1// Copyright 2009-2021 Intel Corporation
2// SPDX-License-Identifier: Apache-2.0
3
4#pragma once
5
6#include "../math/math.h"
7
8#define vboolf vboolf_impl
9#define vboold vboold_impl
10#define vint vint_impl
11#define vuint vuint_impl
12#define vllong vllong_impl
13#define vfloat vfloat_impl
14#define vdouble vdouble_impl
15
16namespace embree
17{
18 /* 4-wide SSE integer type */
19 template<>
20 struct vuint<4>
21 {
22 ALIGNED_STRUCT_(16);
23
24 typedef vboolf4 Bool;
25 typedef vuint4 Int;
26 typedef vfloat4 Float;
27
28 enum { size = 4 }; // number of SIMD elements
29 union { __m128i v; unsigned int i[4]; }; // data
30
31 ////////////////////////////////////////////////////////////////////////////////
32 /// Constructors, Assignment & Cast Operators
33 ////////////////////////////////////////////////////////////////////////////////
34
35 __forceinline vuint() {}
36 __forceinline vuint(const vuint4& a) { v = a.v; }
37 __forceinline vuint4& operator =(const vuint4& a) { v = a.v; return *this; }
38
39 __forceinline vuint(const __m128i a) : v(a) {}
40 __forceinline operator const __m128i&() const { return v; }
41 __forceinline operator __m128i&() { return v; }
42
43
44 __forceinline vuint(unsigned int a) : v(_mm_set1_epi32(i: a)) {}
45 __forceinline vuint(unsigned int a, unsigned int b, unsigned int c, unsigned int d) : v(_mm_set_epi32(i3: d, i2: c, i1: b, i0: a)) {}
46
47#if defined(__AVX512VL__)
48 __forceinline explicit vuint(__m128 a) : v(_mm_cvtps_epu32(a)) {}
49#endif
50
51#if defined(__AVX512VL__)
52 __forceinline explicit vuint(const vboolf4& a) : v(_mm_movm_epi32(a)) {}
53#else
54 __forceinline explicit vuint(const vboolf4& a) : v(_mm_castps_si128(a: (__m128)a)) {}
55#endif
56
57 ////////////////////////////////////////////////////////////////////////////////
58 /// Constants
59 ////////////////////////////////////////////////////////////////////////////////
60
61 __forceinline vuint(ZeroTy) : v(_mm_setzero_si128()) {}
62 __forceinline vuint(OneTy) : v(_mm_set1_epi32(i: 1)) {}
63 __forceinline vuint(PosInfTy) : v(_mm_set1_epi32(i: unsigned(pos_inf))) {}
64 __forceinline vuint(StepTy) : v(_mm_set_epi32(i3: 3, i2: 2, i1: 1, i0: 0)) {}
65 __forceinline vuint(TrueTy) { v = _mm_cmpeq_epi32(a: v,b: v); }
66 __forceinline vuint(UndefinedTy) : v(_mm_castps_si128(a: _mm_undefined_ps())) {}
67
68 ////////////////////////////////////////////////////////////////////////////////
69 /// Loads and Stores
70 ////////////////////////////////////////////////////////////////////////////////
71
72 static __forceinline vuint4 load (const void* a) { return _mm_load_si128(p: (__m128i*)a); }
73 static __forceinline vuint4 loadu(const void* a) { return _mm_loadu_si128(p: (__m128i*)a); }
74
75 static __forceinline void store (void* ptr, const vuint4& v) { _mm_store_si128(p: (__m128i*)ptr,b: v); }
76 static __forceinline void storeu(void* ptr, const vuint4& v) { _mm_storeu_si128(p: (__m128i*)ptr,b: v); }
77
78#if defined(__AVX512VL__)
79 static __forceinline vuint4 load (const vboolf4& mask, const void* ptr) { return _mm_mask_load_epi32 (_mm_setzero_si128(),mask,ptr); }
80 static __forceinline vuint4 loadu(const vboolf4& mask, const void* ptr) { return _mm_mask_loadu_epi32(_mm_setzero_si128(),mask,ptr); }
81
82 static __forceinline void store (const vboolf4& mask, void* ptr, const vuint4& v) { _mm_mask_store_epi32 (ptr,mask,v); }
83 static __forceinline void storeu(const vboolf4& mask, void* ptr, const vuint4& v) { _mm_mask_storeu_epi32(ptr,mask,v); }
84#elif defined(__AVX__)
85 static __forceinline vuint4 load (const vbool4& mask, const void* a) { return _mm_castps_si128(_mm_maskload_ps((float*)a,mask)); }
86 static __forceinline vuint4 loadu(const vbool4& mask, const void* a) { return _mm_castps_si128(_mm_maskload_ps((float*)a,mask)); }
87
88 static __forceinline void store (const vboolf4& mask, void* ptr, const vuint4& i) { _mm_maskstore_ps((float*)ptr,(__m128i)mask,_mm_castsi128_ps(i)); }
89 static __forceinline void storeu(const vboolf4& mask, void* ptr, const vuint4& i) { _mm_maskstore_ps((float*)ptr,(__m128i)mask,_mm_castsi128_ps(i)); }
90#else
91 static __forceinline vuint4 load (const vbool4& mask, const void* a) { return _mm_and_si128(a: _mm_load_si128 (p: (__m128i*)a),b: mask); }
92 static __forceinline vuint4 loadu(const vbool4& mask, const void* a) { return _mm_and_si128(a: _mm_loadu_si128(p: (__m128i*)a),b: mask); }
93
94 static __forceinline void store (const vboolf4& mask, void* ptr, const vuint4& i) { store (ptr,v: select(m: mask,t: i,f: load (a: ptr))); }
95 static __forceinline void storeu(const vboolf4& mask, void* ptr, const vuint4& i) { storeu(ptr,v: select(m: mask,t: i,f: loadu(a: ptr))); }
96#endif
97
98#if defined(__SSE4_1__)
99 static __forceinline vuint4 load(const unsigned char* ptr) {
100 return _mm_cvtepu8_epi32(_mm_loadl_epi64((__m128i*)ptr));
101 }
102
103 static __forceinline vuint4 loadu(const unsigned char* ptr) {
104 return _mm_cvtepu8_epi32(_mm_loadl_epi64((__m128i*)ptr));
105 }
106
107#endif
108
109 static __forceinline vuint4 load(const unsigned short* ptr) {
110#if defined (__SSE4_1__)
111 return _mm_cvtepu16_epi32(_mm_loadu_si128((__m128i*)ptr));
112#else
113 return vuint4(ptr[0],ptr[1],ptr[2],ptr[3]);
114#endif
115 }
116
117 static __forceinline vuint4 load_nt(void* ptr) {
118#if defined(__SSE4_1__)
119 return _mm_stream_load_si128((__m128i*)ptr);
120#else
121 return _mm_load_si128(p: (__m128i*)ptr);
122#endif
123 }
124
125 static __forceinline void store_nt(void* ptr, const vuint4& v) {
126#if defined(__SSE4_1__)
127 _mm_stream_ps((float*)ptr,_mm_castsi128_ps(v));
128#else
129 _mm_store_si128(p: (__m128i*)ptr,b: v);
130#endif
131 }
132
133 template<int scale = 4>
134 static __forceinline vuint4 gather(const unsigned int* ptr, const vint4& index) {
135#if defined(__AVX2__)
136 return _mm_i32gather_epi32((const int*)ptr, index, scale);
137#else
138 return vuint4(
139 *(unsigned int*)(((char*)ptr)+scale*index[0]),
140 *(unsigned int*)(((char*)ptr)+scale*index[1]),
141 *(unsigned int*)(((char*)ptr)+scale*index[2]),
142 *(unsigned int*)(((char*)ptr)+scale*index[3]));
143#endif
144 }
145
146 template<int scale = 4>
147 static __forceinline vuint4 gather(const vboolf4& mask, const unsigned int* ptr, const vint4& index) {
148 vuint4 r = zero;
149#if defined(__AVX512VL__)
150 return _mm_mmask_i32gather_epi32(r, mask, index, ptr, scale);
151#elif defined(__AVX2__)
152 return _mm_mask_i32gather_epi32(r, (const int*)ptr, index, mask, scale);
153#else
154 if (likely(mask[0])) r[0] = *(unsigned int*)(((char*)ptr)+scale*index[0]);
155 if (likely(mask[1])) r[1] = *(unsigned int*)(((char*)ptr)+scale*index[1]);
156 if (likely(mask[2])) r[2] = *(unsigned int*)(((char*)ptr)+scale*index[2]);
157 if (likely(mask[3])) r[3] = *(unsigned int*)(((char*)ptr)+scale*index[3]);
158 return r;
159#endif
160 }
161
162 ////////////////////////////////////////////////////////////////////////////////
163 /// Array Access
164 ////////////////////////////////////////////////////////////////////////////////
165
166 __forceinline const unsigned int& operator [](size_t index) const { assert(index < 4); return i[index]; }
167 __forceinline unsigned int& operator [](size_t index) { assert(index < 4); return i[index]; }
168
169 friend __forceinline vuint4 select(const vboolf4& m, const vuint4& t, const vuint4& f) {
170#if defined(__AVX512VL__)
171 return _mm_mask_blend_epi32(m, (__m128i)f, (__m128i)t);
172#elif defined(__SSE4_1__)
173 return _mm_castps_si128(_mm_blendv_ps(_mm_castsi128_ps(f), _mm_castsi128_ps(t), m));
174#else
175 return _mm_or_si128(a: _mm_and_si128(a: m, b: t), b: _mm_andnot_si128(a: m, b: f));
176#endif
177 }
178 };
179
180 ////////////////////////////////////////////////////////////////////////////////
181 /// Unary Operators
182 ////////////////////////////////////////////////////////////////////////////////
183
184#if defined(__AVX512VL__)
185 __forceinline vboolf4 asBool(const vuint4& a) { return _mm_movepi32_mask(a); }
186#else
187 __forceinline vboolf4 asBool(const vuint4& a) { return _mm_castsi128_ps(a: a); }
188#endif
189
190 __forceinline vuint4 operator +(const vuint4& a) { return a; }
191 __forceinline vuint4 operator -(const vuint4& a) { return _mm_sub_epi32(a: _mm_setzero_si128(), b: a); }
192
193 ////////////////////////////////////////////////////////////////////////////////
194 /// Binary Operators
195 ////////////////////////////////////////////////////////////////////////////////
196
197 __forceinline vuint4 operator +(const vuint4& a, const vuint4& b) { return _mm_add_epi32(a: a, b: b); }
198 __forceinline vuint4 operator +(const vuint4& a, unsigned int b) { return a + vuint4(b); }
199 __forceinline vuint4 operator +(unsigned int a, const vuint4& b) { return vuint4(a) + b; }
200
201 __forceinline vuint4 operator -(const vuint4& a, const vuint4& b) { return _mm_sub_epi32(a: a, b: b); }
202 __forceinline vuint4 operator -(const vuint4& a, unsigned int b) { return a - vuint4(b); }
203 __forceinline vuint4 operator -(unsigned int a, const vuint4& b) { return vuint4(a) - b; }
204
205//#if defined(__SSE4_1__)
206// __forceinline vuint4 operator *(const vuint4& a, const vuint4& b) { return _mm_mullo_epu32(a, b); }
207//#else
208// __forceinline vuint4 operator *(const vuint4& a, const vuint4& b) { return vuint4(a[0]*b[0],a[1]*b[1],a[2]*b[2],a[3]*b[3]); }
209//#endif
210// __forceinline vuint4 operator *(const vuint4& a, unsigned int b) { return a * vuint4(b); }
211// __forceinline vuint4 operator *(unsigned int a, const vuint4& b) { return vuint4(a) * b; }
212
213 __forceinline vuint4 operator &(const vuint4& a, const vuint4& b) { return _mm_and_si128(a: a, b: b); }
214 __forceinline vuint4 operator &(const vuint4& a, unsigned int b) { return a & vuint4(b); }
215 __forceinline vuint4 operator &(unsigned int a, const vuint4& b) { return vuint4(a) & b; }
216
217 __forceinline vuint4 operator |(const vuint4& a, const vuint4& b) { return _mm_or_si128(a: a, b: b); }
218 __forceinline vuint4 operator |(const vuint4& a, unsigned int b) { return a | vuint4(b); }
219 __forceinline vuint4 operator |(unsigned int a, const vuint4& b) { return vuint4(a) | b; }
220
221 __forceinline vuint4 operator ^(const vuint4& a, const vuint4& b) { return _mm_xor_si128(a: a, b: b); }
222 __forceinline vuint4 operator ^(const vuint4& a, unsigned int b) { return a ^ vuint4(b); }
223 __forceinline vuint4 operator ^(unsigned int a, const vuint4& b) { return vuint4(a) ^ b; }
224
225 __forceinline vuint4 operator <<(const vuint4& a, unsigned int n) { return _mm_slli_epi32(a: a, count: n); }
226 __forceinline vuint4 operator >>(const vuint4& a, unsigned int n) { return _mm_srli_epi32(a: a, count: n); }
227
228 __forceinline vuint4 sll (const vuint4& a, unsigned int b) { return _mm_slli_epi32(a: a, count: b); }
229 __forceinline vuint4 sra (const vuint4& a, unsigned int b) { return _mm_srai_epi32(a: a, count: b); }
230 __forceinline vuint4 srl (const vuint4& a, unsigned int b) { return _mm_srli_epi32(a: a, count: b); }
231
232 ////////////////////////////////////////////////////////////////////////////////
233 /// Assignment Operators
234 ////////////////////////////////////////////////////////////////////////////////
235
236 __forceinline vuint4& operator +=(vuint4& a, const vuint4& b) { return a = a + b; }
237 __forceinline vuint4& operator +=(vuint4& a, unsigned int b) { return a = a + b; }
238
239 __forceinline vuint4& operator -=(vuint4& a, const vuint4& b) { return a = a - b; }
240 __forceinline vuint4& operator -=(vuint4& a, unsigned int b) { return a = a - b; }
241
242//#if defined(__SSE4_1__)
243// __forceinline vuint4& operator *=(vuint4& a, const vuint4& b) { return a = a * b; }
244// __forceinline vuint4& operator *=(vuint4& a, unsigned int b) { return a = a * b; }
245//#endif
246
247 __forceinline vuint4& operator &=(vuint4& a, const vuint4& b) { return a = a & b; }
248 __forceinline vuint4& operator &=(vuint4& a, unsigned int b) { return a = a & b; }
249
250 __forceinline vuint4& operator |=(vuint4& a, const vuint4& b) { return a = a | b; }
251 __forceinline vuint4& operator |=(vuint4& a, unsigned int b) { return a = a | b; }
252
253 __forceinline vuint4& operator <<=(vuint4& a, unsigned int b) { return a = a << b; }
254 __forceinline vuint4& operator >>=(vuint4& a, unsigned int b) { return a = a >> b; }
255
256 ////////////////////////////////////////////////////////////////////////////////
257 /// Comparison Operators + Select
258 ////////////////////////////////////////////////////////////////////////////////
259
260#if defined(__AVX512VL__)
261 __forceinline vboolf4 operator ==(const vuint4& a, const vuint4& b) { return _mm_cmp_epu32_mask(a,b,_MM_CMPINT_EQ); }
262 __forceinline vboolf4 operator !=(const vuint4& a, const vuint4& b) { return _mm_cmp_epu32_mask(a,b,_MM_CMPINT_NE); }
263 //__forceinline vboolf4 operator < (const vuint4& a, const vuint4& b) { return _mm_cmp_epu32_mask(a,b,_MM_CMPINT_LT); }
264 //__forceinline vboolf4 operator >=(const vuint4& a, const vuint4& b) { return _mm_cmp_epu32_mask(a,b,_MM_CMPINT_GE); }
265 //__forceinline vboolf4 operator > (const vuint4& a, const vuint4& b) { return _mm_cmp_epu32_mask(a,b,_MM_CMPINT_GT); }
266 //__forceinline vboolf4 operator <=(const vuint4& a, const vuint4& b) { return _mm_cmp_epu32_mask(a,b,_MM_CMPINT_LE); }
267#else
268 __forceinline vboolf4 operator ==(const vuint4& a, const vuint4& b) { return _mm_castsi128_ps(a: _mm_cmpeq_epi32(a: a, b: b)); }
269 __forceinline vboolf4 operator !=(const vuint4& a, const vuint4& b) { return !(a == b); }
270 //__forceinline vboolf4 operator < (const vuint4& a, const vuint4& b) { return _mm_castsi128_ps(_mm_cmplt_epu32(a, b)); }
271 //__forceinline vboolf4 operator >=(const vuint4& a, const vuint4& b) { return !(a < b); }
272 //__forceinline vboolf4 operator > (const vuint4& a, const vuint4& b) { return _mm_castsi128_ps(_mm_cmpgt_epu32(a, b)); }
273 //__forceinline vboolf4 operator <=(const vuint4& a, const vuint4& b) { return !(a > b); }
274#endif
275
276 __forceinline vboolf4 operator ==(const vuint4& a, unsigned int b) { return a == vuint4(b); }
277 __forceinline vboolf4 operator ==(unsigned int a, const vuint4& b) { return vuint4(a) == b; }
278
279 __forceinline vboolf4 operator !=(const vuint4& a, unsigned int b) { return a != vuint4(b); }
280 __forceinline vboolf4 operator !=(unsigned int a, const vuint4& b) { return vuint4(a) != b; }
281
282 //__forceinline vboolf4 operator < (const vuint4& a, unsigned int b) { return a < vuint4(b); }
283 //__forceinline vboolf4 operator < (unsigned int a, const vuint4& b) { return vuint4(a) < b; }
284
285 //__forceinline vboolf4 operator >=(const vuint4& a, unsigned int b) { return a >= vuint4(b); }
286 //__forceinline vboolf4 operator >=(unsigned int a, const vuint4& b) { return vuint4(a) >= b; }
287
288 //__forceinline vboolf4 operator > (const vuint4& a, unsigned int b) { return a > vuint4(b); }
289 //__forceinline vboolf4 operator > (unsigned int a, const vuint4& b) { return vuint4(a) > b; }
290
291 //__forceinline vboolf4 operator <=(const vuint4& a, unsigned int b) { return a <= vuint4(b); }
292 //__forceinline vboolf4 operator <=(unsigned int a, const vuint4& b) { return vuint4(a) <= b; }
293
294 __forceinline vboolf4 eq(const vuint4& a, const vuint4& b) { return a == b; }
295 __forceinline vboolf4 ne(const vuint4& a, const vuint4& b) { return a != b; }
296 //__forceinline vboolf4 lt(const vuint4& a, const vuint4& b) { return a < b; }
297 //__forceinline vboolf4 ge(const vuint4& a, const vuint4& b) { return a >= b; }
298 //__forceinline vboolf4 gt(const vuint4& a, const vuint4& b) { return a > b; }
299 //__forceinline vboolf4 le(const vuint4& a, const vuint4& b) { return a <= b; }
300
301#if defined(__AVX512VL__)
302 __forceinline vboolf4 eq(const vboolf4& mask, const vuint4& a, const vuint4& b) { return _mm_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_EQ); }
303 __forceinline vboolf4 ne(const vboolf4& mask, const vuint4& a, const vuint4& b) { return _mm_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_NE); }
304 //__forceinline vboolf4 lt(const vboolf4& mask, const vuint4& a, const vuint4& b) { return _mm_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_LT); }
305 //__forceinline vboolf4 ge(const vboolf4& mask, const vuint4& a, const vuint4& b) { return _mm_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_GE); }
306 //__forceinline vboolf4 gt(const vboolf4& mask, const vuint4& a, const vuint4& b) { return _mm_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_GT); }
307 //__forceinline vboolf4 le(const vboolf4& mask, const vuint4& a, const vuint4& b) { return _mm_mask_cmp_epu32_mask(mask, a, b, _MM_CMPINT_LE); }
308#else
309 __forceinline vboolf4 eq(const vboolf4& mask, const vuint4& a, const vuint4& b) { return mask & (a == b); }
310 __forceinline vboolf4 ne(const vboolf4& mask, const vuint4& a, const vuint4& b) { return mask & (a != b); }
311 //__forceinline vboolf4 lt(const vboolf4& mask, const vuint4& a, const vuint4& b) { return mask & (a < b); }
312 //__forceinline vboolf4 ge(const vboolf4& mask, const vuint4& a, const vuint4& b) { return mask & (a >= b); }
313 //__forceinline vboolf4 gt(const vboolf4& mask, const vuint4& a, const vuint4& b) { return mask & (a > b); }
314 //__forceinline vboolf4 le(const vboolf4& mask, const vuint4& a, const vuint4& b) { return mask & (a <= b); }
315#endif
316
317 template<int mask>
318 __forceinline vuint4 select(const vuint4& t, const vuint4& f) {
319#if defined(__SSE4_1__)
320 return _mm_castps_si128(_mm_blend_ps(_mm_castsi128_ps(f), _mm_castsi128_ps(t), mask));
321#else
322 return select(m: vboolf4(mask), t, f);
323#endif
324 }
325
326/*#if defined(__SSE4_1__)
327 __forceinline vuint4 min(const vuint4& a, const vuint4& b) { return _mm_min_epu32(a, b); }
328 __forceinline vuint4 max(const vuint4& a, const vuint4& b) { return _mm_max_epu32(a, b); }
329
330#else
331 __forceinline vuint4 min(const vuint4& a, const vuint4& b) { return select(a < b,a,b); }
332 __forceinline vuint4 max(const vuint4& a, const vuint4& b) { return select(a < b,b,a); }
333#endif
334
335 __forceinline vuint4 min(const vuint4& a, unsigned int b) { return min(a,vuint4(b)); }
336 __forceinline vuint4 min(unsigned int a, const vuint4& b) { return min(vuint4(a),b); }
337 __forceinline vuint4 max(const vuint4& a, unsigned int b) { return max(a,vuint4(b)); }
338 __forceinline vuint4 max(unsigned int a, const vuint4& b) { return max(vuint4(a),b); }*/
339
340 ////////////////////////////////////////////////////////////////////////////////
341 // Movement/Shifting/Shuffling Functions
342 ////////////////////////////////////////////////////////////////////////////////
343
344 __forceinline vuint4 unpacklo(const vuint4& a, const vuint4& b) { return _mm_castps_si128(a: _mm_unpacklo_ps(a: _mm_castsi128_ps(a: a), b: _mm_castsi128_ps(a: b))); }
345 __forceinline vuint4 unpackhi(const vuint4& a, const vuint4& b) { return _mm_castps_si128(a: _mm_unpackhi_ps(a: _mm_castsi128_ps(a: a), b: _mm_castsi128_ps(a: b))); }
346
347 template<int i0, int i1, int i2, int i3>
348 __forceinline vuint4 shuffle(const vuint4& v) {
349 return _mm_shuffle_epi32(v, _MM_SHUFFLE(i3, i2, i1, i0));
350 }
351
352 template<int i0, int i1, int i2, int i3>
353 __forceinline vuint4 shuffle(const vuint4& a, const vuint4& b) {
354 return _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(a), _mm_castsi128_ps(b), _MM_SHUFFLE(i3, i2, i1, i0)));
355 }
356
357#if defined(__SSE3__)
358 template<> __forceinline vuint4 shuffle<0, 0, 2, 2>(const vuint4& v) { return _mm_castps_si128(_mm_moveldup_ps(_mm_castsi128_ps(v))); }
359 template<> __forceinline vuint4 shuffle<1, 1, 3, 3>(const vuint4& v) { return _mm_castps_si128(_mm_movehdup_ps(_mm_castsi128_ps(v))); }
360 template<> __forceinline vuint4 shuffle<0, 1, 0, 1>(const vuint4& v) { return _mm_castpd_si128(_mm_movedup_pd (_mm_castsi128_pd(v))); }
361#endif
362
363 template<int i>
364 __forceinline vuint4 shuffle(const vuint4& v) {
365 return shuffle<i,i,i,i>(v);
366 }
367
368#if defined(__SSE4_1__)
369 template<int src> __forceinline unsigned int extract(const vuint4& b) { return _mm_extract_epi32(b, src); }
370 template<int dst> __forceinline vuint4 insert(const vuint4& a, const unsigned b) { return _mm_insert_epi32(a, b, dst); }
371#else
372 template<int src> __forceinline unsigned int extract(const vuint4& b) { return b[src&3]; }
373 template<int dst> __forceinline vuint4 insert(const vuint4& a, const unsigned b) { vuint4 c = a; c[dst&3] = b; return c; }
374#endif
375
376
377 template<> __forceinline unsigned int extract<0>(const vuint4& b) { return _mm_cvtsi128_si32(a: b); }
378
379 __forceinline unsigned int toScalar(const vuint4& v) { return _mm_cvtsi128_si32(a: v); }
380
381 ////////////////////////////////////////////////////////////////////////////////
382 /// Reductions
383 ////////////////////////////////////////////////////////////////////////////////
384
385#if 0
386#if defined(__SSE4_1__)
387
388 __forceinline vuint4 vreduce_min(const vuint4& v) { vuint4 h = min(shuffle<1,0,3,2>(v),v); return min(shuffle<2,3,0,1>(h),h); }
389 __forceinline vuint4 vreduce_max(const vuint4& v) { vuint4 h = max(shuffle<1,0,3,2>(v),v); return max(shuffle<2,3,0,1>(h),h); }
390 __forceinline vuint4 vreduce_add(const vuint4& v) { vuint4 h = shuffle<1,0,3,2>(v) + v ; return shuffle<2,3,0,1>(h) + h ; }
391
392 __forceinline unsigned int reduce_min(const vuint4& v) { return toScalar(vreduce_min(v)); }
393 __forceinline unsigned int reduce_max(const vuint4& v) { return toScalar(vreduce_max(v)); }
394 __forceinline unsigned int reduce_add(const vuint4& v) { return toScalar(vreduce_add(v)); }
395
396 __forceinline size_t select_min(const vuint4& v) { return bsf(movemask(v == vreduce_min(v))); }
397 __forceinline size_t select_max(const vuint4& v) { return bsf(movemask(v == vreduce_max(v))); }
398
399 //__forceinline size_t select_min(const vboolf4& valid, const vuint4& v) { const vuint4 a = select(valid,v,vuint4(pos_inf)); return bsf(movemask(valid & (a == vreduce_min(a)))); }
400 //__forceinline size_t select_max(const vboolf4& valid, const vuint4& v) { const vuint4 a = select(valid,v,vuint4(neg_inf)); return bsf(movemask(valid & (a == vreduce_max(a)))); }
401
402#else
403
404 __forceinline unsigned int reduce_min(const vuint4& v) { return min(v[0],v[1],v[2],v[3]); }
405 __forceinline unsigned int reduce_max(const vuint4& v) { return max(v[0],v[1],v[2],v[3]); }
406 __forceinline unsigned int reduce_add(const vuint4& v) { return v[0]+v[1]+v[2]+v[3]; }
407
408#endif
409#endif
410
411 ////////////////////////////////////////////////////////////////////////////////
412 /// Output Operators
413 ////////////////////////////////////////////////////////////////////////////////
414
415 __forceinline embree_ostream operator <<(embree_ostream cout, const vuint4& a) {
416 return cout << "<" << a[0] << ", " << a[1] << ", " << a[2] << ", " << a[3] << ">";
417 }
418}
419
420#undef vboolf
421#undef vboold
422#undef vint
423#undef vuint
424#undef vllong
425#undef vfloat
426#undef vdouble
427

source code of qtquick3d/src/3rdparty/embree/common/simd/vuint4_sse2.h