Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
godotengine
GitHub Repository: godotengine/godot
 Path: blob/master/thirdparty/embree/common/simd/vuint8_avx2.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
      unsigned int i[8];

30
    }; 

31


32
    ////////////////////////////////////////////////////////////////////////////////

33
    /// Constructors, Assignment & Cast Operators

34
    ////////////////////////////////////////////////////////////////////////////////

35


36
    __forceinline vuint() {}

37
    __forceinline vuint(const vuint8& a) { v = a.v; }

38
    __forceinline vuint8& operator =(const vuint8& a) { v = a.v; return *this; }

39


40
    __forceinline vuint(__m256i a) : v(a) {}

41
    __forceinline operator const __m256i&() const { return v; }

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

43


44
    __forceinline explicit vuint(const vuint4& a) : v(_mm256_insertf128_si256(_mm256_castsi128_si256(a),a,1)) {}

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

46
    __forceinline vuint(const __m128i& a, const __m128i& b) : v(_mm256_insertf128_si256(_mm256_castsi128_si256(a),b,1)) {}

47
 

48
    __forceinline explicit vuint(const unsigned int* a) : v(_mm256_castps_si256(_mm256_loadu_ps((const float*)a))) {}

49
    __forceinline vuint(unsigned int a) : v(_mm256_set1_epi32(a)) {}

50
    __forceinline vuint(unsigned int a, unsigned int b) : v(_mm256_set_epi32(b, a, b, a, b, a, b, a)) {}

51
    __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)) {}

52
    __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 h) : v(_mm256_set_epi32(h, g, f, e, d, c, b, a)) {}

53


54
    __forceinline explicit vuint(__m256 a) : v(_mm256_cvtps_epi32(a)) {}

55


56
#if defined(__AVX512VL__)

57
    __forceinline explicit vuint(const vboolf8& a) : v(_mm256_movm_epi32(a)) {}

58
#else

59
    __forceinline explicit vuint(const vboolf8& a) : v(_mm256_castps_si256((__m256)a)) {}

60
#endif

61


62
    ////////////////////////////////////////////////////////////////////////////////

63
    /// Constants

64
    ////////////////////////////////////////////////////////////////////////////////

65


66
    __forceinline vuint(ZeroTy)   : v(_mm256_setzero_si256()) {}

67
    __forceinline vuint(OneTy)    : v(_mm256_set1_epi32(1)) {}

68
    __forceinline vuint(PosInfTy) : v(_mm256_set1_epi32(pos_inf)) {}

69
    __forceinline vuint(NegInfTy) : v(_mm256_set1_epi32(neg_inf)) {}

70
    __forceinline vuint(StepTy)   : v(_mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0)) {}

71
    __forceinline vuint(UndefinedTy) : v(_mm256_undefined_si256()) {}

72


73
    ////////////////////////////////////////////////////////////////////////////////

74
    /// Loads and Stores

75
    ////////////////////////////////////////////////////////////////////////////////

76


77
    static __forceinline vuint8 load(const unsigned char* ptr)  { return _mm256_cvtepu8_epi32(_mm_loadl_epi64((__m128i*)ptr)); }

78
    static __forceinline vuint8 loadu(const unsigned char* ptr) { return _mm256_cvtepu8_epi32(_mm_loadl_epi64((__m128i*)ptr)); }

79
    static __forceinline vuint8 load(const unsigned short* ptr)  { return _mm256_cvtepu16_epi32(_mm_load_si128((__m128i*)ptr)); }

80
    static __forceinline vuint8 loadu(const unsigned short* ptr) { return _mm256_cvtepu16_epi32(_mm_loadu_si128((__m128i*)ptr)); }

81


82
    static __forceinline vuint8 load(const void* ptr) { return _mm256_load_si256((__m256i*)ptr); }

83
    static __forceinline vuint8 loadu(const void* ptr) { return _mm256_loadu_si256((__m256i*)ptr); }

84


85
    static __forceinline void store (void* ptr, const vuint8& v) { _mm256_store_si256((__m256i*)ptr,v); }

86
    static __forceinline void storeu(void* ptr, const vuint8& v) { _mm256_storeu_ps((float*)ptr,_mm256_castsi256_ps(v)); }

87


88
#if defined(__AVX512VL__)

89


90
    static __forceinline vuint8 compact(const vboolf8& mask, vuint8 &v) {

91
      return _mm256_mask_compress_epi32(v, mask, v);

92
    }

93
    static __forceinline vuint8 compact(const vboolf8& mask, vuint8 &a, const vuint8& b) {

94
      return _mm256_mask_compress_epi32(a, mask, b);

95
    }

96


97
    static __forceinline vuint8 load (const vboolf8& mask, const void* ptr) { return _mm256_mask_load_epi32 (_mm256_setzero_si256(),mask,ptr); }

98
    static __forceinline vuint8 loadu(const vboolf8& mask, const void* ptr) { return _mm256_mask_loadu_epi32(_mm256_setzero_si256(),mask,ptr); }

99


100
    static __forceinline void store (const vboolf8& mask, void* ptr, const vuint8& v) { _mm256_mask_store_epi32 (ptr,mask,v); }

101
    static __forceinline void storeu(const vboolf8& mask, void* ptr, const vuint8& v) { _mm256_mask_storeu_epi32(ptr,mask,v); }

102
#else

103
    static __forceinline vuint8 load (const vboolf8& mask, const void* ptr) { return _mm256_castps_si256(_mm256_maskload_ps((float*)ptr,mask)); }

104
    static __forceinline vuint8 loadu(const vboolf8& mask, const void* ptr) { return _mm256_castps_si256(_mm256_maskload_ps((float*)ptr,mask)); }

105


106
    static __forceinline void store (const vboolf8& mask, void* ptr, const vuint8& v) { _mm256_maskstore_epi32((int*)ptr,mask,v); }

107
    static __forceinline void storeu(const vboolf8& mask, void* ptr, const vuint8& v) { _mm256_maskstore_epi32((int*)ptr,mask,v); }

108
#endif

109
    

110
    static __forceinline vuint8 load_nt(void* ptr) {

111
      return _mm256_stream_load_si256((__m256i*)ptr);

112
    }

113


114
    static __forceinline void store_nt(void* ptr, const vuint8& v) {

115
      _mm256_stream_ps((float*)ptr,_mm256_castsi256_ps(v));

116
    }

117


118
    static __forceinline void store(unsigned char* ptr, const vuint8& i)

119
    {

120
      for (size_t j=0; j<8; j++)

121
        ptr[j] = i[j];

122
    }

123


124
    static __forceinline void store(unsigned short* ptr, const vuint8& v) {

125
      for (size_t i=0;i<8;i++)

126
        ptr[i] = (unsigned short)v[i];

127
    }

128


129
    template<int scale = 4>

130
    static __forceinline vuint8 gather(const unsigned int *const ptr, const vint8& index) {

131
      return _mm256_i32gather_epi32((const int*) ptr, index, scale);

132
    }

133


134
    template<int scale = 4>

135
    static __forceinline vuint8 gather(const vboolf8& mask, const unsigned int *const ptr, const vint8& index) {

136
      vuint8 r = zero;

137
#if defined(__AVX512VL__)

138
      return _mm256_mmask_i32gather_epi32(r, mask, index, (const int*) ptr, scale);

139
#else

140
      return _mm256_mask_i32gather_epi32(r, (const int*) ptr, index, mask, scale);

141
#endif

142
    }

143


144
    template<int scale = 4>

145
    static __forceinline void scatter(void* ptr, const vint8& ofs, const vuint8& v)

146
    {

147
#if defined(__AVX512VL__)

148
      _mm256_i32scatter_epi32((int*)ptr, ofs, v, scale);

149
#else

150
      *(unsigned int*)(((char*)ptr)+scale*ofs[0]) = v[0];

151
      *(unsigned int*)(((char*)ptr)+scale*ofs[1]) = v[1];

152
      *(unsigned int*)(((char*)ptr)+scale*ofs[2]) = v[2];

153
      *(unsigned int*)(((char*)ptr)+scale*ofs[3]) = v[3];

154
      *(unsigned int*)(((char*)ptr)+scale*ofs[4]) = v[4];

155
      *(unsigned int*)(((char*)ptr)+scale*ofs[5]) = v[5];

156
      *(unsigned int*)(((char*)ptr)+scale*ofs[6]) = v[6];

157
      *(unsigned int*)(((char*)ptr)+scale*ofs[7]) = v[7];

158
#endif

159
    }

160


161
    template<int scale = 4>

162
    static __forceinline void scatter(const vboolf8& mask, void* ptr, const vint8& ofs, const vuint8& v)

163
    {

164
#if defined(__AVX512VL__)

165
      _mm256_mask_i32scatter_epi32((int*)ptr, mask, ofs, v, scale);

166
#else

167
      if (likely(mask[0])) *(unsigned int*)(((char*)ptr)+scale*ofs[0]) = v[0];

168
      if (likely(mask[1])) *(unsigned int*)(((char*)ptr)+scale*ofs[1]) = v[1];

169
      if (likely(mask[2])) *(unsigned int*)(((char*)ptr)+scale*ofs[2]) = v[2];

170
      if (likely(mask[3])) *(unsigned int*)(((char*)ptr)+scale*ofs[3]) = v[3];

171
      if (likely(mask[4])) *(unsigned int*)(((char*)ptr)+scale*ofs[4]) = v[4];

172
      if (likely(mask[5])) *(unsigned int*)(((char*)ptr)+scale*ofs[5]) = v[5];

173
      if (likely(mask[6])) *(unsigned int*)(((char*)ptr)+scale*ofs[6]) = v[6];

174
      if (likely(mask[7])) *(unsigned int*)(((char*)ptr)+scale*ofs[7]) = v[7];

175
#endif

176
    }

177


178
    static __forceinline vuint8 broadcast64(const long long &a) { return _mm256_set1_epi64x(a); }

179
    

180
    ////////////////////////////////////////////////////////////////////////////////

181
    /// Array Access

182
    ////////////////////////////////////////////////////////////////////////////////

183


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

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

186
  };

187


188
  ////////////////////////////////////////////////////////////////////////////////

189
  /// Unary Operators

190
  ////////////////////////////////////////////////////////////////////////////////

191


192
#if defined(__AVX512VL__)

193
  __forceinline vboolf8 asBool(const vuint8& a) { return _mm256_movepi32_mask(a); }

194
#else

195
  __forceinline vboolf8 asBool(const vuint8& a) { return _mm256_castsi256_ps(a); }

196
#endif

197


198
  __forceinline vuint8 operator +(const vuint8& a) { return a; }

199


200
  ////////////////////////////////////////////////////////////////////////////////

201
  /// Binary Operators

202
  ////////////////////////////////////////////////////////////////////////////////

203


204
  __forceinline vuint8 operator +(const vuint8& a, const vuint8& b) { return _mm256_add_epi32(a, b); }

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

206
  __forceinline vuint8 operator +(unsigned int          a, const vuint8& b) { return vuint8(a) + b; }

207


208
  __forceinline vuint8 operator -(const vuint8& a, const vuint8& b) { return _mm256_sub_epi32(a, b); }

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

210
  __forceinline vuint8 operator -(unsigned int          a, const vuint8& b) { return vuint8(a) - b; }

211


212
  //__forceinline vuint8 operator *(const vuint8& a, const vuint8& b) { return _mm256_mullo_epu32(a, b); }

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

214
  //__forceinline vuint8 operator *(unsigned int          a, const vuint8& b) { return vuint8(a) * b; }

215


216
  __forceinline vuint8 operator &(const vuint8& a, const vuint8& b) { return _mm256_and_si256(a, b); }

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

218
  __forceinline vuint8 operator &(unsigned int          a, const vuint8& b) { return vuint8(a) & b; }

219


220
  __forceinline vuint8 operator |(const vuint8& a, const vuint8& b) { return _mm256_or_si256(a, b); }

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

222
  __forceinline vuint8 operator |(unsigned int          a, const vuint8& b) { return vuint8(a) | b; }

223


224
  __forceinline vuint8 operator ^(const vuint8& a, const vuint8& b) { return _mm256_xor_si256(a, b); }

225
  __forceinline vuint8 operator ^(const vuint8& a, unsigned int          b) { return a ^ vuint8(b); }

226
  __forceinline vuint8 operator ^(unsigned int          a, const vuint8& b) { return vuint8(a) ^ b; }

227


228
  __forceinline vuint8 operator <<(const vuint8& a, unsigned int n) { return _mm256_slli_epi32(a, n); }

229
  __forceinline vuint8 operator >>(const vuint8& a, unsigned int n) { return _mm256_srli_epi32(a, n); }

230


231
  __forceinline vuint8 operator <<(const vuint8& a, const vuint8& n) { return _mm256_sllv_epi32(a, n); }

232
  __forceinline vuint8 operator >>(const vuint8& a, const vuint8& n) { return _mm256_srlv_epi32(a, n); }

233


234
  __forceinline vuint8 sll(const vuint8& a, unsigned int b) { return _mm256_slli_epi32(a, b); }

235
  __forceinline vuint8 sra(const vuint8& a, unsigned int b) { return _mm256_srai_epi32(a, b); }

236
  __forceinline vuint8 srl(const vuint8& a, unsigned int b) { return _mm256_srli_epi32(a, b); }

237


238
  __forceinline vuint8 sll(const vuint8& a, const vuint8& b) { return _mm256_sllv_epi32(a, b); }

239
  __forceinline vuint8 sra(const vuint8& a, const vuint8& b) { return _mm256_srav_epi32(a, b); }

240
  __forceinline vuint8 srl(const vuint8& a, const vuint8& b) { return _mm256_srlv_epi32(a, b); }

241
  

242
  __forceinline vuint8 min(const vuint8& a, const vuint8& b) { return _mm256_min_epu32(a, b); }

243
  __forceinline vuint8 min(const vuint8& a, unsigned int          b) { return min(a,vuint8(b)); }

244
  __forceinline vuint8 min(unsigned int          a, const vuint8& b) { return min(vuint8(a),b); }

245


246
  __forceinline vuint8 max(const vuint8& a, const vuint8& b) { return _mm256_max_epu32(a, b); }

247
  __forceinline vuint8 max(const vuint8& a, unsigned int          b) { return max(a,vuint8(b)); }

248
  __forceinline vuint8 max(unsigned int          a, const vuint8& b) { return max(vuint8(a),b); }

249


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

251
  /// Assignment Operators

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

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, const vuint8& b) { return a = a - b; }

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

259
  

260
  //__forceinline vuint8& operator *=(vuint8& a, const vuint8& b) { return a = a * b; }

261
  //__forceinline vuint8& operator *=(vuint8& a, unsigned int          b) { return a = a * b; }

262
  

263
  __forceinline vuint8& operator &=(vuint8& a, const vuint8& b) { return a = a & b; }

264
  __forceinline vuint8& operator &=(vuint8& a, unsigned int          b) { return a = a & b; }

265
  

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

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

268
  

269
  __forceinline vuint8& operator <<=(vuint8& a, const unsigned int b) { return a = a << b; }

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

271


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

273
  /// Comparison Operators + Select

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

275


276
#if defined(__AVX512VL__)

277
  __forceinline vboolf8 operator ==(const vuint8& a, const vuint8& b) { return _mm256_cmp_epu32_mask(a,b,_MM_CMPINT_EQ); }

278
  __forceinline vboolf8 operator !=(const vuint8& a, const vuint8& b) { return _mm256_cmp_epu32_mask(a,b,_MM_CMPINT_NE); }

279
  __forceinline vboolf8 operator < (const vuint8& a, const vuint8& b) { return _mm256_cmp_epu32_mask(a,b,_MM_CMPINT_LT); }

280
  __forceinline vboolf8 operator >=(const vuint8& a, const vuint8& b) { return _mm256_cmp_epu32_mask(a,b,_MM_CMPINT_GE); }

281
  __forceinline vboolf8 operator > (const vuint8& a, const vuint8& b) { return _mm256_cmp_epu32_mask(a,b,_MM_CMPINT_GT); }

282
  __forceinline vboolf8 operator <=(const vuint8& a, const vuint8& b) { return _mm256_cmp_epu32_mask(a,b,_MM_CMPINT_LE); }

283


284
  __forceinline vuint8 select(const vboolf8& m, const vuint8& t, const vuint8& f) {

285
    return _mm256_mask_blend_epi32(m, (__m256i)f, (__m256i)t);

286
  }

287
#else

288
  __forceinline vboolf8 operator ==(const vuint8& a, const vuint8& b) { return _mm256_castsi256_ps(_mm256_cmpeq_epi32(a, b)); }

289
  __forceinline vboolf8 operator !=(const vuint8& a, const vuint8& b) { return !(a == b); }

290
  //__forceinline vboolf8 operator < (const vuint8& a, const vuint8& b) { return _mm256_castsi256_ps(_mm256_cmpgt_epu32(b, a)); }

291
  //__forceinline vboolf8 operator >=(const vuint8& a, const vuint8& b) { return !(a <  b); }

292
  //__forceinline vboolf8 operator > (const vuint8& a, const vuint8& b) { return _mm256_castsi256_ps(_mm256_cmpgt_epu32(a, b)); }

293
  //__forceinline vboolf8 operator <=(const vuint8& a, const vuint8& b) { return !(a >  b); }

294


295
  __forceinline vuint8 select(const vboolf8& m, const vuint8& t, const vuint8& f) {

296
    return _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(f), _mm256_castsi256_ps(t), m));

297
  }

298
#endif

299


300
  template<int mask>

301
  __forceinline vuint8 select(const vuint8& t, const vuint8& f) {

302
    return _mm256_blend_epi32(f, t, mask);

303
  }

304


305
  __forceinline vboolf8 operator ==(const vuint8& a, unsigned int          b) { return a == vuint8(b); }

306
  __forceinline vboolf8 operator ==(unsigned int          a, const vuint8& b) { return vuint8(a) == b; }

307


308
  __forceinline vboolf8 operator !=(const vuint8& a, unsigned int          b) { return a != vuint8(b); }

309
  __forceinline vboolf8 operator !=(unsigned int          a, const vuint8& b) { return vuint8(a) != b; }

310


311
  //__forceinline vboolf8 operator < (const vuint8& a, unsigned int          b) { return a <  vuint8(b); }

312
  //__forceinline vboolf8 operator < (unsigned int          a, const vuint8& b) { return vuint8(a) <  b; }

313


314
  //__forceinline vboolf8 operator >=(const vuint8& a, unsigned int          b) { return a >= vuint8(b); }

315
  //__forceinline vboolf8 operator >=(unsigned int          a, const vuint8& b) { return vuint8(a) >= b; }

316


317
  //__forceinline vboolf8 operator > (const vuint8& a, unsigned int          b) { return a >  vuint8(b); }

318
  //__forceinline vboolf8 operator > (unsigned int          a, const vuint8& b) { return vuint8(a) >  b; }

319


320
  //__forceinline vboolf8 operator <=(const vuint8& a, unsigned int          b) { return a <= vuint8(b); }

321
  //__forceinline vboolf8 operator <=(unsigned int          a, const vuint8& b) { return vuint8(a) <= b; }

322


323
  __forceinline vboolf8 eq(const vuint8& a, const vuint8& b) { return a == b; }

324
  __forceinline vboolf8 ne(const vuint8& a, const vuint8& b) { return a != b; }

325
  //__forceinline vboolf8 lt(const vuint8& a, const vuint8& b) { return a <  b; }

326
  //__forceinline vboolf8 ge(const vuint8& a, const vuint8& b) { return a >= b; }

327
  //__forceinline vboolf8 gt(const vuint8& a, const vuint8& b) { return a >  b; }

328
  //__forceinline vboolf8 le(const vuint8& a, const vuint8& b) { return a <= b; }

329


330
#if defined(__AVX512VL__)

331
  __forceinline vboolf8 eq(const vboolf8& mask, const vuint8& a, const vuint8& b) { return _mm256_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_EQ); }

332
  __forceinline vboolf8 ne(const vboolf8& mask, const vuint8& a, const vuint8& b) { return _mm256_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_NE); }

333
  __forceinline vboolf8 lt(const vboolf8& mask, const vuint8& a, const vuint8& b) { return _mm256_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_LT); }

334
  __forceinline vboolf8 ge(const vboolf8& mask, const vuint8& a, const vuint8& b) { return _mm256_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_GE); }

335
  __forceinline vboolf8 gt(const vboolf8& mask, const vuint8& a, const vuint8& b) { return _mm256_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_GT); }

336
  __forceinline vboolf8 le(const vboolf8& mask, const vuint8& a, const vuint8& b) { return _mm256_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_LE); }

337
#else

338
  __forceinline vboolf8 eq(const vboolf8& mask, const vuint8& a, const vuint8& b) { return mask & (a == b); }

339
  __forceinline vboolf8 ne(const vboolf8& mask, const vuint8& a, const vuint8& b) { return mask & (a != b); }

340
  //__forceinline vboolf8 lt(const vboolf8& mask, const vuint8& a, const vuint8& b) { return mask & (a <  b); }

341
  //__forceinline vboolf8 ge(const vboolf8& mask, const vuint8& a, const vuint8& b) { return mask & (a >= b); }

342
  //__forceinline vboolf8 gt(const vboolf8& mask, const vuint8& a, const vuint8& b) { return mask & (a >  b); }

343
  //__forceinline vboolf8 le(const vboolf8& mask, const vuint8& a, const vuint8& b) { return mask & (a <= b); }

344
#endif

345


346
  ////////////////////////////////////////////////////////////////////////////////

347
  /// Movement/Shifting/Shuffling Functions

348
  ////////////////////////////////////////////////////////////////////////////////

349


350
  __forceinline vuint8 unpacklo(const vuint8& a, const vuint8& b) { return _mm256_unpacklo_epi32(a, b); }

351
  __forceinline vuint8 unpackhi(const vuint8& a, const vuint8& b) { return _mm256_unpackhi_epi32(a, b); }

352


353
  template<int i>

354
  __forceinline vuint8 shuffle(const vuint8& v) {

355
    return _mm256_castps_si256(_mm256_permute_ps(_mm256_castsi256_ps(v), _MM_SHUFFLE(i, i, i, i)));

356
  }

357


358
  template<int i0, int i1>

359
  __forceinline vuint8 shuffle4(const vuint8& v) {

360
    return _mm256_permute2f128_si256(v, v, (i1 << 4) | (i0 << 0));

361
  }

362


363
  template<int i0, int i1>

364
  __forceinline vuint8 shuffle4(const vuint8& a, const vuint8& b) {

365
    return _mm256_permute2f128_si256(a, b, (i1 << 4) | (i0 << 0));

366
  }

367


368
  template<int i0, int i1, int i2, int i3>

369
  __forceinline vuint8 shuffle(const vuint8& v) {

370
    return _mm256_castps_si256(_mm256_permute_ps(_mm256_castsi256_ps(v), _MM_SHUFFLE(i3, i2, i1, i0)));

371
  }

372


373
  template<int i0, int i1, int i2, int i3>

374
  __forceinline vuint8 shuffle(const vuint8& a, const vuint8& b) {

375
    return _mm256_castps_si256(_mm256_shuffle_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b), _MM_SHUFFLE(i3, i2, i1, i0)));

376
  }

377


378
  template<> __forceinline vuint8 shuffle<0, 0, 2, 2>(const vuint8& v) { return _mm256_castps_si256(_mm256_moveldup_ps(_mm256_castsi256_ps(v))); }

379
  template<> __forceinline vuint8 shuffle<1, 1, 3, 3>(const vuint8& v) { return _mm256_castps_si256(_mm256_movehdup_ps(_mm256_castsi256_ps(v))); }

380
  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))))); }

381


382
  template<int i> __forceinline vuint8 insert4(const vuint8& a, const vuint4& b) { return _mm256_insertf128_si256(a, b, i); }

383
  template<int i> __forceinline vuint4 extract4(const vuint8& a) { return _mm256_extractf128_si256(a, i); }

384
  template<> __forceinline vuint4 extract4<0>(const vuint8& a) { return _mm256_castsi256_si128(a); }

385


386
  __forceinline int toScalar(const vuint8& v) { return _mm_cvtsi128_si32(_mm256_castsi256_si128(v)); }

387


388
#if !defined(__aarch64__)

389
  __forceinline vuint8 permute(const vuint8& v, const __m256i& index) {

390
    return _mm256_permutevar8x32_epi32(v, index);

391
  }

392


393
  __forceinline vuint8 shuffle(const vuint8& v, const __m256i& index) {

394
    return _mm256_castps_si256(_mm256_permutevar_ps(_mm256_castsi256_ps(v), index));

395
  }

396


397
  template<int i>

398
  __forceinline vuint8 align_shift_right(const vuint8& a, const vuint8& b) {

399
#if defined(__AVX512VL__)

400
    return _mm256_alignr_epi32(a, b, i);    

401
#else

402
    return _mm256_alignr_epi8(a, b, 4*i);

403
#endif

404
  }  

405
#endif // !defined(__aarch64__)

406


407
  ////////////////////////////////////////////////////////////////////////////////

408
  /// Reductions

409
  ////////////////////////////////////////////////////////////////////////////////

410


411
  //__forceinline vuint8 vreduce_min2(const vuint8& v) { return min(v,shuffle<1,0,3,2>(v)); }

412
  //__forceinline vuint8 vreduce_min4(const vuint8& v) { vuint8 v1 = vreduce_min2(v); return min(v1,shuffle<2,3,0,1>(v1)); }

413
  //__forceinline vuint8 vreduce_min (const vuint8& v) { vuint8 v1 = vreduce_min4(v); return min(v1,shuffle4<1,0>(v1)); }

414


415
  //__forceinline vuint8 vreduce_max2(const vuint8& v) { return max(v,shuffle<1,0,3,2>(v)); }

416
  //__forceinline vuint8 vreduce_max4(const vuint8& v) { vuint8 v1 = vreduce_max2(v); return max(v1,shuffle<2,3,0,1>(v1)); }

417
  //__forceinline vuint8 vreduce_max (const vuint8& v) { vuint8 v1 = vreduce_max4(v); return max(v1,shuffle4<1,0>(v1)); }

418


419
  __forceinline vuint8 vreduce_add2(const vuint8& v) { return v + shuffle<1,0,3,2>(v); }

420
  __forceinline vuint8 vreduce_add4(const vuint8& v) { vuint8 v1 = vreduce_add2(v); return v1 + shuffle<2,3,0,1>(v1); }

421
  __forceinline vuint8 vreduce_add (const vuint8& v) { vuint8 v1 = vreduce_add4(v); return v1 + shuffle4<1,0>(v1); }

422


423
  //__forceinline int reduce_min(const vuint8& v) { return toScalar(vreduce_min(v)); }

424
  //__forceinline int reduce_max(const vuint8& v) { return toScalar(vreduce_max(v)); }

425
  __forceinline int reduce_add(const vuint8& v) { return toScalar(vreduce_add(v)); }

426


427
  //__forceinline size_t select_min(const vuint8& v) { return bsf(movemask(v == vreduce_min(v))); }

428
  //__forceinline size_t select_max(const vuint8& v) { return bsf(movemask(v == vreduce_max(v))); }

429


430
  //__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)))); }

431
  //__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)))); }

432


433
  ////////////////////////////////////////////////////////////////////////////////

434
  /// Output Operators

435
  ////////////////////////////////////////////////////////////////////////////////

436


437
  __forceinline embree_ostream operator <<(embree_ostream cout, const vuint8& a) {

438
    return cout << "<" << a[0] << ", " << a[1] << ", " << a[2] << ", " << a[3] << ", " << a[4] << ", " << a[5] << ", " << a[6] << ", " << a[7] << ">";

439
  }

440
}

441


442
#undef vboolf

443
#undef vboold

444
#undef vint

445
#undef vuint

446
#undef vllong

447
#undef vfloat

448
#undef vdouble

449


450