icXML is the name of our project to dramatically accelerate
Xerces performance on modern commodity processors by taking
advantage of SIMD and multicore capabilities and parallel bit
stream technology.

We are interested in feedback and interest from developers and
potential users.   We are also interested in identifying a potential
Champion who could help put us on track to become an official
Xerces subproject.

Version 0.8 of icXML has been released and is available together
with a development version on the costar.sfu.ca server.
svn co http://parabix.costar.sfu.ca/svn/icXML/icXML-0.8
svn co http://parabix.costar.sfu.ca/svn/icXML/icXML-devel
Trac browser: http://parabix.costar.sfu.ca/browser/icXML

To get an idea of the performance prospects, here are end-to-end
figures using Xerces-C 3.1.1 and icXML with a GML-to-SVG conversion
application.

Xerces-C 3.1.1

 Performance counter stats for './gml2svg_3_1_1 ../../data/layer/gml-10 out_3':

    24,444,713,630 instructions:u            #    1.83  insns per
cycle         [83.35%]
    13,344,529,298 cycles:u                  #    0.000 GHz
         [83.33%]
        41,915,991 branch-misses:u           #    0.70% of all
branches         [83.33%]
     6,013,112,976 branches:u
         [83.34%]
        81,290,233 L1-dcache-misses:u
         [83.33%]
       153,198,046 L1-icache-misses:u
         [66.73%]

       3.764054961 seconds time elapsed

icXML

 Performance counter stats for './gml2svg_icx ../../data/layer/gml-10 out_3':

    16,470,263,948 instructions:u            #    1.89  insns per
cycle         [83.33%]
     8,707,613,130 cycles:u                  #    0.000 GHz
         [83.33%]
        13,912,341 branch-misses:u           #    0.43% of all
branches         [83.35%]
     3,244,282,034 branches:u
         [83.33%]
        67,380,609 L1-dcache-misses:u
         [83.33%]
        32,141,837 L1-icache-misses:u
         [66.66%]

       2.554010404 seconds time elapsed

icXML experimental version with 2-thread pipeline parallelism

 Performance counter stats for './gml2svg_icx_pipeline
../../data/layer/gml-10 out_3':

    16,544,368,151 instructions:u            #    1.37  insns per
cycle         [84.11%]
    12,060,226,476 cycles:u                  #    0.000 GHz
         [83.93%]
        13,212,826 branch-misses:u           #    0.39% of all
branches         [83.92%]
     3,357,152,226 branches:u
         [83.78%]
        77,941,092 L1-dcache-misses:u
         [83.10%]
        25,757,287 L1-icache-misses:u
         [67.36%]

       2.180680680 seconds time elapsed



Rob Cameron,
CTO, International Characters, Inc
Professor of Computing Science, Simon Fraser University
http://www.international-characters.com/
http://parabix.costar.sfu.ca/

---------------------------------------------------------------------
To unsubscribe, e-mail: c-dev-unsubscr...@xerces.apache.org
For additional commands, e-mail: c-dev-h...@xerces.apache.org

Reply via email to