On Sat, Aug 03, 2002 at 10:51:41AM -0400, Ken Fox wrote: > Nicholas Clark wrote: > >I can write more a efficient implementation of abs_i ... things will > >go slightly faster > > The law of diminishing returns is broken for a VM. Eventually you > reach a point where adding more ops actually decreases total > performance. Instead of the change in performance tending towards > zero, it actually becomes negative. No surprise there -- it's > the heart of the RISC v CISC debate. > > Parrot has a RISC-like 3 operand design. IMHO if all ops are > consistent with that design, it will be easier/faster to generate > code. You can always add abs_i later if abs_i_i turns out to be a > performance problem.
But there do seem already to be arguably duplicate 2 operand versions of many ops. Hence I was surprised at the lack of consistency. To be "orthogonal" there are currently many $foo_i_ic_ic ops, that are entirely redundant, merely being complex ways to express set_i_ic while chewing CPU and RAM. Independent of my abs_i_i question (and a few other ops, including logical and bitwise that could be added), shouldn't all these _ic_ic ops go? Nicholas Clark -- Even better than the real thing: http://nms-cgi.sourceforge.net/
