Thou shalt have no other gods before the ANSI C standard 1413

D. J. Bernstein

OK I'll read it.

OIC. You keep them around to obfuscate the code.

No. Just how many FPUs do you think typical computers have?

On typical computers copying within the FP register file does not involve the FPU so the cost is only the instruction fetch and decoding which I would expect to be fully overlapped with the other data manipulation. Subtraction does require a pbutt through the FPU.

Of course. That's why I said "repeated". Doing it once makes sense. Repeating the subtraction rather than copying within the FP register file does not make sense.

In the code you presented for porting that variable appears in positions were numbers are allowed. Your reference to an "input language" implies that this code was machine generated. So you are asking me to port the intermediate code rather than the source code. Does my writing suggest to you that I am really that stupid?

If the code was machine generated than that generator is perfectly situated to perform architectural tweaking to address alternate architectures. So you have basically proven my point.

It is my intention to continue this exercise just to see how far I get. But I interpret your submission of so specialized a routine as support for my overall thesis that general purpose software need not be architecture specific. So the outcome of my efforts are kind of moot.

This is extremely low-level code. Even if it really was source code, what fraction of general purpose software do you believe merits so intense an effort at hand optimization?

tj3