Re: [f-cpu] Control signals on the Xbar

[Yann Guidon <whygee@f-cpu.org>]

hi again,

yet another remark :
when the pipeline is stalled (well, the decoder and the Xbar),
then the old value is kept.

In your ROP2 version i read something like :

if ( ! stalled )
  tmp_ROP2_function = ROP2_function ;
else
  tmp_ROP2_function = 0;

In fact, there should be no "else" clause because
the old value must be preserved. This is important
for the physical implementation because the CMOS
circuits consume energy when a data is changed.
If there is a 1-cycle stall, then the input
of the units will go from the old value to
0, and then from 0 to the new value. This consumes
2x more power than keeping the old value, even though
it's a bit more complex to implement in HW ("clock gating"
or a MUX per latched bit). but it's worth it !

WHYGEE
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/