summaryrefslogtreecommitdiff
path: root/thirdparty/embree-aarch64/kernels/subdiv/patch_eval_simd.h
blob: 28016d9e20d902ef55f392ea14de4c12a28a0f3b (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
// Copyright 2009-2020 Intel Corporation
// SPDX-License-Identifier: Apache-2.0

#pragma once

#include "patch.h"
#include "feature_adaptive_eval_simd.h"

namespace embree
{
  namespace isa
  {
    template<typename vbool, typename vint, typename vfloat, typename Vertex, typename Vertex_t = Vertex>
      struct PatchEvalSimd
      {
      public:
        
        typedef PatchT<Vertex,Vertex_t> Patch;
        typedef typename Patch::Ref Ref;
        typedef CatmullClarkPatchT<Vertex,Vertex_t> CatmullClarkPatch;

        PatchEvalSimd (SharedLazyTessellationCache::CacheEntry& entry, size_t commitCounter, 
                       const HalfEdge* edge, const char* vertices, size_t stride, const vbool& valid0, const vfloat& u, const vfloat& v, 
                       float* P, float* dPdu, float* dPdv, float* ddPdudu, float* ddPdvdv, float* ddPdudv, const size_t dstride, const size_t N)
        : P(P), dPdu(dPdu), dPdv(dPdv), ddPdudu(ddPdudu), ddPdvdv(ddPdvdv), ddPdudv(ddPdudv), dstride(dstride), N(N)
        {
          /* conservative time for the very first allocation */
          auto time = SharedLazyTessellationCache::sharedLazyTessellationCache.getTime(commitCounter);

          Ref patch = SharedLazyTessellationCache::lookup(entry,commitCounter,[&] () {
              auto alloc = [](size_t bytes) { return SharedLazyTessellationCache::malloc(bytes); };
              return Patch::create(alloc,edge,vertices,stride);
            }, true);

          auto curTime = SharedLazyTessellationCache::sharedLazyTessellationCache.getTime(commitCounter);
          const bool allAllocationsValid = SharedLazyTessellationCache::validTime(time,curTime);
          
          patch = allAllocationsValid ? patch : nullptr;

          /* use cached data structure for calculations */
          const vbool valid1 = patch ? eval(valid0,patch,u,v,1.0f,0) : vbool(false);
          SharedLazyTessellationCache::unlock();
          const vbool valid2 = valid0 & !valid1;
          if (any(valid2)) {
            FeatureAdaptiveEvalSimd<vbool,vint,vfloat,Vertex,Vertex_t>(edge,vertices,stride,valid2,u,v,P,dPdu,dPdv,ddPdudu,ddPdvdv,ddPdudv,dstride,N);
          }
        }
        
        vbool eval_quad(const vbool& valid, const typename Patch::SubdividedQuadPatch* This, const vfloat& u, const vfloat& v, const float dscale, const size_t depth)
        {
          vbool ret = false;
          const vbool u0_mask = u < 0.5f, u1_mask = u >= 0.5f;
          const vbool v0_mask = v < 0.5f, v1_mask = v >= 0.5f;
          const vbool u0v0_mask = valid & u0_mask & v0_mask;
          const vbool u0v1_mask = valid & u0_mask & v1_mask;
          const vbool u1v0_mask = valid & u1_mask & v0_mask;
          const vbool u1v1_mask = valid & u1_mask & v1_mask;
          if (any(u0v0_mask)) ret |= eval(u0v0_mask,This->child[0],2.0f*u,2.0f*v,2.0f*dscale,depth+1);
          if (any(u1v0_mask)) ret |= eval(u1v0_mask,This->child[1],2.0f*u-1.0f,2.0f*v,2.0f*dscale,depth+1);
          if (any(u1v1_mask)) ret |= eval(u1v1_mask,This->child[2],2.0f*u-1.0f,2.0f*v-1.0f,2.0f*dscale,depth+1);
          if (any(u0v1_mask)) ret |= eval(u0v1_mask,This->child[3],2.0f*u,2.0f*v-1.0f,2.0f*dscale,depth+1);
          return ret;
        }
        
        vbool eval_general(const vbool& valid, const typename Patch::SubdividedGeneralPatch* patch, const vfloat& U, const vfloat& V, const size_t depth)
        {
          vbool ret = false;
          const vint l = (vint)floor(0.5f*U); const vfloat u = 2.0f*frac(0.5f*U)-0.5f; 
          const vint h = (vint)floor(0.5f*V); const vfloat v = 2.0f*frac(0.5f*V)-0.5f; 
          const vint i = (h<<2)+l; assert(all(valid,i<patch->N));
          foreach_unique(valid,i,[&](const vbool& valid, const int i) {
              ret |= eval(valid,patch->child[i],u,v,1.0f,depth+1);
            });
          return ret;
        }
        
        vbool eval(const vbool& valid, Ref This, const vfloat& u, const vfloat& v, const float dscale, const size_t depth) 
        {
          if (!This) return false;
          switch (This.type()) 
          {
          case Patch::BILINEAR_PATCH: {
            ((typename Patch::BilinearPatch*)This.object())->patch.eval(valid,u,v,P,dPdu,dPdv,ddPdudu,ddPdvdv,ddPdudv,dscale,dstride,N); 
            return valid;
          }
          case Patch::BSPLINE_PATCH: {
            ((typename Patch::BSplinePatch*)This.object())->patch.eval(valid,u,v,P,dPdu,dPdv,ddPdudu,ddPdvdv,ddPdudv,dscale,dstride,N);
            return valid;
          }
          case Patch::BEZIER_PATCH: {
            ((typename Patch::BezierPatch*)This.object())->patch.eval(valid,u,v,P,dPdu,dPdv,ddPdudu,ddPdvdv,ddPdudv,dscale,dstride,N);
            return valid;
          }
          case Patch::GREGORY_PATCH: {
            ((typename Patch::GregoryPatch*)This.object())->patch.eval(valid,u,v,P,dPdu,dPdv,ddPdudu,ddPdvdv,ddPdudv,dscale,dstride,N); 
            return valid;
          }
          case Patch::SUBDIVIDED_QUAD_PATCH: {
            return eval_quad(valid,((typename Patch::SubdividedQuadPatch*)This.object()),u,v,dscale,depth);
          }
          case Patch::SUBDIVIDED_GENERAL_PATCH: { 
            assert(dscale == 1.0f); 
            return eval_general(valid,((typename Patch::SubdividedGeneralPatch*)This.object()),u,v,depth); 
          }
          case Patch::EVAL_PATCH: { 
            CatmullClarkPatch patch; patch.deserialize(This.object());
            FeatureAdaptiveEvalSimd<vbool,vint,vfloat,Vertex,Vertex_t>(patch,valid,u,v,dscale,depth,P,dPdu,dPdv,ddPdudu,ddPdvdv,ddPdudv,dstride,N);
            return valid;
          }
          default: 
            assert(false); 
            return false;
          }
        }

      private:
        float* const P;
        float* const dPdu;
        float* const dPdv;
        float* const ddPdudu;
        float* const ddPdvdv;
        float* const ddPdudv;
        const size_t dstride;
        const size_t N;
      };
  }
}