[Bf-blender-cvs] SVN commit: /data/svn/bf-blender [28833] branches/render25/source/blender/ blenkernel/intern/implicit.c: Render Branch: implicit solver now only uses openmp on matrix multiplication

[Brecht Van Lommel]

Revision: 28833
          
http://projects.blender.org/plugins/scmsvn/viewcvs.php?view=rev&root=bf-blender&revision=28833
Author:   blendix
Date:     2010-05-18 17:03:45 +0200 (Tue, 18 May 2010)

Log Message:
-----------
Render Branch: implicit solver now only uses openmp on matrix multiplication
and dot product if num vertices > 25. This may help for cases where there are
few vertices and thread overhead actually slows things down, but mostly it's
to avoid useless threads in threads once I commit the code to solve each hair
individually.

Modified Paths:
--------------
    branches/render25/source/blender/blenkernel/intern/implicit.c

Modified: branches/render25/source/blender/blenkernel/intern/implicit.c
===================================================================
--- branches/render25/source/blender/blenkernel/intern/implicit.c       
2010-05-18 15:02:39 UTC (rev 28832)
+++ branches/render25/source/blender/blenkernel/intern/implicit.c       
2010-05-18 15:03:45 UTC (rev 28833)
@@ -37,6 +37,11 @@
 #include "BKE_global.h"
 #include "BKE_utildefines.h"
 
+#include "BLI_threads.h"
+
+//#define CLOTH_GOAL_ORIGINAL
+#define CLOTH_OPENMP_LIMIT 25
+
 #ifdef _WIN32
 #include <windows.h>
 static LARGE_INTEGER _itstart, _itend;
@@ -234,7 +239,7 @@
 // due to non-commutative nature of floating point ops this makes the sim give
 // different results each time you run it!
 // schedule(guided, 2)
-//#pragma omp parallel for reduction(+: temp)
+//#pragma omp parallel for reduction(+: temp) if(verts > CLOTH_OPENMP_LIMIT)
        for(i = 0; i < (long)verts; i++)
        {
                temp += INPR(fLongVectorA[i], fLongVectorB[i]);
@@ -580,11 +585,12 @@
 DO_INLINE void mul_bfmatrix_lfvector( float (*to)[3], fmatrix3x3 *from, 
lfVector *fLongVector)
 {
        unsigned int i = 0;
-       lfVector *temp = create_lfvector(from[0].vcount);
+       unsigned int vcount = from[0].vcount;
+       lfVector *temp = create_lfvector(vcount);
        
-       zero_lfvector(to, from[0].vcount);
+       zero_lfvector(to, vcount);
 
-#pragma omp parallel sections private(i)
+#pragma omp parallel sections private(i) if(vcount > CLOTH_OPENMP_LIMIT)
        {
 #pragma omp section
                {
@@ -965,7 +971,7 @@
        unsigned int i = 0;
        
        // Take only the diagonal blocks of A
-// #pragma omp parallel for private(i)
+// #pragma omp parallel for private(i) if(lA[0].vcount > CLOTH_OPENMP_LIMIT)
        for(i = 0; i<lA[0].vcount; i++)
        {
                // block diagonalizer
@@ -1592,14 +1598,14 @@
        if(effectors)
        {       
                // 0 = force, 1 = normalized force
-               winvec = create_lfvector(cloth->numverts);
+               winvec = create_lfvector(numverts);
                
                if(!winvec)
                        printf("winvec: out of memory in implicit.c\n");
                
                // precalculate wind forces
-               #pragma omp parallel for private(i)
-               for(i = 0; i < cloth->numverts; i++)
+               #pragma omp parallel for private(i) if(numverts > 
CLOTH_OPENMP_LIMIT)
+               for(i = 0; i < numverts; i++)
                {       
                        pd_point_from_loc(clmd->scene, (float*)lX[i], 
(float*)lV[i], i, &epoint);
                        pdDoEffectors(effectors, NULL, 
clmd->sim_parms->effector_weights, &epoint, winvec[i], NULL);


_______________________________________________
Bf-blender-cvs mailing list
Bf-blender-cvs@blender.org
http://lists.blender.org/mailman/listinfo/bf-blender-cvs