multiple asychronous resets

[Ruth]

Hi,

Are there any problems with having more than one asynchronous reset? I
assumed that the synthesis tool just added an 'or' gate in front of the
reset on the FF, but I've ben told recently that this is not necessarily the
case and that it can cause problems. Can anyone shed any light on what
these might be please?

process (reset, clk, o_mreset)
begin
if (reset = '1') or (o_mreset = '1') then
dummy <= '0';
elsif rising_edge(slow_clk) then
if control = '1' then
dummy <= i_dummy;
end if;
end if;
end process;

Thanks

Ruth

 

[Tim Hubberstey]

Hi,

Are there any problems with having more than one asynchronous reset? I
assumed that the synthesis tool just added an 'or' gate in front of the
reset on the FF, but I've ben told recently that this is not necessarily the
case and that it can cause problems. Can anyone shed any light on what
these might be please?

process (reset, clk, o_mreset)
begin
if (reset = '1') or (o_mreset = '1') then
dummy <= '0';
elsif rising_edge(slow_clk) then
if control = '1' then
dummy <= i_dummy;
end if;
end if;
end process;

The result of this depends on your target architecture.

If you're targeting an ASIC, you'll probably get exactly what you
expect.

However, FPGAs can be quite different. Xilinx (and other) FPGAs have a
global reset net that is routed directly to all FFs. However, if you use
*any* logic on the reset line, other than inversions, the synthesis
tools will kick all resets off the global net and onto regular routing
resources. There are Xilinx app notes and/or answer records that discuss
this. Xilinx recommends against using the dedicated reset net in bigger
parts due to timing issues but in many designs this is not an issue and
it does save routing resources.

This is really an implementation issue not a VHDL issue and you should
ask in comp.arch.fpga instead.