• 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) 2018 Gael Guennebaud <gael.guennebaud@inria.fr>
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_AVX512_H
11 #define EIGEN_COMPLEX_AVX512_H
12 
13 namespace Eigen {
14 
15 namespace internal {
16 
17 //---------- float ----------
18 struct Packet8cf
19 {
Packet8cfPacket8cf20   EIGEN_STRONG_INLINE Packet8cf() {}
Packet8cfPacket8cf21   EIGEN_STRONG_INLINE explicit Packet8cf(const __m512& a) : v(a) {}
22   __m512  v;
23 };
24 
25 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
26 {
27   typedef Packet8cf type;
28   typedef Packet4cf half;
29   enum {
30     Vectorizable = 1,
31     AlignedOnScalar = 1,
32     size = 8,
33     HasHalfPacket = 1,
34 
35     HasAdd    = 1,
36     HasSub    = 1,
37     HasMul    = 1,
38     HasDiv    = 1,
39     HasNegate = 1,
40     HasSqrt   = 1,
41     HasAbs    = 0,
42     HasAbs2   = 0,
43     HasMin    = 0,
44     HasMax    = 0,
45     HasSetLinear = 0
46   };
47 };
48 
49 template<> struct unpacket_traits<Packet8cf> {
50   typedef std::complex<float> type;
51   typedef Packet4cf half;
52   typedef Packet16f as_real;
53   enum {
54     size = 8,
55     alignment=unpacket_traits<Packet16f>::alignment,
56     vectorizable=true,
57     masked_load_available=false,
58     masked_store_available=false
59   };
60 };
61 
62 template<> EIGEN_STRONG_INLINE Packet8cf ptrue<Packet8cf>(const Packet8cf& a) { return Packet8cf(ptrue(Packet16f(a.v))); }
63 template<> EIGEN_STRONG_INLINE Packet8cf padd<Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_add_ps(a.v,b.v)); }
64 template<> EIGEN_STRONG_INLINE Packet8cf psub<Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_sub_ps(a.v,b.v)); }
65 template<> EIGEN_STRONG_INLINE Packet8cf pnegate(const Packet8cf& a)
66 {
67   return Packet8cf(pnegate(a.v));
68 }
69 template<> EIGEN_STRONG_INLINE Packet8cf pconj(const Packet8cf& a)
70 {
71   const __m512 mask = _mm512_castsi512_ps(_mm512_setr_epi32(
72     0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,
73     0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000));
74   return Packet8cf(pxor(a.v,mask));
75 }
76 
77 template<> EIGEN_STRONG_INLINE Packet8cf pmul<Packet8cf>(const Packet8cf& a, const Packet8cf& b)
78 {
79   __m512 tmp2 = _mm512_mul_ps(_mm512_movehdup_ps(a.v), _mm512_permute_ps(b.v, _MM_SHUFFLE(2,3,0,1)));
80   return Packet8cf(_mm512_fmaddsub_ps(_mm512_moveldup_ps(a.v), b.v, tmp2));
81 }
82 
83 template<> EIGEN_STRONG_INLINE Packet8cf pand   <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(pand(a.v,b.v)); }
84 template<> EIGEN_STRONG_INLINE Packet8cf por    <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(por(a.v,b.v)); }
85 template<> EIGEN_STRONG_INLINE Packet8cf pxor   <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(pxor(a.v,b.v)); }
86 template<> EIGEN_STRONG_INLINE Packet8cf pandnot<Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(pandnot(a.v,b.v)); }
87 
88 template <>
89 EIGEN_STRONG_INLINE Packet8cf pcmp_eq(const Packet8cf& a, const Packet8cf& b) {
90   __m512 eq = pcmp_eq<Packet16f>(a.v, b.v);
91   return Packet8cf(pand(eq, _mm512_permute_ps(eq, 0xB1)));
92 }
93 
94 template<> EIGEN_STRONG_INLINE Packet8cf pload <Packet8cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet8cf(pload<Packet16f>(&numext::real_ref(*from))); }
95 template<> EIGEN_STRONG_INLINE Packet8cf ploadu<Packet8cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet8cf(ploadu<Packet16f>(&numext::real_ref(*from))); }
96 
97 
98 template<> EIGEN_STRONG_INLINE Packet8cf pset1<Packet8cf>(const std::complex<float>& from)
99 {
100   return Packet8cf(_mm512_castpd_ps(pload1<Packet8d>((const double*)(const void*)&from)));
101 }
102 
103 template<> EIGEN_STRONG_INLINE Packet8cf ploaddup<Packet8cf>(const std::complex<float>* from)
104 {
105   return Packet8cf( _mm512_castpd_ps( ploaddup<Packet8d>((const double*)(const void*)from )) );
106 }
107 template<> EIGEN_STRONG_INLINE Packet8cf ploadquad<Packet8cf>(const std::complex<float>* from)
108 {
109   return Packet8cf( _mm512_castpd_ps( ploadquad<Packet8d>((const double*)(const void*)from )) );
110 }
111 
112 template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float>* to, const Packet8cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); }
113 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float>* to, const Packet8cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); }
114 
115 template<> EIGEN_DEVICE_FUNC inline Packet8cf pgather<std::complex<float>, Packet8cf>(const std::complex<float>* from, Index stride)
116 {
117   return Packet8cf(_mm512_castpd_ps(pgather<double,Packet8d>((const double*)(const void*)from, stride)));
118 }
119 
120 template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet8cf>(std::complex<float>* to, const Packet8cf& from, Index stride)
121 {
122   pscatter((double*)(void*)to, _mm512_castps_pd(from.v), stride);
123 }
124 
125 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet8cf>(const Packet8cf& a)
126 {
127   return pfirst(Packet2cf(_mm512_castps512_ps128(a.v)));
128 }
129 
130 template<> EIGEN_STRONG_INLINE Packet8cf preverse(const Packet8cf& a) {
131   return Packet8cf(_mm512_castsi512_ps(
132             _mm512_permutexvar_epi64( _mm512_set_epi32(0, 0, 0, 1, 0, 2, 0, 3, 0, 4, 0, 5, 0, 6, 0, 7),
133                                       _mm512_castps_si512(a.v))));
134 }
135 
136 template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet8cf>(const Packet8cf& a)
137 {
138   return predux(padd(Packet4cf(extract256<0>(a.v)),
139                      Packet4cf(extract256<1>(a.v))));
140 }
141 
142 template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet8cf>(const Packet8cf& a)
143 {
144   return predux_mul(pmul(Packet4cf(extract256<0>(a.v)),
145                          Packet4cf(extract256<1>(a.v))));
146 }
147 
148 template <>
149 EIGEN_STRONG_INLINE Packet4cf predux_half_dowto4<Packet8cf>(const Packet8cf& a) {
150   __m256 lane0 = extract256<0>(a.v);
151   __m256 lane1 = extract256<1>(a.v);
152   __m256 res = _mm256_add_ps(lane0, lane1);
153   return Packet4cf(res);
154 }
155 
156 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet8cf,Packet16f)
157 
158 template<> EIGEN_STRONG_INLINE Packet8cf pdiv<Packet8cf>(const Packet8cf& a, const Packet8cf& b)
159 {
160   Packet8cf num = pmul(a, pconj(b));
161   __m512 tmp = _mm512_mul_ps(b.v, b.v);
162   __m512 tmp2    = _mm512_shuffle_ps(tmp,tmp,0xB1);
163   __m512 denom = _mm512_add_ps(tmp, tmp2);
164   return Packet8cf(_mm512_div_ps(num.v, denom));
165 }
166 
167 template<> EIGEN_STRONG_INLINE Packet8cf pcplxflip<Packet8cf>(const Packet8cf& x)
168 {
169   return Packet8cf(_mm512_shuffle_ps(x.v, x.v, _MM_SHUFFLE(2, 3, 0 ,1)));
170 }
171 
172 //---------- double ----------
173 struct Packet4cd
174 {
175   EIGEN_STRONG_INLINE Packet4cd() {}
176   EIGEN_STRONG_INLINE explicit Packet4cd(const __m512d& a) : v(a) {}
177   __m512d  v;
178 };
179 
180 template<> struct packet_traits<std::complex<double> >  : default_packet_traits
181 {
182   typedef Packet4cd type;
183   typedef Packet2cd half;
184   enum {
185     Vectorizable = 1,
186     AlignedOnScalar = 0,
187     size = 4,
188     HasHalfPacket = 1,
189 
190     HasAdd    = 1,
191     HasSub    = 1,
192     HasMul    = 1,
193     HasDiv    = 1,
194     HasNegate = 1,
195     HasSqrt   = 1,
196     HasAbs    = 0,
197     HasAbs2   = 0,
198     HasMin    = 0,
199     HasMax    = 0,
200     HasSetLinear = 0
201   };
202 };
203 
204 template<> struct unpacket_traits<Packet4cd> {
205   typedef std::complex<double> type;
206   typedef Packet2cd half;
207   typedef Packet8d as_real;
208   enum {
209     size = 4,
210     alignment = unpacket_traits<Packet8d>::alignment,
211     vectorizable=true,
212     masked_load_available=false,
213     masked_store_available=false
214   };
215 };
216 
217 template<> EIGEN_STRONG_INLINE Packet4cd padd<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_add_pd(a.v,b.v)); }
218 template<> EIGEN_STRONG_INLINE Packet4cd psub<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_sub_pd(a.v,b.v)); }
219 template<> EIGEN_STRONG_INLINE Packet4cd pnegate(const Packet4cd& a) { return Packet4cd(pnegate(a.v)); }
220 template<> EIGEN_STRONG_INLINE Packet4cd pconj(const Packet4cd& a)
221 {
222   const __m512d mask = _mm512_castsi512_pd(
223           _mm512_set_epi32(0x80000000,0x0,0x0,0x0,0x80000000,0x0,0x0,0x0,
224                            0x80000000,0x0,0x0,0x0,0x80000000,0x0,0x0,0x0));
225   return Packet4cd(pxor(a.v,mask));
226 }
227 
228 template<> EIGEN_STRONG_INLINE Packet4cd pmul<Packet4cd>(const Packet4cd& a, const Packet4cd& b)
229 {
230   __m512d tmp1 = _mm512_shuffle_pd(a.v,a.v,0x0);
231   __m512d tmp2 = _mm512_shuffle_pd(a.v,a.v,0xFF);
232   __m512d tmp3 = _mm512_shuffle_pd(b.v,b.v,0x55);
233   __m512d odd  = _mm512_mul_pd(tmp2, tmp3);
234   return Packet4cd(_mm512_fmaddsub_pd(tmp1, b.v, odd));
235 }
236 
237 template<> EIGEN_STRONG_INLINE Packet4cd ptrue<Packet4cd>(const Packet4cd& a) { return Packet4cd(ptrue(Packet8d(a.v))); }
238 template<> EIGEN_STRONG_INLINE Packet4cd pand   <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(pand(a.v,b.v)); }
239 template<> EIGEN_STRONG_INLINE Packet4cd por    <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(por(a.v,b.v)); }
240 template<> EIGEN_STRONG_INLINE Packet4cd pxor   <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(pxor(a.v,b.v)); }
241 template<> EIGEN_STRONG_INLINE Packet4cd pandnot<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(pandnot(a.v,b.v)); }
242 
243 template <>
244 EIGEN_STRONG_INLINE Packet4cd pcmp_eq(const Packet4cd& a, const Packet4cd& b) {
245   __m512d eq = pcmp_eq<Packet8d>(a.v, b.v);
246   return Packet4cd(pand(eq, _mm512_permute_pd(eq, 0x55)));
247 }
248 
249 template<> EIGEN_STRONG_INLINE Packet4cd pload <Packet4cd>(const std::complex<double>* from)
250 { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cd(pload<Packet8d>((const double*)from)); }
251 template<> EIGEN_STRONG_INLINE Packet4cd ploadu<Packet4cd>(const std::complex<double>* from)
252 { EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cd(ploadu<Packet8d>((const double*)from)); }
253 
254 template<> EIGEN_STRONG_INLINE Packet4cd pset1<Packet4cd>(const std::complex<double>& from)
255 {
256   #ifdef EIGEN_VECTORIZE_AVX512DQ
257   return Packet4cd(_mm512_broadcast_f64x2(pset1<Packet1cd>(from).v));
258   #else
259   return Packet4cd(_mm512_castps_pd(_mm512_broadcast_f32x4( _mm_castpd_ps(pset1<Packet1cd>(from).v))));
260   #endif
261 }
262 
263 template<> EIGEN_STRONG_INLINE Packet4cd ploaddup<Packet4cd>(const std::complex<double>* from) {
264   return Packet4cd(_mm512_insertf64x4(
265           _mm512_castpd256_pd512(ploaddup<Packet2cd>(from).v), ploaddup<Packet2cd>(from+1).v, 1));
266 }
267 
268 template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet4cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
269 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet4cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
270 
271 template<> EIGEN_DEVICE_FUNC inline Packet4cd pgather<std::complex<double>, Packet4cd>(const std::complex<double>* from, Index stride)
272 {
273   return Packet4cd(_mm512_insertf64x4(_mm512_castpd256_pd512(
274             _mm256_insertf128_pd(_mm256_castpd128_pd256(ploadu<Packet1cd>(from+0*stride).v), ploadu<Packet1cd>(from+1*stride).v,1)),
275             _mm256_insertf128_pd(_mm256_castpd128_pd256(ploadu<Packet1cd>(from+2*stride).v), ploadu<Packet1cd>(from+3*stride).v,1), 1));
276 }
277 
278 template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet4cd>(std::complex<double>* to, const Packet4cd& from, Index stride)
279 {
280   __m512i fromi = _mm512_castpd_si512(from.v);
281   double* tod = (double*)(void*)to;
282   _mm_storeu_pd(tod+0*stride, _mm_castsi128_pd(_mm512_extracti32x4_epi32(fromi,0)) );
283   _mm_storeu_pd(tod+2*stride, _mm_castsi128_pd(_mm512_extracti32x4_epi32(fromi,1)) );
284   _mm_storeu_pd(tod+4*stride, _mm_castsi128_pd(_mm512_extracti32x4_epi32(fromi,2)) );
285   _mm_storeu_pd(tod+6*stride, _mm_castsi128_pd(_mm512_extracti32x4_epi32(fromi,3)) );
286 }
287 
288 template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet4cd>(const Packet4cd& a)
289 {
290   __m128d low = extract128<0>(a.v);
291   EIGEN_ALIGN16 double res[2];
292   _mm_store_pd(res, low);
293   return std::complex<double>(res[0],res[1]);
294 }
295 
296 template<> EIGEN_STRONG_INLINE Packet4cd preverse(const Packet4cd& a) {
297   return Packet4cd(_mm512_shuffle_f64x2(a.v, a.v, (shuffle_mask<3,2,1,0>::mask)));
298 }
299 
300 template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet4cd>(const Packet4cd& a)
301 {
302   return predux(padd(Packet2cd(_mm512_extractf64x4_pd(a.v,0)),
303                      Packet2cd(_mm512_extractf64x4_pd(a.v,1))));
304 }
305 
306 template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet4cd>(const Packet4cd& a)
307 {
308   return predux_mul(pmul(Packet2cd(_mm512_extractf64x4_pd(a.v,0)),
309                          Packet2cd(_mm512_extractf64x4_pd(a.v,1))));
310 }
311 
312 template<> struct conj_helper<Packet4cd, Packet4cd, false,true>
313 {
314   EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const
315   { return padd(pmul(x,y),c); }
316 
317   EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const
318   {
319     return internal::pmul(a, pconj(b));
320   }
321 };
322 
323 template<> struct conj_helper<Packet4cd, Packet4cd, true,false>
324 {
325   EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const
326   { return padd(pmul(x,y),c); }
327 
328   EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const
329   {
330     return internal::pmul(pconj(a), b);
331   }
332 };
333 
334 template<> struct conj_helper<Packet4cd, Packet4cd, true,true>
335 {
336   EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const
337   { return padd(pmul(x,y),c); }
338 
339   EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const
340   {
341     return pconj(internal::pmul(a, b));
342   }
343 };
344 
345 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cd,Packet8d)
346 
347 template<> EIGEN_STRONG_INLINE Packet4cd pdiv<Packet4cd>(const Packet4cd& a, const Packet4cd& b)
348 {
349   Packet4cd num = pmul(a, pconj(b));
350   __m512d tmp = _mm512_mul_pd(b.v, b.v);
351   __m512d denom =  padd(_mm512_permute_pd(tmp,0x55), tmp);
352   return Packet4cd(_mm512_div_pd(num.v, denom));
353 }
354 
355 template<> EIGEN_STRONG_INLINE Packet4cd pcplxflip<Packet4cd>(const Packet4cd& x)
356 {
357   return Packet4cd(_mm512_permute_pd(x.v,0x55));
358 }
359 
360 EIGEN_DEVICE_FUNC inline void
361 ptranspose(PacketBlock<Packet8cf,4>& kernel) {
362   PacketBlock<Packet8d,4> pb;
363 
364   pb.packet[0] = _mm512_castps_pd(kernel.packet[0].v);
365   pb.packet[1] = _mm512_castps_pd(kernel.packet[1].v);
366   pb.packet[2] = _mm512_castps_pd(kernel.packet[2].v);
367   pb.packet[3] = _mm512_castps_pd(kernel.packet[3].v);
368   ptranspose(pb);
369   kernel.packet[0].v = _mm512_castpd_ps(pb.packet[0]);
370   kernel.packet[1].v = _mm512_castpd_ps(pb.packet[1]);
371   kernel.packet[2].v = _mm512_castpd_ps(pb.packet[2]);
372   kernel.packet[3].v = _mm512_castpd_ps(pb.packet[3]);
373 }
374 
375 EIGEN_DEVICE_FUNC inline void
376 ptranspose(PacketBlock<Packet8cf,8>& kernel) {
377   PacketBlock<Packet8d,8> pb;
378 
379   pb.packet[0] = _mm512_castps_pd(kernel.packet[0].v);
380   pb.packet[1] = _mm512_castps_pd(kernel.packet[1].v);
381   pb.packet[2] = _mm512_castps_pd(kernel.packet[2].v);
382   pb.packet[3] = _mm512_castps_pd(kernel.packet[3].v);
383   pb.packet[4] = _mm512_castps_pd(kernel.packet[4].v);
384   pb.packet[5] = _mm512_castps_pd(kernel.packet[5].v);
385   pb.packet[6] = _mm512_castps_pd(kernel.packet[6].v);
386   pb.packet[7] = _mm512_castps_pd(kernel.packet[7].v);
387   ptranspose(pb);
388   kernel.packet[0].v = _mm512_castpd_ps(pb.packet[0]);
389   kernel.packet[1].v = _mm512_castpd_ps(pb.packet[1]);
390   kernel.packet[2].v = _mm512_castpd_ps(pb.packet[2]);
391   kernel.packet[3].v = _mm512_castpd_ps(pb.packet[3]);
392   kernel.packet[4].v = _mm512_castpd_ps(pb.packet[4]);
393   kernel.packet[5].v = _mm512_castpd_ps(pb.packet[5]);
394   kernel.packet[6].v = _mm512_castpd_ps(pb.packet[6]);
395   kernel.packet[7].v = _mm512_castpd_ps(pb.packet[7]);
396 }
397 
398 EIGEN_DEVICE_FUNC inline void
399 ptranspose(PacketBlock<Packet4cd,4>& kernel) {
400   __m512d T0 = _mm512_shuffle_f64x2(kernel.packet[0].v, kernel.packet[1].v, (shuffle_mask<0,1,0,1>::mask)); // [a0 a1 b0 b1]
401   __m512d T1 = _mm512_shuffle_f64x2(kernel.packet[0].v, kernel.packet[1].v, (shuffle_mask<2,3,2,3>::mask)); // [a2 a3 b2 b3]
402   __m512d T2 = _mm512_shuffle_f64x2(kernel.packet[2].v, kernel.packet[3].v, (shuffle_mask<0,1,0,1>::mask)); // [c0 c1 d0 d1]
403   __m512d T3 = _mm512_shuffle_f64x2(kernel.packet[2].v, kernel.packet[3].v, (shuffle_mask<2,3,2,3>::mask)); // [c2 c3 d2 d3]
404 
405   kernel.packet[3] = Packet4cd(_mm512_shuffle_f64x2(T1, T3, (shuffle_mask<1,3,1,3>::mask))); // [a3 b3 c3 d3]
406   kernel.packet[2] = Packet4cd(_mm512_shuffle_f64x2(T1, T3, (shuffle_mask<0,2,0,2>::mask))); // [a2 b2 c2 d2]
407   kernel.packet[1] = Packet4cd(_mm512_shuffle_f64x2(T0, T2, (shuffle_mask<1,3,1,3>::mask))); // [a1 b1 c1 d1]
408   kernel.packet[0] = Packet4cd(_mm512_shuffle_f64x2(T0, T2, (shuffle_mask<0,2,0,2>::mask))); // [a0 b0 c0 d0]
409 }
410 
411 template<> EIGEN_STRONG_INLINE Packet4cd psqrt<Packet4cd>(const Packet4cd& a) {
412   return psqrt_complex<Packet4cd>(a);
413 }
414 
415 template<> EIGEN_STRONG_INLINE Packet8cf psqrt<Packet8cf>(const Packet8cf& a) {
416   return psqrt_complex<Packet8cf>(a);
417 }
418 
419 } // end namespace internal
420 } // end namespace Eigen
421 
422 #endif // EIGEN_COMPLEX_AVX512_H
423