Just saw that the JDK1.4.2 beta has been posted. Get it at http://java.sun.com/j2se/1.4.2/download.html
The bugfixes page seems to be down. Anyone have a good experience wit hthis jre? Or are there any compelling reasons to download this beta?
According to the docs, they’ve fixed a lot of the full screen crashes. That’s something I guess.
Incremental download option looks nice… only problem is that I constantly hit server errors for ANY download, be it incremental or full. Probably java.sun.com got slashdotted 
Wow, check out the Release Notes thoroughly…
Some amazing new features in there!
SSE = Streaming SIMD Extensions! (tear rolls down cheek)
.
.
.
Server HotSpot Compiler
The following are enhancements to the server HotSpot compiler in 1.4.2. Also see Virtual Machine release notes for issues.
* Use of SSE and SSE2 instructions on IA32 platforms when they are supported by the processor and operating system.
* Faster integer division by constants.
* Faster floating-point division by powers of 2.
* Faster NullPointerException handling.WOW. i’m seeing some impressive results with the server VM. i’m freaking out. lots of array accesses, integer calcs and floating point calcs happening in something i’m working on. is it SSE/SSE2? are there any array access optimizations? why isn’t SSE in the client?
however, using the server VM, i got an ArrayIndexOutOfBoundsException that doesn’t occur using the client VM. it’s after a long code path with lots of calcs, so being a busy person i’m not sure i’ll be able to track down the problem and submit a bug report soon.
sun guys, is there a way to turn off SSE/SSE2 in the server VM ? anyone know of any other new VM arguments? can anyone give more info on the SSE support?
Some noteworthy changes: General startup time shall be 10%-20% faster. Swing apps start noticable faster. When using the Windows LAF on Windows XP, you’ll get a quite good (better than usual) emulation of the XP look (thanks to some native code it uses the real images right from the Windows XP skinning engine). Client hotspot shall have a lower memory foot print.
After several short (but complex) microbenchmarks I have found a 100% speed increase ( i.e. ~ half the time ) with regards to Matrix4f in the vecmath API as well as with similiar custom float-point operations.
As expected the operations must be mults with adds that are in a series and have independant (i.e. different) variables for there data sources. This would be in line with SIMD ops.
I am hoping set-up a pure C++ vs. Java matrix test benchmark to get an idea of regular Java, Java SIMD, regular C++, and C++ with SIMD vector if I can get my hands on the Intel compiler.
I was just recently researching setting up the Intel SIMD instructions in a special JNI vec api, I wonder if there will little to no gain now! (hope hope)
More when I have it
But this is still server VM only??
This is not installed with a common JRE, right?
Hm, our prototype works well with 1.4.2, but only with client VM. Very jerky with server VM and not faster in any way with server.
Shawn, we’ve got a vector JNI API in the LWJGL but if JDK1.4.2 delivers as promised and fixes floating point performance we’ll be able to ditch it. There were plans to do SIMD in it (an Altivec version was in progress too but still no Mac port of LWJGL in general…)
Cas
Hmm Theme support in xp is a bit flaky ?
http://www.matzon.dk/brian/Random%20Junk/java-theme-screwup-1.png
http://www.matzon.dk/brian/Random%20Junk/java-theme-screwup-2.png
PLEEEEAAASE tell me that’s going to be the new Metal theme! That looks awesome!
morebrackeen:
http://java.sun.com/j2se/1.4.2/relnotes.html#hotspot, says somehing about “-XX:UseSSE=0”. Looks like what you were looking for.
/M