* The Hurricane python module was not compiled at all. The name of
Python module wrappers is a bit inconsistent:
isobar --> Hurricane
crlcore --> CRL
anabatic --> Anabatic
katana --> Katana
* Use "py_mod_deps" and not only "py_deps" so the __PYTHON_MODULE__
flag is used for py.extension_module().
* In cumulus, the "cumulus" directory is not used for the installation
in site-packages/coriolis/.
* In designflow/pnr.py, temporarily disable Tramontana & Tutorial,
until they are ported to meson.
* Change: In GCell::stepNetDesaturate(), never move up a segment in
non-preferred direction as they are usually attached to terminals,
so they won't reduce the GCell density anyway.
* Change: In LayerAssign::RpsInRow::slacken(), do not slacken horizontals
from M1 vertical terminals when they are tall enough. Arbitrarily
choose above 6 H-tracks (should be made a configuration parameter).
* Change: In DataNegociate::update(), when processing perpandicular,
we were taking into account "fixed axis" even when they were
reduced (so not in tracks). Now do not.
In CoreToChip, the "iolib" I/O pad library an alternative to the "cmos"
one. "cmos" uses the pxlib library while iolib uses a symbolic abstract
version of the C4M (real) I/O pad.
The initialization is a bit split as the "cmos" library are part of
Alliance, while "iolib" is in alliance-check-toolkit. So "iolib" is
added at designflow.technos.setupCMOS().
Setup additional configuration parameter directly in:
core2chip.niolib.CoreToChip.__init__().
* Bug: In CRL/technos.node180.gf180mcu_c4m.iolib.py, remove the VDD and
VSS ring terminals in the pad as only *some* of the have it.
Assume that it is a bug from GF. The power rail will still be ok
as it connect by abutment (with the filler & other I/O pads).
* New: In cumulus.plugins.block.configuration.py, added support for
iterable I/O pad specifications in ioPads argument.
* New: In cumulus.plugins.core2chip.core2chip.py, add support for
any number of control signals on I/O pads. Not fully implemented
yet, as we only allow to hard-wire them either to one or zero.
Raise an error if _connect() fails to find a master net, so
we don't fail strangely later...
Jean-Paul Chaput
Rob Taylor
Gabriel Gouvine