Dual port Ram - for beginners

[dwerdna]

Hi all

Not having thought it was necessary to use a dual port ram until now I
was after a bit of advice.. I'm getting a few warnings on ModelSim,
and though I dont think I'm doing anything critically wrong I would
like to know if it can be improved.

Im targetting a Spartan 2E, and instantiating ramb4_s8_s8

I have a sampler which updates a RAM addr every second. My RAM is 64
addresses deep, therefore storing the previous minutes results which
the CPU can retrieve and graph at its leisure. I actaully have 8
different sampler channels, so the RAM is being fully utilised

I have decided to use a DP RAM as I felt it was a good way for 2
completely different controllers to access the same information
independently, which would also reduce my coding lines to make a
simpler system.

Both clocks into the RAM are from the same source, so I am getting a
warning that there is no phase difference between the rising edges of
the clocks (apparently it needs 0.1ns). Is this something I need to
modify in some way??

It is also coming up with a warning that I should not be addressing the
same address with both ports (currently the ram_enable is hardwired
'1'), but due to only one side writing and the other side reading I am
comfortable with this. I will probably implement some arbitration here
so that I dont get a unknown value for the read if both sides are
accessing the same address. Again, do I need to worry about this at
all?? Xilinx does advertise it as 'true dual port ram'.. My read
sample might be unknown but that is the worst that could happen isnt
it??

Thanks for your time

Andrew

 

[Mike Treseler]

Not having thought it was necessary to use a dual port ram until now I
was after a bit of advice.. I'm getting a few warnings on ModelSim,
and though I dont think I'm doing anything critically wrong I would
like to know if it can be improved.

Dual port RAM is not really a VHDL - for beginners topic.
Consider trying a design using registers first.

Im targetting a Spartan 2E, and instantiating ramb4_s8_s8
I have a sampler which updates a RAM addr every second. My RAM is 64
addresses deep, therefore storing the previous minutes results which
the CPU can retrieve and graph at its leisure.

Sounds you could get by with
one read port and one write port.
This can be inferred from a vhdl template,
and requires no arbitration.
http://groups-beta.google.com/groups?q=vhdl+mem_type+dual+port

-- Mike Treseler

 

[dwerdna]

Hi Mike

I would suggest that im in the 'VHDL - for intermediates' catergory,
looking for bit of info on dual port ram.

Since I very well could still be in the beginners group, could you
explain what the templates are all about please?? I had already looked
over the posts which come up under your suggested search string,
previous to my original question, and most of those posts are talking
about instantiating primitives, which is what I am doing..

So the original questions in summary were along the lines of:

Is there any particular issue with having the same clock - with no
phase difference - going into both ports of the RAM??

What is the worst thing that will happen if I am referencing the same
address, but ONLY writing to one address at that time?? From the
Xilinx info I would suggest that I will just get an unknown value on my
read port..

Thanks for your time

Andrew

 

[Mike Treseler]

Since I very well could still be in the beginners group, could you
explain what the templates are all about please?? I had already looked
over the posts which come up under your suggested search string,
previous to my original question, and most of those posts are talking
about instantiating primitives, which is what I am doing..

And others show template inference.
Synthesis manuals also show RAM templates.

Is there any particular issue with having the same clock - with no
phase difference - going into both ports of the RAM??

No. Make sure it is on a global net.

What is the worst thing that will happen if I am referencing the same
address, but ONLY writing to one address at that time?? From the
Xilinx info I would suggest that I will just get an unknown value on my
read port..

I don't know. Consider arbitration.


-- Mike Treseler

 

[Duane Clark]

What is the worst thing that will happen if I am referencing the same
address, but ONLY writing to one address at that time?? From the
Xilinx info I would suggest that I will just get an unknown value on my
read port..

The error message generated when this happens in VHDL simulation says
(for example):

# ** Warning: Memory Collision Error on RAMB16_S4_S4: at simulation
time 92425 ns.
# A read was performed on address fff (hex) of port B while a write was
requested to the same address on Port A The write will be successful
however the read value is unknown until the next CLKB cycle
# Time: 92425 ns Iteration: 2 Instance:
/bd_top/uut/data_buf_e/ram2_e/art_ram_g__7/art_ram_e

So there you go. One of the most detailed warning messages I have ever
seen

 

[dwerdna]

Thanks for your help Mike and Duane

 

[Jonathan Bromley]

What is the worst thing that will happen if I am referencing the same
address, but ONLY writing to one address at that time?? From the
Xilinx info I would suggest that I will just get an unknown value on my
read port..

You could consider "write-forwarding" around the DPRAM: add an
equality comparator that detects (write address)==(read address);
if equal, read data is taken from the write data pins instead of
from the RAM. Obviously the detailed timing needs a lot of
care, but the basic idea is straightforward enough.
--
Jonathan Bromley, Consultant

DOULOS - Developing Design Know-how
VHDL, Verilog, SystemC, Perl, Tcl/Tk, Verification, Project Services

Doulos Ltd. Church Hatch, 22 Market Place, Ringwood, BH24 1AW, UK
Tel: +44 (0)1425 471223 mail:[email protected]
Fax: +44 (0)1425 471573 Web: http://www.doulos.com

The contents of this message may contain personal views which
are not the views of Doulos Ltd., unless specifically stated.

 

[dwerdna]

Thanks Jonathan, thats a good suggestion.

I've solved my problem, so its all good, but I havent looked into why a
read from an address which is being written to will produce "X" when
you go by the principal that a register can be read the same time it is
written to. I guess the "X" is just descriptive to say that the value
might not be what you expect, and this would be a more likely case is
the clocks were at different frequencies.

Anyway just a point of interest, nothing more

Thanks again