harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Egor Pasko <egor.pa...@gmail.com>
Subject Re: [drlvm][jit][opt][abcd] Two-state Inequality Graph for both Lower and Upper problems
Date Wed, 18 Jul 2007 10:30:01 GMT
On the 0x317 day of Apache Harmony Vladimir Strigun wrote:
> On 18 Jul 2007 13:58:46 +0400, Egor Pasko <egor.pasko@gmail.com> wrote:
> > On the 0x317 day of Apache Harmony Vladimir Strigun wrote:
> > > Egor,
> > >
> > > On 17 Jul 2007 02:44:25 +0400, Egor Pasko <egor.pasko@gmail.com> wrote:
> > > > Guys,
> > > >
> > > > I am not very stuck with ABCD. I am not. Not at all. Not am I.
> > > > Ehm.
> > > >
> > > > Lickily, I finished digging into the implementation and making sure it
> > > > is correct. Now I am pretty confident that classic_abcd does the right
> > > > thing! (no guarantees, you know, it's software) Had to refactor the
> > > > code a bit to fill in the gaps of my poor understanding. I think, we
> > > > should commit the changes...
> > > >
> > > > featuring:
> > > > * two-state Inequality Graph, dot printing is just beautiful
> > > > * better readability
> > > > * unit-like-tests against the new functionality
> > > > * option: -XX:jit.arg.dump_abcd_stats=true to dump stats (total/eliminated)
> > > > * same amount of checks eliminated as before
> > > > * well-known tests breaking oldish ABCD _passed_, of course
> > > >
> > > > in all, HARMONY-4476 (more details in JIRA)
> > > >
> > > > Given that Windows is not what I am lucky with today, if a JIT guru
> > > > (Pavel, Mikhail, George?) had time to take a look at the patch and run
> > > > 'build test' on Windows that would be really-really great!
> > > >
> > > > And now the ugly porn:
> > > >
> > > > 1. I could not run almost all of DaCapo benches for various reasons, so
> > > >   tested only on hsqldb, wow, anybody aware of it or is it just me
> > > >   ugly little creature? had little time, sorry
> > >
> > > Could you please describe what problems do you have with running Dacapo?
> >
> > for example:
> >
> > [pasko 13:51:19 dacapo]$ $HARMONY -showversion -XX:jit.arg.dump_abcd_stats=true
-Xem:server -jar dacapo-2006-10-MR2.jar antlr
> > Apache Harmony Launcher : (c) Copyright 1991, 2006 The Apache Software Foundation
or its licensors, as applicable.
> > java version "1.5.0"
> > pre-alpha : not complete or compatible
> > svn = r554009, (Jul 17 2007), Linux/ia32/gcc 4.0.3, debug build
> > http://harmony.apache.org
> > ===== DaCapo antlr starting =====
> > Running antlr on grammar antlr/cpp/calc.g
> > ANTLR Parser Generator   Version 2.7.5 (20050128)   1989-2005 jGuru.com
> > [pasko 13:52:03 dacapo]$
> >
> > this is a failure ("PASSED" was not printed)
> > ..but the version is rather old.
> >
> > I am going to file the bugs as soon as I manage to check the latest version.
> 
> Yes, it's well known issue (HARMONY-1906). The problem here that
> Dacapo uses patched version of antlr, while antlr.jar included to
> bootclasspath for vm. So, the workaround here is to remove antlr.jar
> from bootclasspath.properties file.

Very interesting!

Vladimir, as an expert in running benchmarks, could you, please, share
the knowledge about all tricks that you know (including the just
mentioned one) that are necessary to run DaCapo on Wiki?

and publishing performance measurements on Linux would be also good to
start to track progress here. (but this is another subject to discuss)

now on DaCapo eclipse:

$ $HARMONY -showversion -XX:jit.arg.dump_abcd_stats=true -Xem:server -jar dacapo-2006-10-MR2.jar
eclipse
Apache Harmony Launcher : (c) Copyright 1991, 2006 The Apache Software Foundation or its licensors,
as applicable.
java version "1.5.0"
pre-alpha : not complete or compatible
svn = r554009, (Jul 17 2007), Linux/ia32/gcc 4.0.3, debug build
http://harmony.apache.org
The GC did not provide gc_add_weak_root_set_entry()
===== DaCapo eclipse starting =====
Uncaught exception in main:
java.lang.reflect.InvocationTargetException
        at java.lang.reflect.VMReflection.invokeMethod(VMReflection.java)
        at java.lang.reflect.Method.invoke(Method.java:317)
        at org.apache.harmony.vm.JarRunner.main(JarRunner.java:80)
Caused by: java.lang.NoClassDefFoundError: org/eclipse/jdt/internal/launching/StandardVMType
        at dacapo.eclipse.HarnessRunner.run(HarnessRunner.java:19)
        at org.eclipse.core.internal.runtime.PlatformActivator$1.run(PlatformActivator.java)
        at org.eclipse.core.runtime.adaptor.EclipseStarter.run(EclipseStarter.java)
        at dacapo.eclipse.EclipseHarness.iterate(EclipseHarness.java:35)
        at dacapo.Benchmark.run(Benchmark.java)
        at dacapo.TestHarness.runBenchmark(TestHarness.java:302)
        at dacapo.TestHarness.main(TestHarness.java:242)
        at Harness.main(Harness.java:5)
        at java.lang.reflect.VMReflection.invokeMethod(VMReflection.java)
        ... 2 more
Caused by: java.lang.ClassNotFoundException: org.eclipse.jdt.internal.launching.StandardVMType
        at org.eclipse.osgi.framework.internal.core.BundleLoader.findClass(BundleLoader.java:405)
        at org.eclipse.osgi.framework.internal.core.BundleLoader.findClass(BundleLoader.java)
        at org.eclipse.osgi.framework.adaptor.core.AbstractClassLoader.loadClass(AbstractClassLoader.java)
        at java.lang.ClassLoader.loadClass(ClassLoader.java)
        at dacapo.eclipse.HarnessRunner.run(HarnessRunner.java:19)
        ... 10 more

is it known?

> 
> Thanks.
> Vladimir.
> 
> > > I've checked your patch on all dacapo benches and I haven't see any
> > > boost or degradation (release build rev 556923).
> >
> > That is good! There should be no boost or degradation. These changes
> > were made for better maintainability and to more or less prove
> > correctness of the optimization.
> >
> > > Thanks.
> > > Vladimir.
> > >
> > > > 2. ~10% bounds checks removed in hot methods and 8% in total, with
> > > >   ABCD innocent and many other optimizations very very guilty or
> > > >   absent. OMG!
> > > >
> > > > 3. "memopt" is probably the ugliest!! In my example:
> > > >   1. array A lies in a non-volatile field
> > > >   2. A.length is computed right in the method entrance
> > > >   3. A is then loaded via "ldfield" bytecode instruction in the
> > > >      middle of the method
> > > >   4. nothing writes to the field, just accesses elements of A
> > > >
> > > >   And who could imagine that A.length would be computed twice with
> > > >   "memopt" having nothing to do with second appearance A.length?
> > > >   Thus, not eliminating the second appearance of this sequence:
> > > >
> > > >   I9:ldflda    [t1.BidirectionalBubbleSort2::a] -) t4:ref:int32[]
> > > >   I10:ldind.unc:[]  [t4] ((t2,t3)) -) t5:int32[]
> > > >   I11:chknull   t5 -) t6:tau
> > > >   I12:tauhastype      t5,int32[] -) t7:tau
> > > >   I13:arraylen  t5 ((t6,t7)) -) t8:int32
> > > >
> > > >   this is what should be optimized out definitely! and the thing that
> > > >   breaks two A.length-s apart killing the idea of ABCD.
> > > >
> > > > 4. "loop versioning" is not implemented, and this is what I would like
> > > >   to take. I already wrote some 2-component-nullstone-like
> > > >   performance tests to detect how good a JVM deals with loop
> > > >   versioning. Will post the bunch of them soon.
> > > >
> > > > JIT gurus,
> > > >
> > > > given the ugly (1) - (4), rather critical for performance do you like
> > > > the idea to fight them in a high priority? Could you share your
> > > > vision, please?
> > > >
> > > >
> > > > --
> > > > Egor Pasko
> > > >
> > > >
> > >
> >
> > --
> > Egor Pasko
> >
> >
> 

-- 
Egor Pasko


Mime
View raw message