GraphOps

Basic operations on graphs.

Documentation

addNode :: Uniquable k => k -> Node k cls color -> Graph k cls color -> Graph k cls color

Add a node to the graph, linking up its edges

delNode :: Uniquable k => k -> Graph k cls color -> Maybe (Graph k cls color)

Delete a node and all its edges from the graph.

getNode :: Uniquable k => Graph k cls color -> k -> Node k cls color

Get a node from the graph, throwing an error if it's not there

lookupNode :: Uniquable k => Graph k cls color -> k -> Maybe (Node k cls color)

Lookup a node from the graph.

modNode :: Uniquable k => (Node k cls color -> Node k cls color) -> k -> Graph k cls color -> Maybe (Graph k cls color)

Modify a node in the graph. returns Nothing if the node isn't present.

size :: Graph k cls color -> Int

Get the size of the graph, O(n)

union :: Graph k cls color -> Graph k cls color -> Graph k cls color

Union two graphs together.

addConflict :: Uniquable k => (k, cls) -> (k, cls) -> Graph k cls color -> Graph k cls color

Add a conflict between nodes to the graph, creating the nodes required. Conflicts are virtual regs which need to be colored differently.

delConflict :: Uniquable k => k -> k -> Graph k cls color -> Maybe (Graph k cls color)

Delete a conflict edge. k1 -> k2 returns Nothing if the node isn't in the graph

addConflicts :: Uniquable k => UniqSet k -> (k -> cls) -> Graph k cls color -> Graph k cls color

Add some conflicts to the graph, creating nodes if required. All the nodes in the set are taken to conflict with each other.

addCoalesce :: Uniquable k => (k, cls) -> (k, cls) -> Graph k cls color -> Graph k cls color

Add a coalescence edge to the graph, creating nodes if requried. It is considered adventageous to assign the same color to nodes in a coalesence.

delCoalesce :: Uniquable k => k -> k -> Graph k cls color -> Maybe (Graph k cls color)

Delete a coalescence edge (k1 -> k2) from the graph.

addExclusion :: (Uniquable k, Uniquable color) => k -> (k -> cls) -> color -> Graph k cls color -> Graph k cls color

Add an exclusion to the graph, creating nodes if required. These are extra colors that the node cannot use.

addExclusions :: (Uniquable k, Uniquable color) => k -> (k -> cls) -> [color] -> Graph k cls color -> Graph k cls color

addPreference :: Uniquable k => (k, cls) -> color -> Graph k cls color -> Graph k cls color

Add a color preference to the graph, creating nodes if required. The most recently added preference is the most prefered. The algorithm tries to assign a node it's prefered color if possible.

coalesceNodes

Arguments

:: (Uniquable k, Ord k, Eq cls)

=> Bool

If True, coalesce nodes even if this might make the graph less colorable (aggressive coalescing)

-> Triv k cls color

-> Graph k cls color

-> (k, k)

keys of the nodes to be coalesced

-> (Graph k cls color, Maybe (k, k))

Coalesce this pair of nodes unconditionally / aggressively. The resulting node is the one with the least key.

returns: Just the pair of keys if the nodes were coalesced the second element of the pair being the least one

Nothing if either of the nodes weren't in the graph

coalesceGraph

Arguments

:: (Uniquable k, Ord k, Eq cls, Outputable k)

=> Bool

If True, coalesce nodes even if this might make the graph less colorable (aggressive coalescing)

-> Triv k cls color

-> Graph k cls color

-> (Graph k cls color, [(k, k)])

Do aggressive coalescing on this graph. returns the new graph and the list of pairs of nodes that got coalesced together. for each pair, the resulting node will have the least key and be second in the pair.

freezeNode

Arguments

:: Uniquable k

=> k

key of the node to freeze

-> Graph k cls color

the graph

-> Graph k cls color

graph with that node frozen

Freeze a node This is for the iterative coalescer. By freezing a node we give up on ever coalescing it. Move all its coalesce edges into the frozen set - and update back edges from other nodes.

freezeOneInGraph :: Uniquable k => Graph k cls color -> (Graph k cls color, Bool)

Freeze one node in the graph This if for the iterative coalescer. Look for a move related node of low degree and freeze it.

We probably don't need to scan the whole graph looking for the node of absolute lowest degree. Just sample the first few and choose the one with the lowest degree out of those. Also, we don't make any distinction between conflicts of different classes.. this is just a heuristic, after all.

IDEA: freezing a node might free it up for Simplify.. would be good to check for triv right here, and add it to a worklist if known triv/non-move nodes.

freezeAllInGraph :: Uniquable k => Graph k cls color -> Graph k cls color

Freeze all the nodes in the graph for debugging the iterative allocator.

scanGraph :: (Node k cls color -> Bool) -> Graph k cls color -> [Node k cls color]

Find all the nodes in the graph that meet some criteria

setColor :: Uniquable k => k -> color -> Graph k cls color -> Graph k cls color

Set the color of a certain node

validateGraph

Arguments

:: (Uniquable k, Outputable k, Eq color)

=> SDoc

extra debugging info to display on error

-> Bool

whether this graph is supposed to be colored.

-> Graph k cls color

graph to validate

-> Graph k cls color

validated graph

validate the internal structure of a graph all its edges should point to valid nodes If they don't then throw an error

slurpNodeConflictCount

Arguments

:: Graph k cls color

-> UniqFM (Int, Int)

(conflict neighbours, num nodes with that many conflicts)

Slurp out a map of how many nodes had a certain number of conflict neighbours