Commit: a08e2179f17569abed814f734dadfebf591e7024
Author: Sergey Sharybin
Date: Thu Jul 7 12:23:13 2016 +0200
Branches: master
https://developer.blender.org/rBa08e2179f17569abed814f734dadfebf591e7024
Cycles: Implement unaligned nodes BVH traversal
This commit implements traversal of unaligned BVH nodes.
QBVH traversal is fully SIMD optimized and calculates orientation
for all 4 children at a time, regular BVH might probably be optimized
a bit more.
===================================================================
M intern/cycles/kernel/CMakeLists.txt
M intern/cycles/kernel/geom/geom_bvh.h
A intern/cycles/kernel/geom/geom_bvh_nodes.h
M intern/cycles/kernel/geom/geom_bvh_shadow.h
M intern/cycles/kernel/geom/geom_bvh_subsurface.h
M intern/cycles/kernel/geom/geom_bvh_traversal.h
M intern/cycles/kernel/geom/geom_bvh_volume.h
M intern/cycles/kernel/geom/geom_bvh_volume_all.h
M intern/cycles/kernel/geom/geom_qbvh.h
M intern/cycles/kernel/geom/geom_qbvh_shadow.h
M intern/cycles/kernel/geom/geom_qbvh_subsurface.h
M intern/cycles/kernel/geom/geom_qbvh_traversal.h
M intern/cycles/kernel/geom/geom_qbvh_volume.h
M intern/cycles/kernel/geom/geom_qbvh_volume_all.h
===================================================================
diff --git a/intern/cycles/kernel/CMakeLists.txt
b/intern/cycles/kernel/CMakeLists.txt
index f0adbc0..3c2f774 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -141,6 +141,7 @@ set(SRC_GEOM_HEADERS
geom/geom.h
geom/geom_attribute.h
geom/geom_bvh.h
+ geom/geom_bvh_nodes.h
geom/geom_bvh_shadow.h
geom/geom_bvh_subsurface.h
geom/geom_bvh_traversal.h
diff --git a/intern/cycles/kernel/geom/geom_bvh.h
b/intern/cycles/kernel/geom/geom_bvh.h
index d0eedd3..f8d563f 100644
--- a/intern/cycles/kernel/geom/geom_bvh.h
+++ b/intern/cycles/kernel/geom/geom_bvh.h
@@ -77,6 +77,8 @@ CCL_NAMESPACE_BEGIN
/* Regular BVH traversal */
+#include "geom_bvh_nodes.h"
+
#define BVH_FUNCTION_NAME bvh_intersect
#define BVH_FUNCTION_FEATURES 0
#include "geom_bvh_traversal.h"
@@ -109,13 +111,13 @@ CCL_NAMESPACE_BEGIN
#if defined(__SUBSURFACE__)
# define BVH_FUNCTION_NAME bvh_intersect_subsurface
-# define BVH_FUNCTION_FEATURES 0
+# define BVH_FUNCTION_FEATURES BVH_HAIR
# include "geom_bvh_subsurface.h"
#endif
#if defined(__SUBSURFACE__) && defined(__OBJECT_MOTION__)
# define BVH_FUNCTION_NAME bvh_intersect_subsurface_motion
-# define BVH_FUNCTION_FEATURES BVH_MOTION
+# define BVH_FUNCTION_FEATURES BVH_MOTION|BVH_HAIR
# include "geom_bvh_subsurface.h"
#endif
@@ -123,19 +125,19 @@ CCL_NAMESPACE_BEGIN
#if defined(__VOLUME__)
# define BVH_FUNCTION_NAME bvh_intersect_volume
-# define BVH_FUNCTION_FEATURES 0
+# define BVH_FUNCTION_FEATURES BVH_HAIR
# include "geom_bvh_volume.h"
#endif
#if defined(__VOLUME__) && defined(__INSTANCING__)
# define BVH_FUNCTION_NAME bvh_intersect_volume_instancing
-# define BVH_FUNCTION_FEATURES BVH_INSTANCING
+# define BVH_FUNCTION_FEATURES BVH_INSTANCING|BVH_HAIR
# include "geom_bvh_volume.h"
#endif
#if defined(__VOLUME__) && defined(__OBJECT_MOTION__)
# define BVH_FUNCTION_NAME bvh_intersect_volume_motion
-# define BVH_FUNCTION_FEATURES BVH_INSTANCING|BVH_MOTION
+# define BVH_FUNCTION_FEATURES BVH_INSTANCING|BVH_MOTION|BVH_HAIR
# include "geom_bvh_volume.h"
#endif
@@ -175,19 +177,19 @@ CCL_NAMESPACE_BEGIN
#if defined(__VOLUME_RECORD_ALL__)
# define BVH_FUNCTION_NAME bvh_intersect_volume_all
-# define BVH_FUNCTION_FEATURES 0
+# define BVH_FUNCTION_FEATURES BVH_HAIR
# include "geom_bvh_volume_all.h"
#endif
#if defined(__VOLUME_RECORD_ALL__) && defined(__INSTANCING__)
# define BVH_FUNCTION_NAME bvh_intersect_volume_all_instancing
-# define BVH_FUNCTION_FEATURES BVH_INSTANCING
+# define BVH_FUNCTION_FEATURES BVH_INSTANCING|BVH_HAIR
# include "geom_bvh_volume_all.h"
#endif
#if defined(__VOLUME_RECORD_ALL__) && defined(__OBJECT_MOTION__)
# define BVH_FUNCTION_NAME bvh_intersect_volume_all_motion
-# define BVH_FUNCTION_FEATURES BVH_INSTANCING|BVH_MOTION
+# define BVH_FUNCTION_FEATURES BVH_INSTANCING|BVH_MOTION|BVH_HAIR
# include "geom_bvh_volume_all.h"
#endif
diff --git a/intern/cycles/kernel/geom/geom_bvh_nodes.h
b/intern/cycles/kernel/geom/geom_bvh_nodes.h
new file mode 100644
index 0000000..deb91ec
--- /dev/null
+++ b/intern/cycles/kernel/geom/geom_bvh_nodes.h
@@ -0,0 +1,659 @@
+/*
+ * Copyright 2011-2016, Blender Foundation.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+// TODO(sergey): Look into avoid use of full Transform and use 3x3 matrix and
+// 3-vector which might be faster.
+ccl_device_inline Transform bvh_unaligned_node_fetch_space(KernelGlobals *kg,
+ int nodeAddr,
+ int child)
+{
+ Transform space;
+ const int child_addr = nodeAddr + child * 3;
+ space.x = kernel_tex_fetch(__bvh_nodes, child_addr+1);
+ space.y = kernel_tex_fetch(__bvh_nodes, child_addr+2);
+ space.z = kernel_tex_fetch(__bvh_nodes, child_addr+3);
+ space.w = make_float4(0.0f, 0.0f, 0.0f, 1.0f);
+ return space;
+}
+
+#if !defined(__KERNEL_SSE2__)
+ccl_device_inline int bvh_aligned_node_intersect(KernelGlobals *kg,
+ const float3 P,
+ const float3 idir,
+ const float t,
+ const int nodeAddr,
+ const uint visibility,
+ float *dist)
+{
+
+ /* fetch node data */
+ float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
+ float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr+1);
+ float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr+2);
+ float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr+3);
+
+ /* intersect ray against child nodes */
+ float c0lox = (node0.x - P.x) * idir.x;
+ float c0hix = (node0.z - P.x) * idir.x;
+ float c0loy = (node1.x - P.y) * idir.y;
+ float c0hiy = (node1.z - P.y) * idir.y;
+ float c0loz = (node2.x - P.z) * idir.z;
+ float c0hiz = (node2.z - P.z) * idir.z;
+ float c0min = max4(min(c0lox, c0hix), min(c0loy, c0hiy), min(c0loz,
c0hiz), 0.0f);
+ float c0max = min4(max(c0lox, c0hix), max(c0loy, c0hiy), max(c0loz,
c0hiz), t);
+
+ float c1lox = (node0.y - P.x) * idir.x;
+ float c1hix = (node0.w - P.x) * idir.x;
+ float c1loy = (node1.y - P.y) * idir.y;
+ float c1hiy = (node1.w - P.y) * idir.y;
+ float c1loz = (node2.y - P.z) * idir.z;
+ float c1hiz = (node2.w - P.z) * idir.z;
+ float c1min = max4(min(c1lox, c1hix), min(c1loy, c1hiy), min(c1loz,
c1hiz), 0.0f);
+ float c1max = min4(max(c1lox, c1hix), max(c1loy, c1hiy), max(c1loz,
c1hiz), t);
+
+ dist[0] = c0min;
+ dist[1] = c1min;
+
+#ifdef __VISIBILITY_FLAG__
+ /* this visibility test gives a 5% performance hit, how to solve? */
+ return (((c0max >= c0min) && (__float_as_uint(cnodes.x) & visibility))?
1: 0) |
+ (((c1max >= c1min) && (__float_as_uint(cnodes.y) & visibility))?
2: 0);
+#else
+ return ((c0max >= c0min)? 1: 0) |
+ ((c1max >= c1min)? 2: 0);
+#endif
+}
+
+ccl_device_inline int bvh_aligned_node_intersect_robust(KernelGlobals *kg,
+ const float3 P,
+ const float3 idir,
+ const float t,
+ const float difl,
+ const float extmax,
+ const int nodeAddr,
+ const uint visibility,
+ float *dist)
+{
+
+ /* fetch node data */
+ float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
+ float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr+1);
+ float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr+2);
+ float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr+3);
+
+ /* intersect ray against child nodes */
+ float c0lox = (node0.x - P.x) * idir.x;
+ float c0hix = (node0.z - P.x) * idir.x;
+ float c0loy = (node1.x - P.y) * idir.y;
+ float c0hiy = (node1.z - P.y) * idir.y;
+ float c0loz = (node2.x - P.z) * idir.z;
+ float c0hiz = (node2.z - P.z) * idir.z;
+ float c0min = max4(min(c0lox, c0hix), min(c0loy, c0hiy), min(c0loz,
c0hiz), 0.0f);
+ float c0max = min4(max(c0lox, c0hix), max(c0loy, c0hiy), max(c0loz,
c0hiz), t);
+
+ float c1lox = (node0.y - P.x) * idir.x;
+ float c1hix = (node0.w - P.x) * idir.x;
+ float c1loy = (node1.y - P.y) * idir.y;
+ float c1hiy = (node1.w - P.y) * idir.y;
+ float c1loz = (node2.y - P.z) * idir.z;
+ float c1hiz = (node2.w - P.z) * idir.z;
+ float c1min = max4(min(c1lox, c1hix), min(c1loy, c1hiy), min(c1loz,
c1hiz), 0.0f);
+ float c1max = min4(max(c1lox, c1hix), max(c1loy, c1hiy), max(c1loz,
c1hiz), t);
+
+ if(difl != 0.0f) {
+ float hdiff = 1.0f + difl;
+ float ldiff = 1.0f - difl;
+ if(__float_as_int(cnodes.z) & PATH_RAY_CURVE) {
+ c0min = max(ldiff * c0min, c0min - extmax);
+ c0max = min(hdiff * c0max, c0max + extmax);
+ }
+ if(__float_as_int(cnodes.w) & PATH_RAY_CURVE) {
+ c1min = max(ldiff * c1min, c1min - extmax);
+ c1max = min(hdiff * c1max, c1max + extmax);
+ }
+ }
+
+ dist[0] = c0min;
+ dist[1] = c1min;
+
+#ifdef __VISIBILITY_FLAG__
+ /* this visibility test gives a 5% performance hit, how to solve? */
+ return (((c0max >= c0min) && (__float_as_uint(cnodes.x) & visibility))?
1: 0) |
+ (((c1max >= c1min) && (__float_as_uint(cnodes.y) & visibility))?
2: 0);
+#else
+ return ((c0max >= c0min)? 1: 0) |
+ ((c1max >= c1min)? 2: 0);
+#endif
+}
+
+ccl_device_inline bool bvh_unaligned_node_intersect_child(
+ KernelGlobals *kg,
+ const float3 P,
+ const float3 dir,
+ const float t,
+ int nodeAddr,
+ int child,
+ float *dist)
+{
+ Transform space = bvh_unaligned_node_fetch_space(kg, nodeAddr, child);
+ float3 aligned_dir = transform_direction(&space, dir);
+ float3 aligned_P = transform_point(&space, P);
+ float3 nrdir = -bvh_inverse_direction(aligned_dir);
+ float3 tLowerXYZ = aligned_P * nrdir;
+ float3 tUpperXYZ = tLowerXYZ - nrdir;
+ const float tNearX = min(tLowerXYZ.x, tUpperXYZ.x);
+ const float tNearY = min(tLowerXYZ.y, tUpperXYZ.y);
+ const float tNearZ = min(tLowerXYZ.z, tUpperXYZ.z);
+ const float tFarX = max(tLowerXYZ.x, tUpperXYZ.x);
+ const float tFarY = max(tLowerXYZ.y, tUpperXYZ.y);
+ const float tFarZ = max(tLowerXYZ.z, tUpperXYZ.z);
+ const float tNear = max4(0.0f, tNearX, tNearY, tNearZ);
+ const float tFar = min4(t, tFarX, tFarY, tFarZ);
+ *dist = tNear;
+ return tNear <= tFar;
+}
+
+ccl_device_inline bool bvh_unaligned_node_intersect_child_robust(
+ KernelGlobals *kg,
+
@@ Diff output truncated at 10240 characters. @@
_______________________________________________
Bf-blender-cvs mailing list
[email protected]
https://lists.blender.org/mailman/listinfo/bf-blender-cvs
