From eb16ec27cb6baf07c235c86b6a474f3ecc6e9e2a Mon Sep 17 00:00:00 2001 From: supermerill Date: Tue, 11 Jun 2024 14:08:07 +0200 Subject: [PATCH] improved concurrency for perimeter & infill (doesn't block the complicated stuff on only one thread) and various little modification (types, debug test) --- src/libslic3r/Fill/Fill.cpp | 24 +++++++++++++++++++----- src/libslic3r/GCode.cpp | 12 ++++++++++++ src/libslic3r/MultiPoint.cpp | 10 +++++----- src/libslic3r/MultiPoint.hpp | 2 +- src/libslic3r/Point.hpp | 10 +++++----- src/libslic3r/PrintObject.cpp | 27 ++++++++++++++++++--------- 6 files changed, 60 insertions(+), 25 deletions(-) diff --git a/src/libslic3r/Fill/Fill.cpp b/src/libslic3r/Fill/Fill.cpp index 6577395f17e..99f5d1ec925 100644 --- a/src/libslic3r/Fill/Fill.cpp +++ b/src/libslic3r/Fill/Fill.cpp @@ -716,11 +716,25 @@ void Layer::make_fills(FillAdaptive::Octree* adaptive_fill_octree, FillAdaptive: for(auto &t : temp) real_surface += t.area(); assert(compute_volume.volume < unscaled(unscaled(surface_fill.surface.area())) * surface_fill.params.layer_height + EPSILON); double area = unscaled(unscaled(real_surface)); - assert(compute_volume.volume <= area * surface_fill.params.layer_height * 1.001 || f->debug_verify_flow_mult <= 0.8); - if(compute_volume.volume > 0) //can fail for thin regions - assert(compute_volume.volume >= area * surface_fill.params.layer_height * 0.999 || f->debug_verify_flow_mult >= 1.3 || f->debug_verify_flow_mult == 0 // sawtooth output more filament,as it's 3D (debug_verify_flow_mult==0) - || area < std::max(1.,surface_fill.params.config->solid_infill_below_area.value) || area < std::max(1.,surface_fill.params.config->solid_infill_below_layer_area.value)); - } + if(surface_fill.surface.has_pos_top()) + area *= surface_fill.params.config->fill_top_flow_ratio.value; + if(surface_fill.surface.has_pos_bottom() && f->layer_id == 0) + area *= surface_fill.params.config->first_layer_flow_ratio.value; + if(surface_fill.surface.has_mod_bridge() && f->layer_id == 0) + area *= surface_fill.params.config->first_layer_flow_ratio.value; + //TODO: over-bridge mod + if(surface_fill.params.config->over_bridge_flow_ratio.value == 1){ + assert(compute_volume.volume <= area * surface_fill.params.layer_height * 1.001 || f->debug_verify_flow_mult <= 0.8); + if(compute_volume.volume > 0) //can fail for thin regions + assert( + compute_volume.volume >= area * surface_fill.params.layer_height * 0.999 || + f->debug_verify_flow_mult >= 1.3 || + f->debug_verify_flow_mult == + 0 // sawtooth output more filament,as it's 3D (debug_verify_flow_mult==0) + || area < std::max(1., surface_fill.params.config->solid_infill_below_area.value) || + area < std::max(1., surface_fill.params.config->solid_infill_below_layer_area.value)); + } + } #endif } } diff --git a/src/libslic3r/GCode.cpp b/src/libslic3r/GCode.cpp index 213ccee2774..35b072198cb 100644 --- a/src/libslic3r/GCode.cpp +++ b/src/libslic3r/GCode.cpp @@ -4142,6 +4142,18 @@ void GCode::split_at_seam_pos(ExtrusionLoop& loop, bool was_clockwise) if (loop.paths.empty()) return; +#if _DEBUG + ExtrusionLoop old_loop = loop; + for (const ExtrusionPath &path : loop.paths) + for (int i = 1; i < path.polyline.get_points().size(); ++i) + assert(!path.polyline.get_points()[i - 1].coincides_with_epsilon(path.polyline.get_points()[i])); + for (auto it = std::next(loop.paths.begin()); it != loop.paths.end(); ++it) { + assert(it->polyline.size() >= 2); + assert(std::prev(it)->polyline.back() == it->polyline.front()); + } + assert(loop.first_point() == loop.last_point()); +#endif + // SeamPosition seam_position = m_config.seam_position; // if (loop.loop_role() == elrSkirt) // seam_position = spNearest; diff --git a/src/libslic3r/MultiPoint.cpp b/src/libslic3r/MultiPoint.cpp index 313c96f6497..3dfaa2165d4 100644 --- a/src/libslic3r/MultiPoint.cpp +++ b/src/libslic3r/MultiPoint.cpp @@ -63,16 +63,16 @@ int MultiPoint::find_point(const Point &point) const return -1; // not found } -int MultiPoint::find_point(const Point &point, double scaled_epsilon) const +int MultiPoint::find_point(const Point &point, coordf_t scaled_epsilon) const { if (scaled_epsilon == 0) return this->find_point(point); - auto dist2_min = std::numeric_limits::max(); - auto eps2 = scaled_epsilon * scaled_epsilon; - int idx_min = -1; + coordf_t dist2_min = std::numeric_limits::max(); + coordf_t eps2 = scaled_epsilon * scaled_epsilon; + int idx_min = -1; for (const Point &pt : this->points) { - double d2 = (pt - point).cast().squaredNorm(); + coordf_t d2 = pt.distance_to_square(point); //(pt - point).cast().squaredNorm(); if (d2 < dist2_min) { idx_min = int(&pt - &this->points.front()); dist2_min = d2; diff --git a/src/libslic3r/MultiPoint.hpp b/src/libslic3r/MultiPoint.hpp index c65e9807cba..33edd409738 100644 --- a/src/libslic3r/MultiPoint.hpp +++ b/src/libslic3r/MultiPoint.hpp @@ -50,7 +50,7 @@ class MultiPoint int find_point(const Point &point) const; // Return index of the closest point to point closer than scaled_epsilon. // Return -1 if no such point exists. - int find_point(const Point &point, const double scaled_epsilon) const; + int find_point(const Point &point, const coordf_t scaled_epsilon) const; bool has_boundary_point(const Point &point) const; int closest_point_index(const Point &point) const { int idx = -1; diff --git a/src/libslic3r/Point.hpp b/src/libslic3r/Point.hpp index 0d5b8eec490..51ef3d5d2ae 100644 --- a/src/libslic3r/Point.hpp +++ b/src/libslic3r/Point.hpp @@ -179,13 +179,13 @@ class Point : public Vec2crd Point projection_onto(const Line &line) const; Point interpolate(const double percent, const Point &p) const; - double distance_to(const Point &point) const { return (point - *this).cast().norm(); } - double distance_to_square(const Point &point) const { - double dx = double(point.x() - this->x()); - double dy = double(point.y() - this->y()); + coordf_t distance_to(const Point &point) const { return (point - *this).cast().norm(); } + coordf_t distance_to_square(const Point &point) const { + coordf_t dx = double(point.x() - this->x()); + coordf_t dy = double(point.y() - this->y()); return dx*dx + dy*dy; } - double distance_to(const Line &line) const; + coordf_t distance_to(const Line &line) const; bool coincides_with(const Point &point) const { return this->x() == point.x() && this->y() == point.y(); } bool coincides_with_epsilon(const Point &point) const { return std::abs(this->x() - point.x()) < SCALED_EPSILON/2 && std::abs(this->y() - point.y()) < SCALED_EPSILON/2; diff --git a/src/libslic3r/PrintObject.cpp b/src/libslic3r/PrintObject.cpp index 8d06e30f780..d981f8c9c98 100644 --- a/src/libslic3r/PrintObject.cpp +++ b/src/libslic3r/PrintObject.cpp @@ -150,16 +150,20 @@ namespace Slic3r { // but we don't generate any extra perimeter if fill density is zero, as they would be floating // inside the object - infill_only_where_needed should be the method of choice for printing // hollow objects - for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) { - const PrintRegion ®ion = this->printing_region(region_id); + for (size_t region_id = 0; region_id < this->num_printing_regions(); ++ region_id) { + const PrintRegion ®ion = this->printing_region(region_id); if (!region.config().extra_perimeters || region.config().perimeters == 0 || region.config().fill_density == 0 || this->layer_count() < 2) continue; - + // use an antomic idx instead of the range, to avoid a thread being very late because it's on the difficult layers. + //TODO: sort the layers by difficulty (difficult first) (number of points, region, surfaces, .. ?) + std::atomic_size_t next_layer_idx(0); BOOST_LOG_TRIVIAL(debug) << "Generating extra perimeters for region " << region_id << " in parallel - start"; tbb::parallel_for( tbb::blocked_range(0, m_layers.size() - 1), - [this, ®ion, region_id](const tbb::blocked_range& range) { - for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++layer_idx) { + [this, ®ion, region_id, &next_layer_idx](const tbb::blocked_range& range) { + //TODO: find a better wya to just fire the threads. + //for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++layer_idx) { + for (size_t layer_idx = next_layer_idx++; layer_idx < m_layers.size() - 1; layer_idx = next_layer_idx++) { m_print->throw_if_canceled(); LayerRegion &layerm = *m_layers[layer_idx]->get_region(region_id); const LayerRegion &upper_layerm = *m_layers[layer_idx+1]->get_region(region_id); @@ -214,10 +218,12 @@ namespace Slic3r { } BOOST_LOG_TRIVIAL(debug) << "Generating perimeters in parallel - start"; + std::atomic_size_t next_layer_idx(0); tbb::parallel_for( tbb::blocked_range(0, m_layers.size()), - [this, &atomic_count, nb_layers_update](const tbb::blocked_range& range) { - for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++layer_idx) { + [this, &atomic_count, nb_layers_update, &next_layer_idx](const tbb::blocked_range& range) { + //for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++layer_idx) { + for (size_t layer_idx = next_layer_idx++; layer_idx < m_layers.size(); layer_idx = next_layer_idx++) { std::chrono::time_point start_make_perimeter = std::chrono::system_clock::now(); m_print->throw_if_canceled(); m_layers[layer_idx]->make_perimeters(); @@ -456,10 +462,13 @@ namespace Slic3r { const int nb_layers_update = std::max(1, (int)m_layers.size() / 20); BOOST_LOG_TRIVIAL(debug) << "Filling layers in parallel - start"; + std::atomic_size_t next_layer_idx(0); tbb::parallel_for( tbb::blocked_range(0, m_layers.size()), - [this, &adaptive_fill_octree = adaptive_fill_octree, &support_fill_octree = support_fill_octree, &lightning_generator, &atomic_count, nb_layers_update](const tbb::blocked_range& range) { - for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++layer_idx) { + [this, &adaptive_fill_octree = adaptive_fill_octree, &support_fill_octree = support_fill_octree, &lightning_generator, &atomic_count, nb_layers_update, &next_layer_idx] + (const tbb::blocked_range& range) { + //for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++layer_idx) { + for (size_t layer_idx = next_layer_idx++; layer_idx < m_layers.size(); layer_idx = next_layer_idx++) { std::chrono::time_point start_make_fill = std::chrono::system_clock::now(); m_print->throw_if_canceled(); m_layers[layer_idx]->make_fills(adaptive_fill_octree.get(), support_fill_octree.get(), lightning_generator.get());