generated instance names
romi
below is legal (I think it is based on the LRM)? I'm using a mixture
of Cadence and Synopsys tools and they seem to support the label, but
depending on the tool create different names for the hierarchy. I'm
seeing mod_inst0 from some tools and label.mod_inst0 from others. I
think it should be label.mod_inst0, agree?
For other reasons, I need the the hierarchical names to be the same
both with and w/o PARAM set which is why I used the label "label" on
both parts of the conditional.
generate if (PARAM=0) begin:label
some_mod mod_inst0 ();
end else begin:label
some_other_mod mod_inst0 ();
end endgenerate
sh...@cadence.com
romi wrote:
> Hi, I'm wondering if the label "label" put on the generate conditional
> below is legal (I think it is based on the LRM)? I'm using a mixture
> of Cadence and Synopsys tools and they seem to support the label, but
> depending on the tool create different names for the hierarchy. I'm
> seeing mod_inst0 from some tools and label.mod_inst0 from others. I
> think it should be label.mod_inst0, agree?
Yes.
> For other reasons, I need the the hierarchical names to be the same
> both with and w/o PARAM set which is why I used the label "label" on
> both parts of the conditional.
Which is why that was explicitly made legal in this situation, where
the two generated scopes are mutually exclusive. This should be fine.
The Verilog-2001 LRM did not specify the creation of a scope for a
conditional generate (though it did specify it for generate for-loops),
and this created a variety of problems. This was fixed for the 2005
LRM, and implementors were advised to follow the 2005 version in their
2001 implementations. The 2005 version allowed leaving off the
explicit label, which still produced a scope but with a
compiler-generated name.
I would guess that some of your tools tried to implement the 2001
specification, but also ignored the named scope that you explicitly
specified, which they should not have done even in the 2001
specification. However, one of the problems with the 2001
specification was that it was unclear about the scoping associated with
generates, so this may be understandable.