Comments on Eamon Nerbonne: .NET numeric performance disappointing

> Using the array's built-in Length propert...

2009-02-09T21:13:00.000+01:00

> Using the array's built-in Length property is actually somewhat faster than using a constant, but the difference is negligible.

This should work the way it does because, when the JIT knows you're comparing the index against the array upper bound in a loop, it can optimize away the range checks. It is surprising that doesn't work on x64, though!

For the record, an equivalent matlab loop takes 13...

2009-02-04T15:06:00.000+01:00

For the record, an equivalent matlab loop takes 13 seconds on the same machine.

I should test that sometime indeed; but based on h...

2009-02-04T12:45:00.000+01:00

I should test that sometime indeed; but based on http://flyingfrogblog.blogspot.com/2009/01/mono-22.html
which caused me to experiment in the first place, I'd expect mono to be even slower.

Of course, it's quite likely unsafe C# code with pointers will perform better - but it's frustrating that the cost of higher level constructs is so great.

Interesting! I wonder how your test would fare on ...

2009-02-04T04:48:00.000+01:00

Interesting! I wonder how your test would fare on Mono.