>As the experience of many in this group shows, TYA is in no imminent
>danger of being put out of business by Sun's JIT efforts.
>
I don't understand what everybody's perception is here, but I've grabbed
an old benchmark (CaffeineMark3) and the results are obvious:
JDK1.1.7:
TYA 1.2v4: 811, 1559, 3011, 1753, 1074, 927, 626, 0, 62
JDK1.2v1
TYA 1.3v2: 731, 1422, 2626, 1196, 980, 687, 417, 0, 31
SUNWjit: 1808, 3115, 9149, 1077, 1773, 2355, 457, 0, 32
none: 398, 447, 459, 665, 411, 407, 479, 0, 31
unset = SUNWjit:
JDK1.2v3
SUNWjit: 2038, 3384,10034, 1618, 1915, 2492, 491, 0, 49
(figures for respectively Sieve, Loop, Logic, String, Float, Method,
Graphics, Image and Dialog; larger is better, beats me why the Image
test always returns 0. Native threads, 233Mhz P/MMX).
Looks like the SUNWjit beats TYA by a large margin (and, without putting
down the tremendous effort that TYA represents, I more-or-less expected
that because my feeling is that JIT optimization is for a large part a
matter of applying a lot of raw manpower...).
The perceived differences may be due to the JIT doing the JIT thingy:
analyzing and compiling classes, thus slowing down class loading and user
interface response. If TYA covers the middle ground between interpreted
and a JIT that's fast in execution but slow in analyzing, it may well be
the JIT of choice for UI software, with SUNWjit/Hotspot/... speeding up
long-running server processes.
As usual, the only thing that I'm really saying here is YMMV :-)
--
Cees de Groot http://www.cdegroot.com <[EMAIL PROTECTED]>
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
