Parallelize other cost functions

src/mpl2/CMakeLists.txt

@@ -33,9 +33,10 @@
 
 include("openroad")
 
-
+set(THREADS_PREFER_PTHREAD_FLAG ON)
 
 find_package(ortools REQUIRED)
+find_package(OpenMP REQUIRED)
 
 add_library(mpl2_lib
   src/rtl_mp.cpp
@@ -60,6 +61,8 @@ target_link_libraries(mpl2_lib
     ortools::ortools
     dl
     par_lib
+    Threads::Threads
+    OpenMP::OpenMP_CXX
 )
 
 swig_lib(NAME      mpl2

src/mpl2/src/SACoreSoftMacro.cpp

@@ -379,13 +379,15 @@ void SACoreSoftMacro::calBoundaryPenalty()
   }
 
   int tot_num_macros = 0;
+  #pragma omp parallel for reduction(+:tot_num_macros)
   for (const auto& macro : macros_) {
     tot_num_macros += macro.getNumMacro();
   }
   if (tot_num_macros <= 0) {
     return;
   }
 
+  #pragma omp parallel for reduction(+:boundary_penalty_)
   for (const auto& macro : macros_) {
     if (macro.getNumMacro() > 0) {
       const float lx = macro.getX();
@@ -419,6 +421,7 @@ void SACoreSoftMacro::calMacroBlockagePenalty()
     return;
   }
 
+  #pragma omp parallel for reduction(+:macro_blockage_penalty_)
   for (auto& bbox : blockages_) {
     for (const auto& macro : macros_) {
       if (macro.getNumMacro() > 0) {
@@ -440,8 +443,8 @@ void SACoreSoftMacro::calMacroBlockagePenalty()
             = std::abs((region_uy + region_ly) / 2.0 - (uy + ly) / 2.0);
         x_dist = std::max(width - x_dist, 0.0f) / width;
         y_dist = std::max(height - y_dist, 0.0f) / height;
-        macro_blockage_penalty_
-            += (x_dist * x_dist + y_dist * y_dist) * macro.getNumMacro();
+        macro_blockage_penalty_ +=
+            (x_dist * x_dist + y_dist * y_dist) * macro.getNumMacro();
       }
     }
   }

src/mpl2/src/SimulatedAnnealingCore.cpp

@@ -244,6 +244,7 @@ void SimulatedAnnealingCore<T>::calWirelength()
 
   // calculate the total net weight
   float tot_net_weight = 0.0;
+  #pragma omp parallel for reduction(+:tot_net_weight)
   for (const auto& net : nets_) {
     tot_net_weight += net.weight;
   }
@@ -252,6 +253,7 @@ void SimulatedAnnealingCore<T>::calWirelength()
     return;
   }
 
+  #pragma omp parallel for reduction(+:wirelength_)
   for (const auto& net : nets_) {
     const float x1 = macros_[net.terminals.first].getPinX();
     const float y1 = macros_[net.terminals.first].getPinY();
@@ -304,7 +306,7 @@ void SimulatedAnnealingCore<T>::calFencePenalty()
     float height = y_dist <= max_y_dist ? 0.0 : (y_dist - max_y_dist);
     width = width / outline_width_;
     height = height / outline_height_;
-    fence_penalty_ += width * width + height * height;
+    fence_penalty_ = fence_penalty_ + width * width + height * height;
   }
   // normalization
   fence_penalty_ = fence_penalty_ / fences_.size();
@@ -338,7 +340,7 @@ void SimulatedAnnealingCore<T>::calGuidancePenalty()
                             - (bbox.yMax() + bbox.yMin()) / 2.0);
     x_dist = std::max(x_dist - width, 0.0f) / width;
     y_dist = std::max(y_dist - height, 0.0f) / height;
-    guidance_penalty_ += x_dist * x_dist + y_dist * y_dist;
+    guidance_penalty_ = guidance_penalty_ + x_dist * x_dist + y_dist * y_dist;
   }
   guidance_penalty_ = guidance_penalty_ / guides_.size();
   if (graphics_) {