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
|
/*
* 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.
*/
/* This is a template BVH traversal function for subsurface scattering, 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_MOTION: motion blur rendering
*
*/
#define FEATURE(f) (((BVH_FUNCTION_FEATURES) & (f)) != 0)
ccl_device uint BVH_FUNCTION_NAME(KernelGlobals *kg, const Ray *ray, Intersection *isect_array,
int subsurface_object, uint *lcg_state, int max_hits)
{
/* todo:
* - test if pushing distance on the stack helps (for non shadow rays)
* - separate version for shadow rays
* - likely and unlikely for if() statements
* - SSE for hair
* - test restrict attribute for pointers
*/
/* traversal stack in CUDA thread-local memory */
int traversalStack[BVH_STACK_SIZE];
traversalStack[0] = ENTRYPOINT_SENTINEL;
/* traversal variables in registers */
int stackPtr = 0;
int nodeAddr = 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;
float isect_t = ray->t;
const uint visibility = PATH_RAY_ALL_VISIBILITY;
uint num_hits = 0;
#if FEATURE(BVH_MOTION)
Transform ob_tfm;
#endif
#if defined(__KERNEL_SSE2__)
const shuffle_swap_t shuf_identity = shuffle_swap_identity();
const shuffle_swap_t shuf_swap = shuffle_swap_swap();
const __m128 pn = _mm_castsi128_ps(_mm_set_epi32(0x80000000, 0x80000000, 0, 0));
__m128 Psplat[3], idirsplat[3];
shuffle_swap_t shufflexyz[3];
Psplat[0] = _mm_set_ps1(P.x);
Psplat[1] = _mm_set_ps1(P.y);
Psplat[2] = _mm_set_ps1(P.z);
__m128 tsplat = _mm_set_ps(-isect_t, -isect_t, 0.0f, 0.0f);
gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
#endif
/* traversal loop */
do {
do
{
/* traverse internal nodes */
while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL)
{
bool traverseChild0, traverseChild1;
int nodeAddrChild1;
#if !defined(__KERNEL_SSE2__)
/* Intersect two child bounding boxes, non-SSE version */
float t = isect_t;
/* fetch node data */
float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+0);
float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+1);
float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+2);
float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+3);
/* intersect ray against child nodes */
NO_EXTENDED_PRECISION float c0lox = (node0.x - P.x) * idir.x;
NO_EXTENDED_PRECISION float c0hix = (node0.z - P.x) * idir.x;
NO_EXTENDED_PRECISION float c0loy = (node1.x - P.y) * idir.y;
NO_EXTENDED_PRECISION float c0hiy = (node1.z - P.y) * idir.y;
NO_EXTENDED_PRECISION float c0loz = (node2.x - P.z) * idir.z;
NO_EXTENDED_PRECISION float c0hiz = (node2.z - P.z) * idir.z;
NO_EXTENDED_PRECISION float c0min = max4(min(c0lox, c0hix), min(c0loy, c0hiy), min(c0loz, c0hiz), 0.0f);
NO_EXTENDED_PRECISION float c0max = min4(max(c0lox, c0hix), max(c0loy, c0hiy), max(c0loz, c0hiz), t);
NO_EXTENDED_PRECISION float c1lox = (node0.y - P.x) * idir.x;
NO_EXTENDED_PRECISION float c1hix = (node0.w - P.x) * idir.x;
NO_EXTENDED_PRECISION float c1loy = (node1.y - P.y) * idir.y;
NO_EXTENDED_PRECISION float c1hiy = (node1.w - P.y) * idir.y;
NO_EXTENDED_PRECISION float c1loz = (node2.y - P.z) * idir.z;
NO_EXTENDED_PRECISION float c1hiz = (node2.w - P.z) * idir.z;
NO_EXTENDED_PRECISION float c1min = max4(min(c1lox, c1hix), min(c1loy, c1hiy), min(c1loz, c1hiz), 0.0f);
NO_EXTENDED_PRECISION float c1max = min4(max(c1lox, c1hix), max(c1loy, c1hiy), max(c1loz, c1hiz), t);
/* decide which nodes to traverse next */
#ifdef __VISIBILITY_FLAG__
/* this visibility test gives a 5% performance hit, how to solve? */
traverseChild0 = (c0max >= c0min) && (__float_as_uint(cnodes.z) & visibility);
traverseChild1 = (c1max >= c1min) && (__float_as_uint(cnodes.w) & visibility);
#else
traverseChild0 = (c0max >= c0min);
traverseChild1 = (c1max >= c1min);
#endif
#else // __KERNEL_SSE2__
/* Intersect two child bounding boxes, SSE3 version adapted from Embree */
/* fetch node data */
const __m128 *bvh_nodes = (__m128*)kg->__bvh_nodes.data + nodeAddr*BVH_NODE_SIZE;
const float4 cnodes = ((float4*)bvh_nodes)[3];
/* intersect ray against child nodes */
const __m128 tminmaxx = _mm_mul_ps(_mm_sub_ps(shuffle_swap(bvh_nodes[0], shufflexyz[0]), Psplat[0]), idirsplat[0]);
const __m128 tminmaxy = _mm_mul_ps(_mm_sub_ps(shuffle_swap(bvh_nodes[1], shufflexyz[1]), Psplat[1]), idirsplat[1]);
const __m128 tminmaxz = _mm_mul_ps(_mm_sub_ps(shuffle_swap(bvh_nodes[2], shufflexyz[2]), Psplat[2]), idirsplat[2]);
const __m128 tminmax = _mm_xor_ps(_mm_max_ps(_mm_max_ps(tminmaxx, tminmaxy), _mm_max_ps(tminmaxz, tsplat)), pn);
const __m128 lrhit = _mm_cmple_ps(tminmax, shuffle<2, 3, 0, 1>(tminmax));
/* decide which nodes to traverse next */
#ifdef __VISIBILITY_FLAG__
/* this visibility test gives a 5% performance hit, how to solve? */
traverseChild0 = (_mm_movemask_ps(lrhit) & 1) && (__float_as_uint(cnodes.z) & visibility);
traverseChild1 = (_mm_movemask_ps(lrhit) & 2) && (__float_as_uint(cnodes.w) & visibility);
#else
traverseChild0 = (_mm_movemask_ps(lrhit) & 1);
traverseChild1 = (_mm_movemask_ps(lrhit) & 2);
#endif
#endif // __KERNEL_SSE2__
nodeAddr = __float_as_int(cnodes.x);
nodeAddrChild1 = __float_as_int(cnodes.y);
if(traverseChild0 && traverseChild1) {
/* both children were intersected, push the farther one */
#if !defined(__KERNEL_SSE2__)
bool closestChild1 = (c1min < c0min);
#else
union { __m128 m128; float v[4]; } uminmax;
uminmax.m128 = tminmax;
bool closestChild1 = uminmax.v[1] < uminmax.v[0];
#endif
if(closestChild1) {
int tmp = nodeAddr;
nodeAddr = nodeAddrChild1;
nodeAddrChild1 = tmp;
}
++stackPtr;
traversalStack[stackPtr] = nodeAddrChild1;
}
else {
/* one child was intersected */
if(traverseChild1) {
nodeAddr = nodeAddrChild1;
}
else if(!traverseChild0) {
/* neither child was intersected */
nodeAddr = traversalStack[stackPtr];
--stackPtr;
}
}
}
/* if node is leaf, fetch triangle list */
if(nodeAddr < 0) {
float4 leaf = kernel_tex_fetch(__bvh_nodes, (-nodeAddr-1)*BVH_NODE_SIZE+(BVH_NODE_SIZE-1));
int primAddr = __float_as_int(leaf.x);
#if FEATURE(BVH_INSTANCING)
if(primAddr >= 0) {
#endif
int primAddr2 = __float_as_int(leaf.y);
/* pop */
nodeAddr = traversalStack[stackPtr];
--stackPtr;
/* primitive intersection */
for(; primAddr < primAddr2; primAddr++) {
/* only primitives from the same object */
uint tri_object = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, primAddr): object;
if(tri_object != subsurface_object)
continue;
/* intersect ray against primitive */
uint type = kernel_tex_fetch(__prim_type, primAddr);
switch(type & PRIMITIVE_ALL) {
case PRIMITIVE_TRIANGLE: {
triangle_intersect_subsurface(kg, isect_array, P, dir, object, primAddr, isect_t, &num_hits, lcg_state, max_hits);
break;
}
#if FEATURE(BVH_MOTION)
case PRIMITIVE_MOTION_TRIANGLE: {
motion_triangle_intersect_subsurface(kg, isect_array, P, dir, ray->time, object, primAddr, isect_t, &num_hits, lcg_state, max_hits);
break;
}
#endif
default: {
break;
}
}
}
}
#if FEATURE(BVH_INSTANCING)
else {
/* instance push */
if(subsurface_object == kernel_tex_fetch(__prim_object, -primAddr-1)) {
object = subsurface_object;
#if FEATURE(BVH_MOTION)
bvh_instance_motion_push(kg, object, ray, &P, &dir, &idir, &isect_t, &ob_tfm);
#else
bvh_instance_push(kg, object, ray, &P, &dir, &idir, &isect_t);
#endif
#if defined(__KERNEL_SSE2__)
Psplat[0] = _mm_set_ps1(P.x);
Psplat[1] = _mm_set_ps1(P.y);
Psplat[2] = _mm_set_ps1(P.z);
tsplat = _mm_set_ps(-isect_t, -isect_t, 0.0f, 0.0f);
gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
#endif
++stackPtr;
traversalStack[stackPtr] = ENTRYPOINT_SENTINEL;
nodeAddr = kernel_tex_fetch(__object_node, object);
}
else {
/* pop */
nodeAddr = traversalStack[stackPtr];
--stackPtr;
}
}
}
#endif
} while(nodeAddr != ENTRYPOINT_SENTINEL);
#if FEATURE(BVH_INSTANCING)
if(stackPtr >= 0) {
kernel_assert(object != OBJECT_NONE);
/* instance pop */
#if FEATURE(BVH_MOTION)
bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, &isect_t, &ob_tfm);
#else
bvh_instance_pop(kg, object, ray, &P, &dir, &idir, &isect_t);
#endif
#if defined(__KERNEL_SSE2__)
Psplat[0] = _mm_set_ps1(P.x);
Psplat[1] = _mm_set_ps1(P.y);
Psplat[2] = _mm_set_ps1(P.z);
tsplat = _mm_set_ps(-isect_t, -isect_t, 0.0f, 0.0f);
gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
#endif
object = OBJECT_NONE;
nodeAddr = traversalStack[stackPtr];
--stackPtr;
}
#endif
} while(nodeAddr != ENTRYPOINT_SENTINEL);
return num_hits;
}
#undef FEATURE
#undef BVH_FUNCTION_NAME
#undef BVH_FUNCTION_FEATURES
|