> > Today, all RISC processors have a one-step shift implementation for their
base
> > data type (like 32 bit or 64 bit word).
>
> "All" is a strong word. I haven't kept up with the various chips,
> but surely there still ought to be (someplace?) an old-fashioned
> CPU design that has an on-chip "barrel shifter" - which allows
> multi-step shifting in one clock cycle.
>
> The problem is not RISC, but economics. It takes YY thousand gates
> to implement a "barrel shifter". Chip designers often find someplace
> else where using an additional YY thousand gates will improve the
> chip's SPECmark - and that is something that can be advertised!
> (Note: shift instructions are rarely used in the actual SPEC tests.)
> So the designers devote their YY thousand gates to where they perceive
> the greatest payback.
The SPARC processor can shift a number by n positions in one step. This is -
what I estimate - implemented by direct wired ways, and the rest is filled with
zeros. The idea of RISC was to make simple but fast instructions (and to
increase the potetial of parallelity *). You can see this by the huge number of
transistor in a RISC chip. The number is up to 10x higher than in the Pentium.
> Again I don't know the actual numbers. But I was under the impression
> that divide instructions on some RISC chips were only "several" times
> slower - that is, whereas most instructions take less than one cycle
> (because of pipelining and multiple parallel execution units), divide
> instructions take a multiple (but not large) number of cycles. Contrast
> that with a shift of 12 bit positions - the duration of that single
> instruction will be at least 11 clock cycles (for those 11 additional
> 1-step shifts!).
>
> mikus
>
Most instructions take one step because is in their nature to be fast (like bit
operations). Parallel operation execution (with the help of *) have few to do
with the execution length of an operation.
I stop now with discussing on this topic because it belongs to processor designs
(an interesting theme) and with chaning of technology strategies of speeding up
a process will change. We can discuss for ever about that.
Bojan
