E027C8C27>0 D<70F8F8F87005057C8E0E>I<C00003E0
000770000E38001C1C00380E00700700E00381C001C38000E700007E00003C00003C00007E0000
E70001C3800381C00700E00E00701C003838001C70000EE00007C000031818799727>I
E /Fi 4 62 df<00200040008001000300060004000C000C001800180030003000300070006000
60006000E000E000E000E000E000E000E000E000E000E000E000E000E000E00060006000600070
00300030003000180018000C000C0004000600030001000080004000200B327CA413>40
D<800040002000100018000C000400060006000300030001800180018001C000C000C000C000E0
00E000E000E000E000E000E000E000E000E000E000E000E000E000C000C000C001C00180018001
80030003000600060004000C00180010002000400080000B327DA413>I<000180000001800000
018000000180000001800000018000000180000001800000018000000180000001800000018000
00018000000180000001800000018000FFFFFFFEFFFFFFFE000180000001800000018000000180
000001800000018000000180000001800000018000000180000001800000018000000180000001
800000018000000180001F227D9C26>43 D<FFFFFFFEFFFFFFFE00000000000000000000000000
00000000000000000000000000000000000000FFFFFFFEFFFFFFFE1F0C7D9126>61
D E /Fj 16 111 df<003F000000E180000380C020070060400E0070401C0070403C0070803C00
3880780039007800390078003A00F0003A00F0003C00F0003800F0003800700038007000780030
00B800380338401C1C188007E00F001B157E941F>11 D<70F8FCFC740404040408081010204006
0F7C840E>59 D<00000080000000018000000001C000000003C000000003C000000007C0000000
0BC00000000BC000000013C000000033C000000023C000000043C000000043E000000081E00000
0181E000000101E000000201E000000201E000000401E000000C01E000000801E000001001E000
001FFFF000002000F000006000F000004000F000008000F000008000F000010000F000030000F0
00020000F000040000F8000C0000F8001E0000F800FF800FFF8021237EA225>65
D<007FFFF8000007800F00000780078000078003C0000F0001C0000F0001C0000F0001E0000F00
01E0001E0001C0001E0003C0001E0003C0001E000780003C000F00003C001E00003C003C00003C
01F000007FFFE00000780078000078003C000078001E0000F0001E0000F0000E0000F0000F0000
F0000F0001E0001E0001E0001E0001E0001E0001E0003C0003C0003C0003C000780003C000F000
03C001C00007C00F8000FFFFFC000023227EA125>I<007FFFF0000007801C000007800F000007
800700000F000380000F000380000F000380000F000380001E000780001E000780001E00078000
1E000F00003C000F00003C001E00003C003C00003C007000007801E000007FFF00000078000000
007800000000F000000000F000000000F000000000F000000001E000000001E000000001E00000
0001E000000003C000000003C000000003C000000003C000000007C0000000FFFC00000021227E
A11F>80 D<007FFFE0000007803C000007800E000007800700000F000780000F000380000F0003
C0000F0003C0001E000780001E000780001E000780001E000F00003C001E00003C003C00003C00
7000003C01C000007FFE00000078078000007801C000007801E00000F000F00000F000F00000F0
00F00000F000F00001E001E00001E001E00001E001E00001E001E00003C003C00003C003C04003
C003C04003C001C08007C001C080FFFC00E3000000003C0022237EA125>82
D<3FFE01FF8003C0003C0003C000300003C0001000078000200007800020000780002000078000
20000F000040000F000040000F000040000F000040001E000080001E000080001E000080001E00
0080003C000100003C000100003C000100003C0001000078000200007800020000780002000078
000200007000040000F000040000F0000800007000080000700010000070002000003800400000
38008000001C01000000060600000001F800000021237DA121>85 D<FFF8007FC00F80000F000F
00000C000F000008000F000010000F800010000780002000078000600007800040000780008000
07800080000780010000078002000007C002000003C004000003C00C000003C008000003C01000
0003C010000003C020000003E040000003E040000001E080000001E180000001E100000001E200
000001E200000001E400000001F800000000F800000000F000000000E000000000E000000000C0
00000000C000000022237DA11C>I<007FFC03FF0007E000F80007C000E00003C000800003E001
000001E002000001F006000001F00C000000F018000000F81000000078200000007C400000007C
800000003D000000003E000000001E000000001F000000001F000000002F000000006F80000000
C78000000187C000000103C000000203C000000403E000000801E000001001F000002000F00000
4000F800008000F80001800078000300007C000F8000FC00FFE007FFC028227FA128>88
D<00001E00000063800000C7800001C7800001C300000180000003800000038000000380000003
80000007000000070000000700000007000000FFF800000E0000000E0000000E0000000E000000
0E0000000E0000001C0000001C0000001C0000001C0000001C0000003800000038000000380000
0038000000380000007000000070000000700000007000000060000000E0000000E0000000E000
0000C0000070C00000F1800000F1000000620000003C000000192D7EA218>102
D<000F0C00389C00605C00C03801C0380380380780380700700F00700F00700F00701E00E01E00
E01E00E01E00E01E01C00E01C00E03C00605C0031B8001E3800003800003800007000007000007
00700E00F00C00F018006070003FC000161F809417>I<00F0000FE00000E00000E00000E00001
C00001C00001C00001C000038000038000038000038000070000071F0007218007C0C00F00E00F
00E00E00E00E00E01C01C01C01C01C01C01C01C0380380380380380700380704700708700E0870
0E08700610E006206003C016237DA21C>I<00E000E001E000C000000000000000000000000000
00000000001E0023004380438083808380870007000E000E000E001C001C003800382038407040
7040308031001E000B227EA111>I<0000E00001E00001E00000C0000000000000000000000000
000000000000000000000000000000001E00002300004380008380010380010380010380000700
000700000700000700000E00000E00000E00000E00001C00001C00001C00001C00003800003800
00380000380000700000700000700070E000F0C000F180006300003E0000132C81A114>I<00F0
000FE00000E00000E00000E00001C00001C00001C00001C0000380000380000380000380000700
000700F00703080704380E08780E10780E20300E40001C80001F00001FC0001C70003838003838
00381C00381C10703820703820703820701840E00C8060070015237DA219>I<3C07C046186047
20308740388780388700388700380E00700E00700E00700E00701C00E01C00E01C01C01C01C138
01C23803823803823801847001883000F018157E941D>110 D E /Fk 134[21
1[30 1[21 12 16 14 1[21 21 21 32 12 2[12 21 21 14 18 21 18
21 18 3[14 1[14 17[14 5[28 8[12 21 21 5[21 21 1[12 10 14 45[{}32
41.666668 /Times-Roman rf /Fl 203[15 15 15 15 49[{}4 29.166668
/Times-Roman rf /Fm 203[17 17 17 17 17 48[{}5 33.333332 /Times-Roman
rf /Fn 138[39 23 27 31 1[39 35 39 59 20 39 1[20 1[35 23 31
39 31 39 35 9[71 4[51 1[43 6[27 2[43 1[51 51 11[35 35 35 35
35 35 35 49[{}32 70.833336 /Times-Bold rf /Fo 69[22 8[25 1[28
28 3[22 47[22 25 25 36 25 25 14 19 17 25 25 25 25 39 14 25
14 14 25 25 17 22 25 22 25 22 3[17 1[17 30 2[47 36 36 30 28
33 1[28 36 36 44 30 36 19 17 36 36 28 30 36 33 33 36 3[28 1[14
14 25 25 25 25 25 25 25 25 25 25 1[12 17 12 2[17 17 17 39[{}75
50.000000 /Times-Roman rf /Fp 139[17 19 22 14[22 28 25 31[36
65[{}7 50.000000 /Times-Bold rf /Fq 2 104 df<0000F80003C0000F00001E00003C0000
780000780000780000780000780000780000780000780000780000780000780000780000780000
780000780000780000780000780000780000780000F00000F00001E000078000FE0000FE000007
800001E00000F00000F00000780000780000780000780000780000780000780000780000780000
7800007800007800007800007800007800007800007800007800007800007800003C00001E0000
0F000003C00000F8153C7CAC1E>102 D<F800000F000003C00001E00000F00000780000780000
780000780000780000780000780000780000780000780000780000780000780000780000780000
7800007800007800007800007800003C00003C00001E000007000001F80001F8000700001E0000
3C00003C0000780000780000780000780000780000780000780000780000780000780000780000
780000780000780000780000780000780000780000780000780000F00001E00003C0000F0000F8
0000153C7CAC1E>I E /Fr 134[29 2[29 29 16 23 19 1[29 29 29 45
16 29 1[16 29 29 19 26 29 26 29 26 11[42 36 32 5[52 7[36 42
39 1[42 54 5[16 4[29 29 2[29 2[15 19 15 44[{}37 58.333336 /Times-Roman
rf /Fs 134[42 3[46 28 32 37 1[46 42 46 69 23 2[23 46 42 1[37
46 37 46 42 13[46 2[51 2[78 8[60 60 67[{}23 83.333336 /Times-Bold
rf end
%%EndProlog
%%BeginSetup
%%Feature: *Resolution 300dpi
TeXDict begin

%%EndSetup
%%Page: 1 1
1 0 bop 80 177 a Fs(Computation)19 b(and)i(Data)e(Partitioning)g(on)h
(Scalable)341 280 y(Shar)o(ed)g(Memory)g(Multipr)o(ocessors)403
451 y Fr(Sudarsan)15 b(T)l(andri)29 b(and)g(T)l(arek)14 b(S.)g(Abdelrahman)
316 526 y(Department)h(of)g(Electrical)f(and)h(Computer)g(Engineering)284
601 y(The)f(University)h(of)g(T)l(oronto,)f(T)l(oronto,)g(Canada,)f(M5S)i
(1A4)478 675 y(e-mail:)g Fq(f)p Fr(tandri,tsa)p Fq(g)p Fr(@eecg.toronto.edu)
833 865 y Fp(Abstract)217 945 y Fo(In)g(this)h(paper)f(we)h(identify)f(the)h
(factors)f(that)h(af)o(fect)f(the)h(derivation)e(of)i(com-)217
999 y(putation)10 b(and)h(data)g(partitions)g(on)g(scalable)g(shared)g
(memory)g(multiprocessors)217 1053 y(\(SSMMs\).)18 b(W)l(e)12
b(show)h(that)f(these)h(factors)f(necessitate)i(an)e(SSMM-conscious)217
1107 y(approach.)17 b(In)10 b(addition)g(to)g(remote)g(memory)f(access,)k
(which)d(is)h(the)f(sole)h(factor)217 1161 y(on)19 b(distributed)g(memory)f
(multiprocessors,)k(cache)d(af)o(\256nity)m(,)i(memory)e(con-)217
1216 y(tention)12 b(and)h(false)g(sharing)f(are)h(important)f(factors)g(that)
h(must)g(be)g(considered.)217 1270 y(Experimental)g(evidence)h(is)g
(presented)g(to)g(demonstrate)f(the)h(impact)f(of)h(these)217
1324 y(factors)i(on)g(performance)g(using)g(three)h(applications)f(on)h(the)f
(KSR1)h(and)f(the)217 1378 y(Hector)c(multiprocessors.)4 1540
y Fn(1)71 b(Intr)o(oduction)4 1667 y Fo(Scalable)12 b(shared)g(memory)f
(multiprocessors)g(\(SSMMs\))g(are)h(becoming)f(increasingly)h(popular)f(and)
h(a)4 1721 y(viable)e(alternative)f(to)h(distributed)f(memory)g
(multiprocessors)h(\(DMMs\).)17 b(The)11 b(Stanford)e(DASH)g([20],)4
1775 y(FLASH)i([14)o(],)h(the)f(KSR1)f([24],)h(T)m(oronto')m(s)f(Hector)h
([26)o(],)h(NUMAchine)f([1)o(],)h(and)f(the)f(Cray)h(T3D)h([23)o(])4
1830 y(are)d(some)g(SSMMs)g(currently)e(in)i(use)g(or)f(under)g(development.)
17 b(Processors)9 b(in)f(a)h(SSMM)g(share)g(a)g(single)4 1884
y(coherent)f(address)g(space.)17 b(However)n(,)9 b(shared)f(memory)g(is)g
(physically)g(distributed)g(to)f(allow)h(scalability)l(,)4
1938 y(as)17 b(shown)f(in)g(Figure)f(1.)29 b(This)17 b(distribution)e(of)g
(shared)i(memory)e(results)h(in)g(non-uniform)e(memory)4 1992
y(access)f(latencies,)g(depending)f(on)f(the)h(distance)h(between)f(a)g
(processor)f(and)h(memory)m(.)17 b(Consequently)m(,)4 2046
y(careful)12 b(placement)g(and)g(management)g(of)g(data)h(is)g(essential)g
(for)e(scaling)i(performance.)77 2122 y(W)l(e)i(believe)f(that)g(data)g
(distribution)732 2104 y Fm(1)764 2122 y Fo(is)g(a)h(good)f(paradigm)f(for)h
(managing)f(data)i(in)f(data-parallel)4 2176 y(applications)h(on)g(SSMMs)g
([3)o(,)h(21].)25 b(The)16 b(division)e(of)h(array)f(data)h(allows)g(a)g
(compiler)f(to)h(place)g(data)4 2230 y(in)g(the)g(physical)f(memory)g(of)h
(the)g(processor)f(that)h(uses)h(it)e(the)h(most,)h(and)f(also)g(allows)g
(the)g(compiler)4 2284 y(to)k(partition)f(the)h(computations)g(of)f(parallel)
h(loops.)38 b(W)l(e)19 b(have)g(experimented)g(with)f(programmer)4
2339 y(speci\256ed)12 b(data)f(distributions)g(on)g(the)h(Hector)f
(multiprocessor)f(and)i(have)g(found)e(them)h(to)h(be)f(ef)o(fective)4
2393 y(in)e(improving)e(performance.)16 b(However)n(,)10 b(the)e(task)h(of)g
(selecting)g(a)g(good)f(data)h(distribution)f(requires)g(the)4
2447 y(programmer)i(to)h(understand)g(both)f(the)i(parallel)e(machine)h
(architecture)g(and)g(the)g(data)g(access)i(patterns)4 2501
y(in)19 b(the)f(program.)37 b(Porting)17 b(programs)h(to)h(various)g
(machines)g(and)f(tuning)h(them)f(for)g(performance)4 2555
y(becomes)g(a)f(tedious)g(and)g(laborious)g(process.)33 b(Consequently)m(,)19
b(it)e(is)h(desirable)f(to)g(derive)f(data)i(and)4 2609 y(computation)h
(partitions)g(automatically)h(using)g(a)g(compiler)m(.)40 b(The)21
b(objective)e(of)h(this)g(paper)g(is)g(to)4 2664 y(describe)13
b(the)f(factors)g(that)g(af)o(fect)g(the)g(derivation)g(of)g(computation)f
(and)i(data)f(partitions)g(on)g(SSMMs.)77 2739 y(On)19 b(DMMs,)k(the)c(main)g
(factor)f(that)i(af)o(fects)f(the)g(performance)f(of)h(an)g(application)g(is)
g(the)g(cost)4 2793 y(of)d(interprocessor)f(communication.)28
b(Consequently)m(,)17 b(scalable)g(performance)e(can)h(be)g(achieved)g(by)p
4 2838 737 2 v 62 2869 a Fl(1)79 2884 y Fk(In)10 b(this)f(paper)i(we)g(use)f
(the)g(terms)h(data)g(distributi)o(ons)c(and)k(data)f(partitions)f
(interchangeably)m(.)p eop
%%Page: 2 2
2 1 bop 175 533 a @beginspecial 114 @llx 408 @lly 476 @urx
553 @ury 3600 @rwi @setspecial
%%BeginDocument: numaarch1.ps

/arrowHeight 10 def
/arrowWidth 5 def

/IdrawDict 51 dict def
IdrawDict begin

/reencodeISO {
dup dup findfont dup length dict begin
{ 1 index /FID ne { def }{ pop pop } ifelse } forall
/Encoding ISOLatin1Encoding def
currentdict end definefont
} def

/ISOLatin1Encoding [
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/space/exclam/quotedbl/numbersign/dollar/percent/ampersand/quoteright
/parenleft/parenright/asterisk/plus/comma/minus/period/slash
/zero/one/two/three/four/five/six/seven/eight/nine/colon/semicolon
/less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N
/O/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright
/asciicircum/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m
/n/o/p/q/r/s/t/u/v/w/x/y/z/braceleft/bar/braceright/asciitilde
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/dotlessi/grave/acute/circumflex/tilde/macron/breve
/dotaccent/dieresis/.notdef/ring/cedilla/.notdef/hungarumlaut
/ogonek/caron/space/exclamdown/cent/sterling/currency/yen/brokenbar
/section/dieresis/copyright/ordfeminine/guillemotleft/logicalnot
/hyphen/registered/macron/degree/plusminus/twosuperior/threesuperior
/acute/mu/paragraph/periodcentered/cedilla/onesuperior/ordmasculine
/guillemotright/onequarter/onehalf/threequarters/questiondown
/Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE/Ccedilla
/Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex
/Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis
/multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute
/Thorn/germandbls/agrave/aacute/acircumflex/atilde/adieresis
/aring/ae/ccedilla/egrave/eacute/ecircumflex/edieresis/igrave
/iacute/icircumflex/idieresis/eth/ntilde/ograve/oacute/ocircumflex
/otilde/odieresis/divide/oslash/ugrave/uacute/ucircumflex/udieresis
/yacute/thorn/ydieresis
] def
/Times-Roman reencodeISO def

/none null def
/numGraphicParameters 17 def
/stringLimit 65535 def

/Begin {
save
numGraphicParameters dict begin
} def

/End {
end
restore
} def

/SetB {
dup type /nulltype eq {
pop
false /brushRightArrow idef
false /brushLeftArrow idef
true /brushNone idef
} {
/brushDashOffset idef
/brushDashArray idef
0 ne /brushRightArrow idef
0 ne /brushLeftArrow idef
/brushWidth idef
false /brushNone idef
} ifelse
} def

/SetCFg {
/fgblue idef
/fggreen idef
/fgred idef
} def

/SetCBg {
/bgblue idef
/bggreen idef
/bgred idef
} def

/SetF {
/printSize idef
/printFont idef
} def

/SetP {
dup type /nulltype eq {
pop true /patternNone idef
} {
dup -1 eq {
/patternGrayLevel idef
/patternString idef
} {
/patternGrayLevel idef
} ifelse
false /patternNone idef
} ifelse
} def

/BSpl {
0 begin
storexyn
newpath
n 1 gt {
0 0 0 0 0 0 1 1 true subspline
n 2 gt {
0 0 0 0 1 1 2 2 false subspline
1 1 n 3 sub {
/i exch def
i 1 sub dup i dup i 1 add dup i 2 add dup false subspline
} for
n 3 sub dup n 2 sub dup n 1 sub dup 2 copy false subspline
} if
n 2 sub dup n 1 sub dup 2 copy 2 copy false subspline
patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if
brushNone not { istroke } if
0 0 1 1 leftarrow
n 2 sub dup n 1 sub dup rightarrow
} if
end
} dup 0 4 dict put def

/Circ {
newpath
0 360 arc
patternNone not { ifill } if
brushNone not { istroke } if
} def

/CBSpl {
0 begin
dup 2 gt {
storexyn
newpath
n 1 sub dup 0 0 1 1 2 2 true subspline
1 1 n 3 sub {
/i exch def
i 1 sub dup i dup i 1 add dup i 2 add dup false subspline
} for
n 3 sub dup n 2 sub dup n 1 sub dup 0 0 false subspline
n 2 sub dup n 1 sub dup 0 0 1 1 false subspline
patternNone not { ifill } if
brushNone not { istroke } if
} {
Poly
} ifelse
end
} dup 0 4 dict put def

/Elli {
0 begin
newpath
4 2 roll
translate
scale
0 0 1 0 360 arc
patternNone not { ifill } if
brushNone not { istroke } if
end
} dup 0 1 dict put def

/Line {
0 begin
2 storexyn
newpath
x 0 get y 0 get moveto
x 1 get y 1 get lineto
brushNone not { istroke } if
0 0 1 1 leftarrow
0 0 1 1 rightarrow
end
} dup 0 4 dict put def

/MLine {
0 begin
storexyn
newpath
n 1 gt {
x 0 get y 0 get moveto
1 1 n 1 sub {
/i exch def
x i get y i get lineto
} for
patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if
brushNone not { istroke } if
0 0 1 1 leftarrow
n 2 sub dup n 1 sub dup rightarrow
} if
end
} dup 0 4 dict put def

/Poly {
3 1 roll
newpath
moveto
-1 add
{ lineto } repeat
closepath
patternNone not { ifill } if
brushNone not { istroke } if
} def

/Rect {
0 begin
/t exch def
/r exch def
/b exch def
/l exch def
newpath
l b moveto
l t lineto
r t lineto
r b lineto
closepath
patternNone not { ifill } if
brushNone not { istroke } if
end
} dup 0 4 dict put def

/Text {
ishow
} def

/idef {
dup where { pop pop pop } { exch def } ifelse
} def

/ifill {
0 begin
gsave
patternGrayLevel -1 ne {
fgred bgred fgred sub patternGrayLevel mul add
fggreen bggreen fggreen sub patternGrayLevel mul add
fgblue bgblue fgblue sub patternGrayLevel mul add setrgbcolor
eofill
} {
eoclip
originalCTM setmatrix
pathbbox /t exch def /r exch def /b exch def /l exch def
/w r l sub ceiling cvi def
/h t b sub ceiling cvi def
/imageByteWidth w 8 div ceiling cvi def
/imageHeight h def
bgred bggreen bgblue setrgbcolor
eofill
fgred fggreen fgblue setrgbcolor
w 0 gt h 0 gt and {
l b translate w h scale
w h true [w 0 0 h neg 0 h] { patternproc } imagemask
} if
} ifelse
grestore
end
} dup 0 8 dict put def

/istroke {
gsave
brushDashOffset -1 eq {
[] 0 setdash
1 setgray
} {
brushDashArray brushDashOffset setdash
fgred fggreen fgblue setrgbcolor
} ifelse
brushWidth setlinewidth
originalCTM setmatrix
stroke
grestore
} def

/ishow {
0 begin
gsave
fgred fggreen fgblue setrgbcolor
/fontDict printFont printSize scalefont dup setfont def
/descender fontDict begin 0 [FontBBox] 1 get FontMatrix end
transform exch pop def
/vertoffset 1 printSize sub descender sub def {
0 vertoffset moveto show
/vertoffset vertoffset printSize sub def
} forall
grestore
end
} dup 0 3 dict put def
/patternproc {
0 begin
/patternByteLength patternString length def
/patternHeight patternByteLength 8 mul sqrt cvi def
/patternWidth patternHeight def
/patternByteWidth patternWidth 8 idiv def
/imageByteMaxLength imageByteWidth imageHeight mul
stringLimit patternByteWidth sub min def
/imageMaxHeight imageByteMaxLength imageByteWidth idiv patternHeight idiv
patternHeight mul patternHeight max def
/imageHeight imageHeight imageMaxHeight sub store
/imageString imageByteWidth imageMaxHeight mul patternByteWidth add string def
0 1 imageMaxHeight 1 sub {
/y exch def
/patternRow y patternByteWidth mul patternByteLength mod def
/patternRowString patternString patternRow patternByteWidth getinterval def
/imageRow y imageByteWidth mul def
0 patternByteWidth imageByteWidth 1 sub {
/x exch def
imageString imageRow x add patternRowString putinterval
} for
} for
imageString
end
} dup 0 12 dict put def

/min {
dup 3 2 roll dup 4 3 roll lt { exch } if pop
} def

/max {
dup 3 2 roll dup 4 3 roll gt { exch } if pop
} def

/midpoint {
0 begin
/y1 exch def
/x1 exch def
/y0 exch def
/x0 exch def
x0 x1 add 2 div
y0 y1 add 2 div
end
} dup 0 4 dict put def

/thirdpoint {
0 begin
/y1 exch def
/x1 exch def
/y0 exch def
/x0 exch def
x0 2 mul x1 add 3 div
y0 2 mul y1 add 3 div
end
} dup 0 4 dict put def

/subspline {
0 begin
/movetoNeeded exch def
y exch get /y3 exch def
x exch get /x3 exch def
y exch get /y2 exch def
x exch get /x2 exch def
y exch get /y1 exch def
x exch get /x1 exch def
y exch get /y0 exch def
x exch get /x0 exch def
x1 y1 x2 y2 thirdpoint
/p1y exch def
/p1x exch def
x2 y2 x1 y1 thirdpoint
/p2y exch def
/p2x exch def
x1 y1 x0 y0 thirdpoint
p1x p1y midpoint
/p0y exch def
/p0x exch def
x2 y2 x3 y3 thirdpoint
p2x p2y midpoint
/p3y exch def
/p3x exch def
movetoNeeded { p0x p0y moveto } if
p1x p1y p2x p2y p3x p3y curveto
end
} dup 0 17 dict put def

/storexyn {
/n exch def
/y n array def
/x n array def
n 1 sub -1 0 {
/i exch def
y i 3 2 roll put
x i 3 2 roll put
} for
} def

/SSten {
fgred fggreen fgblue setrgbcolor
dup true exch 1 0 0 -1 0 6 -1 roll matrix astore
} def

/FSten {
dup 3 -1 roll dup 4 1 roll exch
newpath
0 0 moveto
dup 0 exch lineto
exch dup 3 1 roll exch lineto
0 lineto
closepath
bgred bggreen bgblue setrgbcolor
eofill
SSten
} def

/Rast {
exch dup 3 1 roll 1 0 0 -1 0 6 -1 roll matrix astore
} def


/arrowhead {
0 begin
transform originalCTM itransform
/taily exch def
/tailx exch def
transform originalCTM itransform
/tipy exch def
/tipx exch def
/dy tipy taily sub def
/dx tipx tailx sub def
/angle dx 0 ne dy 0 ne or { dy dx atan } { 90 } ifelse def
gsave
originalCTM setmatrix
tipx tipy translate
angle rotate
newpath
arrowHeight neg arrowWidth 2 div moveto
0 0 lineto
arrowHeight neg arrowWidth 2 div neg lineto
patternNone not {
originalCTM setmatrix
/padtip arrowHeight 2 exp 0.25 arrowWidth 2 exp mul add sqrt brushWidth mul
arrowWidth div def
/padtail brushWidth 2 div def
tipx tipy translate
angle rotate
padtip 0 translate
arrowHeight padtip add padtail add arrowHeight div dup scale
arrowheadpath
ifill
} if
brushNone not {
originalCTM setmatrix
tipx tipy translate
angle rotate
arrowheadpath
istroke
} if
grestore
end
} dup 0 9 dict put def

/arrowheadpath {
newpath
arrowHeight neg arrowWidth 2 div moveto
0 0 lineto
arrowHeight neg arrowWidth 2 div neg lineto
} def

/leftarrow {
0 begin
y exch get /taily exch def
x exch get /tailx exch def
y exch get /tipy exch def
x exch get /tipx exch def
brushLeftArrow { tipx tipy tailx taily arrowhead } if
end
} dup 0 4 dict put def

/rightarrow {
0 begin
y exch get /tipy exch def
x exch get /tipx exch def
y exch get /taily exch def
x exch get /tailx exch def
brushRightArrow { tipx tipy tailx taily arrowhead } if
end
} dup 0 4 dict put def




Begin
[ 0.799705 0 0 0.799705 0 0 ] concat
/originalCTM matrix currentmatrix def

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 433.125 504.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 433.125 552.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 265.125 504.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 265.125 552.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 137.125 504.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 137.125 552.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 321.125 600.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 489.125 600.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 193.125 600.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 144 410 ] concat
453 529 448 32 Elli
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 486.625 598.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 505.272 606.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 318.625 598.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 337.272 606.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 190.625 598.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 209.272 606.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 134.625 502.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 153.272 510.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.125 -0 -0 0.125 262.625 502.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.125 -0 -0 0.125 430.625 502.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 362.875 611.625 ] concat
617 99 16 16 Elli
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 378.875 611.625 ] concat
617 99 16 16 Elli
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 133.5 439.5 ] concat
117 369 181 369 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 261.5 439.5 ] concat
117 369 181 369 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 429.5 439.5 ] concat
117 369 181 369 Line
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 179.5 532 ] concat
[
(Procr)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 307.5 532 ] concat
[
(Procr)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 475.5 532 ] concat
[
(Procr)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 235.5 628 ] concat
[
(Mem)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 363.5 628 ] concat
[
(Mem)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 531.5 628 ] concat
[
(Mem)
] Text
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 346.875 611.625 ] concat
617 99 16 16 Elli
End

Begin %I BSpl
1 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 19.5 393.5 ] concat
457 333
473 381
441 365
457 413
4 BSpl
End

Begin %I BSpl
1 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 147.5 393.5 ] concat
457 333
473 381
441 365
457 413
4 BSpl
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 360 555 ] concat
[
(Remote)
(memory)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 232 555 ] concat
[
(Local)
(memory)
] Text
End

Begin %I BSpl
1 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 315.5 393.5 ] concat
457 333
473 381
441 365
457 413
4 BSpl
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 528 555 ] concat
[
(Remote)
(memory)
] Text
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 134.625 550.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 153.272 558.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.125 -0 -0 0.125 262.625 550.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.125 -0 -0 0.125 430.625 550.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 177 580 ] concat
[
(Cache)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 305 580 ] concat
[
(Cache)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 473 580 ] concat
[
(Cache)
] Text
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 60 364 ] concat
132 180 132 196 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 188 364 ] concat
132 180 132 196 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 356 364 ] concat
132 180 132 196 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 49 237 ] concat
143 355 143 435 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 49 237 ] concat
439 355 439 427 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 49 237 ] concat
271 355 271 427 Line
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.5 -0 -0 0.5 141.5 407.5 ] concat
453 529 448 32 Elli
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 306.5 676 ] concat
[
(Interconnection Network)
] Text
End

End %I eop

showpage


end
%%EndDocument
 @endspecial 295 598 a Fo(Figure)11 b(1:)18 b(Scalable)12 b(shared-memory)f
(multiprocessor)h(architecture.)4 719 y(partitioning)g(data)i(and)g
(computations)f(in)g(a)h(way)f(that)h(minimizes)f(interprocessor)g
(communications.)4 773 y(On)f(SSMMs,)h(processors)f(communicate)f(through)g
(shared)g(memory)m(,)h(and)f(the)h(cost)g(of)f(interprocessor)4
827 y(communications)h(\(i.e.,)i(remote)e(memory)f(access\))j(is)f
(relatively)e(inexpensive.)19 b(W)l(e)13 b(show)g(that)f(cache)4
881 y(af)o(\256nity)m(,)i(memory)f(contention)h(and)g(false)g(sharing)g(are)g
(additional)g(factors)g(that)f(must)i(be)f(considered)4 935
y(in)i(the)g(selection)g(of)f(data)h(distributions.)28 b(Furthermore,)16
b(the)g(presence)g(of)f(a)h(single)g(shared)g(address)4 989
y(space)i(allows)g(\257exibility)f(in)g(the)h(selection)g(of)f(a)h
(computation)e(partition.)33 b(Speci\256cally)m(,)19 b(we)f(show)4
1044 y(that)h(relaxing)g(the)h(commonly)e(used)i(owner)o(-computes)f(rule)g
([15)o(])h(has)g(performance)e(advantages.)4 1098 y(W)l(e)d(present)g
(experimental)f(results)i(to)e(support)h(our)f(conclusions)i(using)f(three)f
(applications)h(on)g(two)4 1152 y(SSMMs,)e(the)g(Hector)f(and)g(the)g(KSR1)h
(multiprocessors.)77 1228 y(The)g(remainder)f(of)g(this)g(paper)g(is)h(or)o
(ganized)f(as)h(follows.)18 b(Section)12 b(2)h(presents)f(an)h(overview)f
(data)4 1282 y(distributions.)35 b(Section)18 b(3)g(describes)h(the)f
(factors)f(that)h(impact)g(on)g(the)h(selection)f(of)g(computation)4
1336 y(and)i(data)g(partitions)f(on)g(SSMMs.)41 b(Section)19
b(4)h(gives)g(experimental)f(evidence)h(of)f(the)h(impact)f(of)4
1390 y(cache)e(af)o(\256nity)e(and)h(false)g(sharing)g(on)g(the)g(choice)h
(of)e(data)h(partitions.)29 b(Section)16 b(5)g(presents)h(results)4
1444 y(to)d(show)g(that)g(the)g(\257exibility)f(in)h(selecting)h(the)f
(computation)f(partitioning)g(can)h(be)g(used)h(to)f(improve)4
1498 y(performance.)j(Section)9 b(6)i(reviews)f(related)g(work.)17
b(Finally)m(,)11 b(Section)e(7)i(presents)f(concluding)g(remarks)4
1553 y(and)j(directions)e(for)h(future)f(work.)4 1734 y Fn(2)71
b(Data)19 b(Distributions)4 1861 y Fo(Data)10 b(distribution)f([15)o(,)i(16])
e(is)i(achieved)f(by)g(specifying)f(a)h(partitioning)f(scheme)h(for)f(each)i
(array)e(in)h(the)4 1915 y(program)h(and)i(by)f(specifying)g(a)g(processor)h
(geometry)e(to)i(which)f(array)g(partitions)f(map.)18 b(A)13
b(processor)4 1970 y(geometry)g(is)i(an)f Fj(n)p Fo(-dimensional)f(Cartesian)
h(grid)f(of)h(virtual)f(processors)h Fi(\()p Fj(V)1385 1977
y Fm(0)1404 1970 y Fj(;)8 b(V)1454 1977 y Fm(1)1473 1970 y
Fj(;)g Fh(\001)g(\001)g(\001)g Fj(;)g(V)1611 1977 y Fg(n)p
Ff(\000)p Fm(1)1679 1970 y Fi(\))p Fo(,)15 b(where)4 2024 y
Fj(V)32 2031 y Fg(i)63 2024 y Fo(is)i(the)f(number)g(of)g(processors)h(in)f
(the)g Fj(i)793 2006 y Fg(th)844 2024 y Fo(dimension)g(of)g(the)h(grid,)g
(and)f Fj(V)1430 2031 y Fm(0)1463 2024 y Fh(\002)d Fj(V)1543
2031 y Fm(1)1575 2024 y Fh(\002)g(\001)8 b(\001)g(\001)14 b(\002)f
Fj(V)1779 2031 y Fg(n)p Ff(\000)p Fm(1)4 2078 y Fo(=)i Fj(P)7
b Fo(,)16 b(the)f(total)f(number)h(of)f(processors.)26 b(A)15
b(partitioning)e(scheme)i(assigns)h(a)f Fe(partitioning)f(attribute)4
2132 y Fo(to)k(each)g(dimension)g(the)g(array)m(.)34 b(There)18
b(are)g(four)f(partitioning)f(attributes.)35 b(The)18 b Fd(Block)g
Fo(attribute)4 2186 y(divides)f(the)g(corresponding)g(dimension)g(of)f(the)h
(array)g(in)g(approximately)f(equal)h(size)h(blocks)f(such)4
2240 y(that)j(a)g(processor)g(owns)h(a)f(contiguous)g(range)g(of)f(that)h
(dimension)g(of)g(the)g(array)m(.)41 b(The)20 b Fd(Cyclic)4
2295 y Fo(attribute)11 b(divides)h(the)h(corresponding)e(array)g(dimension)h
(by)g(distributing)f(the)h(array)f(elements)i(in)f(this)4 2349
y(dimension)g(to)g(processors)h(in)f(a)h(round-robin)d(fashion.)18
b(The)13 b Fd(BlockCyclic)f Fo(attribute)f(\256rst)h(groups)4
2403 y(array)f(elements)h(in)f(the)g(corresponding)g(dimension)g(in)g
(contiguous)g(blocks)h(of)f(a)h(given)f(size,)h(and)g(then)4
2457 y(assigns)k(the)f(blocks)f(to)h(processors)g(in)g(a)g(round-robin)d
(fashion.)26 b(The)15 b(block)f(size,)j(called)d(the)h Fe(block-)4
2511 y(cyclic)10 b(factor)p Fo(,)h(is)e(supplied)h(by)f(the)h(programmer)m(.)
16 b(Finally)m(,)9 b(the)h Fd(*)f Fo(attribute)g(is)h(used)g(to)f(indicate)g
(that)h(the)4 2565 y(corresponding)f(dimension)g(of)g(the)h(array)f(is)h(not)
f(distributed.)17 b(The)10 b(processor)g(geometry)f(on)g(which)h(the)4
2620 y(array)h(is)i(mapped)e(determines)h(the)g(number)f(of)g(processors)i
(assigned)f(to)g(each)g(distributed)f(dimension)4 2674 y(of)h(the)f(array)m
(.)18 b(For)11 b(example,)h(distributing)e(a)i(two)g(dimensional)f(array)h
(using)f(the)h Fd(\(Block,Block\))4 2728 y Fo(attributes)g(onto)h(a)g(two)f
(dimensional)h(processor)f(geometry)g(of)h(\(2,4\),)f(distributes)h(the)f
(array)h(on)f(to)h(the)4 2782 y(8)k(processors,)i(assigning)f(2)f(processors)
g(to)g(the)g(\256rst)g(dimension)g(and)g(4)g(processors)g(to)g(the)g(second)4
2836 y(dimension.)p eop
%%Page: 3 3
3 2 bop 4 -21 a Fn(3)71 b(Performance)21 b(Factors)4 106 y
Fo(The)15 b(main)g(factor)f(that)g(af)o(fects)h(the)f(performance)g(of)g(a)h
(parallel)f(application)g(on)h(a)g(DMM)g(is)g(the)g(rel-)4
160 y(atively)i(high)f(cost)h(of)g(interprocessor)f(communication.)30
b(For)17 b(example,)h(the)f(latency)f(for)g(a)h(remote)4 215
y(memory)e(access)i(on)e(the)h(CM5)g(multiprocessor)f(is)h(approximately)e
(2560)h(processor)h(cycles)1699 196 y Fm(2)1718 215 y Fo(.)28
b(This)4 269 y(necessitates)16 b(the)e(selection)h(of)f(computation)f(and)h
(data)h(partitions)f(that)g(minimize)f(the)i(cost)f(of)g(com-)4
323 y(munication.)27 b(In)15 b(contrast,)h(on)f(SSMMs,)i(processors)f
(communicate)e(through)h(shared)g(memory)g(and)4 377 y(the)j(cost)h(of)f
(remote)f(memory)h(access)h(is)g(relatively)e(small.)36 b(For)17
b(example,)j(the)f(cost)f(of)g(a)g(remote)4 431 y(read)11 b(operation)g(on)g
(the)h(KSR1)f(is)h(approximately)e(170)h(processor)g(cycles)h([24].)17
b(Consequently)m(,)12 b(other)4 485 y(factors)h(come)f(into)h(play)g(in)f
(the)h(selection)g(of)f(computation)g(and)h(data)g(partitions.)19
b(In)13 b(this)g(section)g(we)4 540 y(elaborate)h(on)g(these)g(factors)g(and)
g(on)g(how)g(they)g(af)o(fect)g(performance,)f(and)h(consequently)m(,)h(af)o
(fect)f(the)4 594 y(choice)f(of)f(data)g(and)g(computation)g(partitions.)4
755 y Fc(3.1)58 b(Cache)14 b(Af\256nity)4 853 y Fo(Caches)j(are)e(used)h(in)f
(SSMMs)h(to)g(reduce)f(ef)o(fective)g(memory)f(access)j(time)e(and)h(reduce)f
(contention)4 907 y(in)e(the)h(interconnection)e(network.)21
b(Data)14 b(is)g(transferred)e(between)i(cache)g(and)f(memory)g(in)g(units)g
(of)h(a)4 961 y Fe(cache)g(line)p Fo(,)h(typically)e(a)h(multiple)f(of)g(the)
h(processor)g(word)f(size.)24 b Fe(Spatial)13 b(r)n(euse)i
Fo(occurs)e(when)h(other)4 1015 y(words)h(on)g(the)g(same)g(line)g(are)g
(used)g(by)g(the)g(processor)g(before)f(the)h(line)g(is)g(\257ushed)g(from)f
(the)h(cache.)4 1070 y(Analogously)m(,)g Fe(temporal)f(r)n(euse)i
Fo(occurs)e(when)g(data)h(on)f(a)g(cache)h(line)f(is)h(used)g(again)f(before)
g(the)g(line)4 1124 y(is)i(evicted)g(from)e(the)i(cache.)29
b(The)16 b(performance)e(of)i(an)f(application)h(depends)g(to)f(a)h(lar)o(ge)
f(extent)h(on)4 1178 y(the)g(ability)g(of)g(the)g(caches)h(to)f(exploit)g
(spatial)h(and)f(temporal)f(reuse.)31 b(In)16 b(some)g(cases,)j(this)d(may)h
(be)4 1232 y(dif)o(\256cult)9 b(because)i(of)f(the)g(limited)f(capacity)h
(and)g(associativity)g(of)g(caches.)18 b(Data)10 b(brought)f(into)h(a)g
(cache)4 1286 y(by)16 b(a)g(reference)f(or)h(a)g(prefetch)f(may)h(be)g
(evicted)f(before)h(being)f(used)h(or)g(reused,)h(because)g(of)e(either)4
1340 y(a)i(capacity)g(or)g(a)g(con\257ict)f(miss)i(caused)f(by)g(a)g
(subsequent)h(reference.)31 b(Cache)18 b(misses)f(on)g(SSMMs)4
1395 y(adversely)f(af)o(fect)g(performance,)g(since)h(evicted)f(data)g(must)g
(be)g(retrieved)f(from)g(its)i(home)e(memory)m(,)4 1449 y(which)k(may)g(be)g
(remote)f(to)h(the)f(processor)m(.)38 b(Caches)20 b(play)f(less)h(of)e(an)h
(important)f(role)g(in)h(DMMs)4 1503 y(because)g(cache)f(misses)h(result)e
(exclusively)h(in)f(local)h(memory)f(accesses,)k(which)d(are)g(inexpensive)4
1557 y(relative)12 b(to)g(interprocessor)g(communications.)4
1718 y Fc(3.2)58 b(False)14 b(Sharing)4 1816 y Fo(In)g(SSMMs)h(data)f(on)h
(the)f(same)h(cache)g(line)f(may)g(be)h(shared)f(by)h(more)e(than)i(one)f
(processor)n(,)h(and)g(the)4 1870 y(line)j(may)g(exit)g(in)g(more)g(than)g
(one)g(processor)r(')m(s)g(cache)h(at)f(the)g(same)h(time.)35
b(Hardware)18 b(is)g(used)h(to)4 1925 y(maintain)13 b(the)f(consistency)i(of)
e(the)h(multiple)g(copies)g(of)f(the)h(line,)h(typically)e(using)h(a)g
(write-invalidate)4 1979 y(protocol)e([24,)h(14].)18 b Fe(T)m(rue)12
b(sharing)g Fo(occurs)g(when)g(two)g(or)f(more)g(processors)i(access)g(the)f
(same)g(data)g(on)4 2033 y(a)k(cache)f(line,)i(and)e(it)g(re\257ects)g
(necessary)h(data)f(communications)g(in)g(an)g(application.)27
b(On)15 b(the)g(other)4 2087 y(hand,)h Fe(false)e(sharing)h
Fo(occurs)f(when)h(two)f(processors)h(access)h(dif)o(ferent)d(pieces)i(of)f
(data)h(on)f(the)g(same)4 2141 y(cache)e(line.)18 b(If)11 b(processors)h
(write)g(to)f(the)h(same)g(cache)g(line,)g(the)g(cache)g(consistency)h
(hardware)e(causes)4 2195 y(the)j(cache)g(line)g(to)g(be)g(transferred)f
(back)h(and)g(forth)f(between)h(processors)g(leading)g(to)g(a)g
(\252ping-pong\272)4 2250 y(ef)o(fect)h([8)o(].)27 b(False)16
b(sharing)f(causes)h(extensive)g(invalidation)e(traf)o(\256c)g(and)i(can)f
(considerably)g(degrade)4 2304 y(performance.)i(False)c(sharing)f(is)h
(non-existent)e(on)i(DMMs.)4 2465 y Fc(3.3)58 b(Memory)14 b(Contention)4
2563 y Fo(Memory)i(contention)g(occurs)g(when)g(many)g(processors)h(access)h
(data)e(in)g(a)g(single)h(memory)e(module)4 2617 y(at)j(the)g(same)h(time.)35
b(Since)18 b(the)g(communication)f(protocol)g(in)h(SSMMs)g(is)g(receiver)o
(-initiated,)h(and)4 2671 y(transfers)i(data)g(in)f(units)h(of)g(relatively)f
(small)h(cache)g(lines,)j(a)d(lar)o(ge)g(number)f(of)h(requests)g(to)g(the)4
2725 y(same)12 b(memory)f(can)h(over\257ow)f(memory)g(buf)o(fers)g(and)h
(cause)g(excessive)h(delays)f(in)g(memory)e(response)4 2780
y(time)20 b([13].)42 b(Contention)20 b(has)h(been)g(considered)g(less)g(of)f
(a)h(performance)e(bottleneck)h(on)h(DMMs)p 4 2825 737 2 v
62 2855 a Fl(2)79 2870 y Fk(Calculated)10 b(based)h(on)f(the)g(elapsed)h
(time)f(for)g(a)g(send-reply)g(message)i(of)e(128)g(bytes)g([19)o(].)p
eop
%%Page: 4 4
4 3 bop 4 -27 a Fo(because)16 b(a)g(sender)o(-initiated)e(communication)h
(protocol)f(is)i(employed,)g(and)g(because)g(programmers)4
27 y(typically)f(communicate)f(data)i(in)f(lar)o(ge)g(infrequent)f(messages.)
28 b(Applications)15 b(on)g(DMMs)h(also)f(use)4 82 y(collective)d
(communications)g([15)o(])g(that)h(further)e(reduce)h(contention.)4
243 y Fc(3.4)58 b(Over)o(head)14 b(of)g(Parallelism)4 341 y
Fo(In)g(DMM,)i(synchronization)e(is)h(achieved)f(through)g(data)g
(communication.)24 b(However)n(,)15 b(on)g(SSMMs,)4 395 y(synchronization)9
b(is)h(explicit)e(and)i(is)g(independent)f(of)f(data)i(communication.)16
b(The)10 b(resulting)f(overhead)4 449 y(can)14 b(become)f(a)h(performance)e
(bottleneck)h([27)o(],)h(and)f(must)h(be)f(minimized.)21 b(The)14
b(performance)e(of)h(an)4 503 y(application)e(is)h(also)h(af)o(fected)e(by)h
(the)f(overhead)h(involved)f(in)h(parallelizing)e(loops,)j(manifested)e(in)h
(the)4 557 y(form)h(of)h(computation)f(partitioning)f(tests)j([25)o(].)23
b(These)15 b(tests)g(can)f(be)g(eliminated)g(in)f(some)i(cases)g(by)4
612 y(compiler)g(analysis,)i(but)d(when)i(not)f(possible,)h(can)g(degrade)f
(performance.)26 b(This)15 b(overhead)g(though)4 666 y(also)d(present)g(in)f
(the)h(case)g(of)f(DMMs,)j(is)e(not)f(considered)h(signi\256cant)f(because)h
(of)g(the)f(predominantly)4 720 y(high)h(cost)h(of)f(remote)g(memory)f
(access.)4 902 y Fn(4)71 b(Impact)19 b(on)f(Data)h(Distribution)4
1029 y Fo(In)e(this)g(section)g(we)g(use)g(two)g(applications,)h
Fd(Multigrid)e Fo(and)h Fd(Tred2)p Fo(,)h(to)f(illustrate)f(the)h(impact)4
1083 y(of)f(cache)h(af)o(\256nity)f(and)h(false)g(sharing)f(on)h(the)f
(choice)h(of)f(a)h(data)g(distribution.)30 b(The)17 b(KSR1)f(system)4
1137 y(is)f(used)f(because)h(of)f(its)h(lar)o(ge)f(cache)g(size,)i(and)e
(because)h(of)f(the)g(presence)h(of)f(monitoring)e(hardware)4
1191 y(that)i(enables)h(the)g(measurement)f(of)g(the)g(number)g(of)f
(non-local)h(memory)g(accesses)i(and)e(the)g(number)4 1245
y(of)e(caches)h(misses)h(for)d(a)i(processor)m(.)77 1321 y(The)j(KSR1)e(is)h
(a)g(Cache)g(only)g(Memory)f(Architecture)g(\(COMA\))g(con\256gured)g(as)h(a)
g(hierarchy)f(of)4 1375 y(slotted)c(rings)g(with)g(processing)g(cells)h(on)f
(the)g(leaf-level)f(rings.)18 b(The)10 b(local)g(portion)g(of)f(shared)i
(memory)4 1429 y(associated)g(with)e(a)i(processor)e(is)i(or)o(ganized)e(as)i
(a)f(cache.)18 b(There)10 b(is)g(no)g(home)g(location)f(for)g(data,)i(rather)
n(,)4 1483 y(data)k(may)g(exist)f(in)h(more)f(than)h(one)f(local)h(memory)m
(.)24 b(The)16 b(hardware)e(maintains)g(the)h(consistency)g(of)4
1538 y(possible)e(multiple)e(copies)i(of)f(the)g(data.)77 1613
y(The)e(KSR1)g(implicitly)e(implements)i(the)f(owner)o(-computes)g(rule,)h
(since)g(data)g(written)f(by)g(a)h(proces-)4 1667 y(sor)j(must)f(exclusively)
g(reside)h(in)f(the)h(processor)r(')m(s)f(local)g(portion)g(of)g(the)g
(shared)h(memory)m(.)k(Hardware)4 1722 y(automatically)j(migrates)g(data)h
(to)g(the)f(processor)h(that)f(requests)h(the)g(data)f(in)h(units)g(of)f
Fe(subpages)p Fo(.)4 1776 y(Hence,)13 b(the)f(computation)g(partitioning)e
(of)i(a)g(loop)g(dictates)h(the)f(residence)g(of)g(a)g(data)h(item)e(and)i
(hence)4 1830 y(the)k(distribution)f(of)h(the)g(arrays)g(in)g(the)g(loop.)33
b(Data)17 b(which)g(is)h(read)f(by)g(the)g(processors)h(may)f(exist)4
1884 y(in)e(multiple)e(local)i(memories,)g(and)g(read)f(requests)h(to)g(this)
g(data)f(from)g(dif)o(ferent)f(processors)i(may)g(be)4 1938
y(satis\256ed)e(from)e(dif)o(ferent)g(portions)h(of)g(the)g(shared)h(memory)m
(.)4 2099 y Fc(4.1)58 b(Cache-Conscious)13 b(Data)i(Distribution)4
2197 y Fo(The)j Fd(Multigrid)e Fo(application)g(from)g(the)h(NAS)f(suite)h
(of)g(benchmarks)f(illustrates)h(how)g(data)g(dis-)4 2252 y(tributions)d
(must)h(be)g(cache-conscious.)27 b Fd(Multigrid)14 b Fo(is)h(a)g(three)g
(dimensional)f(solver)h(calculating)4 2306 y(the)j(potential)f(\256eld)h(on)f
(a)h(cubical)g(grid.)34 b(W)l(e)18 b(focus)g(on)f(the)h(subroutine)f
Fd(psinv)h Fo(which)f(uses)i(two)4 2360 y(3-dimensional)13
b(arrays)h Fj(U)20 b Fo(and)14 b Fj(R)p Fo(.)25 b(The)14 b(subroutine)g
(mainly)f(performs)h(the)g(following)f(computation)4 2414 y(inside)i(a)h
(triply)e(nested)h(loop:)23 b Fj(U)5 b Fi(\()p Fj(i;)23 b(j;)h(k)r
Fi(\))15 b(+)j(=)33 b Fj(\013)p Fi(\()15 b Fj(R)p Fi(\()p Fj(f)5
b Fi(\()p Fj(i)p Fi(\))p Fj(;)24 b(g)r Fi(\()p Fj(j)s Fi(\))p
Fj(;)f(h)p Fi(\()p Fj(k)r Fi(\)\)\))p Fo(,)16 b(where)f Fj(f)5
b Fi(\()p Fj(i)p Fi(\))15 b Fo(=)h Fj(i)c Fh(\000)g Fo(1,)4
2468 y Fj(i)18 b Fo(or)g Fj(i)13 b Fi(+)i Fo(1,)20 b(as)e(are)g(the)g
(functions)g Fj(g)i Fo(and)e Fj(h)p Fo(.)36 b(The)18 b(loop)g(nest)g(is)h
(fully)e(parallel.)35 b(The)18 b(application)4 2522 y(has)e(nearest)g
(neighbor)e(communications)h(along)g(all)g(three)g(dimensions,)i(which)e(is)h
(typical)f(of)g(many)4 2577 y(scienti\256c)d(applications.)77
2652 y(In)d(this)g(application,)g(we)g(choose)h(not)e(to)h(parallelize)f(the)
h(innermost)g(loop)f(to)h(avoid)g(cache)g(line)g(false)4 2706
y(sharing)k(and)g(cache)h(interference;)e(successive)j(iterations)e(of)f
(this)i(loop)f(access)h(successive)h(elements)4 2761 y(on)h(the)f(same)i
(cache)f(line.)28 b(Hence)16 b(we)g(use)g(a)g(two)g(dimensional)f(grid)g(for)
g(the)h(processor)f(geometry)m(.)4 2815 y(Since)10 b(the)g(application)g(has)
h(nearest)f(neighbor)f(communications,)i Fd(Block)f Fo(distribution)f
(performs)g(the)4 2869 y(best.)18 b(The)10 b(restriction)e(of)h(the)h
(innermost)f(loop)g(to)g(be)h(sequential)f(requires)g(the)g(arrays)h(to)f(be)
h(distributed)p eop
%%Page: 5 5
5 4 bop 503 532 a @beginspecial 50 @llx 50 @lly 410 @urx 302
@ury 2057 @rwi @setspecial
%%BeginDocument: mg.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.100 0.100 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 473 M
2817 0 V
LTb
600 473 M
63 0 V
2754 0 R
-63 0 V
540 473 M
(96) Rshow
LTa
600 916 M
2817 0 V
LTb
600 916 M
63 0 V
2754 0 R
-63 0 V
540 916 M
(98) Rshow
LTa
600 1360 M
2817 0 V
LTb
600 1360 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(100) Rshow
LTa
600 1804 M
2817 0 V
LTb
600 1804 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(102) Rshow
LTa
600 2247 M
2817 0 V
LTb
600 2247 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(104) Rshow
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(\(16,1\)) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(8,2\)) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(4,4\)) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(2,8\)) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(1,16\)) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
340 1360 M
currentpoint gsave translate 90 rotate 0 0 M
(Normalized Execution Time \(w.r.t \(16,1\)\)) Cshow
grestore
2008 51 M
(Processor Geometry - 16 Processors                    ) Cshow
LT1
1654 2106 M
(160x160x160) Rshow
1714 2106 M
180 0 V
1523 341 R
2713 2048 L
2009 1826 L
1304 939 L
600 1360 L
1774 2106 A
3417 2447 A
2713 2048 A
2009 1826 A
1304 939 A
600 1360 A
LT2
1654 2006 M
(144x144x144) Rshow
1714 2006 M
180 0 V
1523 241 R
2713 1715 L
-705 -67 V
1304 850 L
600 1360 L
1774 2006 B
3417 2247 B
2713 1715 B
2009 1648 B
1304 850 B
600 1360 B
LT4
1654 1906 M
(64x64x64) Rshow
1714 1906 M
180 0 V
1523 -590 R
2713 1160 L
2009 340 L
1304 495 L
600 1360 L
1774 1906 T
3417 1316 T
2713 1160 T
2009 340 T
1304 495 T
600 1360 T
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 377 612 a Fo(Figure)11 b(2:)18 b(Normalized)11
b(Execution)i(time)f(of)g Fd(Multigrid)p Fo(.)4 733 y(with)17
b Fd(\(*,Block,Block\))f Fo(since)h(the)g(arrays)g(are)g(assumed)h(to)f(be)h
(stored)f(using)g(column)f(major)4 787 y(ordering.)31 b(W)n(ith)16
b(16)h(processors,)h(it)f(is)g(possible)g(to)g(choose)g(one)g(of)f(the)h
(\(16,1\),)h(\(8,2\),)f(\(4,4\),)h(\(2,8\))4 841 y(and)d(\(1,16\))g
(processor)g(geometries.)27 b(The)15 b(choice)h(of)e(the)i(processor)f
(geometry)f(af)o(fects)h(the)g(number)4 895 y(of)h(processors)g(that)g
(execute)g(each)g(parallel)f(loop.)29 b(For)15 b(example,)i(a)f(processor)g
(geometry)f(of)h(\(8,2\),)4 949 y(implies)11 b(8)g(processors)h(assigned)g
(to)e(the)i(inner)e(parallel)h(loop)g(and)g(2)g(processors)g(assigned)h(to)f
(the)g(outer)4 1004 y(parallel)h(loop.)77 1079 y(Figure)17
b(2)g(shows)h(the)f(execution)g(time)g(of)g(the)h(application)e(for)h
(various)g(processor)g(geometries)4 1133 y(with)e(the)g Fd(\(*,Block,Block\))
e Fo(distribution)h(for)g(the)h(arrays)g(on)g(the)g(KSR1)f(with)h(16)g
(processors,)4 1188 y(normalized)d(with)h(respect)g(to)g(the)g(\(16,1\))f
(processor)h(geometry)m(.)19 b(For)12 b(a)h(small)g(data)g(size)h
(\(64x64x64\),)4 1242 y(execution)22 b(time)f(is)h(minimized)e(by)i(a)g
(distribution)e(with)h(equal)h(number)f(of)g(processors)h(in)f(each)4
1296 y(dimension,)15 b(i.e.,)i(\(4,4\).)24 b(This)16 b(is)f(the)f(same)h
(distribution)e(scheme)j(suggested)f(in)f(the)h(Syracuse)f(High)4
1350 y(Performance)9 b(Fortran)h(applications)g(suite)772 1332
y Fm(3)801 1350 y Fo(for)g(DMMs.)19 b(However)n(,)11 b(when)f(the)h(data)g
(size)g(is)g(lar)o(ge,)g(the)4 1404 y(processor)h(geometry)f(\(4,4\))h(no)g
(longer)f(performs)g(the)h(best.)19 b(The)12 b(execution)g(time)g(is)g
(minimized)f(with)4 1458 y(a)i(processor)f(geometry)g(of)g(\(8,2\).)77
1534 y(The)20 b(impact)e(of)h(processor)g(geometry)f(on)g(performance)g(is)h
(due)g(to)g(cache)g(af)o(\256nity)m(,)h(as)g(can)f(be)4 1588
y(deduced)12 b(from)f(Figures)h(3)g(and)g(4.)19 b(Figure)11
b(3)h(shows)h(the)f(measured)g(number)g(of)f(cache)i(lines)f(accessed)4
1642 y(from)17 b(remote)g(memory)f(modules,)j(normalized)e(with)g(respect)h
(to)f(the)h(processor)f(geometry)g(\(16,1\).)4 1697 y(The)h(number)e(of)h
(remote)f(memory)g(accesses)j(is)e(minimal)g(when)g(the)g(processor)g
(geometry)f(is)h(\(4,4\))4 1751 y(for)h(all)g(data)h(sizes.)38
b(Figure)17 b(4)i(shows)g(the)g(average)f(measured)h(number)e(of)i(cache)g
(misses)g(from)f(a)4 1805 y(processor)c(cache,)h(again)e(normalized)g(with)g
(respect)h(to)g(the)f(processor)h(geometry)f(\(16,1\).)21 b(When)14
b(the)4 1859 y(data)e(size)h(is)f(small)g(\(64x64x64\),)g(the)g(data)g(used)g
(by)g(a)h(processor)f(\256ts)g(into)f(the)h(256k)g(processor)g(cache)4
1913 y(and)19 b(the)g(misses)h(from)e(the)h(cache)h(in)f(this)g(case)h
(re\257ect)f(remote)f(memory)g(accesses)j(that)e(occur)g(in)4
1967 y(the)13 b(parallel)g(program.)19 b(Hence,)14 b(the)f(predominant)f
(factor)h(af)o(fecting)f(performance)g(is)h(interprocessor)4
2022 y(communication,)f(and)g(the)h(best)f(performance)g(is)g(attained)g
(using)h(the)f(\(4,4\))g(geometry)m(.)77 2097 y(However)n(,)17
b(when)f(the)g(arrays)g(are)g(relatively)f(lar)o(ge)h(\(144x144x144\),)g(the)
g(cache)g(capacity)h(is)f(no)4 2151 y(longer)g(suf)o(\256cient)h(to)g(hold)f
(data)i(from)d(successive)k(iterations)d(of)h(the)g(outer)f(parallel)h(loop,)
h(and)f(the)4 2206 y(number)10 b(of)h(cache)g(misses)h(increases.)19
b(When)11 b(the)g(number)f(of)h(processors)g(assigned)h(to)f(the)g(outer)f
(loop)4 2260 y(increases,)j(the)f(number)f(of)h(misses)h(from)d(the)i(cache)h
(also)f(increases.)19 b(The)12 b(\(4,4\))g(processor)g(geometry)4
2314 y(minimizes)d(the)f(amount)h(of)f(remote)g(memory)g(access,)k(but)c(the)
h(\(16,1\))f(processor)h(geometry)f(minimizes)4 2368 y(the)k(amount)f(of)g
(cache)h(misses.)19 b(The)12 b(distribution)e(with)i(\(8,2\))f(processor)g
(geometry)g(strikes)h(a)g(balance)4 2422 y(between)17 b(the)g(cost)g(of)g
(remote)f(memory)g(access)i(and)f(the)g(cost)g(of)g(cache)g(misses,)i
(resulting)e(in)f(best)4 2476 y(overall)c(performance,)g(in)g(spite)g(of)g
(higher)g(interprocessor)g(communication)f(cost.)4 2638 y Fc(4.2)58
b(False)14 b(Sharing)g(Conscious)g(Data)h(Distribution)4 2736
y Fo(The)d(programs)f Fd(Tred2)h Fo(\(which)f(is)h(part)f(of)g(Eispack\),)i
Fd(mdg)p Fo(,)f(and)g Fd(trfd)f Fo(\(which)g(are)h(both)f(part)h(of)f(the)4
2790 y(Perfect)f(Club)h(Benchmark)f(Suite\))g(exhibit)h(parallelism)f(which)h
(result)f(in)h(considerable)g(false)g(sharing.)p 4 2835 737
2 v 62 2865 a Fl(3)79 2880 y Fk(http://www)m(.npac.syr)n(.edu/hpfa/)c(.)p
eop
%%Page: 6 6
6 5 bop 47 586 a @beginspecial 50 @llx 50 @lly 230 @urx 176
@ury 2057 @rwi @setspecial
%%BeginDocument: spmiss.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.050 0.050 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
540 251 M
(30) Rshow
LTa
600 568 M
2817 0 V
LTb
600 568 M
63 0 V
2754 0 R
-63 0 V
540 568 M
(40) Rshow
LTa
600 885 M
2817 0 V
LTb
600 885 M
63 0 V
2754 0 R
-63 0 V
540 885 M
(50) Rshow
LTa
600 1202 M
2817 0 V
LTb
600 1202 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(60) Rshow
LTa
600 1518 M
2817 0 V
LTb
600 1518 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(70) Rshow
LTa
600 1835 M
2817 0 V
LTb
600 1835 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(80) Rshow
LTa
600 2152 M
2817 0 V
LTb
600 2152 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(90) Rshow
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(100) Rshow
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(\(16,1\)) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(8,2\)) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(4,4\)) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(2,8\)) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(1,16\)) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
340 1360 M
currentpoint gsave translate 90 rotate 0 0 M
(Normalized subpage misses \(w.r.t. \(16,1\)\)) Cshow
grestore
2008 51 M
(Processor Geometry - 16 Processors            ) Cshow
LT1
1654 2106 M
(160x160x160) Rshow
1714 2106 M
180 0 V
600 2469 M
1304 1154 L
2009 495 L
705 646 V
3417 2453 L
1774 2106 A
600 2469 A
1304 1154 A
2009 495 A
2713 1141 A
3417 2453 A
LT2
1654 2006 M
(144x144x144) Rshow
1714 2006 M
180 0 V
600 2469 M
1304 1072 L
2009 473 L
705 567 V
3417 2387 L
1774 2006 B
600 2469 B
1304 1072 B
2009 473 B
2713 1040 B
3417 2387 B
LT4
1654 1906 M
(64x64x64) Rshow
1714 1906 M
180 0 V
600 2469 M
1304 1113 L
2009 543 L
705 649 V
3417 2444 L
1774 1906 T
600 2469 T
1304 1113 T
2009 543 T
2713 1192 T
3417 2444 T
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 124 640 a Fo(Figure)12 b(3.)18 b(Remote)12 b(Memory)g(Access.)
899 586 y @beginspecial 50 @llx 50 @lly 230 @urx 176 @ury 2057
@rwi @setspecial
%%BeginDocument: datac.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.050 0.050 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
540 251 M
(90) Rshow
LTa
600 528 M
2817 0 V
LTb
600 528 M
63 0 V
2754 0 R
-63 0 V
540 528 M
(100) Rshow
LTa
600 806 M
2817 0 V
LTb
600 806 M
63 0 V
2754 0 R
-63 0 V
540 806 M
(110) Rshow
LTa
600 1083 M
2817 0 V
LTb
600 1083 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(120) Rshow
LTa
600 1360 M
2817 0 V
LTb
600 1360 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(130) Rshow
LTa
600 1637 M
2817 0 V
LTb
600 1637 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(140) Rshow
LTa
600 1915 M
2817 0 V
LTb
600 1915 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(150) Rshow
LTa
600 2192 M
2817 0 V
LTb
600 2192 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(160) Rshow
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(170) Rshow
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(\(16,1\)) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(8,2\)) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(4,4\)) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(2,8\)) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(1,16\)) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
340 1360 M
currentpoint gsave translate 90 rotate 0 0 M
(Normalized cache misses \(w.r.t \(16,1\)\)) Cshow
grestore
2008 51 M
(Processor Geometry - 16 Processors         ) Cshow
LT1
1654 2106 M
(160x160x160) Rshow
1714 2106 M
180 0 V
600 528 M
1304 817 L
705 626 V
705 721 V
704 250 V
1774 2106 A
600 528 A
1304 817 A
2009 1443 A
2713 2164 A
3417 2414 A
LT2
1654 2006 M
(144x144x144) Rshow
1714 2006 M
180 0 V
600 528 M
1304 678 L
705 488 V
705 610 V
704 333 V
1774 2006 B
600 528 B
1304 678 B
2009 1166 B
2713 1776 B
3417 2109 B
LT4
1654 1906 M
(64x64x64) Rshow
1714 1906 M
180 0 V
600 528 M
1304 329 L
705 -36 V
705 133 V
704 97 V
1774 1906 T
600 528 T
1304 329 T
2009 293 T
2713 426 T
3417 523 T
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 1085 640 a(Figure)g(4.)18 b(Cache)13 b(Misses.)47
1297 y @beginspecial 50 @llx 50 @lly 410 @urx 302 @ury 2057
@rwi @setspecial
%%BeginDocument: tred2.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.100 0.100 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
540 251 M
(1e+06) Rshow
LTb
600 2469 M
63 0 V
2754 0 R
-63 0 V
540 2469 M
(4e+06) Rshow
LTa
600 1360 M
2817 0 V
LTb
600 1360 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2009 M
2817 0 V
LTb
600 2009 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
31 0 V
2786 0 R
-31 0 V
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(0) Cshow
LTa
952 251 M
0 2218 V
LTb
952 251 M
0 63 V
0 2155 R
0 -63 V
952 151 M
(2) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(4) Cshow
LTa
1656 251 M
0 2218 V
LTb
1656 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(6) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(8) Cshow
LTa
2361 251 M
0 2218 V
LTb
2361 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(10) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(12) Cshow
LTa
3065 251 M
0 2218 V
LTb
3065 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(14) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(16) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
340 1260 M
currentpoint gsave translate 90 rotate 0 0 M
(Execution Time \(Micro Seconds\)) Cshow
grestore
2008 51 M
(Number of Processors) Cshow
LT0
3054 2306 M
("Cyclic") Rshow
3114 2306 M
180 0 V
776 910 M
176 318 V
352 533 V
705 -387 V
704 -149 V
704 707 V
3174 2306 D
776 910 D
952 1228 D
1304 1761 D
2009 1374 D
2713 1225 D
3417 1932 D
LT1
3054 2206 M
("BlockCyclic") Rshow
3114 2206 M
180 0 V
776 942 M
952 817 L
1304 662 L
705 -34 V
704 110 V
704 1480 V
3174 2206 A
776 942 A
952 817 A
1304 662 A
2009 628 A
2713 738 A
3417 2218 A
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 140 1351 a(Figure)f(5.)18 b(Ef)o(fect)12 b(of)g(False)h
(Sharing.)899 1297 y @beginspecial 50 @llx 50 @lly 410 @urx
302 @ury 2057 @rwi @setspecial
%%BeginDocument: tred2c.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.100 0.100 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(40000) Rshow
LTa
600 251 M
2817 0 V
LTb
600 251 M
31 0 V
2786 0 R
-31 0 V
LTa
600 497 M
2817 0 V
LTb
600 497 M
31 0 V
2786 0 R
-31 0 V
LTa
600 697 M
2817 0 V
LTb
600 697 M
31 0 V
2786 0 R
-31 0 V
LTa
600 867 M
2817 0 V
LTb
600 867 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1014 M
2817 0 V
LTb
600 1014 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1144 M
2817 0 V
LTb
600 1144 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1260 M
2817 0 V
LTb
600 1260 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(100000) Rshow
LTa
600 2023 M
2817 0 V
LTb
600 2023 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
31 0 V
2786 0 R
-31 0 V
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(0) Cshow
LTa
952 251 M
0 2218 V
LTb
952 251 M
0 63 V
0 2155 R
0 -63 V
952 151 M
(2) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(4) Cshow
LTa
1656 251 M
0 2218 V
LTb
1656 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(6) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(8) Cshow
LTa
2361 251 M
0 2218 V
LTb
2361 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(10) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(12) Cshow
LTa
3065 251 M
0 2218 V
LTb
3065 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(14) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(16) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
400 1660 M
currentpoint gsave translate 90 rotate 0 0 M
(Cache Misses) Cshow
grestore
2008 51 M
(Number of Processors) Cshow
LT0
3054 2306 M
("Cyclic") Rshow
3114 2306 M
180 0 V
123 -856 R
-704 -3 V
-704 172 V
-705 449 V
952 1885 L
776 2180 L
3174 2306 D
3417 1450 D
2713 1447 D
2009 1619 D
1304 2068 D
952 1885 D
776 2180 D
LT1
3054 2206 M
("BlockCyclic") Rshow
3114 2206 M
180 0 V
3417 1029 M
2713 642 L
2009 475 L
-705 601 V
952 1596 L
776 2166 L
3174 2206 A
3417 1029 A
2713 642 A
2009 475 A
1304 1076 A
952 1596 A
776 2166 A
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 1085 1351 a(Figure)f(6.)18 b(Cache)13 b(Misses.)4
1510 y(These)18 b(programs)f(have)g(triangular)f(iteration)g(spaces)i(which)f
(necessitate)h(cyclical)f(distribution)f(for)4 1564 y(load)c(balancing.)19
b(The)13 b(choice)f(of)g(this)h(distribution)e(combined)h(with)g(the)h
(storage)f(order)g(of)g(the)g(arrays)4 1619 y(cause)17 b(more)f(than)g(one)g
(processor)g(to)g(share)g(the)g(same)h(cache)f(line,)i(leading)e(to)g(false)g
(sharing.)29 b(The)4 1673 y(impact)17 b(of)f(this)h(false)h(sharing)e(is)i
(shown)f(in)g(Figure)f(5)h(for)f(the)h Fd(Tred2)g Fo(application)f(on)h(the)g
(KSR1)4 1727 y(multiprocessor)m(.)37 b(The)19 b(\256gure)f(shows)h(the)g
(execution)f(time)g(of)g(the)h(application)f(for)g Fd(Cyclic)g
Fo(and)4 1781 y Fd(BlockCyclic)12 b Fo(distributions)g(using)i(1)f(to)g(16)g
(processors.)20 b(The)14 b(use)g(of)e(the)h Fd(Cyclic)g Fo(distribution)4
1835 y(results)f(in)g(a)h(lar)o(ge)f(number)f(of)h(cache)h(misses,)g(as)g
(can)f(be)g(seen)h(in)f(Figure)g(6.)18 b(The)13 b(resulting)e(overhead)4
1889 y(causes)20 b(execution)f(time)g(to)g(increase)g(as)g(the)g(number)g(of)
f(processors)i(increases.)39 b(The)19 b(arrays)g(are)4 1944
y(distributed)c(using)g(a)g Fd(BlockCyclic)f Fo(distribution,)h(where)g(the)g
(size)h(of)f(the)g(block)g(is)h(equal)f(to)g(the)4 1998 y(size)22
b(of)e(the)h(cache)g(line,)i(which)e(ef)o(fectively)f(eliminates)h(false)g
(sharing.)43 b(When)21 b(the)g(number)f(of)4 2052 y(processors)14
b(is)g(small,)g(the)f(load)h(is)g(relatively)e(well-balanced,)i(and)f(the)h
(elimination)e(of)h(false)h(sharing)4 2106 y(improves)h(performance.)25
b(However)n(,)15 b(as)h(the)f(number)f(of)h(processors)g(increases,)i(the)e
(load)f(becomes)4 2160 y(increasingly)f(imbalanced,)h(and)f(the)h(negative)f
(impact)g(of)g(this)g(load)g(imbalance)h(begins)f(to)g(outweigh)4
2214 y(the)h(bene\256ts)h(of)e(eliminating)h(false)g(sharing.)24
b(A)14 b(compiler)g(for)f(SSMM)h(must)g(consider)h(this)f(tradeof)o(f)4
2269 y(between)f(load)f(imbalance)g(and)g(false)h(sharing)f(when)g
(determining)g(data)g(distributions.)4 2450 y Fn(5)71 b(Impact)19
b(on)f(Computation)i(Partitioning)4 2577 y Fo(The)12 b(owner)o(-computes)f
(rule)g(has)h(been)f(the)h(computation)f(partitioner)f(of)h(choice)g(for)g
(compiling)g(HPF-)4 2631 y(type)17 b(languages)g(on)g(DMMs)h([16].)32
b(The)17 b(owner)o(-computes)f(rule)h(maps)g(a)g(statement)h(such)f(that)g
(the)4 2686 y(the)h(computation)e(is)i(executed)g(on)g(the)f(processor)h(on)f
(which)h(the)f(data)h(element)f(that)h(is)g(written)e(is)4
2740 y(local.)27 b(All)15 b(the)g(data)g(elements)g(that)g(are)g(required)f
(to)h(compute)g(the)g(result)g(\(which)g(may)g(be)g(remote\))4
2794 y(are)h(communicated)f(to)h(the)g(processor)m(.)29 b(A)16
b(strict)g(rule)f(such)i(as)f(owner)o(-computes)f(is)h(not)g(necessary)4
2848 y(on)h(a)f(SSMM)h(because)g(message)h(passing)f(code)g(is)g(not)f
(generated)g(at)h(compile)f(time)g([3].)30 b(In)17 b(some)p
eop
%%Page: 7 7
7 6 bop 482 311 a @beginspecial 127 @llx 520 @lly 393 @urx
632 @ury 2160 @rwi @setspecial
%%BeginDocument: adi.idraw

/arrowhead {
0 begin
transform originalCTM itransform
/taily exch def
/tailx exch def
transform originalCTM itransform
/tipy exch def
/tipx exch def
/dy tipy taily sub def
/dx tipx tailx sub def
/angle dx 0 ne dy 0 ne or { dy dx atan } { 90 } ifelse def
gsave
originalCTM setmatrix
tipx tipy translate
angle rotate
newpath
arrowHeight neg arrowWidth 2 div moveto
0 0 lineto
arrowHeight neg arrowWidth 2 div neg lineto
patternNone not {
originalCTM setmatrix
/padtip arrowHeight 2 exp 0.25 arrowWidth 2 exp mul add sqrt brushWidth mul
arrowWidth div def
/padtail brushWidth 2 div def
tipx tipy translate
angle rotate
padtip 0 translate
arrowHeight padtip add padtail add arrowHeight div dup scale
arrowheadpath
ifill
} if
brushNone not {
originalCTM setmatrix
tipx tipy translate
angle rotate
arrowheadpath
istroke
} if
grestore
end
} dup 0 9 dict put def

/arrowheadpath {
newpath
arrowHeight neg arrowWidth 2 div moveto
0 0 lineto
arrowHeight neg arrowWidth 2 div neg lineto
} def

/leftarrow {
0 begin
y exch get /taily exch def
x exch get /tailx exch def
y exch get /tipy exch def
x exch get /tipx exch def
brushLeftArrow { tipx tipy tailx taily arrowhead } if
end
} dup 0 4 dict put def

/rightarrow {
0 begin
y exch get /tipy exch def
x exch get /tipx exch def
y exch get /taily exch def
x exch get /tailx exch def
brushRightArrow { tipx tipy tailx taily arrowhead } if
end
} dup 0 4 dict put def


/arrowHeight 10 def
/arrowWidth 5 def

/IdrawDict 51 dict def
IdrawDict begin

/reencodeISO {
dup dup findfont dup length dict begin
{ 1 index /FID ne { def }{ pop pop } ifelse } forall
/Encoding ISOLatin1Encoding def
currentdict end definefont
} def

/ISOLatin1Encoding [
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/space/exclam/quotedbl/numbersign/dollar/percent/ampersand/quoteright
/parenleft/parenright/asterisk/plus/comma/minus/period/slash
/zero/one/two/three/four/five/six/seven/eight/nine/colon/semicolon
/less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N
/O/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright
/asciicircum/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m
/n/o/p/q/r/s/t/u/v/w/x/y/z/braceleft/bar/braceright/asciitilde
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/dotlessi/grave/acute/circumflex/tilde/macron/breve
/dotaccent/dieresis/.notdef/ring/cedilla/.notdef/hungarumlaut
/ogonek/caron/space/exclamdown/cent/sterling/currency/yen/brokenbar
/section/dieresis/copyright/ordfeminine/guillemotleft/logicalnot
/hyphen/registered/macron/degree/plusminus/twosuperior/threesuperior
/acute/mu/paragraph/periodcentered/cedilla/onesuperior/ordmasculine
/guillemotright/onequarter/onehalf/threequarters/questiondown
/Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE/Ccedilla
/Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex
/Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis
/multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute
/Thorn/germandbls/agrave/aacute/acircumflex/atilde/adieresis
/aring/ae/ccedilla/egrave/eacute/ecircumflex/edieresis/igrave
/iacute/icircumflex/idieresis/eth/ntilde/ograve/oacute/ocircumflex
/otilde/odieresis/divide/oslash/ugrave/uacute/ucircumflex/udieresis
/yacute/thorn/ydieresis
] def
/Helvetica reencodeISO def

/none null def
/numGraphicParameters 17 def
/stringLimit 65535 def

/Begin {
save
numGraphicParameters dict begin
} def

/End {
end
restore
} def

/SetB {
dup type /nulltype eq {
pop
false /brushRightArrow idef
false /brushLeftArrow idef
true /brushNone idef
} {
/brushDashOffset idef
/brushDashArray idef
0 ne /brushRightArrow idef
0 ne /brushLeftArrow idef
/brushWidth idef
false /brushNone idef
} ifelse
} def

/SetCFg {
/fgblue idef
/fggreen idef
/fgred idef
} def

/SetCBg {
/bgblue idef
/bggreen idef
/bgred idef
} def

/SetF {
/printSize idef
/printFont idef
} def

/SetP {
dup type /nulltype eq {
pop true /patternNone idef
} {
dup -1 eq {
/patternGrayLevel idef
/patternString idef
} {
/patternGrayLevel idef
} ifelse
false /patternNone idef
} ifelse
} def

/BSpl {
0 begin
storexyn
newpath
n 1 gt {
0 0 0 0 0 0 1 1 true subspline
n 2 gt {
0 0 0 0 1 1 2 2 false subspline
1 1 n 3 sub {
/i exch def
i 1 sub dup i dup i 1 add dup i 2 add dup false subspline
} for
n 3 sub dup n 2 sub dup n 1 sub dup 2 copy false subspline
} if
n 2 sub dup n 1 sub dup 2 copy 2 copy false subspline
patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if
brushNone not { istroke } if
0 0 1 1 leftarrow
n 2 sub dup n 1 sub dup rightarrow
} if
end
} dup 0 4 dict put def

/Circ {
newpath
0 360 arc
patternNone not { ifill } if
brushNone not { istroke } if
} def

/CBSpl {
0 begin
dup 2 gt {
storexyn
newpath
n 1 sub dup 0 0 1 1 2 2 true subspline
1 1 n 3 sub {
/i exch def
i 1 sub dup i dup i 1 add dup i 2 add dup false subspline
} for
n 3 sub dup n 2 sub dup n 1 sub dup 0 0 false subspline
n 2 sub dup n 1 sub dup 0 0 1 1 false subspline
patternNone not { ifill } if
brushNone not { istroke } if
} {
Poly
} ifelse
end
} dup 0 4 dict put def

/Elli {
0 begin
newpath
4 2 roll
translate
scale
0 0 1 0 360 arc
patternNone not { ifill } if
brushNone not { istroke } if
end
} dup 0 1 dict put def

/Line {
0 begin
2 storexyn
newpath
x 0 get y 0 get moveto
x 1 get y 1 get lineto
brushNone not { istroke } if
0 0 1 1 leftarrow
0 0 1 1 rightarrow
end
} dup 0 4 dict put def

/MLine {
0 begin
storexyn
newpath
n 1 gt {
x 0 get y 0 get moveto
1 1 n 1 sub {
/i exch def
x i get y i get lineto
} for
patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if
brushNone not { istroke } if
0 0 1 1 leftarrow
n 2 sub dup n 1 sub dup rightarrow
} if
end
} dup 0 4 dict put def

/Poly {
3 1 roll
newpath
moveto
-1 add
{ lineto } repeat
closepath
patternNone not { ifill } if
brushNone not { istroke } if
} def

/Rect {
0 begin
/t exch def
/r exch def
/b exch def
/l exch def
newpath
l b moveto
l t lineto
r t lineto
r b lineto
closepath
patternNone not { ifill } if
brushNone not { istroke } if
end
} dup 0 4 dict put def

/Text {
ishow
} def

/idef {
dup where { pop pop pop } { exch def } ifelse
} def

/ifill {
0 begin
gsave
patternGrayLevel -1 ne {
fgred bgred fgred sub patternGrayLevel mul add
fggreen bggreen fggreen sub patternGrayLevel mul add
fgblue bgblue fgblue sub patternGrayLevel mul add setrgbcolor
eofill
} {
eoclip
originalCTM setmatrix
pathbbox /t exch def /r exch def /b exch def /l exch def
/w r l sub ceiling cvi def
/h t b sub ceiling cvi def
/imageByteWidth w 8 div ceiling cvi def
/imageHeight h def
bgred bggreen bgblue setrgbcolor
eofill
fgred fggreen fgblue setrgbcolor
w 0 gt h 0 gt and {
l w add b translate w neg h scale
w h true [w 0 0 h neg 0 h] { patternproc } imagemask
} if
} ifelse
grestore
end
} dup 0 8 dict put def

/istroke {
gsave
brushDashOffset -1 eq {
[] 0 setdash
1 setgray
} {
brushDashArray brushDashOffset setdash
fgred fggreen fgblue setrgbcolor
} ifelse
brushWidth setlinewidth
originalCTM setmatrix
stroke
grestore
} def

/ishow {
0 begin
gsave
fgred fggreen fgblue setrgbcolor
/fontDict printFont printSize scalefont dup setfont def
/descender fontDict begin 0 [FontBBox] 1 get FontMatrix end
transform exch pop def
/vertoffset 1 printSize sub descender sub def {
0 vertoffset moveto show
/vertoffset vertoffset printSize sub def
} forall
grestore
end
} dup 0 3 dict put def
/patternproc {
0 begin
/patternByteLength patternString length def
/patternHeight patternByteLength 8 mul sqrt cvi def
/patternWidth patternHeight def
/patternByteWidth patternWidth 8 idiv def
/imageByteMaxLength imageByteWidth imageHeight mul
stringLimit patternByteWidth sub min def
/imageMaxHeight imageByteMaxLength imageByteWidth idiv patternHeight idiv
patternHeight mul patternHeight max def
/imageHeight imageHeight imageMaxHeight sub store
/imageString imageByteWidth imageMaxHeight mul patternByteWidth add string def
0 1 imageMaxHeight 1 sub {
/y exch def
/patternRow y patternByteWidth mul patternByteLength mod def
/patternRowString patternString patternRow patternByteWidth getinterval def
/imageRow y imageByteWidth mul def
0 patternByteWidth imageByteWidth 1 sub {
/x exch def
imageString imageRow x add patternRowString putinterval
} for
} for
imageString
end
} dup 0 12 dict put def

/min {
dup 3 2 roll dup 4 3 roll lt { exch } if pop
} def

/max {
dup 3 2 roll dup 4 3 roll gt { exch } if pop
} def

/midpoint {
0 begin
/y1 exch def
/x1 exch def
/y0 exch def
/x0 exch def
x0 x1 add 2 div
y0 y1 add 2 div
end
} dup 0 4 dict put def

/thirdpoint {
0 begin
/y1 exch def
/x1 exch def
/y0 exch def
/x0 exch def
x0 2 mul x1 add 3 div
y0 2 mul y1 add 3 div
end
} dup 0 4 dict put def

/subspline {
0 begin
/movetoNeeded exch def
y exch get /y3 exch def
x exch get /x3 exch def
y exch get /y2 exch def
x exch get /x2 exch def
y exch get /y1 exch def
x exch get /x1 exch def
y exch get /y0 exch def
x exch get /x0 exch def
x1 y1 x2 y2 thirdpoint
/p1y exch def
/p1x exch def
x2 y2 x1 y1 thirdpoint
/p2y exch def
/p2x exch def
x1 y1 x0 y0 thirdpoint
p1x p1y midpoint
/p0y exch def
/p0x exch def
x2 y2 x3 y3 thirdpoint
p2x p2y midpoint
/p3y exch def
/p3x exch def
movetoNeeded { p0x p0y moveto } if
p1x p1y p2x p2y p3x p3y curveto
end
} dup 0 17 dict put def

/storexyn {
/n exch def
/y n array def
/x n array def
n 1 sub -1 0 {
/i exch def
y i 3 2 roll put
x i 3 2 roll put
} for
} def

/SSten {
fgred fggreen fgblue setrgbcolor
dup true exch 1 0 0 -1 0 6 -1 roll matrix astore
} def

/FSten {
dup 3 -1 roll dup 4 1 roll exch
newpath
0 0 moveto
dup 0 exch lineto
exch dup 3 1 roll exch lineto
0 lineto
closepath
bgred bggreen bgblue setrgbcolor
eofill
SSten
} def

/Rast {
exch dup 3 1 roll 1 0 0 -1 0 6 -1 roll matrix astore
} def




Begin
[ 0.799705 0 0 0.799705 0 0 ] concat
/originalCTM matrix currentmatrix def

Begin %I Pict

Begin %I Pict

Begin %I Line
0 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 97 141 ] concat
87 643 119 643 Line
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 224 787 ] concat
[
(Phase 1)
] Text
End

End %I eop

Begin %I Pict

Begin %I Line
0 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 97 141 ] concat
87 643 87 619 Line
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 6.12303e-17 1 -1 6.12303e-17 176.5 714.5 ] concat
[
(Phase2)
] Text
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -96 48 ] concat

Begin %I Pict
[ 1 0 0 1 -8 0 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 344 715 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 224 280 ] concat
79 419 175 443 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -8 8 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 344 683 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 224 248 ] concat
79 419 175 443 Rect
End

End %I eop

Begin %I Pict

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 336 667 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 216 232 ] concat
79 419 175 443 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 0 8 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 336 635 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 216 200 ] concat
79 419 175 443 Rect
End

End %I eop

End %I eop

End %I eop

Begin %I Pict
[ 1 0 0 1 15 -1 ] concat

Begin %I Pict
[ 1 0 0 1 176 192 ] concat

Begin %I Pict

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 184 571 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 92 144 ] concat
87 411 111 435 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -168 -72 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 376 643 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
271 483 295 507 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -200 8 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 432 563 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
327 403 351 427 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -280 104 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 536 467 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
431 307 455 331 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 72 -24 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 184 571 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 92 144 ] concat
87 411 111 435 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 48 -48 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 184 571 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 92 144 ] concat
87 411 111 435 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 24 -72 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 184 571 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 92 144 ] concat
87 411 111 435 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -192 -96 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 376 643 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
271 483 295 507 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -120 -120 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 376 643 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
271 483 295 507 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -144 -144 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 376 643 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
271 483 295 507 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -224 -16 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 432 563 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
327 403 351 427 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -248 -40 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 432 563 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
327 403 351 427 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -176 -64 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 432 563 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
327 403 351 427 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -304 80 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 536 467 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
431 307 455 331 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -352 32 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 536 467 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
431 307 455 331 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -328 56 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 536 467 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
431 307 455 331 Rect
End

End %I eop

End %I eop

Begin %I Pict
[ 1 0 0 1 160 0 ] concat

Begin %I Line
0 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 97 141 ] concat
87 643 87 619 Line
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 6.12303e-17 1 -1 6.12303e-17 176.5 714.5 ] concat
[
(Phase2)
] Text
End

End %I eop

Begin %I Pict
[ 1 0 0 1 160 0 ] concat

Begin %I Line
0 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 97 141 ] concat
87 643 119 643 Line
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 224 787 ] concat
[
(Phase 1)
] Text
End

End %I eop

End %I eop

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 160.25 662.724 ] concat
[
(\(a\) Row Block Distribution.)
] Text
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 323.75 664 ] concat
[
(\(b\) Our Proposed Distribution.)
] Text
End

End %I eop

showpage


end
%%EndDocument
 @endspecial 269 391 a Fo(Figure)11 b(7:)18 b(Data)13 b(Distribution)e(used)i
(to)f(alleviate)g(Memory)g(Contention.)503 1045 y @beginspecial
50 @llx 50 @lly 410 @urx 302 @ury 2057 @rwi @setspecial
%%BeginDocument: 256.adi.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.100 0.100 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
540 251 M
(1000) Rshow
LTa
600 585 M
2817 0 V
LTb
600 585 M
31 0 V
2786 0 R
-31 0 V
LTa
600 780 M
2817 0 V
LTb
600 780 M
31 0 V
2786 0 R
-31 0 V
LTa
600 919 M
2817 0 V
LTb
600 919 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1026 M
2817 0 V
LTb
600 1026 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1114 M
2817 0 V
LTb
600 1114 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1188 M
2817 0 V
LTb
600 1188 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1253 M
2817 0 V
LTb
600 1253 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1309 M
2817 0 V
LTb
600 1309 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1360 M
2817 0 V
LTb
600 1360 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(10000) Rshow
LTa
600 1694 M
2817 0 V
LTb
600 1694 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1889 M
2817 0 V
LTb
600 1889 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2028 M
2817 0 V
LTb
600 2028 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2135 M
2817 0 V
LTb
600 2135 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2223 M
2817 0 V
LTb
600 2223 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2297 M
2817 0 V
LTb
600 2297 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2362 M
2817 0 V
LTb
600 2362 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2418 M
2817 0 V
LTb
600 2418 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(100000) Rshow
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(0) Cshow
LTa
952 251 M
0 2218 V
LTb
952 251 M
0 63 V
0 2155 R
0 -63 V
952 151 M
(2) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(4) Cshow
LTa
1656 251 M
0 2218 V
LTb
1656 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(6) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(8) Cshow
LTa
2361 251 M
0 2218 V
LTb
2361 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(10) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(12) Cshow
LTa
3065 251 M
0 2218 V
LTb
3065 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(14) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(16) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
220 1260 M
currentpoint gsave translate 90 rotate 0 0 M
(Execution Time \(milli sec\)) Cshow
grestore
2008 51 M
(Number of Processors) Cshow
LT0
2009 1253 M
(Owner Computes\(Block, Block\)) Rshow
2069 1253 M
180 0 V
952 2200 M
352 -24 V
705 86 V
1408 89 V
2129 1253 D
952 2200 D
1304 2176 D
2009 2262 D
3417 2351 D
LT1
2009 1153 M
(Sequential) Rshow
2069 1153 M
180 0 V
776 2146 M
2129 1153 A
776 2146 A
LT2
2009 1053 M
(No Distributions) Rshow
2069 1053 M
180 0 V
3417 2056 M
-704 3 V
-704 6 V
-705 8 V
-352 61 V
2129 1053 B
3417 2056 B
2713 2059 B
2009 2065 B
1304 2073 B
952 2134 B
LT3
2009 953 M
(\(*,Cyclic\)) Rshow
2069 953 M
180 0 V
952 2090 M
352 -33 V
705 -48 V
704 25 V
704 18 V
2129 953 C
952 2090 C
1304 2057 C
2009 2009 C
2713 2034 C
3417 2052 C
LT4
2009 853 M
(\(*,Block\)) Rshow
2069 853 M
180 0 V
952 2070 M
352 -53 V
705 -45 V
704 23 V
704 9 V
2129 853 T
952 2070 T
1304 2017 T
2009 1972 T
2713 1995 T
3417 2004 T
LT5
2009 753 M
(Owner Computes\(*,Cyclic\)) Rshow
2069 753 M
180 0 V
952 2080 M
352 -252 V
705 -78 V
352 216 V
352 20 V
704 128 V
2129 753 S
952 2080 S
1304 1828 S
2009 1750 S
2361 1966 S
2713 1986 S
3417 2114 S
LT6
2009 653 M
(Owner Computes\(*,Block\)) Rshow
2069 653 M
180 0 V
3417 1820 M
2713 1611 L
-704 -55 V
-705 118 V
952 1954 L
2129 653 D
3417 1820 D
2713 1611 D
2009 1556 D
1304 1674 D
952 1954 D
LT7
2009 553 M
(\(Block,Block\)) Rshow
2069 553 M
180 0 V
1168 547 R
-704 222 V
-704 85 V
-705 249 V
952 1951 L
2129 553 A
3417 1100 A
2713 1322 A
2009 1407 A
1304 1656 A
952 1951 A
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 532 1111 a(Figure)f(8:)18 b(ADI)12 b(Performance)f(\(256x256\).)
4 1246 y(cases)18 b(adhering)d(to)i(owner)o(-computes)e(rule)h(can)h(incur)e
(severe)i(synchronization)f(or)g(ownership)f(test)4 1300 y(overhead)c(which)f
(exceeds)h(the)g(cost)g(of)f(accessing)i(remote)e(memory)m(.)17
b(W)l(e)11 b(use)g(the)g(Altering)f(Direction)4 1355 y(Integration)i(\()p
Fd(ADI)p Fo(\))f(to)i(illustrate)f(that)h(the)g(shared)f(address)i(space)f
(provides)g(\257exibility)e(in)i(the)g(choice)4 1409 y(of)i(computation)f
(partitions,)h(reducing)f(contention)g(and)h(synchronization)f(overhead,)i
(and)e(resulting)4 1463 y(in)e(signi\256cant)g(performance)g(improvements.)77
1539 y(W)l(e)k(use)f(the)g(Hector)n(,)h(a)f(Non-Uniform)e(Memory)i(Access)h
(multiprocessor)n(,)f(as)g(an)h(experimental)4 1593 y(platform.)21
b(Hector)13 b(consists)h(of)f(4)h(sets)g(of)f(processor)o(-memory)g(pairs)g
(connected)h(by)f(a)h(bus)g(to)f(form)g(a)4 1647 y(station;)g(4)g(stations)g
(are)f(connected)h(by)g(a)g(local)f(ring)g(to)h(form)f(a)h(cluster;)f(4)h
(local)g(rings)f(are)h(connected)4 1701 y(by)g(a)g(global)g(ring.)19
b(W)l(e)14 b(use)f(a)g(system)h(with)e(one)h(cluster)m(.)21
b(Each)13 b(processor)o(-memory)f(pair)g(consists)i(of)4 1755
y(a)f(Motorola)f(MC88100)h(CPU,)g(a)g(16)f(KB)h(instruction)f(cache,)i(a)f
(16)f(KB)h(data)g(cache)g(and)g(4)f(MB)i(of)e(the)4 1809 y(globally)i
(addressable)g(memory)m(.)23 b(The)15 b(hardware)e(provides)h(no)g(support)g
(for)f(cache)i(coherence.)23 b(The)4 1864 y(coherence)12 b(of)f(data)h(is)g
(maintained)f(by)h(software)f(at)h(cache)g(line)f(granularity)g([10)o(].)18
b(Data)12 b(distributions)4 1918 y(are)g(implemented)g(using)g(the)h(array)e
(allocation)h(techniques)h(described)f(in)g([21,)h(3].)4 2079
y Fc(5.1)58 b(Contention)15 b(and)f(Synchr)n(onization)h(Conscious)e
(Distribution)4 2177 y Fo(The)21 b Fd(ADI)e Fo(program)g(has)h(two)g(phases)h
(with)e(parallelism)g(along)h(orthogonal)f(dimensions)h(in)f(each)4
2231 y(phase.)k(It)13 b(operates)g(on)h(three)f(2-dimensional)g(arrays)g
Fj(A)p Fo(,)h Fj(B)i Fo(and)e Fj(X)t Fo(.)22 b(A)14 b(single)g(iteration)e
(of)i(an)f(outer)4 2285 y(sequentially)d(iterated)f(loop)h(consists)g(of)g(a)
g(forward)f(and)g(a)i(backward)e(sweep)i(phase)f(along)g(the)f(rows)h(of)4
2339 y(three)f(arrays,)h(followed)e(by)h(another)g(forward)e(and)i(backward)g
(sweep)h(phase)f(along)g(the)g(columns)g(of)g(the)4 2394 y(arrays)g([18].)17
b(This)10 b(application)f(is)h(typical)f(of)g(other)g(programs)g(such)g(as)h
Fd(2D-FFT)f Fo(and)h Fd(Erlebacher)4 2448 y Fo(that)i(have)h(parallelism)f
(in)g(orthogonal)f(directions)h(in)g(dif)o(ferent)f(phases)j(of)e(the)g
(program.)77 2523 y(The)k(best)g(data)f(distribution)f(scheme)i(for)e
Fd(ADI)h Fo(remains)g(an)g(issue)h(of)f(debate)h([18)o(,)g(4].)26
b(The)16 b(two)4 2578 y(proposed)h(schemes)g(partition)f(arrays)g(along)h(a)f
(single)h(dimension,)h(either)e(in)h(blocks)g(or)f(cyclically)m(.)4
2632 y(These)f(distributions,)g(in)e(conjunction)h(with)g(the)g(owner)o
(-computes)f(rule)h(result)f(in)h(a)h(wavefront)e(type)4 2686
y(computation,)e(leading)g(to)g(heavy)g(synchronization)g(overhead)g(in)g
(one)g(of)f(the)i(phases)g(of)e(the)h(program.)4 2740 y(Figure)i(7\(a\))g
(shows)i(a)f Fd(Block)f Fo(distribution)g(of)g(the)h(rows)g(of)f(the)h
(arrays.)23 b(W)n(ith)13 b(such)h(a)g(distribution,)4 2794
y(during)g(the)h(\256rst)g(phase)g(of)g(the)g(program)e(all)i(the)g
(processors)g(access)i(data)e(that)f(is)i(local)f(and)g(require)4
2848 y(no)g(communication.)25 b(During)14 b(the)h(second)g(phase,)h(however)n
(,)g(the)f(parallelism)f(is)h(orthogonal)f(to)h(the)p eop
%%Page: 8 8
8 7 bop 35 1 a Fo(T)m(able)12 b(1:)18 b(Performance)11 b(Bottlenecks)i(for)e
(various)h(data)h(and)f(computation)g(partitioning)e(for)i(ADI.)p
244 33 1363 2 v 243 84 2 51 v 252 84 V 277 69 a Fb(Data)g(Distribution)p
620 84 V 77 w(Compute)f(Rule)p 993 84 V 135 w(Performance)i(Bottleneck)p
1597 84 V 1606 84 V 244 85 1363 2 v 243 136 2 51 v 252 136
V 387 121 a(None)p 620 136 V 247 w(Relaxed)p 993 136 V 228
w(Memory)e(Contention)p 1597 136 V 1606 136 V 243 187 V 252
187 V 344 172 a(\(*,)h(Block\))p 620 187 V 117 w(Owner)o(-Computes)p
993 187 V 125 w(High)f(Synchronization)p 1597 187 V 1606 187
V 243 238 V 252 238 V 344 223 a(\(*,)h(Block\))p 620 238 V
204 w(Relaxed)p 993 238 V 228 w(Memory)f(Contention)p 1597
238 V 1606 238 V 243 289 V 252 289 V 339 273 a(\(*,)h(Cyclic\))p
620 289 V 112 w(Owner)o(-Computes)p 993 289 V 125 w(High)f(Synchronization)p
1597 289 V 1606 289 V 243 340 V 252 340 V 339 324 a(\(*,)h(Cyclic\))p
620 340 V 199 w(Relaxed)p 993 340 V 228 w(Memory)f(Contention)p
1597 340 V 1606 340 V 243 390 V 252 390 V 301 375 a(\(Block,)h(Block\))p
620 390 V 74 w(Owner)o(-Computes)p 993 390 V 180 w(Ownership)f(tests)p
1597 390 V 1606 390 V 243 441 V 252 441 V 301 426 a(\(Block,)h(Block\))p
620 441 V 161 w(Relaxed)p 993 441 V 137 w(High)f(Remote)g(Memory)g(Access)p
1597 441 V 1606 441 V 244 443 1363 2 v 4 606 a Fo(direction)17
b(of)g(distribution.)32 b(Strict)16 b(adherence)h(to)g(the)h(owner)o
(-computes)e(rule)h(implies)g(ordering)f(of)4 660 y(the)d(computations)f(by)g
(processors)h(on)f(the)g(corresponding)g(chunk)g(of)g(the)h(columns)f(they)g
(own.)19 b(Thus,)4 715 y(processor)10 b Fj(i)g Fo(has)h(to)f(wait)f(for)h
(processor)g Fj(i)c Fh(\000)g Fo(1)j(to)h(\256nish)g(the)g(computation)f(on)h
(its)h(chunk)f(of)f(the)h(column)4 769 y(before)g(proceeding.)17
b(A)10 b(lar)o(ger)f(number)h(of)f(synchronizations)h(are)g(required)f(to)h
(maintain)g(the)g(ordering)4 823 y(involved)i(in)g(the)g(wavefront)g
(computation.)77 899 y(The)i(synchronization)e(overhead)h(can)g(be)g
(eliminated)f(by)h(relaxing)f(the)h(owner)o(-computes)f(rule)h(in)4
953 y(the)18 b(second)g(phase)h(and)f(allowing)f(the)h(processor)g(to)f
(write)h(the)f(results)i(to)e(remote)h(memory)f(mod-)4 1007
y(ules.)24 b(This)15 b(eliminates)f(synchronization)f(overhead)h(at)g(the)g
(expense)g(of)g(increased)g(remote)g(memory)4 1061 y(accesses.)26
b(However)n(,)15 b(the)g(use)g(of)f(this)g(relaxed)g(compute)g(rule)g(with)g
(the)g Fd(\(*,Block\))g Fo(distribution)4 1115 y(results)9
b(in)g(heavy)g(contention.)17 b(Each)9 b(processor)g(is)g(responsible)g(for)f
(computing)g(a)i(column,)f(and)g(hence,)4 1169 y(each)14 b(processor)g
(accesses)h(every)e(memory)g(module)g(in)g(sequence.)23 b(Thus,)15
b(a)e(given)h(memory)e(module)4 1224 y(is)h(accessed)h(by)e(every)g
(processor)g(at)h(the)f(same)h(time,)f(leading)g(to)h(contention.)77
1299 y(The)k(data)e(distribution)g(scheme)h(depicted)f(in)h(Figure)f(7\(b\))
1149 1281 y Fm(4)1182 1299 y Fo(eliminates)h(contention)f(and)h(results)4
1353 y(in)21 b(the)g(best)h(possible)f(performance)f(with)h(the)g(relaxed)g
(compute)g(rule.)44 b(W)n(ith)21 b(this)g(distribution,)4 1408
y(processors)13 b(access)g(data)g(from)e(remote)g(memory)h(modules)g(in)g
(both)g(phases)h(of)f(the)g(program.)17 b(In)12 b(both)4 1462
y(phases,)h(processors)f(start)g(working)e(on)i(the)f(columns)h(assigned)g
(to)g(them)f(by)g(accessing)i(data)f(that)f(is)h(in)4 1516
y(dif)o(ferent)f(memory)f(modules)i(thus)g(avoiding)f(contention.)18
b(There)12 b(is)g(no)f(wavefront)g(type)h(parallelism,)4 1570
y(and)h(hence)f(no)g(overhead)g(involved)g(due)h(to)f(synchronization.)77
1646 y(The)19 b(use)f(of)f(owner)o(-computes)g(rule)h(with)f(the)h
(distribution)f(of)g(Figure)g(7\(b\))g(will)h(not)f(result)h(in)4
1700 y(good)f(performance.)31 b(Either)17 b(ownership)g(tests)h(must)f(be)g
(introduced)f(in)h(the)g(body)g(of)g(the)g(loops)g(to)4 1754
y(enforce)c(the)g(owner)o(-computes)f(rule,)i(or)e(the)i(loops)f(must)g(be)g
(rewritten)g(with)f(additional)h(strip-mined)4 1808 y(controlling)f(loops)h
(to)g(schedule)h(the)f(computations)f(on)h(sub-blocks)g(of)g(the)g(array)m(.)
20 b(The)14 b(former)d(leads)4 1862 y(to)h(overhead)g(and)h(the)f(latter)g
(introduces)g(synchronization)g(similar)f(to)i(the)f(wavefront)f
(computation.)77 1938 y(The)j(result)g(of)f(executing)g(the)h
Fd(ADI)f Fo(application)g(on)h(the)f(Hector)g(multiprocessor)g(for)g(a)h
(data)f(size)4 1992 y(of)18 b(256x256)f(with)h(various)g(data)g
(distributions)g(and)g(compute)g(rules)g(is)g(shown)g(in)g(Figure)g(8.)35
b(The)4 2046 y Fd(\(Block,Block\))17 b Fo(data)h(distribution)f(that)h
(relaxes)g(the)g(owner)o(-computes)f(rule)g(outperforms)g(all)4
2101 y(data)d(distribution)e(schemes)i(that)f(adhere)g(to)g(the)g(rule.)21
b(The)14 b(\256gure)f(also)g(indicates)h(that)f(the)g(overhead)4
2155 y(due)j(to)g(the)f(ownership)h(tests)g(when)g(using)g(the)f(owner)o
(-computes)g(rule)h(with)f(a)h Fd(\(Block,Block\))4 2209 y
Fo(distribution)d(degrades)h(performance.)21 b(It)14 b(is)g(also)g(clear)g
(that)f(the)h(use)g(of)g(data)g(distribution)e(improves)4 2263
y(performance)i(over)h(the)g(use)h(of)f(operating)f(system)i(policies)f(to)g
(manage)g(data)h(\(the)e(no)i(distributions)4 2317 y(curve\).)35
b(The)19 b(performance)e(bottlenecks)h(of)g(various)g(distributions)g(for)f
Fd(ADI)h Fo(are)g(summarized)g(in)4 2371 y(T)m(able)13 b(1.)p
4 2406 737 2 v 62 2437 a Fl(4)79 2452 y Fb(This)f(is)h(equivalent)g(to)f
(!HPF$)i(PROCESSORS)i(PROCS\(N\))g(with)c(!HPF$)i(DISTRIBUTE)h(B\(BLOCK,)4
2503 y(BLOCK\),)10 b(X\(BLOCK,)g(BLOCK\))g(ON)e(PROCS)j(in)d
Fa(HPF)p Fb(.)i(In)f(the)f(current)h Fa(HPF)h Fb(speci\256cation,)f(this)f
(distribution)4 2554 y(is)18 b(not)g(valid;)k(the)c(rank)h(of)g(each)g
(distributee)f(must)f(equal)i(the)g(rank)f(of)h(the)g(named)f(processor)h
(grid)f([16].)4 2604 y(Distributions)7 b(in)i(which)g(this)g(is)f(not)h(the)g
(case)h(introduce)f(additional)g(complexity)e(on)j(DMMs)e([17].)16
b(In)10 b(contrast,)4 2655 y(SSMMs)h(provide)g(the)g(\257exibility)f(to)h
(implement)f(these)h(distributions.)p eop
%%Page: 9 9
9 8 bop 4 -21 a Fn(6)71 b(Related)19 b(W)l(ork)4 106 y Fo(Several)12
b(researchers)g(have)g(focused)g(on)g(the)g(problem)f(of)g(deriving)g(data)h
(distributions)f(automatically)4 160 y(for)g(DMMs.)20 b(Li)12
b(and)g(Chen)h([22)o(],)f(Gupta)g(and)g(Banerjee)h([12)o(],)f(Zima)h(et)f
(al.)h([9)o(])f(and)g(Garcia)g(et)g(al.)h([11)o(])4 215 y(follow)e(the)h
(approach)g(of)f(\256nding)h(the)f(alignment)h(constraints)g(between)g(dif)o
(ferent)e(dimensions)i(of)g(the)4 269 y(arrays)g(and)g(derive)g(a)g(data)g
(distribution)f(that)h(minimizes)g(interprocessor)g(communication.)17
b(T)m(o)12 b(avoid)4 323 y(a)f(heuristic)g(approach,)g(Bixby)g(et)g(al.)h([7)
o(])f(formulate)e(a)j(0-1)e(integer)g(programming)g(problem)g(for)g(deriv-)4
377 y(ing)k(data)g(distributions.)21 b(Their)14 b(approach)g(relies)g(on)f
(the)h(assumption)g(that)g(a)g(good)f(data)h(distribution)4
431 y(for)h(the)i(entire)e(program)g(can)i(be)f(found)f(by)i(mer)o(ging)e
(the)h(data)g(distributions)g(of)f(smaller)h(segments)4 485
y(of)g(the)g(program.)27 b(They)17 b(minimize)e(the)h(interprocessor)f
(communication)g(using)h(the)g(\252performance)4 540 y(estimator)r(\272)c
(developed)h(by)g(Balasundaram)g(et)g(al.)g([6)o(].)20 b(Anderson)12
b([5])g(presents)i(an)e(algebraic)h(frame-)4 594 y(work)g(for)g(determining)f
(data)h(and)h(computation)e(partitions)h(by)g(minimizing)g(communication)f
(across)4 648 y(processors.)28 b(Data)16 b(transformations)e(are)i(then)f
(applied)h(so)f(that)h(the)f(processors)h(access)h(contiguous)4
702 y(data)g(regions)f(to)h(reduce)g(false)g(sharing.)31 b(This)17
b(technique)g(is)g(oblivious)f(to)h(SSMM)g(speci\256c)g(issues)4
756 y(such)c(as)g(contention)f(and)g(cache)h(af)o(\256nity)m(.)4
938 y Fn(7)71 b(Concluding)19 b(Remarks)4 1065 y Fo(Although)9
b(lar)o(ge)g(SSMMs)i(are)e(built)g(based)h(on)g(an)g(architecture)e(with)i
(distributed)f(memory)m(,)g(the)h(shared)4 1119 y(memory)15
b(paradigm)g(introduces)g(performance)g(issues)i(that)e(are)h(dif)o(ferent)f
(from)f(those)i(encountered)4 1173 y(in)e(DMMs.)24 b(The)14
b(high)f(cost)i(of)e(interprocessor)g(communication)g(in)h(distributed)f
(memory)f(multipro-)4 1227 y(cessors)18 b(makes)e(the)h(minimization)e(of)h
(communication)g(the)g(predominant)g(issue)h(in)f(selecting)h(data)4
1282 y(distributions)h(and)i(in)e(partitioning)g(computations.)38
b(On)19 b(SSMMs,)j(a)d(methodology)f(for)h(selecting)4 1336
y(data)14 b(distributions)g(must)g(also)g(consider)g(cache)h(af)o(\256nity)m
(,)f(memory)f(contention)h(and)g(false)g(sharing)g(in)4 1390
y(addition)d(to)g(the)g(cost)h(of)f(interprocessor)g(communication.)17
b(Furthermore,)10 b(the)h(single)h(shared)f(address)4 1444
y(space)j(present)f(in)g(SSMMs)g(provides)g(\257exibility)f(in)h(the)g
(selection)g(of)f(computation)h(partitions.)19 b(This)4 1498
y(should)e(be)f(exploited)g(in)g(applications)h(in)f(which)g(owner)o
(-computes)g(results)g(in)h(poor)f(performance.)4 1552 y(The)f
Fe(Jasmine)g Fo(compiler)f(project)g([2)o(])h(is)g(investigating)f(the)g
(issues)i(discussed)f(in)g(this)f(paper)h(through)4 1607 y(the)d(development)
g(of)g(a)h(framework)e(for)g(automatically)h(deriving)f(data)i(distributions)
e(on)i(SSMMs.)4 1777 y Fn(Refer)o(ences)29 1896 y Fo([1])24
b(T)l(.S.)17 b(Abdelrahman)f(et)g(al.)29 b(An)16 b(overview)f(of)h(the)g
(NUMAchine)h(multiprocessor)e(project.)28 b(In)112 1943 y Fe(Pr)n(oc.)13
b(of)g(the)f(Canadian)g(Super)n(computing)g(Conf.)p Fo(,)h(pages)g
(283\261295,)f(1994.)29 2032 y([2])24 b(T)l(.S.)12 b(Abdelrahman,)f(N.)h
(Manjikian,)g(and)f(S.)g(T)m(andri.)16 b(The)11 b(Jasmine)h(Compiler.)k(In)10
b(preparation.)29 2120 y([3])24 b(T)l(.S.)e(Abdelrahman)e(and)h(T)l(.N.)h(W)l
(ong.)41 b(Distributed)20 b(array)g(data)h(management)g(on)f(NUMA)112
2167 y(multiprocessors.)d(In)12 b Fe(Pr)n(oc.)i(of)e(SHPCC)p
Fo(,)i(pages)f(551\261559,)f(1994.)29 2256 y([4])24 b(S.P)-6
b(.)10 b(Amarasinghe,)g(J.M.)h(Anderson,)f(M.S.)h(Lam,)g(and)e(A.W)-5
b(.)11 b(Lim.)j(An)9 b(overview)g(of)g(a)h(compiler)112 2303
y(for)k(scalable)j(parallel)e(machines.)27 b(In)15 b Fe(Languages)h(and)f
(Compilers)i(for)f(Parallel)g(Computing)p Fo(,)112 2350 y(pages)c
(253\261272.)h(Springer)o(-V)-6 b(erlag)10 b(LNCS-768,)j(1993.)29
2438 y([5])24 b(J.M.)13 b(Anderson.)j(Demonstration)11 b(of)g(automatic)g
(data)h(and)f(computation)g(decomposition)g(tech-)112 2485
y(niques.)f(In)e Fe(Pr)n(oc.)g(of)g(the)g(W)-5 b(orkshop)8
b(on)g(Automatic)g(Data)g(Layout)g(and)g(Performance)g(Pr)n(ediction)p
Fo(,)112 2532 y(1995.)29 2620 y([6])24 b(V)-6 b(.)12 b(Balasundaram,)i(G.)f
(Fox,)g(K.)g(Kennedy)m(,)g(and)f(U.)i(Kremer)m(.)k(A)13 b(static)g
(performance)e(estimator)112 2667 y(to)h(guide)g(data)g(partitioning)f
(decisions.)19 b(In)12 b Fe(Pr)n(oc.)i(of)e(PPOPP)p Fo(,)j(pages)e
(213\261223,)f(1991.)29 2756 y([7])24 b(R.)9 b(Bixby)m(,)h(K.)g(Kennedy)m(,)f
(and)h(U.)f(Kremer)m(.)j(Automatic)d(data)g(layout)f(using)h(0-1)g(integer)f
(program-)112 2803 y(ming.)15 b(In)c Fe(Pr)n(oc.)i(of)e(the)g(Int'l)f(Conf.)i
(on)f(Parallel)g(Ar)n(chitectur)n(es)i(and)e(Compilation)g(T)-5
b(echniques)p Fo(,)112 2850 y(pages)12 b(111\261122,)h(1994.)p
eop
%%Page: 10 10
10 9 bop 29 -27 a Fo([8])24 b(W)-5 b(.J.)19 b(Bolosky)f(and)g(M.L.)h(Scott.)
32 b(False)18 b(sharing)f(and)h(its)g(ef)o(fect)f(on)g(shared)h(memory)f
(multi-)112 20 y(processors.)k(In)13 b Fe(Pr)n(oc.)j(of)d(4th)g(Symp.)h(on)g
(Experiences)h(with)e(Distributed)h(and)f(Multipr)n(ocessor)112
67 y(Systems)p Fo(,)g(pages)g(57\26171,)f(1993.)29 155 y([9])24
b(B.M.)14 b(Chapman,)g(T)l(.)h(Fahringer)n(,)e(and)g(H.)h(Zima.)21
b(Automatic)12 b(support)h(for)f(data)i(distribution)e(on)112
202 y(distributed)g(memory)g(multiprocessor)g(systems.)22 b(In)12
b Fe(Languages)h(and)g(Compilers)h(for)f(Parallel)112 249 y(Computing)p
Fo(,)f(pages)h(184\261199.)f(Springer)o(-V)-6 b(erlag)11 b(LNCS-768,)h(1993.)
4 337 y([10])24 b(B.)11 b(Gamsa.)k(Region-oriented)9 b(main)h(memory)f
(management)h(in)g(shared-memory)f(NUMA)i(mul-)112 384 y(tiprocessors.)19
b(Master)r(')m(s)13 b(thesis,)h(Department)d(of)i(Computer)f(Science,)h
(University)f(of)g(T)m(oronto,)112 431 y(T)m(oronto,)f(CANADA,)i(1992.)4
519 y([11])24 b(J.)f(Garcia,)h(E.)f(A)-5 b(yguade,)26 b(and)c(J.)h(Labarta.)
44 b(A)22 b(novel)g(approach)g(towards)g(automatic)g(data)112
566 y(distribution.)33 b(In)18 b Fe(Pr)n(oc.)i(of)e(the)h(W)-5
b(orkshop)20 b(on)e(Automatic)g(Data)g(Layout)g(and)h(Performance)112
613 y(Pr)n(ediction)p Fo(,)13 b(1995.)4 701 y([12])24 b(M.)16
b(Gupta)f(and)h(P)-6 b(.)17 b(Banerjee.)27 b(Automatic)15 b(data)g
(partitioning)g(on)g(distributed)g(memory)g(multi-)112 748
y(processors.)j Fe(IEEE)c(T)m(rans.)f(on)f(Parallel)h(and)f(Distributed)h
(Systems)p Fo(,)g(3\(2\):179\261193,)e(1992.)4 836 y([13])24
b(K.)15 b(Harzallah)g(and)g(K.C.)h(Sevcik.)25 b(Hot)15 b(spot)g(analysis)g
(in)g(lar)o(ge)g(scale)h(shared)f(memory)f(multi-)112 883 y(processors.)k(In)
12 b Fe(Pr)n(oc.)i(of)e(Super)n(computing'93)p Fo(,)g(pages)h(895\261905.)f
(ACM,)i(1993.)4 971 y([14])24 b(M.)11 b(Heinrich)f(et)h(al.)16
b(The)11 b(Stanford)f(FLASH)g(Multiprocessor.)16 b(In)10 b
Fe(Pr)n(oc.)i(of)f(the)g(21st)g(Int'l)e(Symp.)112 1018 y(on)j(Computer)g(Ar)n
(chitectur)n(e)p Fo(,)j(pages)e(302\261313,)f(1994.)4 1106
y([15])24 b(S.)15 b(Hiranandani,)i(K.)f(Kennedy)m(,)g(and)g(C.)g(T)m(seng.)27
b(Compiler)15 b(optimizations)g(for)g(Fortran)f(D)i(on)112
1153 y(MIMD)f(distributed-memory)e(machines.)25 b(In)15 b Fe(Pr)n(oc.)h(of)f
(Super)n(computing'91)p Fo(,)g(pages)h(86\261100,)112 1200
y(Albuquerque,)c(NM,)h(1991.)4 1288 y([16])24 b(HPF)l(.)33
b(High)17 b(Performance)g(Fortran)g(Language)i(Speci\256cation)e(\(High)g
(Performance)g(Fortran)112 1335 y(Forum\).)f(T)m(echnical)d(report)e
(CRPC-TR92225,)i(Rice)g(University)m(,)f(1994.)4 1424 y([17])24
b(C.)13 b(Koelbel.)18 b(HPF)12 b(constraints.)18 b(Personal)12
b(Communications,)g(1995.)4 1512 y([18])24 b(U.)14 b(Kremer)m(.)23
b(Automatic)14 b(data)g(layout)g(for)g(distributed-memory)e(multiprocessors.)
23 b(T)m(echnical)112 1559 y(report)11 b(CRPC-TR93229-S,)h(Center)h(for)e
(Research)i(on)f(Parallel)g(Computation,)g(1993.)4 1647 y([19])24
b(T)l(.T)l(.)15 b(Kwan,)f(B.K.)h(T)m(otty)m(,)e(and)g(D.A.)h(Reed.)21
b(Communication)12 b(and)i(computation)e(performance)112 1694
y(of)f(the)i(CM5.)19 b(In)12 b Fe(Pr)n(oc.)h(of)g(Super)n(computing'93)p
Fo(,)f(pages)h(192\261201.)f(ACM,)h(1993.)4 1782 y([20])24
b(D.)15 b(Lenoski)h(et)f(al.)26 b(The)15 b(Stanford)f(DASH)h(multiprocessor)m
(.)25 b Fe(IEEE)16 b(Computer)p Fo(,)h(25\(3\):63\26179,)112
1829 y(1992.)4 1917 y([21])24 b(H.)12 b(Li)g(and)g(K.C.)h(Sevcik.)k
(Numacros:)g(Data)12 b(parallel)f(programming)f(on)i(NUMA)g(multiproces-)112
1964 y(sors.)h(In)c Fe(Pr)n(oc.)i(of)e(4th)g(Symp.)h(on)g(Experiences)g(with)
g(Distributed)f(and)g(Multipr)n(ocessor)j(Systems)p Fo(,)112
2011 y(pages)g(247\261263,)h(1993.)4 2099 y([22])24 b(J.)11
b(Li)h(and)f(M.)h(Chen.)k(Compiling)10 b(communication-ef)o(\256cient)f
(programs)i(for)f(massively)h(parallel)112 2146 y(machines.)18
b Fe(Journal)12 b(of)h(Parallel)f(and)h(Distributed)f(Computing)p
Fo(,)g(2\(3\):361\261376,)f(1991.)4 2234 y([23])24 b(Cray)14
b(Research.)25 b(The)16 b(Cray)e(Research)h(Massively)h(Parallel)e(Processor)
g(System)h(-)f(Cray)h(T3D.)112 2281 y(T)m(echnical)d(report)f(80922,)i
(Munchen,)g(Germany)m(,)f(1993.)4 2369 y([24])24 b(Kendall)12
b(Square)f(Research.)19 b Fe(KSR1)13 b(Principles)h(of)e(Operation)p
Fo(.)18 b(W)l(altham,)13 b(MA,)g(1991.)4 2457 y([25])24 b(J.)12
b(T)m(orres,)g(E.)h(A)-5 b(yguade,)13 b(J.)g(Labarta,)f(and)g(M.)h(V)-6
b(alero.)17 b(Align)12 b(and)g(distribute-based)f(linear)g(loop)112
2504 y(transformations.)16 b(In)11 b Fe(Languages)g(and)h(Compilers)h(for)f
(Parallel)g(Computing)p Fo(,)g(pages)g(321\261339.)112 2551
y(Springer)o(-V)-6 b(erlag)10 b(LNCS-768,)j(1993.)4 2639 y([26])24
b(Z.)17 b(V)m(ranesic,)h(M.)f(Stumm,)f(R.)i(White,)f(and)f(D.)h(Lewis.)30
b(The)16 b(Hector)g(Multiprocessor.)29 b Fe(IEEE)112 2686 y(Computer)p
Fo(,)13 b(24\(1\):72\26180,)e(1991.)4 2774 y([27])24 b(R.W)-5
b(.)12 b(W)n(isniewski,)g(L.I.)g(Kontothanassis,)h(and)e(M.L.)h(Scott.)k
(High)11 b(performance)e(synchroniza-)112 2821 y(tion)i(algorithms)h(for)g
(multiprogrammed)e(multiprocessors.)18 b(In)12 b Fe(Pr)n(oc.)h(of)g(PPOPP)p
Fo(,)h(1995.)p eop
%%Trailer
end
userdict /end-hook known{end-hook}if
%%EOF

%!PS-Adobe-2.0
%%Creator: dvips 5.512 Copyright 1986, 1993 Radical Eye Software
%%Title: pdpta.dvi
%%CreationDate: Thu Nov 23 17:27:55 1995
%%Pages: 10
%%PageOrder: Ascend
%%BoundingBox: 0 0 612 792
%%DocumentFonts: Times-Bold Times-Roman Times-Italic Courier
%%EndComments
%DVIPSCommandLine: dvips -o pdpta.ps pdpta.dvi
%DVIPSSource:  TeX output 1995.08.11:1234
%%BeginProcSet: tex.pro
/TeXDict 250 dict def TeXDict begin /N{def}def /B{bind def}N /S{exch}N /X{S N}
B /TR{translate}N /isls false N /vsize 11 72 mul N /@rigin{isls{[0 -1 1 0 0 0]
concat}if 72 Resolution div 72 VResolution div neg scale isls{Resolution hsize
-72 div mul 0 TR}if Resolution VResolution vsize -72 div 1 add mul TR matrix
currentmatrix dup dup 4 get round 4 exch put dup dup 5 get round 5 exch put
setmatrix}N /@landscape{/isls true N}B /@manualfeed{statusdict /manualfeed
true put}B /@copies{/#copies X}B /FMat[1 0 0 -1 0 0]N /FBB[0 0 0 0]N /nn 0 N
/IE 0 N /ctr 0 N /df-tail{/nn 8 dict N nn begin /FontType 3 N /FontMatrix
fntrx N /FontBBox FBB N string /base X array /BitMaps X /BuildChar{
CharBuilder}N /Encoding IE N end dup{/foo setfont}2 array copy cvx N load 0 nn
put /ctr 0 N[}B /df{/sf 1 N /fntrx FMat N df-tail}B /dfs{div /sf X /fntrx[sf 0
0 sf neg 0 0]N df-tail}B /E{pop nn dup definefont setfont}B /ch-width{ch-data
dup length 5 sub get}B /ch-height{ch-data dup length 4 sub get}B /ch-xoff{128
ch-data dup length 3 sub get sub}B /ch-yoff{ch-data dup length 2 sub get 127
sub}B /ch-dx{ch-data dup length 1 sub get}B /ch-image{ch-data dup type
/stringtype ne{ctr get /ctr ctr 1 add N}if}B /id 0 N /rw 0 N /rc 0 N /gp 0 N
/cp 0 N /G 0 N /sf 0 N /CharBuilder{save 3 1 roll S dup /base get 2 index get
S /BitMaps get S get /ch-data X pop /ctr 0 N ch-dx 0 ch-xoff ch-yoff ch-height
sub ch-xoff ch-width add ch-yoff setcachedevice ch-width ch-height true[1 0 0
-1 -.1 ch-xoff sub ch-yoff .1 add]{ch-image}imagemask restore}B /D{/cc X dup
type /stringtype ne{]}if nn /base get cc ctr put nn /BitMaps get S ctr S sf 1
ne{dup dup length 1 sub dup 2 index S get sf div put}if put /ctr ctr 1 add N}
B /I{cc 1 add D}B /bop{userdict /bop-hook known{bop-hook}if /SI save N @rigin
0 0 moveto /V matrix currentmatrix dup 1 get dup mul exch 0 get dup mul add
.99 lt{/QV}{/RV}ifelse load def pop pop}N /eop{SI restore showpage userdict
/eop-hook known{eop-hook}if}N /@start{userdict /start-hook known{start-hook}
if pop /VResolution X /Resolution X 1000 div /DVImag X /IE 256 array N 0 1 255
{IE S 1 string dup 0 3 index put cvn put}for 65781.76 div /vsize X 65781.76
div /hsize X}N /p{show}N /RMat[1 0 0 -1 0 0]N /BDot 260 string N /rulex 0 N
/ruley 0 N /v{/ruley X /rulex X V}B /V{}B /RV statusdict begin /product where{
pop product dup length 7 ge{0 7 getinterval dup(Display)eq exch 0 4
getinterval(NeXT)eq or}{pop false}ifelse}{false}ifelse end{{gsave TR -.1 -.1
TR 1 1 scale rulex ruley false RMat{BDot}imagemask grestore}}{{gsave TR -.1
-.1 TR rulex ruley scale 1 1 false RMat{BDot}imagemask grestore}}ifelse B /QV{
gsave transform round exch round exch itransform moveto rulex 0 rlineto 0
ruley neg rlineto rulex neg 0 rlineto fill grestore}B /a{moveto}B /delta 0 N
/tail{dup /delta X 0 rmoveto}B /M{S p delta add tail}B /b{S p tail}B /c{-4 M}
B /d{-3 M}B /e{-2 M}B /f{-1 M}B /g{0 M}B /h{1 M}B /i{2 M}B /j{3 M}B /k{4 M}B
/w{0 rmoveto}B /l{p -4 w}B /m{p -3 w}B /n{p -2 w}B /o{p -1 w}B /q{p 1 w}B /r{
p 2 w}B /s{p 3 w}B /t{p 4 w}B /x{0 S rmoveto}B /y{3 2 roll p a}B /bos{/SS save
N}B /eos{SS restore}B end
%%EndProcSet
%%BeginProcSet: texps.pro
TeXDict begin /rf{findfont dup length 1 add dict begin{1 index /FID ne 2 index
/UniqueID ne and{def}{pop pop}ifelse}forall[1 index 0 6 -1 roll exec 0 exch 5
-1 roll VResolution Resolution div mul neg 0 0]/Metrics exch def dict begin
Encoding{exch dup type /integertype ne{pop pop 1 sub dup 0 le{pop}{[}ifelse}{
FontMatrix 0 get div Metrics 0 get div def}ifelse}forall Metrics /Metrics
currentdict end def[2 index currentdict end definefont 3 -1 roll makefont
/setfont load]cvx def}def /ObliqueSlant{dup sin S cos div neg}B /SlantFont{4
index mul add}def /ExtendFont{3 -1 roll mul exch}def /ReEncodeFont{/Encoding
exch def}def end
%%EndProcSet
%%BeginProcSet: special.pro
TeXDict begin /SDict 200 dict N SDict begin /@SpecialDefaults{/hs 612 N /vs
792 N /ho 0 N /vo 0 N /hsc 1 N /vsc 1 N /ang 0 N /CLIP 0 N /rwiSeen false N
/rhiSeen false N /letter{}N /note{}N /a4{}N /legal{}N}B /@scaleunit 100 N
/@hscale{@scaleunit div /hsc X}B /@vscale{@scaleunit div /vsc X}B /@hsize{/hs
X /CLIP 1 N}B /@vsize{/vs X /CLIP 1 N}B /@clip{/CLIP 2 N}B /@hoffset{/ho X}B
/@voffset{/vo X}B /@angle{/ang X}B /@rwi{10 div /rwi X /rwiSeen true N}B /@rhi
{10 div /rhi X /rhiSeen true N}B /@llx{/llx X}B /@lly{/lly X}B /@urx{/urx X}B
/@ury{/ury X}B /magscale true def end /@MacSetUp{userdict /md known{userdict
/md get type /dicttype eq{userdict begin md length 10 add md maxlength ge{/md
md dup length 20 add dict copy def}if end md begin /letter{}N /note{}N /legal{
}N /od{txpose 1 0 mtx defaultmatrix dtransform S atan/pa X newpath clippath
mark{transform{itransform moveto}}{transform{itransform lineto}}{6 -2 roll
transform 6 -2 roll transform 6 -2 roll transform{itransform 6 2 roll
itransform 6 2 roll itransform 6 2 roll curveto}}{{closepath}}pathforall
newpath counttomark array astore /gc xdf pop ct 39 0 put 10 fz 0 fs 2
F/|______Courier fnt invertflag{PaintBlack}if}N /txpose{pxs pys scale ppr
aload pop por{noflips{pop S neg S TR pop 1 -1 scale}if xflip yflip and{pop S
neg S TR 180 rotate 1 -1 scale ppr 3 get ppr 1 get neg sub neg ppr 2 get ppr 0
get neg sub neg TR}if xflip yflip not and{pop S neg S TR pop 180 rotate ppr 3
get ppr 1 get neg sub neg 0 TR}if yflip xflip not and{ppr 1 get neg ppr 0 get
neg TR}if}{noflips{TR pop pop 270 rotate 1 -1 scale}if xflip yflip and{TR pop
pop 90 rotate 1 -1 scale ppr 3 get ppr 1 get neg sub neg ppr 2 get ppr 0 get
neg sub neg TR}if xflip yflip not and{TR pop pop 90 rotate ppr 3 get ppr 1 get
neg sub neg 0 TR}if yflip xflip not and{TR pop pop 270 rotate ppr 2 get ppr 0
get neg sub neg 0 S TR}if}ifelse scaleby96{ppr aload pop 4 -1 roll add 2 div 3
1 roll add 2 div 2 copy TR .96 dup scale neg S neg S TR}if}N /cp{pop pop
showpage pm restore}N end}if}if}N /normalscale{Resolution 72 div VResolution
72 div neg scale magscale{DVImag dup scale}if 0 setgray}N /psfts{S 65781.76
div N}N /startTexFig{/psf$SavedState save N userdict maxlength dict begin
/magscale false def normalscale currentpoint TR /psf$ury psfts /psf$urx psfts
/psf$lly psfts /psf$llx psfts /psf$y psfts /psf$x psfts currentpoint /psf$cy X
/psf$cx X /psf$sx psf$x psf$urx psf$llx sub div N /psf$sy psf$y psf$ury
psf$lly sub div N psf$sx psf$sy scale psf$cx psf$sx div psf$llx sub psf$cy
psf$sy div psf$ury sub TR /showpage{}N /erasepage{}N /copypage{}N /p 3 def
@MacSetUp}N /doclip{psf$llx psf$lly psf$urx psf$ury currentpoint 6 2 roll
newpath 4 copy 4 2 roll moveto 6 -1 roll S lineto S lineto S lineto closepath
clip newpath moveto}N /endTexFig{end psf$SavedState restore}N /@beginspecial{
SDict begin /SpecialSave save N gsave normalscale currentpoint TR
@SpecialDefaults count /ocount X /dcount countdictstack N}N /@setspecial{CLIP
1 eq{newpath 0 0 moveto hs 0 rlineto 0 vs rlineto hs neg 0 rlineto closepath
clip}if ho vo TR hsc vsc scale ang rotate rwiSeen{rwi urx llx sub div rhiSeen{
rhi ury lly sub div}{dup}ifelse scale llx neg lly neg TR}{rhiSeen{rhi ury lly
sub div dup scale llx neg lly neg TR}if}ifelse CLIP 2 eq{newpath llx lly
moveto urx lly lineto urx ury lineto llx ury lineto closepath clip}if
/showpage{}N /erasepage{}N /copypage{}N newpath}N /@endspecial{count ocount
sub{pop}repeat countdictstack dcount sub{end}repeat grestore SpecialSave
restore end}N /@defspecial{SDict begin}N /@fedspecial{end}B /li{lineto}B /rl{
rlineto}B /rc{rcurveto}B /np{/SaveX currentpoint /SaveY X N 1 setlinecap
newpath}N /st{stroke SaveX SaveY moveto}N /fil{fill SaveX SaveY moveto}N
/ellipse{/endangle X /startangle X /yrad X /xrad X /savematrix matrix
currentmatrix N TR xrad yrad scale 0 0 1 startangle endangle arc savematrix
setmatrix}N end
%%EndProcSet
TeXDict begin 40258431 52099146 1000 300 300
(/stumm/a0/tandri/pdpta/pdpta.dvi) @start /Fa 175[27 7[27 1[27
70[{}3 45.833332 /Courier rf /Fb 80[25 25 51[20 23 23 33 23
23 13 18 15 23 23 23 23 36 13 23 1[13 23 23 15 20 23 20 23
20 3[15 1[15 2[33 2[33 28 25 30 1[25 33 33 41 28 33 1[15 33
1[25 28 33 30 30 33 5[13 3[23 23 4[23 2[11 15 11 1[23 15 15
3[23 2[15 33[{}60 45.833332 /Times-Roman rf /Fc 81[29 51[23
26 2[26 29 16 23 23 2[29 29 42 16 2[16 29 29 16 26 29 26 29
29 13[29 2[36 42 1[48 6[36 1[42 39 1[36 11[29 29 29 29 29 2[15
19 45[{}36 58.333336 /Times-Italic rf /Fd 134[30 2[30 30 30
30 30 1[30 30 30 30 30 30 1[30 30 30 30 30 30 30 30 30 12[30
6[30 3[30 2[30 30 30 30 30 30 14[30 4[30 30 1[30 30 30 40[{}36
50.000000 /Courier rf /Fe 134[22 22 33 1[25 14 19 19 25 25
25 25 36 14 22 1[14 25 25 14 22 25 22 25 25 9[41 2[28 25 30
1[30 36 1[41 28 33 22 17 36 2[30 36 33 1[30 7[25 4[25 25 25
25 2[12 17 5[17 39[{}47 50.000000 /Times-Italic rf /Ff 1 1
df<FFFFF0FFFFF014027D881B>0 D E /Fg 4 117 df<1F0006000600060006000C000C000C00
0C0018F01B181C08180838183018301830306030603160616062C022C03C10177E9614>104
D<0300038003000000000000000000000000001C002400460046008C000C001800180018003100
3100320032001C0009177F960C>I<383C0044C6004702004602008E06000C06000C06000C0C00
180C00180C40181840181880300880300F00120E7F8D15>110 D<030003000600060006000600
FFC00C000C000C001800180018001800300030803080310031001E000A147F930D>116
D E /Fh 3 3 df<FFFFFFFCFFFFFFFC1E027C8C27>0 D<70F8F8F87005057C8E0E>I<C00003E0
000770000E38001C1C00380E00700700E00381C001C38000E700007E00003C00003C00007E0000
E70001C3800381C00700E00E00701C003838001C70000EE00007C000031818799727>I
E /Fi 4 62 df<00200040008001000300060004000C000C001800180030003000300070006000
60006000E000E000E000E000E000E000E000E000E000E000E000E000E000E00060006000600070
00300030003000180018000C000C0004000600030001000080004000200B327CA413>40
D<800040002000100018000C000400060006000300030001800180018001C000C000C000C000E0
00E000E000E000E000E000E000E000E000E000E000E000E000E000C000C000C001C00180018001
80030003000600060004000C00180010002000400080000B327DA413>I<000180000001800000
018000000180000001800000018000000180000001800000018000000180000001800000018000
00018000000180000001800000018000FFFFFFFEFFFFFFFE000180000001800000018000000180
000001800000018000000180000001800000018000000180000001800000018000000180000001
800000018000000180001F227D9C26>43 D<FFFFFFFEFFFFFFFE00000000000000000000000000
00000000000000000000000000000000000000FFFFFFFEFFFFFFFE1F0C7D9126>61
D E /Fj 16 111 df<003F000000E180000380C020070060400E0070401C0070403C0070803C00
3880780039007800390078003A00F0003A00F0003C00F0003800F0003800700038007000780030
00B800380338401C1C188007E00F001B157E941F>11 D<70F8FCFC740404040408081010204006
0F7C840E>59 D<00000080000000018000000001C000000003C000000003C000000007C0000000
0BC00000000BC000000013C000000033C000000023C000000043C000000043E000000081E00000
0181E000000101E000000201E000000201E000000401E000000C01E000000801E000001001E000
001FFFF000002000F000006000F000004000F000008000F000008000F000010000F000030000F0
00020000F000040000F8000C0000F8001E0000F800FF800FFF8021237EA225>65
D<007FFFF8000007800F00000780078000078003C0000F0001C0000F0001C0000F0001E0000F00
01E0001E0001C0001E0003C0001E0003C0001E000780003C000F00003C001E00003C003C00003C
01F000007FFFE00000780078000078003C000078001E0000F0001E0000F0000E0000F0000F0000
F0000F0001E0001E0001E0001E0001E0001E0001E0003C0003C0003C0003C000780003C000F000
03C001C00007C00F8000FFFFFC000023227EA125>I<007FFFF0000007801C000007800F000007
800700000F000380000F000380000F000380000F000380001E000780001E000780001E00078000
1E000F00003C000F00003C001E00003C003C00003C007000007801E000007FFF00000078000000
007800000000F000000000F000000000F000000000F000000001E000000001E000000001E00000
0001E000000003C000000003C000000003C000000003C000000007C0000000FFFC00000021227E
A11F>80 D<007FFFE0000007803C000007800E000007800700000F000780000F000380000F0003
C0000F0003C0001E000780001E000780001E000780001E000F00003C001E00003C003C00003C00
7000003C01C000007FFE00000078078000007801C000007801E00000F000F00000F000F00000F0
00F00000F000F00001E001E00001E001E00001E001E00001E001E00003C003C00003C003C04003
C003C04003C001C08007C001C080FFFC00E3000000003C0022237EA125>82
D<3FFE01FF8003C0003C0003C000300003C0001000078000200007800020000780002000078000
20000F000040000F000040000F000040000F000040001E000080001E000080001E000080001E00
0080003C000100003C000100003C000100003C0001000078000200007800020000780002000078
000200007000040000F000040000F0000800007000080000700010000070002000003800400000
38008000001C01000000060600000001F800000021237DA121>85 D<FFF8007FC00F80000F000F
00000C000F000008000F000010000F800010000780002000078000600007800040000780008000
07800080000780010000078002000007C002000003C004000003C00C000003C008000003C01000
0003C010000003C020000003E040000003E040000001E080000001E180000001E100000001E200
000001E200000001E400000001F800000000F800000000F000000000E000000000E000000000C0
00000000C000000022237DA11C>I<007FFC03FF0007E000F80007C000E00003C000800003E001
000001E002000001F006000001F00C000000F018000000F81000000078200000007C400000007C
800000003D000000003E000000001E000000001F000000001F000000002F000000006F80000000
C78000000187C000000103C000000203C000000403E000000801E000001001F000002000F00000
4000F800008000F80001800078000300007C000F8000FC00FFE007FFC028227FA128>88
D<00001E00000063800000C7800001C7800001C300000180000003800000038000000380000003
80000007000000070000000700000007000000FFF800000E0000000E0000000E0000000E000000
0E0000000E0000001C0000001C0000001C0000001C0000001C0000003800000038000000380000
0038000000380000007000000070000000700000007000000060000000E0000000E0000000E000
0000C0000070C00000F1800000F1000000620000003C000000192D7EA218>102
D<000F0C00389C00605C00C03801C0380380380780380700700F00700F00700F00701E00E01E00
E01E00E01E00E01E01C00E01C00E03C00605C0031B8001E3800003800003800007000007000007
00700E00F00C00F018006070003FC000161F809417>I<00F0000FE00000E00000E00000E00001
C00001C00001C00001C000038000038000038000038000070000071F0007218007C0C00F00E00F
00E00E00E00E00E01C01C01C01C01C01C01C01C0380380380380380700380704700708700E0870
0E08700610E006206003C016237DA21C>I<00E000E001E000C000000000000000000000000000
00000000001E0023004380438083808380870007000E000E000E001C001C003800382038407040
7040308031001E000B227EA111>I<0000E00001E00001E00000C0000000000000000000000000
000000000000000000000000000000001E00002300004380008380010380010380010380000700
000700000700000700000E00000E00000E00000E00001C00001C00001C00001C00003800003800
00380000380000700000700000700070E000F0C000F180006300003E0000132C81A114>I<00F0
000FE00000E00000E00000E00001C00001C00001C00001C0000380000380000380000380000700
000700F00703080704380E08780E10780E20300E40001C80001F00001FC0001C70003838003838
00381C00381C10703820703820703820701840E00C8060070015237DA219>I<3C07C046186047
20308740388780388700388700380E00700E00700E00700E00701C00E01C00E01C01C01C01C138
01C23803823803823801847001883000F018157E941D>110 D E /Fk 134[21
1[30 1[21 12 16 14 1[21 21 21 32 12 2[12 21 21 14 18 21 18
21 18 3[14 1[14 17[14 5[28 8[12 21 21 5[21 21 1[12 10 14 45[{}32
41.666668 /Times-Roman rf /Fl 203[15 15 15 15 49[{}4 29.166668
/Times-Roman rf /Fm 203[17 17 17 17 17 48[{}5 33.333332 /Times-Roman
rf /Fn 138[39 23 27 31 1[39 35 39 59 20 39 1[20 1[35 23 31
39 31 39 35 9[71 4[51 1[43 6[27 2[43 1[51 51 11[35 35 35 35
35 35 35 49[{}32 70.833336 /Times-Bold rf /Fo 69[22 8[25 1[28
28 3[22 47[22 25 25 36 25 25 14 19 17 25 25 25 25 39 14 25
14 14 25 25 17 22 25 22 25 22 3[17 1[17 30 2[47 36 36 30 28
33 1[28 36 36 44 30 36 19 17 36 36 28 30 36 33 33 36 3[28 1[14
14 25 25 25 25 25 25 25 25 25 25 1[12 17 12 2[17 17 17 39[{}75
50.000000 /Times-Roman rf /Fp 139[17 19 22 14[22 28 25 31[36
65[{}7 50.000000 /Times-Bold rf /Fq 2 104 df<0000F80003C0000F00001E00003C0000
780000780000780000780000780000780000780000780000780000780000780000780000780000
780000780000780000780000780000780000780000F00000F00001E000078000FE0000FE000007
800001E00000F00000F00000780000780000780000780000780000780000780000780000780000
7800007800007800007800007800007800007800007800007800007800007800003C00001E0000
0F000003C00000F8153C7CAC1E>102 D<F800000F000003C00001E00000F00000780000780000
780000780000780000780000780000780000780000780000780000780000780000780000780000
7800007800007800007800007800003C00003C00001E000007000001F80001F8000700001E0000
3C00003C0000780000780000780000780000780000780000780000780000780000780000780000
780000780000780000780000780000780000780000780000780000F00001E00003C0000F0000F8
0000153C7CAC1E>I E /Fr 134[29 2[29 29 16 23 19 1[29 29 29 45
16 29 1[16 29 29 19 26 29 26 29 26 11[42 36 32 5[52 7[36 42
39 1[42 54 5[16 4[29 29 2[29 2[15 19 15 44[{}37 58.333336 /Times-Roman
rf /Fs 134[42 3[46 28 32 37 1[46 42 46 69 23 2[23 46 42 1[37
46 37 46 42 13[46 2[51 2[78 8[60 60 67[{}23 83.333336 /Times-Bold
rf end
%%EndProlog
%%BeginSetup
%%Feature: *Resolution 300dpi
TeXDict begin

%%EndSetup
%%Page: 1 1
1 0 bop 80 177 a Fs(Computation)19 b(and)i(Data)e(Partitioning)g(on)h
(Scalable)341 280 y(Shar)o(ed)g(Memory)g(Multipr)o(ocessors)403
451 y Fr(Sudarsan)15 b(T)l(andri)29 b(and)g(T)l(arek)14 b(S.)g(Abdelrahman)
316 526 y(Department)h(of)g(Electrical)f(and)h(Computer)g(Engineering)284
601 y(The)f(University)h(of)g(T)l(oronto,)f(T)l(oronto,)g(Canada,)f(M5S)i
(1A4)478 675 y(e-mail:)g Fq(f)p Fr(tandri,tsa)p Fq(g)p Fr(@eecg.toronto.edu)
833 865 y Fp(Abstract)217 945 y Fo(In)g(this)h(paper)f(we)h(identify)f(the)h
(factors)f(that)h(af)o(fect)f(the)h(derivation)e(of)i(com-)217
999 y(putation)10 b(and)h(data)g(partitions)g(on)g(scalable)g(shared)g
(memory)g(multiprocessors)217 1053 y(\(SSMMs\).)18 b(W)l(e)12
b(show)h(that)f(these)h(factors)f(necessitate)i(an)e(SSMM-conscious)217
1107 y(approach.)17 b(In)10 b(addition)g(to)g(remote)g(memory)f(access,)k
(which)d(is)h(the)f(sole)h(factor)217 1161 y(on)19 b(distributed)g(memory)f
(multiprocessors,)k(cache)d(af)o(\256nity)m(,)i(memory)e(con-)217
1216 y(tention)12 b(and)h(false)g(sharing)f(are)h(important)f(factors)g(that)
h(must)g(be)g(considered.)217 1270 y(Experimental)g(evidence)h(is)g
(presented)g(to)g(demonstrate)f(the)h(impact)f(of)h(these)217
1324 y(factors)i(on)g(performance)g(using)g(three)h(applications)f(on)h(the)f
(KSR1)h(and)f(the)217 1378 y(Hector)c(multiprocessors.)4 1540
y Fn(1)71 b(Intr)o(oduction)4 1667 y Fo(Scalable)12 b(shared)g(memory)f
(multiprocessors)g(\(SSMMs\))g(are)h(becoming)f(increasingly)h(popular)f(and)
h(a)4 1721 y(viable)e(alternative)f(to)h(distributed)f(memory)g
(multiprocessors)h(\(DMMs\).)17 b(The)11 b(Stanford)e(DASH)g([20],)4
1775 y(FLASH)i([14)o(],)h(the)f(KSR1)f([24],)h(T)m(oronto')m(s)f(Hector)h
([26)o(],)h(NUMAchine)f([1)o(],)h(and)f(the)f(Cray)h(T3D)h([23)o(])4
1830 y(are)d(some)g(SSMMs)g(currently)e(in)i(use)g(or)f(under)g(development.)
17 b(Processors)9 b(in)f(a)h(SSMM)g(share)g(a)g(single)4 1884
y(coherent)f(address)g(space.)17 b(However)n(,)9 b(shared)f(memory)g(is)g
(physically)g(distributed)g(to)f(allow)h(scalability)l(,)4
1938 y(as)17 b(shown)f(in)g(Figure)f(1.)29 b(This)17 b(distribution)e(of)g
(shared)i(memory)e(results)h(in)g(non-uniform)e(memory)4 1992
y(access)f(latencies,)g(depending)f(on)f(the)h(distance)h(between)f(a)g
(processor)f(and)h(memory)m(.)17 b(Consequently)m(,)4 2046
y(careful)12 b(placement)g(and)g(management)g(of)g(data)h(is)g(essential)g
(for)e(scaling)i(performance.)77 2122 y(W)l(e)i(believe)f(that)g(data)g
(distribution)732 2104 y Fm(1)764 2122 y Fo(is)g(a)h(good)f(paradigm)f(for)h
(managing)f(data)i(in)f(data-parallel)4 2176 y(applications)h(on)g(SSMMs)g
([3)o(,)h(21].)25 b(The)16 b(division)e(of)h(array)f(data)h(allows)g(a)g
(compiler)f(to)h(place)g(data)4 2230 y(in)g(the)g(physical)f(memory)g(of)h
(the)g(processor)f(that)h(uses)h(it)e(the)h(most,)h(and)f(also)g(allows)g
(the)g(compiler)4 2284 y(to)k(partition)f(the)h(computations)g(of)f(parallel)
h(loops.)38 b(W)l(e)19 b(have)g(experimented)g(with)f(programmer)4
2339 y(speci\256ed)12 b(data)f(distributions)g(on)g(the)h(Hector)f
(multiprocessor)f(and)i(have)g(found)e(them)h(to)h(be)f(ef)o(fective)4
2393 y(in)e(improving)e(performance.)16 b(However)n(,)10 b(the)e(task)h(of)g
(selecting)g(a)g(good)f(data)h(distribution)f(requires)g(the)4
2447 y(programmer)i(to)h(understand)g(both)f(the)i(parallel)e(machine)h
(architecture)g(and)g(the)g(data)g(access)i(patterns)4 2501
y(in)19 b(the)f(program.)37 b(Porting)17 b(programs)h(to)h(various)g
(machines)g(and)f(tuning)h(them)f(for)g(performance)4 2555
y(becomes)g(a)f(tedious)g(and)g(laborious)g(process.)33 b(Consequently)m(,)19
b(it)e(is)h(desirable)f(to)g(derive)f(data)i(and)4 2609 y(computation)h
(partitions)g(automatically)h(using)g(a)g(compiler)m(.)40 b(The)21
b(objective)e(of)h(this)g(paper)g(is)g(to)4 2664 y(describe)13
b(the)f(factors)g(that)g(af)o(fect)g(the)g(derivation)g(of)g(computation)f
(and)i(data)f(partitions)g(on)g(SSMMs.)77 2739 y(On)19 b(DMMs,)k(the)c(main)g
(factor)f(that)i(af)o(fects)f(the)g(performance)f(of)h(an)g(application)g(is)
g(the)g(cost)4 2793 y(of)d(interprocessor)f(communication.)28
b(Consequently)m(,)17 b(scalable)g(performance)e(can)h(be)g(achieved)g(by)p
4 2838 737 2 v 62 2869 a Fl(1)79 2884 y Fk(In)10 b(this)f(paper)i(we)g(use)f
(the)g(terms)h(data)g(distributi)o(ons)c(and)k(data)f(partitions)f
(interchangeably)m(.)p eop
%%Page: 2 2
2 1 bop 175 533 a @beginspecial 114 @llx 408 @lly 476 @urx
553 @ury 3600 @rwi @setspecial
%%BeginDocument: numaarch1.ps

/arrowHeight 10 def
/arrowWidth 5 def

/IdrawDict 51 dict def
IdrawDict begin

/reencodeISO {
dup dup findfont dup length dict begin
{ 1 index /FID ne { def }{ pop pop } ifelse } forall
/Encoding ISOLatin1Encoding def
currentdict end definefont
} def

/ISOLatin1Encoding [
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/space/exclam/quotedbl/numbersign/dollar/percent/ampersand/quoteright
/parenleft/parenright/asterisk/plus/comma/minus/period/slash
/zero/one/two/three/four/five/six/seven/eight/nine/colon/semicolon
/less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N
/O/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright
/asciicircum/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m
/n/o/p/q/r/s/t/u/v/w/x/y/z/braceleft/bar/braceright/asciitilde
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/dotlessi/grave/acute/circumflex/tilde/macron/breve
/dotaccent/dieresis/.notdef/ring/cedilla/.notdef/hungarumlaut
/ogonek/caron/space/exclamdown/cent/sterling/currency/yen/brokenbar
/section/dieresis/copyright/ordfeminine/guillemotleft/logicalnot
/hyphen/registered/macron/degree/plusminus/twosuperior/threesuperior
/acute/mu/paragraph/periodcentered/cedilla/onesuperior/ordmasculine
/guillemotright/onequarter/onehalf/threequarters/questiondown
/Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE/Ccedilla
/Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex
/Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis
/multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute
/Thorn/germandbls/agrave/aacute/acircumflex/atilde/adieresis
/aring/ae/ccedilla/egrave/eacute/ecircumflex/edieresis/igrave
/iacute/icircumflex/idieresis/eth/ntilde/ograve/oacute/ocircumflex
/otilde/odieresis/divide/oslash/ugrave/uacute/ucircumflex/udieresis
/yacute/thorn/ydieresis
] def
/Times-Roman reencodeISO def

/none null def
/numGraphicParameters 17 def
/stringLimit 65535 def

/Begin {
save
numGraphicParameters dict begin
} def

/End {
end
restore
} def

/SetB {
dup type /nulltype eq {
pop
false /brushRightArrow idef
false /brushLeftArrow idef
true /brushNone idef
} {
/brushDashOffset idef
/brushDashArray idef
0 ne /brushRightArrow idef
0 ne /brushLeftArrow idef
/brushWidth idef
false /brushNone idef
} ifelse
} def

/SetCFg {
/fgblue idef
/fggreen idef
/fgred idef
} def

/SetCBg {
/bgblue idef
/bggreen idef
/bgred idef
} def

/SetF {
/printSize idef
/printFont idef
} def

/SetP {
dup type /nulltype eq {
pop true /patternNone idef
} {
dup -1 eq {
/patternGrayLevel idef
/patternString idef
} {
/patternGrayLevel idef
} ifelse
false /patternNone idef
} ifelse
} def

/BSpl {
0 begin
storexyn
newpath
n 1 gt {
0 0 0 0 0 0 1 1 true subspline
n 2 gt {
0 0 0 0 1 1 2 2 false subspline
1 1 n 3 sub {
/i exch def
i 1 sub dup i dup i 1 add dup i 2 add dup false subspline
} for
n 3 sub dup n 2 sub dup n 1 sub dup 2 copy false subspline
} if
n 2 sub dup n 1 sub dup 2 copy 2 copy false subspline
patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if
brushNone not { istroke } if
0 0 1 1 leftarrow
n 2 sub dup n 1 sub dup rightarrow
} if
end
} dup 0 4 dict put def

/Circ {
newpath
0 360 arc
patternNone not { ifill } if
brushNone not { istroke } if
} def

/CBSpl {
0 begin
dup 2 gt {
storexyn
newpath
n 1 sub dup 0 0 1 1 2 2 true subspline
1 1 n 3 sub {
/i exch def
i 1 sub dup i dup i 1 add dup i 2 add dup false subspline
} for
n 3 sub dup n 2 sub dup n 1 sub dup 0 0 false subspline
n 2 sub dup n 1 sub dup 0 0 1 1 false subspline
patternNone not { ifill } if
brushNone not { istroke } if
} {
Poly
} ifelse
end
} dup 0 4 dict put def

/Elli {
0 begin
newpath
4 2 roll
translate
scale
0 0 1 0 360 arc
patternNone not { ifill } if
brushNone not { istroke } if
end
} dup 0 1 dict put def

/Line {
0 begin
2 storexyn
newpath
x 0 get y 0 get moveto
x 1 get y 1 get lineto
brushNone not { istroke } if
0 0 1 1 leftarrow
0 0 1 1 rightarrow
end
} dup 0 4 dict put def

/MLine {
0 begin
storexyn
newpath
n 1 gt {
x 0 get y 0 get moveto
1 1 n 1 sub {
/i exch def
x i get y i get lineto
} for
patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if
brushNone not { istroke } if
0 0 1 1 leftarrow
n 2 sub dup n 1 sub dup rightarrow
} if
end
} dup 0 4 dict put def

/Poly {
3 1 roll
newpath
moveto
-1 add
{ lineto } repeat
closepath
patternNone not { ifill } if
brushNone not { istroke } if
} def

/Rect {
0 begin
/t exch def
/r exch def
/b exch def
/l exch def
newpath
l b moveto
l t lineto
r t lineto
r b lineto
closepath
patternNone not { ifill } if
brushNone not { istroke } if
end
} dup 0 4 dict put def

/Text {
ishow
} def

/idef {
dup where { pop pop pop } { exch def } ifelse
} def

/ifill {
0 begin
gsave
patternGrayLevel -1 ne {
fgred bgred fgred sub patternGrayLevel mul add
fggreen bggreen fggreen sub patternGrayLevel mul add
fgblue bgblue fgblue sub patternGrayLevel mul add setrgbcolor
eofill
} {
eoclip
originalCTM setmatrix
pathbbox /t exch def /r exch def /b exch def /l exch def
/w r l sub ceiling cvi def
/h t b sub ceiling cvi def
/imageByteWidth w 8 div ceiling cvi def
/imageHeight h def
bgred bggreen bgblue setrgbcolor
eofill
fgred fggreen fgblue setrgbcolor
w 0 gt h 0 gt and {
l b translate w h scale
w h true [w 0 0 h neg 0 h] { patternproc } imagemask
} if
} ifelse
grestore
end
} dup 0 8 dict put def

/istroke {
gsave
brushDashOffset -1 eq {
[] 0 setdash
1 setgray
} {
brushDashArray brushDashOffset setdash
fgred fggreen fgblue setrgbcolor
} ifelse
brushWidth setlinewidth
originalCTM setmatrix
stroke
grestore
} def

/ishow {
0 begin
gsave
fgred fggreen fgblue setrgbcolor
/fontDict printFont printSize scalefont dup setfont def
/descender fontDict begin 0 [FontBBox] 1 get FontMatrix end
transform exch pop def
/vertoffset 1 printSize sub descender sub def {
0 vertoffset moveto show
/vertoffset vertoffset printSize sub def
} forall
grestore
end
} dup 0 3 dict put def
/patternproc {
0 begin
/patternByteLength patternString length def
/patternHeight patternByteLength 8 mul sqrt cvi def
/patternWidth patternHeight def
/patternByteWidth patternWidth 8 idiv def
/imageByteMaxLength imageByteWidth imageHeight mul
stringLimit patternByteWidth sub min def
/imageMaxHeight imageByteMaxLength imageByteWidth idiv patternHeight idiv
patternHeight mul patternHeight max def
/imageHeight imageHeight imageMaxHeight sub store
/imageString imageByteWidth imageMaxHeight mul patternByteWidth add string def
0 1 imageMaxHeight 1 sub {
/y exch def
/patternRow y patternByteWidth mul patternByteLength mod def
/patternRowString patternString patternRow patternByteWidth getinterval def
/imageRow y imageByteWidth mul def
0 patternByteWidth imageByteWidth 1 sub {
/x exch def
imageString imageRow x add patternRowString putinterval
} for
} for
imageString
end
} dup 0 12 dict put def

/min {
dup 3 2 roll dup 4 3 roll lt { exch } if pop
} def

/max {
dup 3 2 roll dup 4 3 roll gt { exch } if pop
} def

/midpoint {
0 begin
/y1 exch def
/x1 exch def
/y0 exch def
/x0 exch def
x0 x1 add 2 div
y0 y1 add 2 div
end
} dup 0 4 dict put def

/thirdpoint {
0 begin
/y1 exch def
/x1 exch def
/y0 exch def
/x0 exch def
x0 2 mul x1 add 3 div
y0 2 mul y1 add 3 div
end
} dup 0 4 dict put def

/subspline {
0 begin
/movetoNeeded exch def
y exch get /y3 exch def
x exch get /x3 exch def
y exch get /y2 exch def
x exch get /x2 exch def
y exch get /y1 exch def
x exch get /x1 exch def
y exch get /y0 exch def
x exch get /x0 exch def
x1 y1 x2 y2 thirdpoint
/p1y exch def
/p1x exch def
x2 y2 x1 y1 thirdpoint
/p2y exch def
/p2x exch def
x1 y1 x0 y0 thirdpoint
p1x p1y midpoint
/p0y exch def
/p0x exch def
x2 y2 x3 y3 thirdpoint
p2x p2y midpoint
/p3y exch def
/p3x exch def
movetoNeeded { p0x p0y moveto } if
p1x p1y p2x p2y p3x p3y curveto
end
} dup 0 17 dict put def

/storexyn {
/n exch def
/y n array def
/x n array def
n 1 sub -1 0 {
/i exch def
y i 3 2 roll put
x i 3 2 roll put
} for
} def

/SSten {
fgred fggreen fgblue setrgbcolor
dup true exch 1 0 0 -1 0 6 -1 roll matrix astore
} def

/FSten {
dup 3 -1 roll dup 4 1 roll exch
newpath
0 0 moveto
dup 0 exch lineto
exch dup 3 1 roll exch lineto
0 lineto
closepath
bgred bggreen bgblue setrgbcolor
eofill
SSten
} def

/Rast {
exch dup 3 1 roll 1 0 0 -1 0 6 -1 roll matrix astore
} def


/arrowhead {
0 begin
transform originalCTM itransform
/taily exch def
/tailx exch def
transform originalCTM itransform
/tipy exch def
/tipx exch def
/dy tipy taily sub def
/dx tipx tailx sub def
/angle dx 0 ne dy 0 ne or { dy dx atan } { 90 } ifelse def
gsave
originalCTM setmatrix
tipx tipy translate
angle rotate
newpath
arrowHeight neg arrowWidth 2 div moveto
0 0 lineto
arrowHeight neg arrowWidth 2 div neg lineto
patternNone not {
originalCTM setmatrix
/padtip arrowHeight 2 exp 0.25 arrowWidth 2 exp mul add sqrt brushWidth mul
arrowWidth div def
/padtail brushWidth 2 div def
tipx tipy translate
angle rotate
padtip 0 translate
arrowHeight padtip add padtail add arrowHeight div dup scale
arrowheadpath
ifill
} if
brushNone not {
originalCTM setmatrix
tipx tipy translate
angle rotate
arrowheadpath
istroke
} if
grestore
end
} dup 0 9 dict put def

/arrowheadpath {
newpath
arrowHeight neg arrowWidth 2 div moveto
0 0 lineto
arrowHeight neg arrowWidth 2 div neg lineto
} def

/leftarrow {
0 begin
y exch get /taily exch def
x exch get /tailx exch def
y exch get /tipy exch def
x exch get /tipx exch def
brushLeftArrow { tipx tipy tailx taily arrowhead } if
end
} dup 0 4 dict put def

/rightarrow {
0 begin
y exch get /tipy exch def
x exch get /tipx exch def
y exch get /taily exch def
x exch get /tailx exch def
brushRightArrow { tipx tipy tailx taily arrowhead } if
end
} dup 0 4 dict put def




Begin
[ 0.799705 0 0 0.799705 0 0 ] concat
/originalCTM matrix currentmatrix def

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 433.125 504.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 433.125 552.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 265.125 504.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 265.125 552.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 137.125 504.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 137.125 552.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 321.125 600.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 489.125 600.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 193.125 600.625 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 144 410 ] concat
453 529 448 32 Elli
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 486.625 598.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 505.272 606.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 318.625 598.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 337.272 606.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 190.625 598.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 209.272 606.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 134.625 502.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 153.272 510.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.125 -0 -0 0.125 262.625 502.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.125 -0 -0 0.125 430.625 502.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 362.875 611.625 ] concat
617 99 16 16 Elli
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 378.875 611.625 ] concat
617 99 16 16 Elli
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 133.5 439.5 ] concat
117 369 181 369 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 261.5 439.5 ] concat
117 369 181 369 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 429.5 439.5 ] concat
117 369 181 369 Line
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 179.5 532 ] concat
[
(Procr)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 307.5 532 ] concat
[
(Procr)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 475.5 532 ] concat
[
(Procr)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 235.5 628 ] concat
[
(Mem)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 363.5 628 ] concat
[
(Mem)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 531.5 628 ] concat
[
(Mem)
] Text
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.125 -0 -0 0.125 346.875 611.625 ] concat
617 99 16 16 Elli
End

Begin %I BSpl
1 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 19.5 393.5 ] concat
457 333
473 381
441 365
457 413
4 BSpl
End

Begin %I BSpl
1 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 147.5 393.5 ] concat
457 333
473 381
441 365
457 413
4 BSpl
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 360 555 ] concat
[
(Remote)
(memory)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 232 555 ] concat
[
(Local)
(memory)
] Text
End

Begin %I BSpl
1 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 0.5 -0 -0 0.5 315.5 393.5 ] concat
457 333
473 381
441 365
457 413
4 BSpl
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 528 555 ] concat
[
(Remote)
(memory)
] Text
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0.75 SetP
[ 0.125 -0 -0 0.125 134.625 550.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
none SetB %I b n
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.084375 -0 -0 0.084375 153.272 558.534 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.125 -0 -0 0.125 262.625 550.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I CBSpl
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.125 -0 -0 0.125 430.625 550.125 ] concat
267 271
267 335
331 335
587 335
651 335
651 271
651 143
651 79
587 79
331 79
267 79
267 143
12 CBSpl
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 177 580 ] concat
[
(Cache)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 305 580 ] concat
[
(Cache)
] Text
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 473 580 ] concat
[
(Cache)
] Text
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 60 364 ] concat
132 180 132 196 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 188 364 ] concat
132 180 132 196 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 356 364 ] concat
132 180 132 196 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 49 237 ] concat
143 355 143 435 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 49 237 ] concat
439 355 439 427 Line
End

Begin %I Line
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 49 237 ] concat
271 355 271 427 Line
End

Begin %I Elli
1 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
1 SetP
[ 0.5 -0 -0 0.5 141.5 407.5 ] concat
453 529 448 32 Elli
End

Begin %I Text
0 0 0 SetCFg
Times-Roman 12 SetF
[ 1 0 0 1 306.5 676 ] concat
[
(Interconnection Network)
] Text
End

End %I eop

showpage


end
%%EndDocument
 @endspecial 295 598 a Fo(Figure)11 b(1:)18 b(Scalable)12 b(shared-memory)f
(multiprocessor)h(architecture.)4 719 y(partitioning)g(data)i(and)g
(computations)f(in)g(a)h(way)f(that)h(minimizes)f(interprocessor)g
(communications.)4 773 y(On)f(SSMMs,)h(processors)f(communicate)f(through)g
(shared)g(memory)m(,)h(and)f(the)h(cost)g(of)f(interprocessor)4
827 y(communications)h(\(i.e.,)i(remote)e(memory)f(access\))j(is)f
(relatively)e(inexpensive.)19 b(W)l(e)13 b(show)g(that)f(cache)4
881 y(af)o(\256nity)m(,)i(memory)f(contention)h(and)g(false)g(sharing)g(are)g
(additional)g(factors)g(that)f(must)i(be)f(considered)4 935
y(in)i(the)g(selection)g(of)f(data)h(distributions.)28 b(Furthermore,)16
b(the)g(presence)g(of)f(a)h(single)g(shared)g(address)4 989
y(space)i(allows)g(\257exibility)f(in)g(the)h(selection)g(of)f(a)h
(computation)e(partition.)33 b(Speci\256cally)m(,)19 b(we)f(show)4
1044 y(that)h(relaxing)g(the)h(commonly)e(used)i(owner)o(-computes)f(rule)g
([15)o(])h(has)g(performance)e(advantages.)4 1098 y(W)l(e)d(present)g
(experimental)f(results)i(to)e(support)h(our)f(conclusions)i(using)f(three)f
(applications)h(on)g(two)4 1152 y(SSMMs,)e(the)g(Hector)f(and)g(the)g(KSR1)h
(multiprocessors.)77 1228 y(The)g(remainder)f(of)g(this)g(paper)g(is)h(or)o
(ganized)f(as)h(follows.)18 b(Section)12 b(2)h(presents)f(an)h(overview)f
(data)4 1282 y(distributions.)35 b(Section)18 b(3)g(describes)h(the)f
(factors)f(that)h(impact)g(on)g(the)h(selection)f(of)g(computation)4
1336 y(and)i(data)g(partitions)f(on)g(SSMMs.)41 b(Section)19
b(4)h(gives)g(experimental)f(evidence)h(of)f(the)h(impact)f(of)4
1390 y(cache)e(af)o(\256nity)e(and)h(false)g(sharing)g(on)g(the)g(choice)h
(of)e(data)h(partitions.)29 b(Section)16 b(5)g(presents)h(results)4
1444 y(to)d(show)g(that)g(the)g(\257exibility)f(in)h(selecting)h(the)f
(computation)f(partitioning)g(can)h(be)g(used)h(to)f(improve)4
1498 y(performance.)j(Section)9 b(6)i(reviews)f(related)g(work.)17
b(Finally)m(,)11 b(Section)e(7)i(presents)f(concluding)g(remarks)4
1553 y(and)j(directions)e(for)h(future)f(work.)4 1734 y Fn(2)71
b(Data)19 b(Distributions)4 1861 y Fo(Data)10 b(distribution)f([15)o(,)i(16])
e(is)i(achieved)f(by)g(specifying)f(a)h(partitioning)f(scheme)h(for)f(each)i
(array)e(in)h(the)4 1915 y(program)h(and)i(by)f(specifying)g(a)g(processor)h
(geometry)e(to)i(which)f(array)g(partitions)f(map.)18 b(A)13
b(processor)4 1970 y(geometry)g(is)i(an)f Fj(n)p Fo(-dimensional)f(Cartesian)
h(grid)f(of)h(virtual)f(processors)h Fi(\()p Fj(V)1385 1977
y Fm(0)1404 1970 y Fj(;)8 b(V)1454 1977 y Fm(1)1473 1970 y
Fj(;)g Fh(\001)g(\001)g(\001)g Fj(;)g(V)1611 1977 y Fg(n)p
Ff(\000)p Fm(1)1679 1970 y Fi(\))p Fo(,)15 b(where)4 2024 y
Fj(V)32 2031 y Fg(i)63 2024 y Fo(is)i(the)f(number)g(of)g(processors)h(in)f
(the)g Fj(i)793 2006 y Fg(th)844 2024 y Fo(dimension)g(of)g(the)h(grid,)g
(and)f Fj(V)1430 2031 y Fm(0)1463 2024 y Fh(\002)d Fj(V)1543
2031 y Fm(1)1575 2024 y Fh(\002)g(\001)8 b(\001)g(\001)14 b(\002)f
Fj(V)1779 2031 y Fg(n)p Ff(\000)p Fm(1)4 2078 y Fo(=)i Fj(P)7
b Fo(,)16 b(the)f(total)f(number)h(of)f(processors.)26 b(A)15
b(partitioning)e(scheme)i(assigns)h(a)f Fe(partitioning)f(attribute)4
2132 y Fo(to)k(each)g(dimension)g(the)g(array)m(.)34 b(There)18
b(are)g(four)f(partitioning)f(attributes.)35 b(The)18 b Fd(Block)g
Fo(attribute)4 2186 y(divides)f(the)g(corresponding)g(dimension)g(of)f(the)h
(array)g(in)g(approximately)f(equal)h(size)h(blocks)f(such)4
2240 y(that)j(a)g(processor)g(owns)h(a)f(contiguous)g(range)g(of)f(that)h
(dimension)g(of)g(the)g(array)m(.)41 b(The)20 b Fd(Cyclic)4
2295 y Fo(attribute)11 b(divides)h(the)h(corresponding)e(array)g(dimension)h
(by)g(distributing)f(the)h(array)f(elements)i(in)f(this)4 2349
y(dimension)g(to)g(processors)h(in)f(a)h(round-robin)d(fashion.)18
b(The)13 b Fd(BlockCyclic)f Fo(attribute)f(\256rst)h(groups)4
2403 y(array)f(elements)h(in)f(the)g(corresponding)g(dimension)g(in)g
(contiguous)g(blocks)h(of)f(a)h(given)f(size,)h(and)g(then)4
2457 y(assigns)k(the)f(blocks)f(to)h(processors)g(in)g(a)g(round-robin)d
(fashion.)26 b(The)15 b(block)f(size,)j(called)d(the)h Fe(block-)4
2511 y(cyclic)10 b(factor)p Fo(,)h(is)e(supplied)h(by)f(the)h(programmer)m(.)
16 b(Finally)m(,)9 b(the)h Fd(*)f Fo(attribute)g(is)h(used)g(to)f(indicate)g
(that)h(the)4 2565 y(corresponding)f(dimension)g(of)g(the)h(array)f(is)h(not)
f(distributed.)17 b(The)10 b(processor)g(geometry)f(on)g(which)h(the)4
2620 y(array)h(is)i(mapped)e(determines)h(the)g(number)f(of)g(processors)i
(assigned)f(to)g(each)g(distributed)f(dimension)4 2674 y(of)h(the)f(array)m
(.)18 b(For)11 b(example,)h(distributing)e(a)i(two)g(dimensional)f(array)h
(using)f(the)h Fd(\(Block,Block\))4 2728 y Fo(attributes)g(onto)h(a)g(two)f
(dimensional)h(processor)f(geometry)g(of)h(\(2,4\),)f(distributes)h(the)f
(array)h(on)f(to)h(the)4 2782 y(8)k(processors,)i(assigning)f(2)f(processors)
g(to)g(the)g(\256rst)g(dimension)g(and)g(4)g(processors)g(to)g(the)g(second)4
2836 y(dimension.)p eop
%%Page: 3 3
3 2 bop 4 -21 a Fn(3)71 b(Performance)21 b(Factors)4 106 y
Fo(The)15 b(main)g(factor)f(that)g(af)o(fects)h(the)f(performance)g(of)g(a)h
(parallel)f(application)g(on)h(a)g(DMM)g(is)g(the)g(rel-)4
160 y(atively)i(high)f(cost)h(of)g(interprocessor)f(communication.)30
b(For)17 b(example,)h(the)f(latency)f(for)g(a)h(remote)4 215
y(memory)e(access)i(on)e(the)h(CM5)g(multiprocessor)f(is)h(approximately)e
(2560)h(processor)h(cycles)1699 196 y Fm(2)1718 215 y Fo(.)28
b(This)4 269 y(necessitates)16 b(the)e(selection)h(of)f(computation)f(and)h
(data)h(partitions)f(that)g(minimize)f(the)i(cost)f(of)g(com-)4
323 y(munication.)27 b(In)15 b(contrast,)h(on)f(SSMMs,)i(processors)f
(communicate)e(through)h(shared)g(memory)g(and)4 377 y(the)j(cost)h(of)f
(remote)f(memory)h(access)h(is)g(relatively)e(small.)36 b(For)17
b(example,)j(the)f(cost)f(of)g(a)g(remote)4 431 y(read)11 b(operation)g(on)g
(the)h(KSR1)f(is)h(approximately)e(170)h(processor)g(cycles)h([24].)17
b(Consequently)m(,)12 b(other)4 485 y(factors)h(come)f(into)h(play)g(in)f
(the)h(selection)g(of)f(computation)g(and)h(data)g(partitions.)19
b(In)13 b(this)g(section)g(we)4 540 y(elaborate)h(on)g(these)g(factors)g(and)
g(on)g(how)g(they)g(af)o(fect)g(performance,)f(and)h(consequently)m(,)h(af)o
(fect)f(the)4 594 y(choice)f(of)f(data)g(and)g(computation)g(partitions.)4
755 y Fc(3.1)58 b(Cache)14 b(Af\256nity)4 853 y Fo(Caches)j(are)e(used)h(in)f
(SSMMs)h(to)g(reduce)f(ef)o(fective)g(memory)f(access)j(time)e(and)h(reduce)f
(contention)4 907 y(in)e(the)h(interconnection)e(network.)21
b(Data)14 b(is)g(transferred)e(between)i(cache)g(and)f(memory)g(in)g(units)g
(of)h(a)4 961 y Fe(cache)g(line)p Fo(,)h(typically)e(a)h(multiple)f(of)g(the)
h(processor)g(word)f(size.)24 b Fe(Spatial)13 b(r)n(euse)i
Fo(occurs)e(when)h(other)4 1015 y(words)h(on)g(the)g(same)g(line)g(are)g
(used)g(by)g(the)g(processor)g(before)f(the)h(line)g(is)g(\257ushed)g(from)f
(the)h(cache.)4 1070 y(Analogously)m(,)g Fe(temporal)f(r)n(euse)i
Fo(occurs)e(when)g(data)h(on)f(a)g(cache)h(line)f(is)h(used)g(again)f(before)
g(the)g(line)4 1124 y(is)i(evicted)g(from)e(the)i(cache.)29
b(The)16 b(performance)e(of)i(an)f(application)h(depends)g(to)f(a)h(lar)o(ge)
f(extent)h(on)4 1178 y(the)g(ability)g(of)g(the)g(caches)h(to)f(exploit)g
(spatial)h(and)f(temporal)f(reuse.)31 b(In)16 b(some)g(cases,)j(this)d(may)h
(be)4 1232 y(dif)o(\256cult)9 b(because)i(of)f(the)g(limited)f(capacity)h
(and)g(associativity)g(of)g(caches.)18 b(Data)10 b(brought)f(into)h(a)g
(cache)4 1286 y(by)16 b(a)g(reference)f(or)h(a)g(prefetch)f(may)h(be)g
(evicted)f(before)h(being)f(used)h(or)g(reused,)h(because)g(of)e(either)4
1340 y(a)i(capacity)g(or)g(a)g(con\257ict)f(miss)i(caused)f(by)g(a)g
(subsequent)h(reference.)31 b(Cache)18 b(misses)f(on)g(SSMMs)4
1395 y(adversely)f(af)o(fect)g(performance,)g(since)h(evicted)f(data)g(must)g
(be)g(retrieved)f(from)g(its)i(home)e(memory)m(,)4 1449 y(which)k(may)g(be)g
(remote)f(to)h(the)f(processor)m(.)38 b(Caches)20 b(play)f(less)h(of)e(an)h
(important)f(role)g(in)h(DMMs)4 1503 y(because)g(cache)f(misses)h(result)e
(exclusively)h(in)f(local)h(memory)f(accesses,)k(which)d(are)g(inexpensive)4
1557 y(relative)12 b(to)g(interprocessor)g(communications.)4
1718 y Fc(3.2)58 b(False)14 b(Sharing)4 1816 y Fo(In)g(SSMMs)h(data)f(on)h
(the)f(same)h(cache)g(line)f(may)g(be)h(shared)f(by)h(more)e(than)i(one)f
(processor)n(,)h(and)g(the)4 1870 y(line)j(may)g(exit)g(in)g(more)g(than)g
(one)g(processor)r(')m(s)g(cache)h(at)f(the)g(same)h(time.)35
b(Hardware)18 b(is)g(used)h(to)4 1925 y(maintain)13 b(the)f(consistency)i(of)
e(the)h(multiple)g(copies)g(of)f(the)h(line,)h(typically)e(using)h(a)g
(write-invalidate)4 1979 y(protocol)e([24,)h(14].)18 b Fe(T)m(rue)12
b(sharing)g Fo(occurs)g(when)g(two)g(or)f(more)g(processors)i(access)g(the)f
(same)g(data)g(on)4 2033 y(a)k(cache)f(line,)i(and)e(it)g(re\257ects)g
(necessary)h(data)f(communications)g(in)g(an)g(application.)27
b(On)15 b(the)g(other)4 2087 y(hand,)h Fe(false)e(sharing)h
Fo(occurs)f(when)h(two)f(processors)h(access)h(dif)o(ferent)d(pieces)i(of)f
(data)h(on)f(the)g(same)4 2141 y(cache)e(line.)18 b(If)11 b(processors)h
(write)g(to)f(the)h(same)g(cache)g(line,)g(the)g(cache)g(consistency)h
(hardware)e(causes)4 2195 y(the)j(cache)g(line)g(to)g(be)g(transferred)f
(back)h(and)g(forth)f(between)h(processors)g(leading)g(to)g(a)g
(\252ping-pong\272)4 2250 y(ef)o(fect)h([8)o(].)27 b(False)16
b(sharing)f(causes)h(extensive)g(invalidation)e(traf)o(\256c)g(and)i(can)f
(considerably)g(degrade)4 2304 y(performance.)i(False)c(sharing)f(is)h
(non-existent)e(on)i(DMMs.)4 2465 y Fc(3.3)58 b(Memory)14 b(Contention)4
2563 y Fo(Memory)i(contention)g(occurs)g(when)g(many)g(processors)h(access)h
(data)e(in)g(a)g(single)h(memory)e(module)4 2617 y(at)j(the)g(same)h(time.)35
b(Since)18 b(the)g(communication)f(protocol)g(in)h(SSMMs)g(is)g(receiver)o
(-initiated,)h(and)4 2671 y(transfers)i(data)g(in)f(units)h(of)g(relatively)f
(small)h(cache)g(lines,)j(a)d(lar)o(ge)g(number)f(of)h(requests)g(to)g(the)4
2725 y(same)12 b(memory)f(can)h(over\257ow)f(memory)g(buf)o(fers)g(and)h
(cause)g(excessive)h(delays)f(in)g(memory)e(response)4 2780
y(time)20 b([13].)42 b(Contention)20 b(has)h(been)g(considered)g(less)g(of)f
(a)h(performance)e(bottleneck)h(on)h(DMMs)p 4 2825 737 2 v
62 2855 a Fl(2)79 2870 y Fk(Calculated)10 b(based)h(on)f(the)g(elapsed)h
(time)f(for)g(a)g(send-reply)g(message)i(of)e(128)g(bytes)g([19)o(].)p
eop
%%Page: 4 4
4 3 bop 4 -27 a Fo(because)16 b(a)g(sender)o(-initiated)e(communication)h
(protocol)f(is)i(employed,)g(and)g(because)g(programmers)4
27 y(typically)f(communicate)f(data)i(in)f(lar)o(ge)g(infrequent)f(messages.)
28 b(Applications)15 b(on)g(DMMs)h(also)f(use)4 82 y(collective)d
(communications)g([15)o(])g(that)h(further)e(reduce)h(contention.)4
243 y Fc(3.4)58 b(Over)o(head)14 b(of)g(Parallelism)4 341 y
Fo(In)g(DMM,)i(synchronization)e(is)h(achieved)f(through)g(data)g
(communication.)24 b(However)n(,)15 b(on)g(SSMMs,)4 395 y(synchronization)9
b(is)h(explicit)e(and)i(is)g(independent)f(of)f(data)i(communication.)16
b(The)10 b(resulting)f(overhead)4 449 y(can)14 b(become)f(a)h(performance)e
(bottleneck)h([27)o(],)h(and)f(must)h(be)f(minimized.)21 b(The)14
b(performance)e(of)h(an)4 503 y(application)e(is)h(also)h(af)o(fected)e(by)h
(the)f(overhead)h(involved)f(in)h(parallelizing)e(loops,)j(manifested)e(in)h
(the)4 557 y(form)h(of)h(computation)f(partitioning)f(tests)j([25)o(].)23
b(These)15 b(tests)g(can)f(be)g(eliminated)g(in)f(some)i(cases)g(by)4
612 y(compiler)g(analysis,)i(but)d(when)i(not)f(possible,)h(can)g(degrade)f
(performance.)26 b(This)15 b(overhead)g(though)4 666 y(also)d(present)g(in)f
(the)h(case)g(of)f(DMMs,)j(is)e(not)f(considered)h(signi\256cant)f(because)h
(of)g(the)f(predominantly)4 720 y(high)h(cost)h(of)f(remote)g(memory)f
(access.)4 902 y Fn(4)71 b(Impact)19 b(on)f(Data)h(Distribution)4
1029 y Fo(In)e(this)g(section)g(we)g(use)g(two)g(applications,)h
Fd(Multigrid)e Fo(and)h Fd(Tred2)p Fo(,)h(to)f(illustrate)f(the)h(impact)4
1083 y(of)f(cache)h(af)o(\256nity)f(and)h(false)g(sharing)f(on)h(the)f
(choice)h(of)f(a)h(data)g(distribution.)30 b(The)17 b(KSR1)f(system)4
1137 y(is)f(used)f(because)h(of)f(its)h(lar)o(ge)f(cache)g(size,)i(and)e
(because)h(of)f(the)g(presence)h(of)f(monitoring)e(hardware)4
1191 y(that)i(enables)h(the)g(measurement)f(of)g(the)g(number)g(of)f
(non-local)h(memory)g(accesses)i(and)e(the)g(number)4 1245
y(of)e(caches)h(misses)h(for)d(a)i(processor)m(.)77 1321 y(The)j(KSR1)e(is)h
(a)g(Cache)g(only)g(Memory)f(Architecture)g(\(COMA\))g(con\256gured)g(as)h(a)
g(hierarchy)f(of)4 1375 y(slotted)c(rings)g(with)g(processing)g(cells)h(on)f
(the)g(leaf-level)f(rings.)18 b(The)10 b(local)g(portion)g(of)f(shared)i
(memory)4 1429 y(associated)g(with)e(a)i(processor)e(is)i(or)o(ganized)e(as)i
(a)f(cache.)18 b(There)10 b(is)g(no)g(home)g(location)f(for)g(data,)i(rather)
n(,)4 1483 y(data)k(may)g(exist)f(in)h(more)f(than)h(one)f(local)h(memory)m
(.)24 b(The)16 b(hardware)e(maintains)g(the)h(consistency)g(of)4
1538 y(possible)e(multiple)e(copies)i(of)f(the)g(data.)77 1613
y(The)e(KSR1)g(implicitly)e(implements)i(the)f(owner)o(-computes)g(rule,)h
(since)g(data)g(written)f(by)g(a)h(proces-)4 1667 y(sor)j(must)f(exclusively)
g(reside)h(in)f(the)h(processor)r(')m(s)f(local)g(portion)g(of)g(the)g
(shared)h(memory)m(.)k(Hardware)4 1722 y(automatically)j(migrates)g(data)h
(to)g(the)f(processor)h(that)f(requests)h(the)g(data)f(in)h(units)g(of)f
Fe(subpages)p Fo(.)4 1776 y(Hence,)13 b(the)f(computation)g(partitioning)e
(of)i(a)g(loop)g(dictates)h(the)f(residence)g(of)g(a)g(data)h(item)e(and)i
(hence)4 1830 y(the)k(distribution)f(of)h(the)g(arrays)g(in)g(the)g(loop.)33
b(Data)17 b(which)g(is)h(read)f(by)g(the)g(processors)h(may)f(exist)4
1884 y(in)e(multiple)e(local)i(memories,)g(and)g(read)f(requests)h(to)g(this)
g(data)f(from)g(dif)o(ferent)f(processors)i(may)g(be)4 1938
y(satis\256ed)e(from)e(dif)o(ferent)g(portions)h(of)g(the)g(shared)h(memory)m
(.)4 2099 y Fc(4.1)58 b(Cache-Conscious)13 b(Data)i(Distribution)4
2197 y Fo(The)j Fd(Multigrid)e Fo(application)g(from)g(the)h(NAS)f(suite)h
(of)g(benchmarks)f(illustrates)h(how)g(data)g(dis-)4 2252 y(tributions)d
(must)h(be)g(cache-conscious.)27 b Fd(Multigrid)14 b Fo(is)h(a)g(three)g
(dimensional)f(solver)h(calculating)4 2306 y(the)j(potential)f(\256eld)h(on)f
(a)h(cubical)g(grid.)34 b(W)l(e)18 b(focus)g(on)f(the)h(subroutine)f
Fd(psinv)h Fo(which)f(uses)i(two)4 2360 y(3-dimensional)13
b(arrays)h Fj(U)20 b Fo(and)14 b Fj(R)p Fo(.)25 b(The)14 b(subroutine)g
(mainly)f(performs)h(the)g(following)f(computation)4 2414 y(inside)i(a)h
(triply)e(nested)h(loop:)23 b Fj(U)5 b Fi(\()p Fj(i;)23 b(j;)h(k)r
Fi(\))15 b(+)j(=)33 b Fj(\013)p Fi(\()15 b Fj(R)p Fi(\()p Fj(f)5
b Fi(\()p Fj(i)p Fi(\))p Fj(;)24 b(g)r Fi(\()p Fj(j)s Fi(\))p
Fj(;)f(h)p Fi(\()p Fj(k)r Fi(\)\)\))p Fo(,)16 b(where)f Fj(f)5
b Fi(\()p Fj(i)p Fi(\))15 b Fo(=)h Fj(i)c Fh(\000)g Fo(1,)4
2468 y Fj(i)18 b Fo(or)g Fj(i)13 b Fi(+)i Fo(1,)20 b(as)e(are)g(the)g
(functions)g Fj(g)i Fo(and)e Fj(h)p Fo(.)36 b(The)18 b(loop)g(nest)g(is)h
(fully)e(parallel.)35 b(The)18 b(application)4 2522 y(has)e(nearest)g
(neighbor)e(communications)h(along)g(all)g(three)g(dimensions,)i(which)e(is)h
(typical)f(of)g(many)4 2577 y(scienti\256c)d(applications.)77
2652 y(In)d(this)g(application,)g(we)g(choose)h(not)e(to)h(parallelize)f(the)
h(innermost)g(loop)f(to)h(avoid)g(cache)g(line)g(false)4 2706
y(sharing)k(and)g(cache)h(interference;)e(successive)j(iterations)e(of)f
(this)i(loop)f(access)h(successive)h(elements)4 2761 y(on)h(the)f(same)i
(cache)f(line.)28 b(Hence)16 b(we)g(use)g(a)g(two)g(dimensional)f(grid)g(for)
g(the)h(processor)f(geometry)m(.)4 2815 y(Since)10 b(the)g(application)g(has)
h(nearest)f(neighbor)f(communications,)i Fd(Block)f Fo(distribution)f
(performs)g(the)4 2869 y(best.)18 b(The)10 b(restriction)e(of)h(the)h
(innermost)f(loop)g(to)g(be)h(sequential)f(requires)g(the)g(arrays)h(to)f(be)
h(distributed)p eop
%%Page: 5 5
5 4 bop 503 532 a @beginspecial 50 @llx 50 @lly 410 @urx 302
@ury 2057 @rwi @setspecial
%%BeginDocument: mg.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.100 0.100 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 473 M
2817 0 V
LTb
600 473 M
63 0 V
2754 0 R
-63 0 V
540 473 M
(96) Rshow
LTa
600 916 M
2817 0 V
LTb
600 916 M
63 0 V
2754 0 R
-63 0 V
540 916 M
(98) Rshow
LTa
600 1360 M
2817 0 V
LTb
600 1360 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(100) Rshow
LTa
600 1804 M
2817 0 V
LTb
600 1804 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(102) Rshow
LTa
600 2247 M
2817 0 V
LTb
600 2247 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(104) Rshow
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(\(16,1\)) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(8,2\)) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(4,4\)) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(2,8\)) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(1,16\)) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
340 1360 M
currentpoint gsave translate 90 rotate 0 0 M
(Normalized Execution Time \(w.r.t \(16,1\)\)) Cshow
grestore
2008 51 M
(Processor Geometry - 16 Processors                    ) Cshow
LT1
1654 2106 M
(160x160x160) Rshow
1714 2106 M
180 0 V
1523 341 R
2713 2048 L
2009 1826 L
1304 939 L
600 1360 L
1774 2106 A
3417 2447 A
2713 2048 A
2009 1826 A
1304 939 A
600 1360 A
LT2
1654 2006 M
(144x144x144) Rshow
1714 2006 M
180 0 V
1523 241 R
2713 1715 L
-705 -67 V
1304 850 L
600 1360 L
1774 2006 B
3417 2247 B
2713 1715 B
2009 1648 B
1304 850 B
600 1360 B
LT4
1654 1906 M
(64x64x64) Rshow
1714 1906 M
180 0 V
1523 -590 R
2713 1160 L
2009 340 L
1304 495 L
600 1360 L
1774 1906 T
3417 1316 T
2713 1160 T
2009 340 T
1304 495 T
600 1360 T
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 377 612 a Fo(Figure)11 b(2:)18 b(Normalized)11
b(Execution)i(time)f(of)g Fd(Multigrid)p Fo(.)4 733 y(with)17
b Fd(\(*,Block,Block\))f Fo(since)h(the)g(arrays)g(are)g(assumed)h(to)f(be)h
(stored)f(using)g(column)f(major)4 787 y(ordering.)31 b(W)n(ith)16
b(16)h(processors,)h(it)f(is)g(possible)g(to)g(choose)g(one)g(of)f(the)h
(\(16,1\),)h(\(8,2\),)f(\(4,4\),)h(\(2,8\))4 841 y(and)d(\(1,16\))g
(processor)g(geometries.)27 b(The)15 b(choice)h(of)e(the)i(processor)f
(geometry)f(af)o(fects)h(the)g(number)4 895 y(of)h(processors)g(that)g
(execute)g(each)g(parallel)f(loop.)29 b(For)15 b(example,)i(a)f(processor)g
(geometry)f(of)h(\(8,2\),)4 949 y(implies)11 b(8)g(processors)h(assigned)g
(to)e(the)i(inner)e(parallel)h(loop)g(and)g(2)g(processors)g(assigned)h(to)f
(the)g(outer)4 1004 y(parallel)h(loop.)77 1079 y(Figure)17
b(2)g(shows)h(the)f(execution)g(time)g(of)g(the)h(application)e(for)h
(various)g(processor)g(geometries)4 1133 y(with)e(the)g Fd(\(*,Block,Block\))
e Fo(distribution)h(for)g(the)h(arrays)g(on)g(the)g(KSR1)f(with)h(16)g
(processors,)4 1188 y(normalized)d(with)h(respect)g(to)g(the)g(\(16,1\))f
(processor)h(geometry)m(.)19 b(For)12 b(a)h(small)g(data)g(size)h
(\(64x64x64\),)4 1242 y(execution)22 b(time)f(is)h(minimized)e(by)i(a)g
(distribution)e(with)h(equal)h(number)f(of)g(processors)h(in)f(each)4
1296 y(dimension,)15 b(i.e.,)i(\(4,4\).)24 b(This)16 b(is)f(the)f(same)h
(distribution)e(scheme)j(suggested)f(in)f(the)h(Syracuse)f(High)4
1350 y(Performance)9 b(Fortran)h(applications)g(suite)772 1332
y Fm(3)801 1350 y Fo(for)g(DMMs.)19 b(However)n(,)11 b(when)f(the)h(data)g
(size)g(is)g(lar)o(ge,)g(the)4 1404 y(processor)h(geometry)f(\(4,4\))h(no)g
(longer)f(performs)g(the)h(best.)19 b(The)12 b(execution)g(time)g(is)g
(minimized)f(with)4 1458 y(a)i(processor)f(geometry)g(of)g(\(8,2\).)77
1534 y(The)20 b(impact)e(of)h(processor)g(geometry)f(on)g(performance)g(is)h
(due)g(to)g(cache)g(af)o(\256nity)m(,)h(as)g(can)f(be)4 1588
y(deduced)12 b(from)f(Figures)h(3)g(and)g(4.)19 b(Figure)11
b(3)h(shows)h(the)f(measured)g(number)g(of)f(cache)i(lines)f(accessed)4
1642 y(from)17 b(remote)g(memory)f(modules,)j(normalized)e(with)g(respect)h
(to)f(the)h(processor)f(geometry)g(\(16,1\).)4 1697 y(The)h(number)e(of)h
(remote)f(memory)g(accesses)j(is)e(minimal)g(when)g(the)g(processor)g
(geometry)f(is)h(\(4,4\))4 1751 y(for)h(all)g(data)h(sizes.)38
b(Figure)17 b(4)i(shows)g(the)g(average)f(measured)h(number)e(of)i(cache)g
(misses)g(from)f(a)4 1805 y(processor)c(cache,)h(again)e(normalized)g(with)g
(respect)h(to)g(the)f(processor)h(geometry)f(\(16,1\).)21 b(When)14
b(the)4 1859 y(data)e(size)h(is)f(small)g(\(64x64x64\),)g(the)g(data)g(used)g
(by)g(a)h(processor)f(\256ts)g(into)f(the)h(256k)g(processor)g(cache)4
1913 y(and)19 b(the)g(misses)h(from)e(the)h(cache)h(in)f(this)g(case)h
(re\257ect)f(remote)f(memory)g(accesses)j(that)e(occur)g(in)4
1967 y(the)13 b(parallel)g(program.)19 b(Hence,)14 b(the)f(predominant)f
(factor)h(af)o(fecting)f(performance)g(is)h(interprocessor)4
2022 y(communication,)f(and)g(the)h(best)f(performance)g(is)g(attained)g
(using)h(the)f(\(4,4\))g(geometry)m(.)77 2097 y(However)n(,)17
b(when)f(the)g(arrays)g(are)g(relatively)f(lar)o(ge)h(\(144x144x144\),)g(the)
g(cache)g(capacity)h(is)f(no)4 2151 y(longer)g(suf)o(\256cient)h(to)g(hold)f
(data)i(from)d(successive)k(iterations)d(of)h(the)g(outer)f(parallel)h(loop,)
h(and)f(the)4 2206 y(number)10 b(of)h(cache)g(misses)h(increases.)19
b(When)11 b(the)g(number)f(of)h(processors)g(assigned)h(to)f(the)g(outer)f
(loop)4 2260 y(increases,)j(the)f(number)f(of)h(misses)h(from)d(the)i(cache)h
(also)f(increases.)19 b(The)12 b(\(4,4\))g(processor)g(geometry)4
2314 y(minimizes)d(the)f(amount)h(of)f(remote)g(memory)g(access,)k(but)c(the)
h(\(16,1\))f(processor)h(geometry)f(minimizes)4 2368 y(the)k(amount)f(of)g
(cache)h(misses.)19 b(The)12 b(distribution)e(with)i(\(8,2\))f(processor)g
(geometry)g(strikes)h(a)g(balance)4 2422 y(between)17 b(the)g(cost)g(of)g
(remote)f(memory)g(access)i(and)f(the)g(cost)g(of)g(cache)g(misses,)i
(resulting)e(in)f(best)4 2476 y(overall)c(performance,)g(in)g(spite)g(of)g
(higher)g(interprocessor)g(communication)f(cost.)4 2638 y Fc(4.2)58
b(False)14 b(Sharing)g(Conscious)g(Data)h(Distribution)4 2736
y Fo(The)d(programs)f Fd(Tred2)h Fo(\(which)f(is)h(part)f(of)g(Eispack\),)i
Fd(mdg)p Fo(,)f(and)g Fd(trfd)f Fo(\(which)g(are)h(both)f(part)h(of)f(the)4
2790 y(Perfect)f(Club)h(Benchmark)f(Suite\))g(exhibit)h(parallelism)f(which)h
(result)f(in)h(considerable)g(false)g(sharing.)p 4 2835 737
2 v 62 2865 a Fl(3)79 2880 y Fk(http://www)m(.npac.syr)n(.edu/hpfa/)c(.)p
eop
%%Page: 6 6
6 5 bop 47 586 a @beginspecial 50 @llx 50 @lly 230 @urx 176
@ury 2057 @rwi @setspecial
%%BeginDocument: spmiss.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.050 0.050 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
540 251 M
(30) Rshow
LTa
600 568 M
2817 0 V
LTb
600 568 M
63 0 V
2754 0 R
-63 0 V
540 568 M
(40) Rshow
LTa
600 885 M
2817 0 V
LTb
600 885 M
63 0 V
2754 0 R
-63 0 V
540 885 M
(50) Rshow
LTa
600 1202 M
2817 0 V
LTb
600 1202 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(60) Rshow
LTa
600 1518 M
2817 0 V
LTb
600 1518 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(70) Rshow
LTa
600 1835 M
2817 0 V
LTb
600 1835 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(80) Rshow
LTa
600 2152 M
2817 0 V
LTb
600 2152 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(90) Rshow
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(100) Rshow
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(\(16,1\)) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(8,2\)) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(4,4\)) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(2,8\)) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(1,16\)) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
340 1360 M
currentpoint gsave translate 90 rotate 0 0 M
(Normalized subpage misses \(w.r.t. \(16,1\)\)) Cshow
grestore
2008 51 M
(Processor Geometry - 16 Processors            ) Cshow
LT1
1654 2106 M
(160x160x160) Rshow
1714 2106 M
180 0 V
600 2469 M
1304 1154 L
2009 495 L
705 646 V
3417 2453 L
1774 2106 A
600 2469 A
1304 1154 A
2009 495 A
2713 1141 A
3417 2453 A
LT2
1654 2006 M
(144x144x144) Rshow
1714 2006 M
180 0 V
600 2469 M
1304 1072 L
2009 473 L
705 567 V
3417 2387 L
1774 2006 B
600 2469 B
1304 1072 B
2009 473 B
2713 1040 B
3417 2387 B
LT4
1654 1906 M
(64x64x64) Rshow
1714 1906 M
180 0 V
600 2469 M
1304 1113 L
2009 543 L
705 649 V
3417 2444 L
1774 1906 T
600 2469 T
1304 1113 T
2009 543 T
2713 1192 T
3417 2444 T
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 124 640 a Fo(Figure)12 b(3.)18 b(Remote)12 b(Memory)g(Access.)
899 586 y @beginspecial 50 @llx 50 @lly 230 @urx 176 @ury 2057
@rwi @setspecial
%%BeginDocument: datac.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.050 0.050 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
540 251 M
(90) Rshow
LTa
600 528 M
2817 0 V
LTb
600 528 M
63 0 V
2754 0 R
-63 0 V
540 528 M
(100) Rshow
LTa
600 806 M
2817 0 V
LTb
600 806 M
63 0 V
2754 0 R
-63 0 V
540 806 M
(110) Rshow
LTa
600 1083 M
2817 0 V
LTb
600 1083 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(120) Rshow
LTa
600 1360 M
2817 0 V
LTb
600 1360 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(130) Rshow
LTa
600 1637 M
2817 0 V
LTb
600 1637 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(140) Rshow
LTa
600 1915 M
2817 0 V
LTb
600 1915 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(150) Rshow
LTa
600 2192 M
2817 0 V
LTb
600 2192 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(160) Rshow
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(170) Rshow
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(\(16,1\)) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(8,2\)) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(4,4\)) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(2,8\)) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(\(1,16\)) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
340 1360 M
currentpoint gsave translate 90 rotate 0 0 M
(Normalized cache misses \(w.r.t \(16,1\)\)) Cshow
grestore
2008 51 M
(Processor Geometry - 16 Processors         ) Cshow
LT1
1654 2106 M
(160x160x160) Rshow
1714 2106 M
180 0 V
600 528 M
1304 817 L
705 626 V
705 721 V
704 250 V
1774 2106 A
600 528 A
1304 817 A
2009 1443 A
2713 2164 A
3417 2414 A
LT2
1654 2006 M
(144x144x144) Rshow
1714 2006 M
180 0 V
600 528 M
1304 678 L
705 488 V
705 610 V
704 333 V
1774 2006 B
600 528 B
1304 678 B
2009 1166 B
2713 1776 B
3417 2109 B
LT4
1654 1906 M
(64x64x64) Rshow
1714 1906 M
180 0 V
600 528 M
1304 329 L
705 -36 V
705 133 V
704 97 V
1774 1906 T
600 528 T
1304 329 T
2009 293 T
2713 426 T
3417 523 T
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 1085 640 a(Figure)g(4.)18 b(Cache)13 b(Misses.)47
1297 y @beginspecial 50 @llx 50 @lly 410 @urx 302 @ury 2057
@rwi @setspecial
%%BeginDocument: tred2.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.100 0.100 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
540 251 M
(1e+06) Rshow
LTb
600 2469 M
63 0 V
2754 0 R
-63 0 V
540 2469 M
(4e+06) Rshow
LTa
600 1360 M
2817 0 V
LTb
600 1360 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2009 M
2817 0 V
LTb
600 2009 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
31 0 V
2786 0 R
-31 0 V
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(0) Cshow
LTa
952 251 M
0 2218 V
LTb
952 251 M
0 63 V
0 2155 R
0 -63 V
952 151 M
(2) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(4) Cshow
LTa
1656 251 M
0 2218 V
LTb
1656 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(6) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(8) Cshow
LTa
2361 251 M
0 2218 V
LTb
2361 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(10) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(12) Cshow
LTa
3065 251 M
0 2218 V
LTb
3065 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(14) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(16) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
340 1260 M
currentpoint gsave translate 90 rotate 0 0 M
(Execution Time \(Micro Seconds\)) Cshow
grestore
2008 51 M
(Number of Processors) Cshow
LT0
3054 2306 M
("Cyclic") Rshow
3114 2306 M
180 0 V
776 910 M
176 318 V
352 533 V
705 -387 V
704 -149 V
704 707 V
3174 2306 D
776 910 D
952 1228 D
1304 1761 D
2009 1374 D
2713 1225 D
3417 1932 D
LT1
3054 2206 M
("BlockCyclic") Rshow
3114 2206 M
180 0 V
776 942 M
952 817 L
1304 662 L
705 -34 V
704 110 V
704 1480 V
3174 2206 A
776 942 A
952 817 A
1304 662 A
2009 628 A
2713 738 A
3417 2218 A
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 140 1351 a(Figure)f(5.)18 b(Ef)o(fect)12 b(of)g(False)h
(Sharing.)899 1297 y @beginspecial 50 @llx 50 @lly 410 @urx
302 @ury 2057 @rwi @setspecial
%%BeginDocument: tred2c.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.100 0.100 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(40000) Rshow
LTa
600 251 M
2817 0 V
LTb
600 251 M
31 0 V
2786 0 R
-31 0 V
LTa
600 497 M
2817 0 V
LTb
600 497 M
31 0 V
2786 0 R
-31 0 V
LTa
600 697 M
2817 0 V
LTb
600 697 M
31 0 V
2786 0 R
-31 0 V
LTa
600 867 M
2817 0 V
LTb
600 867 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1014 M
2817 0 V
LTb
600 1014 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1144 M
2817 0 V
LTb
600 1144 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1260 M
2817 0 V
LTb
600 1260 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(100000) Rshow
LTa
600 2023 M
2817 0 V
LTb
600 2023 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
31 0 V
2786 0 R
-31 0 V
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(0) Cshow
LTa
952 251 M
0 2218 V
LTb
952 251 M
0 63 V
0 2155 R
0 -63 V
952 151 M
(2) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(4) Cshow
LTa
1656 251 M
0 2218 V
LTb
1656 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(6) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(8) Cshow
LTa
2361 251 M
0 2218 V
LTb
2361 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(10) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(12) Cshow
LTa
3065 251 M
0 2218 V
LTb
3065 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(14) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(16) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
400 1660 M
currentpoint gsave translate 90 rotate 0 0 M
(Cache Misses) Cshow
grestore
2008 51 M
(Number of Processors) Cshow
LT0
3054 2306 M
("Cyclic") Rshow
3114 2306 M
180 0 V
123 -856 R
-704 -3 V
-704 172 V
-705 449 V
952 1885 L
776 2180 L
3174 2306 D
3417 1450 D
2713 1447 D
2009 1619 D
1304 2068 D
952 1885 D
776 2180 D
LT1
3054 2206 M
("BlockCyclic") Rshow
3114 2206 M
180 0 V
3417 1029 M
2713 642 L
2009 475 L
-705 601 V
952 1596 L
776 2166 L
3174 2206 A
3417 1029 A
2713 642 A
2009 475 A
1304 1076 A
952 1596 A
776 2166 A
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 1085 1351 a(Figure)f(6.)18 b(Cache)13 b(Misses.)4
1510 y(These)18 b(programs)f(have)g(triangular)f(iteration)g(spaces)i(which)f
(necessitate)h(cyclical)f(distribution)f(for)4 1564 y(load)c(balancing.)19
b(The)13 b(choice)f(of)g(this)h(distribution)e(combined)h(with)g(the)h
(storage)f(order)g(of)g(the)g(arrays)4 1619 y(cause)17 b(more)f(than)g(one)g
(processor)g(to)g(share)g(the)g(same)h(cache)f(line,)i(leading)e(to)g(false)g
(sharing.)29 b(The)4 1673 y(impact)17 b(of)f(this)h(false)h(sharing)e(is)i
(shown)f(in)g(Figure)f(5)h(for)f(the)h Fd(Tred2)g Fo(application)f(on)h(the)g
(KSR1)4 1727 y(multiprocessor)m(.)37 b(The)19 b(\256gure)f(shows)h(the)g
(execution)f(time)g(of)g(the)h(application)f(for)g Fd(Cyclic)g
Fo(and)4 1781 y Fd(BlockCyclic)12 b Fo(distributions)g(using)i(1)f(to)g(16)g
(processors.)20 b(The)14 b(use)g(of)e(the)h Fd(Cyclic)g Fo(distribution)4
1835 y(results)f(in)g(a)h(lar)o(ge)f(number)f(of)h(cache)h(misses,)g(as)g
(can)f(be)g(seen)h(in)f(Figure)g(6.)18 b(The)13 b(resulting)e(overhead)4
1889 y(causes)20 b(execution)f(time)g(to)g(increase)g(as)g(the)g(number)g(of)
f(processors)i(increases.)39 b(The)19 b(arrays)g(are)4 1944
y(distributed)c(using)g(a)g Fd(BlockCyclic)f Fo(distribution,)h(where)g(the)g
(size)h(of)f(the)g(block)g(is)h(equal)f(to)g(the)4 1998 y(size)22
b(of)e(the)h(cache)g(line,)i(which)e(ef)o(fectively)f(eliminates)h(false)g
(sharing.)43 b(When)21 b(the)g(number)f(of)4 2052 y(processors)14
b(is)g(small,)g(the)f(load)h(is)g(relatively)e(well-balanced,)i(and)f(the)h
(elimination)e(of)h(false)h(sharing)4 2106 y(improves)h(performance.)25
b(However)n(,)15 b(as)h(the)f(number)f(of)h(processors)g(increases,)i(the)e
(load)f(becomes)4 2160 y(increasingly)f(imbalanced,)h(and)f(the)h(negative)f
(impact)g(of)g(this)g(load)g(imbalance)h(begins)f(to)g(outweigh)4
2214 y(the)h(bene\256ts)h(of)e(eliminating)h(false)g(sharing.)24
b(A)14 b(compiler)g(for)f(SSMM)h(must)g(consider)h(this)f(tradeof)o(f)4
2269 y(between)f(load)f(imbalance)g(and)g(false)h(sharing)f(when)g
(determining)g(data)g(distributions.)4 2450 y Fn(5)71 b(Impact)19
b(on)f(Computation)i(Partitioning)4 2577 y Fo(The)12 b(owner)o(-computes)f
(rule)g(has)h(been)f(the)h(computation)f(partitioner)f(of)h(choice)g(for)g
(compiling)g(HPF-)4 2631 y(type)17 b(languages)g(on)g(DMMs)h([16].)32
b(The)17 b(owner)o(-computes)f(rule)h(maps)g(a)g(statement)h(such)f(that)g
(the)4 2686 y(the)h(computation)e(is)i(executed)g(on)g(the)f(processor)h(on)f
(which)h(the)f(data)h(element)f(that)h(is)g(written)e(is)4
2740 y(local.)27 b(All)15 b(the)g(data)g(elements)g(that)g(are)g(required)f
(to)h(compute)g(the)g(result)g(\(which)g(may)g(be)g(remote\))4
2794 y(are)h(communicated)f(to)h(the)g(processor)m(.)29 b(A)16
b(strict)g(rule)f(such)i(as)f(owner)o(-computes)f(is)h(not)g(necessary)4
2848 y(on)h(a)f(SSMM)h(because)g(message)h(passing)f(code)g(is)g(not)f
(generated)g(at)h(compile)f(time)g([3].)30 b(In)17 b(some)p
eop
%%Page: 7 7
7 6 bop 482 311 a @beginspecial 127 @llx 520 @lly 393 @urx
632 @ury 2160 @rwi @setspecial
%%BeginDocument: adi.idraw

/arrowhead {
0 begin
transform originalCTM itransform
/taily exch def
/tailx exch def
transform originalCTM itransform
/tipy exch def
/tipx exch def
/dy tipy taily sub def
/dx tipx tailx sub def
/angle dx 0 ne dy 0 ne or { dy dx atan } { 90 } ifelse def
gsave
originalCTM setmatrix
tipx tipy translate
angle rotate
newpath
arrowHeight neg arrowWidth 2 div moveto
0 0 lineto
arrowHeight neg arrowWidth 2 div neg lineto
patternNone not {
originalCTM setmatrix
/padtip arrowHeight 2 exp 0.25 arrowWidth 2 exp mul add sqrt brushWidth mul
arrowWidth div def
/padtail brushWidth 2 div def
tipx tipy translate
angle rotate
padtip 0 translate
arrowHeight padtip add padtail add arrowHeight div dup scale
arrowheadpath
ifill
} if
brushNone not {
originalCTM setmatrix
tipx tipy translate
angle rotate
arrowheadpath
istroke
} if
grestore
end
} dup 0 9 dict put def

/arrowheadpath {
newpath
arrowHeight neg arrowWidth 2 div moveto
0 0 lineto
arrowHeight neg arrowWidth 2 div neg lineto
} def

/leftarrow {
0 begin
y exch get /taily exch def
x exch get /tailx exch def
y exch get /tipy exch def
x exch get /tipx exch def
brushLeftArrow { tipx tipy tailx taily arrowhead } if
end
} dup 0 4 dict put def

/rightarrow {
0 begin
y exch get /tipy exch def
x exch get /tipx exch def
y exch get /taily exch def
x exch get /tailx exch def
brushRightArrow { tipx tipy tailx taily arrowhead } if
end
} dup 0 4 dict put def


/arrowHeight 10 def
/arrowWidth 5 def

/IdrawDict 51 dict def
IdrawDict begin

/reencodeISO {
dup dup findfont dup length dict begin
{ 1 index /FID ne { def }{ pop pop } ifelse } forall
/Encoding ISOLatin1Encoding def
currentdict end definefont
} def

/ISOLatin1Encoding [
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/space/exclam/quotedbl/numbersign/dollar/percent/ampersand/quoteright
/parenleft/parenright/asterisk/plus/comma/minus/period/slash
/zero/one/two/three/four/five/six/seven/eight/nine/colon/semicolon
/less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N
/O/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright
/asciicircum/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m
/n/o/p/q/r/s/t/u/v/w/x/y/z/braceleft/bar/braceright/asciitilde
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/dotlessi/grave/acute/circumflex/tilde/macron/breve
/dotaccent/dieresis/.notdef/ring/cedilla/.notdef/hungarumlaut
/ogonek/caron/space/exclamdown/cent/sterling/currency/yen/brokenbar
/section/dieresis/copyright/ordfeminine/guillemotleft/logicalnot
/hyphen/registered/macron/degree/plusminus/twosuperior/threesuperior
/acute/mu/paragraph/periodcentered/cedilla/onesuperior/ordmasculine
/guillemotright/onequarter/onehalf/threequarters/questiondown
/Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE/Ccedilla
/Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex
/Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis
/multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute
/Thorn/germandbls/agrave/aacute/acircumflex/atilde/adieresis
/aring/ae/ccedilla/egrave/eacute/ecircumflex/edieresis/igrave
/iacute/icircumflex/idieresis/eth/ntilde/ograve/oacute/ocircumflex
/otilde/odieresis/divide/oslash/ugrave/uacute/ucircumflex/udieresis
/yacute/thorn/ydieresis
] def
/Helvetica reencodeISO def

/none null def
/numGraphicParameters 17 def
/stringLimit 65535 def

/Begin {
save
numGraphicParameters dict begin
} def

/End {
end
restore
} def

/SetB {
dup type /nulltype eq {
pop
false /brushRightArrow idef
false /brushLeftArrow idef
true /brushNone idef
} {
/brushDashOffset idef
/brushDashArray idef
0 ne /brushRightArrow idef
0 ne /brushLeftArrow idef
/brushWidth idef
false /brushNone idef
} ifelse
} def

/SetCFg {
/fgblue idef
/fggreen idef
/fgred idef
} def

/SetCBg {
/bgblue idef
/bggreen idef
/bgred idef
} def

/SetF {
/printSize idef
/printFont idef
} def

/SetP {
dup type /nulltype eq {
pop true /patternNone idef
} {
dup -1 eq {
/patternGrayLevel idef
/patternString idef
} {
/patternGrayLevel idef
} ifelse
false /patternNone idef
} ifelse
} def

/BSpl {
0 begin
storexyn
newpath
n 1 gt {
0 0 0 0 0 0 1 1 true subspline
n 2 gt {
0 0 0 0 1 1 2 2 false subspline
1 1 n 3 sub {
/i exch def
i 1 sub dup i dup i 1 add dup i 2 add dup false subspline
} for
n 3 sub dup n 2 sub dup n 1 sub dup 2 copy false subspline
} if
n 2 sub dup n 1 sub dup 2 copy 2 copy false subspline
patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if
brushNone not { istroke } if
0 0 1 1 leftarrow
n 2 sub dup n 1 sub dup rightarrow
} if
end
} dup 0 4 dict put def

/Circ {
newpath
0 360 arc
patternNone not { ifill } if
brushNone not { istroke } if
} def

/CBSpl {
0 begin
dup 2 gt {
storexyn
newpath
n 1 sub dup 0 0 1 1 2 2 true subspline
1 1 n 3 sub {
/i exch def
i 1 sub dup i dup i 1 add dup i 2 add dup false subspline
} for
n 3 sub dup n 2 sub dup n 1 sub dup 0 0 false subspline
n 2 sub dup n 1 sub dup 0 0 1 1 false subspline
patternNone not { ifill } if
brushNone not { istroke } if
} {
Poly
} ifelse
end
} dup 0 4 dict put def

/Elli {
0 begin
newpath
4 2 roll
translate
scale
0 0 1 0 360 arc
patternNone not { ifill } if
brushNone not { istroke } if
end
} dup 0 1 dict put def

/Line {
0 begin
2 storexyn
newpath
x 0 get y 0 get moveto
x 1 get y 1 get lineto
brushNone not { istroke } if
0 0 1 1 leftarrow
0 0 1 1 rightarrow
end
} dup 0 4 dict put def

/MLine {
0 begin
storexyn
newpath
n 1 gt {
x 0 get y 0 get moveto
1 1 n 1 sub {
/i exch def
x i get y i get lineto
} for
patternNone not brushLeftArrow not brushRightArrow not and and { ifill } if
brushNone not { istroke } if
0 0 1 1 leftarrow
n 2 sub dup n 1 sub dup rightarrow
} if
end
} dup 0 4 dict put def

/Poly {
3 1 roll
newpath
moveto
-1 add
{ lineto } repeat
closepath
patternNone not { ifill } if
brushNone not { istroke } if
} def

/Rect {
0 begin
/t exch def
/r exch def
/b exch def
/l exch def
newpath
l b moveto
l t lineto
r t lineto
r b lineto
closepath
patternNone not { ifill } if
brushNone not { istroke } if
end
} dup 0 4 dict put def

/Text {
ishow
} def

/idef {
dup where { pop pop pop } { exch def } ifelse
} def

/ifill {
0 begin
gsave
patternGrayLevel -1 ne {
fgred bgred fgred sub patternGrayLevel mul add
fggreen bggreen fggreen sub patternGrayLevel mul add
fgblue bgblue fgblue sub patternGrayLevel mul add setrgbcolor
eofill
} {
eoclip
originalCTM setmatrix
pathbbox /t exch def /r exch def /b exch def /l exch def
/w r l sub ceiling cvi def
/h t b sub ceiling cvi def
/imageByteWidth w 8 div ceiling cvi def
/imageHeight h def
bgred bggreen bgblue setrgbcolor
eofill
fgred fggreen fgblue setrgbcolor
w 0 gt h 0 gt and {
l w add b translate w neg h scale
w h true [w 0 0 h neg 0 h] { patternproc } imagemask
} if
} ifelse
grestore
end
} dup 0 8 dict put def

/istroke {
gsave
brushDashOffset -1 eq {
[] 0 setdash
1 setgray
} {
brushDashArray brushDashOffset setdash
fgred fggreen fgblue setrgbcolor
} ifelse
brushWidth setlinewidth
originalCTM setmatrix
stroke
grestore
} def

/ishow {
0 begin
gsave
fgred fggreen fgblue setrgbcolor
/fontDict printFont printSize scalefont dup setfont def
/descender fontDict begin 0 [FontBBox] 1 get FontMatrix end
transform exch pop def
/vertoffset 1 printSize sub descender sub def {
0 vertoffset moveto show
/vertoffset vertoffset printSize sub def
} forall
grestore
end
} dup 0 3 dict put def
/patternproc {
0 begin
/patternByteLength patternString length def
/patternHeight patternByteLength 8 mul sqrt cvi def
/patternWidth patternHeight def
/patternByteWidth patternWidth 8 idiv def
/imageByteMaxLength imageByteWidth imageHeight mul
stringLimit patternByteWidth sub min def
/imageMaxHeight imageByteMaxLength imageByteWidth idiv patternHeight idiv
patternHeight mul patternHeight max def
/imageHeight imageHeight imageMaxHeight sub store
/imageString imageByteWidth imageMaxHeight mul patternByteWidth add string def
0 1 imageMaxHeight 1 sub {
/y exch def
/patternRow y patternByteWidth mul patternByteLength mod def
/patternRowString patternString patternRow patternByteWidth getinterval def
/imageRow y imageByteWidth mul def
0 patternByteWidth imageByteWidth 1 sub {
/x exch def
imageString imageRow x add patternRowString putinterval
} for
} for
imageString
end
} dup 0 12 dict put def

/min {
dup 3 2 roll dup 4 3 roll lt { exch } if pop
} def

/max {
dup 3 2 roll dup 4 3 roll gt { exch } if pop
} def

/midpoint {
0 begin
/y1 exch def
/x1 exch def
/y0 exch def
/x0 exch def
x0 x1 add 2 div
y0 y1 add 2 div
end
} dup 0 4 dict put def

/thirdpoint {
0 begin
/y1 exch def
/x1 exch def
/y0 exch def
/x0 exch def
x0 2 mul x1 add 3 div
y0 2 mul y1 add 3 div
end
} dup 0 4 dict put def

/subspline {
0 begin
/movetoNeeded exch def
y exch get /y3 exch def
x exch get /x3 exch def
y exch get /y2 exch def
x exch get /x2 exch def
y exch get /y1 exch def
x exch get /x1 exch def
y exch get /y0 exch def
x exch get /x0 exch def
x1 y1 x2 y2 thirdpoint
/p1y exch def
/p1x exch def
x2 y2 x1 y1 thirdpoint
/p2y exch def
/p2x exch def
x1 y1 x0 y0 thirdpoint
p1x p1y midpoint
/p0y exch def
/p0x exch def
x2 y2 x3 y3 thirdpoint
p2x p2y midpoint
/p3y exch def
/p3x exch def
movetoNeeded { p0x p0y moveto } if
p1x p1y p2x p2y p3x p3y curveto
end
} dup 0 17 dict put def

/storexyn {
/n exch def
/y n array def
/x n array def
n 1 sub -1 0 {
/i exch def
y i 3 2 roll put
x i 3 2 roll put
} for
} def

/SSten {
fgred fggreen fgblue setrgbcolor
dup true exch 1 0 0 -1 0 6 -1 roll matrix astore
} def

/FSten {
dup 3 -1 roll dup 4 1 roll exch
newpath
0 0 moveto
dup 0 exch lineto
exch dup 3 1 roll exch lineto
0 lineto
closepath
bgred bggreen bgblue setrgbcolor
eofill
SSten
} def

/Rast {
exch dup 3 1 roll 1 0 0 -1 0 6 -1 roll matrix astore
} def




Begin
[ 0.799705 0 0 0.799705 0 0 ] concat
/originalCTM matrix currentmatrix def

Begin %I Pict

Begin %I Pict

Begin %I Line
0 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 97 141 ] concat
87 643 119 643 Line
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 224 787 ] concat
[
(Phase 1)
] Text
End

End %I eop

Begin %I Pict

Begin %I Line
0 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 97 141 ] concat
87 643 87 619 Line
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 6.12303e-17 1 -1 6.12303e-17 176.5 714.5 ] concat
[
(Phase2)
] Text
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -96 48 ] concat

Begin %I Pict
[ 1 0 0 1 -8 0 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 344 715 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 224 280 ] concat
79 419 175 443 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -8 8 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 344 683 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 224 248 ] concat
79 419 175 443 Rect
End

End %I eop

Begin %I Pict

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 336 667 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 216 232 ] concat
79 419 175 443 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 0 8 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 336 635 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 216 200 ] concat
79 419 175 443 Rect
End

End %I eop

End %I eop

End %I eop

Begin %I Pict
[ 1 0 0 1 15 -1 ] concat

Begin %I Pict
[ 1 0 0 1 176 192 ] concat

Begin %I Pict

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 184 571 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 92 144 ] concat
87 411 111 435 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -168 -72 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 376 643 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
271 483 295 507 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -200 8 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 432 563 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
327 403 351 427 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -280 104 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 536 467 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
431 307 455 331 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 72 -24 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 184 571 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 92 144 ] concat
87 411 111 435 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 48 -48 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 184 571 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 92 144 ] concat
87 411 111 435 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 24 -72 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 184 571 ] concat
[
(P0)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 92 144 ] concat
87 411 111 435 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -192 -96 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 376 643 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
271 483 295 507 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -120 -120 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 376 643 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
271 483 295 507 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -144 -144 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 376 643 ] concat
[
(P1)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
271 483 295 507 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -224 -16 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 432 563 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
327 403 351 427 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -248 -40 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 432 563 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
327 403 351 427 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -176 -64 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 432 563 ] concat
[
(P2)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
327 403 351 427 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -304 80 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 536 467 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
431 307 455 331 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -352 32 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 536 467 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
431 307 455 331 Rect
End

End %I eop

Begin %I Pict
[ 1 0 0 1 -328 56 ] concat

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 536 467 ] concat
[
(P3)
] Text
End

Begin %I Rect
0 0 0 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
none SetP %I p n
[ 1 -0 -0 1 100 144 ] concat
431 307 455 331 Rect
End

End %I eop

End %I eop

Begin %I Pict
[ 1 0 0 1 160 0 ] concat

Begin %I Line
0 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 97 141 ] concat
87 643 87 619 Line
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 6.12303e-17 1 -1 6.12303e-17 176.5 714.5 ] concat
[
(Phase2)
] Text
End

End %I eop

Begin %I Pict
[ 1 0 0 1 160 0 ] concat

Begin %I Line
0 0 1 [] 0 SetB
0 0 0 SetCFg
1 1 1 SetCBg
0 SetP
[ 1 -0 -0 1 97 141 ] concat
87 643 119 643 Line
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 224 787 ] concat
[
(Phase 1)
] Text
End

End %I eop

End %I eop

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 160.25 662.724 ] concat
[
(\(a\) Row Block Distribution.)
] Text
End

Begin %I Text
0 0 0 SetCFg
Helvetica 12 SetF
[ 1 0 0 1 323.75 664 ] concat
[
(\(b\) Our Proposed Distribution.)
] Text
End

End %I eop

showpage


end
%%EndDocument
 @endspecial 269 391 a Fo(Figure)11 b(7:)18 b(Data)13 b(Distribution)e(used)i
(to)f(alleviate)g(Memory)g(Contention.)503 1045 y @beginspecial
50 @llx 50 @lly 410 @urx 302 @ury 2057 @rwi @setspecial
%%BeginDocument: 256.adi.ps
/gnudict 40 dict def
gnudict begin
/Color false def
/Solid false def
/gnulinewidth 5.000 def
/vshift -33 def
/dl {10 mul} def
/hpt 31.5 def
/vpt 31.5 def
/M {moveto} bind def
/L {lineto} bind def
/R {rmoveto} bind def
/V {rlineto} bind def
/vpt2 vpt 2 mul def
/hpt2 hpt 2 mul def
/Lshow { currentpoint stroke M
  0 vshift R show } def
/Rshow { currentpoint stroke M
  dup stringwidth pop neg vshift R show } def
/Cshow { currentpoint stroke M
  dup stringwidth pop -2 div vshift R show } def
/DL { Color {setrgbcolor Solid {pop []} if 0 setdash }
 {pop pop pop Solid {pop []} if 0 setdash} ifelse } def
/BL { stroke gnulinewidth 2 mul setlinewidth } def
/AL { stroke gnulinewidth 2 div setlinewidth } def
/PL { stroke gnulinewidth setlinewidth } def
/LTb { BL [] 0 0 0 DL } def
/LTa { AL [1 dl 2 dl] 0 setdash 0 0 0 setrgbcolor } def
/LT0 { PL [] 0 1 0 DL } def
/LT1 { PL [4 dl 2 dl] 0 0 1 DL } def
/LT2 { PL [2 dl 3 dl] 1 0 0 DL } def
/LT3 { PL [1 dl 1.5 dl] 1 0 1 DL } def
/LT4 { PL [5 dl 2 dl 1 dl 2 dl] 0 1 1 DL } def
/LT5 { PL [4 dl 3 dl 1 dl 3 dl] 1 1 0 DL } def
/LT6 { PL [2 dl 2 dl 2 dl 4 dl] 0 0 0 DL } def
/LT7 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 1 0.3 0 DL } def
/LT8 { PL [2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 2 dl 4 dl] 0.5 0.5 0.5 DL } def
/P { stroke [] 0 setdash
  currentlinewidth 2 div sub M
  0 currentlinewidth V stroke } def
/D { stroke [] 0 setdash 2 copy vpt add M
  hpt neg vpt neg V hpt vpt neg V
  hpt vpt V hpt neg vpt V closepath stroke
  P } def
/A { stroke [] 0 setdash vpt sub M 0 vpt2 V
  currentpoint stroke M
  hpt neg vpt neg R hpt2 0 V stroke
  } def
/B { stroke [] 0 setdash 2 copy exch hpt sub exch vpt add M
  0 vpt2 neg V hpt2 0 V 0 vpt2 V
  hpt2 neg 0 V closepath stroke
  P } def
/C { stroke [] 0 setdash exch hpt sub exch vpt add M
  hpt2 vpt2 neg V currentpoint stroke M
  hpt2 neg 0 R hpt2 vpt2 V stroke } def
/T { stroke [] 0 setdash 2 copy vpt 1.12 mul add M
  hpt neg vpt -1.62 mul V
  hpt 2 mul 0 V
  hpt neg vpt 1.62 mul V closepath stroke
  P  } def
/S { 2 copy A C} def
end
gnudict begin
gsave
50 50 translate
0.100 0.100 scale
0 setgray
/Times-Roman findfont 100 scalefont setfont
newpath
LTa
600 251 M
0 2218 V
LTb
LTa
600 251 M
2817 0 V
LTb
600 251 M
63 0 V
2754 0 R
-63 0 V
540 251 M
(1000) Rshow
LTa
600 585 M
2817 0 V
LTb
600 585 M
31 0 V
2786 0 R
-31 0 V
LTa
600 780 M
2817 0 V
LTb
600 780 M
31 0 V
2786 0 R
-31 0 V
LTa
600 919 M
2817 0 V
LTb
600 919 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1026 M
2817 0 V
LTb
600 1026 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1114 M
2817 0 V
LTb
600 1114 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1188 M
2817 0 V
LTb
600 1188 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1253 M
2817 0 V
LTb
600 1253 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1309 M
2817 0 V
LTb
600 1309 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1360 M
2817 0 V
LTb
600 1360 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(10000) Rshow
LTa
600 1694 M
2817 0 V
LTb
600 1694 M
31 0 V
2786 0 R
-31 0 V
LTa
600 1889 M
2817 0 V
LTb
600 1889 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2028 M
2817 0 V
LTb
600 2028 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2135 M
2817 0 V
LTb
600 2135 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2223 M
2817 0 V
LTb
600 2223 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2297 M
2817 0 V
LTb
600 2297 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2362 M
2817 0 V
LTb
600 2362 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2418 M
2817 0 V
LTb
600 2418 M
31 0 V
2786 0 R
-31 0 V
LTa
600 2469 M
2817 0 V
LTb
600 2469 M
63 0 V
2754 0 R
-63 0 V
-2814 0 R
(100000) Rshow
LTa
600 251 M
0 2218 V
LTb
600 251 M
0 63 V
0 2155 R
0 -63 V
600 151 M
(0) Cshow
LTa
952 251 M
0 2218 V
LTb
952 251 M
0 63 V
0 2155 R
0 -63 V
952 151 M
(2) Cshow
LTa
1304 251 M
0 2218 V
LTb
1304 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(4) Cshow
LTa
1656 251 M
0 2218 V
LTb
1656 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(6) Cshow
LTa
2009 251 M
0 2218 V
LTb
2009 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(8) Cshow
LTa
2361 251 M
0 2218 V
LTb
2361 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(10) Cshow
LTa
2713 251 M
0 2218 V
LTb
2713 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(12) Cshow
LTa
3065 251 M
0 2218 V
LTb
3065 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(14) Cshow
LTa
3417 251 M
0 2218 V
LTb
3417 251 M
0 63 V
0 2155 R
0 -63 V
0 -2255 R
(16) Cshow
600 251 M
2817 0 V
0 2218 V
-2817 0 V
600 251 L
220 1260 M
currentpoint gsave translate 90 rotate 0 0 M
(Execution Time \(milli sec\)) Cshow
grestore
2008 51 M
(Number of Processors) Cshow
LT0
2009 1253 M
(Owner Computes\(Block, Block\)) Rshow
2069 1253 M
180 0 V
952 2200 M
352 -24 V
705 86 V
1408 89 V
2129 1253 D
952 2200 D
1304 2176 D
2009 2262 D
3417 2351 D
LT1
2009 1153 M
(Sequential) Rshow
2069 1153 M
180 0 V
776 2146 M
2129 1153 A
776 2146 A
LT2
2009 1053 M
(No Distributions) Rshow
2069 1053 M
180 0 V
3417 2056 M
-704 3 V
-704 6 V
-705 8 V
-352 61 V
2129 1053 B
3417 2056 B
2713 2059 B
2009 2065 B
1304 2073 B
952 2134 B
LT3
2009 953 M
(\(*,Cyclic\)) Rshow
2069 953 M
180 0 V
952 2090 M
352 -33 V
705 -48 V
704 25 V
704 18 V
2129 953 C
952 2090 C
1304 2057 C
2009 2009 C
2713 2034 C
3417 2052 C
LT4
2009 853 M
(\(*,Block\)) Rshow
2069 853 M
180 0 V
952 2070 M
352 -53 V
705 -45 V
704 23 V
704 9 V
2129 853 T
952 2070 T
1304 2017 T
2009 1972 T
2713 1995 T
3417 2004 T
LT5
2009 753 M
(Owner Computes\(*,Cyclic\)) Rshow
2069 753 M
180 0 V
952 2080 M
352 -252 V
705 -78 V
352 216 V
352 20 V
704 128 V
2129 753 S
952 2080 S
1304 1828 S
2009 1750 S
2361 1966 S
2713 1986 S
3417 2114 S
LT6
2009 653 M
(Owner Computes\(*,Block\)) Rshow
2069 653 M
180 0 V
3417 1820 M
2713 1611 L
-704 -55 V
-705 118 V
952 1954 L
2129 653 D
3417 1820 D
2713 1611 D
2009 1556 D
1304 1674 D
952 1954 D
LT7
2009 553 M
(\(Block,Block\)) Rshow
2069 553 M
180 0 V
1168 547 R
-704 222 V
-704 85 V
-705 249 V
952 1951 L
2129 553 A
3417 1100 A
2713 1322 A
2009 1407 A
1304 1656 A
952 1951 A
stroke
grestore
end
showpage
%%EndDocument
 @endspecial 532 1111 a(Figure)f(8:)18 b(ADI)12 b(Performance)f(\(256x256\).)
4 1246 y(cases)18 b(adhering)d(to)i(owner)o(-computes)e(rule)h(can)h(incur)e
(severe)i(synchronization)f(or)g(ownership)f(test)4 1300 y(overhead)c(which)f
(exceeds)h(the)g(cost)g(of)f(accessing)i(remote)e(memory)m(.)17
b(W)l(e)11 b(use)g(the)g(Altering)f(Direction)4 1355 y(Integration)i(\()p
Fd(ADI)p Fo(\))f(to)i(illustrate)f(that)h(the)g(shared)f(address)i(space)f
(provides)g(\257exibility)e(in)i(the)g(choice)4 1409 y(of)i(computation)f
(partitions,)h(reducing)f(contention)g(and)h(synchronization)f(overhead,)i
(and)e(resulting)4 1463 y(in)e(signi\256cant)g(performance)g(improvements.)77
1539 y(W)l(e)k(use)f(the)g(Hector)n(,)h(a)f(Non-Uniform)e(Memory)i(Access)h
(multiprocessor)n(,)f(as)g(an)h(experimental)4 1593 y(platform.)21
b(Hector)13 b(consists)h(of)f(4)h(sets)g(of)f(processor)o(-memory)g(pairs)g
(connected)h(by)f(a)h(bus)g(to)f(form)g(a)4 1647 y(station;)g(4)g(stations)g
(are)f(connected)h(by)g(a)g(local)f(ring)g(to)h(form)f(a)h(cluster;)f(4)h
(local)g(rings)f(are)h(connected)4 1701 y(by)g(a)g(global)g(ring.)19
b(W)l(e)14 b(use)f(a)g(system)h(with)e(one)h(cluster)m(.)21
b(Each)13 b(processor)o(-memory)f(pair)g(consists)i(of)4 1755
y(a)f(Motorola)f(MC88100)h(CPU,)g(a)g(16)f(KB)h(instruction)f(cache,)i(a)f
(16)f(KB)h(data)g(cache)g(and)g(4)f(MB)i(of)e(the)4 1809 y(globally)i
(addressable)g(memory)m(.)23 b(The)15 b(hardware)e(provides)h(no)g(support)g
(for)f(cache)i(coherence.)23 b(The)4 1864 y(coherence)12 b(of)f(data)h(is)g
(maintained)f(by)h(software)f(at)h(cache)g(line)f(granularity)g([10)o(].)18
b(Data)12 b(distributions)4 1918 y(are)g(implemented)g(using)g(the)h(array)e
(allocation)h(techniques)h(described)f(in)g([21,)h(3].)4 2079
y Fc(5.1)58 b(Contention)15 b(and)f(Synchr)n(onization)h(Conscious)e
(Distribution)4 2177 y Fo(The)21 b Fd(ADI)e Fo(program)g(has)h(two)g(phases)h
(with)e(parallelism)g(along)h(orthogonal)f(dimensions)h(in)f(each)4
2231 y(phase.)k(It)13 b(operates)g(on)h(three)f(2-dimensional)g(arrays)g
Fj(A)p Fo(,)h Fj(B)i Fo(and)e Fj(X)t Fo(.)22 b(A)14 b(single)g(iteration)e
(of)i(an)f(outer)4 2285 y(sequentially)d(iterated)f(loop)h(consists)g(of)g(a)
g(forward)f(and)g(a)i(backward)e(sweep)i(phase)f(along)g(the)f(rows)h(of)4
2339 y(three)f(arrays,)h(followed)e(by)h(another)g(forward)e(and)i(backward)g
(sweep)h(phase)f(along)g(the)g(columns)g(of)g(the)4 2394 y(arrays)g([18].)17
b(This)10 b(application)f(is)h(typical)f(of)g(other)g(programs)g(such)g(as)h
Fd(2D-FFT)f Fo(and)h Fd(Erlebacher)4 2448 y Fo(that)i(have)h(parallelism)f
(in)g(orthogonal)f(directions)h(in)g(dif)o(ferent)f(phases)j(of)e(the)g
(program.)77 2523 y(The)k(best)g(data)f(distribution)f(scheme)i(for)e
Fd(ADI)h Fo(remains)g(an)g(issue)h(of)f(debate)h([18)o(,)g(4].)26
b(The)16 b(two)4 2578 y(proposed)h(schemes)g(partition)f(arrays)g(along)h(a)f
(single)h(dimension,)h(either)e(in)h(blocks)g(or)f(cyclically)m(.)4
2632 y(These)f(distributions,)g(in)e(conjunction)h(with)g(the)g(owner)o
(-computes)f(rule)h(result)f(in)h(a)h(wavefront)e(type)4 2686
y(computation,)e(leading)g(to)g(heavy)g(synchronization)g(overhead)g(in)g
(one)g(of)f(the)i(phases)g(of)e(the)h(program.)4 2740 y(Figure)i(7\(a\))g
(shows)i(a)f Fd(Block)f Fo(distribution)g(of)g(the)h(rows)g(of)f(the)h
(arrays.)23 b(W)n(ith)13 b(such)h(a)g(distribution,)4 2794
y(during)g(the)h(\256rst)g(phase)g(of)g(the)g(program)e(all)i(the)g
(processors)g(access)i(data)e(that)f(is)i(local)f(and)g(require)4
2848 y(no)g(communication.)25 b(During)14 b(the)h(second)g(phase,)h(however)n
(,)g(the)f(parallelism)f(is)h(orthogonal)f(to)h(the)p eop
%%Page: 8 8
8 7 bop 35 1 a Fo(T)m(able)12 b(1:)18 b(Performance)11 b(Bottlenecks)i(for)e
(various)h(data)h(and)f(computation)g(partitioning)e(for)i(ADI.)p
244 33 1363 2 v 243 84 2 51 v 252 84 V 277 69 a Fb(Data)g(Distribution)p
620 84 V 77 w(Compute)f(Rule)p 993 84 V 135 w(Performance)i(Bottleneck)p
1597 84 V 1606 84 V 244 85 1363 2 v 243 136 2 51 v 252 136
V 387 121 a(None)p 620 136 V 247 w(Relaxed)p 993 136 V 228
w(Memory)e(Contention)p 1597 136 V 1606 136 V 243 187 V 252
187 V 344 172 a(\(*,)h(Block\))p 620 187 V 117 w(Owner)o(-Computes)p
993 187 V 125 w(High)f(Synchronization)p 1597 187 V 1606 187
V 243 238 V 252 238 V 344 223 a(\(*,)h(Block\))p 620 238 V
204 w(Relaxed)p 993 238 V 228 w(Memory)f(Contention)p 1597
238 V 1606 238 V 243 289 V 252 289 V 339 273 a(\(*,)h(Cyclic\))p
620 289 V 112 w(Owner)o(-Computes)p 993 289 V 125 w(High)f(Synchronization)p
1597 289 V 1606 289 V 243 340 V 252 340 V 339 324 a(\(*,)h(Cyclic\))p
620 340 V 199 w(Relaxed)p 993 340 V 228 w(Memory)f(Contention)p
1597 340 V 1606 340 V 243 390 V 252 390 V 301 375 a(\(Block,)h(Block\))p
620 390 V 74 w(Owner)o(-Computes)p 993 390 V 180 w(Ownership)f(tests)p
1597 390 V 1606 390 V 243 441 V 252 441 V 301 426 a(\(Block,)h(Block\))p
620 441 V 161 w(Relaxed)p 993 441 V 137 w(High)f(Remote)g(Memory)g(Access)p
1597 441 V 1606 441 V 244 443 1363 2 v 4 606 a Fo(direction)17
b(of)g(distribution.)32 b(Strict)16 b(adherence)h(to)g(the)h(owner)o
(-computes)e(rule)h(implies)g(ordering)f(of)4 660 y(the)d(computations)f(by)g
(processors)h(on)f(the)g(corresponding)g(chunk)g(of)g(the)h(columns)f(they)g
(own.)19 b(Thus,)4 715 y(processor)10 b Fj(i)g Fo(has)h(to)f(wait)f(for)h
(processor)g Fj(i)c Fh(\000)g Fo(1)j(to)h(\256nish)g(the)g(computation)f(on)h
(its)h(chunk)f(of)f(the)h(column)4 769 y(before)g(proceeding.)17
b(A)10 b(lar)o(ger)f(number)h(of)f(synchronizations)h(are)g(required)f(to)h
(maintain)g(the)g(ordering)4 823 y(involved)i(in)g(the)g(wavefront)g
(computation.)77 899 y(The)i(synchronization)e(overhead)h(can)g(be)g
(eliminated)f(by)h(relaxing)f(the)h(owner)o(-computes)f(rule)h(in)4
953 y(the)18 b(second)g(phase)h(and)f(allowing)f(the)h(processor)g(to)f
(write)h(the)f(results)i(to)e(remote)h(memory)f(mod-)4 1007
y(ules.)24 b(This)15 b(eliminates)f(synchronization)f(overhead)h(at)g(the)g
(expense)g(of)g(increased)g(remote)g(memory)4 1061 y(accesses.)26
b(However)n(,)15 b(the)g(use)g(of)f(this)g(relaxed)g(compute)g(rule)g(with)g
(the)g Fd(\(*,Block\))g Fo(distribution)4 1115 y(results)9
b(in)g(heavy)g(contention.)17 b(Each)9 b(processor)g(is)g(responsible)g(for)f
(computing)g(a)i(column,)f(and)g(hence,)4 1169 y(each)14 b(processor)g
(accesses)h(every)e(memory)g(module)g(in)g(sequence.)23 b(Thus,)15
b(a)e(given)h(memory)e(module)4 1224 y(is)h(accessed)h(by)e(every)g
(processor)g(at)h(the)f(same)h(time,)f(leading)g(to)h(contention.)77
1299 y(The)k(data)e(distribution)g(scheme)h(depicted)f(in)h(Figure)f(7\(b\))
1149 1281 y Fm(4)1182 1299 y Fo(eliminates)h(contention)f(and)h(results)4
1353 y(in)21 b(the)g(best)h(possible)f(performance)f(with)h(the)g(relaxed)g
(compute)g(rule.)44 b(W)n(ith)21 b(this)g(distribution,)4 1408
y(processors)13 b(access)g(data)g(from)e(remote)g(memory)h(modules)g(in)g
(both)g(phases)h(of)f(the)g(program.)17 b(In)12 b(both)4 1462
y(phases,)h(processors)f(start)g(working)e(on)i(the)f(columns)h(assigned)g
(to)g(them)f(by)g(accessing)i(data)f(that)f(is)h(in)4 1516
y(dif)o(ferent)f(memory)f(modules)i(thus)g(avoiding)f(contention.)18
b(There)12 b(is)g(no)f(wavefront)g(type)h(parallelism,)4 1570
y(and)h(hence)f(no)g(overhead)g(involved)g(due)h(to)f(synchronization.)77
1646 y(The)19 b(use)f(of)f(owner)o(-computes)g(rule)h(with)f(the)h
(distribution)f(of)g(Figure)g(7\(b\))g(will)h(not)f(result)h(in)4
1700 y(good)f(performance.)31 b(Either)17 b(ownership)g(tests)h(must)f(be)g
(introduced)f(in)h(the)g(body)g(of)g(the)g(loops)g(to)4 1754
y(enforce)c(the)g(owner)o(-computes)f(rule,)i(or)e(the)i(loops)f(must)g(be)g
(rewritten)g(with)f(additional)h(strip-mined)4 1808 y(controlling)f(loops)h
(to)g(schedule)h(the)f(computations)f(on)h(sub-blocks)g(of)g(the)g(array)m(.)
20 b(The)14 b(former)d(leads)4 1862 y(to)h(overhead)g(and)h(the)f(latter)g
(introduces)g(synchronization)g(similar)f(to)i(the)f(wavefront)f
(computation.)77 1938 y(The)j(result)g(of)f(executing)g(the)h
Fd(ADI)f Fo(application)g(on)h(the)f(Hector)g(multiprocessor)g(for)g(a)h
(data)f(size)4 1992 y(of)18 b(256x256)f(with)h(various)g(data)g
(distributions)g(and)g(compute)g(rules)g(is)g(shown)g(in)g(Figure)g(8.)35
b(The)4 2046 y Fd(\(Block,Block\))17 b Fo(data)h(distribution)f(that)h
(relaxes)g(the)g(owner)o(-computes)f(rule)g(outperforms)g(all)4
2101 y(data)d(distribution)e(schemes)i(that)f(adhere)g(to)g(the)g(rule.)21
b(The)14 b(\256gure)f(also)g(indicates)h(that)f(the)g(overhead)4
2155 y(due)j(to)g(the)f(ownership)h(tests)g(when)g(using)g(the)f(owner)o
(-computes)g(rule)h(with)f(a)h Fd(\(Block,Block\))4 2209 y
Fo(distribution)d(degrades)h(performance.)21 b(It)14 b(is)g(also)g(clear)g
(that)f(the)h(use)g(of)g(data)g(distribution)e(improves)4 2263
y(performance)i(over)h(the)g(use)h(of)f(operating)f(system)i(policies)f(to)g
(manage)g(data)h(\(the)e(no)i(distributions)4 2317 y(curve\).)35
b(The)19 b(performance)e(bottlenecks)h(of)g(various)g(distributions)g(for)f
Fd(ADI)h Fo(are)g(summarized)g(in)4 2371 y(T)m(able)13 b(1.)p
4 2406 737 2 v 62 2437 a Fl(4)79 2452 y Fb(This)f(is)h(equivalent)g(to)f
(!HPF$)i(PROCESSORS)i(PROCS\(N\))g(with)c(!HPF$)i(DISTRIBUTE)h(B\(BLOCK,)4
2503 y(BLOCK\),)10 b(X\(BLOCK,)g(BLOCK\))g(ON)e(PROCS)j(in)d
Fa(HPF)p Fb(.)i(In)f(the)f(current)h Fa(HPF)h Fb(speci\256cation,)f(this)f
(distribution)4 2554 y(is)18 b(not)g(valid;)k(the)c(rank)h(of)g(each)g
(distributee)f(must)f(equal)i(the)g(rank)f(of)h(the)g(named)f(processor)h
(grid)f([16].)4 2604 y(Distributions)7 b(in)i(which)g(this)g(is)f(not)h(the)g
(case)h(introduce)f(additional)g(complexity)e(on)j(DMMs)e([17].)16
b(In)10 b(contrast,)4 2655 y(SSMMs)h(provide)g(the)g(\257exibility)f(to)h
(implement)f(these)h(distributions.)p eop
%%Page: 9 9
9 8 bop 4 -21 a Fn(6)71 b(Related)19 b(W)l(ork)4 106 y Fo(Several)12
b(researchers)g(have)g(focused)g(on)g(the)g(problem)f(of)g(deriving)g(data)h
(distributions)f(automatically)4 160 y(for)g(DMMs.)20 b(Li)12
b(and)g(Chen)h([22)o(],)f(Gupta)g(and)g(Banerjee)h([12)o(],)f(Zima)h(et)f
(al.)h([9)o(])f(and)g(Garcia)g(et)g(al.)h([11)o(])4 215 y(follow)e(the)h
(approach)g(of)f(\256nding)h(the)f(alignment)h(constraints)g(between)g(dif)o
(ferent)e(dimensions)i(of)g(the)4 269 y(arrays)g(and)g(derive)g(a)g(data)g
(distribution)f(that)h(minimizes)g(interprocessor)g(communication.)17
b(T)m(o)12 b(avoid)4 323 y(a)f(heuristic)g(approach,)g(Bixby)g(et)g(al.)h([7)
o(])f(formulate)e(a)j(0-1)e(integer)g(programming)g(problem)g(for)g(deriv-)4
377 y(ing)k(data)g(distributions.)21 b(Their)14 b(approach)g(relies)g(on)f
(the)h(assumption)g(that)g(a)g(good)f(data)h(distribution)4
431 y(for)h(the)i(entire)e(program)g(can)i(be)f(found)f(by)i(mer)o(ging)e
(the)h(data)g(distributions)g(of)f(smaller)h(segments)4 485
y(of)g(the)g(program.)27 b(They)17 b(minimize)e(the)h(interprocessor)f
(communication)g(using)h(the)g(\252performance)4 540 y(estimator)r(\272)c
(developed)h(by)g(Balasundaram)g(et)g(al.)g([6)o(].)20 b(Anderson)12
b([5])g(presents)i(an)e(algebraic)h(frame-)4 594 y(work)g(for)g(determining)f
(data)h(and)h(computation)e(partitions)h(by)g(minimizing)g(communication)f
(across)4 648 y(processors.)28 b(Data)16 b(transformations)e(are)i(then)f
(applied)h(so)f(that)h(the)f(processors)h(access)h(contiguous)4
702 y(data)g(regions)f(to)h(reduce)g(false)g(sharing.)31 b(This)17
b(technique)g(is)g(oblivious)f(to)h(SSMM)g(speci\256c)g(issues)4
756 y(such)c(as)g(contention)f(and)g(cache)h(af)o(\256nity)m(.)4
938 y Fn(7)71 b(Concluding)19 b(Remarks)4 1065 y Fo(Although)9
b(lar)o(ge)g(SSMMs)i(are)e(built)g(based)h(on)g(an)g(architecture)e(with)i
(distributed)f(memory)m(,)g(the)h(shared)4 1119 y(memory)15
b(paradigm)g(introduces)g(performance)g(issues)i(that)e(are)h(dif)o(ferent)f
(from)f(those)i(encountered)4 1173 y(in)e(DMMs.)24 b(The)14
b(high)f(cost)i(of)e(interprocessor)g(communication)g(in)h(distributed)f
(memory)f(multipro-)4 1227 y(cessors)18 b(makes)e(the)h(minimization)e(of)h
(communication)g(the)g(predominant)g(issue)h(in)f(selecting)h(data)4
1282 y(distributions)h(and)i(in)e(partitioning)g(computations.)38
b(On)19 b(SSMMs,)j(a)d(methodology)f(for)h(selecting)4 1336
y(data)14 b(distributions)g(must)g(also)g(consider)g(cache)h(af)o(\256nity)m
(,)f(memory)f(contention)h(and)g(false)g(sharing)g(in)4 1390
y(addition)d(to)g(the)g(cost)h(of)f(interprocessor)g(communication.)17
b(Furthermore,)10 b(the)h(single)h(shared)f(address)4 1444
y(space)j(present)f(in)g(SSMMs)g(provides)g(\257exibility)f(in)h(the)g
(selection)g(of)f(computation)h(partitions.)19 b(This)4 1498
y(should)e(be)f(exploited)g(in)g(applications)h(in)f(which)g(owner)o
(-computes)g(results)g(in)h(poor)f(performance.)4 1552 y(The)f
Fe(Jasmine)g Fo(compiler)f(project)g([2)o(])h(is)g(investigating)f(the)g
(issues)i(discussed)f(in)g(this)f(paper)h(through)4 1607 y(the)d(development)
g(of)g(a)h(framework)e(for)g(automatically)h(deriving)f(data)i(distributions)
e(on)i(SSMMs.)4 1777 y Fn(Refer)o(ences)29 1896 y Fo([1])24
b(T)l(.S.)17 b(Abdelrahman)f(et)g(al.)29 b(An)16 b(overview)f(of)h(the)g
(NUMAchine)h(multiprocessor)e(project.)28 b(In)112 1943 y Fe(Pr)n(oc.)13
b(of)g(the)f(Canadian)g(Super)n(computing)g(Conf.)p Fo(,)h(pages)g
(283\261295,)f(1994.)29 2032 y([2])24 b(T)l(.S.)12 b(Abdelrahman,)f(N.)h
(Manjikian,)g(and)f(S.)g(T)m(andri.)16 b(The)11 b(Jasmine)h(Compiler.)k(In)10
b(preparation.)29 2120 y([3])24 b(T)l(.S.)e(Abdelrahman)e(and)h(T)l(.N.)h(W)l
(ong.)41 b(Distributed)20 b(array)g(data)h(management)g(on)f(NUMA)112
2167 y(multiprocessors.)d(In)12 b Fe(Pr)n(oc.)i(of)e(SHPCC)p
Fo(,)i(pages)f(551\261559,)f(1994.)29 2256 y([4])24 b(S.P)-6
b(.)10 b(Amarasinghe,)g(J.M.)h(Anderson,)f(M.S.)h(Lam,)g(and)e(A.W)-5
b(.)11 b(Lim.)j(An)9 b(overview)g(of)g(a)h(compiler)112 2303
y(for)k(scalable)j(parallel)e(machines.)27 b(In)15 b Fe(Languages)h(and)f
(Compilers)i(for)f(Parallel)g(Computing)p Fo(,)112 2350 y(pages)c
(253\261272.)h(Springer)o(-V)-6 b(erlag)10 b(LNCS-768,)j(1993.)29
2438 y([5])24 b(J.M.)13 b(Anderson.)j(Demonstration)11 b(of)g(automatic)g
(data)h(and)f(computation)g(decomposition)g(tech-)112 2485
y(niques.)f(In)e Fe(Pr)n(oc.)g(of)g(the)g(W)-5 b(orkshop)8
b(on)g(Automatic)g(Data)g(Layout)g(and)g(Performance)g(Pr)n(ediction)p
Fo(,)112 2532 y(1995.)29 2620 y([6])24 b(V)-6 b(.)12 b(Balasundaram,)i(G.)f
(Fox,)g(K.)g(Kennedy)m(,)g(and)f(U.)i(Kremer)m(.)k(A)13 b(static)g
(performance)e(estimator)112 2667 y(to)h(guide)g(data)g(partitioning)f
(decisions.)19 b(In)12 b Fe(Pr)n(oc.)i(of)e(PPOPP)p Fo(,)j(pages)e
(213\261223,)f(1991.)29 2756 y([7])24 b(R.)9 b(Bixby)m(,)h(K.)g(Kennedy)m(,)f
(and)h(U.)f(Kremer)m(.)j(Automatic)d(data)g(layout)f(using)h(0-1)g(integer)f
(program-)112 2803 y(ming.)15 b(In)c Fe(Pr)n(oc.)i(of)e(the)g(Int'l)f(Conf.)i
(on)f(Parallel)g(Ar)n(chitectur)n(es)i(and)e(Compilation)g(T)-5
b(echniques)p Fo(,)112 2850 y(pages)12 b(111\261122,)h(1994.)p
eop
%%Page: 10 10
10 9 bop 29 -27 a Fo([8])24 b(W)-5 b(.J.)19 b(Bolosky)f(and)g(M.L.)h(Scott.)
32 b(False)18 b(sharing)f(and)h(its)g(ef)o(fect)f(on)g(shared)h(memory)f
(multi-)112 20 y(processors.)k(In)13 b Fe(Pr)n(oc.)j(of)d(4th)g(Symp.)h(on)g
(Experiences)h(with)e(Distributed)h(and)f(Multipr)n(ocessor)112
67 y(Systems)p Fo(,)g(pages)g(57\26171,)f(1993.)29 155 y([9])24
b(B.M.)14 b(Chapman,)g(T)l(.)h(Fahringer)n(,)e(and)g(H.)h(Zima.)21
b(Automatic)12 b(support)h(for)f(data)i(distribution)e(on)112
202 y(distributed)g(memory)g(multiprocessor)g(systems.)22 b(In)12
b Fe(Languages)h(and)g(Compilers)h(for)f(Parallel)112 249 y(Computing)p
Fo(,)f(pages)h(184\261199.)f(Springer)o(-V)-6 b(erlag)11 b(LNCS-768,)h(1993.)
4 337 y([10])24 b(B.)11 b(Gamsa.)k(Region-oriented)9 b(main)h(memory)f
(management)h(in)g(shared-memory)f(NUMA)i(mul-)112 384 y(tiprocessors.)19
b(Master)r(')m(s)13 b(thesis,)h(Department)d(of)i(Computer)f(Science,)h
(University)f(of)g(T)m(oronto,)112 431 y(T)m(oronto,)f(CANADA,)i(1992.)4
519 y([11])24 b(J.)f(Garcia,)h(E.)f(A)-5 b(yguade,)26 b(and)c(J.)h(Labarta.)
44 b(A)22 b(novel)g(approach)g(towards)g(automatic)g(data)112
566 y(distribution.)33 b(In)18 b Fe(Pr)n(oc.)i(of)e(the)h(W)-5
b(orkshop)20 b(on)e(Automatic)g(Data)g(Layout)g(and)h(Performance)112
613 y(Pr)n(ediction)p Fo(,)13 b(1995.)4 701 y([12])24 b(M.)16
b(Gupta)f(and)h(P)-6 b(.)17 b(Banerjee.)27 b(Automatic)15 b(data)g
(partitioning)g(on)g(distributed)g(memory)g(multi-)112 748
y(processors.)j Fe(IEEE)c(T)m(rans.)f(on)f(Parallel)h(and)f(Distributed)h
(Systems)p Fo(,)g(3\(2\):179\261193,)e(1992.)4 836 y([13])24
b(K.)15 b(Harzallah)g(and)g(K.C.)h(Sevcik.)25 b(Hot)15 b(spot)g(analysis)g
(in)g(lar)o(ge)g(scale)h(shared)f(memory)f(multi-)112 883 y(processors.)k(In)
12 b Fe(Pr)n(oc.)i(of)e(Super)n(computing'93)p Fo(,)g(pages)h(895\261905.)f
(ACM,)i(1993.)4 971 y([14])24 b(M.)11 b(Heinrich)f(et)h(al.)16
b(The)11 b(Stanford)f(FLASH)g(Multiprocessor.)16 b(In)10 b
Fe(Pr)n(oc.)i(of)f(the)g(21st)g(Int'l)e(Symp.)112 1018 y(on)j(Computer)g(Ar)n
(chitectur)n(e)p Fo(,)j(pages)e(302\261313,)f(1994.)4 1106
y([15])24 b(S.)15 b(Hiranandani,)i(K.)f(Kennedy)m(,)g(and)g(C.)g(T)m(seng.)27
b(Compiler)15 b(optimizations)g(for)g(Fortran)f(D)i(on)112
1153 y(MIMD)f(distributed-memory)e(machines.)25 b(In)15 b Fe(Pr)n(oc.)h(of)f
(Super)n(computing'91)p Fo(,)g(pages)h(86\261100,)112 1200
y(Albuquerque,)c(NM,)h(1991.)4 1288 y([16])24 b(HPF)l(.)33
b(High)17 b(Performance)g(Fortran)g(Language)i(Speci\256cation)e(\(High)g
(Performance)g(Fortran)112 1335 y(Forum\).)f(T)m(echnical)d(report)e
(CRPC-TR92225,)i(Rice)g(University)m(,)f(1994.)4 1424 y([17])24
b(C.)13 b(Koelbel.)18 b(HPF)12 b(constraints.)18 b(Personal)12
b(Communications,)g(1995.)4 1512 y([18])24 b(U.)14 b(Kremer)m(.)23
b(Automatic)14 b(data)g(layout)g(for)g(distributed-memory)e(multiprocessors.)
23 b(T)m(echnical)112 1559 y(report)11 b(CRPC-TR93229-S,)h(Center)h(for)e
(Research)i(on)f(Parallel)g(Computation,)g(1993.)4 1647 y([19])24
b(T)l(.T)l(.)15 b(Kwan,)f(B.K.)h(T)m(otty)m(,)e(and)g(D.A.)h(Reed.)21
b(Communication)12 b(and)i(computation)e(performance)112 1694
y(of)f(the)i(CM5.)19 b(In)12 b Fe(Pr)n(oc.)h(of)g(Super)n(computing'93)p
Fo(,)f(pages)h(192\261201.)f(ACM,)h(1993.)4 1782 y([20])24
b(D.)15 b(Lenoski)h(et)f(al.)26 b(The)15 b(Stanford)f(DASH)h(multiprocessor)m
(.)25 b Fe(IEEE)16 b(Computer)p Fo(,)h(25\(3\):63\26179,)112
1829 y(1992.)4 1917 y([21])24 b(H.)12 b(Li)g(and)g(K.C.)h(Sevcik.)k
(Numacros:)g(Data)12 b(parallel)f(programming)f(on)i(NUMA)g(multiproces-)112
1964 y(sors.)h(In)c Fe(Pr)n(oc.)i(of)e(4th)g(Symp.)h(on)g(Experiences)g(with)
g(Distributed)f(and)g(Multipr)n(ocessor)j(Systems)p Fo(,)112
2011 y(pages)g(247\261263,)h(1993.)4 2099 y([22])24 b(J.)11
b(Li)h(and)f(M.)h(Chen.)k(Compiling)10 b(communication-ef)o(\256cient)f
(programs)i(for)f(massively)h(parallel)112 2146 y(machines.)18
b Fe(Journal)12 b(of)h(Parallel)f(and)h(Distributed)f(Computing)p
Fo(,)g(2\(3\):361\261376,)f(1991.)4 2234 y([23])24 b(Cray)14
b(Research.)25 b(The)16 b(Cray)e(Research)h(Massively)h(Parallel)e(Processor)
g(System)h(-)f(Cray)h(T3D.)112 2281 y(T)m(echnical)d(report)f(80922,)i
(Munchen,)g(Germany)m(,)f(1993.)4 2369 y([24])24 b(Kendall)12
b(Square)f(Research.)19 b Fe(KSR1)13 b(Principles)h(of)e(Operation)p
Fo(.)18 b(W)l(altham,)13 b(MA,)g(1991.)4 2457 y([25])24 b(J.)12
b(T)m(orres,)g(E.)h(A)-5 b(yguade,)13 b(J.)g(Labarta,)f(and)g(M.)h(V)-6
b(alero.)17 b(Align)12 b(and)g(distribute-based)f(linear)g(loop)112
2504 y(transformations.)16 b(In)11 b Fe(Languages)g(and)h(Compilers)h(for)f
(Parallel)g(Computing)p Fo(,)g(pages)g(321\261339.)112 2551
y(Springer)o(-V)-6 b(erlag)10 b(LNCS-768,)j(1993.)4 2639 y([26])24
b(Z.)17 b(V)m(ranesic,)h(M.)f(Stumm,)f(R.)i(White,)f(and)f(D.)h(Lewis.)30
b(The)16 b(Hector)g(Multiprocessor.)29 b Fe(IEEE)112 2686 y(Computer)p
Fo(,)13 b(24\(1\):72\26180,)e(1991.)4 2774 y([27])24 b(R.W)-5
b(.)12 b(W)n(isniewski,)g(L.I.)g(Kontothanassis,)h(and)e(M.L.)h(Scott.)k
(High)11 b(performance)e(synchroniza-)112 2821 y(tion)i(algorithms)h(for)g
(multiprogrammed)e(multiprocessors.)18 b(In)12 b Fe(Pr)n(oc.)h(of)g(PPOPP)p
Fo(,)h(1995.)p eop
%%Trailer
end
userdict /end-hook known{end-hook}if
%%EOF