Comments on: Quick initial Shark JIT on ARM benchmark http://labb.zafena.se/?p=282 Wed, 08 Sep 2010 23:25:25 +0000 http://wordpress.org/?v=2.6.1 By: Damon Hart-Davis http://labb.zafena.se/?p=282#comment-5081 Damon Hart-Davis Thu, 05 Nov 2009 16:37:50 +0000 http://labb.zafena.se/?p=282#comment-5081 Testing started... Please fit protecting clothing and retreat to underground bunkers! B^> Rgds Damon Testing started… Please fit protecting clothing and retreat to underground bunkers! B^>

Rgds

Damon

]]> By: xerxes http://labb.zafena.se/?p=282#comment-5080 xerxes Thu, 05 Nov 2009 16:16:08 +0000 http://labb.zafena.se/?p=282#comment-5080 I can happily announce that the warm-uptime that i have been worried about have been all removed by combining the Shark JIT with the optimized ASM interpreter on ARM. Edward Nevill are the one to thank that have made all this happen quite quickly, all sources can be found in the icedtea6 sourcetree and mailinglist. I have made some armv5 binarys that can be run on Ubuntu Jaunty systems like the Sheeva Plug and OpenRD-Client. Give it a spin and tell us what you think! http://labb.zafena.se/shark-testing/armv5/ Cheers and thanks Ed! Xerxes I can happily announce that the warm-uptime that i have been worried about have been all removed by combining the Shark JIT with the optimized ASM interpreter on ARM.
Edward Nevill are the one to thank that have made all this happen quite quickly, all sources can be found in the icedtea6 sourcetree and mailinglist.

I have made some armv5 binarys that can be run on Ubuntu Jaunty systems like the Sheeva Plug and OpenRD-Client.
Give it a spin and tell us what you think!
http://labb.zafena.se/shark-testing/armv5/

Cheers and thanks Ed!
Xerxes

]]> By: Damon Hart-Davis http://labb.zafena.se/?p=282#comment-4894 Damon Hart-Davis Sat, 10 Oct 2009 20:12:19 +0000 http://labb.zafena.se/?p=282#comment-4894 As to optimising start-up... The tiered compilation stuff that Steve Goldman of Sun was working on would of course help with that if you got a light C1 quickly and then C2 when needed: http://blogs.sun.com/fatcatair/category/Java But also I'd suggested to him to simply remember which methods had been hot at startup but didn't run long enough after being compiled to get much benefit from compilation. On a subsequent run those would be compiled as soon as (and this before) they were called (C1 level I guess) like an old-fashioned JIT. Much safer and less data and dependencies to store than a code cache, and the worst that can really happen is some excess compilation for things that turned out not to be early/start-up hotspots *this* time, eg because the code was changed. Java could really benefit from having such work folded back in IMHO... Rgds Damon As to optimising start-up… The tiered compilation stuff that Steve Goldman of Sun was working on would of course help with that if you got a light C1 quickly and then C2 when needed:

http://blogs.sun.com/fatcatair/category/Java

But also I’d suggested to him to simply remember which methods had been hot at startup but didn’t run long enough after being compiled to get much benefit from compilation. On a subsequent run those would be compiled as soon as (and this before) they were called (C1 level I guess) like an old-fashioned JIT.

Much safer and less data and dependencies to store than a code cache, and the worst that can really happen is some excess compilation for things that turned out not to be early/start-up hotspots *this* time, eg because the code was changed.

Java could really benefit from having such work folded back in IMHO…

Rgds

Damon

]]> By: Damon Hart-Davis http://labb.zafena.se/?p=282#comment-4892 Damon Hart-Davis Sat, 10 Oct 2009 07:36:56 +0000 http://labb.zafena.se/?p=282#comment-4892 Ah, good work! And luckily I don't think I care about most of the parts that Shark is handling less well unless network I/O gets snarled up the same way in JNI stuff! B^> Rgds Damon Ah, good work! And luckily I don’t think I care about most of the parts that Shark is handling less well unless network I/O gets snarled up the same way in JNI stuff! B^>

Rgds

Damon

]]> By: gbenson.net :: Long overdue update http://labb.zafena.se/?p=282#comment-4887 gbenson.net :: Long overdue update Fri, 09 Oct 2009 16:07:13 +0000 http://labb.zafena.se/?p=282#comment-4887 [...] one last thing I want to do before I step aside for a couple of months. Xerxes Rånby posted some benchmarks of Zero, Shark, and the assembler interpreter on ARM; Shark is gratifyingly faster than everything [...] [...] one last thing I want to do before I step aside for a couple of months. Xerxes Rånby posted some benchmarks of Zero, Shark, and the assembler interpreter on ARM; Shark is gratifyingly faster than everything [...]

]]>