blender/intern/cycles/kernel/geom/geom_motion_triangle.h

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

174 lines
6.7 KiB
C
Raw Normal View History

/*
* Copyright 2011-2013 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.
*/
/* Motion Triangle Primitive
*
* These are stored as regular triangles, plus extra positions and normals at
* times other than the frame center. Computing the triangle vertex positions
* or normals at a given ray time is a matter of interpolation of the two steps
* between which the ray time lies.
*
* The extra positions and normals are stored as ATTR_STD_MOTION_VERTEX_POSITION
* and ATTR_STD_MOTION_VERTEX_NORMAL mesh attributes.
*/
CCL_NAMESPACE_BEGIN
/* Time interpolation of vertex positions and normals */
ccl_device_inline int find_attribute_motion(KernelGlobals *kg,
int object,
uint id,
AttributeElement *elem)
{
/* todo: find a better (faster) solution for this, maybe store offset per object */
uint attr_offset = object_attribute_map_offset(kg, object);
uint4 attr_map = kernel_tex_fetch(__attributes_map, attr_offset);
while (attr_map.x != id) {
attr_offset += ATTR_PRIM_TYPES;
attr_map = kernel_tex_fetch(__attributes_map, attr_offset);
}
*elem = (AttributeElement)attr_map.y;
/* return result */
return (attr_map.y == ATTR_ELEMENT_NONE) ? (int)ATTR_STD_NOT_FOUND : (int)attr_map.z;
}
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_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. Additionally, the order of vertices in that array is aligned with primitives from BVH. This is needed to keep cache as much coherent as possible for BVH traversal. This causes some extra tricks needed to fill the array in and deal with True Displacement but those trickery is fully required to prevent noticeable slowdown. 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. Mainly this loss is caused by extra lookup happening to access vertex coordinate. Fortunately, tricks here and there (i,e, some types changes to avoid casts which are not really coming for free) reduces those losses to an acceptable level. So now they are within couple of percent only, On a positive site we've achieved: - 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. Similar figure for bunny.blend. This memory save was the main goal of this commit to move forward with Hair BVH which required more memory per BVH node. So while this sounds exciting, this memory optimization will become invisible by upcoming Hair BVH work. But again on a positive side, we can add an option to NOT use Hair BVH and then we'll have same-ish render times as we've got currently but will have this 20% memory benefit on hairy scenes.
2016-06-10 14:13:50 +00:00
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 */
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_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. Additionally, the order of vertices in that array is aligned with primitives from BVH. This is needed to keep cache as much coherent as possible for BVH traversal. This causes some extra tricks needed to fill the array in and deal with True Displacement but those trickery is fully required to prevent noticeable slowdown. 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. Mainly this loss is caused by extra lookup happening to access vertex coordinate. Fortunately, tricks here and there (i,e, some types changes to avoid casts which are not really coming for free) reduces those losses to an acceptable level. So now they are within couple of percent only, On a positive site we've achieved: - 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. Similar figure for bunny.blend. This memory save was the main goal of this commit to move forward with Hair BVH which required more memory per BVH node. So while this sounds exciting, this memory optimization will become invisible by upcoming Hair BVH work. But again on a positive side, we can add an option to NOT use Hair BVH and then we'll have same-ish render times as we've got currently but will have this 20% memory benefit on hairy scenes.
2016-06-10 14:13:50 +00:00
verts[0] = float4_to_float3(kernel_tex_fetch(__prim_tri_verts, tri_vindex.w + 0));
verts[1] = float4_to_float3(kernel_tex_fetch(__prim_tri_verts, tri_vindex.w + 1));
verts[2] = float4_to_float3(kernel_tex_fetch(__prim_tri_verts, tri_vindex.w + 2));
}
else {
/* center step not store in this array */
if (step > numsteps)
step--;
offset += step * numverts;
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_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. Additionally, the order of vertices in that array is aligned with primitives from BVH. This is needed to keep cache as much coherent as possible for BVH traversal. This causes some extra tricks needed to fill the array in and deal with True Displacement but those trickery is fully required to prevent noticeable slowdown. 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. Mainly this loss is caused by extra lookup happening to access vertex coordinate. Fortunately, tricks here and there (i,e, some types changes to avoid casts which are not really coming for free) reduces those losses to an acceptable level. So now they are within couple of percent only, On a positive site we've achieved: - 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. Similar figure for bunny.blend. This memory save was the main goal of this commit to move forward with Hair BVH which required more memory per BVH node. So while this sounds exciting, this memory optimization will become invisible by upcoming Hair BVH work. But again on a positive side, we can add an option to NOT use Hair BVH and then we'll have same-ish render times as we've got currently but will have this 20% memory benefit on hairy scenes.
2016-06-10 14:13:50 +00:00
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));
}
}
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_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. Additionally, the order of vertices in that array is aligned with primitives from BVH. This is needed to keep cache as much coherent as possible for BVH traversal. This causes some extra tricks needed to fill the array in and deal with True Displacement but those trickery is fully required to prevent noticeable slowdown. 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. Mainly this loss is caused by extra lookup happening to access vertex coordinate. Fortunately, tricks here and there (i,e, some types changes to avoid casts which are not really coming for free) reduces those losses to an acceptable level. So now they are within couple of percent only, On a positive site we've achieved: - 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. Similar figure for bunny.blend. This memory save was the main goal of this commit to move forward with Hair BVH which required more memory per BVH node. So while this sounds exciting, this memory optimization will become invisible by upcoming Hair BVH work. But again on a positive side, we can add an option to NOT use Hair BVH and then we'll have same-ish render times as we've got currently but will have this 20% memory benefit on hairy scenes.
2016-06-10 14:13:50 +00:00
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 */
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_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. Additionally, the order of vertices in that array is aligned with primitives from BVH. This is needed to keep cache as much coherent as possible for BVH traversal. This causes some extra tricks needed to fill the array in and deal with True Displacement but those trickery is fully required to prevent noticeable slowdown. 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. Mainly this loss is caused by extra lookup happening to access vertex coordinate. Fortunately, tricks here and there (i,e, some types changes to avoid casts which are not really coming for free) reduces those losses to an acceptable level. So now they are within couple of percent only, On a positive site we've achieved: - 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. Similar figure for bunny.blend. This memory save was the main goal of this commit to move forward with Hair BVH which required more memory per BVH node. So while this sounds exciting, this memory optimization will become invisible by upcoming Hair BVH work. But again on a positive side, we can add an option to NOT use Hair BVH and then we'll have same-ish render times as we've got currently but will have this 20% memory benefit on hairy scenes.
2016-06-10 14:13:50 +00:00
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 {
2017-01-16 19:08:19 +00:00
/* center step is not stored in this array */
if (step > numsteps)
step--;
offset += step * numverts;
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_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. Additionally, the order of vertices in that array is aligned with primitives from BVH. This is needed to keep cache as much coherent as possible for BVH traversal. This causes some extra tricks needed to fill the array in and deal with True Displacement but those trickery is fully required to prevent noticeable slowdown. 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. Mainly this loss is caused by extra lookup happening to access vertex coordinate. Fortunately, tricks here and there (i,e, some types changes to avoid casts which are not really coming for free) reduces those losses to an acceptable level. So now they are within couple of percent only, On a positive site we've achieved: - 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. Similar figure for bunny.blend. This memory save was the main goal of this commit to move forward with Hair BVH which required more memory per BVH node. So while this sounds exciting, this memory optimization will become invisible by upcoming Hair BVH work. But again on a positive side, we can add an option to NOT use Hair BVH and then we'll have same-ish render times as we've got currently but will have this 20% memory benefit on hairy scenes.
2016-06-10 14:13:50 +00:00
normals[0] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + tri_vindex.x));
normals[1] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + tri_vindex.y));
normals[2] = float4_to_float3(kernel_tex_fetch(__attributes_float3, offset + tri_vindex.z));
}
}
ccl_device_inline void motion_triangle_vertices(
KernelGlobals *kg, int object, int prim, float time, float3 verts[3])
{
/* get motion info */
int numsteps, numverts;
object_motion_info(kg, object, &numsteps, &numverts, NULL);
/* figure out which steps we need to fetch and their interpolation factor */
int maxstep = numsteps * 2;
int step = min((int)(time * maxstep), maxstep - 1);
float t = time * maxstep - step;
/* find attribute */
AttributeElement elem;
int offset = find_attribute_motion(kg, object, ATTR_STD_MOTION_VERTEX_POSITION, &elem);
kernel_assert(offset != ATTR_STD_NOT_FOUND);
/* fetch vertex coordinates */
float3 next_verts[3];
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_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. Additionally, the order of vertices in that array is aligned with primitives from BVH. This is needed to keep cache as much coherent as possible for BVH traversal. This causes some extra tricks needed to fill the array in and deal with True Displacement but those trickery is fully required to prevent noticeable slowdown. 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. Mainly this loss is caused by extra lookup happening to access vertex coordinate. Fortunately, tricks here and there (i,e, some types changes to avoid casts which are not really coming for free) reduces those losses to an acceptable level. So now they are within couple of percent only, On a positive site we've achieved: - 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. Similar figure for bunny.blend. This memory save was the main goal of this commit to move forward with Hair BVH which required more memory per BVH node. So while this sounds exciting, this memory optimization will become invisible by upcoming Hair BVH work. But again on a positive side, we can add an option to NOT use Hair BVH and then we'll have same-ish render times as we've got currently but will have this 20% memory benefit on hairy scenes.
2016-06-10 14:13:50 +00:00
uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, prim);
motion_triangle_verts_for_step(kg, tri_vindex, offset, numverts, numsteps, step, verts);
motion_triangle_verts_for_step(kg, tri_vindex, offset, numverts, numsteps, step + 1, next_verts);
/* interpolate between steps */
verts[0] = (1.0f - t) * verts[0] + t * next_verts[0];
verts[1] = (1.0f - t) * verts[1] + t * next_verts[1];
verts[2] = (1.0f - t) * verts[2] + t * next_verts[2];
}
ccl_device_inline float3 motion_triangle_smooth_normal(
KernelGlobals *kg, float3 Ng, int object, int prim, float u, float v, float time)
{
/* get motion info */
int numsteps, numverts;
object_motion_info(kg, object, &numsteps, &numverts, NULL);
/* figure out which steps we need to fetch and their interpolation factor */
int maxstep = numsteps * 2;
int step = min((int)(time * maxstep), maxstep - 1);
float t = time * maxstep - step;
/* find attribute */
AttributeElement elem;
int offset = find_attribute_motion(kg, object, ATTR_STD_MOTION_VERTEX_NORMAL, &elem);
kernel_assert(offset != ATTR_STD_NOT_FOUND);
/* fetch normals */
float3 normals[3], next_normals[3];
uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, prim);
motion_triangle_normals_for_step(kg, tri_vindex, offset, numverts, numsteps, step, normals);
motion_triangle_normals_for_step(
kg, tri_vindex, offset, numverts, numsteps, step + 1, next_normals);
/* interpolate between steps */
normals[0] = (1.0f - t) * normals[0] + t * next_normals[0];
normals[1] = (1.0f - t) * normals[1] + t * next_normals[1];
normals[2] = (1.0f - t) * normals[2] + t * next_normals[2];
/* interpolate between vertices */
float w = 1.0f - u - v;
float3 N = safe_normalize(u * normals[0] + v * normals[1] + w * normals[2]);
return is_zero(N) ? Ng : N;
}
CCL_NAMESPACE_END