Sharon Dagan wrote:
It can be educating to watch a functionally-equal piece of code, one written in C++ and compiled into native machine code, and another written in Java and JIT-compiled into native machine code (in-memory).Hi All,
Putting the JIT aside for a moment, consider an arbitrary C/C++ application optimized for Xeon - will it run faster, unconditionally, on Itanium? I guess that 'to get better performance on Itanium' this guy should get an optimized JVM *for* Itanium. In Java, performance varies dramatically depending on the JVM and JIT implementation.
I once tried something similar with C++ (compiled withMS Visual C++ 13.x, optimized but not CPU specific) vs. .NET (on MS .NET Framework 1.1) on Intel P4. I've written a tight loop doing some bit-crunching in a C# unsafe clause (a special clause which allows pointer manipulations) and yet found it running a bit slower than the C++ program. On inspecting the assembly code, I've seen the .NET JIT compiler simply used the common registers (i386 ones, nothing fancy) less efficiently. At that point I understood it's simply a matter of a few revisions to the JIT Compiler until it runs VM code on par with native code.
================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
