• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2014 Benoit Steiner (benoit.steiner.goog@gmail.com)
5 //
6 // This Source Code Form is subject to the terms of the Mozilla
7 // Public License v. 2.0. If a copy of the MPL was not distributed
8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
9 
10 #ifndef EIGEN_COMPLEX_AVX_H
11 #define EIGEN_COMPLEX_AVX_H
12 
13 namespace Eigen {
14 
15 namespace internal {
16 
17 //---------- float ----------
18 struct Packet4cf
19 {
Packet4cfPacket4cf20   EIGEN_STRONG_INLINE Packet4cf() {}
Packet4cfPacket4cf21   EIGEN_STRONG_INLINE explicit Packet4cf(const __m256& a) : v(a) {}
22   __m256  v;
23 };
24 
25 #ifndef EIGEN_VECTORIZE_AVX512
26 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
27 {
28   typedef Packet4cf type;
29   typedef Packet2cf half;
30   enum {
31     Vectorizable = 1,
32     AlignedOnScalar = 1,
33     size = 4,
34     HasHalfPacket = 1,
35 
36     HasAdd    = 1,
37     HasSub    = 1,
38     HasMul    = 1,
39     HasDiv    = 1,
40     HasNegate = 1,
41     HasSqrt   = 1,
42     HasAbs    = 0,
43     HasAbs2   = 0,
44     HasMin    = 0,
45     HasMax    = 0,
46     HasSetLinear = 0
47   };
48 };
49 #endif
50 
51 template<> struct unpacket_traits<Packet4cf> {
52   typedef std::complex<float> type;
53   typedef Packet2cf half;
54   typedef Packet8f as_real;
55   enum {
56     size=4,
57     alignment=Aligned32,
58     vectorizable=true,
59     masked_load_available=false,
60     masked_store_available=false
61   };
62 };
63 
64 template<> EIGEN_STRONG_INLINE Packet4cf padd<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_add_ps(a.v,b.v)); }
65 template<> EIGEN_STRONG_INLINE Packet4cf psub<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_sub_ps(a.v,b.v)); }
66 template<> EIGEN_STRONG_INLINE Packet4cf pnegate(const Packet4cf& a)
67 {
68   return Packet4cf(pnegate(a.v));
69 }
70 template<> EIGEN_STRONG_INLINE Packet4cf pconj(const Packet4cf& a)
71 {
72   const __m256 mask = _mm256_castsi256_ps(_mm256_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000));
73   return Packet4cf(_mm256_xor_ps(a.v,mask));
74 }
75 
76 template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
77 {
78   __m256 tmp1 = _mm256_mul_ps(_mm256_moveldup_ps(a.v), b.v);
79   __m256 tmp2 = _mm256_mul_ps(_mm256_movehdup_ps(a.v), _mm256_permute_ps(b.v, _MM_SHUFFLE(2,3,0,1)));
80   __m256 result = _mm256_addsub_ps(tmp1, tmp2);
81   return Packet4cf(result);
82 }
83 
84 template <>
85 EIGEN_STRONG_INLINE Packet4cf pcmp_eq(const Packet4cf& a, const Packet4cf& b) {
86   __m256 eq = _mm256_cmp_ps(a.v, b.v, _CMP_EQ_OQ);
87   return Packet4cf(_mm256_and_ps(eq, _mm256_permute_ps(eq, 0xb1)));
88 }
89 
90 template<> EIGEN_STRONG_INLINE Packet4cf ptrue<Packet4cf>(const Packet4cf& a) { return Packet4cf(ptrue(Packet8f(a.v))); }
91 template<> EIGEN_STRONG_INLINE Packet4cf pand   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_and_ps(a.v,b.v)); }
92 template<> EIGEN_STRONG_INLINE Packet4cf por    <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_or_ps(a.v,b.v)); }
93 template<> EIGEN_STRONG_INLINE Packet4cf pxor   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_xor_ps(a.v,b.v)); }
94 template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(b.v,a.v)); }
95 
96 template<> EIGEN_STRONG_INLINE Packet4cf pload <Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cf(pload<Packet8f>(&numext::real_ref(*from))); }
97 template<> EIGEN_STRONG_INLINE Packet4cf ploadu<Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cf(ploadu<Packet8f>(&numext::real_ref(*from))); }
98 
99 
100 template<> EIGEN_STRONG_INLINE Packet4cf pset1<Packet4cf>(const std::complex<float>& from)
101 {
102   return Packet4cf(_mm256_castpd_ps(_mm256_broadcast_sd((const double*)(const void*)&from)));
103 }
104 
105 template<> EIGEN_STRONG_INLINE Packet4cf ploaddup<Packet4cf>(const std::complex<float>* from)
106 {
107   // FIXME The following might be optimized using _mm256_movedup_pd
108   Packet2cf a = ploaddup<Packet2cf>(from);
109   Packet2cf b = ploaddup<Packet2cf>(from+1);
110   return  Packet4cf(_mm256_insertf128_ps(_mm256_castps128_ps256(a.v), b.v, 1));
111 }
112 
113 template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float>* to, const Packet4cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); }
114 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float>* to, const Packet4cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); }
115 
116 template<> EIGEN_DEVICE_FUNC inline Packet4cf pgather<std::complex<float>, Packet4cf>(const std::complex<float>* from, Index stride)
117 {
118   return Packet4cf(_mm256_set_ps(std::imag(from[3*stride]), std::real(from[3*stride]),
119                                  std::imag(from[2*stride]), std::real(from[2*stride]),
120                                  std::imag(from[1*stride]), std::real(from[1*stride]),
121                                  std::imag(from[0*stride]), std::real(from[0*stride])));
122 }
123 
124 template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet4cf>(std::complex<float>* to, const Packet4cf& from, Index stride)
125 {
126   __m128 low = _mm256_extractf128_ps(from.v, 0);
127   to[stride*0] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(low, low, 0)),
128                                      _mm_cvtss_f32(_mm_shuffle_ps(low, low, 1)));
129   to[stride*1] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(low, low, 2)),
130                                      _mm_cvtss_f32(_mm_shuffle_ps(low, low, 3)));
131 
132   __m128 high = _mm256_extractf128_ps(from.v, 1);
133   to[stride*2] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(high, high, 0)),
134                                      _mm_cvtss_f32(_mm_shuffle_ps(high, high, 1)));
135   to[stride*3] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(high, high, 2)),
136                                      _mm_cvtss_f32(_mm_shuffle_ps(high, high, 3)));
137 
138 }
139 
140 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet4cf>(const Packet4cf& a)
141 {
142   return pfirst(Packet2cf(_mm256_castps256_ps128(a.v)));
143 }
144 
145 template<> EIGEN_STRONG_INLINE Packet4cf preverse(const Packet4cf& a) {
146   __m128 low  = _mm256_extractf128_ps(a.v, 0);
147   __m128 high = _mm256_extractf128_ps(a.v, 1);
148   __m128d lowd  = _mm_castps_pd(low);
149   __m128d highd = _mm_castps_pd(high);
150   low  = _mm_castpd_ps(_mm_shuffle_pd(lowd,lowd,0x1));
151   high = _mm_castpd_ps(_mm_shuffle_pd(highd,highd,0x1));
152   __m256 result = _mm256_setzero_ps();
153   result = _mm256_insertf128_ps(result, low, 1);
154   result = _mm256_insertf128_ps(result, high, 0);
155   return Packet4cf(result);
156 }
157 
158 template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet4cf>(const Packet4cf& a)
159 {
160   return predux(padd(Packet2cf(_mm256_extractf128_ps(a.v,0)),
161                      Packet2cf(_mm256_extractf128_ps(a.v,1))));
162 }
163 
164 template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet4cf>(const Packet4cf& a)
165 {
166   return predux_mul(pmul(Packet2cf(_mm256_extractf128_ps(a.v, 0)),
167                          Packet2cf(_mm256_extractf128_ps(a.v, 1))));
168 }
169 
170 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cf,Packet8f)
171 
172 template<> EIGEN_STRONG_INLINE Packet4cf pdiv<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
173 {
174   Packet4cf num = pmul(a, pconj(b));
175   __m256 tmp = _mm256_mul_ps(b.v, b.v);
176   __m256 tmp2    = _mm256_shuffle_ps(tmp,tmp,0xB1);
177   __m256 denom = _mm256_add_ps(tmp, tmp2);
178   return Packet4cf(_mm256_div_ps(num.v, denom));
179 }
180 
181 template<> EIGEN_STRONG_INLINE Packet4cf pcplxflip<Packet4cf>(const Packet4cf& x)
182 {
183   return Packet4cf(_mm256_shuffle_ps(x.v, x.v, _MM_SHUFFLE(2, 3, 0 ,1)));
184 }
185 
186 //---------- double ----------
187 struct Packet2cd
188 {
189   EIGEN_STRONG_INLINE Packet2cd() {}
190   EIGEN_STRONG_INLINE explicit Packet2cd(const __m256d& a) : v(a) {}
191   __m256d  v;
192 };
193 
194 #ifndef EIGEN_VECTORIZE_AVX512
195 template<> struct packet_traits<std::complex<double> >  : default_packet_traits
196 {
197   typedef Packet2cd type;
198   typedef Packet1cd half;
199   enum {
200     Vectorizable = 1,
201     AlignedOnScalar = 0,
202     size = 2,
203     HasHalfPacket = 1,
204 
205     HasAdd    = 1,
206     HasSub    = 1,
207     HasMul    = 1,
208     HasDiv    = 1,
209     HasNegate = 1,
210     HasSqrt   = 1,
211     HasAbs    = 0,
212     HasAbs2   = 0,
213     HasMin    = 0,
214     HasMax    = 0,
215     HasSetLinear = 0
216   };
217 };
218 #endif
219 
220 template<> struct unpacket_traits<Packet2cd> {
221   typedef std::complex<double> type;
222   typedef Packet1cd half;
223   typedef Packet4d as_real;
224   enum {
225     size=2,
226     alignment=Aligned32,
227     vectorizable=true,
228     masked_load_available=false,
229     masked_store_available=false
230   };
231 };
232 
233 template<> EIGEN_STRONG_INLINE Packet2cd padd<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_add_pd(a.v,b.v)); }
234 template<> EIGEN_STRONG_INLINE Packet2cd psub<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_sub_pd(a.v,b.v)); }
235 template<> EIGEN_STRONG_INLINE Packet2cd pnegate(const Packet2cd& a) { return Packet2cd(pnegate(a.v)); }
236 template<> EIGEN_STRONG_INLINE Packet2cd pconj(const Packet2cd& a)
237 {
238   const __m256d mask = _mm256_castsi256_pd(_mm256_set_epi32(0x80000000,0x0,0x0,0x0,0x80000000,0x0,0x0,0x0));
239   return Packet2cd(_mm256_xor_pd(a.v,mask));
240 }
241 
242 template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
243 {
244   __m256d tmp1 = _mm256_shuffle_pd(a.v,a.v,0x0);
245   __m256d even = _mm256_mul_pd(tmp1, b.v);
246   __m256d tmp2 = _mm256_shuffle_pd(a.v,a.v,0xF);
247   __m256d tmp3 = _mm256_shuffle_pd(b.v,b.v,0x5);
248   __m256d odd  = _mm256_mul_pd(tmp2, tmp3);
249   return Packet2cd(_mm256_addsub_pd(even, odd));
250 }
251 
252 template <>
253 EIGEN_STRONG_INLINE Packet2cd pcmp_eq(const Packet2cd& a, const Packet2cd& b) {
254   __m256d eq = _mm256_cmp_pd(a.v, b.v, _CMP_EQ_OQ);
255   return Packet2cd(pand(eq, _mm256_permute_pd(eq, 0x5)));
256 }
257 
258 template<> EIGEN_STRONG_INLINE Packet2cd ptrue<Packet2cd>(const Packet2cd& a) { return Packet2cd(ptrue(Packet4d(a.v))); }
259 template<> EIGEN_STRONG_INLINE Packet2cd pand   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_and_pd(a.v,b.v)); }
260 template<> EIGEN_STRONG_INLINE Packet2cd por    <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_or_pd(a.v,b.v)); }
261 template<> EIGEN_STRONG_INLINE Packet2cd pxor   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_xor_pd(a.v,b.v)); }
262 template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(b.v,a.v)); }
263 
264 template<> EIGEN_STRONG_INLINE Packet2cd pload <Packet2cd>(const std::complex<double>* from)
265 { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cd(pload<Packet4d>((const double*)from)); }
266 template<> EIGEN_STRONG_INLINE Packet2cd ploadu<Packet2cd>(const std::complex<double>* from)
267 { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cd(ploadu<Packet4d>((const double*)from)); }
268 
269 template<> EIGEN_STRONG_INLINE Packet2cd pset1<Packet2cd>(const std::complex<double>& from)
270 {
271   // in case casting to a __m128d* is really not safe, then we can still fallback to this version: (much slower though)
272 //   return Packet2cd(_mm256_loadu2_m128d((const double*)&from,(const double*)&from));
273     return Packet2cd(_mm256_broadcast_pd((const __m128d*)(const void*)&from));
274 }
275 
276 template<> EIGEN_STRONG_INLINE Packet2cd ploaddup<Packet2cd>(const std::complex<double>* from) { return pset1<Packet2cd>(*from); }
277 
278 template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet2cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
279 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet2cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
280 
281 template<> EIGEN_DEVICE_FUNC inline Packet2cd pgather<std::complex<double>, Packet2cd>(const std::complex<double>* from, Index stride)
282 {
283   return Packet2cd(_mm256_set_pd(std::imag(from[1*stride]), std::real(from[1*stride]),
284 				 std::imag(from[0*stride]), std::real(from[0*stride])));
285 }
286 
287 template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet2cd>(std::complex<double>* to, const Packet2cd& from, Index stride)
288 {
289   __m128d low = _mm256_extractf128_pd(from.v, 0);
290   to[stride*0] = std::complex<double>(_mm_cvtsd_f64(low), _mm_cvtsd_f64(_mm_shuffle_pd(low, low, 1)));
291   __m128d high = _mm256_extractf128_pd(from.v, 1);
292   to[stride*1] = std::complex<double>(_mm_cvtsd_f64(high), _mm_cvtsd_f64(_mm_shuffle_pd(high, high, 1)));
293 }
294 
295 template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet2cd>(const Packet2cd& a)
296 {
297   __m128d low = _mm256_extractf128_pd(a.v, 0);
298   EIGEN_ALIGN16 double res[2];
299   _mm_store_pd(res, low);
300   return std::complex<double>(res[0],res[1]);
301 }
302 
303 template<> EIGEN_STRONG_INLINE Packet2cd preverse(const Packet2cd& a) {
304   __m256d result = _mm256_permute2f128_pd(a.v, a.v, 1);
305   return Packet2cd(result);
306 }
307 
308 template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet2cd>(const Packet2cd& a)
309 {
310   return predux(padd(Packet1cd(_mm256_extractf128_pd(a.v,0)),
311                      Packet1cd(_mm256_extractf128_pd(a.v,1))));
312 }
313 
314 template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet2cd>(const Packet2cd& a)
315 {
316   return predux(pmul(Packet1cd(_mm256_extractf128_pd(a.v,0)),
317                      Packet1cd(_mm256_extractf128_pd(a.v,1))));
318 }
319 
320 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cd,Packet4d)
321 
322 template<> EIGEN_STRONG_INLINE Packet2cd pdiv<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
323 {
324   Packet2cd num = pmul(a, pconj(b));
325   __m256d tmp = _mm256_mul_pd(b.v, b.v);
326   __m256d denom = _mm256_hadd_pd(tmp, tmp);
327   return Packet2cd(_mm256_div_pd(num.v, denom));
328 }
329 
330 template<> EIGEN_STRONG_INLINE Packet2cd pcplxflip<Packet2cd>(const Packet2cd& x)
331 {
332   return Packet2cd(_mm256_shuffle_pd(x.v, x.v, 0x5));
333 }
334 
335 EIGEN_DEVICE_FUNC inline void
336 ptranspose(PacketBlock<Packet4cf,4>& kernel) {
337   __m256d P0 = _mm256_castps_pd(kernel.packet[0].v);
338   __m256d P1 = _mm256_castps_pd(kernel.packet[1].v);
339   __m256d P2 = _mm256_castps_pd(kernel.packet[2].v);
340   __m256d P3 = _mm256_castps_pd(kernel.packet[3].v);
341 
342   __m256d T0 = _mm256_shuffle_pd(P0, P1, 15);
343   __m256d T1 = _mm256_shuffle_pd(P0, P1, 0);
344   __m256d T2 = _mm256_shuffle_pd(P2, P3, 15);
345   __m256d T3 = _mm256_shuffle_pd(P2, P3, 0);
346 
347   kernel.packet[1].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T0, T2, 32));
348   kernel.packet[3].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T0, T2, 49));
349   kernel.packet[0].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T1, T3, 32));
350   kernel.packet[2].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T1, T3, 49));
351 }
352 
353 EIGEN_DEVICE_FUNC inline void
354 ptranspose(PacketBlock<Packet2cd,2>& kernel) {
355   __m256d tmp = _mm256_permute2f128_pd(kernel.packet[0].v, kernel.packet[1].v, 0+(2<<4));
356   kernel.packet[1].v = _mm256_permute2f128_pd(kernel.packet[0].v, kernel.packet[1].v, 1+(3<<4));
357  kernel.packet[0].v = tmp;
358 }
359 
360 template<> EIGEN_STRONG_INLINE Packet2cd psqrt<Packet2cd>(const Packet2cd& a) {
361   return psqrt_complex<Packet2cd>(a);
362 }
363 
364 template<> EIGEN_STRONG_INLINE Packet4cf psqrt<Packet4cf>(const Packet4cf& a) {
365   return psqrt_complex<Packet4cf>(a);
366 }
367 
368 } // end namespace internal
369 
370 } // end namespace Eigen
371 
372 #endif // EIGEN_COMPLEX_AVX_H
373