performance improvements through OpenMP

git-svn-id: https://phat.googlecode.com/svn/trunk@88 8e3bb3c2-eed4-f18f-5264-0b6c94e6926d

2 files changed, 7 insertions, 4 deletions

diff --git a/include/phat/boundary_matrix.h b/include/phat/boundary_matrix.h
index 7be14d8..1bdee61 100644
--- a/include/phat/boundary_matrix.h
+++ b/include/phat/boundary_matrix.h
@@ -176,6 +176,7 @@ namespace phat {
             const index nr_of_columns = (index)input_matrix.size();

             this->set_num_cols( nr_of_columns );

             column temp_col;

+            #pragma omp parallel for private( temp_col )

             for( index cur_col = 0; cur_col <  nr_of_columns; cur_col++ ) {

                 this->set_dim( cur_col, (dimension)input_dims[ cur_col ] );

                 

diff --git a/include/phat/helpers/dualize.h b/include/phat/helpers/dualize.h
index e3ae9ed..3ffedf8 100644
--- a/include/phat/helpers/dualize.h
+++ b/include/phat/helpers/dualize.h
@@ -40,13 +40,13 @@ namespace phat {
         for( index cur_col = 0; cur_col < nr_of_columns; cur_col++ ) {

             boundary_matrix.get_col( cur_col, temp_col );

             for( index idx = 0; idx < (index)temp_col.size(); idx++)

-                dual_sizes[ nr_of_columns - 1 - temp_col[ idx ]  ]++;

+                    dual_sizes[ nr_of_columns - 1 - temp_col[ idx ]  ]++;

         }

  

-        for( index cur_col = 0; cur_col < nr_of_columns; cur_col++ ) {

+        #pragma omp parallel for 

+        for( index cur_col = 0; cur_col < nr_of_columns; cur_col++ )

             dual_matrix[cur_col].reserve(dual_sizes[cur_col]);

-        }

-        

+

         for( index cur_col = 0; cur_col < nr_of_columns; cur_col++ ) {

             boundary_matrix.get_col( cur_col, temp_col );

             for( index idx = 0; idx < (index)temp_col.size(); idx++)

@@ -54,9 +54,11 @@ namespace phat {
         }

 

         const dimension max_dim = boundary_matrix.get_max_dim();

+        #pragma omp parallel for

         for( index cur_col = 0; cur_col < nr_of_columns; cur_col++ )

             dual_dims[ nr_of_columns - 1 - cur_col ] = max_dim - boundary_matrix.get_dim( cur_col );

 

+        #pragma omp parallel for

         for( index cur_col = 0; cur_col < nr_of_columns; cur_col++ )

             std::reverse( dual_matrix[ cur_col ].begin(), dual_matrix[ cur_col ].end() );