speedup, internalize libop

1 files changed, 12 insertions, 7 deletions

diff --git a/model/ranksvmtn.cpp b/model/ranksvmtn.cpp
index 7f292f5..ba2d5a3 100644
--- a/model/ranksvmtn.cpp
+++ b/model/ranksvmtn.cpp
@@ -9,23 +9,26 @@ using namespace Eigen;
 void cal_Dw(RidList &D,const VectorXd &w, VectorXd &Dw)
 {
     int n = D.getSize();
-    #pragma omp parallel for
+    // static chunk size of 1 to interleave the iterations
+    #pragma omp parallel for schedule(static,1)
     for (int i=0;i<n;++i)
-        Dw(i) = D.getVec(i).dot(w);
+        Dw(i) = D.getVecDot(i,w);
 }
 
 void cal_Dtw(RidList &D,const VectorXd &w, VectorXd &Dw)
 {
     int n = D.getSize();
-    #pragma omp parallel
+    int fsize = D.getfSize();
+    #pragma omp parallel shared(D,Dw)
     {
         VectorXd Dw_private = VectorXd::Zero(D.getfSize());
         #pragma omp for nowait
         for (int i=0;i<n;++i)
-            Dw_private = Dw_private + D.getVec(i)*w(i);
+            D.addVecw(i,w(i),Dw_private);
         #pragma omp critical
         {
-            Dw=Dw+Dw_private;
+            for (int i=0;i<fsize;++i)
+                Dw(i) = Dw(i) + Dw_private(i);
         }
     }
 }
@@ -55,8 +58,10 @@ int cal_Hs(RidList &D,const vector<int> &rank,const VectorXd &corr,const VectorX
             else
                 g+=Ds[rank[i+j]];
     }
-    VectorXd tmp = alpha.cwiseProduct(Ds)-gamma;
-    VectorXd res = VectorXd::Zero(D.getSize());
+    VectorXd tmp(n);
+    for (int i=0;i<n;++i)
+        tmp(i) = alpha(i)*Ds(i)-gamma(i);
+    VectorXd res = VectorXd::Zero(D.getfSize());
     cal_Dtw(D,tmp,res);
     Hs = s + C*res;
     return 0;