a bug : the VIAs where too close of each others in ALU4,ALU5 nero discarting
silently one of them. With this modification we no longer respect the double
pitch for upper layers at the edge of the routing grid.
* reordering of functions in Pirouz' code
* fixed bug : variable declared as extern in .h and as static in .c
* adding some missing inclusions of string.h, stdlib.h (this suppresses some nasty warnings)
* fixed gcc version detection in ocp
- Bug : the Darwin link editor seems to be more strict, or having a
different policy about implicit static/extern variables. We have
to do a lot of cleanup to avoid symbols redefinition at link time.
We do hope it may solve the earlier bison problem with asimut...
- New Feature : to allow routing reuse, NeRo is now able to build the
abstract of a mega-cell. That is, adds the relevant CALUx and TALUx
describing the interface at top-level.
- Bug : in the abstract generation, the intermediates TALUx metals of
a stacked VIA where forgotten. For example, when there is a VIA34
on top of a VIA23 you must have obstacles in TALU4 and TALU2 but
also a "dot" obstacle in TALU3. "Dot" obstacle are added only when
we process a layer in the preferred routing direction, otherwise
we would create a dot at each routing grid point...
- Bug : the size of the non-default width VIA was not correctly computed.
The result was bigger obstacles cluttering the routing grid and
forcing unneeded detours.
pour pouvoir faire make sans make install
- struct timeb et ftime sont depreciees et non definies sous OSX,
appels remplaces par gettimeofday.
patch propose par Nicolas Pouillon.
- Bug : the global qualification of a net was independent of the RBox
choice. The bug occurs for designs about the size of the local/
global choice : when the routing is still local but some nets
have a size that made them global. In that cases terminals are
extented (for global routing) outside of the routing box :
in z=3 where z <= 2, which leads to immediate memory corruption.
Franck Wajsburt
Jean-Paul Chaput