Merge back a few cloth solver fixes from the render branch:

* Disable openmp for dot product, this gives different results each
  time due to non-commutative floating point add.
* Disable openmp with few vertices, the extra thread overhead only
  slows things down then.
* Replace the hack that would divide stepsPerFrame and then set it
  back, now it simply uses the timescale in the collision function.
  This was incorrect because stepsPerFrame is an int, but we don't
  want this to be rounded.
* Extra out of bounds check for hair velocity smoothing grid.

1 files changed, 34 insertions, 31 deletions

diff --git a/source/blender/blenkernel/intern/implicit.c b/source/blender/blenkernel/intern/implicit.c
index c625fb28840..902965bd2f6 100644
--- a/source/blender/blenkernel/intern/implicit.c
+++ b/source/blender/blenkernel/intern/implicit.c
@@ -37,6 +37,10 @@
 #include "BKE_global.h"
 #include "BKE_utildefines.h"
 
+#include "BLI_threads.h"
+
+#define CLOTH_OPENMP_LIMIT 25
+
 #ifdef _WIN32
 #include <windows.h>
 static LARGE_INTEGER _itstart, _itend;
@@ -230,8 +234,11 @@ DO_INLINE float dot_lfvector(float (*fLongVectorA)[3], float (*fLongVectorB)[3],
 {
 	long i = 0;
 	float temp = 0.0;
+// XXX brecht, disabled this for now (first schedule line was already disabled),
+// 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]);
@@ -577,11 +584,12 @@ DO_INLINE void mul_bfmatrix_S(fmatrix3x3 *matrix, float scalar)
 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
 		{
@@ -962,7 +970,7 @@ DO_INLINE void BuildPPinv(fmatrix3x3 *lA, fmatrix3x3 *P, fmatrix3x3 *Pinv)
 	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
@@ -1460,6 +1468,8 @@ static void hair_velocity_smoothing(ClothModifierData *clmd, lfVector *lF, lfVec
 		i = HAIR_GRID_INDEX(lX[v], gmin, gmax, 0);
 		j = HAIR_GRID_INDEX(lX[v], gmin, gmax, 1);
 		k = HAIR_GRID_INDEX(lX[v], gmin, gmax, 2);
+		if (i < 0 || j < 0 || k < 0 || i > 10 || j >= 10 || k >= 10)
+			continue;
 
 		grid[i][j][k].velocity[0] += lV[v][0];
 		grid[i][j][k].velocity[1] += lV[v][1];
@@ -1523,6 +1533,8 @@ static void hair_velocity_smoothing(ClothModifierData *clmd, lfVector *lF, lfVec
 		i = HAIR_GRID_INDEX(lX[v], gmin, gmax, 0);
 		j = HAIR_GRID_INDEX(lX[v], gmin, gmax, 1);
 		k = HAIR_GRID_INDEX(lX[v], gmin, gmax, 2);
+		if (i < 0 || j < 0 || k < 0 || i > 10 || j >= 10 || k >= 10)
+			continue;
 
 		lF[v][0] += smoothfac * (grid[i][j][k].velocity[0] - lV[v][0]);
 		lF[v][1] += smoothfac * (grid[i][j][k].velocity[1] - lV[v][1]);
@@ -1537,6 +1549,7 @@ static void hair_velocity_smoothing(ClothModifierData *clmd, lfVector *lF, lfVec
 
 	free_collider_cache(&colliders);
 }
+
 static void cloth_calc_force(ClothModifierData *clmd, float frame, lfVector *lF, lfVector *lX, lfVector *lV, fmatrix3x3 *dFdV, fmatrix3x3 *dFdX, ListBase *effectors, float time, fmatrix3x3 *M)
 {
 	/* Collect forces and derivatives:  F,dFdX,dFdV */
@@ -1731,9 +1744,10 @@ int implicit_solver (Object *ob, float frame, ClothModifierData *clmd, ListBase
 	ClothVertex *verts = cloth->verts;
 	unsigned int numverts = cloth->numverts;
 	float dt = clmd->sim_parms->timescale / clmd->sim_parms->stepsPerFrame;
+	float spf = (float)clmd->sim_parms->stepsPerFrame / clmd->sim_parms->timescale;
 	Implicit_Data *id = cloth->implicit;
-	int result = 0;
-	
+	int do_extra_solve;
+
 	if(clmd->sim_parms->flags & CLOTH_SIMSETTINGS_FLAG_GOAL) /* do goal stuff */
 	{
 		for(i = 0; i < numverts; i++)
@@ -1778,60 +1792,50 @@ int implicit_solver (Object *ob, float frame, ClothModifierData *clmd, ListBase
 
 		if(clmd->coll_parms->flags & CLOTH_COLLSETTINGS_FLAG_ENABLED && clmd->clothObject->bvhtree)
 		{
-			float temp = clmd->sim_parms->stepsPerFrame;
-			/* not too nice hack, but collisions need this correction -jahka */
-			clmd->sim_parms->stepsPerFrame /= clmd->sim_parms->timescale;
-
 			// collisions 
 			// itstart();
 			
 			// update verts to current positions
 			for(i = 0; i < numverts; i++)
-			{	
+			{
 				VECCOPY(verts[i].tx, id->Xnew[i]);
-				
+
 				VECSUB(verts[i].tv, verts[i].tx, verts[i].txold);
 				VECCOPY(verts[i].v, verts[i].tv);
 			}
-			
+
 			// call collision function
 			// TODO: check if "step" or "step+dt" is correct - dg
-			result = cloth_bvh_objcollision(ob, clmd, step/clmd->sim_parms->timescale, dt/clmd->sim_parms->timescale);
-			
-			// correct velocity again, just to be sure we had to change it due to adaptive collisions
-			for(i = 0; i < numverts; i++)
-			{
-				VECSUB(verts[i].tv, verts[i].tx, id->X[i]);
-			}
+			do_extra_solve = cloth_bvh_objcollision(ob, clmd, step/clmd->sim_parms->timescale, dt/clmd->sim_parms->timescale);
 			
 			// copy corrected positions back to simulation
 			for(i = 0; i < numverts; i++)
 			{		
-				if(result)
+				// correct velocity again, just to be sure we had to change it due to adaptive collisions
+				VECSUB(verts[i].tv, verts[i].tx, id->X[i]);
+
+				if(do_extra_solve)
 				{
 					
 					if((clmd->sim_parms->flags & CLOTH_SIMSETTINGS_FLAG_GOAL) && (verts [i].flags & CLOTH_VERT_FLAG_PINNED))
 						continue;
-					
+
 					VECCOPY(id->Xnew[i], verts[i].tx);
 					VECCOPY(id->Vnew[i], verts[i].tv);
-					mul_v3_fl(id->Vnew[i], clmd->sim_parms->stepsPerFrame);
+					mul_v3_fl(id->Vnew[i], spf);
 				}
 			}
 			
-			/* restore original stepsPerFrame */
-			clmd->sim_parms->stepsPerFrame = temp;
-			
 			// X = Xnew;
 			cp_lfvector(id->X, id->Xnew, numverts);
-			
+
 			// if there were collisions, advance the velocity from v_n+1/2 to v_n+1
 			
-			if(result)
+			if(do_extra_solve)
 			{
 				// V = Vnew;
 				cp_lfvector(id->V, id->Vnew, numverts);
-				
+
 				// calculate 
 				cloth_calc_force(clmd, frame, id->F, id->X, id->V, id->dFdV, id->dFdX, effectors, step+dt, id->M);	
 				
@@ -1851,7 +1855,6 @@ int implicit_solver (Object *ob, float frame, ClothModifierData *clmd, ListBase
 		cp_lfvector(id->V, id->Vnew, numverts);
 		
 		step += dt;
-		
 	}
 
 	for(i = 0; i < numverts; i++)