Re: Cycles per ASM instruction

On Feb 13, 6:46 am, "Roded" <rod...@[EMAIL PROTECTED]
> wrote:
> Hello,
> Anyone know where I can find the number of cycles per ASM
> instruction for the z990 and the z9 computers?
> Any help would be greatly appreciated.
> Thanks,
> Roded

About 15 years ago, whenever I asked the design folks in Poughkeepsie
about whether instruction X was faster than instruction Y, the usual
response was "they're both 1 cycle". Of course, memory-memory
instructions would take longer, as would esoteric instructions
implemented in micro-code.

Speaking of memory-memory -- on a 370/168, a colleague sped up a text-
processing program by about 10-15% by recoding all MVCL instructions
as MVC loops. YMMV.

Timing manuals are notoriously difficult to get right. I once
optimised some inner-loop code on an RS/6000 (predecessor of the
PowerPC, which has a 360-ish feel to the instruction set) -- the Red
Book implied that a particular loop would perform quite poorly but in
reality it did quite well. IBM's specs were very conservative and I
suspect that the CPU had some pipelining that wasn't do***ented.

Anyway, if you want to get a feeling of what's involved in instruction
timing, you could do worse than look at some of the RS/6000 (PowerPC)
performance do***ents.

- peter