Jun 25, 2014, 6:50:45 PM (5 years ago)

Clean out old style files; check in .bbl file with vfill/eject

1 edited


  • docs/Working/re/conclusion.tex

    r3889 r3898  
    3737on acceleration of matching for a single input stream.
    3838Mytkowicz use the SIMD byte-shuffle capabilities found, for example,
    39 in the SSE3 instruction sets to perform small-table parallel lookups for
     39in the SSSE3 instruction sets to perform small-table parallel lookups for
    4040multiple potentially active states of a FSM.    Data parallelism is
    4141achieved by initially considering all possible states at the
    4949However, the Mytkowicz approach is still fundamentally considering
    5050input elements one byte at a time and would be hard pressed to
    51 compete with our reported results of 1-3 CPU cycles per input byte.
     51compete with our reported results of 1-4 CPU cycles per input byte.
    5252It is also dependent on the availability of the SIMD byte-shuffle
    5353operation, which is unavailable in SIMD instructions sets such as
    9292of single characters (bytes).   Each such extension would
    9393replace while-loop iteration with addition and bitwise logic.
    94 The UTF-8 example shows this is possible for repetitions of
    95 variable-length byte sequences having particular synchronizing
    96 properties, for example.
     94For example, 
     95the repetitions of
     96variable-length UTF-8 byte sequences
     97can be compiled without while loops by taking advantage
     98of synchronizing properties of UTF-8.
    98100Future work also includes the development of multicore versions of the
Note: See TracChangeset for help on using the changeset viewer.