"Thomas F. Burdick" <[EMAIL PROTECTED]> writes: > I'm quite happy with it. It's readible, not littered with type > information, and runs 1.08x the speed of GCC's code (1.1 for Sun's C > compiler). Note that it does the exact same computations as the C > version. There are still some things in the disassembly that are > suboptimal, but in all it's as good as the C.
Indeed. That's very nice code. My times are now 1.169s for the C-code, 1.22s for Gerd's version, and 1.49s for yours. Since the disassembly does not look so nice, I have to admit that Eric was right (the memory bandwidth should be the essential factor). Thanks again, Nicolas.
