1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
|
/*
* Adapted from code Copyright 2009-2010 NVIDIA Corporation,
* and code copyright 2009-2012 Intel Corporation
*
* Modifications Copyright 2011-2014, 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.
*/
/* 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
*
*/
ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(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
)
{
/* TODO(sergey):
* - 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. */
QBVHStackItem traversalStack[BVH_QSTACK_SIZE];
traversalStack[0].addr = ENTRYPOINT_SENTINEL;
traversalStack[0].dist = -FLT_MAX;
/* Traversal variables in registers. */
int stackPtr = 0;
int nodeAddr = kernel_data.bvh.root;
float nodeDist = -FLT_MAX;
/* 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
#ifndef __KERNEL_SSE41__
if(!isfinite(P.x)) {
return false;
}
#endif
isect->t = ray->t;
isect->u = 0.0f;
isect->v = 0.0f;
isect->prim = PRIM_NONE;
isect->object = OBJECT_NONE;
#if defined(__KERNEL_DEBUG__)
isect->num_traversal_steps = 0;
isect->num_traversed_instances = 0;
#endif
ssef tnear(0.0f), tfar(ray->t);
sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z));
#ifdef __KERNEL_AVX2__
float3 P_idir = P*idir;
sse3f P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
#else
sse3f org = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
#endif
/* Offsets to select the side that becomes the lower or upper bound. */
int near_x, near_y, near_z;
int far_x, far_y, far_z;
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
IsectPrecalc isect_precalc;
triangle_intersect_precalc(dir, &isect_precalc);
/* Traversal loop. */
do {
do {
/* Traverse internal nodes. */
while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL) {
if(UNLIKELY(nodeDist > isect->t)) {
/* Pop. */
nodeAddr = traversalStack[stackPtr].addr;
nodeDist = traversalStack[stackPtr].dist;
--stackPtr;
continue;
}
int traverseChild;
ssef dist;
#if defined(__KERNEL_DEBUG__)
isect->num_traversal_steps++;
#endif
#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
if(difl != 0.0f) {
/* NOTE: We extend all the child BB instead of fetching
* and checking visibility flags for each of the,
*
* Need to test if doing opposite would be any faster.
*/
traverseChild = qbvh_node_intersect_robust(kg,
tnear,
tfar,
#ifdef __KERNEL_AVX2__
P_idir4,
#else
org,
#endif
idir4,
near_x, near_y, near_z,
far_x, far_y, far_z,
nodeAddr,
difl,
&dist);
}
else
#endif
{
traverseChild = qbvh_node_intersect(kg,
tnear,
tfar,
#ifdef __KERNEL_AVX2__
P_idir4,
#else
org,
#endif
idir4,
near_x, near_y, near_z,
far_x, far_y, far_z,
nodeAddr,
&dist);
}
if(traverseChild != 0) {
float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_QNODE_SIZE+6);
/* One child is hit, continue with that child. */
int r = __bscf(traverseChild);
float d0 = ((float*)&dist)[r];
if(traverseChild == 0) {
nodeAddr = __float_as_int(cnodes[r]);
nodeDist = d0;
continue;
}
/* Two children are hit, push far child, and continue with
* closer child.
*/
int c0 = __float_as_int(cnodes[r]);
r = __bscf(traverseChild);
int c1 = __float_as_int(cnodes[r]);
float d1 = ((float*)&dist)[r];
if(traverseChild == 0) {
if(d1 < d0) {
nodeAddr = c1;
nodeDist = d1;
++stackPtr;
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
traversalStack[stackPtr].addr = c0;
traversalStack[stackPtr].dist = d0;
continue;
}
else {
nodeAddr = c0;
nodeDist = d0;
++stackPtr;
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
traversalStack[stackPtr].addr = c1;
traversalStack[stackPtr].dist = d1;
continue;
}
}
/* Here starts the slow path for 3 or 4 hit children. We push
* all nodes onto the stack to sort them there.
*/
++stackPtr;
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
traversalStack[stackPtr].addr = c1;
traversalStack[stackPtr].dist = d1;
++stackPtr;
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
traversalStack[stackPtr].addr = c0;
traversalStack[stackPtr].dist = d0;
/* Three children are hit, push all onto stack and sort 3
* stack items, continue with closest child.
*/
r = __bscf(traverseChild);
int c2 = __float_as_int(cnodes[r]);
float d2 = ((float*)&dist)[r];
if(traverseChild == 0) {
++stackPtr;
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
traversalStack[stackPtr].addr = c2;
traversalStack[stackPtr].dist = d2;
qbvh_stack_sort(&traversalStack[stackPtr],
&traversalStack[stackPtr - 1],
&traversalStack[stackPtr - 2]);
nodeAddr = traversalStack[stackPtr].addr;
nodeDist = traversalStack[stackPtr].dist;
--stackPtr;
continue;
}
/* Four children are hit, push all onto stack and sort 4
* stack items, continue with closest child.
*/
r = __bscf(traverseChild);
int c3 = __float_as_int(cnodes[r]);
float d3 = ((float*)&dist)[r];
++stackPtr;
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
traversalStack[stackPtr].addr = c3;
traversalStack[stackPtr].dist = d3;
++stackPtr;
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
traversalStack[stackPtr].addr = c2;
traversalStack[stackPtr].dist = d2;
qbvh_stack_sort(&traversalStack[stackPtr],
&traversalStack[stackPtr - 1],
&traversalStack[stackPtr - 2],
&traversalStack[stackPtr - 3]);
}
nodeAddr = traversalStack[stackPtr].addr;
nodeDist = traversalStack[stackPtr].dist;
--stackPtr;
}
/* If node is leaf, fetch triangle list. */
if(nodeAddr < 0) {
float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_QNODE_LEAF_SIZE);
#ifdef __VISIBILITY_FLAG__
if(UNLIKELY((nodeDist > isect->t) || ((__float_as_uint(leaf.z) & visibility) == 0)))
#else
if(UNLIKELY((nodeDist > isect->t)))
#endif
{
/* Pop. */
nodeAddr = traversalStack[stackPtr].addr;
nodeDist = traversalStack[stackPtr].dist;
--stackPtr;
continue;
}
int primAddr = __float_as_int(leaf.x);
#if BVH_FEATURE(BVH_INSTANCING)
if(primAddr >= 0) {
#endif
int primAddr2 = __float_as_int(leaf.y);
const uint type = __float_as_int(leaf.w);
/* Pop. */
nodeAddr = traversalStack[stackPtr].addr;
nodeDist = traversalStack[stackPtr].dist;
--stackPtr;
/* Primitive intersection. */
switch(type & PRIMITIVE_ALL) {
case PRIMITIVE_TRIANGLE: {
for(; primAddr < primAddr2; primAddr++) {
#if defined(__KERNEL_DEBUG__)
isect->num_traversal_steps++;
#endif
kernel_assert(kernel_tex_fetch(__prim_type, primAddr) == type);
if(triangle_intersect(kg, &isect_precalc, isect, P, visibility, object, primAddr)) {
tfar = ssef(isect->t);
/* Shadow ray early termination. */
if(visibility == PATH_RAY_SHADOW_OPAQUE)
return true;
}
}
break;
}
#if BVH_FEATURE(BVH_MOTION)
case PRIMITIVE_MOTION_TRIANGLE: {
for(; primAddr < primAddr2; primAddr++) {
#if defined(__KERNEL_DEBUG__)
isect->num_traversal_steps++;
#endif
kernel_assert(kernel_tex_fetch(__prim_type, primAddr) == type);
if(motion_triangle_intersect(kg, isect, P, dir, ray->time, visibility, object, primAddr)) {
tfar = ssef(isect->t);
/* Shadow ray early termination. */
if(visibility == PATH_RAY_SHADOW_OPAQUE)
return true;
}
}
break;
}
#endif /* BVH_FEATURE(BVH_MOTION) */
#if BVH_FEATURE(BVH_HAIR)
case PRIMITIVE_CURVE:
case PRIMITIVE_MOTION_CURVE: {
for(; primAddr < primAddr2; primAddr++) {
#if defined(__KERNEL_DEBUG__)
isect->num_traversal_steps++;
#endif
kernel_assert(kernel_tex_fetch(__prim_type, primAddr) == type);
bool hit;
if(kernel_data.curve.curveflags & CURVE_KN_INTERPOLATE)
hit = bvh_cardinal_curve_intersect(kg, isect, P, dir, visibility, object, primAddr, ray->time, type, lcg_state, difl, extmax);
else
hit = bvh_curve_intersect(kg, isect, P, dir, visibility, object, primAddr, ray->time, type, lcg_state, difl, extmax);
if(hit) {
tfar = ssef(isect->t);
/* Shadow ray early termination. */
if(visibility == PATH_RAY_SHADOW_OPAQUE)
return true;
}
}
break;
}
#endif /* BVH_FEATURE(BVH_HAIR) */
}
}
#if BVH_FEATURE(BVH_INSTANCING)
else {
/* Instance push. */
object = kernel_tex_fetch(__prim_object, -primAddr-1);
#if BVH_FEATURE(BVH_MOTION)
qbvh_instance_motion_push(kg, object, ray, &P, &dir, &idir, &isect->t, &nodeDist, &ob_itfm);
#else
qbvh_instance_push(kg, object, ray, &P, &dir, &idir, &isect->t, &nodeDist);
#endif
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
tfar = ssef(isect->t);
idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z));
#ifdef __KERNEL_AVX2__
P_idir = P*idir;
P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
#else
org = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
#endif
triangle_intersect_precalc(dir, &isect_precalc);
++stackPtr;
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
traversalStack[stackPtr].addr = ENTRYPOINT_SENTINEL;
traversalStack[stackPtr].dist = -FLT_MAX;
nodeAddr = kernel_tex_fetch(__object_node, object);
#if defined(__KERNEL_DEBUG__)
isect->num_traversed_instances++;
#endif
}
}
#endif /* FEATURE(BVH_INSTANCING) */
} while(nodeAddr != ENTRYPOINT_SENTINEL);
#if BVH_FEATURE(BVH_INSTANCING)
if(stackPtr >= 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
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
tfar = ssef(isect->t);
idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z));
#ifdef __KERNEL_AVX2__
P_idir = P*idir;
P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
#else
org = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
#endif
triangle_intersect_precalc(dir, &isect_precalc);
object = OBJECT_NONE;
nodeAddr = traversalStack[stackPtr].addr;
nodeDist = traversalStack[stackPtr].dist;
--stackPtr;
}
#endif /* FEATURE(BVH_INSTANCING) */
} while(nodeAddr != ENTRYPOINT_SENTINEL);
return (isect->prim != PRIM_NONE);
}
|