[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[f-cpu] CMP Execution Unit



We've been talking about this some days ago on the web. It contains the
currently missing parts from EU_INC (cmp, min/max/sort, msb0 and msb1)
and may be merged with EU_INC one day (that is, both units may share
I/O ports).

The implementation is rather simple; there's still some unnecessary
code duplication, and the layout could be improved as well (e.g. by
`interleaving' the trees for different chunk sizes). But it works :)

The timing is pretty tight: I had do violate the Six Gate Rule in the
second stage (but there are only and/or gates and muxes - I guess it's
not a problem). Synthesis reports are welcome, as usual.

For the record: all operations have a latency of two clock cycles.
It is possible to perform 8-bit compares in a single cycle, but that
will require an additional unit which can do nothing else, or a more
serious violation of the Six Gate Rule.

Sometimes I think we might do better with a `Twelve Gate Rule', half
the clock frequency, and shorter pipelines - at least when I look at
benchmark results of the Pentium 3 and 4 families... the P4 clocks are
much higher, but the CPUs aren't much faster.

-- 
 Michael "Tired" Riepe <Michael.Riepe@stud.uni-hannover.de>
 "All I wanna do is have a little fun before I die"

fcpu-cmp-mr-20020726.tar.gz