Added dependencies on the Intel Thread Building Blocks.

Changed the C++ parallelization code to Intel Thread Building Blocks.

xs/Build.PL

@@ -8,8 +8,18 @@ use ExtUtils::CppGuess;
 use Module::Build::WithXSpp;
 
 my $cpp_guess = ExtUtils::CppGuess->new;
+my $lib_ext = ${$cpp_guess}{config}{lib_ext};
 my $mswin = $^O eq 'MSWin32';
 
+# Library paths to search for boost, thread building blocks and such.
+# On Windows, there is really no standard. On Unices, this is a bit better.
+my @library_path_prefixes = ();
+if ($mswin) {
+    @library_path_prefixes = ("C:\\", "C:\\dev\\", "C:\\local\\", "D:\\", "D:\\dev\\", "D:\\local\\");
+} else {
+    @library_path_prefixes = qw(/opt/local/ /usr/local/ /opt/ /usr/);
+}
+
 # _GLIBCXX_USE_C99 : to get the long long type for g++
 # HAS_BOOL         : stops Perl/lib/CORE/handy.h from doing "#  define bool char" for MSVC
 # NOGDI            : prevents inclusion of wingdi.h which defines functions Polygon() and Polyline() in global namespace
@@ -76,24 +86,23 @@ if (defined $ENV{BOOST_INCLUDEDIR}) {
     # Boost library was not defined by the environment.
     # Try to guess at some default paths.
     if ($mswin) {
-        for my $path (glob('C:\dev\boost*\include'), glob ('C:\boost*\include')) {
+        for my $path (map glob($_ . 'boost*\include'), @library_path_prefixes) {
             push @boost_include, $path;
         }
         if (! @boost_include) {
             # No boost\include. Try to include the boost root.
-            for my $path (glob('C:\dev\boost*'), glob ('C:\boost*')) {
+            for my $path (map glob($_ . 'boost*'), @library_path_prefixes) {
                 push @boost_include, $path;
             }
         }
     } else {
-        push @boost_include, grep { -d $_ }
-            qw(/opt/local/include /usr/local/include /opt/include /usr/include);
+        @boost_include = grep { -d $_ . '/boost' } map { $_ . 'include' } @library_path_prefixes;
     }
 }
 
 my @boost_libs = ();
 if (defined $ENV{BOOST_LIBRARYDIR}) {
-    push @boost_libs, $ENV{BOOST_LIBRARYDIR}
+    push @boost_libs, $ENV{BOOST_LIBRARYDIR};
 } elsif (defined $ENV{BOOST_DIR}) {
     my $subdir = $ENV{BOOST_DIR} . ($mswin ? '\stage\lib' : '/stage/lib');
     if (-d $subdir) {
@@ -105,14 +114,11 @@ if (defined $ENV{BOOST_LIBRARYDIR}) {
     # Boost library was not defined by the environment.
     # Try to guess at some default paths.
     if ($mswin) {
-        for my $path (
-            glob('C:\dev\boost*\lib'),       glob ('C:\boost*\lib'), 
-            glob('C:\dev\boost*\stage\lib'), glob ('C:\boost*\stage\lib')) {
+        for my $path (map (glob($_ . 'boost*\lib'), glob($_ . 'boost*\stage\lib')), @library_path_prefixes) {
             push @boost_libs, $path;
         }
     } else {
-        push @boost_libs, grep { -d $_ }
-            qw(/opt/local/lib /usr/local/lib /opt/lib /usr/lib /lib);
+        push @boost_libs, grep { -d $_ } map $_ . 'lib', @library_path_prefixes;
     }
 }
 
@@ -138,7 +144,6 @@ if ($have_boost) {
 } else {
     # Either static linking, or check_lib could not be used to find the boost libraries.
     my $lib_prefix = 'libboost_';
-    my $lib_ext = ${$cpp_guess}{config}{lib_ext};
     PATH: foreach my $path (@boost_libs) {
         # Try to find the boost system library.
         my @files = glob "$path/${lib_prefix}system*$lib_ext";
@@ -185,6 +190,110 @@ is handy, if you have built Boost libraries with mutliple settings.
 
 EOF
 
+# Search for the Intel Thread Building Blocks.
+my @tbb_include = ();
+if (defined $ENV{TBB_INCLUDEDIR}) {
+    push @tbb_include, $ENV{TBB_INCLUDEDIR}
+} elsif (defined $ENV{TBB_DIR}) {
+    my $subdir = $ENV{TBB_DIR} . (($mswin == 1) ? '\include' : '/include');
+    push @tbb_include, $subdir if (-d $subdir);
+} else {
+    # Thread Building Blocks library was not defined by the environment.
+    # Try to guess at some default paths.
+    if ($mswin) {
+        for my $path (map glob($_ . 'tbb*\include'), @library_path_prefixes) {
+            push @tbb_include, $path;
+        }
+    } else {
+        @tbb_include = grep { -d $_ . '/tbb' } map { $_ . 'include' } @library_path_prefixes;
+    }
+}
+
+my @tbb_libs = ();
+if (defined $ENV{TBB_LIBRARYDIR}) {
+    push @tbb_libs, $ENV{TBB_LIBRARYDIR}
+} elsif (defined $ENV{TBB_DIR}) {
+    my $subdir = $ENV{TBB_DIR} . ($mswin ? '\lib' : '/lib');
+    push @tbb_libs, $subdir if (-d $subdir);
+} else {
+    # Thread Building Blocks library was not defined by the environment.
+    # Try to guess at some default paths.
+    if ($mswin) {
+        for my $path (map { glob($_ . 'tbb*\lib') } @library_path_prefixes) {
+            push @tbb_libs, $path;
+        }
+    } else {
+        @tbb_libs = grep { -d $_ } map { $_ . 'lib' } @library_path_prefixes;
+    }
+}
+
+# In order to generate the -l switches we need to know how Thread Building Blocks libraries are named
+my $have_tbb = 0;
+#my @tbb_libraries = qw(tbb tbbmalloc tbbmalloc_proxy); # we need these
+my @tbb_libraries = qw(tbb); # we need these
+
+if (!$ENV{SLIC3R_STATIC}) {
+    # Dynamic linking of Thread Building Blocks libraries.
+    if (! $mswin) {
+        # Check without explicit lib path (works on Linux and OSX).
+        $have_tbb = 1
+            if check_lib(
+                lib     => [ @tbb_libraries ],
+            );
+    }
+}
+
+if ($have_tbb) {
+    # The Thread Building Blocks library was detected by check_lib on Linux.
+    push @LIBS, map "-l${_}", @tbb_libraries;
+} else {
+    # Either static linking, or check_lib could not be used to find the Thread Building Blocks libraries.
+    my $lib_prefix = $cpp_guess->is_msvc ? '' : 'lib';
+    PATH: foreach my $path (@tbb_libs) {
+        # Try to find the Thread Building Blocks system library.
+        my @files = glob "$path/${lib_prefix}tbb*$lib_ext";
+        next if !@files;
+        if ($files[0] =~ /\Q${lib_prefix}tbb\E([^.]*)\Q$lib_ext\E$/) {
+            # Suffix contains the version number, the build type etc.
+            my $suffix = $1;
+            # Verify existence of all required TBB libraries at $path.
+            for my $lib (map "${lib_prefix}${_}${suffix}${lib_ext}", @tbb_libraries) {
+                # If the library file does not exist, try next library path.
+                -f "$path/$lib" or next PATH;
+            }
+            if (! $cpp_guess->is_msvc) {
+                # Test the correctness of TBB libraries by linking them to a minimal C program.
+                check_lib(
+                    lib     => [ map "${_}${suffix}", @tbb_libraries ],
+                    INC     => join(' ', map "-I$_", @INC,  @tbb_include),
+                    LIBS    => "-L$path",
+                ) or next;
+            }
+            push @INC, (map " -I$_", @tbb_include);  # TODO: only use the one related to the chosen lib path
+            if ($ENV{SLIC3R_STATIC} || $cpp_guess->is_msvc) {
+                push @LIBS, map "${path}/${_}_static.lib", @tbb_libraries;
+            } else {
+                push @LIBS, " -L$path", (map " -l$_$suffix", @tbb_libraries);
+            }
+            $have_tbb = 1;
+            last;
+        }
+    }
+}
+die <<'EOF' if !$have_tbb;
+Slic3r requires the Intel Thread Building Blocks libraries. Please make sure the library is installed.
+
+If the Intel Thread Building Blocks library is installed, this script should be able to locate them in several
+standard locations. If this is not the case, you might want to supply a path to the library 
+through the TBB_DIR environment variable:
+
+    TBB_DIR=/path/to/TBB perl Build.PL
+
+Or you may specify TBB_INCLUDEPATH and TBB_LIBRARYPATH separatly, which
+is handy, if you have built the Thread Building Blocks libraries with mutliple settings.
+
+EOF
+
 # Add the OpenGL and GLU libraries.
 if ($ENV{SLIC3R_GUI}) {
     if ($mswin) {
@@ -206,15 +315,6 @@ if ($ENV{SLIC3R_DEBUG}) {
     # Disable asserts in the release builds.
     push @cflags, '-DNDEBUG';
 }
-if ($cpp_guess->is_gcc) {
-    # check whether we're dealing with a buggy GCC version
-    # see https://github.com/alexrj/Slic3r/issues/1965
-    if (`cc --version` =~ m/ 4\.7\.[012]/) {
-        # Workaround suggested by Boost devs:
-        # https://svn.boost.org/trac/boost/ticket/8695
-        push @cflags, qw(-fno-inline-small-functions);
-    }
-}
 
 print "\n";
 print 'With @cflags: ', join(', ', map "\"$_\"", @cflags), "\n";

xs/src/libslic3r/GCode.cpp

@@ -1029,6 +1029,7 @@ GCode::needs_retraction(const Polyline &travel, ExtrusionRole role)
     
     if (role == erSupportMaterial) {
         const SupportLayer* support_layer = dynamic_cast<const SupportLayer*>(this->layer);
+        //FIXME support_layer->support_islands.contains should use some search structure!
         if (support_layer != NULL && support_layer->support_islands.contains(travel)) {
             // skip retraction if this is a travel move inside a support material island
             return false;

xs/src/libslic3r/PrintObject.cpp

@@ -7,6 +7,9 @@
 #include <utility>
 #include <boost/log/trivial.hpp>
 
+#include <tbb/parallel_for.h>
+#include <tbb/atomic.h>
+
 #include <Shiny/Shiny.h>
 
 #ifdef SLIC3R_DEBUG_SLICE_PROCESSING
@@ -632,7 +635,7 @@ PrintObject::discover_vertical_shells()
                         LayerRegion &neighbor_region = *neighbor_layer.get_region(int(idx_region));
                         Polygons newholes;
                         for (size_t idx_region = 0; idx_region < this->_print->regions.size(); ++ idx_region)
-                            polygons_append(newholes, to_polygons(neighbor_layer.get_region(idx_region)->fill_expolygons));
+                            polygons_append(newholes, to_polygons(neighbor_layer.regions[idx_region]->fill_expolygons));
                         if (hole_first) {
                             hole_first = false;
                             polygons_append(holes, STDMOVE(newholes));
@@ -1267,13 +1270,16 @@ PrintObject::_make_perimeters()
         }
     }
     
-    BOOST_LOG_TRIVIAL(debug) << "Generating perimeters in parallel";
-    parallelize<Layer*>(
-        std::queue<Layer*>(std::deque<Layer*>(this->layers.begin(), this->layers.end())),  // cast LayerPtrs to std::queue<Layer*>
-        boost::bind(&Slic3r::Layer::make_perimeters, _1),
-        this->_print->config.threads.value
+    BOOST_LOG_TRIVIAL(debug) << "Generating perimeters in parallel - start";
+    tbb::parallel_for(
+        tbb::blocked_range<size_t>(0, this->layers.size()),
+        [this](const tbb::blocked_range<size_t>& range) {
+            for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++ layer_idx)
+                this->layers[layer_idx]->make_perimeters();
+        }
     );
-    
+    BOOST_LOG_TRIVIAL(debug) << "Generating perimeters in parallel - end";
+
     /*
         simplify slices (both layer and region slices),
         we only need the max resolution for perimeters
@@ -1290,13 +1296,16 @@ PrintObject::_infill()
     if (this->state.is_done(posInfill)) return;
     this->state.set_started(posInfill);
     
-    BOOST_LOG_TRIVIAL(debug) << "Filling layers in parallel";
-    parallelize<Layer*>(
-        std::queue<Layer*>(std::deque<Layer*>(this->layers.begin(), this->layers.end())),  // cast LayerPtrs to std::queue<Layer*>
-        boost::bind(&Slic3r::Layer::make_fills, _1),
-        this->_print->config.threads.value
+    BOOST_LOG_TRIVIAL(debug) << "Filling layers in parallel - start";
+    tbb::parallel_for(
+        tbb::blocked_range<size_t>(0, this->layers.size()),
+        [this](const tbb::blocked_range<size_t>& range) {
+            for (size_t layer_idx = range.begin(); layer_idx < range.end(); ++ layer_idx)
+                this->layers[layer_idx]->make_fills();
+        }
     );
-    
+    BOOST_LOG_TRIVIAL(debug) << "Filling layers in parallel - end";
+
     /*  we could free memory now, but this would make this step not idempotent
     ### $_->fill_surfaces->clear for map @{$_->regions}, @{$object->layers};
     */

xs/src/libslic3r/TriangleMesh.cpp

@@ -13,6 +13,8 @@
 
 #include <boost/log/trivial.hpp>
 
+#include <tbb/parallel_for.h>
+
 #if 0
     #define DEBUG
     #define _DEBUG
@@ -672,10 +674,12 @@ TriangleMeshSlicer::slice(const std::vector<float> &z, std::vector<Polygons>* la
     std::vector<IntersectionLines> lines(z.size());
     {
         boost::mutex lines_mutex;
-        parallelize<int>(
-            0,
-            this->mesh->stl.stats.number_of_facets-1,
-            boost::bind(&TriangleMeshSlicer::_slice_do, this, _1, &lines, &lines_mutex, z)
+        tbb::parallel_for(
+            tbb::blocked_range<int>(0,this->mesh->stl.stats.number_of_facets),
+            [&lines, &lines_mutex, &z, this](const tbb::blocked_range<int>& range) {
+                for (int facet_idx = range.begin(); facet_idx < range.end(); ++ facet_idx)
+                    this->_slice_do(facet_idx, &lines, &lines_mutex, z);
+            }
         );
     }
     
@@ -684,10 +688,12 @@ TriangleMeshSlicer::slice(const std::vector<float> &z, std::vector<Polygons>* la
     // build loops
     BOOST_LOG_TRIVIAL(trace) << "TriangleMeshSlicer::_make_loops_do";
     layers->resize(z.size());
-    parallelize<size_t>(
-        0,
-        lines.size()-1,
-        boost::bind(&TriangleMeshSlicer::_make_loops_do, this, _1, &lines, layers)
+    tbb::parallel_for(
+        tbb::blocked_range<size_t>(0, lines.size()),
+        [&lines, &layers, this](const tbb::blocked_range<size_t>& range) {
+            for (size_t line_idx = range.begin(); line_idx < range.end(); ++ line_idx)
+                this->make_loops(lines[line_idx], &(*layers)[line_idx]);
+        }
     );
     BOOST_LOG_TRIVIAL(trace) << "TriangleMeshSlicer::slice finished";
 }
@@ -873,12 +879,6 @@ bool TriangleMeshSlicer::slice_facet(
     return false;
 }
 
-void
-TriangleMeshSlicer::_make_loops_do(size_t i, std::vector<IntersectionLines>* lines, std::vector<Polygons>* layers) const
-{
-    this->make_loops((*lines)[i], &(*layers)[i]);
-}
-
 void TriangleMeshSlicer::make_loops(std::vector<IntersectionLine> &lines, Polygons* loops) const
 {
     // Remove tangent edges.

xs/src/libslic3r/TriangleMesh.hpp

@@ -132,7 +132,6 @@ private:
     std::vector<stl_vertex>  v_scaled_shared;
 
     void _slice_do(size_t facet_idx, std::vector<IntersectionLines>* lines, boost::mutex* lines_mutex, const std::vector<float> &z) const;
-    void _make_loops_do(size_t i, std::vector<IntersectionLines>* lines, std::vector<Polygons>* layers) const;
     void make_loops(std::vector<IntersectionLine> &lines, Polygons* loops) const;
     void make_expolygons(const Polygons &loops, ExPolygons* slices) const;
     void make_expolygons_simple(std::vector<IntersectionLine> &lines, ExPolygons* slices) const;

xs/src/libslic3r/libslic3r.h

@@ -14,7 +14,7 @@
 #include <boost/thread.hpp>
 
 #define SLIC3R_FORK_NAME "Slic3r Prusa Edition"
-#define SLIC3R_VERSION "1.31.6"
+#define SLIC3R_VERSION "1.33.8.devel"
 #define SLIC3R_BUILD "UNKNOWN"
 
 //FIXME This epsilon value is used for many non-related purposes:
@@ -99,53 +99,6 @@ inline void append_to(std::vector<T> &dst, const std::vector<T> &src)
     dst.insert(dst.end(), src.begin(), src.end());
 }
 
-template <class T> void
-_parallelize_do(std::queue<T>* queue, boost::mutex* queue_mutex, boost::function<void(T)> func)
-{
-    //std::cout << "THREAD STARTED: " << boost::this_thread::get_id() << std::endl;
-    while (true) {
-        T i;
-        {
-            boost::lock_guard<boost::mutex> l(*queue_mutex);
-            if (queue->empty()) return;
-            i = queue->front();
-            queue->pop();
-        }
-        //std::cout << "  Thread " << boost::this_thread::get_id() << " processing item " << i << std::endl;
-        func(i);
-        boost::this_thread::interruption_point();
-    }
-}
-
-template <class T> void
-parallelize(std::queue<T> queue, boost::function<void(T)> func,
-    int threads_count = boost::thread::hardware_concurrency())
-{
-#ifdef SLIC3R_PROFILE
-    while (! queue.empty()) {
-        func(queue.front());
-        queue.pop();
-    }
-#else
-    if (threads_count == 0)
-        threads_count = 2;
-    boost::mutex queue_mutex;
-    boost::thread_group workers;
-    for (int i = 0; i < std::min(threads_count, int(queue.size())); ++ i)
-        workers.add_thread(new boost::thread(&_parallelize_do<T>, &queue, &queue_mutex, func));
-    workers.join_all();
-#endif
-}
-
-template <class T> void
-parallelize(T start, T end, boost::function<void(T)> func,
-    int threads_count = boost::thread::hardware_concurrency())
-{
-    std::queue<T> queue;
-    for (T i = start; i <= end; ++i) queue.push(i);
-    parallelize(queue, func, threads_count);
-}
-
 template <typename T>
 void append(std::vector<T>& dest, const std::vector<T>& src)
 {