[Bf-blender-cvs] [03f2855] master: Cycles: Implement QBVH tree traversal
Sergey Sharybin
noreply at git.blender.org
Wed Dec 24 22:56:27 CET 2014
Commit: 03f28553ff07e9d79a92834188017d93de555ccb
Author: Sergey Sharybin
Date: Tue Dec 16 20:11:37 2014 +0500
Branches: master
https://developer.blender.org/rB03f28553ff07e9d79a92834188017d93de555ccb
Cycles: Implement QBVH tree traversal
This commit implements traversal for QBVH tree, which is based on the old loop
code for traversal itself and Embree for node intersection.
This commit also does some changes to the loop inspired by Embree:
- Visibility flags are only checked for primitives.
Doing visibility check for every node cost quite reasonable amount of time
and in most cases those checks are true-positive.
Other idea here would be to do visibility checks for leaf nodes only, but
this would need to be investigated further.
- For minimum hair width we extend all the nodes' bounding boxes.
Again doing curve visibility check is quite costly for each of the nodes and
those checks returns truth for most of the hierarchy anyway.
There are number of possible optimization still, but current state is good
enough in terms it makes rendering faster a little bit after recent watertight
commit.
Currently QBVH is only implemented for CPU with SSE2 support at least. All
other devices would need to be supported later (if that'd make sense from
performance point of view).
The code is enabled for compilation in kernel. but blender wouldn't use it
still.
===================================================================
M intern/cycles/bvh/bvh.cpp
M intern/cycles/kernel/CMakeLists.txt
M intern/cycles/kernel/geom/geom.h
M intern/cycles/kernel/geom/geom_bvh.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
A intern/cycles/kernel/geom/geom_qbvh.h
A intern/cycles/kernel/geom/geom_qbvh_shadow.h
A intern/cycles/kernel/geom/geom_qbvh_subsurface.h
A intern/cycles/kernel/geom/geom_qbvh_traversal.h
A intern/cycles/kernel/geom/geom_qbvh_volume.h
M intern/cycles/kernel/kernel_compat_cpu.h
M intern/cycles/kernel/kernel_types.h
M intern/cycles/render/mesh.cpp
M intern/cycles/render/scene.h
===================================================================
diff --git a/intern/cycles/bvh/bvh.cpp b/intern/cycles/bvh/bvh.cpp
index b591d59..e7141c9 100644
--- a/intern/cycles/bvh/bvh.cpp
+++ b/intern/cycles/bvh/bvh.cpp
@@ -479,8 +479,11 @@ void BVH::pack_instances(size_t nodes_size)
pack_nodes[pack_nodes_offset + nsize_bbox] = data;
- if(use_qbvh)
- pack_nodes[pack_nodes_offset + nsize_bbox+1] = bvh_nodes[i + nsize_bbox+1];
+ if(use_qbvh) {
+ memcpy(&pack_nodes[pack_nodes_offset + nsize_bbox+1],
+ &bvh_nodes[i + nsize_bbox+1],
+ sizeof(int4) * (nsize - (nsize_bbox+1)));
+ }
pack_nodes_offset += nsize;
}
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index ca1065f..f8d2ee6 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -123,6 +123,11 @@ set(SRC_GEOM_HEADERS
geom/geom_motion_triangle.h
geom/geom_object.h
geom/geom_primitive.h
+ geom/geom_qbvh.h
+ geom/geom_qbvh_shadow.h
+ geom/geom_qbvh_subsurface.h
+ geom/geom_qbvh_traversal.h
+ geom/geom_qbvh_volume.h
geom/geom_triangle.h
geom/geom_triangle_intersect.h
geom/geom_volume.h
diff --git a/intern/cycles/kernel/geom/geom.h b/intern/cycles/kernel/geom/geom.h
index 3a768f3..38fd785 100644
--- a/intern/cycles/kernel/geom/geom.h
+++ b/intern/cycles/kernel/geom/geom.h
@@ -21,6 +21,7 @@
/* 64 object BVH + 64 mesh BVH + 64 object node splitting */
#define BVH_STACK_SIZE 192
#define BVH_NODE_SIZE 4
+#define BVH_QNODE_SIZE 7
#define TRI_NODE_SIZE 3
/* silly workaround for float extended precision that happens when compiling
diff --git a/intern/cycles/kernel/geom/geom_bvh.h b/intern/cycles/kernel/geom/geom_bvh.h
index a989267..c0eefcd 100644
--- a/intern/cycles/kernel/geom/geom_bvh.h
+++ b/intern/cycles/kernel/geom/geom_bvh.h
@@ -48,6 +48,11 @@ CCL_NAMESPACE_BEGIN
#define BVH_FEATURE(f) (((BVH_FUNCTION_FEATURES) & (f)) != 0)
+/* Common QBVH functions. */
+#ifdef __QBVH__
+#include "geom_qbvh.h"
+#endif
+
/* Regular BVH traversal */
#define BVH_FUNCTION_NAME bvh_intersect
diff --git a/intern/cycles/kernel/geom/geom_bvh_shadow.h b/intern/cycles/kernel/geom/geom_bvh_shadow.h
index 4bdfc74..d605602 100644
--- a/intern/cycles/kernel/geom/geom_bvh_shadow.h
+++ b/intern/cycles/kernel/geom/geom_bvh_shadow.h
@@ -17,6 +17,10 @@
* limitations under the License.
*/
+#ifdef __QBVH__
+#include "geom_qbvh_shadow.h"
+#endif
+
/* This is a template BVH traversal function, where various features can be
* enabled/disabled. This way we can compile optimized versions for each case
* without new features slowing things down.
@@ -380,11 +384,23 @@ ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg,
const uint max_hits,
uint *num_hits)
{
- return BVH_FUNCTION_FULL_NAME(BVH)(kg,
- ray,
- isect_array,
- max_hits,
- num_hits);
+#ifdef __QBVH__
+ if(kernel_data.bvh.use_qbvh) {
+ return BVH_FUNCTION_FULL_NAME(QBVH)(kg,
+ ray,
+ isect_array,
+ max_hits,
+ num_hits);
+ }
+ else
+#endif
+ {
+ return BVH_FUNCTION_FULL_NAME(BVH)(kg,
+ ray,
+ isect_array,
+ max_hits,
+ num_hits);
+ }
}
#undef BVH_FUNCTION_NAME
diff --git a/intern/cycles/kernel/geom/geom_bvh_subsurface.h b/intern/cycles/kernel/geom/geom_bvh_subsurface.h
index 90cbbc0..ff46214 100644
--- a/intern/cycles/kernel/geom/geom_bvh_subsurface.h
+++ b/intern/cycles/kernel/geom/geom_bvh_subsurface.h
@@ -17,6 +17,10 @@
* limitations under the License.
*/
+#ifdef __QBVH__
+#include "geom_qbvh_subsurface.h"
+#endif
+
/* This is a template BVH traversal function for subsurface scattering, where
* various features can be enabled/disabled. This way we can compile optimized
* versions for each case without new features slowing things down.
@@ -300,12 +304,25 @@ ccl_device_inline uint BVH_FUNCTION_NAME(KernelGlobals *kg,
uint *lcg_state,
int max_hits)
{
- return BVH_FUNCTION_FULL_NAME(BVH)(kg,
- ray,
- isect_array,
- subsurface_object,
- lcg_state,
- max_hits);
+#ifdef __QBVH__
+ if(kernel_data.bvh.use_qbvh) {
+ return BVH_FUNCTION_FULL_NAME(QBVH)(kg,
+ ray,
+ isect_array,
+ subsurface_object,
+ lcg_state,
+ max_hits);
+ }
+ else
+#endif
+ {
+ return BVH_FUNCTION_FULL_NAME(BVH)(kg,
+ ray,
+ isect_array,
+ subsurface_object,
+ lcg_state,
+ max_hits);
+ }
}
#undef BVH_FUNCTION_NAME
diff --git a/intern/cycles/kernel/geom/geom_bvh_traversal.h b/intern/cycles/kernel/geom/geom_bvh_traversal.h
index d48eda5..6e5b6ea 100644
--- a/intern/cycles/kernel/geom/geom_bvh_traversal.h
+++ b/intern/cycles/kernel/geom/geom_bvh_traversal.h
@@ -17,6 +17,10 @@
* limitations under the License.
*/
+#ifdef __QBVH__
+#include "geom_qbvh_traversal.h"
+#endif
+
/* This is a template BVH traversal function, where various features can be
* enabled/disabled. This way we can compile optimized versions for each case
* without new features slowing things down.
@@ -381,16 +385,33 @@ ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg,
#endif
)
{
- return BVH_FUNCTION_FULL_NAME(BVH)(kg,
- ray,
- isect,
- visibility
+#ifdef __QBVH__
+ if(kernel_data.bvh.use_qbvh) {
+ return BVH_FUNCTION_FULL_NAME(QBVH)(kg,
+ ray,
+ isect,
+ visibility
+#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
+ , lcg_state,
+ difl,
+ extmax
+#endif
+ );
+ }
+ else
+#endif
+ {
+ return BVH_FUNCTION_FULL_NAME(BVH)(kg,
+ ray,
+ isect,
+ visibility
#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
- , lcg_state,
- difl,
- extmax
+ , lcg_state,
+ difl,
+ extmax
#endif
- );
+ );
+ }
}
#undef BVH_FUNCTION_NAME
diff --git a/intern/cycles/kernel/geom/geom_bvh_volume.h b/intern/cycles/kernel/geom/geom_bvh_volume.h
index bae90c2..8a25b5d 100644
--- a/intern/cycles/kernel/geom/geom_bvh_volume.h
+++ b/intern/cycles/kernel/geom/geom_bvh_volume.h
@@ -17,6 +17,10 @@
* limitations under the License.
*/
+#ifdef __QBVH__
+#include "geom_qbvh_volume.h"
+#endif
+
/* This is a template BVH traversal function for volumes, where
* various features can be enabled/disabled. This way we can compile optimized
* versions for each case without new features slowing things down.
@@ -314,9 +318,19 @@ ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg,
const Ray *ray,
Intersection *isect)
{
- return BVH_FUNCTION_FULL_NAME(BVH)(kg,
- ray,
- isect);
+#ifdef __QBVH__
+ if(kernel_data.bvh.use_qbvh) {
+ return BVH_FUNCTION_FULL_NAME(QBVH)(kg,
+ ray,
+ isect);
+ }
+ else
+#endif
+ {
+ return BVH_FUNCTION_FULL_NAME(BVH)(kg,
+ ray,
+ isect);
+ }
}
#undef BVH_FUNCTION_NAME
diff --git a/intern/cycles/kernel/geom/geom_qbvh.h b/intern/cycles/kernel/geom/geom_qbvh.h
new file mode 100644
index 0000000..a1dd89c
--- /dev/null
+++ b/intern/cycles/kernel/geom/geom_qbvh.h
@@ -0,0 +1,138 @@
+/*
+ * Copyright 2011-2014, 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.
+ */
+
+ccl_device_inline void qbvh_stack_sort(int *__restrict s1,
+ int *__restrict s2,
+ int *__restrict s3,
+ float *__restrict d1,
+ float *__restrict d2,
+ float *__restrict d3)
+{
+ if(*d2 < *d1) { util_swap(s2, s1); util_swap(d2, d1); }
+ if(*d3 < *d2) { util_swap(s3, s2); util_swap(d3, d2); }
+ if(*d2 < *d1) { util_swap(s2, s1); util_swap(d2, d1); }
+}
+
+ccl_device_inline void qbvh_stack_sort(int *__restrict s1,
+ int *__restrict s2,
+ int *__restrict s3,
+
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list