-Dtranslaccel=true slows down AA graphics

[Mikael Grev]

Are anyone at Sun's Java2D team aware of this problem?

If I turn on 'translaccel' graphics primitive rendering slows down by about 30%, if it is anti-aliased.

It is a regression since it's in 1.5 (final) but not 1.4.2_05.

If it's a known problem, I won't nag about it, but if it's not I can make you a test case and investigate a little further. Maybe even have a battle with the bug parade..

I'm on:
Windows XP sp2
ATI FireGL X2 256t
AMD FX-51

Cheers,
Mikael Grev

ps. Filed this on java.sun.com/products/java-media/2d/mail/java2d-comments.html a couple of weeks ago, but there was no answer.

[trembovetski]

>  If I turn on 'translaccel' graphics primitive rendering slows down by about 30%, if it is anti-aliased.

If you're rendering non-accelerated primitives (and AA-rendering is one of those) to an accelerated surface, we have to use our software loops to do therendring, causing us to read-modify-write to video memory, and reading from video memory is very slow.

So we have a scheme for detecting such occasions, which we call the punting mechanism. Once we detect such event, the accelerated surface gets moved off of video memory to system memory, and software rendering to that surface is much faster.

So in your case I suspect what happens is that as a side effect of enabling translaccel this punting mechanism is disabled, so the surface you rendering to never gets punted out of video memory, and reads from video memory cause the slowdown.
The reason for disabling the punting mechanism when translaccel is enabled is a separate issue.. I don't remember it exactly, but I recall that we had to =)

So, basically we know about this problem and there's no way to fix it until we have all our rendering to accelerated surface accelerated, sorry.

BTW, we haven't seen your email on the dev list on this subject..

[Mikael Grev]

> So, basically we know about this problem and there's no way to fix it until we have all our rendering to accelerated surface accelerated, sorry.

Well, hurry up then!    

Cheers

Games published by our own members! Check 'em out!

[Mikael Grev]

> BTW, we haven't seen your email on the dev list on this subject..

Strange, I put it there I'm quite sure, after suggestions from Chet.

Anyway, this forum and javadesktop.org's forum is actually working well for these kinds of things, with good support from you guys. Can't the Swing and Java2D teams raid the bug parade guys, giving them "wedges" until they drag the bug/rfe parade into this century?

Cheers,
Mikael Grev

[trembovetski]

Well, I've read your blog and kind of have an idea of what you think should be done to the bug parade.. File an rfe =)