OpenSolaris Speed VS. Linux. 4194

Yes, well I don't. The last time I optimized a compute intensive operation on sparc solaris (generating RSA key pairs), I was able to achieve a 30% speed up over the basic (max) solaris compiler options by choosing the right amounts of loop unrolling and buttembly tricks and cache alignment either via the sparc compiler or my own hand-unrolling and aligning of the C.

And the bank that was paying said that I had "only" managed a 30% speedup. I guess they wouldn't be impressed by the OPs measly 25% then!

Of course, by changing the algorithm (to avoid certain unhappinesses connected with fake long division), I managed to triple the speed.

This is "situation normal", as you must know if you work with compute intensive problems. What you can get from a compiler is trivial at the best of times - it's you that counts.

There you are. Just as I said - compiler speedup is not worth bothering with, especially given the danger of introducing optimizations you didn't intend - i.e. uncovering your bugs.

You would, with your watch in hand. Sitting at a computer screen you don't. Try it - the difference between 0.5s and 0.25s reaction time is not noticable. Neither is the difference between 0.1s and 0.05s.

It could only manage that on certain good cases. Overall, the effect would likely be diluted. But with the intel cmpiler you lose the chance to cmpile on different architectures - essential for uncovering those bugs that do not appear on your own test, but crop up under optimization ...

Peter