[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [f-cpu] Remarks to most recent (2002/07/04) snapshot
hi again,
Michael Riepe wrote:
> EU_IDU (not ready yet):
> The 8-bit subtractor can be replaced with misc.Generic_Adder.CIAdd.
i'll try to do it.
Cédric is busy with other things, but i'll do it anyway.
> EU_INC:
>
> The `bloc_and' entity and the tree of AND gates can probably
> be replaced with my misc.Bit_Manipulation.cascade_and function.
> Only the SIMD stuff needs to be added.
>
> Since only INC/DEC/LSB is supported, will there be a complementary
> unit for CMP/MSB?
i don't think so, at least yet. INC now is a 1-cycle unit,
i wanted to add a priority encoder but popcount does it nicely.
i also wanted min/max but it can be done with a substract with saturation
followed by a MUX in ROP2.
> EU_POPC:
> Currently uses function "+" from IEEE.Numeric_Std (but that's
> going to change, fortunately).
sure :-)
> Please do NEVER use IEEE.Numeric_Std in your designs
> (it is acceptable in testbenches, however).
that was only a first version that could do the job in simulation.
> I'm also missing delay indications in the new EUs. Do they all satisfy
> the 6G/10T rule? I doubt that e.g. the 8-bit divider from EU_IDU fits
> into a single pipeline stage, no matter how fast the subtractor is.
that's a specific IDIV problem...
Concerning the others, they are indicated in CAPABILITIES.txt.
it's not always up to date but it gives a rough estimation.
POPCOUNT will be updated too.
> Concerning component instantiations:
>
> The VHDL'93 `label : entity blah' instantiation style may be
> more convenient, but it requires blah to be analyzed first.
> With '87 style, you can analyze the source files in any order.
> And it will work with ALL tools (one never knows...).
>
> (Yann, you can put this into the VHDL HOWTO).
i'll do
> In order to speed up simulation and synthesis, it's also a good idea to
> avoid signal assignments whenever possible. Use processes and sequential
> statements, and put temporary values into variables, not signals. Signal
> assignments are only needed for inter-process communication, pipeline
> registers and the like.
ok.
> Michael "Tired" Riepe <Michael.Riepe@stud.uni-hannover.de>
WHYGEE
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu in the body. http://f-cpu.seul.org/