Jikes RVM

Dear Carry -- Almost all performance management like this is a statistical
game of one kind or another. But beyond that, you're essential asking the
question: What is the source of creativity and insight in the human mind?
I'm afraid there's no "turn the crank" answer for that!

-- Eliot

> 
> I'm trying to rebuild Jikes RVM 2.0.3 to revisit some research 
> performed by others. I'm trying to find the required jlibraries, and
> though I've searched through sourceforge, svn, the lists, google, 
> etc. I can't find where they might be hiding. Here is the message 
> I'm getting when running jconfigure: 

Hi Kevin,

        The jlibraries that old versions of Jikes RVM use were based on 
some IBM internal class libraries and were not open source.  Since they 
aren't open source software, when the project moved from IBM's 
developerworks site to sourceforge in 2005 we no longer had any place to 
distribute them from.  But, we kept a complete archive of all the versions 
at Watson with the plan of providing them directly to anyone who needed 
them. I think this is actually the first such request...
 
        In any case, I'll send you a copy of the license agreement (they 
aren't open source, so there's a separate license you need to agree to) 
off list and then once you send that back to me I can send you the 
particular version (R-2002-02-27-22-32-59) of the library you need.

--dave

David P Grove wrote:
>
> Hi,
>
>         SSE2 exploitation has been implemented for Jikes RVM and we're 
> contemplating completely removing all of the old 387 based FP support. 
>  In other words, Jikes RVM on IA32 would require SSE2 and would not 
> run on older (pre 2001 for Intel, 2003 for AMD) IA32 implementations. 
>   Is there anyone who is tracking the current versions of Jikes RVM 
> and needs to run on non-SSE2 capable hardware?  If so, please let us 
> know.
It's probably worth listing some processors and their SSE capabilites [1]:

Intel CPUs:
Pentium, Pentium Pro, Pentium 2, Celeron: no SSE
Pentium 3, Celeron-II: SSE1
Pentium 4/M, Celeron D/M,  Core, Core2: SSE2+

AMD CPUs:
K6, Duron, Athlon: no SSE
Sempron: SSE1
Opteron: SSE2
Athlon 64, Sempron, Turion: SSE3

Most Via CPUs have SSE2 and SSE3
Most Cyrix CPUs don't
Transmeta Efficeon processors have SSE2, not sure for Crusoe

>         The current plan is most likely to make a 2.9.1 release that 
> has SSE2 as the default, but still has unoptimized 387 FP as a 
> build-time option.  After the 2.9.1 release we would remove 387 FP 
> entirely unless people are really still depending on it.

If SSE1 is interesting we could generate SSE1 instructions for floats 
and x87 instructions for doubles. For float to integer casts it would be 
nice to use SSE3 when available. There are a number of situations when 
only x87 will do:
floating point remainders
float/double to 64bit long
 >64bit precision
floating point math intrinsics (sin, cos, ...)

I'm sure there are more. For my needs I do require >64bit precision on 
occasions, for example some times when emulating PowerPC's >64bit 
internal precision.

I'm sure people can add to/correct what I've said. Regards,

Ian

[1] http://en.wikipedia.org/wiki/Category:X86_microprocessors

I would try running "ant very-clean". It looks like you aren't
regenerating your configuration.

As you are aware, assertions are enabled or disabled base on the
configuration you are running.

prototype, prototype-opt, development, Base*** and Full***
configurations will have assertion checking enabled.

production and Fast*** configurations do not.


Cheers,
Daniel.

-----Original Message-----
From: jikesrvm-researchers-bounces@...
[mailto:jikesrvm-researchers-bounces@...] On Behalf Of
viz
Sent: Monday, 25 June 2007 11:14 AM
To: jikesrvm-researchers@...
Subject: [Jikesrvm-researchers] Assertions never seem to be checked


The assertions in any of my code always seem to be skipped at runtime,
which
I have determined to be because VM.VERIFY_ASSERTIONS seems to be false.

I have guarded my assertions in any piece of code I write for the RVM
with
"if (VM.VERIFY_ASSERTIONS)" which results in the assertion being skipped
because VM.VERIFY_ASSERTIONS is false.  If I understand correctly, the
default setting for VM.VERIFY_ASSERTIONS is true if config.assertions is
"normal" or "extreme."  Also, the default should be "normal."  I have
tried
specifying this option at command line and in the file specified by
config.name, but no matter what, VM.VERIFY_ASSERTIONS is set to false in
the
RVM, even for the plans that are distributed with the RVM.

What do I need to do to have assertions checked in my code?

Zvi Effron
--=20
View this message in context:
http://www.nabble.com/Assertions-never-seem-to-be-checked-tf3973833.html
#a11280080
Sent from the jikesrvm-researchers mailing list archive at Nabble.com.


------------------------------------------------------------------------
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
Jikesrvm-researchers mailing list
Jikesrvm-researchers@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers

Daniel Frampton wrote:
> 
> I would try running "ant very-clean". It looks like you aren't
> regenerating your configuration.
> 
I tried running another build with very-clean, and during the build, some
assertions failed, but I fixed those issues.  Reran the build with very
clean, but during runtime, assertions do not seem to be checked.  (I put in
some VM.assertion.fail() statements that never seem to be triggered as well
as some guarded VM.assertions._assert(false) statements)  Am I correct that
assertions should be checked at runtime?  Also, other than assertions, is
there a good method inside the VM to output strings other than
System.out.println() (which is technically a java object)?

Thanks,
Zvi
-- 
View this message in context: http://www.nabble.com/Assertions-never-seem-to-be-checked-tf3973833.html#a11291352
Sent from the jikesrvm-researchers mailing list archive at Nabble.com.

I am wondering if there's a confusion here between VM assertions and
ordinary user assertions? I think these are not the same thing ....

-- Eliot

Eliot Moss wrote:
> 
> I am wondering if there's a confusion here between VM assertions and
> ordinary user assertions? I think these are not the same thing ....
> 
> -- Eliot
> 
That very well could be the issue.  So I guess part of my question is what
are VM assertions supposed to be used to check?

--Zvi Effron

-- 
View this message in context: http://www.nabble.com/Assertions-never-seem-to-be-checked-tf3973833.html#a11291868
Sent from the jikesrvm-researchers mailing list archive at Nabble.com.

viz wrote:
> That very well could be the issue.  So I guess part of my question is what
> are VM assertions supposed to be used to check?
>
> --Zvi Effron
>   
Hi Zvi,

we've not finished the support for Java 1.4 style assertions. Imo it's 
not a big job to make it work, we just need to process a command line 
option and make sure that it propagates into 
Class.desiredAssertionStatus - the easiest way is probably to switch 
from our own java.lang.Class to that in GNU Classpath. In the meantime 
you can modify the behaviour of regular assertions by modifying line 92 
of libraryInterface/Common/src/java/lang/Class.java . By default we say 
the desired assertion status is true, so assertions are checked. Which 
is interesting as it could be costing us runtime performance :-)

Regards,
Ian

On 6/22/07, Matt Horsnell <horsnelm@...> wrote:
>
> I'd like to cross compile jikes to target a powerPC simulator.
>
> I've managed to successfully build jikes from the repository using the
> notes
> at http://docs.codehaus.org/display/RVM/Quick+Start+Guide.
>
> I've got a powerPC crosstool chain compiled and working, and i'd ideally
> like to crosscompile on my ia32-linux box to target a ppc32-linux
> simulator.
>
> a) is this possible?


Not sure. You would need to set the host and target to ppc32-linux and
update build/host/ppc32-linux.properties so that it reflects the cross
compile tool chain.


b) are there any notes? I've tried to follow the cross-compile page
> (http://docs.codehaus.org/display/RVM/Cross-Platform+Building) but get the
> error below.


That is an error when we try to auto build classpath. Try cross-compiling
class path (if it is possible?) then copying the files

classpath.jar         libjavaio.so              libjavanet.so   src.jar
constants.properties  libjavalang.so            libjavanio.so
libgjsmalsa.so        libjavalangmanagement.so  libjavautil.so
libgtkpeer.so         libjavalangreflect.so     libjawt.so

into a directory

${components,dir}/classpath/95/ppc32-linux/lib

adding property "
ppc32-linux.classpath.lib.dir=path/to/components/classpath/95/ppc32-linux/lib"
to ${components,dir}/components.properties

and then starting the build.

Wouldn't mind hearing how it turns out ;)

-- 
Cheers,

Peter Donald

In the past anyway, one did not need special cross-compile tool chains. You
build the RVM image ON the ia32 machine, but FOR ppc (i.e., it has compiled
code in it for the PPC), and then build the boot image runner on the PPC
side. I am aware that people have used cross-compilation tools in the past,
but that was because they had limited platforms that could not build the
boot image runner.

Exactly HOW to do all of this with the new Ant build procedure is a
question I can't answer, but it would not surprise if there are people
doing it all the time ...

-- Eliot

On 6/22/07, Eliot Moss <moss@...> wrote:
>
> In the past anyway, one did not need special cross-compile tool chains.
> You
> build the RVM image ON the ia32 machine, but FOR ppc (i.e., it has
> compiled
> code in it for the PPC), and then build the boot image runner on the PPC
> side. I am aware that people have used cross-compilation tools in the
> past,
> but that was because they had limited platforms that could not build the
> boot image runner.


This is still possible in the current build system via something like

# On host
$ ant -Dconfig.name=prototype-opt -Dhost.name=ia32-linux -
Dtarget.name=ppc32-linux cross-compile-host
# On target
$ ant -Dconfig.name=prototype-opt -Dhost.name=ppc32-linuxcross-compile-target

-- 
Cheers,

Peter Donald

Ah right thanks. I think I can probably just ignore these cases then.

Daniel Frampton wrote:
> These are instructions inserted by the optimizing compiler. 
>
> Looking in org.jikesrvm.compilers.opt.OPT_Constants you can see that -7
> is SYNCHRONIZED_MONITOREXIT_BCI.
>
> So that instruction is a monitor exit, it is inserted by the method:
>
> org.jikesrvm.compilers.opt.ir.OPT_GenerationContext#completeEpilogue(Z)
>
> When generating the epilogue of a synchronized method.
>
> Cheers,
> Daniel. 
>
> -----Original Message-----
> From: jikesrvm-researchers-bounces@...
> [mailto:jikesrvm-researchers-bounces@...] On Behalf Of
> Iain Walsh
> Sent: Wednesday, 20 June 2007 10:56 PM
> To: jikesrvm-researchers@...
> Subject: [Jikesrvm-researchers] Inlining Sequence
>
> For OPT_InlineSequence, why might the toString report the bytecode index
>
> for some of the inlined calls as negative numbers? As currently it's 
> telling me some methods were called from bytecode -7.
>
>
> ------------------------------------------------------------------------
> -
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jikesrvm-researchers@...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jikesrvm-researchers@...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>

FYI. Monitor entry/exits are expanded into the calls you are seeing
inlined in org.jikesrvm.compilers.opt.OPT_ExpandRuntimeServices.

Cheers,
Daniel.

-----Original Message-----
From: jikesrvm-researchers-bounces@...
[mailto:jikesrvm-researchers-bounces@...] On Behalf Of
Iain Walsh
Sent: Wednesday, 20 June 2007 11:03 PM
To: General discussion of Jikes RVM design, implementation, issues, and
plans
Subject: Re: [Jikesrvm-researchers] Inlining Sequence

Ah right thanks. I think I can probably just ignore these cases then.

Daniel Frampton wrote:
> These are instructions inserted by the optimizing compiler.=20
>
> Looking in org.jikesrvm.compilers.opt.OPT_Constants you can see that
-7
> is SYNCHRONIZED_MONITOREXIT_BCI.
>
> So that instruction is a monitor exit, it is inserted by the method:
>
>
org.jikesrvm.compilers.opt.ir.OPT_GenerationContext#completeEpilogue(Z)
>
> When generating the epilogue of a synchronized method.
>
> Cheers,
> Daniel.=20
>
> -----Original Message-----
> From: jikesrvm-researchers-bounces@...
> [mailto:jikesrvm-researchers-bounces@...] On Behalf
Of
> Iain Walsh
> Sent: Wednesday, 20 June 2007 10:56 PM
> To: jikesrvm-researchers@...
> Subject: [Jikesrvm-researchers] Inlining Sequence
>
> For OPT_InlineSequence, why might the toString report the bytecode
index
>
> for some of the inlined calls as negative numbers? As currently it's=20
> telling me some methods were called from bytecode -7.
>
>
>
------------------------------------------------------------------------
> -
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jikesrvm-researchers@...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>
>
------------------------------------------------------------------------
-
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jikesrvm-researchers@...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>  =20


------------------------------------------------------------------------
-
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
Jikesrvm-researchers mailing list
Jikesrvm-researchers@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers

On 6/19/07, viz <zeffron@...> wrote:
>
> I now have a setup so my package will be included in every build of the
> RVM,
> thanks.  If I wanted to make it optionally included, for example, by
> requiring the property config.include.package to be set to true to include
> it, and false otherwise, how would I add the check for that property?


It depends on your particular extension but most likely you would include a
boolean value in
rvm/src-generated/vm_configuration/VM_Configuration.template that is set by
passing in values from build.xml:729, guard all your extra code using this
flag and optionally include the new primordial classes in a way similar to
how we do it for opt compiler in build.xml:993

-- 
Cheers,

Peter Donald