Each architecture must implement the following types and APIs.
Architectures can either inherit from ArchAPI<ArchRanges>, which is a pure virtual description of the architecture API; or BaseArch<ArchRanges> which provides some default implementations described below.
ArchRanges is a struct of usings that allows arches to return custom range types. These ranges can be anything that has a begin() and end() method that return const forward iterators. This can be a std::list<T>, std::vector<T>, a (const) reference to those, or anything else that behaves in a similar way.
The contents of ArchRanges is as follows:
| Type | Range of |
|---|---|
ArchArgsT |
N/A (struct of device params) |
AllBelsRangeT |
BelId |
TileBelsRangeT |
BelId |
BelAttrsRangeT |
std::pair<IdString, std::string> |
BelPinsRangeT |
IdString |
CellBelPinRangeT |
IdString |
AllWiresRangeT |
WireId |
DownhillPipRangeT |
PipId |
UphillPipRangeT |
PipId |
WireBelPinRangeT |
BelPin |
AllPipsRangeT |
PipId |
PipAttrsRangeT |
std::pair<IdString, std::string> |
AllGroupsRangeT |
GroupId |
GroupBelsRangeT |
BelId |
GroupWiresRangeT |
WireId |
GroupPipsRangeT |
PipId |
GroupGroupsRangeT |
GroupId |
DecalGfxRangeT |
GraphicElement |
CellTypeRangeT |
IdString |
BelBucketRangeT |
BelBucketRange |
BucketBelRangeT |
BelId |
The functions that return a particular type are described below. Where a default function implementation is provided, BaseArchRanges (which ArchRanges can inherit from) will set the range type appropriately.
The architecture-specific archdefs.h must define the following types.
With the exception of ArchNetInfo and ArchCellInfo, the following types should be "lightweight" enough so that passing them by value is sensible.
A scalar type that is used to represent delays. May be an integer or float type.
A type representing a bel name. BelId() must construct a unique null-value. Must provide ==, !=, and < operators and a unsigned int hash() const member function.
A type representing a wire name. WireId() must construct a unique null-value. Must provide ==, !=, and < operators and a unsigned int hash() const member function.
A type representing a pip name. PipId() must construct a unique null-value. Must provide ==, !=, and < operators and a unsigned int hash() const member function.
A type representing a bel bucket. BelBucketId() must construct a unique null-value. Must provide ==, !=, and < operators and a unsigned int hash() const member function.
A type representing a group name. GroupId() must construct a unique null-value. Must provide == and != operators and a unsigned int hash() const member function.
A type representing a reference to a graphical decal. DecalId() must construct a unique null-value. Must provide == and != operators and a unsigned int hash() const member function.
A type representing a reference to a constrained cluster of cells. ClusterId() must construct a unique null-value. Must provide == and != operators and unsigned int hash() const member function.
The global NetInfo type derives from this one. Can be used to add arch-specific data (caches of information derived from wire attributes, bound wires and pips, and other net state). Must be declared as empty struct if unused.
The global CellInfo type derives from this one. Can be used to add arch-specific data (caches of information derived from cell attributes and parameters, bound bel, and other cell state). Must be declared as empty struct if unused.
Each architecture must provide their own implementation of the Arch struct in arch.h. Arch must derive from ArchAPI<ArchRanges> or BaseArch<ArchRanges> (see above) and must provide the following methods:
Constructor. ArchArgs is a architecture-specific type (usually a struct also defined in arch.h).
Return a user-friendly string representation of the ArchArgs that was used to construct this object.
Return the ArchArgs used to construct this object.
Return an internal IdString representation of the ArchArgs that was used to construct this object.
BaseArch default: returns empty IdString
Get grid X dimension. All bels and pips must have Y coordinates in the range 0 .. getGridDimX()-1 (inclusive).
Get grid Y dimension. All bels and pips must have Y coordinates in the range 0 .. getGridDimY()-1 (inclusive).
Get Z dimension for the specified tile for bels. All bels with at specified X and Y coordinates must have a Z coordinate in the range 0 .. getTileDimZ(X,Y)-1 (inclusive).
Get Z dimension for the specified tile for pips. All pips with at specified X and Y coordinates must have a Z coordinate in the range 0 .. getTileDimZ(X,Y)-1 (inclusive).
BaseArch default: returns 1
Returns a delimiter that can be used to build up bel, wire and pip names out of hierarchical components (such as tiles and sites) to avoid the high memory usage of storing full names for every object.
BaseArch default: returns ' '
Get list of cell types that this architecture accepts.
BaseArch default: returns list derived from bel types set up by init_cell_types()
Lookup a bel by its name, which is a list of IdStrings joined by getNameDelimiter().
Get the name for a bel. (Bel names must be unique.)
Get the X/Y/Z location of a given bel. Each bel must have a unique X/Y/Z location.
Lookup a bel by its X/Y/Z location.
Return a list of all bels at the give X/Y location.
Returns true if the given bel is a global buffer. A global buffer does not "pull in" other cells it drives to be close to the location of the global buffer.
BaseArch default: returns false
Return a (preferably unique) number that represents this bel. This is used in design state checksum calculations.
BaseArch default: returns bel.hash()
Bind a given bel to a given cell with the given strength.
This method must also update cell->bel and cell->belStrength.
BaseArch default: binds using base_bel2cell
Unbind a bel.
This method must also update CellInfo::bel and CellInfo::belStrength.
BaseArch default: unbinds using base_bel2cell
Returns true if the bel is available. A bel can be unavailable because it is bound, or because it is exclusive to some other resource that is bound.
BaseArch default: returns getBoundBelCell(bel) == nullptr
Return the cell the given bel is bound to, or nullptr if the bel is not bound.
BaseArch default: returns entry in base_bel2cell
If the bel is unavailable, and unbinding a single cell would make it available, then this method must return that cell.
BaseArch default: returns getBoundBelCell(bel)
Return a list of all bels on the device.
Return the type of a given bel.
bool getBelHidden(BelId bel) const
Should this bel be hidden from utilities?
BaseArch default: returns false
Return the attributes for that bel. Bel attributes are only informal. They are displayed by the GUI but are otherwise unused. An implementation may simply return an empty range.
BaseArch default: returns default-constructed range
Return the wire connected to the given bel pin.
Return the type (input/output/inout) of the given bel pin.
Return a list of all pins on that bel.
Return the list of bel pin names that a given cell pin should be routed to. In most cases there will be a single bel pin for each cell pin; and output pins must always have only one bel pin associated with them.
BaseArch default: returns a one-element array containing pin
Lookup a wire by its name, which is a list of IdStrings joined by getNameDelimiter().
Get the name for a wire. (Wire names must be unique.)
Get the type of a wire. The wire type is purely informal and
isn't used by any of the core algorithms. Implementations may
simply return IdString().
BaseArch default: returns empty IdString
Return the attributes for that wire. Wire attributes are only informal. They are displayed by the GUI but are otherwise unused. An implementation may simply return an empty range.
BaseArch default: returns default-constructed range
Return a (preferably unique) number that represents this wire. This is used in design state checksum calculations.
BaseArch default: returns wire.hash()
Bind a wire to a net. This method must be used when binding a wire that is driven by a bel pin. Use bindPip()
when binding a wire that is driven by a pip.
This method must also update net->wires.
BaseArch default: binds using base_wire2net
Unbind a wire. For wires that are driven by a pip, this will also unbind the driving pip.
This method must also update NetInfo::wires.
BaseArch default: unbinds using base_wire2net
Return true if the wire is available, i.e. can be bound to a net.
BaseArch default: returns getBoundWireNet(wire) == nullptr
Return the net a wire is bound to.
BaseArch default: returns entry in base_wire2net
If this returns a non-WireId(), then unbinding that wire will make the given wire available.
BaseArch default: returns wire
If this returns a non-nullptr, then unbinding that entire net will make the given wire available.
BaseArch default: returns getBoundWireNet(wire)
Get the delay for a wire.
Get a list of all wires on the device.
Get a list of all bel pins attached to a given wire.
Get the bounding box required to route an arc, assuming an uncongested chip. There may be significant performance impacts if routing regularly exceeds these bounds by more than a small margin; so an over-estimate of the bounds is almost always better than an under-estimate.
If not an empty string, indicate this wire can be used to source nets
with their constant_value equal to its return value.
BaseArch default: returns IdString()
Lookup a pip by its name, which is a list of IdStrings joined by getNameDelimiter().
Get the name for a pip. (Pip names must be unique.)
Get the type of a pip. Pip types are purely informal and
implementations may simply return IdString().
BaseArch default: returns empty IdString
Return the attributes for that pip. Pip attributes are only informal. They are displayed by the GUI but are otherwise unused. An implementation may simply return an empty range.
BaseArch default: returns default-constructed range
Get the X/Y/Z location of a given pip. Pip locations do not need to be unique, and in most cases they aren't. So for pips a X/Y/Z location refers to a group of pips, not an individual pip.
Return a (preferably unique) number that represents this pip. This is used in design state checksum calculations.
BaseArch default: returns pip.hash()
Bid a pip to a net. This also bind the destination wire of that pip.
This method must also update net->wires.
BaseArch default: binds using base_pip2net and base_wire2net
Unbind a pip and the wire driven by that pip.
This method must also update NetInfo::wires.
BaseArch default: unbinds using base_pip2net and base_wire2net
Returns true if the given pip is available to be bound to a net.
Users must also check if the pip destination wire is available
with checkWireAvail(getPipDstWire(pip)) before binding the
pip to a net.
BaseArch default: returns getBoundPipNet(pip) == nullptr
Returns true if the given pip is available to be bound to a net, or if the pip is already bound to that net.
BaseArch default: returns getBoundPipNet(pip) == nullptr || getBoundPipNet(pip) == net
Return the net this pip is bound to.
BaseArch default: returns entry in base_pip2net
If this returns a non-WireId(), then unbinding that wire will make the given pip available.
BaseArch default: returns empty WireId()
If this returns a non-nullptr, then unbinding that entire net will make the given pip available.
BaseArch default: returns empty getBoundPipNet(pip)
Return a list of all pips on the device.
Get the source wire for a pip.
Get the destination wire for a pip.
Bi-directional switches (transfer gates) are modeled using two anti-parallel pips.
Get the delay for a pip.
Get all pips downhill of a wire, i.e. pips that use this wire as source wire.
Get all pips uphill of a wire, i.e. pips that use this wire as destination wire.
Lookup a group by its name, which is a list of IdStrings joined by getNameDelimiter().
BaseArch default: returns GroupId()
Get the name for a group. (Group names must be unique.)
BaseArch default: returns IdStringList()
Get a list of all groups on the device.
BaseArch default: returns default-constructed range
Get a list of all bels within a group.
BaseArch default: asserts false as unreachable due to there being no groups
Get a list of all wires within a group.
BaseArch default: asserts false as unreachable due to there being no groups
Get a list of all pips within a group.
BaseArch default: asserts false as unreachable due to there being no groups
Get a list of all groups within a group.
BaseArch default: asserts false as unreachable due to there being no groups
Return a rough estimate for the total maxDelay() delay from the given src wire to
the given dst wire.
This should return a low upper bound for the fastest route from src to dst.
Or in other words it should assume an otherwise unused chip (thus "fastest route"). But it only produces an estimate for that fastest route, not an exact result, and for that estimate it is considered more acceptable to return a slightly too high result and it is considered less acceptable to return a too low result (thus "low upper bound").
Return a reasonably good estimate for the total maxDelay() delay for the
given arc. This should return a low upper bound for the fastest route for that arc.
Return a small delay value that can be used as small epsilon during routing. The router will for example not re-calculate cached routing data if faster routes are found when the difference is smaller than this value.
The base penality when calculating delay penalty for ripping up routed nets. The actual penalty used is a multiple of this value (i.e. a weighted version of this value).
Convert an delay_t to an actual real-world delay in nanoseconds.
Convert a real-world delay in nanoseconds to a delay_t.
Convert a delay_t to an integer for checksum calculations.
This allows an arch to provide a more precise method for calculating the delay for a routed arc than
summing wire and pip delays; for example a SPICE simulation. If available, delay should be set and
true returned.
BaseArch default: returns false
Run the packer.
Run the placer.
run the router.
Return the graphic elements that make up a decal.
The same decal must always produce the same list. If the graphics for a design element changes, that element must return another decal.
BaseArch default: returns default-constructed range
Return the decal and X/Y position for the graphics representing a bel.
BaseArch default: returns DecalXY()
Return the decal and X/Y position for the graphics representing a wire.
BaseArch default: returns DecalXY()
Return the decal and X/Y position for the graphics representing a pip.
BaseArch default: returns DecalXY()
Return the decal and X/Y position for the graphics representing a group.
BaseArch default: returns DecalXY()
Returns the delay for the specified path through a cell in the &delay argument. The method returns
false if there is no timing relationship from fromPort to toPort.
BaseArch default: returns false
Return the timing port class of a port. This can be a register or combinational input or output; clock input or output; general startpoint or endpoint; or a port ignored for timing purposes. For register ports, clockInfoCount is set to the number of associated clock edges that can be queried by getPortClockingInfo.
BaseArch default: returns TMG_IGNORE
Return the clocking info (including port name of clock, clock polarity and setup/hold/clock-to-out times) of a
port. Where ports have more than one clock edge associated with them (such as DDR outputs), index can be used to obtain
information for all edges. index must be in [0, clockInfoCount), behaviour is undefined otherwise.
BaseArch default: asserts false as unreachable
Bel buckets are subsets of BelIds and cell types used by analytic placer to seperate types of bels during placement. The buckets should form an exact cover over all BelIds and cell types.
Each bel bucket should be BelIds and cell types that are roughly interchangable during placement. Typical buckets are:
- All LUT bels
- All FF bels
- All multipliers bels
- All block RAM bels
- etc.
The bel buckets will be used during analytic placement for spreading prior to strict legality enforcement. It is not required that all bels within a bucket are strictly equivelant.
Strict legality step will enforce those differences, along with additional
local constraints. isValidBelForCellType, and isBelLocationValid are used
to enforce strict legality checks.
Return a list of all bel buckets on the device.
BaseArch default: the list of buckets created by calling init_bel_buckets(), based on calls to getBelBucketForBel for all bels
Return the name of this bel bucket.
BaseArch default: bucket, if BelBucketId is a typedef of IdString
Return the BelBucketId for the specified bucket name.
BaseArch default: bucket_name, if BelBucketId is a typedef of IdString
Returns the bucket for a particular bel.
BaseArch default: getBelBucketForCellType(getBelType(bel))
Returns the bel bucket for a particular cell type.
BaseArch default: getBelBucketByName(cell_type)
Return the list of bels within a bucket.
BaseArch default: the list of bels in the bucket created by calling init_bel_buckets()
Returns true if the given cell can be bound to the given bel. This check should be fast, compared with isBelLocationValid. This check should always return the same value regardless if other cells are placed within the fabric.
BaseArch default: returns cell_type == getBelType(bel)
Returns true if a bel in the current configuration is legal (for example, a flipflop's clock signal is correctly shared with all bels in a slice.)
If and only if explain_invalid is set to true, then a message using
log_nonfatal_error should be printed explaining why the placement is invalid
to the end user.
BaseArch default: returns true
Name of the default placement algorithm for the architecture, if
--placer isn't specified on the command line.
Name of available placer algorithms for the architecture, used
to provide help for and validate --placer.
Name of the default router algorithm for the architecture, if
--router isn't specified on the command line.
Name of available router algorithms for the architecture, used
to provide help for and validate --router.
Gets the root cell of a cluster, which is used as a datum point when placing the cluster.
Gets an approximate bounding box of the cluster. This is intended for area allocation in the placer and is permitted to occasionally give incorrect estimates, for example due to irregularities in the fabric depending on cluster placement. getClusterPlacement should always be used to get exact locations.
Gets the approximate offset of a cell within its cluster, relative to the root cell. This is intended for global placement usage and is permitted to occasionally give incorrect estimates, for example due to irregularities in the fabric depending on cluster placement. getClusterPlacement should always be used to get exact locations.
The returned x and y coordinates, when added to the root location of the cluster, should give an approximate location where cell will end up placed at.
Returns true if the cell must be placed according to the cluster; for example typical carry chains, and dedicated IO routing. Returns false if the cell can be split from the cluster if placement desires, at the expense of a less optimal result (for example dedicated LUT-FF paths where general routing can also be used).
bool getClusterPlacement(ClusterId cluster, BelId root_bel, std::vector<std::pair<CellInfo *, BelId>> &placement) const
Gets an exact placement of the cluster, with the root cell placed on or near root_bel (and always within the same tile). Returns false if no placement is viable, otherwise returns true and populates placement with a list of cells inside the cluster and bels they should be placed at.
This approach of allowing architectures to define cluster placements enables easier handling of irregular fabrics than requiring strict and constant x, y and z offsets.