Parallel programming model, language and compiler in ACA.

  • Published on
    15-Jul-2015

  • View
    176

  • Download
    3

Embed Size (px)

Transcript

  • A programming model is a collection of program

    abstraction providing a programmer a simplified

    and transparent view of computer H/W and S/W.

    Parallel programming model is designed for vector

    computers.

    Fundamental issues in parallel programming.

    Creation, suspension, reactivation, termination.

  • Five model are designed that exploits

    parallelism-:

    Shared-variable model.

    Message-passing model.

    Data parallel model.

    Object oriented model.

    Functional and logic model.

  • In shared variable model parallelism depends on

    how IPC is implemented.

    IPC implemented in parallel programming by two

    ways.

    IPC using shared variable.

    IPC using message passing.

  • IPC with shared variable

    IPC with message passing

  • Critical section.

    Memory consistency.

    Atomicity with memory operation.

    Fast synchronization.

    Shared data structure.

  • Two process communicate with each other by

    passing message through a network.

    Delay caused by message passing is much longer

    than shared variable model in a same memory.

    Two message passing approach are introduced here.

  • Synchronous message passing-:

    Its synchronizes the sender and receiver process

    with time and space just like telephone call.

    No shared memory.

    No need of mutual exclusion.

    No buffer are used in communication channel.

    It can be blocked by channel being busy.

  • Asynchronous message passing-:

    Does not need to synchronize the sender and

    receiver in time and space.

    Non blocking can be achieved.

    Buffer are used to hold the message along the path

    of connecting channel.

    Message passing programming is gradually

    changing, once the virtual memory from all nodes

    are combined.

  • It require the use of pre-distributed data set.

    Interconnected data structure are also needed to

    facilitate data exchange operation.

    It emphasizes local computation and data routing

    operation such as permutation, replication, reduction

    and parallel prefix.

    It can be implemented on either SIMD or SPMD

    multicomputer, depending on the grain size of

    program.

  • Object are created and manipulated dynamically.

    Processing is performed using object.

    Concurrent programming model are built up from

    low level object such as processes, queue and

    semaphore.

    C-OOP achieve parallelism using three methods.

  • Pipeline concurrency.

    Divide and conquer concurrency.

    Co-operating problem solving.

  • Two language-oriented programming for parallel

    processing are purposed.

    Functional programming model such as LISP,

    SISAL, Strand 88.

    Logic programming model as prolog.

    Based on predicate logic, logic programming is

    suitable for solving large database queries.

  • Language feature for parallel programming into six

    categories according to functionality.

    Optimization features

    Used for program restructuring and compilation

    directives.

    Sequentially coded program into parallel code.

    Automated parallelization.

    Semi-automated parallelization.

  • Availability feature

    Its use to enhance the user- friendliness.

    Make language portable to large class of parallel

    computers.

    Scalability.

    Compatibility.

    Portability.

  • Synchronization/ communication feature

    Shared variable for IPC.

    Single assignment language.

    Send/receive for message passing.

    Logical shared memory such as the row space in

    Linda.

    Remote procedure call.

    Data flow languages such as id.

    .

  • Control of parallelism

    Coarse, medium or fine grain.

    Explicit versus implicit parallelism.

    Loop parallelism in iteration.

    Shared task queue.

    Divide and conquer paradigm.

    Shared abstract data type.

  • Data parallelism feature

    It specified how data are accessed and distributed

    Runtime automatic decomposition.

    Mapping specification.

    Virtual processor support.

    Direct access to shared data.

  • Process management features

    These feature are needed to support the efficient

    creation of parallel processes.

    Implementation of multithreading or multitasking.

    Dynamic process creation at runtime.

    Automatic load balancing.

    Light weight processes.

  • Special language construct and data array

    expression for exploiting parallelism in program.

    First is FORTRAN 90 array notation.

    Parallel flow control is achieve using do across and

    do all type of keyword which is use in the

    FORTRAN 90.

    Same we also use FORK and JOIN method.

  • The role of compiler to remove the burden of

    program optimization and code generation.

    A parallelizing compiler consist of the three major

    phases.

    Flow analysis.

    Optimization.

    Code generation.

  • Compilation phases in parallel code generation

  • THANK YOU

Recommended

View more >