Comments on: JIT Happens http://blog.toadhead.net/index.php/2007/02/24/holy-jit/ Mike Heath's Blog Thu, 10 Nov 2011 18:05:27 +0000 hourly 1 http://wordpress.org/?v=3.0.5 By:   Programming Contest Results by Totally Nerdy Information http://blog.toadhead.net/index.php/2007/02/24/holy-jit/#comment-93   Programming Contest Results by Totally Nerdy Information Fri, 06 Apr 2007 18:43:32 +0000 http://swamp.homelinux.net/blog/?p=13#comment-93 [...] programming contest results are finally in. My C code prevailed and I am the biggest geek in the [...] [...] programming contest results are finally in. My C code prevailed and I am the biggest geek in the [...]

]]> By: mheath http://blog.toadhead.net/index.php/2007/02/24/holy-jit/#comment-12 mheath Fri, 02 Mar 2007 20:39:43 +0000 http://swamp.homelinux.net/blog/?p=13#comment-12 David, I downloaded the Visual C++ Express product but the option to build a DLL was disabled. Perhaps there a way to do it with the command-line compiler and linker but I didn't have the time to dig that deep. Should I be doing something differently? I agree about the microbenchmarks. They're really not that useful but it's still fun to see how different JIT compilers optimized differently. It would be really fun to analyze the assembly produced by each compiler and see what they're doing differently. David,

I downloaded the Visual C++ Express product but the option to build a DLL was disabled. Perhaps there a way to do it with the command-line compiler and linker but I didn’t have the time to dig that deep. Should I be doing something differently?

I agree about the microbenchmarks. They’re really not that useful but it’s still fun to see how different JIT compilers optimized differently. It would be really fun to analyze the assembly produced by each compiler and see what they’re doing differently.

]]> By: David Burggraaf http://blog.toadhead.net/index.php/2007/02/24/holy-jit/#comment-11 David Burggraaf Fri, 02 Mar 2007 19:38:16 +0000 http://swamp.homelinux.net/blog/?p=13#comment-11 I forget to give the URL for the Visual C++ express product: http://msdn.microsoft.com/vstudio/express/downloads/ I forget to give the URL for the Visual C++ express product: http://msdn.microsoft.com/vstudio/express/downloads/

]]> By: David Burggraaf http://blog.toadhead.net/index.php/2007/02/24/holy-jit/#comment-10 David Burggraaf Fri, 02 Mar 2007 19:37:02 +0000 http://swamp.homelinux.net/blog/?p=13#comment-10 Comparing compilers (C++, Java Jit, etc.) on microbenchmarks (small peices of code) is generally not that useful. Optimizations for this sort of code have been around for a long time and any average and above compiler should do fine here (the differences will only be where a specific compiler was tuned for a specific peice of hardware). Where it gets interesting is when you look at large code bases. Here a just-in-time compiler (jitter), trades off more complex optimizations, for runtime jit performance; where as the heavy-weight, build-lab C++ compiler can perform very expensive operations. It is very common for Java and .NET folks to take a tiny sample like this and say "look it's faster":). Regarding you Visual C++ comments, here is a URL where you can download the express SKU. It contains all the tools necessary to build a DLL, or really, any compent that can be built with the full version - though there are some highed optimizations that have been removed, that are very valuable on itanium, and for any large performance sensitive applications. It has been a fun contest:). -David Comparing compilers (C++, Java Jit, etc.) on microbenchmarks (small peices of code) is generally not that useful. Optimizations for this sort of code have been around for a long time and any average and above compiler should do fine here (the differences will only be where a specific compiler was tuned for a specific peice of hardware). Where it gets interesting is when you look at large code bases. Here a just-in-time compiler (jitter), trades off more complex optimizations, for runtime jit performance; where as the heavy-weight, build-lab C++ compiler can perform very expensive operations.

It is very common for Java and .NET folks to take a tiny sample like this and say “look it’s faster”:).

Regarding you Visual C++ comments, here is a URL where you can download the express SKU. It contains all the tools necessary to build a DLL, or really, any compent that can be built with the full version – though there are some highed optimizations that have been removed, that are very valuable on itanium, and for any large performance sensitive applications.

It has been a fun contest:).

-David

]]> By: John http://blog.toadhead.net/index.php/2007/02/24/holy-jit/#comment-9 John Thu, 01 Mar 2007 21:29:41 +0000 http://swamp.homelinux.net/blog/?p=13#comment-9 This doesn't surprise me. Your c code most likely IS faster, however there is a significant amount of overhead involved in JNI calls which probably accounts for the slowdown. This doesn’t surprise me. Your c code most likely IS faster, however there is a significant amount of overhead involved in JNI calls which probably accounts for the slowdown.

]]> By: Dave Wellman http://blog.toadhead.net/index.php/2007/02/24/holy-jit/#comment-8 Dave Wellman Thu, 01 Mar 2007 19:52:07 +0000 http://swamp.homelinux.net/blog/?p=13#comment-8 Nice Mike, So did your C++ solution win the programming contest? :P Dave Nice Mike,

So did your C++ solution win the programming contest? :P

Dave

]]> By: Totally Nerdy Information » Java JIT vs Native Code Performance Matrix http://blog.toadhead.net/index.php/2007/02/24/holy-jit/#comment-7 Totally Nerdy Information » Java JIT vs Native Code Performance Matrix Thu, 01 Mar 2007 04:51:21 +0000 http://swamp.homelinux.net/blog/?p=13#comment-7 [...] The performance results of our programming contest at work are listed below: Dell Lattitude D810 Pentium M 2.13 GHz [...] [...] The performance results of our programming contest at work are listed below: Dell Lattitude D810 Pentium M 2.13 GHz [...]

]]> By: RR http://blog.toadhead.net/index.php/2007/02/24/holy-jit/#comment-5 RR Mon, 26 Feb 2007 17:47:48 +0000 http://swamp.homelinux.net/blog/?p=13#comment-5 I wonder if you had coded this in Emacs if your results would differ. I wonder if you had coded this in Emacs if your results would differ.

]]>