summaryrefslogtreecommitdiff
path: root/thirdparty/embree-aarch64/kernels/bvh/node_intersector_frustum.h
blob: 800ac8b47824e399333d28663857ad711de8d81d (plain)
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
// Copyright 2009-2020 Intel Corporation
// SPDX-License-Identifier: Apache-2.0

#pragma once

#include "node_intersector.h"

namespace embree
{
  namespace isa
  {
    //////////////////////////////////////////////////////////////////////////////////////
    // Frustum structure used in hybrid and stream traversal
    //////////////////////////////////////////////////////////////////////////////////////

    /*
       Optimized frustum test. We calculate t=(p-org)/dir in ray/box
       intersection. We assume the rays are split by octant, thus
       dir intervals are either positive or negative in each
       dimension.

       Case 1: dir.min >= 0 && dir.max >= 0:
         t_min = (p_min - org_max) / dir_max = (p_min - org_max)*rdir_min = p_min*rdir_min - org_max*rdir_min
         t_max = (p_max - org_min) / dir_min = (p_max - org_min)*rdir_max = p_max*rdir_max - org_min*rdir_max

       Case 2: dir.min < 0 && dir.max < 0:
         t_min = (p_max - org_min) / dir_min = (p_max - org_min)*rdir_max = p_max*rdir_max - org_min*rdir_max
         t_max = (p_min - org_max) / dir_max = (p_min - org_max)*rdir_min = p_min*rdir_min - org_max*rdir_min
    */

    template<bool robust>
    struct Frustum;
    
    /* Fast variant */
    template<>
    struct Frustum<false>
    {
      __forceinline Frustum() {}

      template<int K>
      __forceinline Frustum(const vbool<K>& valid, const Vec3vf<K>& org, const Vec3vf<K>& rdir, const vfloat<K>& ray_tnear, const vfloat<K>& ray_tfar, int N)
      {
        init(valid, org, rdir, ray_tnear, ray_tfar, N);
      }

      template<int K>
      __forceinline void init(const vbool<K>& valid, const Vec3vf<K>& org, const Vec3vf<K>& rdir, const vfloat<K>& ray_tnear, const vfloat<K>& ray_tfar, int N)
      {
        const Vec3fa reduced_min_org(reduce_min(select(valid, org.x, pos_inf)),
                                     reduce_min(select(valid, org.y, pos_inf)),
                                     reduce_min(select(valid, org.z, pos_inf)));

        const Vec3fa reduced_max_org(reduce_max(select(valid, org.x, neg_inf)),
                                     reduce_max(select(valid, org.y, neg_inf)),
                                     reduce_max(select(valid, org.z, neg_inf)));

        const Vec3fa reduced_min_rdir(reduce_min(select(valid, rdir.x, pos_inf)),
                                      reduce_min(select(valid, rdir.y, pos_inf)),
                                      reduce_min(select(valid, rdir.z, pos_inf)));

        const Vec3fa reduced_max_rdir(reduce_max(select(valid, rdir.x, neg_inf)),
                                      reduce_max(select(valid, rdir.y, neg_inf)),
                                      reduce_max(select(valid, rdir.z, neg_inf)));

        const float reduced_min_dist = reduce_min(select(valid, ray_tnear, vfloat<K>(pos_inf)));
        const float reduced_max_dist = reduce_max(select(valid, ray_tfar , vfloat<K>(neg_inf)));

        init(reduced_min_org, reduced_max_org, reduced_min_rdir, reduced_max_rdir, reduced_min_dist, reduced_max_dist, N);
      }

      __forceinline void init(const Vec3fa& reduced_min_org,
                              const Vec3fa& reduced_max_org,
                              const Vec3fa& reduced_min_rdir,
                              const Vec3fa& reduced_max_rdir,
                              float reduced_min_dist,
                              float reduced_max_dist,
                              int N)
      {
        const Vec3ba pos_rdir = ge_mask(reduced_min_rdir, Vec3fa(zero));

        min_rdir = select(pos_rdir, reduced_min_rdir, reduced_max_rdir);
        max_rdir = select(pos_rdir, reduced_max_rdir, reduced_min_rdir);

#if defined (__aarch64__)
        neg_min_org_rdir = -(min_rdir * select(pos_rdir, reduced_max_org, reduced_min_org));
        neg_max_org_rdir = -(max_rdir * select(pos_rdir, reduced_min_org, reduced_max_org));
#else
        min_org_rdir = min_rdir * select(pos_rdir, reduced_max_org, reduced_min_org);
        max_org_rdir = max_rdir * select(pos_rdir, reduced_min_org, reduced_max_org);
#endif
        min_dist = reduced_min_dist;
        max_dist = reduced_max_dist;

        nf = NearFarPrecalculations(min_rdir, N);
      }

      template<int K>
      __forceinline void updateMaxDist(const vfloat<K>& ray_tfar)
      {
        max_dist = reduce_max(ray_tfar);
      }

      NearFarPrecalculations nf;

      Vec3fa min_rdir;
      Vec3fa max_rdir;

#if defined (__aarch64__)
      Vec3fa neg_min_org_rdir;
      Vec3fa neg_max_org_rdir;
#else
      Vec3fa min_org_rdir;
      Vec3fa max_org_rdir;
#endif
      float min_dist;
      float max_dist;
    };

    typedef Frustum<false> FrustumFast;

    /* Robust variant */
    template<>
    struct Frustum<true>
    {
      __forceinline Frustum() {}

      template<int K>
      __forceinline Frustum(const vbool<K>& valid, const Vec3vf<K>& org, const Vec3vf<K>& rdir, const vfloat<K>& ray_tnear, const vfloat<K>& ray_tfar, int N)
      {
        init(valid, org, rdir, ray_tnear, ray_tfar, N);
      }

      template<int K>
      __forceinline void init(const vbool<K>& valid, const Vec3vf<K>& org, const Vec3vf<K>& rdir, const vfloat<K>& ray_tnear, const vfloat<K>& ray_tfar, int N)
      {
        const Vec3fa reduced_min_org(reduce_min(select(valid, org.x, pos_inf)),
                                     reduce_min(select(valid, org.y, pos_inf)),
                                     reduce_min(select(valid, org.z, pos_inf)));

        const Vec3fa reduced_max_org(reduce_max(select(valid, org.x, neg_inf)),
                                     reduce_max(select(valid, org.y, neg_inf)),
                                     reduce_max(select(valid, org.z, neg_inf)));

        const Vec3fa reduced_min_rdir(reduce_min(select(valid, rdir.x, pos_inf)),
                                      reduce_min(select(valid, rdir.y, pos_inf)),
                                      reduce_min(select(valid, rdir.z, pos_inf)));

        const Vec3fa reduced_max_rdir(reduce_max(select(valid, rdir.x, neg_inf)),
                                      reduce_max(select(valid, rdir.y, neg_inf)),
                                      reduce_max(select(valid, rdir.z, neg_inf)));

        const float reduced_min_dist = reduce_min(select(valid, ray_tnear, vfloat<K>(pos_inf)));
        const float reduced_max_dist = reduce_max(select(valid, ray_tfar , vfloat<K>(neg_inf)));

        init(reduced_min_org, reduced_max_org, reduced_min_rdir, reduced_max_rdir, reduced_min_dist, reduced_max_dist, N);
      }

      __forceinline void init(const Vec3fa& reduced_min_org,
                              const Vec3fa& reduced_max_org,
                              const Vec3fa& reduced_min_rdir,
                              const Vec3fa& reduced_max_rdir,
                              float reduced_min_dist,
                              float reduced_max_dist,
                              int N)
      {
        const Vec3ba pos_rdir = ge_mask(reduced_min_rdir, Vec3fa(zero));
        min_rdir = select(pos_rdir, reduced_min_rdir, reduced_max_rdir);
        max_rdir = select(pos_rdir, reduced_max_rdir, reduced_min_rdir);

        min_org = select(pos_rdir, reduced_max_org, reduced_min_org);
        max_org = select(pos_rdir, reduced_min_org, reduced_max_org);

        min_dist = reduced_min_dist;
        max_dist = reduced_max_dist;

        nf = NearFarPrecalculations(min_rdir, N);
      }

      template<int K>
      __forceinline void updateMaxDist(const vfloat<K>& ray_tfar)
      {
        max_dist = reduce_max(ray_tfar);
      }

      NearFarPrecalculations nf;

      Vec3fa min_rdir;
      Vec3fa max_rdir;

      Vec3fa min_org;
      Vec3fa max_org;

      float min_dist;
      float max_dist;
    };

    typedef Frustum<true> FrustumRobust;

    //////////////////////////////////////////////////////////////////////////////////////
    // Fast AABBNode intersection
    //////////////////////////////////////////////////////////////////////////////////////

    template<int N, int Nx>
    __forceinline size_t intersectNodeFrustum(const typename BVHN<N>::AABBNode* __restrict__ node,
                                       const FrustumFast& frustum, vfloat<Nx>& dist)
    {
      const vfloat<Nx> bminX = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.nearX);
      const vfloat<Nx> bminY = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.nearY);
      const vfloat<Nx> bminZ = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.nearZ);
      const vfloat<Nx> bmaxX = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.farX);
      const vfloat<Nx> bmaxY = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.farY);
      const vfloat<Nx> bmaxZ = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.farZ);

#if defined (__aarch64__)
      const vfloat<Nx> fminX = madd(bminX, vfloat<Nx>(frustum.min_rdir.x), vfloat<Nx>(frustum.neg_min_org_rdir.x));
      const vfloat<Nx> fminY = madd(bminY, vfloat<Nx>(frustum.min_rdir.y), vfloat<Nx>(frustum.neg_min_org_rdir.y));
      const vfloat<Nx> fminZ = madd(bminZ, vfloat<Nx>(frustum.min_rdir.z), vfloat<Nx>(frustum.neg_min_org_rdir.z));
      const vfloat<Nx> fmaxX = madd(bmaxX, vfloat<Nx>(frustum.max_rdir.x), vfloat<Nx>(frustum.neg_max_org_rdir.x));
      const vfloat<Nx> fmaxY = madd(bmaxY, vfloat<Nx>(frustum.max_rdir.y), vfloat<Nx>(frustum.neg_max_org_rdir.y));
      const vfloat<Nx> fmaxZ = madd(bmaxZ, vfloat<Nx>(frustum.max_rdir.z), vfloat<Nx>(frustum.neg_max_org_rdir.z));
#else
      const vfloat<Nx> fminX = msub(bminX, vfloat<Nx>(frustum.min_rdir.x), vfloat<Nx>(frustum.min_org_rdir.x));
      const vfloat<Nx> fminY = msub(bminY, vfloat<Nx>(frustum.min_rdir.y), vfloat<Nx>(frustum.min_org_rdir.y));
      const vfloat<Nx> fminZ = msub(bminZ, vfloat<Nx>(frustum.min_rdir.z), vfloat<Nx>(frustum.min_org_rdir.z));
      const vfloat<Nx> fmaxX = msub(bmaxX, vfloat<Nx>(frustum.max_rdir.x), vfloat<Nx>(frustum.max_org_rdir.x));
      const vfloat<Nx> fmaxY = msub(bmaxY, vfloat<Nx>(frustum.max_rdir.y), vfloat<Nx>(frustum.max_org_rdir.y));
      const vfloat<Nx> fmaxZ = msub(bmaxZ, vfloat<Nx>(frustum.max_rdir.z), vfloat<Nx>(frustum.max_org_rdir.z));
#endif
      const vfloat<Nx> fmin  = maxi(fminX, fminY, fminZ, vfloat<Nx>(frustum.min_dist));
      dist = fmin;
      const vfloat<Nx> fmax  = mini(fmaxX, fmaxY, fmaxZ, vfloat<Nx>(frustum.max_dist));
      const vbool<Nx> vmask_node_hit = fmin <= fmax;
      size_t m_node = movemask(vmask_node_hit) & (((size_t)1 << N)-1);
      return m_node;
    }

    //////////////////////////////////////////////////////////////////////////////////////
    // Robust AABBNode intersection
    //////////////////////////////////////////////////////////////////////////////////////

    template<int N, int Nx>
    __forceinline size_t intersectNodeFrustum(const typename BVHN<N>::AABBNode* __restrict__ node,
                                       const FrustumRobust& frustum, vfloat<Nx>& dist)
    {
      const vfloat<Nx> bminX = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.nearX);
      const vfloat<Nx> bminY = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.nearY);
      const vfloat<Nx> bminZ = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.nearZ);
      const vfloat<Nx> bmaxX = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.farX);
      const vfloat<Nx> bmaxY = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.farY);
      const vfloat<Nx> bmaxZ = *(const vfloat<N>*)((const char*)&node->lower_x + frustum.nf.farZ);

      const vfloat<Nx> fminX = (bminX - vfloat<Nx>(frustum.min_org.x)) * vfloat<Nx>(frustum.min_rdir.x);
      const vfloat<Nx> fminY = (bminY - vfloat<Nx>(frustum.min_org.y)) * vfloat<Nx>(frustum.min_rdir.y);
      const vfloat<Nx> fminZ = (bminZ - vfloat<Nx>(frustum.min_org.z)) * vfloat<Nx>(frustum.min_rdir.z);
      const vfloat<Nx> fmaxX = (bmaxX - vfloat<Nx>(frustum.max_org.x)) * vfloat<Nx>(frustum.max_rdir.x);
      const vfloat<Nx> fmaxY = (bmaxY - vfloat<Nx>(frustum.max_org.y)) * vfloat<Nx>(frustum.max_rdir.y);
      const vfloat<Nx> fmaxZ = (bmaxZ - vfloat<Nx>(frustum.max_org.z)) * vfloat<Nx>(frustum.max_rdir.z);

      const float round_down = 1.0f-2.0f*float(ulp); // FIXME: use per instruction rounding for AVX512
      const float round_up   = 1.0f+2.0f*float(ulp);
      const vfloat<Nx> fmin  = max(fminX, fminY, fminZ, vfloat<Nx>(frustum.min_dist));
      dist = fmin;
      const vfloat<Nx> fmax  = min(fmaxX, fmaxY, fmaxZ, vfloat<Nx>(frustum.max_dist));
      const vbool<Nx> vmask_node_hit = (round_down*fmin <= round_up*fmax);
      size_t m_node = movemask(vmask_node_hit) & (((size_t)1 << N)-1);
      return m_node;
    }
  }
}