kannonier8/blender
1
0
Fork 0
forked from bartvdbraak/blender
Code Pull Requests Activity
4f03d3290c
blender/intern/cycles/kernel/geom/geom_motion_triangle.h

429 lines
13 KiB
C
Raw Normal View History

Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/*
Cycles: Cleanup file headers Some of the files were wrongly attributing code to some other organizations and in few places proper attribution was missing. This is mainly either a copy-paste error (when new file was created from an existing one and header wasn't updated) or due to some refactor which split non-original-BF code with purely BF code. Should solve some confusion around.
2016-09-29 08:11:40 +00:00
* Copyright 2011-2013 Blender Foundation
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
*
* 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.
*/
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
/* 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.
*/
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
CCL_NAMESPACE_BEGIN
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
/* Time interpolation of vertex positions and normals */
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
ccl_device_inline int find_attribute_motion(KernelGlobals *kg, int object, uint id, AttributeElement *elem)
{
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
/* todo: find a better (faster) solution for this, maybe store offset per object */
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
uint attr_offset = object*kernel_data.bvh.attributes_map_stride;
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])
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
{
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));
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
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: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
}
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])
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
{
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));
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
else {
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
/* center step not stored in this array */
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
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));
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
}
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);
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
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];
}
/* Refine triangle intersection to more precise hit point. For rays that travel
* far the precision is often not so good, this reintersects the primitive from
* a closer distance. */
ccl_device_inline float3 motion_triangle_refine(KernelGlobals *kg, ShaderData *sd, const Intersection *isect, const Ray *ray, float3 verts[3])
{
float3 P = ray->P;
float3 D = ray->D;
float t = isect->t;
#ifdef __INTERSECTION_REFINE__
Cycles code refactor: replace magic ~0 values in the code with defines.
2014-03-29 12:03:47 +00:00
if(isect->object != OBJECT_NONE) {
Cycles: Bring back distance check in re-intersection From more investigation of the numeric failures in the kernel it appears the check was rather correct. But in theory it;s also needed for the motion triangles.
2015-02-10 14:07:55 +00:00
if(UNLIKELY(t == 0.0f)) {
return P;
}
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# ifdef __OBJECT_MOTION__
Cycles: Prepare code for OpenCL camera/motion blur The kernels are now compiling just fine, but there're some issues during rendering. This is still to be investigated.
2015-05-14 13:46:26 +00:00
Transform tfm = ccl_fetch(sd, ob_itfm);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_INVERSE_TRANSFORM);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
P = transform_point(&tfm, P);
D = transform_direction(&tfm, D*t);
D = normalize_len(D, &t);
}
P = P + D*t;
/* compute refined intersection distance */
const float3 e1 = verts[0] - verts[2];
const float3 e2 = verts[1] - verts[2];
const float3 s1 = cross(D, e2);
const float invdivisor = 1.0f/dot(s1, e1);
const float3 d = P - verts[2];
const float3 s2 = cross(d, e1);
float rt = dot(e2, s2)*invdivisor;
/* compute refined position */
P = P + D*rt;
Cycles code refactor: replace magic ~0 values in the code with defines.
2014-03-29 12:03:47 +00:00
if(isect->object != OBJECT_NONE) {
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# ifdef __OBJECT_MOTION__
Cycles: Prepare code for OpenCL camera/motion blur The kernels are now compiling just fine, but there're some issues during rendering. This is still to be investigated.
2015-05-14 13:46:26 +00:00
Transform tfm = ccl_fetch(sd, ob_tfm);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_TRANSFORM);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
P = transform_point(&tfm, P);
}
return P;
#else
return P + D*t;
#endif
}
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
/* Same as above, except that isect->t is assumed to be in object space for instancing */
#ifdef __SUBSURFACE__
Cycles: Tweaks for 32bit CUDA binaries Tweak some inline policies. Not totally crazy yet, and in fact we now have one less ifdef statement now.
2016-02-15 18:11:02 +00:00
# if defined(__KERNEL_CUDA__) && (defined(i386) || defined(_M_IX86))
ccl_device_noinline
# else
ccl_device_inline
# endif
float3 motion_triangle_refine_subsurface(KernelGlobals *kg, ShaderData *sd, const Intersection *isect, const Ray *ray, float3 verts[3])
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
{
float3 P = ray->P;
float3 D = ray->D;
float t = isect->t;
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# ifdef __INTERSECTION_REFINE__
Cycles code refactor: replace magic ~0 values in the code with defines.
2014-03-29 12:03:47 +00:00
if(isect->object != OBJECT_NONE) {
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# ifdef __OBJECT_MOTION__
Cycles: Prepare code for OpenCL camera/motion blur The kernels are now compiling just fine, but there're some issues during rendering. This is still to be investigated.
2015-05-14 13:46:26 +00:00
Transform tfm = ccl_fetch(sd, ob_itfm);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_INVERSE_TRANSFORM);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
P = transform_point(&tfm, P);
D = transform_direction(&tfm, D);
D = normalize(D);
}
P = P + D*t;
/* compute refined intersection distance */
const float3 e1 = verts[0] - verts[2];
const float3 e2 = verts[1] - verts[2];
const float3 s1 = cross(D, e2);
const float invdivisor = 1.0f/dot(s1, e1);
const float3 d = P - verts[2];
const float3 s2 = cross(d, e1);
float rt = dot(e2, s2)*invdivisor;
P = P + D*rt;
Cycles code refactor: replace magic ~0 values in the code with defines.
2014-03-29 12:03:47 +00:00
if(isect->object != OBJECT_NONE) {
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# ifdef __OBJECT_MOTION__
Cycles: Prepare code for OpenCL camera/motion blur The kernels are now compiling just fine, but there're some issues during rendering. This is still to be investigated.
2015-05-14 13:46:26 +00:00
Transform tfm = ccl_fetch(sd, ob_tfm);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_TRANSFORM);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
P = transform_point(&tfm, P);
}
return P;
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
return P + D*t;
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
#endif
/* Setup of motion triangle specific parts of ShaderData, moved into this one
* function to more easily share computation of interpolated positions and
* normals */
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* return 3 triangle vertex normals */
ccl_device_noinline void motion_triangle_shader_setup(KernelGlobals *kg, ShaderData *sd, const Intersection *isect, const Ray *ray, bool subsurface)
{
/* get shader */
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
ccl_fetch(sd, shader) = kernel_tex_fetch(__tri_shader, ccl_fetch(sd, prim));
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* get motion info */
int numsteps, numverts;
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
object_motion_info(kg, ccl_fetch(sd, object), &numsteps, &numverts, NULL);
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* figure out which steps we need to fetch and their interpolation factor */
int maxstep = numsteps*2;
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
int step = min((int)(ccl_fetch(sd, time)*maxstep), maxstep-1);
float t = ccl_fetch(sd, time)*maxstep - step;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* find attribute */
AttributeElement elem;
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
int offset = find_attribute_motion(kg, ccl_fetch(sd, object), ATTR_STD_MOTION_VERTEX_POSITION, &elem);
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
kernel_assert(offset != ATTR_STD_NOT_FOUND);
/* fetch vertex coordinates */
float3 verts[3], 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, ccl_fetch(sd, prim));
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
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];
/* compute refined position */
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
#ifdef __SUBSURFACE__
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
if(!subsurface)
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
#endif
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
ccl_fetch(sd, P) = motion_triangle_refine(kg, sd, isect, ray, verts);
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
#ifdef __SUBSURFACE__
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
else
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
ccl_fetch(sd, P) = motion_triangle_refine_subsurface(kg, sd, isect, ray, verts);
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
#endif
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* compute face normal */
Fix T41219: Cycles backface detection doesn't work properly Root of the issue goes back to the on-fly normals commit and the latest fix for it wasn't actually correct. I've mixed two fixes in there. So the idea here goes back to storing negative scaled object flag and flip runtime-calculated normal if this flag is set, which is pretty much the same as the original fix for the issue from me. The issue with motion blur wasn't caused by the rumtime normals patch and it had issues before, because it already did runtime normals calculation. Now made it so motion triangles takes the negative scale flag into account. This actually makes code more clean imo and avoids rather confusing flipping code in mesh.cpp.
2014-08-13 10:19:12 +00:00
float3 Ng;
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
if(ccl_fetch(sd, flag) & SD_NEGATIVE_SCALE_APPLIED)
Fix T41219: Cycles backface detection doesn't work properly Root of the issue goes back to the on-fly normals commit and the latest fix for it wasn't actually correct. I've mixed two fixes in there. So the idea here goes back to storing negative scaled object flag and flip runtime-calculated normal if this flag is set, which is pretty much the same as the original fix for the issue from me. The issue with motion blur wasn't caused by the rumtime normals patch and it had issues before, because it already did runtime normals calculation. Now made it so motion triangles takes the negative scale flag into account. This actually makes code more clean imo and avoids rather confusing flipping code in mesh.cpp.
2014-08-13 10:19:12 +00:00
Ng = normalize(cross(verts[2] - verts[0], verts[1] - verts[0]));
else
Ng = normalize(cross(verts[1] - verts[0], verts[2] - verts[0]));
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
ccl_fetch(sd, Ng) = Ng;
ccl_fetch(sd, N) = Ng;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* compute derivatives of P w.r.t. uv */
#ifdef __DPDU__
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
ccl_fetch(sd, dPdu) = (verts[0] - verts[2]);
ccl_fetch(sd, dPdv) = (verts[1] - verts[2]);
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
#endif
/* compute smooth normal */
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
if(ccl_fetch(sd, shader) & SHADER_SMOOTH_NORMAL) {
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* find attribute */
AttributeElement elem;
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
int offset = find_attribute_motion(kg, ccl_fetch(sd, object), ATTR_STD_MOTION_VERTEX_NORMAL, &elem);
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
kernel_assert(offset != ATTR_STD_NOT_FOUND);
/* fetch vertex coordinates */
float3 normals[3], next_normals[3];
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 */
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
float u = ccl_fetch(sd, u);
float v = ccl_fetch(sd, v);
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
float w = 1.0f - u - v;
Cycles: OpenCL kernel split This commit contains all the work related on the AMD megakernel split work which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely someone else which we're forgetting to mention. Currently only AMD cards are enabled for the new split kernel, but it is possible to force split opencl kernel to be used by setting the following environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1. Not all the features are supported yet, and that being said no motion blur, camera blur, SSS and volumetrics for now. Also transparent shadows are disabled on AMD device because of some compiler bug. This kernel is also only implements regular path tracing and supporting branched one will take a bit. Branched path tracing is exposed to the interface still, which is a bit misleading and will be hidden there soon. More feature will be enabled once they're ported to the split kernel and tested. Neither regular CPU nor CUDA has any difference, they're generating the same exact code, which means no regressions/improvements there. Based on the research paper: https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf Here's the documentation: https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit Design discussion of the patch: https://developer.blender.org/T44197 Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
ccl_fetch(sd, N) = (u*normals[0] + v*normals[1] + w*normals[2]);
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
}
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
/* Ray intersection. We simply compute the vertex positions at the given ray
* time and do a ray intersection with the resulting triangle */
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
ccl_device_inline bool motion_triangle_intersect(KernelGlobals *kg, Intersection *isect,
Cycles: Cleanup, variable names Use underscore again and also solve confusing part then in BVH smae thing is called prim_addr but in intersection funcitons it was called triAddr.
2016-12-12 11:10:37 +00:00
float3 P, float3 dir, float time, uint visibility, int object, int prim_addr)
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
{
/* primitive index for vertex location lookup */
Cycles: Cleanup, variable names Use underscore again and also solve confusing part then in BVH smae thing is called prim_addr but in intersection funcitons it was called triAddr.
2016-12-12 11:10:37 +00:00
int prim = kernel_tex_fetch(__prim_index, prim_addr);
int fobject = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, prim_addr): object;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* get vertex locations for intersection */
float3 verts[3];
motion_triangle_vertices(kg, fobject, prim, time, verts);
/* ray-triangle intersection, unoptimized */
float t, u, v;
Cycles: avoid 1.0f/(1.0f/x) divisions, which msvc (only) can't optimize. This makes bmw scene in msvc 12 builds 6% faster. It also gives a minor speedup for SSE hair in all compilers.
2014-04-03 18:08:53 +00:00
if(ray_triangle_intersect_uv(P, dir, isect->t, verts[2], verts[0], verts[1], &u, &v, &t)) {
Fix T40320: wrong render layer visibility with cycles deformation motion blur.
2014-05-23 14:10:07 +00:00
#ifdef __VISIBILITY_FLAG__
/* visibility flag test. we do it here under the assumption
* that most triangles are culled by node flags */
Cycles: Cleanup, variable names Use underscore again and also solve confusing part then in BVH smae thing is called prim_addr but in intersection funcitons it was called triAddr.
2016-12-12 11:10:37 +00:00
if(kernel_tex_fetch(__prim_visibility, prim_addr) & visibility)
Fix T40320: wrong render layer visibility with cycles deformation motion blur.
2014-05-23 14:10:07 +00:00
#endif
{
Cycles: set hit values in-order
2014-10-11 09:16:31 +00:00
isect->t = t;
isect->u = u;
isect->v = v;
Cycles: Cleanup, variable names Use underscore again and also solve confusing part then in BVH smae thing is called prim_addr but in intersection funcitons it was called triAddr.
2016-12-12 11:10:37 +00:00
isect->prim = prim_addr;
Fix T40320: wrong render layer visibility with cycles deformation motion blur.
2014-05-23 14:10:07 +00:00
isect->object = object;
isect->type = PRIMITIVE_MOTION_TRIANGLE;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
Fix T40320: wrong render layer visibility with cycles deformation motion blur.
2014-05-23 14:10:07 +00:00
return true;
}
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
return false;
}
/* Special ray intersection routines for subsurface scattering. In that case we
* only want to intersect with primitives in the same object, and if case of
* multiple hits we pick a single random primitive as the intersection point. */
Cycles code refactor: move more code to geom folder, add some comments.
2014-03-29 12:03:48 +00:00
#ifdef __SUBSURFACE__
Cysles: Avoid having ShaderData on the stack This commit introduces a SSS-oriented intersection structure which is replacing old logic of having separate arrays for just intersections and shader data and encapsulates all the data needed for SSS evaluation. This giver a huge stack memory saving on GPU. In own experiments it gave 25% memory usage reduction on GTX560Ti (722MB vs. 946MB). Unfortunately, this gave some performance loss of 20% which only happens on GPU. This is perhaps due to different memory access pattern. Will be solved in the future, hopefully. Famous saying: won in memory - lost in time (which is also valid in other way around).
2015-11-22 10:00:29 +00:00
ccl_device_inline void motion_triangle_intersect_subsurface(
KernelGlobals *kg,
SubsurfaceIntersection *ss_isect,
float3 P,
float3 dir,
float time,
int object,
Cycles: Cleanup, variable names Use underscore again and also solve confusing part then in BVH smae thing is called prim_addr but in intersection funcitons it was called triAddr.
2016-12-12 11:10:37 +00:00
int prim_addr,
Cysles: Avoid having ShaderData on the stack This commit introduces a SSS-oriented intersection structure which is replacing old logic of having separate arrays for just intersections and shader data and encapsulates all the data needed for SSS evaluation. This giver a huge stack memory saving on GPU. In own experiments it gave 25% memory usage reduction on GTX560Ti (722MB vs. 946MB). Unfortunately, this gave some performance loss of 20% which only happens on GPU. This is perhaps due to different memory access pattern. Will be solved in the future, hopefully. Famous saying: won in memory - lost in time (which is also valid in other way around).
2015-11-22 10:00:29 +00:00
float tmax,
uint *lcg_state,
int max_hits)
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
{
/* primitive index for vertex location lookup */
Cycles: Cleanup, variable names Use underscore again and also solve confusing part then in BVH smae thing is called prim_addr but in intersection funcitons it was called triAddr.
2016-12-12 11:10:37 +00:00
int prim = kernel_tex_fetch(__prim_index, prim_addr);
int fobject = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, prim_addr): object;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
/* get vertex locations for intersection */
float3 verts[3];
motion_triangle_vertices(kg, fobject, prim, time, verts);
/* ray-triangle intersection, unoptimized */
float t, u, v;
Cycles: avoid 1.0f/(1.0f/x) divisions, which msvc (only) can't optimize. This makes bmw scene in msvc 12 builds 6% faster. It also gives a minor speedup for SSE hair in all compilers.
2014-04-03 18:08:53 +00:00
if(ray_triangle_intersect_uv(P, dir, tmax, verts[2], verts[0], verts[1], &u, &v, &t)) {
Cycles: Fix SSS with spatial splits and motion blur
2016-07-25 11:53:30 +00:00
for(int i = min(max_hits, ss_isect->num_hits) - 1; i >= 0; --i) {
if(ss_isect->hits[i].t == t) {
return;
}
}
Cysles: Avoid having ShaderData on the stack This commit introduces a SSS-oriented intersection structure which is replacing old logic of having separate arrays for just intersections and shader data and encapsulates all the data needed for SSS evaluation. This giver a huge stack memory saving on GPU. In own experiments it gave 25% memory usage reduction on GTX560Ti (722MB vs. 946MB). Unfortunately, this gave some performance loss of 20% which only happens on GPU. This is perhaps due to different memory access pattern. Will be solved in the future, hopefully. Famous saying: won in memory - lost in time (which is also valid in other way around).
2015-11-22 10:00:29 +00:00
ss_isect->num_hits++;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
int hit;
Cysles: Avoid having ShaderData on the stack This commit introduces a SSS-oriented intersection structure which is replacing old logic of having separate arrays for just intersections and shader data and encapsulates all the data needed for SSS evaluation. This giver a huge stack memory saving on GPU. In own experiments it gave 25% memory usage reduction on GTX560Ti (722MB vs. 946MB). Unfortunately, this gave some performance loss of 20% which only happens on GPU. This is perhaps due to different memory access pattern. Will be solved in the future, hopefully. Famous saying: won in memory - lost in time (which is also valid in other way around).
2015-11-22 10:00:29 +00:00
if(ss_isect->num_hits <= max_hits) {
hit = ss_isect->num_hits - 1;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
else {
/* reservoir sampling: if we are at the maximum number of
* hits, randomly replace element or skip it */
Cysles: Avoid having ShaderData on the stack This commit introduces a SSS-oriented intersection structure which is replacing old logic of having separate arrays for just intersections and shader data and encapsulates all the data needed for SSS evaluation. This giver a huge stack memory saving on GPU. In own experiments it gave 25% memory usage reduction on GTX560Ti (722MB vs. 946MB). Unfortunately, this gave some performance loss of 20% which only happens on GPU. This is perhaps due to different memory access pattern. Will be solved in the future, hopefully. Famous saying: won in memory - lost in time (which is also valid in other way around).
2015-11-22 10:00:29 +00:00
hit = lcg_step_uint(lcg_state) % ss_isect->num_hits;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
if(hit >= max_hits)
return;
}
/* record intersection */
Cysles: Avoid having ShaderData on the stack This commit introduces a SSS-oriented intersection structure which is replacing old logic of having separate arrays for just intersections and shader data and encapsulates all the data needed for SSS evaluation. This giver a huge stack memory saving on GPU. In own experiments it gave 25% memory usage reduction on GTX560Ti (722MB vs. 946MB). Unfortunately, this gave some performance loss of 20% which only happens on GPU. This is perhaps due to different memory access pattern. Will be solved in the future, hopefully. Famous saying: won in memory - lost in time (which is also valid in other way around).
2015-11-22 10:00:29 +00:00
Intersection *isect = &ss_isect->hits[hit];
Cycles: set hit values in-order
2014-10-11 09:16:31 +00:00
isect->t = t;
isect->u = u;
isect->v = v;
Cycles: Cleanup, variable names Use underscore again and also solve confusing part then in BVH smae thing is called prim_addr but in intersection funcitons it was called triAddr.
2016-12-12 11:10:37 +00:00
isect->prim = prim_addr;
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
isect->object = object;
isect->type = PRIMITIVE_MOTION_TRIANGLE;
Cysles: Avoid having ShaderData on the stack This commit introduces a SSS-oriented intersection structure which is replacing old logic of having separate arrays for just intersections and shader data and encapsulates all the data needed for SSS evaluation. This giver a huge stack memory saving on GPU. In own experiments it gave 25% memory usage reduction on GTX560Ti (722MB vs. 946MB). Unfortunately, this gave some performance loss of 20% which only happens on GPU. This is perhaps due to different memory access pattern. Will be solved in the future, hopefully. Famous saying: won in memory - lost in time (which is also valid in other way around).
2015-11-22 10:00:29 +00:00
/* Record geometric normal. */
ss_isect->Ng[hit] = normalize(cross(verts[1] - verts[0],
verts[2] - verts[0]));
Cycles: add support for mesh deformation motion blur.
2014-03-29 12:03:47 +00:00
}
}
#endif
CCL_NAMESPACE_END
Copy Permalink