2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2014,2015,2019,2020, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 #ifndef GMX_SIMD_IMPL_X86_MIC_SIMD4_FLOAT_H
37 #define GMX_SIMD_IMPL_X86_MIC_SIMD4_FLOAT_H
43 #include <immintrin.h>
45 #include "gromacs/utility/basedefinitions.h"
47 #include "impl_x86_mic_simd_float.h"
57 Simd4Float(float f) : simdInternal_(_mm512_set1_ps(f)) {}
59 // Internal utility constructor to simplify return statements
60 Simd4Float(__m512 simd) : simdInternal_(simd) {}
70 // Internal utility constructor to simplify return statements
71 Simd4FBool(__mmask16 simd) : simdInternal_(simd) {}
73 __mmask16 simdInternal_;
76 static inline Simd4Float gmx_simdcall load4(const float* m)
78 assert(size_t(m) % 16 == 0);
79 return { _mm512_mask_extload_ps(
80 _mm512_undefined_ps(), _mm512_int2mask(0xF), m, _MM_UPCONV_PS_NONE, _MM_BROADCAST_4X16, _MM_HINT_NONE) };
83 static inline void gmx_simdcall store4(float* m, Simd4Float a)
85 assert(size_t(m) % 16 == 0);
86 _mm512_mask_packstorelo_ps(m, _mm512_int2mask(0xF), a.simdInternal_);
89 static inline Simd4Float gmx_simdcall load4U(const float* m)
91 return { _mm512_mask_loadunpackhi_ps(
92 _mm512_mask_loadunpacklo_ps(_mm512_undefined_ps(), _mm512_int2mask(0xF), m),
97 static inline void gmx_simdcall store4U(float* m, Simd4Float a)
99 _mm512_mask_packstorelo_ps(m, _mm512_int2mask(0xF), a.simdInternal_);
100 _mm512_mask_packstorehi_ps(m + 16, _mm512_int2mask(0xF), a.simdInternal_);
103 static inline Simd4Float gmx_simdcall simd4SetZeroF()
105 return { _mm512_setzero_ps() };
108 static inline Simd4Float gmx_simdcall operator&(Simd4Float a, Simd4Float b)
110 return { _mm512_castsi512_ps(_mm512_mask_and_epi32(_mm512_undefined_epi32(),
111 _mm512_int2mask(0xF),
112 _mm512_castps_si512(a.simdInternal_),
113 _mm512_castps_si512(b.simdInternal_))) };
116 static inline Simd4Float gmx_simdcall andNot(Simd4Float a, Simd4Float b)
118 return { _mm512_castsi512_ps(_mm512_mask_andnot_epi32(_mm512_undefined_epi32(),
119 _mm512_int2mask(0xF),
120 _mm512_castps_si512(a.simdInternal_),
121 _mm512_castps_si512(b.simdInternal_))) };
124 static inline Simd4Float gmx_simdcall operator|(Simd4Float a, Simd4Float b)
126 return { _mm512_castsi512_ps(_mm512_mask_or_epi32(_mm512_undefined_epi32(),
127 _mm512_int2mask(0xF),
128 _mm512_castps_si512(a.simdInternal_),
129 _mm512_castps_si512(b.simdInternal_))) };
132 static inline Simd4Float gmx_simdcall operator^(Simd4Float a, Simd4Float b)
134 return { _mm512_castsi512_ps(_mm512_mask_xor_epi32(_mm512_undefined_epi32(),
135 _mm512_int2mask(0xF),
136 _mm512_castps_si512(a.simdInternal_),
137 _mm512_castps_si512(b.simdInternal_))) };
140 static inline Simd4Float gmx_simdcall operator+(Simd4Float a, Simd4Float b)
142 return { _mm512_mask_add_ps(
143 _mm512_undefined_ps(), _mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_) };
146 static inline Simd4Float gmx_simdcall operator-(Simd4Float a, Simd4Float b)
148 return { _mm512_mask_sub_ps(
149 _mm512_undefined_ps(), _mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_) };
152 static inline Simd4Float gmx_simdcall operator-(Simd4Float x)
154 return { _mm512_mask_addn_ps(
155 _mm512_undefined_ps(), _mm512_int2mask(0xF), x.simdInternal_, _mm512_setzero_ps()) };
158 static inline Simd4Float gmx_simdcall operator*(Simd4Float a, Simd4Float b)
160 return { _mm512_mask_mul_ps(
161 _mm512_undefined_ps(), _mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_) };
164 static inline Simd4Float gmx_simdcall fma(Simd4Float a, Simd4Float b, Simd4Float c)
166 return { _mm512_mask_fmadd_ps(a.simdInternal_, _mm512_int2mask(0xF), b.simdInternal_, c.simdInternal_) };
169 static inline Simd4Float gmx_simdcall fms(Simd4Float a, Simd4Float b, Simd4Float c)
171 return { _mm512_mask_fmsub_ps(a.simdInternal_, _mm512_int2mask(0xF), b.simdInternal_, c.simdInternal_) };
174 static inline Simd4Float gmx_simdcall fnma(Simd4Float a, Simd4Float b, Simd4Float c)
176 return { _mm512_mask_fnmadd_ps(a.simdInternal_, _mm512_int2mask(0xF), b.simdInternal_, c.simdInternal_) };
179 static inline Simd4Float gmx_simdcall fnms(Simd4Float a, Simd4Float b, Simd4Float c)
181 return { _mm512_mask_fnmsub_ps(a.simdInternal_, _mm512_int2mask(0xF), b.simdInternal_, c.simdInternal_) };
184 static inline Simd4Float gmx_simdcall rsqrt(Simd4Float x)
186 return { _mm512_mask_rsqrt23_ps(_mm512_undefined_ps(), _mm512_int2mask(0xF), x.simdInternal_) };
189 static inline Simd4Float gmx_simdcall abs(Simd4Float x)
191 return { _mm512_castsi512_ps(_mm512_mask_andnot_epi32(_mm512_undefined_epi32(),
192 _mm512_int2mask(0xF),
193 _mm512_castps_si512(_mm512_set1_ps(GMX_FLOAT_NEGZERO)),
194 _mm512_castps_si512(x.simdInternal_))) };
197 static inline Simd4Float gmx_simdcall max(Simd4Float a, Simd4Float b)
199 return { _mm512_mask_gmax_ps(
200 _mm512_undefined_ps(), _mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_) };
203 static inline Simd4Float gmx_simdcall min(Simd4Float a, Simd4Float b)
205 return { _mm512_mask_gmin_ps(
206 _mm512_undefined_ps(), _mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_) };
209 static inline Simd4Float gmx_simdcall round(Simd4Float x)
211 return { _mm512_mask_round_ps(
212 _mm512_undefined_ps(), _mm512_int2mask(0xF), x.simdInternal_, _MM_FROUND_TO_NEAREST_INT, _MM_EXPADJ_NONE) };
215 static inline Simd4Float gmx_simdcall trunc(Simd4Float x)
217 return { _mm512_mask_round_ps(
218 _mm512_undefined_ps(), _mm512_int2mask(0xF), x.simdInternal_, _MM_FROUND_TO_ZERO, _MM_EXPADJ_NONE) };
221 static inline float gmx_simdcall dotProduct(Simd4Float a, Simd4Float b)
223 __m512 x = _mm512_mask_mul_ps(
224 _mm512_setzero_ps(), _mm512_int2mask(0x7), a.simdInternal_, b.simdInternal_);
225 x = _mm512_add_ps(x, _mm512_swizzle_ps(x, _MM_SWIZ_REG_BADC));
226 x = _mm512_add_ps(x, _mm512_swizzle_ps(x, _MM_SWIZ_REG_CDAB));
228 _mm512_mask_packstorelo_ps(&f, _mm512_mask2int(0x1), x);
232 static inline void gmx_simdcall transpose(Simd4Float* v0, Simd4Float* v1, Simd4Float* v2, Simd4Float* v3)
234 v0->simdInternal_ = _mm512_mask_permute4f128_ps(
235 v0->simdInternal_, _mm512_int2mask(0x00F0), v1->simdInternal_, _MM_PERM_AAAA);
236 v2->simdInternal_ = _mm512_mask_permute4f128_ps(
237 v2->simdInternal_, _mm512_int2mask(0x00F0), v3->simdInternal_, _MM_PERM_AAAA);
238 v0->simdInternal_ = _mm512_mask_permute4f128_ps(
239 v0->simdInternal_, _mm512_int2mask(0xFF00), v2->simdInternal_, _MM_PERM_BABA);
240 v0->simdInternal_ = _mm512_castsi512_ps(_mm512_permutevar_epi32(
241 _mm512_set_epi32(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0),
242 _mm512_castps_si512(v0->simdInternal_)));
243 v1->simdInternal_ = _mm512_mask_permute4f128_ps(
244 _mm512_setzero_ps(), _mm512_int2mask(0x000F), v0->simdInternal_, _MM_PERM_BBBB);
245 v2->simdInternal_ = _mm512_mask_permute4f128_ps(
246 _mm512_setzero_ps(), _mm512_int2mask(0x000F), v0->simdInternal_, _MM_PERM_CCCC);
247 v3->simdInternal_ = _mm512_mask_permute4f128_ps(
248 _mm512_setzero_ps(), _mm512_int2mask(0x000F), v0->simdInternal_, _MM_PERM_DDDD);
251 // Picky, picky, picky:
252 // icc-16 complains about "Illegal value of immediate argument to intrinsic"
254 // 1) Ordered-quiet for ==
255 // 2) Unordered-quiet for !=
256 // 3) Ordered-signaling for < and <=
258 static inline Simd4FBool gmx_simdcall operator==(Simd4Float a, Simd4Float b)
260 return { _mm512_mask_cmp_ps_mask(_mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_, _CMP_EQ_OQ) };
263 static inline Simd4FBool gmx_simdcall operator!=(Simd4Float a, Simd4Float b)
265 return { _mm512_mask_cmp_ps_mask(_mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_, _CMP_NEQ_UQ) };
268 static inline Simd4FBool gmx_simdcall operator<(Simd4Float a, Simd4Float b)
270 return { _mm512_mask_cmp_ps_mask(_mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_, _CMP_LT_OS) };
273 static inline Simd4FBool gmx_simdcall operator<=(Simd4Float a, Simd4Float b)
275 return { _mm512_mask_cmp_ps_mask(_mm512_int2mask(0xF), a.simdInternal_, b.simdInternal_, _CMP_LE_OS) };
278 static inline Simd4FBool gmx_simdcall operator&&(Simd4FBool a, Simd4FBool b)
280 return { _mm512_kand(a.simdInternal_, b.simdInternal_) };
283 static inline Simd4FBool gmx_simdcall operator||(Simd4FBool a, Simd4FBool b)
285 return { _mm512_kor(a.simdInternal_, b.simdInternal_) };
288 static inline bool gmx_simdcall anyTrue(Simd4FBool a)
290 return (_mm512_mask2int(a.simdInternal_) & 0xF) != 0;
293 static inline Simd4Float gmx_simdcall selectByMask(Simd4Float a, Simd4FBool m)
295 return { _mm512_mask_mov_ps(_mm512_setzero_ps(), m.simdInternal_, a.simdInternal_) };
298 static inline Simd4Float gmx_simdcall selectByNotMask(Simd4Float a, Simd4FBool m)
300 return { _mm512_mask_mov_ps(_mm512_setzero_ps(), _mm512_knot(m.simdInternal_), a.simdInternal_) };
303 static inline Simd4Float gmx_simdcall blend(Simd4Float a, Simd4Float b, Simd4FBool sel)
305 return { _mm512_mask_blend_ps(sel.simdInternal_, a.simdInternal_, b.simdInternal_) };
308 static inline float gmx_simdcall reduce(Simd4Float a)
310 __m512 x = a.simdInternal_;
311 x = _mm512_add_ps(x, _mm512_swizzle_ps(x, _MM_SWIZ_REG_BADC));
312 x = _mm512_add_ps(x, _mm512_swizzle_ps(x, _MM_SWIZ_REG_CDAB));
314 _mm512_mask_packstorelo_ps(&f, _mm512_mask2int(0x1), x);
320 #endif // GMX_SIMD_IMPL_X86_MIC_SIMD4_FLOAT_H