Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
godotengine
GitHub Repository: godotengine/godot
 Path: blob/master/thirdparty/embree/common/simd/vuint8_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 vuint<8>

19
  {

20
    ALIGNED_STRUCT_(32);   

21


22
    typedef vboolf8 Bool;

23
    typedef vuint8   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
      unsigned int i[8];

31
    }; 

32


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

34
    /// Constructors, Assignment & Cast Operators

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

36


37
    __forceinline vuint() {}

38
    __forceinline vuint(const vuint8& a) { v = a.v; }

39
    __forceinline vuint8& operator =(const vuint8& a) { v = a.v; return *this; }

40


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

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

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

44


45
    __forceinline explicit vuint(const vuint4& a) : v(_mm256_insertf128_si256(_mm256_castsi128_si256(a),a,1)) {}

46
    __forceinline vuint(const vuint4& a, const vuint4& b) : v(_mm256_insertf128_si256(_mm256_castsi128_si256(a),b,1)) {}

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

48
 

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

50
    __forceinline vuint(unsigned int a) : v(_mm256_set1_epi32(a)) {}

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

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

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

54


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

56


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

58
    /// Constants

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

60


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

62
    __forceinline vuint(OneTy)    : v(_mm256_set1_epi32(1)) {}

63
    __forceinline vuint(PosInfTy) : v(_mm256_set1_epi32(0xFFFFFFFF)) {}

64
    __forceinline vuint(StepTy)   : v(_mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0)) {}

65
    __forceinline vuint(UndefinedTy) : v(_mm256_undefined_si256()) {}

66


67
    ////////////////////////////////////////////////////////////////////////////////

68
    /// Loads and Stores

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

70


71
    static __forceinline vuint8 load (const void* a) { return _mm256_castps_si256(_mm256_load_ps((float*)a)); }

72
    static __forceinline vuint8 loadu(const void* a) { return _mm256_castps_si256(_mm256_loadu_ps((float*)a)); }

73


74
    static __forceinline vuint8 load (const vboolf8& mask, const void* a) { return _mm256_castps_si256(_mm256_maskload_ps((float*)a,mask)); }

75
    static __forceinline vuint8 loadu(const vboolf8& mask, const void* a) { return _mm256_castps_si256(_mm256_maskload_ps((float*)a,mask)); }

76


77
    static __forceinline void store (void* ptr, const vuint8& f) { _mm256_store_ps((float*)ptr,_mm256_castsi256_ps(f)); }

78
    static __forceinline void storeu(void* ptr, const vuint8& f) { _mm256_storeu_ps((float*)ptr,_mm256_castsi256_ps(f)); }

79
    

80
    static __forceinline void store (const vboolf8& mask, void* ptr, const vuint8& f) { _mm256_maskstore_ps((float*)ptr,_mm256_castps_si256(mask.v),_mm256_castsi256_ps(f)); }

81
    static __forceinline void storeu(const vboolf8& mask, void* ptr, const vuint8& f) { _mm256_maskstore_ps((float*)ptr,_mm256_castps_si256(mask.v),_mm256_castsi256_ps(f)); }

82


83
    static __forceinline void store_nt(void* ptr, const vuint8& v) {

84
      _mm256_stream_ps((float*)ptr,_mm256_castsi256_ps(v));

85
    }

86


87
    static __forceinline vuint8 load(const unsigned char* ptr) {

88
      vuint4 il = vuint4::load(ptr+0);

89
      vuint4 ih = vuint4::load(ptr+4);

90
      return vuint8(il,ih);

91
    }

92


93
    static __forceinline vuint8 loadu(const unsigned char* ptr) {

94
      vuint4 il = vuint4::loadu(ptr+0);

95
      vuint4 ih = vuint4::loadu(ptr+4);

96
      return vuint8(il,ih);

97
    }

98


99
    static __forceinline vuint8 load(const unsigned short* ptr) {

100
      vuint4 il = vuint4::load(ptr+0);

101
      vuint4 ih = vuint4::load(ptr+4);

102
      return vuint8(il,ih);

103
    }

104


105
    static __forceinline vuint8 loadu(const unsigned short* ptr) {

106
      vuint4 il = vuint4::loadu(ptr+0);

107
      vuint4 ih = vuint4::loadu(ptr+4);

108
      return vuint8(il,ih);

109
    }

110


111
    static __forceinline void store(unsigned char* ptr, const vuint8& i) {

112
      vuint4 il(i.vl);

113
      vuint4 ih(i.vh);

114
      vuint4::store(ptr + 0,il);

115
      vuint4::store(ptr + 4,ih);

116
    }

117


118
    static __forceinline void store(unsigned short* ptr, const vuint8& v) {

119
      for (size_t i=0;i<8;i++)

120
        ptr[i] = (unsigned short)v[i];

121
    }

122


123
    template<int scale = 4>

124
    static __forceinline vuint8 gather(const unsigned int* ptr, const vint8& index) {

125
      return vuint8(

126
          *(unsigned int*)(((char*)ptr)+scale*index[0]),

127
          *(unsigned int*)(((char*)ptr)+scale*index[1]),

128
          *(unsigned int*)(((char*)ptr)+scale*index[2]),

129
          *(unsigned int*)(((char*)ptr)+scale*index[3]),

130
          *(unsigned int*)(((char*)ptr)+scale*index[4]),

131
          *(unsigned int*)(((char*)ptr)+scale*index[5]),

132
          *(unsigned int*)(((char*)ptr)+scale*index[6]),

133
          *(unsigned int*)(((char*)ptr)+scale*index[7]));

134
    }

135


136
    template<int scale = 4>

137
    static __forceinline vuint8 gather(const vboolf8& mask, const unsigned int* ptr, const vint8& index) {

138
      vuint8 r = zero;

139
      if (likely(mask[0])) r[0] = *(unsigned int*)(((char*)ptr)+scale*index[0]);

140
      if (likely(mask[1])) r[1] = *(unsigned int*)(((char*)ptr)+scale*index[1]);

141
      if (likely(mask[2])) r[2] = *(unsigned int*)(((char*)ptr)+scale*index[2]);

142
      if (likely(mask[3])) r[3] = *(unsigned int*)(((char*)ptr)+scale*index[3]);

143
      if (likely(mask[4])) r[4] = *(unsigned int*)(((char*)ptr)+scale*index[4]);

144
      if (likely(mask[5])) r[5] = *(unsigned int*)(((char*)ptr)+scale*index[5]);

145
      if (likely(mask[6])) r[6] = *(unsigned int*)(((char*)ptr)+scale*index[6]);

146
      if (likely(mask[7])) r[7] = *(unsigned int*)(((char*)ptr)+scale*index[7]);

147
      return r;

148
    }

149


150
    template<int scale = 4>

151
    static __forceinline void scatter(void* ptr, const vint8& ofs, const vuint8& v)

152
    {

153
      *(unsigned int*)(((char*)ptr)+scale*ofs[0]) = v[0];

154
      *(unsigned int*)(((char*)ptr)+scale*ofs[1]) = v[1];

155
      *(unsigned int*)(((char*)ptr)+scale*ofs[2]) = v[2];

156
      *(unsigned int*)(((char*)ptr)+scale*ofs[3]) = v[3];

157
      *(unsigned int*)(((char*)ptr)+scale*ofs[4]) = v[4];

158
      *(unsigned int*)(((char*)ptr)+scale*ofs[5]) = v[5];

159
      *(unsigned int*)(((char*)ptr)+scale*ofs[6]) = v[6];

160
      *(unsigned int*)(((char*)ptr)+scale*ofs[7]) = v[7];

161
    }

162


163
    template<int scale = 4>

164
    static __forceinline void scatter(const vboolf8& mask, void* ptr, const vint8& ofs, const vuint8& v)

165
    {

166
      if (likely(mask[0])) *(unsigned int*)(((char*)ptr)+scale*ofs[0]) = v[0];

167
      if (likely(mask[1])) *(unsigned int*)(((char*)ptr)+scale*ofs[1]) = v[1];

168
      if (likely(mask[2])) *(unsigned int*)(((char*)ptr)+scale*ofs[2]) = v[2];

169
      if (likely(mask[3])) *(unsigned int*)(((char*)ptr)+scale*ofs[3]) = v[3];

170
      if (likely(mask[4])) *(unsigned int*)(((char*)ptr)+scale*ofs[4]) = v[4];

171
      if (likely(mask[5])) *(unsigned int*)(((char*)ptr)+scale*ofs[5]) = v[5];

172
      if (likely(mask[6])) *(unsigned int*)(((char*)ptr)+scale*ofs[6]) = v[6];

173
      if (likely(mask[7])) *(unsigned int*)(((char*)ptr)+scale*ofs[7]) = v[7];

174
    }

175


176


177
    static __forceinline vuint8 broadcast64(const long long& a) { return _mm256_set1_epi64x(a); }

178
    

179
    ////////////////////////////////////////////////////////////////////////////////

180
    /// Array Access

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

182


183
    __forceinline const unsigned int& operator [](size_t index) const { assert(index < 8); return i[index]; }

184
    __forceinline       unsigned int& operator [](size_t index)       { assert(index < 8); return i[index]; }

185
  };

186


187
  ////////////////////////////////////////////////////////////////////////////////

188
  /// Unary Operators

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

190


191
  __forceinline vboolf8 asBool(const vuint8& a) { return _mm256_castsi256_ps(a); }

192


193
  __forceinline vuint8 operator +(const vuint8& a) { return a; }

194


195
  ////////////////////////////////////////////////////////////////////////////////

196
  /// Binary Operators

197
  ////////////////////////////////////////////////////////////////////////////////

198


199
  __forceinline vuint8 operator +(const vuint8& a, const vuint8& b) { return vuint8(_mm_add_epi32(a.vl, b.vl), _mm_add_epi32(a.vh, b.vh)); }

200
  __forceinline vuint8 operator +(const vuint8& a, unsigned int          b) { return a + vuint8(b); }

201
  __forceinline vuint8 operator +(unsigned int          a, const vuint8& b) { return vuint8(a) + b; }

202


203
  __forceinline vuint8 operator -(const vuint8& a, const vuint8& b) { return vuint8(_mm_sub_epi32(a.vl, b.vl), _mm_sub_epi32(a.vh, b.vh)); }

204
  __forceinline vuint8 operator -(const vuint8& a, unsigned int          b) { return a - vuint8(b); }

205
  __forceinline vuint8 operator -(unsigned int          a, const vuint8& b) { return vuint8(a) - b; }

206


207
  //__forceinline vuint8 operator *(const vuint8& a, const vuint8& b) { return vuint8(_mm_mullo_epu32(a.vl, b.vl), _mm_mullo_epu32(a.vh, b.vh)); }

208
  //__forceinline vuint8 operator *(const vuint8& a, unsigned int          b) { return a * vuint8(b); }

209
  //__forceinline vuint8 operator *(unsigned int          a, const vuint8& b) { return vuint8(a) * b; }

210


211
  __forceinline vuint8 operator &(const vuint8& a, const vuint8& b) { return _mm256_castps_si256(_mm256_and_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

212
  __forceinline vuint8 operator &(const vuint8& a, unsigned int          b) { return a & vuint8(b); }

213
  __forceinline vuint8 operator &(unsigned int          a, const vuint8& b) { return vuint8(a) & b; }

214


215
  __forceinline vuint8 operator |(const vuint8& a, const vuint8& b) { return _mm256_castps_si256(_mm256_or_ps (_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

216
  __forceinline vuint8 operator |(const vuint8& a, unsigned int          b) { return a | vuint8(b); }

217
  __forceinline vuint8 operator |(unsigned int          a, const vuint8& b) { return vuint8(a) | b; }

218


219
  __forceinline vuint8 operator ^(const vuint8& a, const vuint8& b) { return _mm256_castps_si256(_mm256_xor_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

220
  __forceinline vuint8 operator ^(const vuint8& a, unsigned int          b) { return a ^ vuint8(b); }

221
  __forceinline vuint8 operator ^(unsigned int          a, const vuint8& b) { return vuint8(a) ^ b; }

222


223
  __forceinline vuint8 operator <<(const vuint8& a, unsigned int n) { return vuint8(_mm_slli_epi32(a.vl, n), _mm_slli_epi32(a.vh, n)); }

224
  __forceinline vuint8 operator >>(const vuint8& a, unsigned int n) { return vuint8(_mm_srai_epi32(a.vl, n), _mm_srli_epi32(a.vh, n)); }

225


226
  __forceinline vuint8 sll (const vuint8& a, unsigned int b) { return vuint8(_mm_slli_epi32(a.vl, b), _mm_slli_epi32(a.vh, b)); }

227
  __forceinline vuint8 sra (const vuint8& a, unsigned int b) { return vuint8(_mm_srai_epi32(a.vl, b), _mm_srai_epi32(a.vh, b)); }

228
  __forceinline vuint8 srl (const vuint8& a, unsigned int b) { return vuint8(_mm_srli_epi32(a.vl, b), _mm_srli_epi32(a.vh, b)); }

229
  

230
  __forceinline vuint8 min(const vuint8& a, const vuint8& b) { return vuint8(_mm_min_epu32(a.vl, b.vl), _mm_min_epu32(a.vh, b.vh)); }

231
  __forceinline vuint8 min(const vuint8& a, unsigned int          b) { return min(a,vuint8(b)); }

232
  __forceinline vuint8 min(unsigned int          a, const vuint8& b) { return min(vuint8(a),b); }

233


234
  __forceinline vuint8 max(const vuint8& a, const vuint8& b) { return vuint8(_mm_max_epu32(a.vl, b.vl), _mm_max_epu32(a.vh, b.vh)); }

235
  __forceinline vuint8 max(const vuint8& a, unsigned int          b) { return max(a,vuint8(b)); }

236
  __forceinline vuint8 max(unsigned int          a, const vuint8& b) { return max(vuint8(a),b); }

237


238
  ////////////////////////////////////////////////////////////////////////////////

239
  /// Assignment Operators

240
  ////////////////////////////////////////////////////////////////////////////////

241


242
  __forceinline vuint8& operator +=(vuint8& a, const vuint8& b) { return a = a + b; }

243
  __forceinline vuint8& operator +=(vuint8& a, unsigned int          b) { return a = a + b; }

244
  

245
  __forceinline vuint8& operator -=(vuint8& a, const vuint8& b) { return a = a - b; }

246
  __forceinline vuint8& operator -=(vuint8& a, unsigned int          b) { return a = a - b; }

247
  

248
  //__forceinline vuint8& operator *=(vuint8& a, const vuint8& b) { return a = a * b; }

249
  //__forceinline vuint8& operator *=(vuint8& a, unsigned int          b) { return a = a * b; }

250
  

251
  __forceinline vuint8& operator &=(vuint8& a, const vuint8& b) { return a = a & b; }

252
  __forceinline vuint8& operator &=(vuint8& a, unsigned int          b) { return a = a & b; }

253
  

254
  __forceinline vuint8& operator |=(vuint8& a, const vuint8& b) { return a = a | b; }

255
  __forceinline vuint8& operator |=(vuint8& a, unsigned int          b) { return a = a | b; }

256
  

257
  __forceinline vuint8& operator <<=(vuint8& a, unsigned int b) { return a = a << b; }

258
  __forceinline vuint8& operator >>=(vuint8& a, unsigned int b) { return a = a >> b; }

259


260
  ////////////////////////////////////////////////////////////////////////////////

261
  /// Comparison Operators + Select

262
  ////////////////////////////////////////////////////////////////////////////////

263


264
  __forceinline vboolf8 operator ==(const vuint8& a, const vuint8& b) { return vboolf8(_mm_castsi128_ps(_mm_cmpeq_epi32 (a.vl, b.vl)),

265
                                                                                       _mm_castsi128_ps(_mm_cmpeq_epi32 (a.vh, b.vh))); }

266
  __forceinline vboolf8 operator ==(const vuint8& a, unsigned int          b) { return a == vuint8(b); }

267
  __forceinline vboolf8 operator ==(unsigned int          a, const vuint8& b) { return vuint8(a) == b; }

268
  

269
  __forceinline vboolf8 operator !=(const vuint8& a, const vuint8& b) { return !(a == b); }

270
  __forceinline vboolf8 operator !=(const vuint8& a, unsigned int          b) { return a != vuint8(b); }

271
  __forceinline vboolf8 operator !=(unsigned int          a, const vuint8& b) { return vuint8(a) != b; }

272
  

273
  //__forceinline vboolf8 operator < (const vuint8& a, const vuint8& b) { return vboolf8(_mm_castsi128_ps(_mm_cmplt_epu32 (a.vl, b.vl)),

274
  //                                                                                     _mm_castsi128_ps(_mm_cmplt_epu32 (a.vh, b.vh))); }

275
  //__forceinline vboolf8 operator < (const vuint8& a, unsigned int          b) { return a <  vuint8(b); }

276
  //__forceinline vboolf8 operator < (unsigned int          a, const vuint8& b) { return vuint8(a) <  b; }

277
  

278
  //__forceinline vboolf8 operator >=(const vuint8& a, const vuint8& b) { return !(a <  b); }

279
  //__forceinline vboolf8 operator >=(const vuint8& a, unsigned int          b) { return a >= vuint8(b); }

280
  //__forceinline vboolf8 operator >=(unsigned int          a, const vuint8& b) { return vuint8(a) >= b; }

281


282
  //__forceinline vboolf8 operator > (const vuint8& a, const vuint8& b) { return vboolf8(_mm_castsi128_ps(_mm_cmpgt_epu32 (a.vl, b.vl)),

283
  //                                                                                     _mm_castsi128_ps(_mm_cmpgt_epu32 (a.vh, b.vh))); }

284
  //__forceinline vboolf8 operator > (const vuint8& a, unsigned int          b) { return a >  vuint8(b); }

285
  //__forceinline vboolf8 operator > (unsigned int          a, const vuint8& b) { return vuint8(a) >  b; }

286


287
  //__forceinline vboolf8 operator <=(const vuint8& a, const vuint8& b) { return !(a >  b); }

288
  //__forceinline vboolf8 operator <=(const vuint8& a, unsigned int          b) { return a <= vuint8(b); }

289
  //__forceinline vboolf8 operator <=(unsigned int          a, const vuint8& b) { return vuint8(a) <= b; }

290


291
  __forceinline vboolf8 eq(const vuint8& a, const vuint8& b) { return a == b; }

292
  __forceinline vboolf8 ne(const vuint8& a, const vuint8& b) { return a != b; }

293


294
  __forceinline vboolf8 eq(const vboolf8& mask, const vuint8& a, const vuint8& b) { return mask & (a == b); }

295
  __forceinline vboolf8 ne(const vboolf8& mask, const vuint8& a, const vuint8& b) { return mask & (a != b); }

296


297
  __forceinline vuint8 select(const vboolf8& m, const vuint8& t, const vuint8& f) {

298
    return _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(f), _mm256_castsi256_ps(t), m)); 

299
  }

300


301


302
  ////////////////////////////////////////////////////////////////////////////////

303
  /// Movement/Shifting/Shuffling Functions

304
  ////////////////////////////////////////////////////////////////////////////////

305


306
  __forceinline vuint8 unpacklo(const vuint8& a, const vuint8& b) { return _mm256_castps_si256(_mm256_unpacklo_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

307
  __forceinline vuint8 unpackhi(const vuint8& a, const vuint8& b) { return _mm256_castps_si256(_mm256_unpackhi_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))); }

308


309
  template<int i>

310
  __forceinline vuint8 shuffle(const vuint8& v) {

311
    return _mm256_castps_si256(_mm256_permute_ps(_mm256_castsi256_ps(v), _MM_SHUFFLE(i, i, i, i)));

312
  }

313


314
  template<int i0, int i1>

315
  __forceinline vuint8 shuffle4(const vuint8& v) {

316
    return _mm256_permute2f128_si256(v, v, (i1 << 4) | (i0 << 0));

317
  }

318


319
  template<int i0, int i1>

320
  __forceinline vuint8 shuffle4(const vuint8& a, const vuint8& b) {

321
    return _mm256_permute2f128_si256(a, b, (i1 << 4) | (i0 << 0));

322
  }

323


324
  template<int i0, int i1, int i2, int i3>

325
  __forceinline vuint8 shuffle(const vuint8& v) {

326
    return _mm256_castps_si256(_mm256_permute_ps(_mm256_castsi256_ps(v), _MM_SHUFFLE(i3, i2, i1, i0)));

327
  }

328


329
  template<int i0, int i1, int i2, int i3>

330
  __forceinline vuint8 shuffle(const vuint8& a, const vuint8& b) {

331
    return _mm256_castps_si256(_mm256_shuffle_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b), _MM_SHUFFLE(i3, i2, i1, i0)));

332
  }

333


334
  template<> __forceinline vuint8 shuffle<0, 0, 2, 2>(const vuint8& v) { return _mm256_castps_si256(_mm256_moveldup_ps(_mm256_castsi256_ps(v))); }

335
  template<> __forceinline vuint8 shuffle<1, 1, 3, 3>(const vuint8& v) { return _mm256_castps_si256(_mm256_movehdup_ps(_mm256_castsi256_ps(v))); }

336
  template<> __forceinline vuint8 shuffle<0, 1, 0, 1>(const vuint8& v) { return _mm256_castps_si256(_mm256_castpd_ps(_mm256_movedup_pd(_mm256_castps_pd(_mm256_castsi256_ps(v))))); }

337


338
  template<int i> __forceinline vuint8 insert4(const vuint8& a, const vuint4& b) { return _mm256_insertf128_si256(a, b, i); }

339
  template<int i> __forceinline vuint4 extract4(const vuint8& a) { return _mm256_extractf128_si256(a, i); }

340
  template<> __forceinline vuint4 extract4<0>(const vuint8& a) { return _mm256_castsi256_si128(a); }

341


342
  __forceinline int toScalar(const vuint8& v) { return _mm_cvtsi128_si32(_mm256_castsi256_si128(v)); }

343


344


345
  ////////////////////////////////////////////////////////////////////////////////

346
  /// Reductions

347
  ////////////////////////////////////////////////////////////////////////////////

348


349
  //__forceinline vuint8 vreduce_min2(const vuint8& v) { return min(v,shuffle<1,0,3,2>(v)); }

350
  //__forceinline vuint8 vreduce_min4(const vuint8& v) { vuint8 v1 = vreduce_min2(v); return min(v1,shuffle<2,3,0,1>(v1)); }

351
  //__forceinline vuint8 vreduce_min (const vuint8& v) { vuint8 v1 = vreduce_min4(v); return min(v1,shuffle4<1,0>(v1)); }

352


353
  //__forceinline vuint8 vreduce_max2(const vuint8& v) { return max(v,shuffle<1,0,3,2>(v)); }

354
  //__forceinline vuint8 vreduce_max4(const vuint8& v) { vuint8 v1 = vreduce_max2(v); return max(v1,shuffle<2,3,0,1>(v1)); }

355
  //__forceinline vuint8 vreduce_max (const vuint8& v) { vuint8 v1 = vreduce_max4(v); return max(v1,shuffle4<1,0>(v1)); }

356


357
  __forceinline vuint8 vreduce_add2(const vuint8& v) { return v + shuffle<1,0,3,2>(v); }

358
  __forceinline vuint8 vreduce_add4(const vuint8& v) { vuint8 v1 = vreduce_add2(v); return v1 + shuffle<2,3,0,1>(v1); }

359
  __forceinline vuint8 vreduce_add (const vuint8& v) { vuint8 v1 = vreduce_add4(v); return v1 + shuffle4<1,0>(v1); }

360


361
  //__forceinline int reduce_min(const vuint8& v) { return toScalar(vreduce_min(v)); }

362
  //__forceinline int reduce_max(const vuint8& v) { return toScalar(vreduce_max(v)); }

363
  __forceinline int reduce_add(const vuint8& v) { return toScalar(vreduce_add(v)); }

364


365
  //__forceinline size_t select_min(const vuint8& v) { return bsf(movemask(v == vreduce_min(v))); }

366
  //__forceinline size_t select_max(const vuint8& v) { return bsf(movemask(v == vreduce_max(v))); }

367


368
  //__forceinline size_t select_min(const vboolf8& valid, const vuint8& v) { const vuint8 a = select(valid,v,vuint8(pos_inf)); return bsf(movemask(valid & (a == vreduce_min(a)))); }

369
  //__forceinline size_t select_max(const vboolf8& valid, const vuint8& v) { const vuint8 a = select(valid,v,vuint8(neg_inf)); return bsf(movemask(valid & (a == vreduce_max(a)))); }

370


371
  ////////////////////////////////////////////////////////////////////////////////

372
  /// Output Operators

373
  ////////////////////////////////////////////////////////////////////////////////

374


375
  __forceinline embree_ostream operator <<(embree_ostream cout, const vuint8& a) {

376
    return cout << "<" << a[0] << ", " << a[1] << ", " << a[2] << ", " << a[3] << ", " << a[4] << ", " << a[5] << ", " << a[6] << ", " << a[7] << ">";

377
  }

378
}

379


380
#undef vboolf

381
#undef vboold

382
#undef vint

383
#undef vuint

384
#undef vllong

385
#undef vfloat

386
#undef vdouble

387


388