Frank Deutschmann wrote:
>
> Hello all,
>
> Presuming that I can't get a hold of the source code, could someone
> perhaps tell me how the Java3D VecMath package is implemented?
> Specifically, I'm wondering if it makes use of the SIMD floating point
> instructions on P-3 processors, and, more generally, how efficient it
> is/what sort of performance to expect.
You can get the source if you're willing to sign the Sun Community
Source License. I don't fall into that category, but here are some
general observations on the Linux Java3D implementation (which you can
download from the Blackdown site) based on poking around:
- It appears that no vecmath methods are implemented natively; there are
no native methods in the package. There are a few native methods in the
j3dcore classes, but they don't look (at quick glance) to be relevant to
the needs of the vecmath packages.
- It's unlikely anything implemented natively takes advantage of P-3
features.
- JNI overhead is far from free, and native methods need to do
non-trivial amounts of processing to amortize the cost.
So the news doesn't look very good, unless I'm misreading some of what I
see. The good news is that JNI programming isn't very difficult... if
you've already got some native code capable of doing the heavy lifting,
you may find that wiring it into JNI is a reasonable solution.
Nathan
> I'm looking at using Java for a major numerical analysis package, and
> the "write once run anywhere" aspect is naturally quite apealing;
> however, I'm simply unable to give up the traditional huge measure of
> performance that goes with Java -- but the VecMath package seems quite
> redeeming, provided the overhead is minimal. I'm guessing/presuming
> that VecMath is implemented as efficiently as possible for the native
> hardware; do I presume too much? If VecMath itself is implemented
> "close to the hardware", how much penalty does the requisite Jini
> boundary cost?
>
> Further, could anyone comment on the general
> state/availability/stability of Java3D and the associated VecMath
> package?
>
> Thanks in advance for all comments, etc!
>
> -frank, always riding the bleeding edge......
>
> ----------------------------------------------------------------------
> To UNSUBSCRIBE, email to [EMAIL PROTECTED]
> with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
