2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2014,2015,2016,2017,2018,2019, 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_AVX_512_SIMD_FLOAT_H
37 #define GMX_SIMD_IMPL_X86_AVX_512_SIMD_FLOAT_H
44 #include <immintrin.h>
46 #include "gromacs/math/utilities.h"
47 #include "gromacs/utility/real.h"
49 #include "impl_x86_avx_512_general.h"
59 SimdFloat(float f) : simdInternal_(_mm512_set1_ps(f)) {}
61 // Internal utility constructor to simplify return statements
62 SimdFloat(__m512 simd) : simdInternal_(simd) {}
72 SimdFInt32(std::int32_t i) : simdInternal_(_mm512_set1_epi32(i)) {}
74 // Internal utility constructor to simplify return statements
75 SimdFInt32(__m512i simd) : simdInternal_(simd) {}
77 __m512i simdInternal_;
85 // Internal utility constructor to simplify return statements
86 SimdFBool(__mmask16 simd) : simdInternal_(simd) {}
88 __mmask16 simdInternal_;
96 // Internal utility constructor to simplify return statements
97 SimdFIBool(__mmask16 simd) : simdInternal_(simd) {}
99 __mmask16 simdInternal_;
102 static inline SimdFloat gmx_simdcall simdLoad(const float* m, SimdFloatTag = {})
104 assert(std::size_t(m) % 64 == 0);
105 return { _mm512_load_ps(m) };
108 static inline void gmx_simdcall store(float* m, SimdFloat a)
110 assert(std::size_t(m) % 64 == 0);
111 _mm512_store_ps(m, a.simdInternal_);
114 static inline SimdFloat gmx_simdcall simdLoadU(const float* m, SimdFloatTag = {})
116 return { _mm512_loadu_ps(m) };
119 static inline void gmx_simdcall storeU(float* m, SimdFloat a)
121 _mm512_storeu_ps(m, a.simdInternal_);
124 static inline SimdFloat gmx_simdcall setZeroF()
126 return { _mm512_setzero_ps() };
129 static inline SimdFInt32 gmx_simdcall simdLoad(const std::int32_t* m, SimdFInt32Tag)
131 assert(std::size_t(m) % 64 == 0);
132 return { _mm512_load_si512(m) };
135 static inline void gmx_simdcall store(std::int32_t* m, SimdFInt32 a)
137 assert(std::size_t(m) % 64 == 0);
138 _mm512_store_si512(m, a.simdInternal_);
141 static inline SimdFInt32 gmx_simdcall simdLoadU(const std::int32_t* m, SimdFInt32Tag)
143 return { _mm512_loadu_si512(m) };
146 static inline void gmx_simdcall storeU(std::int32_t* m, SimdFInt32 a)
148 _mm512_storeu_si512(m, a.simdInternal_);
151 static inline SimdFInt32 gmx_simdcall setZeroFI()
153 return { _mm512_setzero_si512() };
157 static inline SimdFloat gmx_simdcall operator&(SimdFloat a, SimdFloat b)
159 return { _mm512_castsi512_ps(_mm512_and_epi32(_mm512_castps_si512(a.simdInternal_),
160 _mm512_castps_si512(b.simdInternal_))) };
163 static inline SimdFloat gmx_simdcall andNot(SimdFloat a, SimdFloat b)
165 return { _mm512_castsi512_ps(_mm512_andnot_epi32(_mm512_castps_si512(a.simdInternal_),
166 _mm512_castps_si512(b.simdInternal_))) };
169 static inline SimdFloat gmx_simdcall operator|(SimdFloat a, SimdFloat b)
171 return { _mm512_castsi512_ps(_mm512_or_epi32(_mm512_castps_si512(a.simdInternal_),
172 _mm512_castps_si512(b.simdInternal_))) };
175 static inline SimdFloat gmx_simdcall operator^(SimdFloat a, SimdFloat b)
177 return { _mm512_castsi512_ps(_mm512_xor_epi32(_mm512_castps_si512(a.simdInternal_),
178 _mm512_castps_si512(b.simdInternal_))) };
181 static inline SimdFloat gmx_simdcall operator+(SimdFloat a, SimdFloat b)
183 return { _mm512_add_ps(a.simdInternal_, b.simdInternal_) };
186 static inline SimdFloat gmx_simdcall operator-(SimdFloat a, SimdFloat b)
188 return { _mm512_sub_ps(a.simdInternal_, b.simdInternal_) };
191 static inline SimdFloat gmx_simdcall operator-(SimdFloat x)
193 return { _mm512_castsi512_ps(_mm512_xor_epi32(_mm512_castps_si512(x.simdInternal_),
194 _mm512_castps_si512(_mm512_set1_ps(GMX_FLOAT_NEGZERO)))) };
197 static inline SimdFloat gmx_simdcall operator*(SimdFloat a, SimdFloat b)
199 return { _mm512_mul_ps(a.simdInternal_, b.simdInternal_) };
202 static inline SimdFloat gmx_simdcall fma(SimdFloat a, SimdFloat b, SimdFloat c)
204 return { _mm512_fmadd_ps(a.simdInternal_, b.simdInternal_, c.simdInternal_) };
207 static inline SimdFloat gmx_simdcall fms(SimdFloat a, SimdFloat b, SimdFloat c)
209 return { _mm512_fmsub_ps(a.simdInternal_, b.simdInternal_, c.simdInternal_) };
212 static inline SimdFloat gmx_simdcall fnma(SimdFloat a, SimdFloat b, SimdFloat c)
214 return { _mm512_fnmadd_ps(a.simdInternal_, b.simdInternal_, c.simdInternal_) };
217 static inline SimdFloat gmx_simdcall fnms(SimdFloat a, SimdFloat b, SimdFloat c)
219 return { _mm512_fnmsub_ps(a.simdInternal_, b.simdInternal_, c.simdInternal_) };
222 // Override for AVX-512-KNL
223 #if GMX_SIMD_X86_AVX_512
224 static inline SimdFloat gmx_simdcall rsqrt(SimdFloat x)
226 return { _mm512_rsqrt14_ps(x.simdInternal_) };
229 static inline SimdFloat gmx_simdcall rcp(SimdFloat x)
231 return { _mm512_rcp14_ps(x.simdInternal_) };
235 static inline SimdFloat gmx_simdcall maskAdd(SimdFloat a, SimdFloat b, SimdFBool m)
237 return { _mm512_mask_add_ps(a.simdInternal_, m.simdInternal_, a.simdInternal_, b.simdInternal_) };
240 static inline SimdFloat gmx_simdcall maskzMul(SimdFloat a, SimdFloat b, SimdFBool m)
242 return { _mm512_maskz_mul_ps(m.simdInternal_, a.simdInternal_, b.simdInternal_) };
245 static inline SimdFloat gmx_simdcall maskzFma(SimdFloat a, SimdFloat b, SimdFloat c, SimdFBool m)
247 return { _mm512_maskz_fmadd_ps(m.simdInternal_, a.simdInternal_, b.simdInternal_, c.simdInternal_) };
250 // Override for AVX-512-KNL
251 #if GMX_SIMD_X86_AVX_512
252 static inline SimdFloat gmx_simdcall maskzRsqrt(SimdFloat x, SimdFBool m)
254 return { _mm512_maskz_rsqrt14_ps(m.simdInternal_, x.simdInternal_) };
257 static inline SimdFloat gmx_simdcall maskzRcp(SimdFloat x, SimdFBool m)
259 return { _mm512_maskz_rcp14_ps(m.simdInternal_, x.simdInternal_) };
263 static inline SimdFloat gmx_simdcall abs(SimdFloat x)
265 return { _mm512_castsi512_ps(_mm512_andnot_epi32(_mm512_castps_si512(_mm512_set1_ps(GMX_FLOAT_NEGZERO)),
266 _mm512_castps_si512(x.simdInternal_))) };
269 static inline SimdFloat gmx_simdcall max(SimdFloat a, SimdFloat b)
271 return { _mm512_max_ps(a.simdInternal_, b.simdInternal_) };
274 static inline SimdFloat gmx_simdcall min(SimdFloat a, SimdFloat b)
276 return { _mm512_min_ps(a.simdInternal_, b.simdInternal_) };
279 static inline SimdFloat gmx_simdcall round(SimdFloat x)
281 return { _mm512_roundscale_ps(x.simdInternal_, 0) };
284 static inline SimdFloat gmx_simdcall trunc(SimdFloat x)
286 #if defined(__INTEL_COMPILER) || defined(__ECC)
287 return { _mm512_trunc_ps(x.simdInternal_) };
289 return { _mm512_cvtepi32_ps(_mm512_cvttps_epi32(x.simdInternal_)) };
293 static inline SimdFloat gmx_simdcall frexp(SimdFloat value, SimdFInt32* exponent)
295 __m512 rExponent = _mm512_getexp_ps(value.simdInternal_);
296 __m512i iExponent = _mm512_cvtps_epi32(rExponent);
298 exponent->simdInternal_ = _mm512_add_epi32(iExponent, _mm512_set1_epi32(1));
300 return { _mm512_getmant_ps(value.simdInternal_, _MM_MANT_NORM_p5_1, _MM_MANT_SIGN_src) };
303 template<MathOptimization opt = MathOptimization::Safe>
304 static inline SimdFloat gmx_simdcall ldexp(SimdFloat value, SimdFInt32 exponent)
306 const __m512i exponentBias = _mm512_set1_epi32(127);
307 __m512i iExponent = _mm512_add_epi32(exponent.simdInternal_, exponentBias);
309 if (opt == MathOptimization::Safe)
311 // Make sure biased argument is not negative
312 iExponent = _mm512_max_epi32(iExponent, _mm512_setzero_epi32());
315 iExponent = _mm512_slli_epi32(iExponent, 23);
317 return { _mm512_mul_ps(value.simdInternal_, _mm512_castsi512_ps(iExponent)) };
320 static inline float gmx_simdcall reduce(SimdFloat a)
322 __m512 x = a.simdInternal_;
323 x = _mm512_add_ps(x, _mm512_shuffle_f32x4(x, x, 0xEE));
324 x = _mm512_add_ps(x, _mm512_shuffle_f32x4(x, x, 0x11));
325 x = _mm512_add_ps(x, _mm512_permute_ps(x, 0xEE));
326 x = _mm512_add_ps(x, _mm512_permute_ps(x, 0x11));
327 return *reinterpret_cast<float*>(&x);
330 static inline SimdFBool gmx_simdcall operator==(SimdFloat a, SimdFloat b)
332 return { _mm512_cmp_ps_mask(a.simdInternal_, b.simdInternal_, _CMP_EQ_OQ) };
335 static inline SimdFBool gmx_simdcall operator!=(SimdFloat a, SimdFloat b)
337 return { _mm512_cmp_ps_mask(a.simdInternal_, b.simdInternal_, _CMP_NEQ_OQ) };
340 static inline SimdFBool gmx_simdcall operator<(SimdFloat a, SimdFloat b)
342 return { _mm512_cmp_ps_mask(a.simdInternal_, b.simdInternal_, _CMP_LT_OQ) };
345 static inline SimdFBool gmx_simdcall operator<=(SimdFloat a, SimdFloat b)
347 return { _mm512_cmp_ps_mask(a.simdInternal_, b.simdInternal_, _CMP_LE_OQ) };
350 static inline SimdFBool gmx_simdcall testBits(SimdFloat a)
352 return { _mm512_test_epi32_mask(_mm512_castps_si512(a.simdInternal_),
353 _mm512_castps_si512(a.simdInternal_)) };
356 static inline SimdFBool gmx_simdcall operator&&(SimdFBool a, SimdFBool b)
358 return { _mm512_kand(a.simdInternal_, b.simdInternal_) };
361 static inline SimdFBool gmx_simdcall operator||(SimdFBool a, SimdFBool b)
363 return { _mm512_kor(a.simdInternal_, b.simdInternal_) };
366 static inline bool gmx_simdcall anyTrue(SimdFBool a)
368 return (avx512Mask2Int(a.simdInternal_) != 0);
371 static inline SimdFloat gmx_simdcall selectByMask(SimdFloat a, SimdFBool m)
373 return { _mm512_mask_mov_ps(_mm512_setzero_ps(), m.simdInternal_, a.simdInternal_) };
376 static inline SimdFloat gmx_simdcall selectByNotMask(SimdFloat a, SimdFBool m)
378 return { _mm512_mask_mov_ps(a.simdInternal_, m.simdInternal_, _mm512_setzero_ps()) };
381 static inline SimdFloat gmx_simdcall blend(SimdFloat a, SimdFloat b, SimdFBool sel)
383 return { _mm512_mask_blend_ps(sel.simdInternal_, a.simdInternal_, b.simdInternal_) };
386 static inline SimdFloat gmx_simdcall copysign(SimdFloat a, SimdFloat b)
388 return { _mm512_castsi512_ps(_mm512_ternarylogic_epi32(_mm512_castps_si512(a.simdInternal_),
389 _mm512_castps_si512(b.simdInternal_),
390 _mm512_set1_epi32(INT32_MIN), 0xD8)) };
393 static inline SimdFInt32 gmx_simdcall operator&(SimdFInt32 a, SimdFInt32 b)
395 return { _mm512_and_epi32(a.simdInternal_, b.simdInternal_) };
398 static inline SimdFInt32 gmx_simdcall andNot(SimdFInt32 a, SimdFInt32 b)
400 return { _mm512_andnot_epi32(a.simdInternal_, b.simdInternal_) };
403 static inline SimdFInt32 gmx_simdcall operator|(SimdFInt32 a, SimdFInt32 b)
405 return { _mm512_or_epi32(a.simdInternal_, b.simdInternal_) };
408 static inline SimdFInt32 gmx_simdcall operator^(SimdFInt32 a, SimdFInt32 b)
410 return { _mm512_xor_epi32(a.simdInternal_, b.simdInternal_) };
413 static inline SimdFInt32 gmx_simdcall operator+(SimdFInt32 a, SimdFInt32 b)
415 return { _mm512_add_epi32(a.simdInternal_, b.simdInternal_) };
418 static inline SimdFInt32 gmx_simdcall operator-(SimdFInt32 a, SimdFInt32 b)
420 return { _mm512_sub_epi32(a.simdInternal_, b.simdInternal_) };
423 static inline SimdFInt32 gmx_simdcall operator*(SimdFInt32 a, SimdFInt32 b)
425 return { _mm512_mullo_epi32(a.simdInternal_, b.simdInternal_) };
428 static inline SimdFIBool gmx_simdcall operator==(SimdFInt32 a, SimdFInt32 b)
430 return { _mm512_cmp_epi32_mask(a.simdInternal_, b.simdInternal_, _MM_CMPINT_EQ) };
433 static inline SimdFIBool gmx_simdcall testBits(SimdFInt32 a)
435 return { _mm512_test_epi32_mask(a.simdInternal_, a.simdInternal_) };
438 static inline SimdFIBool gmx_simdcall operator<(SimdFInt32 a, SimdFInt32 b)
440 return { _mm512_cmp_epi32_mask(a.simdInternal_, b.simdInternal_, _MM_CMPINT_LT) };
443 static inline SimdFIBool gmx_simdcall operator&&(SimdFIBool a, SimdFIBool b)
445 return { _mm512_kand(a.simdInternal_, b.simdInternal_) };
448 static inline SimdFIBool gmx_simdcall operator||(SimdFIBool a, SimdFIBool b)
450 return { _mm512_kor(a.simdInternal_, b.simdInternal_) };
453 static inline bool gmx_simdcall anyTrue(SimdFIBool a)
455 return (avx512Mask2Int(a.simdInternal_) != 0);
458 static inline SimdFInt32 gmx_simdcall selectByMask(SimdFInt32 a, SimdFIBool m)
460 return { _mm512_mask_mov_epi32(_mm512_setzero_epi32(), m.simdInternal_, a.simdInternal_) };
463 static inline SimdFInt32 gmx_simdcall selectByNotMask(SimdFInt32 a, SimdFIBool m)
465 return { _mm512_mask_mov_epi32(a.simdInternal_, m.simdInternal_, _mm512_setzero_epi32()) };
468 static inline SimdFInt32 gmx_simdcall blend(SimdFInt32 a, SimdFInt32 b, SimdFIBool sel)
470 return { _mm512_mask_blend_epi32(sel.simdInternal_, a.simdInternal_, b.simdInternal_) };
473 static inline SimdFInt32 gmx_simdcall cvtR2I(SimdFloat a)
475 return { _mm512_cvtps_epi32(a.simdInternal_) };
478 static inline SimdFInt32 gmx_simdcall cvttR2I(SimdFloat a)
480 return { _mm512_cvttps_epi32(a.simdInternal_) };
483 static inline SimdFloat gmx_simdcall cvtI2R(SimdFInt32 a)
485 return { _mm512_cvtepi32_ps(a.simdInternal_) };
488 static inline SimdFIBool gmx_simdcall cvtB2IB(SimdFBool a)
490 return { a.simdInternal_ };
493 static inline SimdFBool gmx_simdcall cvtIB2B(SimdFIBool a)
495 return { a.simdInternal_ };
500 #endif // GMX_SIMD_IMPL_X86_AVX_512_SIMD_FLOAT_H