xref: /aosp_15_r20/external/eigen/Eigen/src/Core/arch/SSE/Complex.h (revision bf2c37156dfe67e5dfebd6d394bad8b2ab5804d4)
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2010 Gael Guennebaud <[email protected]>
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_SSE_H
11 #define EIGEN_COMPLEX_SSE_H
12 
13 namespace Eigen {
14 
15 namespace internal {
16 
17 //---------- float ----------
18 struct Packet2cf
19 {
Packet2cfPacket2cf20   EIGEN_STRONG_INLINE Packet2cf() {}
Packet2cfPacket2cf21   EIGEN_STRONG_INLINE explicit Packet2cf(const __m128& a) : v(a) {}
22   Packet4f v;
23 };
24 
25 // Use the packet_traits defined in AVX/PacketMath.h instead if we're going
26 // to leverage AVX instructions.
27 #ifndef EIGEN_VECTORIZE_AVX
28 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
29 {
30   typedef Packet2cf type;
31   typedef Packet2cf half;
32   enum {
33     Vectorizable = 1,
34     AlignedOnScalar = 1,
35     size = 2,
36     HasHalfPacket = 0,
37 
38     HasAdd    = 1,
39     HasSub    = 1,
40     HasMul    = 1,
41     HasDiv    = 1,
42     HasNegate = 1,
43     HasSqrt   = 1,
44     HasAbs    = 0,
45     HasAbs2   = 0,
46     HasMin    = 0,
47     HasMax    = 0,
48     HasSetLinear = 0,
49     HasBlend  = 1
50   };
51 };
52 #endif
53 
54 template<> struct unpacket_traits<Packet2cf> {
55   typedef std::complex<float> type;
56   typedef Packet2cf half;
57   typedef Packet4f as_real;
58   enum {
59     size=2,
60     alignment=Aligned16,
61     vectorizable=true,
62     masked_load_available=false,
63     masked_store_available=false
64   };
65 };
66 
67 template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_add_ps(a.v,b.v)); }
68 template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_sub_ps(a.v,b.v)); }
69 
70 template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a)
71 {
72   const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x80000000,0x80000000,0x80000000,0x80000000));
73   return Packet2cf(_mm_xor_ps(a.v,mask));
74 }
75 template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
76 {
77   const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000));
78   return Packet2cf(_mm_xor_ps(a.v,mask));
79 }
80 
81 template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
82 {
83   #ifdef EIGEN_VECTORIZE_SSE3
84   return Packet2cf(_mm_addsub_ps(_mm_mul_ps(_mm_moveldup_ps(a.v), b.v),
85                                  _mm_mul_ps(_mm_movehdup_ps(a.v),
86                                             vec4f_swizzle1(b.v, 1, 0, 3, 2))));
87 //   return Packet2cf(_mm_addsub_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 0, 0, 2, 2), b.v),
88 //                                  _mm_mul_ps(vec4f_swizzle1(a.v, 1, 1, 3, 3),
89 //                                             vec4f_swizzle1(b.v, 1, 0, 3, 2))));
90   #else
91   const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x80000000,0x00000000,0x80000000,0x00000000));
92   return Packet2cf(_mm_add_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 0, 0, 2, 2), b.v),
93                               _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 1, 1, 3, 3),
94                                                     vec4f_swizzle1(b.v, 1, 0, 3, 2)), mask)));
95   #endif
96 }
97 
98 template<> EIGEN_STRONG_INLINE Packet2cf ptrue  <Packet2cf>(const Packet2cf& a) { return Packet2cf(ptrue(Packet4f(a.v))); }
99 template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_and_ps(a.v,b.v)); }
100 template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_or_ps(a.v,b.v)); }
101 template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_xor_ps(a.v,b.v)); }
102 template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_andnot_ps(b.v,a.v)); }
103 
104 template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>(&numext::real_ref(*from))); }
105 template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>(&numext::real_ref(*from))); }
106 
107 template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
108 {
109   Packet2cf res;
110 #ifdef EIGEN_VECTORIZE_SSE3
111   res.v = _mm_castpd_ps(_mm_loaddup_pd(reinterpret_cast<double const*>(&from)));
112 #else
113   res.v = _mm_castpd_ps(_mm_load_sd(reinterpret_cast<double const*>(&from)));
114   res.v = _mm_movelh_ps(res.v, res.v);
115 #endif
116   return res;
117 }
118 
119 template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); }
120 
121 template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), Packet4f(from.v)); }
122 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), Packet4f(from.v)); }
123 
124 
125 template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
126 {
127   return Packet2cf(_mm_set_ps(std::imag(from[1*stride]), std::real(from[1*stride]),
128                               std::imag(from[0*stride]), std::real(from[0*stride])));
129 }
130 
131 template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
132 {
133   to[stride*0] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 0)),
134                                      _mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 1)));
135   to[stride*1] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 2)),
136                                      _mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 3)));
137 }
138 
139 template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
140 
141 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
142 {
143   #if EIGEN_GNUC_AT_MOST(4,3)
144   // Workaround gcc 4.2 ICE - this is not performance wise ideal, but who cares...
145   // This workaround also fix invalid code generation with gcc 4.3
146   EIGEN_ALIGN16 std::complex<float> res[2];
147   _mm_store_ps((float*)res, a.v);
148   return res[0];
149   #else
150   std::complex<float> res;
151   _mm_storel_pi((__m64*)&res, a.v);
152   return res;
153   #endif
154 }
155 
156 template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a) { return Packet2cf(_mm_castpd_ps(preverse(Packet2d(_mm_castps_pd(a.v))))); }
157 
158 template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
159 {
160   return pfirst(Packet2cf(_mm_add_ps(a.v, _mm_movehl_ps(a.v,a.v))));
161 }
162 
163 template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
164 {
165   return pfirst(pmul(a, Packet2cf(_mm_movehl_ps(a.v,a.v))));
166 }
167 
168 EIGEN_STRONG_INLINE Packet2cf pcplxflip/* <Packet2cf> */(const Packet2cf& x)
169 {
170   return Packet2cf(vec4f_swizzle1(x.v, 1, 0, 3, 2));
171 }
172 
173 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
174 
175 template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
176 {
177   // TODO optimize it for SSE3 and 4
178   Packet2cf res = pmul(a, pconj(b));
179   __m128 s = _mm_mul_ps(b.v,b.v);
180   return Packet2cf(_mm_div_ps(res.v,_mm_add_ps(s,vec4f_swizzle1(s, 1, 0, 3, 2))));
181 }
182 
183 
184 
185 //---------- double ----------
186 struct Packet1cd
187 {
188   EIGEN_STRONG_INLINE Packet1cd() {}
189   EIGEN_STRONG_INLINE explicit Packet1cd(const __m128d& a) : v(a) {}
190   Packet2d v;
191 };
192 
193 // Use the packet_traits defined in AVX/PacketMath.h instead if we're going
194 // to leverage AVX instructions.
195 #ifndef EIGEN_VECTORIZE_AVX
196 template<> struct packet_traits<std::complex<double> >  : default_packet_traits
197 {
198   typedef Packet1cd type;
199   typedef Packet1cd half;
200   enum {
201     Vectorizable = 1,
202     AlignedOnScalar = 0,
203     size = 1,
204     HasHalfPacket = 0,
205 
206     HasAdd    = 1,
207     HasSub    = 1,
208     HasMul    = 1,
209     HasDiv    = 1,
210     HasNegate = 1,
211     HasSqrt   = 1,
212     HasAbs    = 0,
213     HasAbs2   = 0,
214     HasMin    = 0,
215     HasMax    = 0,
216     HasSetLinear = 0
217   };
218 };
219 #endif
220 
221 template<> struct unpacket_traits<Packet1cd> {
222   typedef std::complex<double> type;
223   typedef Packet1cd half;
224   typedef Packet2d as_real;
225   enum {
226     size=1,
227     alignment=Aligned16,
228     vectorizable=true,
229     masked_load_available=false,
230     masked_store_available=false
231   };
232 };
233 
234 template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_add_pd(a.v,b.v)); }
235 template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_sub_pd(a.v,b.v)); }
236 template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
237 template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a)
238 {
239   const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0));
240   return Packet1cd(_mm_xor_pd(a.v,mask));
241 }
242 
243 template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
244 {
245   #ifdef EIGEN_VECTORIZE_SSE3
246   return Packet1cd(_mm_addsub_pd(_mm_mul_pd(_mm_movedup_pd(a.v), b.v),
247                                  _mm_mul_pd(vec2d_swizzle1(a.v, 1, 1),
248                                             vec2d_swizzle1(b.v, 1, 0))));
249   #else
250   const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
251   return Packet1cd(_mm_add_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 0, 0), b.v),
252                               _mm_xor_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 1, 1),
253                                                     vec2d_swizzle1(b.v, 1, 0)), mask)));
254   #endif
255 }
256 
257 template<> EIGEN_STRONG_INLINE Packet1cd ptrue  <Packet1cd>(const Packet1cd& a) { return Packet1cd(ptrue(Packet2d(a.v))); }
258 template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_and_pd(a.v,b.v)); }
259 template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_or_pd(a.v,b.v)); }
260 template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_xor_pd(a.v,b.v)); }
261 template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_andnot_pd(b.v,a.v)); }
262 
263 // FIXME force unaligned load, this is a temporary fix
264 template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from)
265 { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); }
266 template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from)
267 { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); }
268 template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>&  from)
269 { /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
270 
271 template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); }
272 
273 // FIXME force unaligned store, this is a temporary fix
274 template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, Packet2d(from.v)); }
275 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, Packet2d(from.v)); }
276 
277 template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
278 
279 template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
280 {
281   EIGEN_ALIGN16 double res[2];
282   _mm_store_pd(res, a.v);
283   return std::complex<double>(res[0],res[1]);
284 }
285 
286 template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
287 
288 template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a)
289 {
290   return pfirst(a);
291 }
292 
293 template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a)
294 {
295   return pfirst(a);
296 }
297 
298 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
299 
300 template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
301 {
302   // TODO optimize it for SSE3 and 4
303   Packet1cd res = pmul(a,pconj(b));
304   __m128d s = _mm_mul_pd(b.v,b.v);
305   return Packet1cd(_mm_div_pd(res.v, _mm_add_pd(s,_mm_shuffle_pd(s, s, 0x1))));
306 }
307 
308 EIGEN_STRONG_INLINE Packet1cd pcplxflip/* <Packet1cd> */(const Packet1cd& x)
309 {
310   return Packet1cd(preverse(Packet2d(x.v)));
311 }
312 
313 EIGEN_DEVICE_FUNC inline void
314 ptranspose(PacketBlock<Packet2cf,2>& kernel) {
315   __m128d w1 = _mm_castps_pd(kernel.packet[0].v);
316   __m128d w2 = _mm_castps_pd(kernel.packet[1].v);
317 
318   __m128 tmp = _mm_castpd_ps(_mm_unpackhi_pd(w1, w2));
319   kernel.packet[0].v = _mm_castpd_ps(_mm_unpacklo_pd(w1, w2));
320   kernel.packet[1].v = tmp;
321 }
322 
323 template<> EIGEN_STRONG_INLINE Packet2cf pcmp_eq(const Packet2cf& a, const Packet2cf& b)
324 {
325   __m128 eq = _mm_cmpeq_ps(a.v, b.v);
326   return Packet2cf(pand<Packet4f>(eq, vec4f_swizzle1(eq, 1, 0, 3, 2)));
327 }
328 
329 template<> EIGEN_STRONG_INLINE Packet1cd pcmp_eq(const Packet1cd& a, const Packet1cd& b)
330 {
331   __m128d eq = _mm_cmpeq_pd(a.v, b.v);
332   return Packet1cd(pand<Packet2d>(eq, vec2d_swizzle1(eq, 1, 0)));
333 }
334 
335 template<>  EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
336   __m128d result = pblend<Packet2d>(ifPacket, _mm_castps_pd(thenPacket.v), _mm_castps_pd(elsePacket.v));
337   return Packet2cf(_mm_castpd_ps(result));
338 }
339 
340 template<> EIGEN_STRONG_INLINE Packet1cd psqrt<Packet1cd>(const Packet1cd& a) {
341   return psqrt_complex<Packet1cd>(a);
342 }
343 
344 template<> EIGEN_STRONG_INLINE Packet2cf psqrt<Packet2cf>(const Packet2cf& a) {
345   return psqrt_complex<Packet2cf>(a);
346 }
347 
348 } // end namespace internal
349 } // end namespace Eigen
350 
351 #endif // EIGEN_COMPLEX_SSE_H
352