* Bug: In Anabatic::AutoHorizontal::_makeDogleg(), the up/down flag
was incorrectly computed when the RoutingGauge RL where not the
lower one. This was leading to making doglegs in non-routable
layers (but present in the gauge for other purposes).
* New: In CRL::RoutingLayerGauge, two new types of gauge are supported:
- Unusable : just do nothing with it, but the layer is stacked.
- BottomPowersupply : can be used for supply routing only, and
is *below* the normal routing layers (instead of on top as
usual).
Both new types must be *below* the real routing layers.
* New: In CRL::RoutingGauge, add a new attribute "firstRoutingLayer"
to give the index (depth) of the first layer usable for routing.
(not Unusable and not BottomPowerSupply)
* New: In Anabatic::Session & Anabatic::NetBuilder, in order to build
the initial wiring, provides (Session) and use (NetBuilder) the
new functions:
- getBuildRoutingLayer(depth)
- getBuildContactLayer(depth)
Thoses functions takes into account (offset) the unusable layers
so depth 0 is the first usable routing layer, and so on.
* Bug: In NetBuilder::construct(), the xG_xM1 topologies where wrongly
redirected towards xG_1M1, so only one M1 was connected, all others
left floating.
Now direct them towards xG_xM1_xM3.
* Bug: In NetBuilderHybridVH, activate topology management for 2G_1M1.
* New: In NetBuilderHybridVH::_do_1G_xM1_1PinM1(), added configuration
to manage pins on the north/south sides for VH,2RL.
* Bug: In NetBuilderHybridVH::doRp_xG_xM1_xM3(), correct misplaced
vertical creation (buildind invalid topologies).
* New: EtesianEngine::toColoquinte(), display histograms of the cells
widths (in pitch) before and after bloating to get a better feeling
of the behavior.
* New: In EtesianEngine, add support for track avoidance. Portions of
tracks to avoid are specified by a Box, which should flat and on
the axis of the request track. This feature is used by the H-Tree
to clear the vertical tracks under the tree from any terminal.
* New: In Etesian::Area, Slice and SubSlice, add support for track
avoidance. Exported to the Python wrapper.
* New: SubSlice::getUsedVTracks() to get a set of tracks blocked by
the cell.
* New: SubSlice::trackAvoid(), shift left/rigth the cell under the
requested vertical track. Try only to move the cell under the
track and not it's neighbor, so it assume that there is sufficient
space left or right of the cell.
* Bug: In cumulus/plugins.block.configuration.BlockConf, the Cfg
parameters may be read too early from the Cfg space into the
various sub-conf objects (like FeedsConf). Delay the reading
of the parameters in a _postInit() functions.
Modify Block and CoreToChip to call _postInit().
* New: In cumulus/plugins.block.configuration.BlockConf._loadRoutingGauge,
allow the cell gauge name to differ from the routing gauge name.
* New: In cumulus/plugins.block.configuration.FeedsConf, allow to
select the default feed to be used with 'etesian.defaultFeed'
parameter.
* New: In cumulus/plugins.block.spares.BufferPool, allow to control
whether or not we want tie to either side of the pool.
(for latch up).
* New: In cumulus/plugins.block.HTree._connectLeaf(), add support
for track avoidance.
* Bug: In cumulus/plugins.block.HTree._connectLeaf(), the TL2 contact,
the one on the *top* auxiliary buffer seemed to have been badly
positioned until now (too low, not using tl2Y).
This is strange because it should have caused disconnections,
but I didn't see it in the wiring and the regressions tests didn't
flag anything wrong. Still a bit weird and worrying.
Previously, the relevant NetBuilder and routing strategies where
directly guessed from the RoutingGauge traits. This is no longer
doable as the combinations increases. Now to configure both the
global and detailed router we need three "parameters" :
1. The routing gauge itself (tells which layers are in which
directions) and how to make the VIAs.
2. The NetBuilder to use, they are identified by strings.
Currently we support:
* "HV,3RL+", for all SxLib derived standard cells.
* "VH,2RL", for hybrid routing (over the cell, but terminals
are also in the first RL).
* "2RL-", for strict channel routing.
* "VH,3RL+", an attempt for FreePDK 45, not optimized enough
to be considered as usable.
3. The routing style, mostly affect the way the GCell grid will be
built.
* VH : first RL is V.
* HV : first RL is H.
* OTH : Run in full over-the-cell mode (needs at least 3RL).
* Channel : Run in *strict* channel routing mode (no routing over
the standard cells).
* Hybrid : Create channels, but can use H tracks over the
standard cells.
Thoses three parameters are partly overlapping and must be sets in
a consistent manner, otherwise strange results may occurs.
* New: CRL::RoutingGauge::getFirstRoutingGauge(), to get the lowest
layer available for routing (not a PinOnly, not a PowerSupply).
* Change: In CRL::RoutingGauge::isHV() and isVH(), were previously
always returning false when the gauge was 2RL only. Now, check
on the first usable RL.
* Bug: In cumulus/plugins.block.configuration._loadRoutingGauge(),
there was a bad computation of the deep RLs when the top layer
was not defined. Occured for 2RL gauges only.
* Bug: In Anabatic::RpsInRow::slacken() (LayerAssign), forgotten curly braces
in the test to skip METAL2 terminals.
* Change: In Etestian::BloatChannel::getDx(), adjust the bloating
policy to converge on Arlet6502. Always ensure that there is
a 50% ratio between terminal used V-tracks and free ones.
If there is more than 80% of terminals, add one more track.
* Bug: In AnabaticEngine & KatanaEngine, KatanaEngine is a derived
class of AnabaticEngine. They uses Anabatic::Configuration
and Katana::Configuration that also derives from each other.
I though I had made one configuration attribute in the base
class that was using the right Configuration. But no. I did
have two configurations attributes, one in AnabaticEngine and
one in KatanaEngine, the later "shadowing" the former. As a
results, parameters modified in AnabaticEngine, *after* the
initial creation of the tool *where never seen* at Katana
level (due to it's own duplicate). What a mess.
Now there is only one attribute in the *base* class Anabatic,
which is created through a new virtual function _createConfiguration()
called in _postCreate() which allocate the right Configuration
according to the dynamic type of the tool (KatanaEngine).
In KatanaEngine, access the configuration through the
attribute (_configuration) and not the accessor (getConfiguration()).
* Bug: In KatanaEngine, no longer directly use the _configuration attribute
(which is not accessible anyway) but the getConfiguration() accessor.
The accessor perform a static_cast from the Super::getConfiguration()
into Katana::Configuration.
Complete cleanup of the various configuration accessors.
* New: AnabaticEngine::setupNetBuilder(), perform an early check
of the requested NetBuilderStyle. The NetBuilderStyle is just a
string that will be matched against the (hard-coded) supported
NetBuilders. Then check the topological characteristics against
the capabilities of the gauge (HV, VH and so on).
Still a bit too hard-coded for now.
This function has been split from AnabaticEngine::_loadGrByNet().
* Change: AnabaticEngine::isChannelStyle() renamed from isChannelMode().
* New: In Anabatic::Configuration, two new attributes to select the
topology and routing style:
- _netBuilderStyle to explicitely select the NetBuilder to use.
It's a string, which is provided by each NetBuilder.
- _routingStyle to define how the overall routing will work.
It's a set of flags (StyleFlags):
* VH : first RL is V.
* HV : first RL is H.
* OTH : Run in full over-the-cell mode (needs at least 3RL).
* Channel : Run in *strict* channel routing mode (no routing over
the standard cells).
* Hybrid : Create channels, but can use H tracks over the
standard cells.
* New: In anabatic/Constants, add StyleFlags to define how the router
should operate (see above).
* Bug: In Anabatic::GCell, in CTOR, no reason to set up the HChannelGCell flag.
* Bug: In Anabatic::GCell::updateDensity(), when computing layers non contiguous
saturation, do not systematically skip RL 0, but only if it's PinOnly.
* Change: In Anabatic::NetBuilder, rename isTwoMetal by isStrictChannel.
* Change: In Anabatic::NetBuilderHV, rename doRp_AccessNorthPin() in
doRp_AccessNorthSouthPin(). More accurate.
* Bug: In NetBuilderHV::_do_1G_xM1_1PinM2(), the wires to connect the M1
terminals where created *twice*. Uterly stupid, there where placed in
overlap by the router!
* New: In AnabaticEngine, new accessors to the NetBuilderStyle and
RoutingStyle, proxies towards Configuration.
* Bug: In Manipulator::relax(), if there are two doglegs to be done, but
they are in the same GCell, only do one (the conflicting interval)
is short.
* Change: In Katana::Session, rename isChannelMode() into isChannelStyle().
* Change: In TrackSegment::isUnbreakable() and isStrap(), return false
when the base segment is a *weak global* (aligned with a global one).
* Change: In Katana::Row::createChannel(), correctly distinguish between
*strict channel* style and *hybrid* style. Tag the GCells as std cells
row or channels only in the former case.
* Change: In AutoSegment::canReduce(), allow reduction, when in channel
mode, for bottom connected jumpers at depth 1. Forbidden otherwise
as it may cause problems with the terminals (in OTC).
* Change: In TrackSegment::canRealign(), allow realignement of reduced
segments. What really needs to be checked is that it do not have
perpandiculars in reduced state.
* New: In Hurricane::RoutingPad, added support if the supporting component
is a Pad. The source/target positions are computed according to the
most likely direction.
Change the _getEntityAsComponent() function into an inline template
_getEntityAs<T>().
Change the flags from an enum to static const uint32_t.
* Change: In Anabatic::Configuration, use _getEntityAs<T>.
* Change: In Anabatic::Dijkstra, use _getEntityAs<T>.
* Change: In Anabatic::NetBuilder, use _getEntityAs<T>.
* Change: In Katabatic::LoadGrByNet, use _getEntityAs<T>.
* Change: In Katana::ProtectRoutingPads, use _getEntityAs<T>.
* Change: In Bora::AnalogDistance, use _getEntityAs<T>.
* Change: In Etesian::EtesianEngine, use _getEntityAs<T>.
* Bug: In Anabatic::AutoSegment::canReduce(), when looking at a fixed
segment we were unconditionally flagging it as reduced if it was
a "jumper", regardless of it's length.
It was wrong, now also check for a length below the P-Pitch.
This was introduced when implementing the channel routing.
* Update: In CRL/node600/phenitec/kite.py, update the routing gauge to
the new format. So now we can use again SxLib-2M (channel routing
SxLib for two metal technologies).
* Change: In CRL::BlifParser, if a master cell is not found in the
AllianceFramework, then try in the Blif supplied libraries.
This is used to load the zero, one and tie cells.
Add a Blif::getCell() static function to look into the Blif
supplied libraries.
* Change: In CRL::LefImport, sometimes there can be discrepencies between
the LEF ROUTING layers and the Coriolis routing gauge. Now ignore
routing layers that are *not* presents in the Coriolis gauge.
* Change: In AnabaticEngine, moved routingMode attribute from KatanaEngine,
as some setup operations needs it.
* Change: In AutoSegment::canReduce(), allow fixed segments to be reduced
if they are "jumpers" (turn+turn and top+top or bot+bot).
This case arise on the edge of routing channels for fixed wires
to connect terminals.
* Change: In AutoSegment::getTopologicalinfos(), compute differently the
(leftBound,rightBound) interval when in channel mode.
In over-the-cell mode, this interval is the one of the whole
GCells under the wire. In channel mode, for fixed wires (that is,
verticals connecteds to cells) this interval is reduced to half
the GCell height, on the connected side of said channel.
This allows Manipulator::_insertToTrack() to issue disantangling
requests (push left/push right) for fixed segments that are face
to face in the channel.
* Change: In Anabatic::Configuration CTOR, allow the cellGauge to have a
different name from the routingGauge. Now if the cell gauge that
should match the routing gauge is not found, fallback to the
name set in "anabatic.cellGauge" parameter.
Case occur when we try to match with CORE sites from LEF files.
* Change: In Etesian::Configuration CTOR, same change as in the
Anabatic configuration.
* Change: In Anabatic::GCell::updateDensity(), never set the GoStraight
flag in channel mode. This flag makes sense when there is at least
4 routing layers (so we have 2 contiguous free of blockages).
* Bug: In Anabatic::Session::_getNearestGridpoint(), sometimes the nearest on
grid point is outside the constraint box. Now force the point
to remains inside constraints even if offgrid.
* Change: In Katana::DataNegociate::update(), perpandiculars that are
either reduced or in non-preferred routing direction should not
trigger a bug message.
* Change: In KatanaEngine::_check(), do not check for fixed, horizontal
non-prefs AutoSegments in channel mode (avoid false bug display).
* Bug: In Manipulator::_forceToTrack(), slighty shrink (-1) the interval
to free. The intersection function of intervals returns true when
the two intervals *exactly* touches (1.vMax == 2.vMin). But in
this specific case, they are not *overlapping* and no action
should be taken...
* Bug: In Manipulator::_insertInTrack(), do not reject the track when
we are overlapping a fixed vertical segment in channel mode.
(Hmm, maybe already corrected by the previous one).
* Change: In Katana::NegociateOverlapCost(), in channel mode, do
not put two overlaping vertical fixed segments into infinite cost.
This happens when two cell connected verticals are face to face in
a channel. We want them negociated the track (by shrinking their
length) instead of excluding it right away.
* Change: In NegociateWindow::createTrackSegment(), in channel mode,
do not attempt to create a track segment over a fixed and reduced
AutoSegment.
Do not attempt to put a non-preferred AutoSegment on a Track
either.
* Bug: In RoutingEvent::revalidate(), the number of availables tracks
was badly computed when in the pure constraint case, when there
was only one it was reporting zero.
* Change: In TrackElements::TrackElements_Perpandicular::Locator,
do not issue a bug when an non-pref or reduced AutoSegment do not
have an associated TrackElement.
* Change: In TrackSegmentCost::update(), do not issue a bug when a
perpandicular is reduded or non-pref and do not have a TrackElement.
* Bug: In AutoSegment::getEndAxes(), as the NonAligned flag may be
wrong, always loop over the aligneds (if any). Then, the
target axis computation was wrong, using a min() instead of
a max().
This was, in turn, wrongly activating isNearMinArea() and
causing non-minimal length segment to have their extremeties
extented. And, in the end, Track overlap.
Bug: In NetBuilderHV::doRp_AutoContact(), when encountering a
punctual RoutingPad in METAL1, create the protection in the
METAL2 layer instead of METAL1 (current layer).
* Bug: In NetBuilder::setStartHook(), trigger the "move up" of
global routing only when encountering RoutingPad of a macro
block and *not* an I/O pin of the whole block (at the edge).
Boils down to look if the RoutingPad is anchored on a Pin
or a Segment (macro block).
* Change: In AutoHorizontal::setDuSource() & ::setDuTarget() (& AutoVertical),
check that the requested "du" is less than a pitch (this is not an
upper bound). Issue a warning if not true.
* Change: In AutoSegment::revalidate(), when passing the previous span
interval to expandToMinLength(), if we are in creation stage, make
it empty because it has no sense yet and can lead to a lock in a
too narrow span.
Change the coupled behavior of expandToMinLength() vs.
unexpandToMinLength(), the call to "unexpand" will be done on
AutoSegments that are flagged with SegAtMinArea, instead of
using the return value of "expand". This way we control across
multiple revalidate() if a segment can be "unexpanded".
* Bug: In AutoSegment::expandToMinLength(), not only try to shift left
if we are beyoond the max bound but also to the right if we are
below the min bound.
the span on the target and source were miscalculated, we must
add the half-minimal distance to get the span inside the Track.
* Change: In AutoSegment::isMiddleStack(), rename into isNearMinArea()
as we check for any *small length* set of AutoSegments and not
only the one part of a middle stack. In some rare instances,
two aligned segments can nevertheless be too short.
* Bug: In AutoSegment::reduceDoglegLayer(), when finding a reduced
segment also reset it's duSource & duTarget because it will
no longer be an isolated strip of metal.
* Change: In AutoHorizontal::updateOrient(), when there is a
source/target swap, no need to exchange the dxSource and
dxTarget extentions.
* Change: In AutoHorizontal::setDuSource() & ::setDuTarget(), check
for du bigger than the pitch, which should never occur and
display a warning.
Same modification in AutoVertical.
* Change: In Track::repair(), now invalidate the shifted segments
to ensure cache coherency with the TrackElement.
Do not take AutoSegment in non-preferred direction into
account, and especially do not try to resize them.
Now, invalidate the corrected segments (see below).
* Change: In KatanaEngine::finalizeLayout(), move the track repair
stage from here into NegociateWidow::run(). This way we avoid
the false warning about segment overlap in the data-base final
check. The false warning was because the AutoSegment where
shifted but not invalidated/revalidated leading to a cache
incoherency in the TrackElement. Now they *are* invalidated
and updated.
* Bug: In Track::checkMinArea(), do not check for minimal area when
a segment is overlapping a same net neighbor. To avoid false
minimum area violation warnings.
* Bug: In AutoContactTurn::updateTopology(), in some case we were making
doglegs on segments in non-preferred routing direction (mostly M2).
This must be never be done. As, by construction, at least one segment
of the turn must be in the preferred routing direction, always use
this one to make the dogleg.
This bug is likely to explain the remaining "bad spin" in the
final self-check of the routing data-structure.
The decoupling of the cell gauge and the routing gauge implies that
the METAL2 & METAL3 terminals of macro blocks cannot be aligned on
the routing tracks anymore. That is, an horizontal METAL2 terminal
will not be on a track axis, but offgrid, so we no longer can use
a METAL2 horizontal segment to connect to it. Making an adjustement
between the offgrid terminal and the on-grid segment has proven
too complex and generating difficult configuration for the router.
Moreover, METLA2 terminal could be fully inside a METAL2 blockage.
So now, when the gauges are decoupled, we connect the METAL2 and
METAL3 the same way we do for METAL1: *from above* in the perpandicular
direction and using a *sliding* VIA. We assume that those kind of
terminals in upper metals are quite long.
* New: In Hurricane::Rectilinear, export the isNonRectangle() method
to the Python interface.
* New: In CRL::RoutingGauge, add function isSuperPitched() with the
associated boolean attribute. Set to true when each pitch of
each layer is independant (not low fractional multiples).
* New: In AnabaticEngine, add the ability to temporarily disable the
canonize() operation (mainly used in dogleg creation).
* New: In AutoSegment::canonize(), do nothing if the operation is
disabled by AnabaticEngine.
* Bug: In Session::_revalidateTopology(), disable the canonization
during the topology updating of a net. Too early canonization
was occuring in makeDogleg() leading to incoherencies when
performing the later canonization stage over the complete net.
Mostly occured in the initial build stage of the net.
* New: In GCell, add function postGlobalAnnotate(), if a layer
is fully blocked (above 0.9), typically, under a blockage,
add a further capacity decrease of 2 on the edges. So we may
handle a modicum of doglegs.
* Bug; In GCell::addBlockage(), removeContact(), removeHSegment()
and removeVSegment(), forgot to set the Invalidated flag.
This may have lead to innacurate densities.
* Change: In GCell::updateDensity(), more complex setting of the
GoStraight flag. This flag is now set if we don't have two
*contiguous* below 60% of density. We need free contiguous
layers to make doglegs.
* New: In NetBuilder, now manage a current state flag along
with the state flag of the *source* GCell. This flag is used
to tell if the GCell needs it's *global* routing to be done
using the upper layers (METAL4 & METAL5) instead of the
lower ones.
* New: In NetBuilder::setStartHook(), set the state flag of the
GCell to ToUpperRouting when processing a global routing
articulation and one of the base layer is obstructed
above 0.9.
In GCell with terminals, also set ToUpperRouting when there
are some in METAL2 / METAL3 and the gauge is not super-pitched.
* New: In NetBuilder, function isInsideBlockage(), to check if a
terminal is completely or partially enclosed in a blockage.
* Change: In NetBuilderHV::doRp_AutoContact(), remove support for
trying to put on grid misaligned METAL2/METAL3.
Instead systematically access them from above.
Do not cover with fixed protection terminals that are already
enclosed in blockages.
* Bug: In NetBuilderHV::doRp_AutoContact(), always add the terminal
contact in the requested GCell and not the target/source one,
in case the terminal span several GCells.
* Change: In NetBuilderHV::doRp_Access(), create the local wiring
according to the RoutingPad layer.
* Change: In NetBuilderHV::_do_xG(), _do_2G(),
create the global wiring in upper layers, according to the
ToUpperRouting flag.
* Change: In NetBuilderHV::_do_xG_xM3(), now delegate to
_do_xG_xM3_baseRouting() and _do_xG_xM3_upperRouting() if the
density at terminal level is above 0.5.
* New: NetBuilderHV::_do_xG_xM3_baseRouting() and
_do_xG_xM3_upperRouting() separated function to manage the
local routing.
* Change: In NetBuilder::_do_globalSegment(), if the currently
processed GCell or it's source is in ToUpperRouting mode,
move up the global segment. Do *not* use the moveUp() function
which would create doglegs unwanted at this stage.
* New: In KatanaEngine::annotateGlobalGraph(), call postGlobalAnnotate()
on the GCell after the blockages have been taken into accound to
add the penalty.
* Bug: In Track::getPrevious(), correctly manage the 0 value for the
index argument. Strange it didn't show earlier.
Same goes for Track::expandFreeInterval().
* Bug: In Anabatic::layerAssign(), during the step of desaturation of
GCells that contains too much terminals (i.e. lot of *local*
congestion), the desaturation threshold was har-coded to 8.
Wich is fine for symbolic cmos but way too low for Flexlib.
End result was that most straight wires where moved towards
the upper layers, creating congestion (imbalance of layer
densities).
Now the parameter:
* "anabatic.saturateRp" (default value:8) is correctly taken
into account.
NOTE: This is likely to explain why we still got overlap in the
track coherency check in very rare occasions.
* Bug: In Katana::DataNegociate::update(), when computing the allowed
free interval for the segment axis deduced from the perpandicularly
connex segments, we account for the extension of the connecting
VIA. Those extension varies according to the kind of VIA and are
given by getExtensionCap().
We were accounting for the source & target extension VIA on the
parallel segments, assuming that source/target would not swap when
the perpandicular is moved. Which is *not* true.
Now account for the extension of the *connecting* VIA on all ends.
* Change: In AutoSegment::getTopologicalInfos(), enrich the list of
perpandicularly connected segment with wether they are connex by
their *source* or *target* contact. Mainly to be used by
DataNegociate::update().
* Change: In AnabaticEngine::setupPrerouteds(), take into account the
number of Pins. Now consider a net containing multiple Pins and, at most,
one segment as *non-routed".
This case may specifically happens for nets with pins on the north
and east side, which are slightly *inside* the abutment box (to be
seen by the router) and draw with them their *outside* direct
connection wire.
* Bug: In cumulus/plugins.chip.CoreWire, no longer put the north or
east side external Pin *exacyly* on the abutment box but *one pitch*
inside so they are correctly seens by the P&R (must be *inside*
the area of a GCell).
* New: In Anabatic::AutoSegment::create(), allow the created segment to
be in any supported routing layer, and not only the bottom H & V.
Modifications impact the *two* overload of the function.
* Change: In Anabatic::NetBuilderHV::doRp_AutoContacts(), for punctual
METAL1, the protection has to be in METAL2. Bump the layer depth
to correctly use the updated verstion of AutoSegment::create().
* Change: In AnabaticEngine::checkPlacement(), for the Pin, check that
it's layer is in the routing gauge before anything else.
* New: In Katana::NegociateWindow::createTrackSegment(), if the track
nearest the segment axis (refTrack) do not exists, call a
breakpoint just before crashing.
* New: In AnabaticEngine::printMeasures(), re-enable runtime & memory
footprint in measurements.
* Bug: In Anabatic::AutoSegment CTOR, correct computation of the initial
number of global segments. We were relying on the "global" flag,
which is *not* set at this stage. Now compare source and target
GCells.
* New: In KatanaEngine::runGlobalRouter(), store H-ovE & V-ovE in
measurements.
* New: In KatanaEngine CTOR, add the number of gates (flattened) to
the measurments.
* New: In KatanaEngine::runNegociate(), add the number of GCells to
the measurements.
* Change: In KatanaEngine::printCompletion(), compute the wirelengths
in either lambdas (symbolic) or *nanometers* (real).
* New: In cumulus/plugins.block.Block, add a setting to request the
Katana measurements dumping: "cfg.katana.dumpMeasures"
* Bug: In all CMakeLists.txt, it seems I was doing a worng use of
target_link_library(). No longer add dependencies to the C++ base
library but instead either to the Python associated module or
to the final binaries. This was inderectly causing the linking
problem related to Python (which was a misdirection).
* Bug: Typo in FindLibexecinfo.cmake, do not use FindLib[E]xecinfo.
* Change: In CRL/ccore/CMakeLists.txt, activate SKIP_AUTOMOC on
bison/flex generated files.
* Bug: In viewer/PyHApplication, do not delete the C++ object in
the Python destroy method.
* Change: In Anabatic::NetBuilder::construct(), add support for up to 5 M1
terminals in calling _do_1G_xM1_1PinM2(). When inserting diodes on
long wires, when there is not enough space underneath, they may end
up gouped at an end.
* Bug: In Anabatic::_desaturate(), the set of invalidated GCells was never
cleared between iterations of desaturated GCells. This was not a
memory leak per se because, in the end, the set was freed. But it did
induce a gigantic bloating in memory. This finally explain the strange
slow down in that stage.
Detected in ls180+cmos45.
* Change: In Anabatic::GCell, add a flag for each depth to avoid
desaturating twice the same GCell (maybe not needed now that the
bug is corrected, but it ensure a no-looping).
* Bug: In Anabatic::AntennaProtect, if the diode cell exists *but* the
maximum wire length for a diode is *not* defined (so zero) cancel
the diode protection instead of making a divide by zero.
* New: Python/C++ API level:
* Write a new C++/template wrapper to get rid of boost::python
* The int & long Python type are now merged. So a C/C++ level,
it became "PyLong_X" (remove "PyInt_X") and at Python code
level, it became "int" (remove "long").
* Change: VLSISAPD finally defunct.
* Configuration is now integrated as a Hurricane component,
makes use of the new C++/template wrapper.
* vlsisapd is now defunct. Keep it in the source for now as
some remaining non essential code may have to be ported in
the future.
* Note: Python code (copy of the migration howto):
* New print function syntax print().
* Changed "dict.has_key(k)" for "k" in dict.
* Changed "except Exception, e" for "except Exception as e".
* The division "/" is now the floating point division, even if
both operand are integers. So 3/2 now gives 1.5 and no longer 1.
The integer division is now "//" : 1 = 3//2. So have to carefully
review the code to update. Most of the time we want to use "//".
We must never change to float for long that, in fact, represents
DbU (exposed as Python int type).
* execfile() must be replaced by exec(open("file").read()).
* iter().__next__() becomes iter(x).__next__().
* __getslice__() has been removed, integrated to __getitem__().
* The formating used for str(type(o)) has changed, so In Stratus,
have to update them ("<class 'MyClass'>" instead of "MyClass").
* the "types" module no longer supply values for default types
like str (types.StringType) or list (types.StringType).
Must use "isinstance()" where they were occuring.
* Remove the 'L' to indicate "long integer" (like "12L"), now
all Python integer are long.
* Change in bootstrap:
* Ported Coriolis builder (ccb) to Python3.
* Ported Coriolis socInstaller.py to Python3.
* Note: In PyQt4+Python3, QVariant no longer exists. Use None or
directly convert using the python syntax: bool(x), int(x), ...
By default, it is a string (str).
* Note: PyQt4 bindings & Python3 under SL7.
* In order to compile user's must upgrade to my own rebuild of
PyQt 4 & 5 bindings 4.19.21-1.el7.soc.
* Bug: In cumulus/plugins.block.htree.HTree.splitNet(), set the root
buffer of the H-Tree to the original signal (mainly: top clock).
Strangely, it was only done when working in full chip mode.
Protections diodes may not be able to play their role if they are
separated from their cluster by upper level layers (METAL4/METAL5).
This seems not to diminish the total number of diodes.
* New: In Anabatic::NetData, add a set of non move up segments in
the object. To tag global wires that are part of a cluster.
* New: In Anabatic::AutoSegment, add support for a SegNoMoveUp
flag. This flag is propagated through _makeDogleg() .
Used in ::canMoveUp() and ::canPivotUp().
* New: In Anabatic::NetBuilder, add NetData to the attributes so
we can extract the NoMoveUp infomation given by the antenna
protect stage.
* New: In NetBuilderHV::_do_globalSegment(), lookup NoMoveUp
information from NetData to put it in AutoSegment.
* New: In AntennaProtect(Net*), flags the RoutingPad clusters wires
as non movable up.
* New: In AnabaticEngine::breatAt(GCell*), propagate the SegNoMoveUp
flags. Based on NetData.
* Bug: In AutoSegment::canReduce(), in the repair stage, segments are
allowed to go beyond their GCell bondaries, so global segments can
end up in zero length. So now allow globals less than one P-Pitch
to be flagged as reduced.
* Bug: In AutoSegment::isMiddleStack(), systematically reject non-canonical
segment. If it is non-canonical, then is aligned with a canonical one.
Then we will perform the check on it.
There seems to be another porblem of update of the length of the
associated TrackElement, the increase of size of the non-canonical is
not taken into account. This is enough to prevent the problem to
arise but we should invsetigate further.
* Hack: In AutoSegment::canMoveUp(), prevent segment id:6378409 to be
moved up so the last antenna effect is avoided. This is not a clean
way to do it.
* Bug: In TrackSegment::isMiddleStack(), formerly, all global segments
where discarted. But in the routing repair stage, segments can go
*outside* their GCell boundaries, allowing globals to became of
null length.
Hence, for global, we now also check the wirelength.
* Change: In AutoSegment::isMiddleStack(), after checking for obvious
non-candidates, relies on axis-to-axis wirelength instead of topological
criterions. We will not be able to account all the topologies that
may cause problem. So consider every segment whose length is below
one perpandicular pitch.
In Anabatic::AntennaProtect, when we cannot insert enough diodes
under a wire cluster. Which makes it likely very long and over an
area where diodes cannot be inserted (chip border close to I/O pads
or over a macro-block). Request extra diode insertion on it's
connecting RoutingPad clusters.
* New: In Anabatic::DiodeCluster, add a "forced diode" counter for
extra diodes inertions. Only used in the DiodeRps derived class.
* New: In Anabatic::DiodeCluster, add support for a cluster to know
it's neighbors. Stored as indexes of the table being built in
Anabatic::antennaProtect(Net*).
* New: In Anabatic::antennaProtect(Net*), when builing the WireCluster,
also find it's neigbors. Store the index of the cluster a segment
belongs to in clusterSegments.
The H-Tree support is now allowed for any net, not only the clocks and
not only top-level nets. This allow to better management of the LS180
internal clock signal.
* New: In Cell::flattenNets(Instance*,set<string>,uint64_t) new overload
of the function to allow the user to select nets that will *not*
be flattened. This makes the NoClockFlatten flag effectively obsolete,
we keep it for backward compatibility.
The net names can be of non top level ones. In that case, they must
use the name an HyperNet will get (the Occurrence name). For example:
"instance1.instance2.deep_net_name".
* New: In PyCell, update the wrapper for the new parameter of flattenNets(),
new utility function pyListToStringSet() to translate a Python list into
a C++ set of names.
* New: In EtesianEngine, add support for a list of nets to be excluded
from the flattening procedure. Those excluded nets will also be
excludeds from the Coloquinte nets *and* HFNS synthesis, as they
are likely to be manageds by a H-Tree.
* Change: In AnabaticEngine::_loadGrByNet(), now also skip nets that are
flagged as manually detailed route.
* New: In AnabaticEngine::antennaProtect(), do not try to insert diodes
on nets that are already fixed or detaled route. This replace the
clock exclusion.
* New: In cumulus/plugins.{block,htree,chip}, replace the concept
of clock-tree by the more generic H-Tree. That is, we can ask the P&R
to create H-Tree on any net of the design, not only the ones matcheds
as clock. The net does not even need to be top-level.
This is to manage the PLL internal clock generated by the PLL in
the LS180 chip.
Start to change all reference to "clock" into "H-Tree".
* Bug: In cumulus/plugins.chip.powerplanes.Builder._connectHTree(),
there was an inversion of the H & V routing gauges to compute the
track into which put the H-Tree center to corona edge wiring.
This was causing tracks to be used twice, seen in the ao68000 test
bench.
* New: In Anabatic::DiodeWire, use "antennaDiodeMaxWL" to compute the number
of diodes to insert in a wire only cluster. Use boxes instead of segments
to define the area as segments can be splitted by the diodes inserteds
at the DiodeRps stage.
* New: In DiodeWire::createDiodes(), specific diode insertion method.
Try to instert first in long horizontal wires.
When a long *horizontal* wire connect to a cluster, an antenna effect
may be created *before* the METAL3 is deposited, if the cluster's diode
is not *directly* connected to the gate through *only* METAL2. So, we
add a "forced halo" where the long horizontal connecting wires will be
broken by a diode *near* the gate. This problem do not occur for long
connecting METAL3, as the diode will be connected by then. Note that
we are hard-coding the gauge routing direction in the algorithm.
With that modification, only one antenna effect remains in LibreSOC
LS180. May be corrected by post-treatement.
* New: In Anabatic::DiodeCluster::mergeForcedhalo() add specific secondary
areas where diode must be insterted in addition to the one of the
RroutingPad cluster. To "isolate" the cluster from long horizontal
wires.
Protecting clusters of sinks is not enough. There can be very long
wires that far exceed the protection capacity of one diode. Instead
of putting a bunch of diodes near the sinks, we choose to put them
regularly along the interconncting wires.
With this approach we are down to 7 antenna violations on LibreSOC
LS180 test chip.
This will get less good results on arlet6502 & ao6800 because of the
core being a long way from the I/O pads. Should create jumpers on thoses,
but it is for later.
First part of the antenna effect protection : diode insertions.
Anabatic::antennaProtect(Net*) and it's supporting infrastructure
has been rewritten & simplificated. Must be used in conjuction
with the "Flexlib" bloat model of Etesian. A cursory description
of the algorithm has been added in the source file.
* New: GCell::hasNet() to tell if net is going through a GCell,
either as a straigth wire or has a local GContact (turn,
branch, terminal).
* New: Etesian::BloatFlexib class (tagged "Flexlib") suited for
flexlib uses. It is derived from "nsxlib".
To have enough space to insert all the wanted diodes, we
enlarge "mx2_x2" & "mx3_x2" of resp. 1 and 2 pitches.
This is an empiric finding, Yosys seems very fond of thoses
gates and we often see them underneath area where no space
was available to put a diode... May need some more fine grained
analysis.
* Change: In AutoSegment::canMoveUp(), hard code for FlexLib a track
reserve of one when trying to move up.
In FlexLib, the router promotes too easily METAL2 towards METAL4,
clogging the higher level. So we end up with the METAL2 level
under-saturated and the METAL4 over saturated, leading to routing
failures.
Now we assume that there is "one track less available" in METAL4.
As for now, solves the problem for LibreSOC.
* Bug: In NetBuilderHV::_do_xG_1PinM2(), for METAL2 pins on the EAST
side, we were computing the track from the XMax position of the
GCell. Now, with complete discoupling of the routing pitches and
GCell size, this is most of the time false.
Replace with a true computation of the last track of routing
area.
* Bug: In NetBuilderHV::_do_xG_1PinM2(), when the M2 pin is on the *east*
side and there are three globals, we must swap htee1 & htee2.
Otherwise, we the SW/NE contacts do not have the right numbers
of free segments attach.
* New: In AnabaticEngine::cleanupGlobal(), check that no global component
is left behind, sure sign that the global to detailed building went
wrong.
Jean-Paul Chaput
Myrtle Shah
Las Safin