[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-java
Subject:    Re: HotSpot (was Re: Linux jdk 1.,2 Jit )
From:       Michael Emmel <memmel () tampabay ! rr ! com>
Date:       1999-02-26 2:50:09
[Download RAW message or body]

Here is a email I sent to Sun concering there licnese. It seems to me
the Sun Community Licens precludes me distributing  my work if I include Sun
source.
I see no easy answer here.

Mike



I have a question.

I have developed a windowing System written almost entirely in Java.
Doing this I replaced most of the platform dependent parts of Sun's java
implementation
with java code.

In doing so I found several areas in the JDK where changes would help my
implementation.
In particular I have a bootstrap driver which uses the AWT for graphics.

This driver requires a small changes to Suns source or the addition of a
new public class.

 Windowing Systems are not exactly a hot seller. Most platforms today
have one.
So the commercial value is fairly low at best the embedded market may be
interested.

The one market where it may find a substantial following is as a
alternative windowing system for linux
and other experimental OS's but this requires me to be able to freely
distribute it.

My concern s that your current license is such that it precludes the
distribution of a technology
such as my  java windowing system to it's target market.

Since I  distribute the system for free in hopes of gaining "mind share"
I'm not in a position to pay royalties.
Also since this is true Java based technology it can not hurt Java
considering the only gui programs which it
runs must be written in java ( for the most part  It can interface with
X11. )

If this project or one like it becomes successful it will only help the
Java platform.

But I see no way to release it under your current license.

Hotspot ???

Extending my all java windowing system down to the graphics layer
requires a high speed Java implementation.
I have already written and tested the performance of Java video drivers
and they are usable but about 10-20 times slower than there
C counterparts.  JIT'S make the drivers usable but certainly not the
fastest or even close to native drivers.
Most of the problems lie in Slow  memory copy calls and lack of a memset
function in java.
Using native code via the JNI does not help much because the overhead of
a JNI call precludes its use.
Most code of the nature  in C contains a lot of macros for inlining. In
fact a lot is still written in assembler.

Hotspot from what I read has the exactly the  characteristics to allow
one to write portable graphics code in java and
still get great performance. I' extremely interested in it. another
project is extending hot spot to allow it to replace java byte code
not only with native machine code but also graphics hardware based
acceleration. This extends hotspot to not only compile java code for
your particular processor but also optimize the graphics part for your
particular graphics subsystem.

I plan on doing this by adding the idea of a signature compiler to
hotspot. This hotspot is aware that the method

graphics.drawLine() is required to meet a certain spec. Thus it can
replace the drawLine method with a hardware accelerated version.
Since both have the same "signature". This is quit different from
translation of the byte code to machine code.
The signature compiler thus accepts on faith the the native code snippet
is a correct replacement for the java code.
Since it contains calls to a foreign processor  the graphics card it can
never compile such code.
Thus the code normally found in lowlevel drivers and graphics libraries
is free to be inlined in the java code.
Also since a basice java implemenation for a dumb linear frame buffer is
provided. The native code need only override
mthods wich can be effectivly acclerated by the particular graphics
hardware. This is similar to the X11 sample server
but removes the eleborate interfaces required to seperate
machine/platform dependent code behind a portability interface.
I beleive that this is a fairly basic and intersing reasearch topic. A
solution in the area of graphics should help the
Java community as a whole in may areas. Currently the dumb frame buffer
design can be ported to a new platform in
about a week. This immediatly gives a fully funtional windowing system.
Assuming fiarly standard keyboard/mose input devices.
I currently try to maintian thre drivers. One ont top of the AWT. On on
X11 and a hardware port to the CirrusLogic 5430.
I plan to look and the new Linux frame buffer devics and a project
called GGI.  None of these solutions allows the
removal of the rigid hardware/graphics lib/java interfaces like a
extended hotspot would.


I do not see that your current license allows me to achieve these goals
unless you have have made exceptions for
free software that enhances the java community goals.

Also I'm very worried that Hotspot will not ever be available  for
research use.  I think that may work may be useful
to the entire java community but I feel that your current license  in
this case restricts the advancement of the java platform
for the benefit of all and I fear that  Sun will not open up the hotspot
technology for research use and thus hinder
my future research goals.

Hopefully I'm mistaken.
If there is any interest at Sun in my Windowing system its called Shark
Windows
It freely available  via CVS at www.gjt.org
cvs directory java/org/gjt/shark.

Thanks for your time.

Michael Emmel






Oliver Fels wrote:

> > Sorry but that whole "hot spot" thing looks like vapor ware. Besides
> > even if Sun is able to get "something" out the door they have already
> > stated that it will be given to "paying customers only". Instead of
>
> Easy answer on this: Letīs wait and see.
>
> > waiting and hoping that Sun will do us a favor why not help out with
> > the free JIT projects like TYA? I have already found a bug in TYA
> As I said, if there is a JIT for Linux which overruns all benchmarks
> and which comes close to what HostSpot is promised to do, I will
> be glad to use it.
>
> > and I plan on making a small test case so that I can track down
> > the problem or send it to the TYA folks. If you help out with the
> > free Java projects then free Java will get better than
> > "paying customers only" Java that Sun wants you to buy into.
> Sun does nothing for pure charity reasons, this is absolutely clear.
> JAVA Linux has to compete with other platforms to be accepted
> and I am not sure if it can do it by relying on free teams only.
> >From my point of view, I am managing projects which heavily rely
> on performance and would love to pay for a Linux HotSpot like
> runtime.
>
> In the meantime, we will give TYA a try as standard JIT for testing
> purposes.
>
> Oliver
>
> ___________________________________________________
> Oliver Fels                    | e-mail:
> Neurotec Hochtechnologie GmbH  | of@neurotec.de
> Team Manager JAVA-/IT-Security | Phone:
> Friedrichshafen, Germany       | +49 7541 3012 351
> ---------------------------------------------------
>
> ----------------------------------------------------------------------
> To UNSUBSCRIBE, email to java-linux-request@java.blackdown.org
> with a subject of "unsubscribe". Trouble? Contact listadm@java.blackdown.org




----------------------------------------------------------------------
To UNSUBSCRIBE, email to java-linux-request@java.blackdown.org
with a subject of "unsubscribe". Trouble? Contact listadm@java.blackdown.org

[prev in list] [next in list] [prev in thread] [next in thread]