JIT optimizations on for loop with ar.Length constraint

re: JIT optimizations on for loop with ar.Length constraint

Drew Marsh — Sun, 24 Apr 2005 07:32:43 GMT

Any reason you don't just suggest foreach() in this kind of scenario? Since, after all, foreach() over an array ends up expanding to the pattern that is recognized by the JIT upon compilation. Obviously if you need the index ya gotta stick with the for(), but that's usually the rarer case.

Just curious,
Drew

re: JIT optimizations on for loop with ar.Length constraint

Henry Boehlert — Sun, 24 Apr 2005 11:44:56 GMT

Thanks, it was helpful indeed. But then again, I'm a little unhappy with that.

It's a area where developers can only guess what the right thing is. In your example, what I would have considered a safe bet was wrong and the opposite was true.
In Ian Hoff's example (on channel 9), a similar call (Array.GetUpperBound) was the cause of a 25% performance hit.

Profiling each and every loop is just too much work. There has to be more consistent guidance to let us make safe 80/20 bets.

Performance Considerations on MSDN is a very good summary but it is probably outdated and way to short on JIT.
Improving Managed Code Performance says:
Avoid repetitive field or property access.

I'd like to call for enhanced FxCop rules on performance, so we could get an automated grip on the problem without having to second guess every little loop.

I understand that the JIT compiler can only perform so much optimization, so wouldn't it be actually better to leave out obscure features if they are impossible to be implemented consistently?

Why can't the C# compiler come to the rescue here? You could put const/readonly propagation and all the special knowledge into it.

During (release) compilation to IL you have (almost) all the time in the world.

re: JIT optimizations on for loop with ar.Length constraint

Lee — Sun, 24 Apr 2005 17:58:53 GMT

Has this been changed in Whidbey? I've made this flawed optimization before specifically after a "measure, measure, measure" step. It wasn't flawed when I did it, as that loop construct became 39% faster. I just tried it again, and it continues to be faster.

Premature optimization is

The Blog — Sun, 24 Apr 2005 18:14:26 GMT

re: JIT optimizations on for loop with ar.Length constraint

Alex Suzuki — Mon, 25 Apr 2005 08:14:16 GMT

Brad: Thanks for the answer. That reader you mentioned was me. ;-)

Lee: I can confirm that. The for loop that caches the length is no longer slower in Whidbey. It seems to perform pretty much identically.

re: JIT optimizations on for loop with ar.Length constraint

ahnan — Mon, 25 Apr 2005 15:19:53 GMT

I don't think that's convenient. Not only for developers but also other languages/compliers. It just confused me.

re: JIT optimizations on for loop with ar.Length constraint

Mike Swaim — Mon, 25 Apr 2005 17:45:31 GMT

I know about that optimization, but what about ArrayLists? I use them a whole lot more than arrays.

Loop Optimizations

Javier G. Lozano — Mon, 25 Apr 2005 22:24:58 GMT

Loop Optimizations

Javier G. Lozano — Mon, 25 Apr 2005 22:27:21 GMT

Not your expected behaviour

the roarty blog — Tue, 03 May 2005 12:52:12 GMT

Brad Abrams explains a JIT optimisation which could catch you out.
This kinda stuff is interesting,...

For Loops and Performance

Greg Young [MVP] — Mon, 12 Jun 2006 01:32:21 GMT

This article discusses looping performance at the lowest level (native). It dispels a few myths that have been circulating the blogosphere for quite some time and makes some general performance recommendations.

An in depth look at for loops

Greg Young [MVP] — Mon, 12 Jun 2006 02:10:18 GMT

This article discusses looping performance at the lowest level (native). It dispels a few myths that have been circulating the blogosphere and makes some general performance recommendations.

Efficient For Loop

jokiz's blog — Sun, 25 Jun 2006 17:35:14 GMT

[code language="C#"]for (int i = 0; i &lt; collection.Count; i++){&nbsp;&nbsp;&nbsp; //do something with...

Misconception on the efficient for loop

jokiz's blog — Sun, 25 Jun 2006 17:54:32 GMT

I have posted an entry on
my blog on how to have an efficient for loops by storing the length
before...

Efficient For Loop

jokiz's blog — Fri, 11 Aug 2006 14:20:33 GMT

[code language="C#"]for (int i = 0; i collection.Count; i++){ //do something with collection}[/code]I’ve

Misconception on the efficient for loop

jokiz's blog — Fri, 11 Aug 2006 14:52:27 GMT

I have posted an entry on my blog on how to have an efficient for loops by storing the length before