Latches problem

Hi guyes,
Just see the process !
RDI output signals are not being latched . At reset '0' , all signals
become low but as soon as reset becomes high , all RDI outputs become
high. and after that they are being latched properly.
Pl suggest what should we do ?
Than in advance.
Rgds
--------------------
PROCESS(reset, a, pcs2) --PROCESS no 2
BEGIN
IF reset = '0' THEN
RDI <= (OTHERS => '0') ;
PDI <= (OTHERS => '0') ;
ELSIF pcs2 = '0' AND a = "00000" THEN
RDI(7 DOWNTO 0) <= D ;
ELSIF pcs2 = '0' AND a = "00001" THEN
RDI(15 DOWNTO <= D ;
ELSIF pcs2 = '0' AND a = "00010" THEN
RDI(23 DOWNTO 16)<= D ;
ELSIF pcs2 = '0' AND a = "00011" THEN
RDI(29 DOWNTO 24)<= D(5 DOWNTO 0) ;
ELSIF pcs2 = '0' AND a = "00100" THEN
PDI(7 DOWNTO 0) <= D ;
ELSIF pcs2 = '0' AND a = "00101" THEN
PDI(15 DOWNTO <= D ;
ELSIF pcs2 = '0' AND a = "00110" THEN
PDI(23 DOWNTO 16)<= D ;
ELSIF pcs2 = '0' AND a = "00111" THEN
PDI(29 DOWNTO 24)<= D(5 DOWNTO 0) ;
END IF ;
END PROCESS ;
------------------------

aaaaaa

aaaaaa wrote:


> RDI output signals are not being latched .

They /are/ latched! You did not assign every bit of RDI and PDI in any case.


> At reset '0' , all signals
> become low but as soon as reset becomes high , all RDI outputs become
> high. and after that they are being latched properly.

Do you talk about simulation of the behavioral or the netlist?

If you talk about netlist simulation then it seems to be the "muxed
latch problem": If there is a mux before a latch and one signal controls
the latch and the mux it may be, that while the latch is beeing closed
the mux changes the output sooner and the new value is loaded into the
latch before it is finally disabled.


> Pl suggest what should we do ?

Did you think about the difficulties that arise while using latches?
Latches are nice, small and power saving, but you have to design them
carefully.
If latches are preferrable for your task, control the mux with a
different signal than the latch-enable. Hold the mux-output stable until
the latch is closed.
Otherwise think about using flipflops.

Ralf

Ralf Hildebrandt

Ralf Hildebrandt wrote:
> aaaaaa wrote:
>
>
>> RDI output signals are not being latched .
>
>
> They /are/ latched! You did not assign every bit of RDI and PDI in any
> case.
>
>
>> At reset '0' , all signals
>> become low but as soon as reset becomes high , all RDI outputs become
>> high. and after that they are being latched properly.
>
>
> Do you talk about simulation of the behavioral or the netlist?
>
> If you talk about netlist simulation then it seems to be the "muxed
> latch problem": If there is a mux before a latch and one signal controls
> the latch and the mux it may be, that while the latch is beeing closed
> the mux changes the output sooner and the new value is loaded into the
> latch before it is finally disabled.
>
>
>> Pl suggest what should we do ?
>
>
> Did you think about the difficulties that arise while using latches?
> Latches are nice, small and power saving, but you have to design them
> carefully.
> If latches are preferrable for your task, control the mux with a
> different signal than the latch-enable. Hold the mux-output stable until
> the latch is closed.
> Otherwise think about using flipflops.
>
> Ralf

The biggest problem with latches is during timing analysis, they are
difficult to analyze because they have both a sequential and a
combinatorial path. AFAIK, utilities like Xilinx TRCE will ignore the
sequential path and report only the combinatorial path. This will make
that path look like it has lots prop delay.

Jim George