Manifoldmath: more performant implementation of project_parallel

spirit-code · Apr 20, 2022 · aca472f · aca472f
1 parent d07ce03
commit aca472f
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 9 deletions.
diff --git a/core/src/engine/Manifoldmath.cpp b/core/src/engine/Manifoldmath.cpp
@@ -22,12 +22,12 @@ namespace Manifoldmath
 {
 void project_parallel( vectorfield & vf1, const vectorfield & vf2 )
 {
-    vectorfield vf3 = vf1;
-    project_orthogonal( vf3, vf2 );
-// TODO: replace the loop with Vectormath Kernel
-#pragma omp parallel for
-    for( unsigned int i = 0; i < vf1.size(); ++i )
-        vf1[i] -= vf3[i];
+    scalar proj = Vectormath::dot(vf1, vf2);
+    Backend::par::apply( vf1.size(), [vf1 = vf1.data(), vf2 = vf2.data(), proj] SPIRIT_LAMBDA (int idx)
+        {
+            vf1[idx] = proj * vf2[idx];
+        } 
+    );
 }
 
 void project_orthogonal( vectorfield & vf1, const vectorfield & vf2 )

diff --git a/core/src/engine/Manifoldmath.cu b/core/src/engine/Manifoldmath.cu
@@ -2,6 +2,7 @@
 
 #include <engine/Vectormath.hpp>
 #include <engine/Manifoldmath.hpp>
+#include <engine/Backend_par.hpp>
 #include <utility/Constants.hpp>
 
 #include <Eigen/Dense>
@@ -18,9 +19,12 @@ namespace Engine
     {
         void project_parallel(vectorfield & vf1, const vectorfield & vf2)
         {
-            vectorfield vf3 = vf1;
-            project_orthogonal(vf3, vf2);
-            Vectormath::add_c_a(-1, vf3, vf1);
+            scalar proj = Vectormath::dot(vf1, vf2);
+            Backend::par::apply( vf1.size(), [vf1 = vf1.data(), vf2 = vf2.data(), proj] SPIRIT_LAMBDA (int idx)
+                {
+                    vf1[idx] = proj * vf2[idx];
+                } 
+            );
         }
 
         __global__ void cu_project_orthogonal(Vector3 *vf1, const Vector3 *vf2, scalar proj, size_t N)