This is the mail archive of the java-discuss@sourceware.cygnus.com mailing list for the GCJ project. See the GCJ home page for more information.
Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
---|---|---|
Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |
KGB Software writes: > Does anyone have any feel for how big a speed hit (relative > to C, say) a GCJ-compiled Java program takes? How does GCJ's > performance compare with other native Java compilers (eg., > TowerJ)? We demoed the gcj compiler at the last Linux World. We were running the Linpack benchmark. Here are the results, on a PII266/Linux box (the Mflops are reported by the benchmark itself): jdk1.1.7: 1.8 Mflops gcj, non optimized: 9 Mflops gcj (-O3 -fomit-frame-pointer): 50 Mflops gcj (-O3 -fomit-frame-pointer -fno-bounds-check): 137 Mflops You can find the source code of the benchmark here: http://www.netlib.org/benchmark/linpackjava/ But be cautious with the results, this is just a benchmark. > I've read one paper on Java performance in scientific apps > (http://wwwipd.ira.uka.de/~phlipp/javaCS.ps.gz) that notes that > IBM's "high performance java" compiler has a flag to disable > array bounds checking, which results in large improvements in > scientific apps' performance. Does/will GCJ also allow > array bounds checking to be turned off? Yes. It's `-fno-bounds-check' > And when will GCJ be available to mere mortals? gcj is already part of egcs. If you download egcs and compile is, you have gcj. .--- Alex (www.cygnus.com/~apbianco, apbianco@cygnus.com) | Alexandre Petit-Bianco, Cygnus Solutions Java Team.