- CN_CONNECTIVITY_ALGO::ForEachItem
move implementation to header, so that a lambda can be inlined by the
compiler. improves iteration performance x2
- CN_ITEM::Net
move implementation to header, so that it can be inlined by the
compiler.
- CONNECTIVITY_DATA::GetNetItems
don't use std::set to filter out duplicates. use std::unique on sorted
vector. use a bitset to perform type check.
With pads, we should check available anchor points for each connection,
increasing probability of finding connection to match fill algorithm
Fixes: lp:1844661
* https://bugs.launchpad.net/kicad/+bug/1844661
Pads not on copper layers now do not have a netname, and are no longer taken in account in connectivity.
Especially it avoid creating useless rats-nests for these pads.
Give the user the option of cancelling a file open if there are
segment zones; otherwise they're converted to polygon fills.
Fixes: lp:1823087
* https://bugs.launchpad.net/kicad/+bug/1823087
It is possible for an async thread to simply be deferred if not
specifically set to async. In which case, until the return value is
requested, the loop may simply wait.
By decoupling from std::thread, we can avoid the wait/sleep cycle in
checking the std::atomic completion variable. The std::future variables
are immediately returning without the additional atomic check cycle.
This minimizes the thread usage, particularly during online actions such
as moving items and routing.
UpdateRatsnest: Minimized the number of atomics by utilizing future
promise. Avoided updating node-less nets. Node-less nets are common in
the dynamic ratsnest because the net vector is created from the board
nets by value but only populated with a subset. So the dynamic ratsnest
calculation always saw every net as dirty and spun up extra threads.
searchConnections: ParallelThreadCount is no longer lower bounded.
Small item counts are handled inline rather than with a separate thread.
Locks protect the std::set in each item. Devolving the mutex to the
CN_ITEM allows multiple threads to make simultaneous connections to
different items where they do not conflict.
The connectivity files were unwieldy. This separates them logically
into data, algo and items where the items classes are those that hold,
surprise, surprise, the items, lists and clusters.
Updating connectivity should not need to iterate over the full item list
in each thread. Instead, we collect the dirty items first and then
iterate only over the dirty list.
SEGZONE types were confusingly named PCB_ZONE_T. Zones in pcbnew are
now _only_ PCB_ZONE_AREA_T, so we name segzone types PCB_SEGZONE_T to be
clear.
This also removes processing of the SEGZONEs from connectivity
calculations.
This is a boyscouting commit to standardize the threading of zone fills.
We do not need to join threads after their completion, instead we simply
allow them to clean up their memory without blocking the user. This
also sets the maximum number of threads that may be created to the
number of zones being filled. More than this will only leave un-used
threads being created and immediately killed.
We also include the connectivity search as a phase in the fill progress
reporter. This was the case before but did not utilize the correct
maxsize, leading to stalled progress bar.
OpenMP is not available for macos, so moving to a std::threads
implementation brings platforms into shared code.
This also reduces the OpenMP overhead when computing connectivity and
ratsnests.
Fixes: lp:1780130
* https://bugs.launchpad.net/kicad/+bug/1780130
This prevents non-copper layers from being included in the connectivity
search. It also limits the layer search in the connectivity RTree to
just the copper layer range.
After refilling zones, we need to unconditionally remove the filled
polys from the map because their fills map have changed. Left-over
filled polygons will not map to current polygons, creating "ghost"
islands in the connectivity. Removing these can overflow the polygon
count.
Fixes: lp:1778454
* https://bugs.launchpad.net/kicad/+bug/1778454
The itemlist and zonelist in connectivity can both contain many items,
so both use the same OpenMP routine. However, we can only clear the
dirty flag when we conduct a full search, including zones. Otherwise we
missing connections to zones when propogating changed items that are
then marked as not dirty.
Fixes: lp:1777993
* https://bugs.launchpad.net/kicad/+bug/1777993
Connectivity searches were slow for large, complex boards. This was
partly due to the need to search all lists for possible connections
between anchors.
We remove this requirement by creating an RTree based on items' bounding
boxes similar to the PNS router. Then we get both the colliding items
and search forward/backward to see if there are connections between the
anchors and the other item.
Because we use RTree, we no longer need the maintenance overhead of multiple
item lists and an anchor list in the connectivity class.
Fixes: lp:1701791
* https://bugs.launchpad.net/kicad/+bug/1701791
Fixes: lp:1769408
* https://bugs.launchpad.net/kicad/+bug/1769408
Fixes: lp:1761989
* https://bugs.launchpad.net/kicad/+bug/1761989
We only need to iterate over the anchors when there are items that are
marked invalid. We check once in the item list and only if there are
invalid items to remove do we trim the anchor lists.
However, the connected items might still need to be trimmed, so we leave
this final step outside of the conditional.
This is a speed commit for large boards. Tracks and pads cannot connect
to elements that are not on the same layer. Rather than checking for
this at the last step, this commit splits the anchor vectors by layer,
limiting the initial search space.
Previously, binary search was hand-coded. This moves the search to a
std::algorithm variant. Also searches bbox by limits rather than
directly iterating.
On OpenMP single core systems only the first thread launched. It means
that only the UI updating thread was running, without the one
responsible for actual computation. It is fixed by enforcing at least
two threads.
Oleg Endo
Seth Hillbrand
Ian McInerney
Jeff Young
Seth Hillbrand
jean-pierre charras
Jon Evans
Tomasz Włostowski
Maciej Suminski
Camille