fixing optimizer and concurrency::reduce

src/libslic3r/Optimize/BruteforceOptimizer.hpp

@@ -1,7 +1,7 @@
 #ifndef BRUTEFORCEOPTIMIZER_HPP
 #define BRUTEFORCEOPTIMIZER_HPP
 
-#include <libslic3r/Optimize/NLoptOptimizer.hpp>
+#include <libslic3r/Optimize/Optimizer.hpp>
 
 namespace Slic3r { namespace opt {
 
@@ -24,19 +24,19 @@ struct AlgBurteForce {
     AlgBurteForce(const StopCriteria &cr, size_t gs): stc{cr}, gridsz{gs} {}
 
     template<int D, size_t N, class Fn, class Cmp>
-    void run(std::array<size_t, N> &idx,
+    bool run(std::array<size_t, N> &idx,
              Result<N> &result,
              const Bounds<N> &bounds,
              Fn &&fn,
              Cmp &&cmp)
     {
-        if (stc.stop_condition()) return;
+        if (stc.stop_condition()) return false;
 
         if constexpr (D < 0) {
             Input<N> inp;
 
             auto max_iter = stc.max_iterations();
-            if (max_iter && num_iter(idx, gridsz) >= max_iter) return;
+            if (max_iter && num_iter(idx, gridsz) >= max_iter) return false;
 
             for (size_t d = 0; d < N; ++d) {
                 const Bound &b = bounds[d];
@@ -46,17 +46,25 @@ struct AlgBurteForce {
 
             auto score = fn(inp);
             if (cmp(score, result.score)) {
+                double absdiff = std::abs(score - result.score);
+
                 result.score = score;
                 result.optimum = inp;
+
+                if (absdiff < stc.abs_score_diff() ||
+                    absdiff < stc.rel_score_diff() * std::abs(score))
+                    return false;
             }
 
         } else {
             for (size_t i = 0; i < gridsz; ++i) {
                 idx[D] = i;
-                run<D - 1>(idx, result, bounds, std::forward<Fn>(fn),
+                if (!run<D - 1>(idx, result, bounds, std::forward<Fn>(fn),
-                           std::forward<Cmp>(cmp));
+                               std::forward<Cmp>(cmp))) return false;
             }
         }
+
+        return true;
     }
 
     template<class Fn, size_t N>

src/libslic3r/SLA/Concurrency.hpp

@@ -43,23 +43,36 @@ template<> struct _ccr<true>
         });
     }
 
-    template<class I, class Fn, class MergeFn, class T>
+    template<class I, class MergeFn, class T, class AccessFn>
     static T reduce(I          from,
                     I          to,
-                    const T & init,
+                    const T   &init,
-                    Fn &&     fn,
                     MergeFn  &&mergefn,
-                    size_t    granularity = 1)
+                    AccessFn &&access,
+                    size_t     granularity = 1
+                    )
     {
         return tbb::parallel_reduce(
             tbb::blocked_range{from, to, granularity}, init,
             [&](const auto &range, T subinit) {
                 T acc = subinit;
-                loop_(range, [&](auto &i) { acc = mergefn(acc, fn(i, acc)); });
+                loop_(range, [&](auto &i) { acc = mergefn(acc, access(i)); });
                 return acc;
             },
             std::forward<MergeFn>(mergefn));
     }
+
+    template<class I, class MergeFn, class T>
+    static IteratorOnly<I, T> reduce(I         from,
+                                     I         to,
+                                     const T & init,
+                                     MergeFn &&mergefn,
+                                     size_t    granularity = 1)
+    {
+        return reduce(
+            from, to, init, std::forward<MergeFn>(mergefn),
+            [](typename I::value_type &i) { return i; }, granularity);
+    }
 };
 
 template<> struct _ccr<false>
@@ -92,18 +105,31 @@ public:
         loop_(from, to, std::forward<Fn>(fn));
     }
 
-    template<class I, class Fn, class MergeFn, class T>
+    template<class I, class MergeFn, class T, class AccessFn>
-    static IntegerOnly<I, T> reduce(I         from,
+    static T reduce(I         from,
                     I         to,
                     const T & init,
-                                    Fn &&     fn,
                     MergeFn  &&mergefn,
-                                    size_t    /*granularity*/ = 1)
+                    AccessFn &&access,
+                    size_t   /*granularity*/ = 1
+                    )
     {
         T acc = init;
-        loop_(from, to, [&](auto &i) { acc = mergefn(acc, fn(i, acc)); });
+        loop_(from, to, [&](auto &i) { acc = mergefn(acc, access(i)); });
         return acc;
     }
+
+    template<class I, class MergeFn, class T>
+    static IteratorOnly<I, T> reduce(I          from,
+                                     I          to,
+                                     const T   &init,
+                                     MergeFn  &&mergefn,
+                                     size_t     /*granularity*/ = 1
+                                     )
+    {
+        return reduce(from, to, init, std::forward<MergeFn>(mergefn),
+                      [](typename I::value_type &i) { return i; });
+    }
 };
 
 using ccr = _ccr<USE_FULL_CONCURRENCY>;

src/libslic3r/SLA/Rotfinder.cpp

@@ -31,7 +31,7 @@ VertexFaceMap create_vertex_face_map(const TriangleMesh &mesh) {
     return vmap;
 }
 
-// Find transformed mesh ground level without copy and with parallell reduce.
+// Find transformed mesh ground level without copy and with parallel reduce.
 double find_ground_level(const TriangleMesh &mesh,
                          const Transform3d & tr,
                          size_t              threads)
@@ -40,15 +40,13 @@ double find_ground_level(const TriangleMesh &mesh,
 
     auto minfn = [](double a, double b) { return std::min(a, b); };
 
-    auto findminz = [&mesh, &tr] (size_t vi, double submin) {
+    auto accessfn = [&mesh, &tr] (size_t vi) {
-        Vec3d v = tr * mesh.its.vertices[vi].template cast<double>();
+        return (tr * mesh.its.vertices[vi].template cast<double>()).z();
-        return std::min(submin, v.z());
     };
 
     double zmin = mesh.its.vertices.front().z();
-
+    size_t granularity = vsize / threads;
-    return ccr_par::reduce(size_t(0), vsize, zmin, findminz, minfn,
+    return ccr_par::reduce(size_t(0), vsize, zmin, minfn, accessfn, granularity);
-                           vsize / threads);
 }
 
 // Try to guess the number of support points needed to support a mesh
@@ -65,7 +63,7 @@ double calculate_model_supportedness(const TriangleMesh & mesh,
 
     double zmin = find_ground_level(mesh, tr, Nthr);
 
-    auto score_mergefn = [&mesh, &tr, zmin](size_t fi, double subscore) {
+    auto accessfn = [&mesh, &tr, zmin](size_t fi) {
 
         static const Vec3d DOWN = {0., 0., -1.};
 
@@ -83,21 +81,18 @@ double calculate_model_supportedness(const TriangleMesh & mesh,
         double zlvl = zmin + 0.1;
         if (p1.z() <= zlvl && p2.z() <= zlvl && p3.z() <= zlvl) {
             //                score += area * POINTS_PER_UNIT_AREA;
-            return subscore;
+            return 0.;
         }
 
         double phi = 1. - std::acos(N.dot(DOWN)) / PI;
-        phi = phi * (phi > 0.5);
+//        phi = phi * (phi > 0.5);
 
         //                    std::cout << "area: " << area << std::endl;
 
-        subscore += area * POINTS_PER_UNIT_AREA * phi;
+        return area * POINTS_PER_UNIT_AREA * phi;
-
-        return subscore;
     };
 
-    double score = ccr_seq::reduce(size_t(0), facesize, 0., score_mergefn,
+    double score = ccr_par::reduce(size_t(0), facesize, 0., std::plus<double>{}, accessfn, facesize / Nthr);
-                                   std::plus<double>{}, facesize / Nthr);
 
     return score / mesh.its.indices.size();
 }
@@ -107,7 +102,7 @@ std::array<double, 2> find_best_rotation(const ModelObject& modelobj,
                                          std::function<void(unsigned)> statuscb,
                                          std::function<bool()> stopcond)
 {
-    static const unsigned MAX_TRIES = 100;
+    static const unsigned MAX_TRIES = 10000;
 
     // return value
     std::array<double, 2> rot;
@@ -158,10 +153,10 @@ std::array<double, 2> find_best_rotation(const ModelObject& modelobj,
                                                   .max_iterations(max_tries)
                                                   .rel_score_diff(1e-6)
                                                   .stop_condition(stopcond),
-                                              10 /*grid size*/);
+                                              100 /*grid size*/);
 
-    // We are searching rotations around the three axes x, y, z. Thus the
+    // We are searching rotations around only two axes x, y. Thus the
-    // problem becomes a 3 dimensional optimization task.
+    // problem becomes a 2 dimensional optimization task.
     // We can specify the bounds for a dimension in the following way:
     auto b = opt::Bound{-PI, PI};
 

tests/libslic3r/CMakeLists.txt

@@ -17,6 +17,7 @@ add_executable(${_TEST_NAME}_tests
 	test_marchingsquares.cpp
 	test_timeutils.cpp
 	test_voronoi.cpp
+    test_optimizers.cpp
     test_png_io.cpp
     test_timeutils.cpp
 	)

tests/libslic3r/test_optimizers.cpp

@@ -0,0 +1,59 @@
+#include <catch2/catch.hpp>
+#include <test_utils.hpp>
+
+#include <libslic3r/Optimize/BruteforceOptimizer.hpp>
+
+#include <libslic3r/Optimize/NLoptOptimizer.hpp>
+
+void check_opt_result(double score, double ref, double abs_err, double rel_err)
+{
+    double abs_diff = std::abs(score - ref);
+    double rel_diff = std::abs(abs_diff / std::abs(ref));
+
+    bool abs_reached = abs_diff < abs_err;
+    bool rel_reached = rel_diff < rel_err;
+    bool precision_reached = abs_reached || rel_reached;
+    REQUIRE(precision_reached);
+}
+
+template<class Opt> void test_sin(Opt &&opt)
+{
+    using namespace Slic3r::opt;
+
+    auto optfunc = [](const auto &in) {
+        auto [phi] = in;
+
+        return std::sin(phi);
+    };
+
+    auto init = initvals({PI});
+    auto optbounds = bounds({ {0., 2 * PI}});
+
+    Result result_min = opt.to_min().optimize(optfunc, init, optbounds);
+    Result result_max = opt.to_max().optimize(optfunc, init, optbounds);
+
+    check_opt_result(result_min.score, -1., 1e-2, 1e-4);
+    check_opt_result(result_max.score,  1., 1e-2, 1e-4);
+}
+
+template<class Opt> void test_sphere_func(Opt &&opt)
+{
+    using namespace Slic3r::opt;
+
+    Result result = opt.to_min().optimize([](const auto &in) {
+        auto [x, y] = in;
+
+        return x * x + y * y + 1.;
+    }, initvals({.6, -0.2}), bounds({{-1., 1.}, {-1., 1.}}));
+
+    check_opt_result(result.score, 1., 1e-2, 1e-4);
+}
+
+TEST_CASE("Test brute force optimzer for basic 1D and 2D functions", "[Opt]") {
+    using namespace Slic3r::opt;
+
+    Optimizer<AlgBruteForce> opt;
+
+    test_sin(opt);
+    test_sphere_func(opt);
+}

tests/sla_print/sla_print_tests.cpp

@@ -5,6 +5,7 @@
 #include "sla_test_utils.hpp"
 
 #include <libslic3r/SLA/SupportTreeMesher.hpp>
+#include <libslic3r/SLA/Concurrency.hpp>
 
 namespace {
 
@@ -239,3 +240,14 @@ TEST_CASE("halfcone test", "[halfcone]") {
     m.require_shared_vertices();
     m.WriteOBJFile("Halfcone.obj");
 }
+
+TEST_CASE("Test concurrency")
+{
+    std::vector<double> vals = grid(0., 100., 10.);
+
+    double ref = std::accumulate(vals.begin(), vals.end(), 0.);
+
+    double s = sla::ccr_par::reduce(vals.begin(), vals.end(), 0., std::plus<double>{});
+
+    REQUIRE(s == Approx(ref));
+}