1 /*
2 * Adapted from code Copyright 2009-2010 NVIDIA Corporation,
3 * and code copyright 2009-2012 Intel Corporation
4 *
5 * Modifications Copyright 2011-2013, Blender Foundation.
6 *
7 * Licensed under the Apache License, Version 2.0 (the "License");
8 * you may not use this file except in compliance with the License.
9 * You may obtain a copy of the License at
10 *
11 * http://www.apache.org/licenses/LICENSE-2.0
12 *
13 * Unless required by applicable law or agreed to in writing, software
14 * distributed under the License is distributed on an "AS IS" BASIS,
15 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 * See the License for the specific language governing permissions and
17 * limitations under the License.
18 */
19
20 #if BVH_FEATURE(BVH_HAIR)
21 # define NODE_INTERSECT bvh_node_intersect
22 #else
23 # define NODE_INTERSECT bvh_aligned_node_intersect
24 #endif
25
26 /* This is a template BVH traversal function, where various features can be
27 * enabled/disabled. This way we can compile optimized versions for each case
28 * without new features slowing things down.
29 *
30 * BVH_HAIR: hair curve rendering
31 * BVH_MOTION: motion blur rendering
32 */
33
34 #ifndef __KERNEL_GPU__
35 ccl_device
36 #else
37 ccl_device_inline
38 #endif
BVH_FUNCTION_FULL_NAME(BVH)39 bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
40 const Ray *ray,
41 Intersection *isect_array,
42 const uint visibility,
43 const uint max_hits,
44 uint *num_hits)
45 {
46 /* todo:
47 * - likely and unlikely for if() statements
48 * - test restrict attribute for pointers
49 */
50
51 /* traversal stack in CUDA thread-local memory */
52 int traversal_stack[BVH_STACK_SIZE];
53 traversal_stack[0] = ENTRYPOINT_SENTINEL;
54
55 /* traversal variables in registers */
56 int stack_ptr = 0;
57 int node_addr = kernel_data.bvh.root;
58
59 /* ray parameters in registers */
60 const float tmax = ray->t;
61 float3 P = ray->P;
62 float3 dir = bvh_clamp_direction(ray->D);
63 float3 idir = bvh_inverse_direction(dir);
64 int object = OBJECT_NONE;
65 float isect_t = tmax;
66
67 #if BVH_FEATURE(BVH_MOTION)
68 Transform ob_itfm;
69 #endif
70
71 int num_hits_in_instance = 0;
72
73 *num_hits = 0;
74 isect_array->t = tmax;
75
76 /* traversal loop */
77 do {
78 do {
79 /* traverse internal nodes */
80 while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) {
81 int node_addr_child1, traverse_mask;
82 float dist[2];
83 float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0);
84
85 traverse_mask = NODE_INTERSECT(kg,
86 P,
87 #if BVH_FEATURE(BVH_HAIR)
88 dir,
89 #endif
90 idir,
91 isect_t,
92 node_addr,
93 visibility,
94 dist);
95
96 node_addr = __float_as_int(cnodes.z);
97 node_addr_child1 = __float_as_int(cnodes.w);
98
99 if (traverse_mask == 3) {
100 /* Both children were intersected, push the farther one. */
101 bool is_closest_child1 = (dist[1] < dist[0]);
102 if (is_closest_child1) {
103 int tmp = node_addr;
104 node_addr = node_addr_child1;
105 node_addr_child1 = tmp;
106 }
107
108 ++stack_ptr;
109 kernel_assert(stack_ptr < BVH_STACK_SIZE);
110 traversal_stack[stack_ptr] = node_addr_child1;
111 }
112 else {
113 /* One child was intersected. */
114 if (traverse_mask == 2) {
115 node_addr = node_addr_child1;
116 }
117 else if (traverse_mask == 0) {
118 /* Neither child was intersected. */
119 node_addr = traversal_stack[stack_ptr];
120 --stack_ptr;
121 }
122 }
123 }
124
125 /* if node is leaf, fetch triangle list */
126 if (node_addr < 0) {
127 float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1));
128 int prim_addr = __float_as_int(leaf.x);
129
130 if (prim_addr >= 0) {
131 const int prim_addr2 = __float_as_int(leaf.y);
132 const uint type = __float_as_int(leaf.w);
133 const uint p_type = type & PRIMITIVE_ALL;
134
135 /* pop */
136 node_addr = traversal_stack[stack_ptr];
137 --stack_ptr;
138
139 /* primitive intersection */
140 while (prim_addr < prim_addr2) {
141 kernel_assert((kernel_tex_fetch(__prim_type, prim_addr) & PRIMITIVE_ALL) == p_type);
142 bool hit;
143
144 /* todo: specialized intersect functions which don't fill in
145 * isect unless needed and check SD_HAS_TRANSPARENT_SHADOW?
146 * might give a few % performance improvement */
147
148 switch (p_type) {
149 case PRIMITIVE_TRIANGLE: {
150 hit = triangle_intersect(kg, isect_array, P, dir, visibility, object, prim_addr);
151 break;
152 }
153 #if BVH_FEATURE(BVH_MOTION)
154 case PRIMITIVE_MOTION_TRIANGLE: {
155 hit = motion_triangle_intersect(
156 kg, isect_array, P, dir, ray->time, visibility, object, prim_addr);
157 break;
158 }
159 #endif
160 #if BVH_FEATURE(BVH_HAIR)
161 case PRIMITIVE_CURVE_THICK:
162 case PRIMITIVE_MOTION_CURVE_THICK:
163 case PRIMITIVE_CURVE_RIBBON:
164 case PRIMITIVE_MOTION_CURVE_RIBBON: {
165 const uint curve_type = kernel_tex_fetch(__prim_type, prim_addr);
166 hit = curve_intersect(
167 kg, isect_array, P, dir, visibility, object, prim_addr, ray->time, curve_type);
168 break;
169 }
170 #endif
171 default: {
172 hit = false;
173 break;
174 }
175 }
176
177 /* shadow ray early termination */
178 if (hit) {
179 /* detect if this surface has a shader with transparent shadows */
180
181 /* todo: optimize so primitive visibility flag indicates if
182 * the primitive has a transparent shadow shader? */
183 int prim = kernel_tex_fetch(__prim_index, isect_array->prim);
184 int shader = 0;
185
186 #ifdef __HAIR__
187 if (kernel_tex_fetch(__prim_type, isect_array->prim) & PRIMITIVE_ALL_TRIANGLE)
188 #endif
189 {
190 shader = kernel_tex_fetch(__tri_shader, prim);
191 }
192 #ifdef __HAIR__
193 else {
194 float4 str = kernel_tex_fetch(__curves, prim);
195 shader = __float_as_int(str.z);
196 }
197 #endif
198 int flag = kernel_tex_fetch(__shaders, (shader & SHADER_MASK)).flags;
199
200 /* if no transparent shadows, all light is blocked */
201 if (!(flag & SD_HAS_TRANSPARENT_SHADOW)) {
202 return true;
203 }
204 /* if maximum number of hits reached, block all light */
205 else if (*num_hits == max_hits) {
206 return true;
207 }
208
209 /* move on to next entry in intersections array */
210 isect_array++;
211 (*num_hits)++;
212 num_hits_in_instance++;
213
214 isect_array->t = isect_t;
215 }
216
217 prim_addr++;
218 }
219 }
220 else {
221 /* instance push */
222 object = kernel_tex_fetch(__prim_object, -prim_addr - 1);
223
224 #if BVH_FEATURE(BVH_MOTION)
225 isect_t = bvh_instance_motion_push(kg, object, ray, &P, &dir, &idir, isect_t, &ob_itfm);
226 #else
227 isect_t = bvh_instance_push(kg, object, ray, &P, &dir, &idir, isect_t);
228 #endif
229
230 num_hits_in_instance = 0;
231 isect_array->t = isect_t;
232
233 ++stack_ptr;
234 kernel_assert(stack_ptr < BVH_STACK_SIZE);
235 traversal_stack[stack_ptr] = ENTRYPOINT_SENTINEL;
236
237 node_addr = kernel_tex_fetch(__object_node, object);
238 }
239 }
240 } while (node_addr != ENTRYPOINT_SENTINEL);
241
242 if (stack_ptr >= 0) {
243 kernel_assert(object != OBJECT_NONE);
244
245 /* Instance pop. */
246 if (num_hits_in_instance) {
247 float t_fac;
248
249 #if BVH_FEATURE(BVH_MOTION)
250 bvh_instance_motion_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac, &ob_itfm);
251 #else
252 bvh_instance_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac);
253 #endif
254
255 /* scale isect->t to adjust for instancing */
256 for (int i = 0; i < num_hits_in_instance; i++) {
257 (isect_array - i - 1)->t *= t_fac;
258 }
259 }
260 else {
261 #if BVH_FEATURE(BVH_MOTION)
262 bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX, &ob_itfm);
263 #else
264 bvh_instance_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX);
265 #endif
266 }
267
268 isect_t = tmax;
269 isect_array->t = isect_t;
270
271 object = OBJECT_NONE;
272 node_addr = traversal_stack[stack_ptr];
273 --stack_ptr;
274 }
275 } while (node_addr != ENTRYPOINT_SENTINEL);
276
277 return false;
278 }
279
BVH_FUNCTION_NAME(KernelGlobals * kg,const Ray * ray,Intersection * isect_array,const uint visibility,const uint max_hits,uint * num_hits)280 ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg,
281 const Ray *ray,
282 Intersection *isect_array,
283 const uint visibility,
284 const uint max_hits,
285 uint *num_hits)
286 {
287 return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect_array, visibility, max_hits, num_hits);
288 }
289
290 #undef BVH_FUNCTION_NAME
291 #undef BVH_FUNCTION_FEATURES
292 #undef NODE_INTERSECT
293