swarm-hackers
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [swarm-hackers] Speed differences: Linux vs. Windows


From: Marcus G. Daniels
Subject: Re: [swarm-hackers] Speed differences: Linux vs. Windows
Date: Fri, 22 Aug 2008 10:40:41 -0600
User-agent: Thunderbird 2.0.0.16 (Windows/20080708)

Steve Railsback wrote:
Before I post these results, do you informed people have anything to say about how much of the differences can be attributed to the 64-bit compiler vs. differences in Swarm code vs. other differences?
Do you think my final conclusion is reasonable?

(In the past, I found negligible speed difference between 32-bit Linux and Cygwin.)

I find it less than helpful to get aggregate scores that combine all sorts of issues into one number. Different compilers, different versions of Swarm, different instruction sets, different runtimes for POSIX semantics. About all it says is how much variance can be expected, but without a good model underlying that number, it's all guesswork what it means. I guess it means that more registers are available on the 64 bit system and thus less cache pressure. To really understand the issues one should run all of this under oprofile for Linux and VTune for Windows and figure out what is happening on the hardware. gprof or valgrind also work ok, but give less detail.

Swarm, like Linux, like Cygwin, like GCC, is all open source so at the end of the day, it is possible to fix anything that may be a bottleneck. I think the best hope for increasing Objective C dispatch performance is probably LLVM Objective C compilers from Apple and link time optimization.

Marcus








reply via email to

[Prev in Thread] Current Thread [Next in Thread]