[Bf-blender-cvs] [1f86434] cycles_bvh: Cycles: Reduce memory usage by de-duplicating triangle storage
Sergey Sharybin
noreply at git.blender.org
Fri Jun 10 16:32:27 CEST 2016
Commit: 1f86434ff1374e45bdf9e90c2b21867a74d2991b
Author: Sergey Sharybin
Date: Fri Jun 10 16:13:50 2016 +0200
Branches: cycles_bvh
https://developer.blender.org/rB1f86434ff1374e45bdf9e90c2b21867a74d2991b
Cycles: Reduce memory usage by de-duplicating triangle storage
There are several internal changes for this:
First idea is to make __tri_verts to behave similar to __tri_storage,
meaning, __tri_verts array now contains all vertices of all triangles
instead of just mesh vertices. This saves some lookup when reading
triangle coordinates in functions like triangle_smooth_normal().
In order to make it efficient needed to store global triangle offset
somewhere. So no __tri_vindex.w contains a global triangle index which
can be used to read triangle vertices.
Next idea was to use this __tri_verts instead of __tri_storage in
intersection code. Unfortunately, this is quite tricky to do without
noticeable speed loss. Current formulation of indices required doing
two level lookup, which destroys cache and leads to poor performance,
causing up to 8% slowdown here. In order to solve this couple of things
were made.
Firstly, added an array to look up global triangle offset directly from
primitive index. This doesn't affect on overall number of textures
because we're getting rid of _-tri_storage anyway.
Secondly, did some types changes to avoid casts which are not really
coming for free.
After doing those tricks it seems slowdown is withing 2% here on my
desktop, but today it behaves flackey and doesn't give totally
consistent results, so need to doublecheck that).
On a positive site we've achived:
- Few percent of memory save with triangle-only scenes. Actual save
in this case is close to size of all vertices.
On a more fine-subdivided scenes this benefit might become more
obvious.
- Huge memory save of hairy scenes. For example, on koro.blend
there is about 20% memory save. On bunny.blend it's hard to measure
against current master because beefy machine is occupied atm, but
let's put it this way: before this change i was unable to render
full bunny on my machine (and it is on 12gig machine, but remember
beefy machine was reporting 9 gig peak) and now the scene renders just
fine and using 3.3gig.
===================================================================
M intern/cycles/bvh/bvh.cpp
M intern/cycles/bvh/bvh.h
M intern/cycles/kernel/geom/geom_motion_triangle.h
M intern/cycles/kernel/geom/geom_triangle.h
M intern/cycles/kernel/geom/geom_triangle_intersect.h
M intern/cycles/kernel/kernel_textures.h
M intern/cycles/render/mesh.cpp
M intern/cycles/render/mesh.h
M intern/cycles/render/scene.h
===================================================================
diff --git a/intern/cycles/bvh/bvh.cpp b/intern/cycles/bvh/bvh.cpp
index 311d729..a8af8e4 100644
--- a/intern/cycles/bvh/bvh.cpp
+++ b/intern/cycles/bvh/bvh.cpp
@@ -121,57 +121,34 @@ void BVH::refit(Progress& progress)
/* Triangles */
-void BVH::pack_triangle(int idx, float4 storage[3])
-{
- int tob = pack.prim_object[idx];
- assert(tob >= 0 && tob < objects.size());
- const Mesh *mesh = objects[tob]->mesh;
-
- int tidx = pack.prim_index[idx];
- Mesh::Triangle t = mesh->get_triangle(tidx);
- const float3* vpos = &mesh->verts[0];
- float3 v0 = vpos[t.v[0]];
- float3 v1 = vpos[t.v[1]];
- float3 v2 = vpos[t.v[2]];
-
- storage[0] = float3_to_float4(v0);
- storage[1] = float3_to_float4(v1);
- storage[2] = float3_to_float4(v2);
-}
-
void BVH::pack_primitives()
{
- int nsize = TRI_NODE_SIZE;
size_t tidx_size = pack.prim_index.size();
- pack.tri_storage.clear();
- pack.tri_storage.resize(tidx_size * nsize);
+ pack.prim_tri_index.clear();
+ pack.prim_tri_index.resize(tidx_size);
pack.prim_visibility.clear();
pack.prim_visibility.resize(tidx_size);
for(unsigned int i = 0; i < tidx_size; i++) {
if(pack.prim_index[i] != -1) {
- float4 storage[3];
+ int tob = pack.prim_object[i];
+ Object *ob = objects[tob];
if(pack.prim_type[i] & PRIMITIVE_TRIANGLE) {
- pack_triangle(i, storage);
+ pack.prim_tri_index[i] = 3 * (pack.prim_index[i] + ob->mesh->tri_offset);
}
else {
- /* Avoid use of uninitialized memory. */
- memset(&storage, 0, sizeof(storage));
+ pack.prim_tri_index[i] = 0;
}
- memcpy(&pack.tri_storage[i * nsize], storage, sizeof(float4)*3);
-
- int tob = pack.prim_object[i];
- Object *ob = objects[tob];
pack.prim_visibility[i] = ob->visibility;
if(pack.prim_type[i] & PRIMITIVE_ALL_CURVE)
pack.prim_visibility[i] |= PATH_RAY_CURVE;
}
else {
- memset(&pack.tri_storage[i * nsize], 0, sizeof(float4)*3);
+ pack.prim_tri_index[i] = 0;
pack.prim_visibility[i] = 0;
}
}
@@ -208,10 +185,8 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
/* reserve */
size_t prim_index_size = pack.prim_index.size();
- size_t tri_storage_size = pack.tri_storage.size();
size_t pack_prim_index_offset = prim_index_size;
- size_t pack_tri_storage_offset = tri_storage_size;
size_t pack_nodes_offset = nodes_size;
size_t pack_leaf_nodes_offset = leaf_nodes_size;
size_t object_offset = 0;
@@ -225,7 +200,6 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
if(mesh->need_build_bvh()) {
if(mesh_map.find(mesh) == mesh_map.end()) {
prim_index_size += bvh->pack.prim_index.size();
- tri_storage_size += bvh->pack.tri_storage.size();
nodes_size += bvh->pack.nodes.size();
leaf_nodes_size += bvh->pack.leaf_nodes.size();
@@ -240,7 +214,7 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
pack.prim_type.resize(prim_index_size);
pack.prim_object.resize(prim_index_size);
pack.prim_visibility.resize(prim_index_size);
- pack.tri_storage.resize(tri_storage_size);
+ pack.prim_tri_index.resize(prim_index_size);
pack.nodes.resize(nodes_size);
pack.leaf_nodes.resize(leaf_nodes_size);
pack.object_node.resize(objects.size());
@@ -249,7 +223,7 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
int *pack_prim_type = (pack.prim_type.size())? &pack.prim_type[0]: NULL;
int *pack_prim_object = (pack.prim_object.size())? &pack.prim_object[0]: NULL;
uint *pack_prim_visibility = (pack.prim_visibility.size())? &pack.prim_visibility[0]: NULL;
- float4 *pack_tri_storage = (pack.tri_storage.size())? &pack.tri_storage[0]: NULL;
+ uint *pack_prim_tri_index = (pack.prim_tri_index.size())? &pack.prim_tri_index[0]: NULL;
int4 *pack_nodes = (pack.nodes.size())? &pack.nodes[0]: NULL;
int4 *pack_leaf_nodes = (pack.leaf_nodes.size())? &pack.leaf_nodes[0]: NULL;
@@ -290,18 +264,23 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
mesh_map[mesh] = pack.object_node[object_offset-1];
- /* merge primitive and object indexes */
+ /* merge primitive, object and triangle indexes */
if(bvh->pack.prim_index.size()) {
size_t bvh_prim_index_size = bvh->pack.prim_index.size();
int *bvh_prim_index = &bvh->pack.prim_index[0];
int *bvh_prim_type = &bvh->pack.prim_type[0];
uint *bvh_prim_visibility = &bvh->pack.prim_visibility[0];
+ uint *bvh_prim_tri_index = &bvh->pack.prim_tri_index[0];
for(size_t i = 0; i < bvh_prim_index_size; i++) {
- if(bvh->pack.prim_type[i] & PRIMITIVE_ALL_CURVE)
+ if(bvh->pack.prim_type[i] & PRIMITIVE_ALL_CURVE) {
pack_prim_index[pack_prim_index_offset] = bvh_prim_index[i] + mesh_curve_offset;
- else
+ pack_prim_tri_index[pack_prim_index_offset] = 0;
+ }
+ else {
pack_prim_index[pack_prim_index_offset] = bvh_prim_index[i] + mesh_tri_offset;
+ pack_prim_tri_index[pack_prim_index_offset] = bvh_prim_tri_index[i];
+ }
pack_prim_type[pack_prim_index_offset] = bvh_prim_type[i];
pack_prim_visibility[pack_prim_index_offset] = bvh_prim_visibility[i];
@@ -310,14 +289,6 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
}
}
- /* merge triangle intersection data */
- if(bvh->pack.tri_storage.size()) {
- memcpy(pack_tri_storage + pack_tri_storage_offset,
- &bvh->pack.tri_storage[0],
- bvh->pack.tri_storage.size()*sizeof(float4));
- pack_tri_storage_offset += bvh->pack.tri_storage.size();
- }
-
/* merge nodes */
if(bvh->pack.leaf_nodes.size()) {
int4 *leaf_nodes_offset = &bvh->pack.leaf_nodes[0];
diff --git a/intern/cycles/bvh/bvh.h b/intern/cycles/bvh/bvh.h
index b1655f0..5dd91fb 100644
--- a/intern/cycles/bvh/bvh.h
+++ b/intern/cycles/bvh/bvh.h
@@ -52,8 +52,8 @@ struct PackedBVH {
array<int4> leaf_nodes;
/* object index to BVH node index mapping for instances */
array<int> object_node;
- /* Aligned triangle storage for fatser lookup in the kernel. */
- array<float4> tri_storage;
+ /* Mapping from pndex from primitive to triangle. */
+ array<uint> prim_tri_index;
/* primitive type - triangle or strand */
array<int> prim_type;
/* visibility visibilitys for primitives */
@@ -93,7 +93,6 @@ protected:
/* triangles and strands*/
void pack_primitives();
- void pack_triangle(int idx, float4 storage[3]);
/* merge instance BVH's */
void pack_instances(size_t nodes_size, size_t leaf_nodes_size);
diff --git a/intern/cycles/kernel/geom/geom_motion_triangle.h b/intern/cycles/kernel/geom/geom_motion_triangle.h
index ffe5552..b1208e1 100644
--- a/intern/cycles/kernel/geom/geom_motion_triangle.h
+++ b/intern/cycles/kernel/geom/geom_motion_triangle.h
@@ -47,13 +47,13 @@ ccl_device_inline int find_attribute_motion(KernelGlobals *kg, int object, uint
return (attr_map.y == ATTR_ELEMENT_NONE) ? (int)ATTR_STD_NOT_FOUND : (int)attr_map.z;
}
-ccl_device_inline void motion_triangle_verts_for_step(KernelGlobals *kg, float3 tri_vindex, int offset, int numverts, int numsteps, int step, float3 verts[3])
+ccl_device_inline void motion_triangle_verts_for_step(KernelGlobals *kg, uint4 tri_vindex, int offset, int numverts, int numsteps, int step, float3 verts[3])
{
if(step == numsteps) {
/* center step: regular vertex location */
- verts[0] = float4_to_float3(kernel_tex_fetch(__tri_verts, __float_as_int(tri_vindex.x)));
- verts[1] = float4_to_float3(kernel_tex_fetch(__tri_verts, __float_as_int(tri_vindex.y)));
- verts[2] = float4_to_float3(kernel_tex_fetch(__tri_verts, __float_as_int(tri_vindex.z)));
+ verts[0] = float4_to_float3(kernel_tex_fetch(__tri_verts, tri_vindex.w+0));
+ verts[1] = float4_to_float3(kernel_tex_fetch(__tri_verts, tri_vindex.w+1));
+ verts[2] = float4_to_float3(kernel_tex_fetch(__tri_verts, tri_vindex.w+2));
}
else {
/* center step not store in this array */
@@ -62,19 +62,19 @@ ccl_device_inline void motion_triangle_verts_for_step(KernelGlobals *kg, float3
offset += step*numverts;
- verts[0] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + __float_as_int(tri_vindex.x)));
- verts[1] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + __float_as_int(tri_vindex.y)));
- verts[2] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + __float_as_int(tri_vindex.z)));
+ verts[0] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + tri_vindex.x));
+ verts[1] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + tri_vindex.y));
+ verts[2] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + tri_vindex.z));
}
}
-ccl_device_inline void motion_triangle_normals_for_step(KernelGlobals *kg, float3 tri_vindex, int offset, int numverts, int numsteps, int step, float3 normals[3])
+ccl_device_inline void motion_triangle_normals_for_step(KernelGlobals *kg, uint4 tri_vindex, int offset, int numverts, int numsteps, int step, float3 normals[3])
{
if(step == numsteps) {
/* center step: regular vertex location */
- normals[0] = float4_to_float3(kernel_tex_fetch(__tri_vnormal, __float_as_int(tri_vindex.x)));
- normals[1] = float4_to_float3(kernel_tex_fetch(__tri_vnormal, __float_as_int(tri_vindex.y)));
- normals[2] = float4_to_float3(kernel_tex_fetch(__tri_vnormal, __float_as_int(tri_vindex.z)));
+ normals[0] = float4_to_float3(kernel_tex_fetch(__tri_vnormal, tri_vindex.x));
+ normals[1] = float4_to_float3(kernel_tex_fetch(__tri_vnormal, tri_vindex.y));
+ normals[2] = float4_to_float3(kernel_tex_fetch(__tri_vnormal, tri_vindex.z));
}
else {
/* center step not stored in this array */
@@ -83,9 +83,9 @@ ccl_device_inline void motion_triangle_normals_for_step(KernelGlobals *kg, float
offset += step*numverts;
- normals[0] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + __float_as_int(tri_vindex.x)));
- normals[1] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + __float_as_int(tri_vindex.y)));
- normals[2] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + __float_as_int(tri_vindex.z)));
+ normals[0] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + tri_vindex.x));
+ normals[1] = float4_to_float3(kernel_tex_fetch(__attributes_floa
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list