You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@harmony.apache.org by "Geir Magnusson Jr." <ge...@pobox.com> on 2006/09/22 05:25:06 UTC

[classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

I've been playing a bit with caffinemark, and our numbers for  
everything except graphics are amazing.

However, our graphics numbers are *terrible*.  I was testing on  
Ubuntu.  We need to fix this.  Does anyone with familiarity with our  
swing/awt impl that could point us in the right direction?

geir


---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by "Geir Magnusson Jr." <ge...@pobox.com>.
On Sep 22, 2006, at 8:44 AM, Mikhail Fursov wrote:

> Why to run benchmarks with debug VM? It's useful only for stability  
> testing.

I probably did run release because the numbers actually did beat Sun  
in some places

But I tend to work in debug, so it wouldn't surprise me if I tested  
using that out of forgetfulness.  It happens to old people :)

geir

>
> On 9/22/06, Geir Magnusson Jr. <ge...@pobox.com> wrote:
>>
>>
>> DRLVM, probably debug, Ubuntu 6, Head as of... tuesday night,
>> Brussels time :)
>>
>> I'll repeat today and start logging the results.
>>
>
> -- 
> Mikhail Fursov


---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by Mikhail Fursov <mi...@gmail.com>.
Why to run benchmarks with debug VM? It's useful only for stability testing.

On 9/22/06, Geir Magnusson Jr. <ge...@pobox.com> wrote:
>
>
> DRLVM, probably debug, Ubuntu 6, Head as of... tuesday night,
> Brussels time :)
>
> I'll repeat today and start logging the results.
>

-- 
Mikhail Fursov

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by "Geir Magnusson Jr." <ge...@pobox.com>.
On Sep 22, 2006, at 7:00 AM, Dmitry Durnev wrote:

> Geir, which VM/build did you manage to complete all CaffeineMark  
> tests on?

DRLVM, probably debug, Ubuntu 6, Head as of... tuesday night,  
Brussels time :)

I'll repeat today and start logging the results.

geir

>
> On 9/22/06, Geir Magnusson Jr. <ge...@pobox.com> wrote:
>> I've been playing a bit with caffinemark, and our numbers for
>> everything except graphics are amazing.
>>
>> However, our graphics numbers are *terrible*.  I was testing on
>> Ubuntu.  We need to fix this.  Does anyone with familiarity with our
>> swing/awt impl that could point us in the right direction?
>>
>> geir
>>
>>
>> ---------------------------------------------------------------------
>> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> For additional commands, e-mail: harmony-dev- 
>> help@incubator.apache.org
>>
>>
>
>
> -- 
>
> Dmitry A. Durnev,
> Intel Middleware Products Division
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>


---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by Dmitry Durnev <dm...@gmail.com>.
Geir, which VM/build did you manage to complete all CaffeineMark tests on?

On 9/22/06, Geir Magnusson Jr. <ge...@pobox.com> wrote:
> I've been playing a bit with caffinemark, and our numbers for
> everything except graphics are amazing.
>
> However, our graphics numbers are *terrible*.  I was testing on
> Ubuntu.  We need to fix this.  Does anyone with familiarity with our
> swing/awt impl that could point us in the right direction?
>
> geir
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>


-- 

Dmitry A. Durnev,
Intel Middleware Products Division

---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by Oleg Khaschansky <ol...@gmail.com>.
They definitely use mitshm, see "Local X Server Performance
Improvements" chapter in [1]. Also they do use xrender, see [2]. Maybe
they also do use something else, like EXA, I don't know...

[1] http://java.sun.com/products/java-media/2D/perf_graphics.html
[2] http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6307603

On 9/22/06, Geir Magnusson Jr. <ge...@pobox.com> wrote:
>
> On Sep 22, 2006, at 9:29 AM, Oleg Khaschansky wrote:
>
> > Current implementation of composite operations (alpha blending) on
> > linux is just a general unacceleratred code, these operations are
> > performed on per-pixel basis and, definitely, they are rather slow.
> > But the main reason is that pixel data should be retrieved from the
> > screen to perform the blending. This screen->client memory data
> > transfer is painfully slow on X. Probably, it is the most significant
> > performance gap of the current linux implementation.
> > Maybe, it is possible to improve it in the most general cases by
> > implementing a codepath which uses xrender extension, which will allow
> > to eliminate screen->memory copying, since xrender supports
> > transparency.
>
> I wonder then how sun goes so fast - the thing is that compared to
> sun, it's slow.
>
> >
> > Probably, it is also possible to accelerate blit operations by adding
> > the code which uses mitshm extension.
> >
> > Basic rendering w/o transparency theoretically should be much faster
> > then the transparent case.
> >
> > These problems are related to linux implementation only.
> >
> > On 9/22/06, Geir Magnusson Jr. <ge...@pobox.com> wrote:
> >> Definitely.  I will do it today and add a page to our "how are we
> >> doing?" page so we can track over time.
> >>
> >> (That and any other benchmark we're allowed to use and publish).
> >>
> >> I think that what we have is mighty impressive. (It's just the
> >> graphics numbers that are bad...)
> >>
> >> geir
> >>
> >> On Sep 22, 2006, at 4:47 AM, Paulex Yang wrote:
> >>
> >> > Geir Magnusson Jr. wrote:
> >> >> I've been playing a bit with caffinemark, and our numbers for
> >> >> everything except graphics are amazing.
> >> > Convenient to post some results vs RI? I believe many ones(includes
> >> > myself of course) would be very interest to see the numbers:)
> >> >>
> >> >> However, our graphics numbers are *terrible*.  I was testing on
> >> >> Ubuntu.  We need to fix this.  Does anyone with familiarity with
> >> >> our swing/awt impl that could point us in the right direction?
> >> >>
> >> >> geir
> >> >>
> >> >>
> >> >>
> >> ---------------------------------------------------------------------
> >> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> >> To unsubscribe, e-mail: harmony-dev-
> >> unsubscribe@incubator.apache.org
> >> >> For additional commands, e-mail: harmony-dev-
> >> >> help@incubator.apache.org
> >> >>
> >> >>
> >> >
> >> >
> >> > --
> >> > Paulex Yang
> >> > China Software Development Lab
> >> > IBM
> >> >
> >> >
> >> >
> >> >
> >> ---------------------------------------------------------------------
> >> > Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> > To unsubscribe, e-mail: harmony-dev-
> >> unsubscribe@incubator.apache.org
> >> > For additional commands, e-mail: harmony-dev-
> >> help@incubator.apache.org
> >> >
> >>
> >>
> >> ---------------------------------------------------------------------
> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> >> For additional commands, e-mail: harmony-dev-
> >> help@incubator.apache.org
> >>
> >>
> >
> > ---------------------------------------------------------------------
> > Terms of use : http://incubator.apache.org/harmony/mailing.html
> > To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> > For additional commands, e-mail: harmony-dev-help@incubator.apache.org
> >
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>

---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by "Geir Magnusson Jr." <ge...@pobox.com>.
On Sep 22, 2006, at 9:29 AM, Oleg Khaschansky wrote:

> Current implementation of composite operations (alpha blending) on
> linux is just a general unacceleratred code, these operations are
> performed on per-pixel basis and, definitely, they are rather slow.
> But the main reason is that pixel data should be retrieved from the
> screen to perform the blending. This screen->client memory data
> transfer is painfully slow on X. Probably, it is the most significant
> performance gap of the current linux implementation.
> Maybe, it is possible to improve it in the most general cases by
> implementing a codepath which uses xrender extension, which will allow
> to eliminate screen->memory copying, since xrender supports
> transparency.

I wonder then how sun goes so fast - the thing is that compared to  
sun, it's slow.

>
> Probably, it is also possible to accelerate blit operations by adding
> the code which uses mitshm extension.
>
> Basic rendering w/o transparency theoretically should be much faster
> then the transparent case.
>
> These problems are related to linux implementation only.
>
> On 9/22/06, Geir Magnusson Jr. <ge...@pobox.com> wrote:
>> Definitely.  I will do it today and add a page to our "how are we
>> doing?" page so we can track over time.
>>
>> (That and any other benchmark we're allowed to use and publish).
>>
>> I think that what we have is mighty impressive. (It's just the
>> graphics numbers that are bad...)
>>
>> geir
>>
>> On Sep 22, 2006, at 4:47 AM, Paulex Yang wrote:
>>
>> > Geir Magnusson Jr. wrote:
>> >> I've been playing a bit with caffinemark, and our numbers for
>> >> everything except graphics are amazing.
>> > Convenient to post some results vs RI? I believe many ones(includes
>> > myself of course) would be very interest to see the numbers:)
>> >>
>> >> However, our graphics numbers are *terrible*.  I was testing on
>> >> Ubuntu.  We need to fix this.  Does anyone with familiarity with
>> >> our swing/awt impl that could point us in the right direction?
>> >>
>> >> geir
>> >>
>> >>
>> >>  
>> ---------------------------------------------------------------------
>> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> >> To unsubscribe, e-mail: harmony-dev- 
>> unsubscribe@incubator.apache.org
>> >> For additional commands, e-mail: harmony-dev-
>> >> help@incubator.apache.org
>> >>
>> >>
>> >
>> >
>> > --
>> > Paulex Yang
>> > China Software Development Lab
>> > IBM
>> >
>> >
>> >
>> >  
>> ---------------------------------------------------------------------
>> > Terms of use : http://incubator.apache.org/harmony/mailing.html
>> > To unsubscribe, e-mail: harmony-dev- 
>> unsubscribe@incubator.apache.org
>> > For additional commands, e-mail: harmony-dev- 
>> help@incubator.apache.org
>> >
>>
>>
>> ---------------------------------------------------------------------
>> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> For additional commands, e-mail: harmony-dev- 
>> help@incubator.apache.org
>>
>>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>


---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by Oleg Khaschansky <ol...@gmail.com>.
Current implementation of composite operations (alpha blending) on
linux is just a general unacceleratred code, these operations are
performed on per-pixel basis and, definitely, they are rather slow.
But the main reason is that pixel data should be retrieved from the
screen to perform the blending. This screen->client memory data
transfer is painfully slow on X. Probably, it is the most significant
performance gap of the current linux implementation.
Maybe, it is possible to improve it in the most general cases by
implementing a codepath which uses xrender extension, which will allow
to eliminate screen->memory copying, since xrender supports
transparency.

Probably, it is also possible to accelerate blit operations by adding
the code which uses mitshm extension.

Basic rendering w/o transparency theoretically should be much faster
then the transparent case.

These problems are related to linux implementation only.

On 9/22/06, Geir Magnusson Jr. <ge...@pobox.com> wrote:
> Definitely.  I will do it today and add a page to our "how are we
> doing?" page so we can track over time.
>
> (That and any other benchmark we're allowed to use and publish).
>
> I think that what we have is mighty impressive. (It's just the
> graphics numbers that are bad...)
>
> geir
>
> On Sep 22, 2006, at 4:47 AM, Paulex Yang wrote:
>
> > Geir Magnusson Jr. wrote:
> >> I've been playing a bit with caffinemark, and our numbers for
> >> everything except graphics are amazing.
> > Convenient to post some results vs RI? I believe many ones(includes
> > myself of course) would be very interest to see the numbers:)
> >>
> >> However, our graphics numbers are *terrible*.  I was testing on
> >> Ubuntu.  We need to fix this.  Does anyone with familiarity with
> >> our swing/awt impl that could point us in the right direction?
> >>
> >> geir
> >>
> >>
> >> ---------------------------------------------------------------------
> >> Terms of use : http://incubator.apache.org/harmony/mailing.html
> >> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> >> For additional commands, e-mail: harmony-dev-
> >> help@incubator.apache.org
> >>
> >>
> >
> >
> > --
> > Paulex Yang
> > China Software Development Lab
> > IBM
> >
> >
> >
> > ---------------------------------------------------------------------
> > Terms of use : http://incubator.apache.org/harmony/mailing.html
> > To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> > For additional commands, e-mail: harmony-dev-help@incubator.apache.org
> >
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>

---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by "Geir Magnusson Jr." <ge...@pobox.com>.
Definitely.  I will do it today and add a page to our "how are we  
doing?" page so we can track over time.

(That and any other benchmark we're allowed to use and publish).

I think that what we have is mighty impressive. (It's just the  
graphics numbers that are bad...)

geir

On Sep 22, 2006, at 4:47 AM, Paulex Yang wrote:

> Geir Magnusson Jr. wrote:
>> I've been playing a bit with caffinemark, and our numbers for  
>> everything except graphics are amazing.
> Convenient to post some results vs RI? I believe many ones(includes  
> myself of course) would be very interest to see the numbers:)
>>
>> However, our graphics numbers are *terrible*.  I was testing on  
>> Ubuntu.  We need to fix this.  Does anyone with familiarity with  
>> our swing/awt impl that could point us in the right direction?
>>
>> geir
>>
>>
>> ---------------------------------------------------------------------
>> Terms of use : http://incubator.apache.org/harmony/mailing.html
>> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
>> For additional commands, e-mail: harmony-dev- 
>> help@incubator.apache.org
>>
>>
>
>
> -- 
> Paulex Yang
> China Software Development Lab
> IBM
>
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>


---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org

Re: [classlib][swing/awt] caffinemark shows that our graphics are sloooowwwww....

Posted by Paulex Yang <pa...@gmail.com>.
Geir Magnusson Jr. wrote:
> I've been playing a bit with caffinemark, and our numbers for 
> everything except graphics are amazing.
Convenient to post some results vs RI? I believe many ones(includes 
myself of course) would be very interest to see the numbers:)
>
> However, our graphics numbers are *terrible*.  I was testing on 
> Ubuntu.  We need to fix this.  Does anyone with familiarity with our 
> swing/awt impl that could point us in the right direction?
>
> geir
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
> For additional commands, e-mail: harmony-dev-help@incubator.apache.org
>
>


-- 
Paulex Yang
China Software Development Lab
IBM



---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org