Ignore:
Timestamp:
Jun 18, 2014, 1:07:02 AM (5 years ago)
Author:
cameron
Message:

Reference cleanups; use GPU instead of GPGPU

File:
1 edited

Legend:

Unmodified
Added
Removed
  • docs/Working/re/conclusion.tex

    r3880 r3883  
    2525continue to scale well up for SIMD vectors up to 4096 bytes
    2626in length based on 64-bit additions.    The model also
    27 supports GPGPU implementation with some additional
     27supports GPU implementation with some additional
    2828effort.
    2929
     
    7676grep version with a dynamic code generator implemented with LLVM
    7777is being developed by another team working with the Parabix
    78 technology (Dale Denis, Nigel Medforth, Nick Sumner and Rob Cameron).
     78technology (Dale Denis, Nick Sumner and Rob Cameron).
    7979An initial version is available at
    8080{\tt http://parabix.costar.sfu.ca/icGREP}.
     
    8888properties, for example.
    8989
    90 Future work also includes the dvelopment of multicore versions of the
     90Future work also includes the development of multicore versions of the
    9191underlying algorithms to further accelerate performance and to
    9292handle regular expression matching problems involving larger rule sets than are
    9393typically encountered in the grep problem.  Such implementations
    9494could have useful application in tokenization and network
    95 intrusion detection for example.   Additional GPGPU implementation
     95intrusion detection for example.   Additional GPU implementation
    9696work could take advantage of specialized instructions available
    9797on particular platforms but not generally avaiable through OpenCL.
    98 For both multicore and GPGPU implementations,
     98For both multicore and GPU implementations,
    9999data-parallel division of input streams could benefit from
    100100techniques such as the principled speculation of Zhao et al
Note: See TracChangeset for help on using the changeset viewer.