Art of Illusion

The JavaApplicationStub that is provided to launch AoI for MacOSX is a PPC code which requires Rosetta when using Intel Macs. Now that Snow Leopard has been released from Cupertino zoo, Rosetta is optional and proposed only if a (usually pretty outdated) program needs it. My 2c is that requiring Rosetta for a launcher is a waste of resource and may deter people from using AoI. I suppose AoI is not the only program using JavaApplicationStub, so presumably a global solution will be proposed ?

Francois.

PS : otherwise this new cat is really faster. upgrading a computer for 30$ really is a deal i'd like to make everyday.

Sigh. This used to be the recommended way for launchers to work. The PPC stub was smart enough to check what CPU you had, and if it was Intel, it would simply launch the proper Intel stub. That way you could have a single launcher that worked on both Intel and PPC.

I'll look into this and see what the new recommended approach is.

Peter

I noticed this in my MacPro. The stub detected intel architecture ok but launched AoI in 32bit mode.

I couldn't get AoI to run in 64bit mode on Java6 using the stub. I got 64bit mode ok by launching the jar file directly, but then JOGL didn't work (I think because the JOGL binaries supplied are 32bit?)!

So I solved this by downloading and putting the 64bit JOGL libraries in the java system extensions library. Now I have AoI running properly on the 64bit JVM on a MacPro with hardware acceleration enabled.

Cool that you have AOI running as 64bit.

I, and I'm sure many others, would be interested to hear how that performs?

My expectation is that the raytracer might get a slight boost, although it uses FLOAT (32bit) much, much more than DOUBLE (64bit). My expectation is that the slowdown from increased memory consumption would outweigh the calculation gains quite early.

Q1: How much RAM do you have, BTW?

Q2: What memory size are you using (the -Xmx option)?

Cheers!

Nik

> My expectation is that the raytracer might get a slight boost, although it > uses FLOAT (32bit) much, much more than DOUBLE (64bit).

Actually it uses double for most things (coordinates, ray directions, etc.) The only thing it uses float for is color components. But this isn't important, because you can use both float and double equally well in either 32 bit or 64 bit mode.

What might make it faster in 64 bit mode is that Intel processors have twice as many accessible registers in 64 bit mode, which can allow the JIT to compile things in a more efficient way. On the other hand, all pointers are twice as large, so that slows things down a bit.

Peter

I stand corrected :o)

I've just done some more reading on this.

The increased number of registers in 64 bit mode (on AMD and Intel chips) would certainly provide a gain - but will that be offset in any way by higher costs in context-switching when switching threads? (And Power chips, at least, have the same number of registers in both 32 and 64 bit modes.)

And of course, the greater memory consumption of 64bit not only increases the required memory for the same dataset, but can also reduce the effectiveness of the CPU caches.

So I guess, the primary reason for going 64-bit with AOI would be less for speed, and more for dramatically increased scalability in modelling and rendering of large models - which is only going to happen if you have serious RAM in the machine - 4GB at a minimum.

I'll try to get some figures in the next month regarding the difference in memory size of AOI data between 32 and 64 bit.

Cheers! Nik

I have 2Gb of RAM installed and I use a max of 512Mb of RAM.

I haven't done any performance comparisons. I just wanted it to work in 64bit mode because it 'should'.

This was reported to work with Freemind : http://www.macosxhints.com/article.php?story=20090915085543375

Could it be as simple as updating the java stub ? However, how is the compatibility with PPC Intel be ensured ? Symlink ?

Francois.

Have you tried the latest early access release? I believe this problem is fixed in it.

Peter

I update the code through SVN repo and usually don't try the 'binary' releases. I've tried the .ea release as you suggested and yes, it works !

Francois.