Page MenuHome
Differential D2226 Diff 7427 intern/cycles/kernel/bvh/bvh_traversal_callback.h

Changeset View

Standalone View

View Options

intern/cycles/kernel/bvh/bvh_traversal_callback.h

  • This file was added.
/*
* Adapted from code Copyright 2009-2010 NVIDIA Corporation,
* and code copyright 2009-2012 Intel Corporation
*
* Modifications 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.
*/
#ifdef __QBVH__
# include "qbvh_traversal_callback.h"
#endif
#if BVH_FEATURE(BVH_HAIR)
# define NODE_INTERSECT bvh_node_intersect
# define NODE_INTERSECT_ROBUST bvh_node_intersect_robust
#else
# define NODE_INTERSECT bvh_aligned_node_intersect
# define NODE_INTERSECT_ROBUST bvh_aligned_node_intersect_robust
#endif
/* This is a template BVH traversal function, where various features can be
* enabled/disabled. This way we can compile optimized versions for each case
* without new features slowing things down.
*
* BVH_INSTANCING: object instancing
* BVH_HAIR: hair curve rendering
* BVH_HAIR_MINIMUM_WIDTH: hair curve rendering with minimum width
* BVH_MOTION: motion blur rendering
*
*/
#ifndef __KERNEL_GPU__
ccl_device
#else
ccl_device_inline
#endif
bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
const Ray *ray,
Intersection *isect,
const uint visibility,
#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
uint *lcg_state,
float difl,
float extmax,
#endif
BVHIntersectionCallback callback,
void *user_data)
{
/* todo:
* - test if pushing distance on the stack helps (for non shadow rays)
* - separate version for shadow rays
* - likely and unlikely for if() statements
* - test restrict attribute for pointers
*/
/* traversal stack in CUDA thread-local memory */
int traversal_stack[BVH_STACK_SIZE];
traversal_stack[0] = ENTRYPOINT_SENTINEL;
/* traversal variables in registers */
int stack_ptr = 0;
int node_addr = kernel_data.bvh.root;
/* ray parameters in registers */
float3 P = ray->P;
float3 dir = bvh_clamp_direction(ray->D);
float3 idir = bvh_inverse_direction(dir);
int object = OBJECT_NONE;
#if BVH_FEATURE(BVH_MOTION)
Transform ob_itfm;
#endif
isect->t = ray->t;
isect->u = 0.0f;
isect->v = 0.0f;
isect->prim = PRIM_NONE;
isect->object = OBJECT_NONE;
BVH_DEBUG_INIT();
#if defined(__KERNEL_SSE2__)
const shuffle_swap_t shuf_identity = shuffle_swap_identity();
const shuffle_swap_t shuf_swap = shuffle_swap_swap();
const ssef pn = cast(ssei(0, 0, 0x80000000, 0x80000000));
ssef Psplat[3], idirsplat[3];
# if BVH_FEATURE(BVH_HAIR)
ssef tnear(0.0f), tfar(isect->t);
# endif
shuffle_swap_t shufflexyz[3];
Psplat[0] = ssef(P.x);
Psplat[1] = ssef(P.y);
Psplat[2] = ssef(P.z);
ssef tsplat(0.0f, 0.0f, -isect->t, -isect->t);
gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
#endif
IsectPrecalc isect_precalc;
triangle_intersect_precalc(dir, &isect_precalc);
/* traversal loop */
do {
do {
/* traverse internal nodes */
while(node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) {
int node_addr_child1, traverse_mask;
float dist[2];
float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr+0);
#if !defined(__KERNEL_SSE2__)
# if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
if(difl != 0.0f) {
traverse_mask = NODE_INTERSECT_ROBUST(kg,
P,
# if BVH_FEATURE(BVH_HAIR)
dir,
# endif
idir,
isect->t,
difl,
extmax,
node_addr,
visibility,
dist);
}
else
# endif
{
traverse_mask = NODE_INTERSECT(kg,
P,
# if BVH_FEATURE(BVH_HAIR)
dir,
# endif
idir,
isect->t,
node_addr,
visibility,
dist);
}
#else // __KERNEL_SSE2__
# if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
if(difl != 0.0f) {
traverse_mask = NODE_INTERSECT_ROBUST(kg,
P,
dir,
# if BVH_FEATURE(BVH_HAIR)
tnear,
tfar,
# endif
tsplat,
Psplat,
idirsplat,
shufflexyz,
difl,
extmax,
node_addr,
visibility,
dist);
}
else
# endif
{
traverse_mask = NODE_INTERSECT(kg,
P,
dir,
# if BVH_FEATURE(BVH_HAIR)
tnear,
tfar,
# endif
tsplat,
Psplat,
idirsplat,
shufflexyz,
node_addr,
visibility,
dist);
}
#endif // __KERNEL_SSE2__
node_addr = __float_as_int(cnodes.z);
node_addr_child1 = __float_as_int(cnodes.w);
if(traverse_mask == 3) {
/* Both children were intersected, push the farther one. */
bool is_closest_child1 = (dist[1] < dist[0]);
if(is_closest_child1) {
int tmp = node_addr;
node_addr = node_addr_child1;
node_addr_child1 = tmp;
}
++stack_ptr;
kernel_assert(stack_ptr < BVH_STACK_SIZE);
traversal_stack[stack_ptr] = node_addr_child1;
}
else {
/* One child was intersected. */
if(traverse_mask == 2) {
node_addr = node_addr_child1;
}
else if(traverse_mask == 0) {
/* Neither child was intersected. */
node_addr = traversal_stack[stack_ptr];
--stack_ptr;
}
}
BVH_DEBUG_NEXT_STEP();
}
/* if node is leaf, fetch triangle list */
if(node_addr < 0) {
float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr-1));
int prim_addr = __float_as_int(leaf.x);
#if BVH_FEATURE(BVH_INSTANCING)
if(prim_addr >= 0) {
#endif
const int prim_addr2 = __float_as_int(leaf.y);
const uint type = __float_as_int(leaf.w);
/* pop */
node_addr = traversal_stack[stack_ptr];
--stack_ptr;
/* primitive intersection */
switch(type & PRIMITIVE_ALL) {
case PRIMITIVE_TRIANGLE: {
for(; prim_addr < prim_addr2; prim_addr++) {
BVH_DEBUG_NEXT_STEP();
kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
if(triangle_intersect(kg,
&isect_precalc,
isect,
P,
visibility,
object,
prim_addr))
{
if(!callback(kg, ray, isect, object, user_data)) {
return true;
}
#if defined(__KERNEL_SSE2__)
tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t);
# if BVH_FEATURE(BVH_HAIR)
tfar = ssef(isect->t);
# endif
#endif
}
}
break;
}
#if BVH_FEATURE(BVH_MOTION)
case PRIMITIVE_MOTION_TRIANGLE: {
for(; prim_addr < prim_addr2; prim_addr++) {
BVH_DEBUG_NEXT_STEP();
kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
if(motion_triangle_intersect(kg,
isect,
P,
dir,
ray->time,
visibility,
object,
prim_addr))
{
if(!callback(kg, ray, isect, object, user_data)) {
return true;
}
# if defined(__KERNEL_SSE2__)
tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t);
# if BVH_FEATURE(BVH_HAIR)
tfar = ssef(isect->t);
# endif
# endif
}
}
break;
}
#endif /* BVH_FEATURE(BVH_MOTION) */
#if BVH_FEATURE(BVH_HAIR)
case PRIMITIVE_CURVE:
case PRIMITIVE_MOTION_CURVE: {
for(; prim_addr < prim_addr2; prim_addr++) {
BVH_DEBUG_NEXT_STEP();
kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
bool hit;
if(kernel_data.curve.curveflags & CURVE_KN_INTERPOLATE) {
hit = bvh_cardinal_curve_intersect(kg,
isect,
P,
dir,
visibility,
object,
prim_addr,
ray->time,
type,
lcg_state,
difl,
extmax);
}
else {
hit = bvh_curve_intersect(kg,
isect,
P,
dir,
visibility,
object,
prim_addr,
ray->time,
type,
lcg_state,
difl,
extmax);
}
if(hit) {
if(!callback(kg, ray, isect, object, user_data)) {
return true;
}
# if defined(__KERNEL_SSE2__)
tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t);
# if BVH_FEATURE(BVH_HAIR)
tfar = ssef(isect->t);
# endif
# endif
}
}
break;
}
#endif /* BVH_FEATURE(BVH_HAIR) */
}
}
#if BVH_FEATURE(BVH_INSTANCING)
else {
/* instance push */
object = kernel_tex_fetch(__prim_object, -prim_addr-1);
# if BVH_FEATURE(BVH_MOTION)
bvh_instance_motion_push(kg, object, ray, &P, &dir, &idir, &isect->t, &ob_itfm);
# else
bvh_instance_push(kg, object, ray, &P, &dir, &idir, &isect->t);
# endif
triangle_intersect_precalc(dir, &isect_precalc);
# if defined(__KERNEL_SSE2__)
Psplat[0] = ssef(P.x);
Psplat[1] = ssef(P.y);
Psplat[2] = ssef(P.z);
tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t);
# if BVH_FEATURE(BVH_HAIR)
tfar = ssef(isect->t);
# endif
gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
# endif
++stack_ptr;
kernel_assert(stack_ptr < BVH_STACK_SIZE);
traversal_stack[stack_ptr] = ENTRYPOINT_SENTINEL;
node_addr = kernel_tex_fetch(__object_node, object);
BVH_DEBUG_NEXT_INSTANCE();
}
}
#endif /* FEATURE(BVH_INSTANCING) */
} while(node_addr != ENTRYPOINT_SENTINEL);
#if BVH_FEATURE(BVH_INSTANCING)
if(stack_ptr >= 0) {
kernel_assert(object != OBJECT_NONE);
/* instance pop */
# if BVH_FEATURE(BVH_MOTION)
bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, &isect->t, &ob_itfm);
# else
bvh_instance_pop(kg, object, ray, &P, &dir, &idir, &isect->t);
# endif
triangle_intersect_precalc(dir, &isect_precalc);
# if defined(__KERNEL_SSE2__)
Psplat[0] = ssef(P.x);
Psplat[1] = ssef(P.y);
Psplat[2] = ssef(P.z);
tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t);
# if BVH_FEATURE(BVH_HAIR)
tfar = ssef(isect->t);
# endif
gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
# endif
object = OBJECT_NONE;
node_addr = traversal_stack[stack_ptr];
--stack_ptr;
}
#endif /* FEATURE(BVH_INSTANCING) */
} while(node_addr != ENTRYPOINT_SENTINEL);
return (isect->prim != PRIM_NONE);
}
ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg,
const Ray *ray,
Intersection *isect,
const uint visibility,
#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
uint *lcg_state,
float difl,
float extmax,
#endif
BVHIntersectionCallback callback,
void *user_data)
{
#ifdef __QBVH__
if(kernel_data.bvh.use_qbvh) {
return BVH_FUNCTION_FULL_NAME(QBVH)(kg,
ray,
isect,
visibility,
#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
lcg_state,
difl,
extmax,
#endif
callback,
user_data);
}
else
#endif
{
kernel_assert(kernel_data.bvh.use_qbvh == false);
return BVH_FUNCTION_FULL_NAME(BVH)(kg,
ray,
isect,
visibility,
#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
lcg_state,
difl,
extmax,
#endif
callback,
user_data);
}
}
#undef BVH_FUNCTION_NAME
#undef BVH_FUNCTION_FEATURES
#undef NODE_INTERSECT
#undef NODE_INTERSECT_ROBUST