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
|
// Copyright 2009-2021 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
#pragma once
#include "node_intersector.h"
namespace embree
{
namespace isa
{
//////////////////////////////////////////////////////////////////////////////////////
// Ray packet structure used in stream traversal
//////////////////////////////////////////////////////////////////////////////////////
template<int K, bool robust>
struct TravRayKStream;
/* Fast variant */
template<int K>
struct TravRayKStream<K, false>
{
__forceinline TravRayKStream() {}
__forceinline TravRayKStream(const Vec3vf<K>& ray_org, const Vec3vf<K>& ray_dir, const vfloat<K>& ray_tnear, const vfloat<K>& ray_tfar)
{
init(ray_org, ray_dir);
tnear = ray_tnear;
tfar = ray_tfar;
}
__forceinline void init(const Vec3vf<K>& ray_org, const Vec3vf<K>& ray_dir)
{
rdir = rcp_safe(ray_dir);
#if defined(__aarch64__)
neg_org_rdir = -(ray_org * rdir);
#else
org_rdir = ray_org * rdir;
#endif
}
Vec3vf<K> rdir;
#if defined(__aarch64__)
Vec3vf<K> neg_org_rdir;
#else
Vec3vf<K> org_rdir;
#endif
vfloat<K> tnear;
vfloat<K> tfar;
};
template<int K>
using TravRayKStreamFast = TravRayKStream<K, false>;
/* Robust variant */
template<int K>
struct TravRayKStream<K, true>
{
__forceinline TravRayKStream() {}
__forceinline TravRayKStream(const Vec3vf<K>& ray_org, const Vec3vf<K>& ray_dir, const vfloat<K>& ray_tnear, const vfloat<K>& ray_tfar)
{
init(ray_org, ray_dir);
tnear = ray_tnear;
tfar = ray_tfar;
}
__forceinline void init(const Vec3vf<K>& ray_org, const Vec3vf<K>& ray_dir)
{
rdir = vfloat<K>(1.0f)/(zero_fix(ray_dir));
org = ray_org;
}
Vec3vf<K> rdir;
Vec3vf<K> org;
vfloat<K> tnear;
vfloat<K> tfar;
};
template<int K>
using TravRayKStreamRobust = TravRayKStream<K, true>;
//////////////////////////////////////////////////////////////////////////////////////
// Fast AABBNode intersection
//////////////////////////////////////////////////////////////////////////////////////
template<int N, int K>
__forceinline size_t intersectNode1(const typename BVHN<N>::AABBNode* __restrict__ node,
const TravRayKStreamFast<K>& ray, size_t k, const NearFarPrecalculations& nf)
{
const vfloat<N> bminX = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.nearX));
const vfloat<N> bminY = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.nearY));
const vfloat<N> bminZ = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.nearZ));
const vfloat<N> bmaxX = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.farX));
const vfloat<N> bmaxY = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.farY));
const vfloat<N> bmaxZ = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.farZ));
#if defined (__aarch64__)
const vfloat<N> rminX = madd(bminX, vfloat<N>(ray.rdir.x[k]), vfloat<N>(ray.neg_org_rdir.x[k]));
const vfloat<N> rminY = madd(bminY, vfloat<N>(ray.rdir.y[k]), vfloat<N>(ray.neg_org_rdir.y[k]));
const vfloat<N> rminZ = madd(bminZ, vfloat<N>(ray.rdir.z[k]), vfloat<N>(ray.neg_org_rdir.z[k]));
const vfloat<N> rmaxX = madd(bmaxX, vfloat<N>(ray.rdir.x[k]), vfloat<N>(ray.neg_org_rdir.x[k]));
const vfloat<N> rmaxY = madd(bmaxY, vfloat<N>(ray.rdir.y[k]), vfloat<N>(ray.neg_org_rdir.y[k]));
const vfloat<N> rmaxZ = madd(bmaxZ, vfloat<N>(ray.rdir.z[k]), vfloat<N>(ray.neg_org_rdir.z[k]));
#else
const vfloat<N> rminX = msub(bminX, vfloat<N>(ray.rdir.x[k]), vfloat<N>(ray.org_rdir.x[k]));
const vfloat<N> rminY = msub(bminY, vfloat<N>(ray.rdir.y[k]), vfloat<N>(ray.org_rdir.y[k]));
const vfloat<N> rminZ = msub(bminZ, vfloat<N>(ray.rdir.z[k]), vfloat<N>(ray.org_rdir.z[k]));
const vfloat<N> rmaxX = msub(bmaxX, vfloat<N>(ray.rdir.x[k]), vfloat<N>(ray.org_rdir.x[k]));
const vfloat<N> rmaxY = msub(bmaxY, vfloat<N>(ray.rdir.y[k]), vfloat<N>(ray.org_rdir.y[k]));
const vfloat<N> rmaxZ = msub(bmaxZ, vfloat<N>(ray.rdir.z[k]), vfloat<N>(ray.org_rdir.z[k]));
#endif
const vfloat<N> rmin = maxi(rminX, rminY, rminZ, vfloat<N>(ray.tnear[k]));
const vfloat<N> rmax = mini(rmaxX, rmaxY, rmaxZ, vfloat<N>(ray.tfar[k]));
const vbool<N> vmask_first_hit = rmin <= rmax;
return movemask(vmask_first_hit) & (((size_t)1 << N)-1);
}
template<int N, int K>
__forceinline size_t intersectNodeK(const typename BVHN<N>::AABBNode* __restrict__ node, size_t i,
const TravRayKStreamFast<K>& ray, const NearFarPrecalculations& nf)
{
char* ptr = (char*)&node->lower_x + i*sizeof(float);
const vfloat<K> bminX = *(const float*)(ptr + nf.nearX);
const vfloat<K> bminY = *(const float*)(ptr + nf.nearY);
const vfloat<K> bminZ = *(const float*)(ptr + nf.nearZ);
const vfloat<K> bmaxX = *(const float*)(ptr + nf.farX);
const vfloat<K> bmaxY = *(const float*)(ptr + nf.farY);
const vfloat<K> bmaxZ = *(const float*)(ptr + nf.farZ);
#if defined (__aarch64__)
const vfloat<K> rminX = madd(bminX, ray.rdir.x, ray.neg_org_rdir.x);
const vfloat<K> rminY = madd(bminY, ray.rdir.y, ray.neg_org_rdir.y);
const vfloat<K> rminZ = madd(bminZ, ray.rdir.z, ray.neg_org_rdir.z);
const vfloat<K> rmaxX = madd(bmaxX, ray.rdir.x, ray.neg_org_rdir.x);
const vfloat<K> rmaxY = madd(bmaxY, ray.rdir.y, ray.neg_org_rdir.y);
const vfloat<K> rmaxZ = madd(bmaxZ, ray.rdir.z, ray.neg_org_rdir.z);
#else
const vfloat<K> rminX = msub(bminX, ray.rdir.x, ray.org_rdir.x);
const vfloat<K> rminY = msub(bminY, ray.rdir.y, ray.org_rdir.y);
const vfloat<K> rminZ = msub(bminZ, ray.rdir.z, ray.org_rdir.z);
const vfloat<K> rmaxX = msub(bmaxX, ray.rdir.x, ray.org_rdir.x);
const vfloat<K> rmaxY = msub(bmaxY, ray.rdir.y, ray.org_rdir.y);
const vfloat<K> rmaxZ = msub(bmaxZ, ray.rdir.z, ray.org_rdir.z);
#endif
const vfloat<K> rmin = maxi(rminX, rminY, rminZ, ray.tnear);
const vfloat<K> rmax = mini(rmaxX, rmaxY, rmaxZ, ray.tfar);
const vbool<K> vmask_first_hit = rmin <= rmax;
return movemask(vmask_first_hit);
}
//////////////////////////////////////////////////////////////////////////////////////
// Robust AABBNode intersection
//////////////////////////////////////////////////////////////////////////////////////
template<int N, int K>
__forceinline size_t intersectNode1(const typename BVHN<N>::AABBNode* __restrict__ node,
const TravRayKStreamRobust<K>& ray, size_t k, const NearFarPrecalculations& nf)
{
const vfloat<N> bminX = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.nearX));
const vfloat<N> bminY = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.nearY));
const vfloat<N> bminZ = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.nearZ));
const vfloat<N> bmaxX = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.farX));
const vfloat<N> bmaxY = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.farY));
const vfloat<N> bmaxZ = vfloat<N>(*(const vfloat<N>*)((const char*)&node->lower_x + nf.farZ));
const vfloat<N> rminX = (bminX - vfloat<N>(ray.org.x[k])) * vfloat<N>(ray.rdir.x[k]);
const vfloat<N> rminY = (bminY - vfloat<N>(ray.org.y[k])) * vfloat<N>(ray.rdir.y[k]);
const vfloat<N> rminZ = (bminZ - vfloat<N>(ray.org.z[k])) * vfloat<N>(ray.rdir.z[k]);
const vfloat<N> rmaxX = (bmaxX - vfloat<N>(ray.org.x[k])) * vfloat<N>(ray.rdir.x[k]);
const vfloat<N> rmaxY = (bmaxY - vfloat<N>(ray.org.y[k])) * vfloat<N>(ray.rdir.y[k]);
const vfloat<N> rmaxZ = (bmaxZ - vfloat<N>(ray.org.z[k])) * vfloat<N>(ray.rdir.z[k]);
const float round_up = 1.0f+3.0f*float(ulp); // FIXME: use per instruction rounding for AVX512
const vfloat<N> rmin = max(rminX, rminY, rminZ, vfloat<N>(ray.tnear[k]));
const vfloat<N> rmax = round_up *min(rmaxX, rmaxY, rmaxZ, vfloat<N>(ray.tfar[k]));
const vbool<N> vmask_first_hit = rmin <= rmax;
return movemask(vmask_first_hit) & (((size_t)1 << N)-1);
}
template<int N, int K>
__forceinline size_t intersectNodeK(const typename BVHN<N>::AABBNode* __restrict__ node, size_t i,
const TravRayKStreamRobust<K>& ray, const NearFarPrecalculations& nf)
{
char *ptr = (char*)&node->lower_x + i*sizeof(float);
const vfloat<K> bminX = *(const float*)(ptr + nf.nearX);
const vfloat<K> bminY = *(const float*)(ptr + nf.nearY);
const vfloat<K> bminZ = *(const float*)(ptr + nf.nearZ);
const vfloat<K> bmaxX = *(const float*)(ptr + nf.farX);
const vfloat<K> bmaxY = *(const float*)(ptr + nf.farY);
const vfloat<K> bmaxZ = *(const float*)(ptr + nf.farZ);
const vfloat<K> rminX = (bminX - ray.org.x) * ray.rdir.x;
const vfloat<K> rminY = (bminY - ray.org.y) * ray.rdir.y;
const vfloat<K> rminZ = (bminZ - ray.org.z) * ray.rdir.z;
const vfloat<K> rmaxX = (bmaxX - ray.org.x) * ray.rdir.x;
const vfloat<K> rmaxY = (bmaxY - ray.org.y) * ray.rdir.y;
const vfloat<K> rmaxZ = (bmaxZ - ray.org.z) * ray.rdir.z;
const float round_up = 1.0f+3.0f*float(ulp);
const vfloat<K> rmin = max(rminX, rminY, rminZ, vfloat<K>(ray.tnear));
const vfloat<K> rmax = round_up * min(rmaxX, rmaxY, rmaxZ, vfloat<K>(ray.tfar));
const vbool<K> vmask_first_hit = rmin <= rmax;
return movemask(vmask_first_hit);
}
}
}
|