If cells are instanced before being defined, causing the GDS parser
to rewind the cell from the top, then the "already defined" error
messages will be suppressed, since it is to be expected that cells
will be seen twice (and ignored the 2nd time). When rewinding, an
output message is issued so that it is clear that the file contains
instances that are used before they are defined, and recommends the
"gds ordering on" setting. Also: Fixed the "gds ordering" command
code so that the command with no third argument returns the state
of the "gds ordering" setting instead of generating a parser error.
found to make ext2spice runtimes very long for large layouts.
The new method is equivalent but doesn't incur the overhead.
Also: Changed a flag check which was causing the substrate node
to be output as a port for certain layouts where the substrate
node connects to no devices, and so should be optimized out.
node representing the global substrate on cells that are abstract
views. Corrected a typecasting issue in ext2spice.c that throws
a compiler warning. Added another check for a cell being editable
when painting, which is a case that was not covered by the
previous code change to address the same issue.
underlying function from fprintf() to fputs(), which was changed for
gcc11 compatibility by Jean-Paul Chaput in github issue #123. Also
corrected a typo from the previous commit.
Implemented a separate check for ports when writing a subcircuit
that cross-checks against the port list in the flattened
extraction. This allows ports that were optimized out during
flattening of the hierarchy to be removed from the cell's port
list, which cuts down on disconnected nodes in the output port
list.
Print the names of the tile types that are illegally overlapping. This
gives us a better idea of what is wrong, eg:
feedback add "Illegal overlap between obsm2 and m2 (types do not connect)" medium
tracing the short path back through stacked contacts. Discovered
a problem with the connectivity search routine (which has been in
the code for a very long time) which will fail to copy contacts
to the selection cell if it has already drawn one of the metal
layers in the same place. This has now been fixed.
stops checking for the "best" path during the feed-forward check
and only enumerates the cost function for every tile in the
selection, moving outward from the source. This keeps the
algorithm efficient.
there was no distinction between a locked file and a new cell
(initial state) before writing to disk. This prevents any new cell
from being saved! Also: Revised the behavior of the "select short"
search, but this still has issues with long run-times on complex
layouts, so this is an ongoing effort.
whenever a process writes a cell to disk, it immediately releases the
file lock it had on that cell, which is clearly not the intent of file
locking. Fixed this issue. On a related topic, revised the "cellname
writeable" command so that it can make a cell editable even if the cell
has an advisory lock and cannot be made writeable. Perhaps there should
be a clearer distinction here between "writeable" and "editable". Also:
Reconsidered the previous commit, which removed the "--disable-locking"
from the configuration options. Because some operating systems may not
implement fnctl()-based file locking (Cygwin, for one, apparently doesn't),
it is still useful to be able to completely remove the function, in case
the operating system will fail to recognize the fnctl() values in the
code. Now, file locking behavior can be permanently removed through the
configuration option, or temporarily disabled from the command line.
post error messages when a GDS library addendum is read. It now
assumes that if a cell is called but not defined in the GDS, and
that cell happens to be in memory already, then this is intentional.
stack routines in utils/stack.c instead of relying on a recursive
routine, which will run out of the computer's stack space pretty
quickly on a large layout.
"defaultareacap" and "defaultperimeter" statements in the technology
file. Now, the parser makes use of the configuration of the
substrate from the "substrate" line to generate a default list of
which types and planes represent the substrate, and which types and
planes represent shielding to the substrate. This solves an issue
with the use of substrate isolation layers (e.g., "isosub" in
sky130A), because its definition and usage created substrate shields
on two planes (well and dwell), while the syntax for "defaultareacap"
and "defaultperimeter" only allow one shielding plane to be defined.
substrate (bulk terminal) and global substrate. Otherwise, the
routine in ext2hier.c that finds the substrate node will find the
first device bulk connection, not the default substrate.
CmdRS.c:1269:22: warning: & has lower precedence than ==; == will be evaluated first [-Wparentheses]
DRCtech.c:2573:16: warning: & has lower precedence than !=; != will be evaluated first [-Wparentheses]
Smith and which had escaped my attention. The commit from 8/3/2019
fixed an obvious fatal error but just replaced it with something
syntactically clean but nonsensical. This commit just removes the
questionable line altogether.
that makes a net a global net if there is a Tcl variable of the
same name. This conflicts with a later use of Tcl variables VDD
and GND to denote power and ground names, which is a completely
different usage.
value 1 after finding a substrate connecting type shielded (by deep
nwell, in the example) from the substrate, thus preventing the
search from processing any remaining substrate types. Solved by
changing the return value to zero to keep the search going.
record to the label structure to hold the port number. One major
issue stemming from this was reported in github issue #203 by Anton
Blanchard. This commit fixes that error.
output for transistors. The problem came from a change made to
fix an issue with capacitors marked as floating nodes because some
nodes are not output as source or drain. But those nodes are output
before the parameters, so when generating parameter output, all
nodes appear to have already been output. Solution: Specify an
additional bit in the "visited" mask for the node having been output
that is separate from the mask for resist classes used by the code
that writes parameter values, and use that bit as a test for whether
the node is connected to some device (not necessarily a FET source
or drain).
client data generated by ext2spice and attached to a node's
nodeClient record; there is an initNodeClient() routine but no
corresponding freeNodeClient() routine. Eventually had to add a
callback function passed to EFDone() and EFFlatDone() to clean up
these entries. After doing that, valgrind reports clean for all
memory allocated within ext2spice (there are other things that are
not freed but not related to a specific command, so do not need to
be treated as leaks).
clean up memory after running "ext2spice". There are apparently
still memory leaks somewhere, difficult to diagnose with valgrind,
but this fix removes the most substantial leakage and allows
"ext2spice" to be run continuously, at least for a while.
used by "topVisit" and "subcktVisit" in ext2spice.c, probably
caused by having different names on the same port number, the
subcktVisit() routine was modified to use exactly the same
enumeration as topVisit() so that they are guaranteed to have
the same result.
this limited ports to 16384, which seemed reasonable at the time.
However, the sky130_sram_macro layouts connect power and ground in a
way that when coupled with "extract unique" can generate tens of
thousands of ports and overrun the bit field, showing that automation
can do the unexpected. The solution was to split out the port number
from the label record as its own 32-bit value.
have already been output; i.e., that have EF_PORT set. However,
since EF_PORT is now set on all implicit ports, it is likely that
this part of the code is no longer exercised at all, and may be
removed.
connections through the substrate as the same node, and so will
not force different nodes names on the soft connection to be
unique. This should probably be selectable behavior. However, as
written, the "extract" command will always merge soft connections,
so giving them unique names just causes problems with "extract".
issue at the change made in revision 214. This was done incorrectly
in two ways, one being a set of statements inside an if() block that
should have been executed always, and the other an incorrect use of
the EF_DEVTERM flag, setting it when it should not have been set.
implicit substrate connections under some conditions were not added
to the subcircuit pin list. When this was corrected, the call to
the subcircuit was missing the implicit substrate port. When that
was corrected, the implicit substrate port printed was the subcircuit's
local node name, not the connection from above in the hierarchy. The
underlying problem was that the substrate was marked as a port in a
node record that was in another (flattened and unused) def and so not
seen when enumerating the def's node list. It's possible that the
better solution is that the efNodeHashTable() should be enumerated to
write subcircuit ports, not def->def_nodes. However, now, by using
EFHNLook(), the corresponding entry in efNodeHashTable() is found and
used.
work around the issue of loading a file containing references to
cells with the same name as cells already loaded. This is probably
going to cause additional headaches until a proper checksum method
is implemented.
Tim Edwards
Jean-Paul Chaput
Anton Blanchard
susinxy
Brad Smith