Zone fill performance improvements

1) better load-balancing for deferred zones
2) sort zones by priority before filling
3) retire BOARD::GetZoneList() which had a horrible performance profile
4) implement a zone bounding box cache
5) better checks for IsCancelled() so long fills can be exited

Fixes https://gitlab.com/kicad/code/kicad/issues/5738
This commit is contained in:
Jeff Young 2020-09-21 20:20:20 +01:00
parent d3dd1c45b6
commit 4badcba4c2
10 changed files with 197 additions and 157 deletions

View File

@ -699,6 +699,8 @@ void ZONE_CONTAINER::TransformShapeWithClearanceToPolygon( SHAPE_POLY_SET& aCorn
return;
aCornerBuffer = m_FilledPolysList.at( aLayer );
aCornerBuffer.Inflate( aClearance, aError );
int numSegs = GetArcToSegmentCount( aClearance, aError, 360.0 );
aCornerBuffer.Inflate( aClearance, numSegs );
aCornerBuffer.Simplify( SHAPE_POLY_SET::PM_STRICTLY_SIMPLE );
}

View File

@ -1701,25 +1701,6 @@ MODULE* BOARD::GetFootprint( const wxPoint& aPosition, PCB_LAYER_ID aActiveLayer
return NULL;
}
std::list<ZONE_CONTAINER*> BOARD::GetZoneList( bool aIncludeZonesInFootprints )
{
std::list<ZONE_CONTAINER*> zones;
for( ZONE_CONTAINER* zone : Zones() )
zones.push_back( zone );
if( aIncludeZonesInFootprints )
{
for( MODULE* mod : m_modules )
{
for( MODULE_ZONE_CONTAINER* zone : mod->Zones() )
zones.push_back( zone );
}
}
return zones;
}
ZONE_CONTAINER* BOARD::AddArea( PICKED_ITEMS_LIST* aNewZonesList, int aNetcode, PCB_LAYER_ID aLayer,
wxPoint aStartPointPosition, ZONE_BORDER_DISPLAY_STYLE aHatch )

View File

@ -879,12 +879,6 @@ public:
return NULL;
}
/**
* Function GetZoneList
* @return a std::list of pointers to all board zones (possibly including zones in footprints)
*/
std::list<ZONE_CONTAINER*> GetZoneList( bool aIncludeZonesInFootprints = false );
/**
* Function GetAreaCount
* @return The number of Areas or ZONE_CONTAINER.

View File

@ -113,11 +113,18 @@ public:
wxString GetZoneName() const { return m_zoneName; }
void SetZoneName( const wxString& aName ) { m_zoneName = aName; }
/** Function GetBoundingBox (virtual)
/**
* Function GetBoundingBox (virtual)
* @return an EDA_RECT that is the bounding box of the zone outline
*/
const EDA_RECT GetBoundingBox() const override;
/**
* ONLY TO BE USED BY CLIENTS WHICH SET UP THE CACHE!
*/
const EDA_RECT GetCachedBoundingBox() const { return m_bboxCache; }
void CacheBoundingBox() { m_bboxCache = GetBoundingBox(); }
/**
* Function GetLocalClearance
* returns any local clearances set in the "classic" (ie: pre-rule) system. These are
@ -912,7 +919,8 @@ protected:
std::map<PCB_LAYER_ID, SHAPE_POLY_SET> m_FilledPolysList;
std::map<PCB_LAYER_ID, SHAPE_POLY_SET> m_RawPolysList;
/// A temp variable used while filling
/// Temp variables used while filling
EDA_RECT m_bboxCache;
std::map<PCB_LAYER_ID, bool> m_fillFlags;
/// A hash value used in zone filling calculations to see if the filled areas are up to date

View File

@ -1019,7 +1019,9 @@ static void export_vrml_zones( MODEL_VRML& aModel, BOARD* aPcb, COMMIT* aCommit
zone->SetFillMode( ZONE_FILL_MODE::POLYGONS ); // use filled polygons
// If the zone fill failed, don't try adding it to the export
if( !filler.Fill( { zone } ) )
std::vector<ZONE_CONTAINER*> toFill = { zone };
if( !filler.Fill( toFill ) )
continue;
}

View File

@ -475,7 +475,7 @@ int PCB_EDITOR_CONTROL::RepairBoard( const TOOL_EVENT& aEvent )
for( BOARD_ITEM* drawing : board()->Drawings() )
processItem( drawing );
for( ZONE_CONTAINER* zone : board()->GetZoneList() )
for( ZONE_CONTAINER* zone : board()->Zones() )
processItem( zone );
for( MARKER_PCB* marker : board()->Markers() )

View File

@ -203,8 +203,9 @@ void ZONE_CREATE_HELPER::commitZone( std::unique_ptr<ZONE_CONTAINER> aZone )
if( !m_params.m_keepout )
{
ZONE_FILLER filler( m_tool.getModel<BOARD>(), &bCommit );
std::vector<ZONE_CONTAINER*> toFill = { aZone.get() };
if( !filler.Fill( { aZone.get() } ) )
if( !filler.Fill( toFill ) )
{
bCommit.Revert();
break;

View File

@ -97,15 +97,15 @@ void ZONE_FILLER_TOOL::FillAllZones( wxWindow* aCaller, PROGRESS_REPORTER* aRepo
BOARD_COMMIT commit( this );
for( auto zone : board()->Zones() )
toFill.push_back(zone);
for( ZONE_CONTAINER* zone : board()->Zones() )
toFill.push_back( zone );
ZONE_FILLER filler( board(), &commit );
if( aReporter )
filler.SetProgressReporter( aReporter );
else
filler.InstallNewProgressReporter( aCaller, _( "Fill All Zones" ), 4 );
filler.InstallNewProgressReporter( aCaller, _( "Fill All Zones" ), 3 );
if( filler.Fill( toFill ) )
{

View File

@ -82,8 +82,7 @@ void ZONE_FILLER::SetProgressReporter( PROGRESS_REPORTER* aReporter )
}
bool ZONE_FILLER::Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck,
wxWindow* aParent )
bool ZONE_FILLER::Fill( std::vector<ZONE_CONTAINER*>& aZones, bool aCheck, wxWindow* aParent )
{
std::vector<std::pair<ZONE_CONTAINER*, PCB_LAYER_ID>> toFill;
std::vector<CN_ZONE_ISOLATED_ISLAND_LIST> islandsList;
@ -109,7 +108,7 @@ bool ZONE_FILLER::Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck,
m_boardOutline.RemoveAllContours();
m_brdOutlinesValid = m_board->GetBoardPolygonOutlines( m_boardOutline );
// Update the bounding box shape caches in the pads to prevent multi-threaded rebuilds
// Update the bounding box and shape caches in the pads to prevent multi-threaded rebuilds.
for( MODULE* module : m_board->Modules() )
{
for( D_PAD* pad : module->Pads() )
@ -119,8 +118,17 @@ bool ZONE_FILLER::Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck,
}
}
// Sort by priority to reduce deferrals waiting on higher priority zones.
std::sort( aZones.begin(), aZones.end(),
[]( const ZONE_CONTAINER* lhs, const ZONE_CONTAINER* rhs )
{
return lhs->GetPriority() > rhs->GetPriority();
} );
for( ZONE_CONTAINER* zone : aZones )
{
zone->CacheBoundingBox();
// Keepout zones are not filled
if( zone->GetIsKeepout() )
continue;
@ -146,84 +154,85 @@ bool ZONE_FILLER::Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck,
zone->SetFillVersion( bds.m_ZoneFillVersion );
}
std::atomic<size_t> nextItem( 0 );
size_t parallelThreadCount = std::min<size_t>( std::thread::hardware_concurrency(),
toFill.size() );
std::vector<std::future<size_t>> returns( parallelThreadCount );
size_t cores = std::thread::hardware_concurrency();
std::atomic<size_t> nextItem;
auto check_fill_dependency =
[&]( ZONE_CONTAINER* aZone, PCB_LAYER_ID aLayer, ZONE_CONTAINER* aOtherZone ) -> bool
{
// Check to see if we have to knock-out the filled areas of a higher-priority
// zone. If so we have to wait until said zone is filled before we can fill.
// If the other zone is already filled then we're good-to-go
if( aOtherZone->GetFillFlag( aLayer ) )
return false;
// Even if keepouts exclude copper pours the exclusion is by outline, not by
// filled area, so we're good-to-go here too.
if( aOtherZone->GetIsKeepout() )
return false;
// If the zones share no common layers
if( !aOtherZone->GetLayerSet().test( aLayer ) )
return false;
if( aOtherZone->GetPriority() <= aZone->GetPriority() )
return false;
// Same-net zones always use outline to produce predictable results
if( aOtherZone->GetNetCode() == aZone->GetNetCode() )
return false;
// A higher priority zone is found: if we intersect and it's not filled yet
// then we have to wait.
EDA_RECT inflatedBBox = aZone->GetCachedBoundingBox();
inflatedBBox.Inflate( worstClearance );
return inflatedBBox.Intersects( aOtherZone->GetCachedBoundingBox() );
};
auto fill_lambda =
[&]( PROGRESS_REPORTER* aReporter ) -> size_t
[&]( PROGRESS_REPORTER* aReporter )
{
std::deque<std::pair<ZONE_CONTAINER*, PCB_LAYER_ID>> deferred;
size_t num = 0;
PCB_LAYER_ID layer = UNDEFINED_LAYER;
ZONE_CONTAINER* zone = nullptr;
while( true )
for( size_t i = nextItem++; i < toFill.size(); i = nextItem++ )
{
size_t i = nextItem++;
if( i < toFill.size() )
{
layer = toFill[i].second;
zone = toFill[i].first;
}
else if( !deferred.empty() )
{
layer = deferred.front().second;
zone = deferred.front().first;
deferred.pop_front();
}
else
{
// all done
break;
}
EDA_RECT zone_boundingbox = zone->GetBoundingBox();
PCB_LAYER_ID layer = toFill[i].second;
ZONE_CONTAINER* zone = toFill[i].first;
bool canFill = true;
zone_boundingbox.Inflate( worstClearance );
// Check for any fill dependencies. If our zone needs to be clipped by
// another zone then we can't fill until that zone is filled.
for( ZONE_CONTAINER* candidate : m_board->GetZoneList( true ) )
for( ZONE_CONTAINER* otherZone : m_board->Zones() )
{
// We don't care about keepouts; even if they exlclude copper pours
// the exclusion is by outline, not by filled area.
if( candidate->GetIsKeepout() )
if( otherZone == zone )
continue;
// If the zones share no common layers
if( !candidate->GetLayerSet().test( layer ) )
continue;
if( candidate->GetPriority() <= zone->GetPriority() )
continue;
// Same-net zones always use outline to produce predictable results
if( candidate->GetNetCode() == zone->GetNetCode() )
continue;
// A higher priority zone is found: if we intersect and it's not
// filled yet then we have to wait.
if( zone_boundingbox.Intersects( candidate->GetBoundingBox() )
&& !candidate->GetFillFlag( layer ) )
if( check_fill_dependency( zone, layer, otherZone ) )
{
canFill = false;
break;
}
}
if( !canFill )
for( MODULE* module : m_board->Modules() )
{
// This isn't ideal from a load-balancing standpoint as another thread
// cannot pick up this thread's deferred zones. However a better
// solution would require a significantly more complex thread-safe queue.
deferred.push_back( { zone, layer } );
continue;
for( ZONE_CONTAINER* otherZone : module->Zones() )
{
if( check_fill_dependency( zone, layer, otherZone ) )
{
canFill = false;
break;
}
}
}
if( m_progressReporter && m_progressReporter->IsCancelled() )
break;
if( !canFill )
continue;
// Now we're ready to fill.
SHAPE_POLY_SET rawPolys, finalPolys;
@ -236,19 +245,21 @@ bool ZONE_FILLER::Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck,
zone->SetFillFlag( layer, true );
if( m_progressReporter )
{
m_progressReporter->AdvanceProgress();
if( m_progressReporter->IsCancelled() )
break;
}
num++;
}
return num;
};
while( !toFill.empty() )
{
size_t parallelThreadCount = std::min( cores, toFill.size() );
std::vector<std::future<size_t>> returns( parallelThreadCount );
nextItem = 0;
if( parallelThreadCount <= 1 )
fill_lambda( m_progressReporter );
else
@ -270,6 +281,17 @@ bool ZONE_FILLER::Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck,
}
}
toFill.erase( std::remove_if( toFill.begin(), toFill.end(),
[&] ( const std::pair<ZONE_CONTAINER*, PCB_LAYER_ID> pair ) -> bool
{
return pair.first->GetFillFlag( pair.second );
} ),
toFill.end() );
if( m_progressReporter && m_progressReporter->IsCancelled() )
break;
}
// Now update the connectivity to check for copper islands
if( m_progressReporter )
{
@ -396,7 +418,7 @@ bool ZONE_FILLER::Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck,
{
m_progressReporter->AdvancePhase();
m_progressReporter->Report( _( "Performing polygon fills..." ) );
m_progressReporter->SetMaxProgress( toFill.size() );
m_progressReporter->SetMaxProgress( islandsList.size() );
}
nextItem = 0;
@ -423,6 +445,9 @@ bool ZONE_FILLER::Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck,
return num;
};
size_t parallelThreadCount = std::min( cores, islandsList.size() );
std::vector<std::future<size_t>> returns( parallelThreadCount );
if( parallelThreadCount <= 1 )
tri_lambda( m_progressReporter );
else
@ -488,7 +513,7 @@ bool hasThermalConnection( D_PAD* pad, const ZONE_CONTAINER* aZone )
int thermalGap = aZone->GetThermalReliefGap( pad );
item_boundingbox.Inflate( thermalGap, thermalGap );
return item_boundingbox.Intersects( aZone->GetBoundingBox() );
return item_boundingbox.Intersects( aZone->GetCachedBoundingBox() );
}
@ -670,7 +695,7 @@ void ZONE_FILLER::buildCopperItemClearances( const ZONE_CONTAINER* aZone, PCB_LA
BOARD_DESIGN_SETTINGS& bds = m_board->GetDesignSettings();
int zone_clearance = aZone->GetLocalClearance();
EDA_RECT zone_boundingbox = aZone->GetBoundingBox();
EDA_RECT zone_boundingbox = aZone->GetCachedBoundingBox();
// Items outside the zone bounding box are skipped, so it needs to be inflated by the
// largest clearance value found in the netclasses and rules
@ -793,48 +818,75 @@ void ZONE_FILLER::buildCopperItemClearances( const ZONE_CONTAINER* aZone, PCB_LA
for( BOARD_ITEM* item : m_board->Drawings() )
doGraphicItem( item );
// Add zones outlines having an higher priority and keepout
// Add keepout zones and higher-priority zones
//
for( ZONE_CONTAINER* zone : m_board->GetZoneList( true ) )
auto knockoutZone =
[&]( ZONE_CONTAINER* aKnockout )
{
// If the zones share no common layers
if( !zone->GetLayerSet().test( aLayer ) )
continue;
if( !aKnockout->GetLayerSet().test( aLayer ) )
return;
if( !zone->GetIsKeepout() && zone->GetPriority() <= aZone->GetPriority() )
continue;
if( zone->GetIsKeepout() && !zone->GetDoNotAllowCopperPour() )
continue;
// A higher priority zone or keepout area is found: remove this area
EDA_RECT item_boundingbox = zone->GetBoundingBox();
if( item_boundingbox.Intersects( zone_boundingbox ) )
if( aKnockout->GetBoundingBox().Intersects( zone_boundingbox ) )
{
if( zone->GetIsKeepout() || aZone->GetNetCode() == zone->GetNetCode() )
if( aKnockout->GetIsKeepout()
|| aZone->GetNetCode() == aKnockout->GetNetCode() )
{
// Keepouts and same-net zones use outline with no clearance
zone->TransformSmoothedOutlineWithClearanceToPolygon( aHoles, 0 );
aKnockout->TransformSmoothedOutlineWithClearanceToPolygon( aHoles, 0 );
}
else
{
int gap = aZone->GetClearance( aLayer, zone );
int gap = aZone->GetClearance( aLayer, aKnockout );
if( bds.m_ZoneFillVersion == 5 )
{
// 5.x used outline with clearance
zone->TransformSmoothedOutlineWithClearanceToPolygon( aHoles, gap );
aKnockout->TransformSmoothedOutlineWithClearanceToPolygon( aHoles, gap );
}
else
{
// 6.0 uses filled areas with clearance
SHAPE_POLY_SET knockout;
zone->TransformShapeWithClearanceToPolygon( knockout, aLayer, gap );
aHoles.Append( knockout );
SHAPE_POLY_SET poly;
aKnockout->TransformShapeWithClearanceToPolygon( poly, aLayer, gap );
aHoles.Append( poly );
}
}
}
};
for( ZONE_CONTAINER* otherZone : m_board->Zones() )
{
if( otherZone == aZone )
continue;
if( otherZone->GetIsKeepout() )
{
if( otherZone->GetDoNotAllowCopperPour() )
knockoutZone( otherZone );
}
else
{
if( otherZone->GetPriority() > aZone->GetPriority() )
knockoutZone( otherZone );
}
}
for( MODULE* module : m_board->Modules() )
{
for( ZONE_CONTAINER* otherZone : module->Zones() )
{
if( otherZone->GetIsKeepout() )
{
if( otherZone->GetDoNotAllowCopperPour() )
knockoutZone( otherZone );
}
else
{
if( otherZone->GetPriority() > aZone->GetPriority() )
knockoutZone( otherZone );
}
}
}
aHoles.Simplify( SHAPE_POLY_SET::PM_FAST );
@ -1087,7 +1139,7 @@ bool ZONE_FILLER::fillSingleZone( ZONE_CONTAINER* aZone, PCB_LAYER_ID aLayer,
void ZONE_FILLER::buildThermalSpokes( const ZONE_CONTAINER* aZone, PCB_LAYER_ID aLayer,
std::deque<SHAPE_LINE_CHAIN>& aSpokesList )
{
auto zoneBB = aZone->GetBoundingBox();
auto zoneBB = aZone->GetCachedBoundingBox();
int zone_clearance = aZone->GetLocalClearance();
int biggest_clearance = m_board->GetDesignSettings().GetBiggestClearanceValue();
biggest_clearance = std::max( biggest_clearance, zone_clearance );
@ -1349,7 +1401,7 @@ void ZONE_FILLER::addHatchFillTypeOnZone( const ZONE_CONTAINER* aZone, PCB_LAYER
// one of the holes. Effectively this means their copper outline needs to be expanded
// to be at least as wide as the gap so that it is guaranteed to touch at least one
// edge.
EDA_RECT zone_boundingbox = aZone->GetBoundingBox();
EDA_RECT zone_boundingbox = aZone->GetCachedBoundingBox();
SHAPE_POLY_SET aprons;
int min_apron_radius = ( aZone->GetHatchGap() * 10 ) / 19;

View File

@ -44,7 +44,7 @@ public:
void SetProgressReporter( PROGRESS_REPORTER* aReporter );
void InstallNewProgressReporter( wxWindow* aParent, const wxString& aTitle, int aNumPhases );
bool Fill( const std::vector<ZONE_CONTAINER*>& aZones, bool aCheck = false,
bool Fill( std::vector<ZONE_CONTAINER*>& aZones, bool aCheck = false,
wxWindow* aParent = nullptr );
private: