Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
godotengine
GitHub Repository: godotengine/godot
 Path: blob/master/thirdparty/embree/common/simd/vint8_avx.h
9912 views
1
// Copyright 2009-2021 Intel Corporation

2
// SPDX-License-Identifier: Apache-2.0

3


4
#pragma once

5


6
#define vboolf vboolf_impl

7
#define vboold vboold_impl

8
#define vint vint_impl

9
#define vuint vuint_impl

10
#define vllong vllong_impl

11
#define vfloat vfloat_impl

12
#define vdouble vdouble_impl

13


14
namespace embree

15
{

16
  /* 8-wide AVX integer type */

17
  template<>

18
  struct vint<8>

19
  {

20
    ALIGNED_STRUCT_(32);

21
    

22
    typedef vboolf8 Bool;

23
    typedef vint8   Int;

24
    typedef vfloat8 Float;

25


26
    enum  { size = 8 };        // number of SIMD elements

27
    union {                    // data

28
      __m256i v;

29
      struct { __m128i vl,vh; };

30
      int i[8];

31
    }; 

32


33
    ////////////////////////////////////////////////////////////////////////////////

34
    /// Constructors, Assignment & Cast Operators

35
    ////////////////////////////////////////////////////////////////////////////////

36


37
    __forceinline vint() {}

38
    __forceinline vint(const vint8& a) { v = a.v; }

39
    __forceinline vint8& operator =(const vint8& a) { v = a.v; return *this; }

40


41
    __forceinline vint(__m256i a) : v(a) {}

42
    __forceinline operator const __m256i&() const { return v; }

43
    __forceinline operator       __m256i&()       { return v; }

44


45
    __forceinline explicit vint(const vint4& a) : v(_mm256_insertf128_si256(_mm256_castsi128_si256(a),a,1)) {}

46
    __forceinline vint(const vint4& a, const vint4& b) : v(_mm256_insertf128_si256(_mm256_castsi128_si256(a),b,1)) {}

47
    __forceinline vint(const __m128i& a, const __m128i& b) : vl(a), vh(b) {}

48
 

49
    __forceinline explicit vint(const int* a) : v(_mm256_castps_si256(_mm256_loadu_ps((const float*)a))) {}

50
    __forceinline vint(int a) : v(_mm256_set1_epi32(a)) {}

51
    __forceinline vint(int a, int b) : v(_mm256_set_epi32(b, a, b, a, b, a, b, a)) {}

52
    __forceinline vint(int a, int b, int c, int d) : v(_mm256_set_epi32(d, c, b, a, d, c, b, a)) {}

53
    __forceinline vint(int a, int b, int c, int d, int e, int f, int g, int vh) : v(_mm256_set_epi32(vh, g, f, e, d, c, b, a)) {}

54


55
    __forceinline explicit vint(__m256 a) : v(_mm256_cvtps_epi32(a)) {}

56


57
    ////////////////////////////////////////////////////////////////////////////////

58
    /// Constants

59
    ////////////////////////////////////////////////////////////////////////////////

60


61
    __forceinline vint(ZeroTy)        : v(_mm256_setzero_si256()) {}

62
    __forceinline vint(OneTy)         : v(_mm256_set_epi32(1,1,1,1,1,1,1,1)) {}

63
    __forceinline vint(PosInfTy)      : v(_mm256_set_epi32(pos_inf,pos_inf,pos_inf,pos_inf,pos_inf,pos_inf,pos_inf,pos_inf)) {}

64
    __forceinline vint(NegInfTy)      : v(_mm256_set_epi32(neg_inf,neg_inf,neg_inf,neg_inf,neg_inf,neg_inf,neg_inf,neg_inf)) {}

65
    __forceinline vint(StepTy)        : v(_mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0)) {}

66
    __forceinline vint(ReverseStepTy) : v(_mm256_set_epi32(0, 1, 2, 3, 4, 5, 6, 7)) {}

67
    __forceinline vint(UndefinedTy)   : v(_mm256_undefined_si256()) {}

68


69
    ////////////////////////////////////////////////////////////////////////////////

70
    /// Loads and Stores

71
    ////////////////////////////////////////////////////////////////////////////////

72


73
    static __forceinline vint8 load (const void* a) { return _mm256_castps_si256(_mm256_load_ps((float*)a)); }

74
    static __forceinline vint8 loadu(const void* a) { return _mm256_castps_si256(_mm256_loadu_ps((float*)a)); }

75


76
    static __forceinline vint8 load (const vboolf8& mask, const void* a) { return _mm256_castps_si256(_mm256_maskload_ps((float*)a,mask)); }

77
    static __forceinline vint8 loadu(const vboolf8& mask, const void* a) { return _mm256_castps_si256(_mm256_maskload_ps((float*)a,mask)); }

78


79
    static __forceinline void store (void* ptr, const vint8& f) { _mm256_store_ps((float*)ptr,_mm256_castsi256_ps(f)); }

80
    static __forceinline void storeu(void* ptr, const vint8& f) { _mm256_storeu_ps((float*)ptr,_mm256_castsi256_ps(f)); }

81
    

82
    static __forceinline void store (const vboolf8& mask, void* ptr, const vint8& f) { _mm256_maskstore_ps((float*)ptr,_mm256_castps_si256(mask.v),_mm256_castsi256_ps(f)); }

83
    static __forceinline void storeu(const vboolf8& mask, void* ptr, const vint8& f) { _mm256_maskstore_ps((float*)ptr,_mm256_castps_si256(mask.v),_mm256_castsi256_ps(f)); }

84


85
    static __forceinline void store_nt(void* ptr, const vint8& v) {

86
      _mm256_stream_ps((float*)ptr,_mm256_castsi256_ps(v));

87
    }

88


89
    static __forceinline vint8 load(const unsigned char* ptr) {

90
      vint4 il = vint4::load(ptr+0);

91
      vint4 ih = vint4::load(ptr+4);

92
      return vint8(il,ih);

93
    }

94


95
    static __forceinline vint8 loadu(const unsigned char* ptr) {

96
      vint4 il = vint4::loadu(ptr+0);

97
      vint4 ih = vint4::loadu(ptr+4);

98
      return vint8(il,ih);

99
    }

100


101
    static __forceinline vint8 load(const unsigned short* ptr) {

102
      vint4 il = vint4::load(ptr+0);

103
      vint4 ih = vint4::load(ptr+4);

104
      return vint8(il,ih);

105
    }

106


107
    static __forceinline vint8 loadu(const unsigned short* ptr) {

108
      vint4 il = vint4::loadu(ptr+0);

109
      vint4 ih = vint4::loadu(ptr+4);

110
      return vint8(il,ih);

111
    }

112


113
    static __forceinline void store(unsigned char* ptr, const vint8& i) {

114
      vint4 il(i.vl);

115
      vint4 ih(i.vh);

116
      vint4::store(ptr + 0,il);

117
      vint4::store(ptr + 4,ih);

118
    }

119


120
    static __forceinline void store(unsigned short* ptr, const vint8& v) {

121
      for (size_t i=0;i<8;i++)

122
        ptr[i] = (unsigned short)v[i];

123
    }

124


125
    template<int scale = 4>

126
    static __forceinline vint8 gather(const int* ptr, const vint8& index) {

127
      return vint8(

128
          *(int*)(((char*)ptr)+scale*index[0]),

129
          *(int*)(((char*)ptr)+scale*index[1]),

130
          *(int*)(((char*)ptr)+scale*index[2]),

131
          *(int*)(((char*)ptr)+scale*index[3]),

132
          *(int*)(((char*)ptr)+scale*index[4]),

133
          *(int*)(((char*)ptr)+scale*index[5]),

134
          *(int*)(((char*)ptr)+scale*index[6]),

135
          *(int*)(((char*)ptr)+scale*index[7]));

136
    }

137


138
    template<int scale = 4>

139
    static __forceinline vint8 gather(const vboolf8& mask, const int* ptr, const vint8& index) {

140
      vint8 r = zero;

141
      if (likely(mask[0])) r[0] = *(int*)(((char*)ptr)+scale*index[0]);

142
      if (likely(mask[1])) r[1] = *(int*)(((char*)ptr)+scale*index[1]);

143
      if (likely(mask[2])) r[2] = *(int*)(((char*)ptr)+scale*index[2]);

144
      if (likely(mask[3])) r[3] = *(int*)(((char*)ptr)+scale*index[3]);

145
      if (likely(mask[4])) r[4] = *(int*)(((char*)ptr)+scale*index[4]);

146
      if (likely(mask[5])) r[5] = *(int*)(((char*)ptr)+scale*index[5]);

147
      if (likely(mask[6])) r[6] = *(int*)(((char*)ptr)+scale*index[6]);

148
      if (likely(mask[7])) r[7] = *(int*)(((char*)ptr)+scale*index[7]);

149
      return r;

150
    }

151


152
    template<int scale = 4>

153
    static __forceinline void scatter(void* ptr, const vint8& ofs, const vint8& v)

154
    {

155
      *(int*)(((char*)ptr)+scale*ofs[0]) = v[0];

156
      *(int*)(((char*)ptr)+scale*ofs[1]) = v[1];

157
      *(int*)(((char*)ptr)+scale*ofs[2]) = v[2];

158
      *(int*)(((char*)ptr)+scale*ofs[3]) = v[3];

159
      *(int*)(((char*)ptr)+scale*ofs[4]) = v[4];

160
      *(int*)(((char*)ptr)+scale*ofs[5]) = v[5];

161
      *(int*)(((char*)ptr)+scale*ofs[6]) = v[6];

162
      *(int*)(((char*)ptr)+scale*ofs[7]) = v[7];

163
    }

164


165
    template<int scale = 4>

166
    static __forceinline void scatter(const vboolf8& mask, void* ptr, const vint8& ofs, const vint8& v)

167
    {

168
      if (likely(mask[0])) *(int*)(((char*)ptr)+scale*ofs[0]) = v[0];

169
      if (likely(mask[1])) *(int*)(((char*)ptr)+scale*ofs[1]) = v[1];

170
      if (likely(mask[2])) *(int*)(((char*)ptr)+scale*ofs[2]) = v[2];

171
      if (likely(mask[3])) *(int*)(((char*)ptr)+scale*ofs[3]) = v[3];

172
      if (likely(mask[4])) *(int*)(((char*)ptr)+scale*ofs[4]) = v[4];

173
      if (likely(mask[5])) *(int*)(((char*)ptr)+scale*ofs[5]) = v[5];

174
      if (likely(mask[6])) *(int*)(((char*)ptr)+scale*ofs[6]) = v[6];

175
      if (likely(mask[7])) *(int*)(((char*)ptr)+scale*ofs[7]) = v[7];

176
    }

177


178


179
    static __forceinline vint8 broadcast64(const long long& a) { return _mm256_set1_epi64x(a); }

180
    

181
    ////////////////////////////////////////////////////////////////////////////////

182
    /// Array Access

183
    ////////////////////////////////////////////////////////////////////////////////

184


185
    __forceinline const int& operator [](size_t index) const { assert(index < 8); return i[index]; }

186
    __forceinline       int& operator [](size_t index)       { assert(index < 8); return i[index]; }

187
  };

188


189
  ////////////////////////////////////////////////////////////////////////////////

190
  /// Unary Operators

191
  ////////////////////////////////////////////////////////////////////////////////

192


193
  __forceinline vboolf8 asBool(const vint8& a) { return _mm256_castsi256_ps(a); }

194


195
  __forceinline vint8 operator +(const vint8& a) { return a; }

196
  __forceinline vint8 operator -(const vint8& a) { return vint8(_mm_sub_epi32(_mm_setzero_si128(), a.vl), _mm_sub_epi32(_mm_setzero_si128(), a.vh)); }

197
  __forceinline vint8 abs       (const vint8& a) { return vint8(_mm_abs_epi32(a.vl), _mm_abs_epi32(a.vh)); }

198


199
  ////////////////////////////////////////////////////////////////////////////////

200
  /// Binary Operators

201
  ////////////////////////////////////////////////////////////////////////////////

202


203
  __forceinline vint8 operator +(const vint8& a, const vint8& b) { return vint8(_mm_add_epi32(a.vl, b.vl), _mm_add_epi32(a.vh, b.vh)); }

204
  __forceinline vint8 operator +(const vint8& a, int          b) { return a + vint8(b); }

205
  __forceinline vint8 operator +(int          a, const vint8& b) { return vint8(a) + b; }

206


207
  __forceinline vint8 operator -(const vint8& a, const vint8& b) { return vint8(_mm_sub_epi32(a.vl, b.vl), _mm_sub_epi32(a.vh, b.vh)); }

208
  __forceinline vint8 operator -(const vint8& a, int          b) { return a - vint8(b); }

209
  __forceinline vint8 operator -(int          a, const vint8& b) { return vint8(a) - b; }

210


211
  __forceinline vint8 operator *(const vint8& a, const vint8& b) { return vint8(_mm_mullo_epi32(a.vl, b.vl), _mm_mullo_epi32(a.vh, b.vh)); }

212
  __forceinline vint8 operator *(const vint8& a, int          b) { return a * vint8(b); }

213
  __forceinline vint8 operator *(int          a, const vint8& b) { return vint8(a) * b; }

214


215
  __forceinline vint8 operator &(const vint8& a, const vint8& b) { return _mm256_castps_si256(_mm256_and_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

216
  __forceinline vint8 operator &(const vint8& a, int          b) { return a & vint8(b); }

217
  __forceinline vint8 operator &(int          a, const vint8& b) { return vint8(a) & b; }

218


219
  __forceinline vint8 operator |(const vint8& a, const vint8& b) { return _mm256_castps_si256(_mm256_or_ps (_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

220
  __forceinline vint8 operator |(const vint8& a, int          b) { return a | vint8(b); }

221
  __forceinline vint8 operator |(int          a, const vint8& b) { return vint8(a) | b; }

222


223
  __forceinline vint8 operator ^(const vint8& a, const vint8& b) { return _mm256_castps_si256(_mm256_xor_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

224
  __forceinline vint8 operator ^(const vint8& a, int          b) { return a ^ vint8(b); }

225
  __forceinline vint8 operator ^(int          a, const vint8& b) { return vint8(a) ^ b; }

226


227
  __forceinline vint8 operator <<(const vint8& a, int n) { return vint8(_mm_slli_epi32(a.vl, n), _mm_slli_epi32(a.vh, n)); }

228
  __forceinline vint8 operator >>(const vint8& a, int n) { return vint8(_mm_srai_epi32(a.vl, n), _mm_srai_epi32(a.vh, n)); }

229


230
  __forceinline vint8 sll (const vint8& a, int b) { return vint8(_mm_slli_epi32(a.vl, b), _mm_slli_epi32(a.vh, b)); }

231
  __forceinline vint8 sra (const vint8& a, int b) { return vint8(_mm_srai_epi32(a.vl, b), _mm_srai_epi32(a.vh, b)); }

232
  __forceinline vint8 srl (const vint8& a, int b) { return vint8(_mm_srli_epi32(a.vl, b), _mm_srli_epi32(a.vh, b)); }

233
  

234
  __forceinline vint8 min(const vint8& a, const vint8& b) { return vint8(_mm_min_epi32(a.vl, b.vl), _mm_min_epi32(a.vh, b.vh)); }

235
  __forceinline vint8 min(const vint8& a, int          b) { return min(a,vint8(b)); }

236
  __forceinline vint8 min(int          a, const vint8& b) { return min(vint8(a),b); }

237


238
  __forceinline vint8 max(const vint8& a, const vint8& b) { return vint8(_mm_max_epi32(a.vl, b.vl), _mm_max_epi32(a.vh, b.vh)); }

239
  __forceinline vint8 max(const vint8& a, int          b) { return max(a,vint8(b)); }

240
  __forceinline vint8 max(int          a, const vint8& b) { return max(vint8(a),b); }

241


242
  __forceinline vint8 umin(const vint8& a, const vint8& b) { return vint8(_mm_min_epu32(a.vl, b.vl), _mm_min_epu32(a.vh, b.vh)); }

243
  __forceinline vint8 umin(const vint8& a, int          b) { return umin(a,vint8(b)); }

244
  __forceinline vint8 umin(int          a, const vint8& b) { return umin(vint8(a),b); }

245


246
  __forceinline vint8 umax(const vint8& a, const vint8& b) { return vint8(_mm_max_epu32(a.vl, b.vl), _mm_max_epu32(a.vh, b.vh)); }

247
  __forceinline vint8 umax(const vint8& a, int          b) { return umax(a,vint8(b)); }

248
  __forceinline vint8 umax(int          a, const vint8& b) { return umax(vint8(a),b); }

249


250
  ////////////////////////////////////////////////////////////////////////////////

251
  /// Assignment Operators

252
  ////////////////////////////////////////////////////////////////////////////////

253


254
  __forceinline vint8& operator +=(vint8& a, const vint8& b) { return a = a + b; }

255
  __forceinline vint8& operator +=(vint8& a, int          b) { return a = a + b; }

256
  

257
  __forceinline vint8& operator -=(vint8& a, const vint8& b) { return a = a - b; }

258
  __forceinline vint8& operator -=(vint8& a, int          b) { return a = a - b; }

259
  

260
  __forceinline vint8& operator *=(vint8& a, const vint8& b) { return a = a * b; }

261
  __forceinline vint8& operator *=(vint8& a, int          b) { return a = a * b; }

262
  

263
  __forceinline vint8& operator &=(vint8& a, const vint8& b) { return a = a & b; }

264
  __forceinline vint8& operator &=(vint8& a, int          b) { return a = a & b; }

265
  

266
  __forceinline vint8& operator |=(vint8& a, const vint8& b) { return a = a | b; }

267
  __forceinline vint8& operator |=(vint8& a, int          b) { return a = a | b; }

268
  

269
  __forceinline vint8& operator <<=(vint8& a, int b) { return a = a << b; }

270
  __forceinline vint8& operator >>=(vint8& a, int b) { return a = a >> b; }

271


272
  ////////////////////////////////////////////////////////////////////////////////

273
  /// Comparison Operators + Select

274
  ////////////////////////////////////////////////////////////////////////////////

275


276
  __forceinline vboolf8 operator ==(const vint8& a, const vint8& b) { return vboolf8(_mm_castsi128_ps(_mm_cmpeq_epi32 (a.vl, b.vl)),

277
                                                                                     _mm_castsi128_ps(_mm_cmpeq_epi32 (a.vh, b.vh))); }

278
  __forceinline vboolf8 operator ==(const vint8& a, int          b) { return a == vint8(b); }

279
  __forceinline vboolf8 operator ==(int          a, const vint8& b) { return vint8(a) == b; }

280
  

281
  __forceinline vboolf8 operator !=(const vint8& a, const vint8& b) { return !(a == b); }

282
  __forceinline vboolf8 operator !=(const vint8& a, int          b) { return a != vint8(b); }

283
  __forceinline vboolf8 operator !=(int          a, const vint8& b) { return vint8(a) != b; }

284
  

285
  __forceinline vboolf8 operator < (const vint8& a, const vint8& b) { return vboolf8(_mm_castsi128_ps(_mm_cmplt_epi32 (a.vl, b.vl)),

286
                                                                                     _mm_castsi128_ps(_mm_cmplt_epi32 (a.vh, b.vh))); }

287
  __forceinline vboolf8 operator < (const vint8& a, int          b) { return a <  vint8(b); }

288
  __forceinline vboolf8 operator < (int          a, const vint8& b) { return vint8(a) <  b; }

289
  

290
  __forceinline vboolf8 operator >=(const vint8& a, const vint8& b) { return !(a <  b); }

291
  __forceinline vboolf8 operator >=(const vint8& a, int          b) { return a >= vint8(b); }

292
  __forceinline vboolf8 operator >=(int          a, const vint8& b) { return vint8(a) >= b; }

293


294
  __forceinline vboolf8 operator > (const vint8& a, const vint8& b) { return vboolf8(_mm_castsi128_ps(_mm_cmpgt_epi32 (a.vl, b.vl)),

295
                                                                                     _mm_castsi128_ps(_mm_cmpgt_epi32 (a.vh, b.vh))); }

296
  __forceinline vboolf8 operator > (const vint8& a, int          b) { return a >  vint8(b); }

297
  __forceinline vboolf8 operator > (int          a, const vint8& b) { return vint8(a) >  b; }

298


299
  __forceinline vboolf8 operator <=(const vint8& a, const vint8& b) { return !(a >  b); }

300
  __forceinline vboolf8 operator <=(const vint8& a, int          b) { return a <= vint8(b); }

301
  __forceinline vboolf8 operator <=(int          a, const vint8& b) { return vint8(a) <= b; }

302


303
  __forceinline vboolf8 eq(const vint8& a, const vint8& b) { return a == b; }

304
  __forceinline vboolf8 ne(const vint8& a, const vint8& b) { return a != b; }

305
  __forceinline vboolf8 lt(const vint8& a, const vint8& b) { return a <  b; }

306
  __forceinline vboolf8 ge(const vint8& a, const vint8& b) { return a >= b; }

307
  __forceinline vboolf8 gt(const vint8& a, const vint8& b) { return a >  b; }

308
  __forceinline vboolf8 le(const vint8& a, const vint8& b) { return a <= b; }

309


310
  __forceinline vboolf8 eq(const vboolf8& mask, const vint8& a, const vint8& b) { return mask & (a == b); }

311
  __forceinline vboolf8 ne(const vboolf8& mask, const vint8& a, const vint8& b) { return mask & (a != b); }

312
  __forceinline vboolf8 lt(const vboolf8& mask, const vint8& a, const vint8& b) { return mask & (a <  b); }

313
  __forceinline vboolf8 ge(const vboolf8& mask, const vint8& a, const vint8& b) { return mask & (a >= b); }

314
  __forceinline vboolf8 gt(const vboolf8& mask, const vint8& a, const vint8& b) { return mask & (a >  b); }

315
  __forceinline vboolf8 le(const vboolf8& mask, const vint8& a, const vint8& b) { return mask & (a <= b); }

316


317
  __forceinline vint8 select(const vboolf8& m, const vint8& t, const vint8& f) {

318
    return _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(f), _mm256_castsi256_ps(t), m)); 

319
  }

320


321
  ////////////////////////////////////////////////////////////////////////////////

322
  /// Movement/Shifting/Shuffling Functions

323
  ////////////////////////////////////////////////////////////////////////////////

324


325
  __forceinline vint8 unpacklo(const vint8& a, const vint8& b) { return _mm256_castps_si256(_mm256_unpacklo_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

326
  __forceinline vint8 unpackhi(const vint8& a, const vint8& b) { return _mm256_castps_si256(_mm256_unpackhi_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

327


328
  template<int i>

329
  __forceinline vint8 shuffle(const vint8& v) {

330
    return _mm256_castps_si256(_mm256_permute_ps(_mm256_castsi256_ps(v), _MM_SHUFFLE(i, i, i, i)));

331
  }

332


333
  template<int i0, int i1>

334
  __forceinline vint8 shuffle4(const vint8& v) {

335
    return _mm256_permute2f128_si256(v, v, (i1 << 4) | (i0 << 0));

336
  }

337


338
  template<int i0, int i1>

339
  __forceinline vint8 shuffle4(const vint8& a, const vint8& b) {

340
    return _mm256_permute2f128_si256(a, b, (i1 << 4) | (i0 << 0));

341
  }

342


343
  template<int i0, int i1, int i2, int i3>

344
  __forceinline vint8 shuffle(const vint8& v) {

345
    return _mm256_castps_si256(_mm256_permute_ps(_mm256_castsi256_ps(v), _MM_SHUFFLE(i3, i2, i1, i0)));

346
  }

347


348
  template<int i0, int i1, int i2, int i3>

349
  __forceinline vint8 shuffle(const vint8& a, const vint8& b) {

350
    return _mm256_castps_si256(_mm256_shuffle_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b), _MM_SHUFFLE(i3, i2, i1, i0)));

351
  }

352


353
  template<> __forceinline vint8 shuffle<0, 0, 2, 2>(const vint8& v) { return _mm256_castps_si256(_mm256_moveldup_ps(_mm256_castsi256_ps(v))); }

354
  template<> __forceinline vint8 shuffle<1, 1, 3, 3>(const vint8& v) { return _mm256_castps_si256(_mm256_movehdup_ps(_mm256_castsi256_ps(v))); }

355
  template<> __forceinline vint8 shuffle<0, 1, 0, 1>(const vint8& v) { return _mm256_castps_si256(_mm256_castpd_ps(_mm256_movedup_pd(_mm256_castps_pd(_mm256_castsi256_ps(v))))); }

356


357
  __forceinline vint8 broadcast(const int* ptr) { return _mm256_castps_si256(_mm256_broadcast_ss((const float*)ptr)); }

358
  template<int i> __forceinline vint8 insert4(const vint8& a, const vint4& b) { return _mm256_insertf128_si256(a, b, i); }

359
  template<int i> __forceinline vint4 extract4(const vint8& a) { return _mm256_extractf128_si256(a, i); }

360
  template<> __forceinline vint4 extract4<0>(const vint8& a) { return _mm256_castsi256_si128(a); }

361


362
  __forceinline int toScalar(const vint8& v) { return _mm_cvtsi128_si32(_mm256_castsi256_si128(v)); }

363


364


365
  ////////////////////////////////////////////////////////////////////////////////

366
  /// Reductions

367
  ////////////////////////////////////////////////////////////////////////////////

368


369
  __forceinline vint8 vreduce_min2(const vint8& v) { return min(v,shuffle<1,0,3,2>(v)); }

370
  __forceinline vint8 vreduce_min4(const vint8& v) { vint8 v1 = vreduce_min2(v); return min(v1,shuffle<2,3,0,1>(v1)); }

371
  __forceinline vint8 vreduce_min (const vint8& v) { vint8 v1 = vreduce_min4(v); return min(v1,shuffle4<1,0>(v1)); }

372


373
  __forceinline vint8 vreduce_max2(const vint8& v) { return max(v,shuffle<1,0,3,2>(v)); }

374
  __forceinline vint8 vreduce_max4(const vint8& v) { vint8 v1 = vreduce_max2(v); return max(v1,shuffle<2,3,0,1>(v1)); }

375
  __forceinline vint8 vreduce_max (const vint8& v) { vint8 v1 = vreduce_max4(v); return max(v1,shuffle4<1,0>(v1)); }

376


377
  __forceinline vint8 vreduce_add2(const vint8& v) { return v + shuffle<1,0,3,2>(v); }

378
  __forceinline vint8 vreduce_add4(const vint8& v) { vint8 v1 = vreduce_add2(v); return v1 + shuffle<2,3,0,1>(v1); }

379
  __forceinline vint8 vreduce_add (const vint8& v) { vint8 v1 = vreduce_add4(v); return v1 + shuffle4<1,0>(v1); }

380


381
  __forceinline int reduce_min(const vint8& v) { return toScalar(vreduce_min(v)); }

382
  __forceinline int reduce_max(const vint8& v) { return toScalar(vreduce_max(v)); }

383
  __forceinline int reduce_add(const vint8& v) { return toScalar(vreduce_add(v)); }

384


385
  __forceinline size_t select_min(const vint8& v) { return bsf(movemask(v == vreduce_min(v))); }

386
  __forceinline size_t select_max(const vint8& v) { return bsf(movemask(v == vreduce_max(v))); }

387


388
  __forceinline size_t select_min(const vboolf8& valid, const vint8& v) { const vint8 a = select(valid,v,vint8(pos_inf)); return bsf(movemask(valid & (a == vreduce_min(a)))); }

389
  __forceinline size_t select_max(const vboolf8& valid, const vint8& v) { const vint8 a = select(valid,v,vint8(neg_inf)); return bsf(movemask(valid & (a == vreduce_max(a)))); }

390


391
  ////////////////////////////////////////////////////////////////////////////////

392
  /// Sorting networks

393
  ////////////////////////////////////////////////////////////////////////////////

394


395
  __forceinline vint8 usort_ascending(const vint8& v)

396
  {

397
    const vint8 a0 = v;

398
    const vint8 b0 = shuffle<1,0,3,2>(a0);

399
    const vint8 c0 = umin(a0,b0);

400
    const vint8 d0 = umax(a0,b0);

401
    const vint8 a1 = select(0x99 /* 0b10011001 */,c0,d0);

402
    const vint8 b1 = shuffle<2,3,0,1>(a1);

403
    const vint8 c1 = umin(a1,b1);

404
    const vint8 d1 = umax(a1,b1);

405
    const vint8 a2 = select(0xc3 /* 0b11000011 */,c1,d1);

406
    const vint8 b2 = shuffle<1,0,3,2>(a2);

407
    const vint8 c2 = umin(a2,b2);

408
    const vint8 d2 = umax(a2,b2);

409
    const vint8 a3 = select(0xa5 /* 0b10100101 */,c2,d2);

410
    const vint8 b3 = shuffle4<1,0>(a3);

411
    const vint8 c3 = umin(a3,b3);

412
    const vint8 d3 = umax(a3,b3);

413
    const vint8 a4 = select(0xf /* 0b00001111 */,c3,d3);

414
    const vint8 b4 = shuffle<2,3,0,1>(a4);

415
    const vint8 c4 = umin(a4,b4);

416
    const vint8 d4 = umax(a4,b4);

417
    const vint8 a5 = select(0x33 /* 0b00110011 */,c4,d4);

418
    const vint8 b5 = shuffle<1,0,3,2>(a5);

419
    const vint8 c5 = umin(a5,b5);

420
    const vint8 d5 = umax(a5,b5);

421
    const vint8 a6 = select(0x55 /* 0b01010101 */,c5,d5);

422
    return a6;

423
  }

424


425
  __forceinline vint8 usort_descending(const vint8& v)

426
  {

427
    const vint8 a0 = v;

428
    const vint8 b0 = shuffle<1,0,3,2>(a0);

429
    const vint8 c0 = umax(a0,b0);

430
    const vint8 d0 = umin(a0,b0);

431
    const vint8 a1 = select(0x99 /* 0b10011001 */,c0,d0);

432
    const vint8 b1 = shuffle<2,3,0,1>(a1);

433
    const vint8 c1 = umax(a1,b1);

434
    const vint8 d1 = umin(a1,b1);

435
    const vint8 a2 = select(0xc3 /* 0b11000011 */,c1,d1);

436
    const vint8 b2 = shuffle<1,0,3,2>(a2);

437
    const vint8 c2 = umax(a2,b2);

438
    const vint8 d2 = umin(a2,b2);

439
    const vint8 a3 = select(0xa5 /* 0b10100101 */,c2,d2);

440
    const vint8 b3 = shuffle4<1,0>(a3);

441
    const vint8 c3 = umax(a3,b3);

442
    const vint8 d3 = umin(a3,b3);

443
    const vint8 a4 = select(0xf /* 0b00001111 */,c3,d3);

444
    const vint8 b4 = shuffle<2,3,0,1>(a4);

445
    const vint8 c4 = umax(a4,b4);

446
    const vint8 d4 = umin(a4,b4);

447
    const vint8 a5 = select(0x33 /* 0b00110011 */,c4,d4);

448
    const vint8 b5 = shuffle<1,0,3,2>(a5);

449
    const vint8 c5 = umax(a5,b5);

450
    const vint8 d5 = umin(a5,b5);

451
    const vint8 a6 = select(0x55 /* 0b01010101 */,c5,d5);

452
    return a6;

453
  }

454


455
  ////////////////////////////////////////////////////////////////////////////////

456
  /// Output Operators

457
  ////////////////////////////////////////////////////////////////////////////////

458


459
  __forceinline embree_ostream operator <<(embree_ostream cout, const vint8& a) {

460
    return cout << "<" << a[0] << ", " << a[1] << ", " << a[2] << ", " << a[3] << ", " << a[4] << ", " << a[5] << ", " << a[6] << ", " << a[7] << ">";

461
  }

462
}

463


464
#undef vboolf

465
#undef vboold

466
#undef vint

467
#undef vuint

468
#undef vllong

469
#undef vfloat

470
#undef vdouble

471


472