Merged Soc 2009 - raytrace optimization [0]

from branch [1] at rev 23647

[0] - http://wiki.blender.org/index.php/User:Jaguarandi/SummerOfCode2009/
[1] - https://svn.blender.org/svnroot/bf-blender/branches/soc-2009-jaguarandi

1 files changed, 249 insertions, 0 deletions

diff --git a/source/blender/render/intern/raytrace/svbvh.h b/source/blender/render/intern/raytrace/svbvh.h
new file mode 100644
index 00000000000..3c733b6b402
--- /dev/null
+++ b/source/blender/render/intern/raytrace/svbvh.h
@@ -0,0 +1,249 @@
+/**
+ * $Id$
+ *
+ * ***** BEGIN GPL LICENSE BLOCK *****
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version. 
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
+ *
+ * The Original Code is Copyright (C) 2009 Blender Foundation.
+ * All rights reserved.
+ *
+ * The Original Code is: all of this file.
+ *
+ * Contributor(s): André Pinto.
+ *
+ * ***** END GPL LICENSE BLOCK *****
+ */
+#ifdef __SSE__
+ 
+#ifndef RE_RAYTRACE_SVBVH_H
+#define RE_RAYTRACE_SVBVH_H
+
+#include "bvh.h"
+#include "BLI_memarena.h"
+#include <stdio.h>
+#include <algorithm>
+
+struct SVBVHNode
+{
+	int nchilds;
+
+	//Array of bb, array of childs
+	float *child_bb;
+	SVBVHNode **child;
+};
+
+template<>
+inline int bvh_node_hit_test<SVBVHNode>(SVBVHNode *node, Isect *isec)
+{
+	return 1;
+}
+
+template<>
+inline void bvh_node_push_childs<SVBVHNode>(SVBVHNode *node, Isect *isec, SVBVHNode **stack, int &stack_pos)
+{
+	int i=0;
+	while(i+4 <= node->nchilds)
+	{
+		int res = test_bb_group4( (__m128*) (node->child_bb+6*i), isec );
+		RE_RC_COUNT(isec->raycounter->simd_bb.test);
+		
+		if(res & 1) { stack[stack_pos++] = node->child[i+0]; RE_RC_COUNT(isec->raycounter->simd_bb.hit); }
+		if(res & 2) { stack[stack_pos++] = node->child[i+1]; RE_RC_COUNT(isec->raycounter->simd_bb.hit); }
+		if(res & 4) { stack[stack_pos++] = node->child[i+2]; RE_RC_COUNT(isec->raycounter->simd_bb.hit); }
+		if(res & 8) { stack[stack_pos++] = node->child[i+3]; RE_RC_COUNT(isec->raycounter->simd_bb.hit); }
+		
+		i += 4;
+	}
+	while(i < node->nchilds)
+	{
+		if(RE_rayobject_bb_intersect_test(isec, (const float*)node->child_bb+6*i))
+			stack[stack_pos++] = node->child[i];
+		i++;
+	}
+}
+
+template<>
+void bvh_node_merge_bb<SVBVHNode>(SVBVHNode *node, float *min, float *max)
+{
+	if(is_leaf(node))
+	{
+		RE_rayobject_merge_bb( (RayObject*)node, min, max);
+	}
+	else
+	{
+		int i=0;
+		while(i+4 <= node->nchilds)
+		{
+			float *res = node->child_bb + 6*i;
+			for(int j=0; j<3; j++)
+			{
+				min[j] = MIN2(min[j], res[4*j+0]);
+				min[j] = MIN2(min[j], res[4*j+1]);
+				min[j] = MIN2(min[j], res[4*j+2]);
+				min[j] = MIN2(min[j], res[4*j+3]);
+			}
+			for(int j=0; j<3; j++)
+			{
+				max[j] = MAX2(max[j], res[4*(j+3)+0]);
+				max[j] = MAX2(max[j], res[4*(j+3)+1]);
+				max[j] = MAX2(max[j], res[4*(j+3)+2]);
+				max[j] = MAX2(max[j], res[4*(j+3)+3]);
+			}
+			
+			i += 4;
+		}
+
+		for(; i<node->nchilds; i++)
+		{
+			DO_MIN(node->child_bb+6*i  , min);
+			DO_MAX(node->child_bb+3+6*i, max);
+		}
+	}
+}
+
+
+
+/*
+ * Builds a SVBVH tree form a VBVHTree
+ */
+template<class OldNode>
+struct Reorganize_SVBVH
+{
+	MemArena *arena;
+
+	float childs_per_node;
+	int nodes_with_childs[16];
+	int useless_bb;
+	int nodes;
+
+	Reorganize_SVBVH(MemArena *a)
+	{
+		arena = a;
+		nodes = 0;
+		childs_per_node = 0;
+		useless_bb = 0;
+		
+		for(int i=0; i<16; i++)
+			nodes_with_childs[i] = 0;
+	}
+	
+	~Reorganize_SVBVH()
+	{
+		printf("%f childs per node\n", childs_per_node / nodes);
+		printf("%d childs BB are useless\n", useless_bb);
+		for(int i=0; i<16; i++)
+			printf("%i childs per node: %d/%d = %f\n", i, nodes_with_childs[i], nodes,  nodes_with_childs[i]/float(nodes));
+	}
+	
+	SVBVHNode *create_node(int nchilds)
+	{
+		SVBVHNode *node = (SVBVHNode*)BLI_memarena_alloc(arena, sizeof(SVBVHNode));
+		node->nchilds = nchilds;
+		node->child_bb   = (float*)BLI_memarena_alloc(arena, sizeof(float)*6*nchilds);
+		node->child= (SVBVHNode**)BLI_memarena_alloc(arena, sizeof(SVBVHNode*)*nchilds);
+
+		return node;
+	}
+	
+	void copy_bb(float *bb, const float *old_bb)
+	{
+		std::copy( old_bb, old_bb+6, bb );
+	}
+	
+	void prepare_for_simd(SVBVHNode *node)
+	{
+		int i=0;
+		while(i+4 <= node->nchilds)
+		{
+			float vec_tmp[4*6];
+			float *res = node->child_bb+6*i;
+			std::copy( res, res+6*4, vec_tmp);
+			
+			for(int j=0; j<6; j++)
+			{
+				res[4*j+0] = vec_tmp[6*0+j];
+				res[4*j+1] = vec_tmp[6*1+j];
+				res[4*j+2] = vec_tmp[6*2+j];
+				res[4*j+3] = vec_tmp[6*3+j];
+			}
+
+			i += 4;
+		}
+	}
+
+	/* amt must be power of two */
+	inline int padup(int num, int amt)
+	{
+		return ((num+(amt-1))&~(amt-1));
+	}
+	
+	SVBVHNode *transform(OldNode *old)
+	{
+		if(is_leaf(old))
+			return (SVBVHNode*)old;
+		if(is_leaf(old->child))
+			return (SVBVHNode*)old->child;
+
+		int nchilds = count_childs(old);
+		int alloc_childs = nchilds;
+		if(nchilds % 4 > 2)
+			alloc_childs = padup(nchilds, 4);
+		
+		SVBVHNode *node = create_node(alloc_childs);
+
+		childs_per_node += nchilds;
+		nodes++;
+		if(nchilds < 16)
+			nodes_with_childs[nchilds]++;
+		
+		useless_bb += alloc_childs-nchilds;
+		while(alloc_childs > nchilds)
+		{
+			const static float def_bb[6] = { FLT_MAX, FLT_MAX, FLT_MAX, FLT_MIN, FLT_MIN, FLT_MIN };
+			alloc_childs--;
+			node->child[alloc_childs] = 0;
+			copy_bb(node->child_bb+alloc_childs*6, def_bb);
+		}
+		
+		int i=nchilds;
+		for(OldNode *o_child = old->child; o_child; o_child = o_child->sibling)
+		{
+			i--;
+			node->child[i] = transform(o_child);
+			if(is_leaf(o_child))
+			{
+				float bb[6];
+				INIT_MINMAX(bb, bb+3);
+				RE_rayobject_merge_bb( (RayObject*)o_child, bb, bb+3);
+				copy_bb(node->child_bb+i*6, bb);
+				break;
+			}
+			else
+			{
+				copy_bb(node->child_bb+i*6, o_child->bb);
+			}
+		}
+		assert( i == 0 );
+
+		prepare_for_simd(node);
+		
+		return node;
+	}	
+};
+
+#endif
+
+#endif //__SSE__
\ No newline at end of file