2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017, 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 * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_avx_128_fma_single.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_single
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: None
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_single
58 (t_nblist
* gmx_restrict nlist
,
59 rvec
* gmx_restrict xx
,
60 rvec
* gmx_restrict ff
,
61 struct t_forcerec
* gmx_restrict fr
,
62 t_mdatoms
* gmx_restrict mdatoms
,
63 nb_kernel_data_t gmx_unused
* gmx_restrict kernel_data
,
64 t_nrnb
* gmx_restrict nrnb
)
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
71 int i_shift_offset
,i_coord_offset
,outeriter
,inneriter
;
72 int j_index_start
,j_index_end
,jidx
,nri
,inr
,ggid
,iidx
;
73 int jnrA
,jnrB
,jnrC
,jnrD
;
74 int jnrlistA
,jnrlistB
,jnrlistC
,jnrlistD
;
75 int j_coord_offsetA
,j_coord_offsetB
,j_coord_offsetC
,j_coord_offsetD
;
76 int *iinr
,*jindex
,*jjnr
,*shiftidx
,*gid
;
78 real
*shiftvec
,*fshift
,*x
,*f
;
79 real
*fjptrA
,*fjptrB
,*fjptrC
,*fjptrD
;
81 __m128 fscal
,rcutoff
,rcutoff2
,jidxall
;
83 __m128 ix1
,iy1
,iz1
,fix1
,fiy1
,fiz1
,iq1
,isai1
;
85 __m128 ix2
,iy2
,iz2
,fix2
,fiy2
,fiz2
,iq2
,isai2
;
87 __m128 ix3
,iy3
,iz3
,fix3
,fiy3
,fiz3
,iq3
,isai3
;
88 int vdwjidx0A
,vdwjidx0B
,vdwjidx0C
,vdwjidx0D
;
89 __m128 jx0
,jy0
,jz0
,fjx0
,fjy0
,fjz0
,jq0
,isaj0
;
90 __m128 dx10
,dy10
,dz10
,rsq10
,rinv10
,rinvsq10
,r10
,qq10
,c6_10
,c12_10
;
91 __m128 dx20
,dy20
,dz20
,rsq20
,rinv20
,rinvsq20
,r20
,qq20
,c6_20
,c12_20
;
92 __m128 dx30
,dy30
,dz30
,rsq30
,rinv30
,rinvsq30
,r30
,qq30
,c6_30
,c12_30
;
93 __m128 velec
,felec
,velecsum
,facel
,crf
,krf
,krf2
;
95 __m128 dummy_mask
,cutoff_mask
;
96 __m128 signbit
= _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
97 __m128 one
= _mm_set1_ps(1.0);
98 __m128 two
= _mm_set1_ps(2.0);
104 jindex
= nlist
->jindex
;
106 shiftidx
= nlist
->shift
;
108 shiftvec
= fr
->shift_vec
[0];
109 fshift
= fr
->fshift
[0];
110 facel
= _mm_set1_ps(fr
->ic
->epsfac
);
111 charge
= mdatoms
->chargeA
;
112 krf
= _mm_set1_ps(fr
->ic
->k_rf
);
113 krf2
= _mm_set1_ps(fr
->ic
->k_rf
*2.0);
114 crf
= _mm_set1_ps(fr
->ic
->c_rf
);
116 /* Setup water-specific parameters */
117 inr
= nlist
->iinr
[0];
118 iq1
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+1]));
119 iq2
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+2]));
120 iq3
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+3]));
122 /* Avoid stupid compiler warnings */
123 jnrA
= jnrB
= jnrC
= jnrD
= 0;
132 for(iidx
=0;iidx
<4*DIM
;iidx
++)
137 /* Start outer loop over neighborlists */
138 for(iidx
=0; iidx
<nri
; iidx
++)
140 /* Load shift vector for this list */
141 i_shift_offset
= DIM
*shiftidx
[iidx
];
143 /* Load limits for loop over neighbors */
144 j_index_start
= jindex
[iidx
];
145 j_index_end
= jindex
[iidx
+1];
147 /* Get outer coordinate index */
149 i_coord_offset
= DIM
*inr
;
151 /* Load i particle coords and add shift vector */
152 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec
+i_shift_offset
,x
+i_coord_offset
+DIM
,
153 &ix1
,&iy1
,&iz1
,&ix2
,&iy2
,&iz2
,&ix3
,&iy3
,&iz3
);
155 fix1
= _mm_setzero_ps();
156 fiy1
= _mm_setzero_ps();
157 fiz1
= _mm_setzero_ps();
158 fix2
= _mm_setzero_ps();
159 fiy2
= _mm_setzero_ps();
160 fiz2
= _mm_setzero_ps();
161 fix3
= _mm_setzero_ps();
162 fiy3
= _mm_setzero_ps();
163 fiz3
= _mm_setzero_ps();
165 /* Reset potential sums */
166 velecsum
= _mm_setzero_ps();
168 /* Start inner kernel loop */
169 for(jidx
=j_index_start
; jidx
<j_index_end
&& jjnr
[jidx
+3]>=0; jidx
+=4)
172 /* Get j neighbor index, and coordinate index */
177 j_coord_offsetA
= DIM
*jnrA
;
178 j_coord_offsetB
= DIM
*jnrB
;
179 j_coord_offsetC
= DIM
*jnrC
;
180 j_coord_offsetD
= DIM
*jnrD
;
182 /* load j atom coordinates */
183 gmx_mm_load_1rvec_4ptr_swizzle_ps(x
+j_coord_offsetA
,x
+j_coord_offsetB
,
184 x
+j_coord_offsetC
,x
+j_coord_offsetD
,
187 /* Calculate displacement vector */
188 dx10
= _mm_sub_ps(ix1
,jx0
);
189 dy10
= _mm_sub_ps(iy1
,jy0
);
190 dz10
= _mm_sub_ps(iz1
,jz0
);
191 dx20
= _mm_sub_ps(ix2
,jx0
);
192 dy20
= _mm_sub_ps(iy2
,jy0
);
193 dz20
= _mm_sub_ps(iz2
,jz0
);
194 dx30
= _mm_sub_ps(ix3
,jx0
);
195 dy30
= _mm_sub_ps(iy3
,jy0
);
196 dz30
= _mm_sub_ps(iz3
,jz0
);
198 /* Calculate squared distance and things based on it */
199 rsq10
= gmx_mm_calc_rsq_ps(dx10
,dy10
,dz10
);
200 rsq20
= gmx_mm_calc_rsq_ps(dx20
,dy20
,dz20
);
201 rsq30
= gmx_mm_calc_rsq_ps(dx30
,dy30
,dz30
);
203 rinv10
= avx128fma_invsqrt_f(rsq10
);
204 rinv20
= avx128fma_invsqrt_f(rsq20
);
205 rinv30
= avx128fma_invsqrt_f(rsq30
);
207 rinvsq10
= _mm_mul_ps(rinv10
,rinv10
);
208 rinvsq20
= _mm_mul_ps(rinv20
,rinv20
);
209 rinvsq30
= _mm_mul_ps(rinv30
,rinv30
);
211 /* Load parameters for j particles */
212 jq0
= gmx_mm_load_4real_swizzle_ps(charge
+jnrA
+0,charge
+jnrB
+0,
213 charge
+jnrC
+0,charge
+jnrD
+0);
215 fjx0
= _mm_setzero_ps();
216 fjy0
= _mm_setzero_ps();
217 fjz0
= _mm_setzero_ps();
219 /**************************
220 * CALCULATE INTERACTIONS *
221 **************************/
223 /* Compute parameters for interactions between i and j atoms */
224 qq10
= _mm_mul_ps(iq1
,jq0
);
226 /* REACTION-FIELD ELECTROSTATICS */
227 velec
= _mm_mul_ps(qq10
,_mm_sub_ps(_mm_macc_ps(krf
,rsq10
,rinv10
),crf
));
228 felec
= _mm_mul_ps(qq10
,_mm_msub_ps(rinv10
,rinvsq10
,krf2
));
230 /* Update potential sum for this i atom from the interaction with this j atom. */
231 velecsum
= _mm_add_ps(velecsum
,velec
);
235 /* Update vectorial force */
236 fix1
= _mm_macc_ps(dx10
,fscal
,fix1
);
237 fiy1
= _mm_macc_ps(dy10
,fscal
,fiy1
);
238 fiz1
= _mm_macc_ps(dz10
,fscal
,fiz1
);
240 fjx0
= _mm_macc_ps(dx10
,fscal
,fjx0
);
241 fjy0
= _mm_macc_ps(dy10
,fscal
,fjy0
);
242 fjz0
= _mm_macc_ps(dz10
,fscal
,fjz0
);
244 /**************************
245 * CALCULATE INTERACTIONS *
246 **************************/
248 /* Compute parameters for interactions between i and j atoms */
249 qq20
= _mm_mul_ps(iq2
,jq0
);
251 /* REACTION-FIELD ELECTROSTATICS */
252 velec
= _mm_mul_ps(qq20
,_mm_sub_ps(_mm_macc_ps(krf
,rsq20
,rinv20
),crf
));
253 felec
= _mm_mul_ps(qq20
,_mm_msub_ps(rinv20
,rinvsq20
,krf2
));
255 /* Update potential sum for this i atom from the interaction with this j atom. */
256 velecsum
= _mm_add_ps(velecsum
,velec
);
260 /* Update vectorial force */
261 fix2
= _mm_macc_ps(dx20
,fscal
,fix2
);
262 fiy2
= _mm_macc_ps(dy20
,fscal
,fiy2
);
263 fiz2
= _mm_macc_ps(dz20
,fscal
,fiz2
);
265 fjx0
= _mm_macc_ps(dx20
,fscal
,fjx0
);
266 fjy0
= _mm_macc_ps(dy20
,fscal
,fjy0
);
267 fjz0
= _mm_macc_ps(dz20
,fscal
,fjz0
);
269 /**************************
270 * CALCULATE INTERACTIONS *
271 **************************/
273 /* Compute parameters for interactions between i and j atoms */
274 qq30
= _mm_mul_ps(iq3
,jq0
);
276 /* REACTION-FIELD ELECTROSTATICS */
277 velec
= _mm_mul_ps(qq30
,_mm_sub_ps(_mm_macc_ps(krf
,rsq30
,rinv30
),crf
));
278 felec
= _mm_mul_ps(qq30
,_mm_msub_ps(rinv30
,rinvsq30
,krf2
));
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velecsum
= _mm_add_ps(velecsum
,velec
);
285 /* Update vectorial force */
286 fix3
= _mm_macc_ps(dx30
,fscal
,fix3
);
287 fiy3
= _mm_macc_ps(dy30
,fscal
,fiy3
);
288 fiz3
= _mm_macc_ps(dz30
,fscal
,fiz3
);
290 fjx0
= _mm_macc_ps(dx30
,fscal
,fjx0
);
291 fjy0
= _mm_macc_ps(dy30
,fscal
,fjy0
);
292 fjz0
= _mm_macc_ps(dz30
,fscal
,fjz0
);
294 fjptrA
= f
+j_coord_offsetA
;
295 fjptrB
= f
+j_coord_offsetB
;
296 fjptrC
= f
+j_coord_offsetC
;
297 fjptrD
= f
+j_coord_offsetD
;
299 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA
,fjptrB
,fjptrC
,fjptrD
,fjx0
,fjy0
,fjz0
);
301 /* Inner loop uses 105 flops */
307 /* Get j neighbor index, and coordinate index */
308 jnrlistA
= jjnr
[jidx
];
309 jnrlistB
= jjnr
[jidx
+1];
310 jnrlistC
= jjnr
[jidx
+2];
311 jnrlistD
= jjnr
[jidx
+3];
312 /* Sign of each element will be negative for non-real atoms.
313 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
314 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
316 dummy_mask
= gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i
*)(jjnr
+jidx
)),_mm_setzero_si128()));
317 jnrA
= (jnrlistA
>=0) ? jnrlistA
: 0;
318 jnrB
= (jnrlistB
>=0) ? jnrlistB
: 0;
319 jnrC
= (jnrlistC
>=0) ? jnrlistC
: 0;
320 jnrD
= (jnrlistD
>=0) ? jnrlistD
: 0;
321 j_coord_offsetA
= DIM
*jnrA
;
322 j_coord_offsetB
= DIM
*jnrB
;
323 j_coord_offsetC
= DIM
*jnrC
;
324 j_coord_offsetD
= DIM
*jnrD
;
326 /* load j atom coordinates */
327 gmx_mm_load_1rvec_4ptr_swizzle_ps(x
+j_coord_offsetA
,x
+j_coord_offsetB
,
328 x
+j_coord_offsetC
,x
+j_coord_offsetD
,
331 /* Calculate displacement vector */
332 dx10
= _mm_sub_ps(ix1
,jx0
);
333 dy10
= _mm_sub_ps(iy1
,jy0
);
334 dz10
= _mm_sub_ps(iz1
,jz0
);
335 dx20
= _mm_sub_ps(ix2
,jx0
);
336 dy20
= _mm_sub_ps(iy2
,jy0
);
337 dz20
= _mm_sub_ps(iz2
,jz0
);
338 dx30
= _mm_sub_ps(ix3
,jx0
);
339 dy30
= _mm_sub_ps(iy3
,jy0
);
340 dz30
= _mm_sub_ps(iz3
,jz0
);
342 /* Calculate squared distance and things based on it */
343 rsq10
= gmx_mm_calc_rsq_ps(dx10
,dy10
,dz10
);
344 rsq20
= gmx_mm_calc_rsq_ps(dx20
,dy20
,dz20
);
345 rsq30
= gmx_mm_calc_rsq_ps(dx30
,dy30
,dz30
);
347 rinv10
= avx128fma_invsqrt_f(rsq10
);
348 rinv20
= avx128fma_invsqrt_f(rsq20
);
349 rinv30
= avx128fma_invsqrt_f(rsq30
);
351 rinvsq10
= _mm_mul_ps(rinv10
,rinv10
);
352 rinvsq20
= _mm_mul_ps(rinv20
,rinv20
);
353 rinvsq30
= _mm_mul_ps(rinv30
,rinv30
);
355 /* Load parameters for j particles */
356 jq0
= gmx_mm_load_4real_swizzle_ps(charge
+jnrA
+0,charge
+jnrB
+0,
357 charge
+jnrC
+0,charge
+jnrD
+0);
359 fjx0
= _mm_setzero_ps();
360 fjy0
= _mm_setzero_ps();
361 fjz0
= _mm_setzero_ps();
363 /**************************
364 * CALCULATE INTERACTIONS *
365 **************************/
367 /* Compute parameters for interactions between i and j atoms */
368 qq10
= _mm_mul_ps(iq1
,jq0
);
370 /* REACTION-FIELD ELECTROSTATICS */
371 velec
= _mm_mul_ps(qq10
,_mm_sub_ps(_mm_macc_ps(krf
,rsq10
,rinv10
),crf
));
372 felec
= _mm_mul_ps(qq10
,_mm_msub_ps(rinv10
,rinvsq10
,krf2
));
374 /* Update potential sum for this i atom from the interaction with this j atom. */
375 velec
= _mm_andnot_ps(dummy_mask
,velec
);
376 velecsum
= _mm_add_ps(velecsum
,velec
);
380 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
382 /* Update vectorial force */
383 fix1
= _mm_macc_ps(dx10
,fscal
,fix1
);
384 fiy1
= _mm_macc_ps(dy10
,fscal
,fiy1
);
385 fiz1
= _mm_macc_ps(dz10
,fscal
,fiz1
);
387 fjx0
= _mm_macc_ps(dx10
,fscal
,fjx0
);
388 fjy0
= _mm_macc_ps(dy10
,fscal
,fjy0
);
389 fjz0
= _mm_macc_ps(dz10
,fscal
,fjz0
);
391 /**************************
392 * CALCULATE INTERACTIONS *
393 **************************/
395 /* Compute parameters for interactions between i and j atoms */
396 qq20
= _mm_mul_ps(iq2
,jq0
);
398 /* REACTION-FIELD ELECTROSTATICS */
399 velec
= _mm_mul_ps(qq20
,_mm_sub_ps(_mm_macc_ps(krf
,rsq20
,rinv20
),crf
));
400 felec
= _mm_mul_ps(qq20
,_mm_msub_ps(rinv20
,rinvsq20
,krf2
));
402 /* Update potential sum for this i atom from the interaction with this j atom. */
403 velec
= _mm_andnot_ps(dummy_mask
,velec
);
404 velecsum
= _mm_add_ps(velecsum
,velec
);
408 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
410 /* Update vectorial force */
411 fix2
= _mm_macc_ps(dx20
,fscal
,fix2
);
412 fiy2
= _mm_macc_ps(dy20
,fscal
,fiy2
);
413 fiz2
= _mm_macc_ps(dz20
,fscal
,fiz2
);
415 fjx0
= _mm_macc_ps(dx20
,fscal
,fjx0
);
416 fjy0
= _mm_macc_ps(dy20
,fscal
,fjy0
);
417 fjz0
= _mm_macc_ps(dz20
,fscal
,fjz0
);
419 /**************************
420 * CALCULATE INTERACTIONS *
421 **************************/
423 /* Compute parameters for interactions between i and j atoms */
424 qq30
= _mm_mul_ps(iq3
,jq0
);
426 /* REACTION-FIELD ELECTROSTATICS */
427 velec
= _mm_mul_ps(qq30
,_mm_sub_ps(_mm_macc_ps(krf
,rsq30
,rinv30
),crf
));
428 felec
= _mm_mul_ps(qq30
,_mm_msub_ps(rinv30
,rinvsq30
,krf2
));
430 /* Update potential sum for this i atom from the interaction with this j atom. */
431 velec
= _mm_andnot_ps(dummy_mask
,velec
);
432 velecsum
= _mm_add_ps(velecsum
,velec
);
436 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
438 /* Update vectorial force */
439 fix3
= _mm_macc_ps(dx30
,fscal
,fix3
);
440 fiy3
= _mm_macc_ps(dy30
,fscal
,fiy3
);
441 fiz3
= _mm_macc_ps(dz30
,fscal
,fiz3
);
443 fjx0
= _mm_macc_ps(dx30
,fscal
,fjx0
);
444 fjy0
= _mm_macc_ps(dy30
,fscal
,fjy0
);
445 fjz0
= _mm_macc_ps(dz30
,fscal
,fjz0
);
447 fjptrA
= (jnrlistA
>=0) ? f
+j_coord_offsetA
: scratch
;
448 fjptrB
= (jnrlistB
>=0) ? f
+j_coord_offsetB
: scratch
;
449 fjptrC
= (jnrlistC
>=0) ? f
+j_coord_offsetC
: scratch
;
450 fjptrD
= (jnrlistD
>=0) ? f
+j_coord_offsetD
: scratch
;
452 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA
,fjptrB
,fjptrC
,fjptrD
,fjx0
,fjy0
,fjz0
);
454 /* Inner loop uses 105 flops */
457 /* End of innermost loop */
459 gmx_mm_update_iforce_3atom_swizzle_ps(fix1
,fiy1
,fiz1
,fix2
,fiy2
,fiz2
,fix3
,fiy3
,fiz3
,
460 f
+i_coord_offset
+DIM
,fshift
+i_shift_offset
);
463 /* Update potential energies */
464 gmx_mm_update_1pot_ps(velecsum
,kernel_data
->energygrp_elec
+ggid
);
466 /* Increment number of inner iterations */
467 inneriter
+= j_index_end
- j_index_start
;
469 /* Outer loop uses 19 flops */
472 /* Increment number of outer iterations */
475 /* Update outer/inner flops */
477 inc_nrnb(nrnb
,eNR_NBKERNEL_ELEC_W4_VF
,outeriter
*19 + inneriter
*105);
480 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_single
481 * Electrostatics interaction: ReactionField
482 * VdW interaction: None
483 * Geometry: Water4-Particle
484 * Calculate force/pot: Force
487 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_single
488 (t_nblist
* gmx_restrict nlist
,
489 rvec
* gmx_restrict xx
,
490 rvec
* gmx_restrict ff
,
491 struct t_forcerec
* gmx_restrict fr
,
492 t_mdatoms
* gmx_restrict mdatoms
,
493 nb_kernel_data_t gmx_unused
* gmx_restrict kernel_data
,
494 t_nrnb
* gmx_restrict nrnb
)
496 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
497 * just 0 for non-waters.
498 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
499 * jnr indices corresponding to data put in the four positions in the SIMD register.
501 int i_shift_offset
,i_coord_offset
,outeriter
,inneriter
;
502 int j_index_start
,j_index_end
,jidx
,nri
,inr
,ggid
,iidx
;
503 int jnrA
,jnrB
,jnrC
,jnrD
;
504 int jnrlistA
,jnrlistB
,jnrlistC
,jnrlistD
;
505 int j_coord_offsetA
,j_coord_offsetB
,j_coord_offsetC
,j_coord_offsetD
;
506 int *iinr
,*jindex
,*jjnr
,*shiftidx
,*gid
;
508 real
*shiftvec
,*fshift
,*x
,*f
;
509 real
*fjptrA
,*fjptrB
,*fjptrC
,*fjptrD
;
511 __m128 fscal
,rcutoff
,rcutoff2
,jidxall
;
513 __m128 ix1
,iy1
,iz1
,fix1
,fiy1
,fiz1
,iq1
,isai1
;
515 __m128 ix2
,iy2
,iz2
,fix2
,fiy2
,fiz2
,iq2
,isai2
;
517 __m128 ix3
,iy3
,iz3
,fix3
,fiy3
,fiz3
,iq3
,isai3
;
518 int vdwjidx0A
,vdwjidx0B
,vdwjidx0C
,vdwjidx0D
;
519 __m128 jx0
,jy0
,jz0
,fjx0
,fjy0
,fjz0
,jq0
,isaj0
;
520 __m128 dx10
,dy10
,dz10
,rsq10
,rinv10
,rinvsq10
,r10
,qq10
,c6_10
,c12_10
;
521 __m128 dx20
,dy20
,dz20
,rsq20
,rinv20
,rinvsq20
,r20
,qq20
,c6_20
,c12_20
;
522 __m128 dx30
,dy30
,dz30
,rsq30
,rinv30
,rinvsq30
,r30
,qq30
,c6_30
,c12_30
;
523 __m128 velec
,felec
,velecsum
,facel
,crf
,krf
,krf2
;
525 __m128 dummy_mask
,cutoff_mask
;
526 __m128 signbit
= _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
527 __m128 one
= _mm_set1_ps(1.0);
528 __m128 two
= _mm_set1_ps(2.0);
534 jindex
= nlist
->jindex
;
536 shiftidx
= nlist
->shift
;
538 shiftvec
= fr
->shift_vec
[0];
539 fshift
= fr
->fshift
[0];
540 facel
= _mm_set1_ps(fr
->ic
->epsfac
);
541 charge
= mdatoms
->chargeA
;
542 krf
= _mm_set1_ps(fr
->ic
->k_rf
);
543 krf2
= _mm_set1_ps(fr
->ic
->k_rf
*2.0);
544 crf
= _mm_set1_ps(fr
->ic
->c_rf
);
546 /* Setup water-specific parameters */
547 inr
= nlist
->iinr
[0];
548 iq1
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+1]));
549 iq2
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+2]));
550 iq3
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+3]));
552 /* Avoid stupid compiler warnings */
553 jnrA
= jnrB
= jnrC
= jnrD
= 0;
562 for(iidx
=0;iidx
<4*DIM
;iidx
++)
567 /* Start outer loop over neighborlists */
568 for(iidx
=0; iidx
<nri
; iidx
++)
570 /* Load shift vector for this list */
571 i_shift_offset
= DIM
*shiftidx
[iidx
];
573 /* Load limits for loop over neighbors */
574 j_index_start
= jindex
[iidx
];
575 j_index_end
= jindex
[iidx
+1];
577 /* Get outer coordinate index */
579 i_coord_offset
= DIM
*inr
;
581 /* Load i particle coords and add shift vector */
582 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec
+i_shift_offset
,x
+i_coord_offset
+DIM
,
583 &ix1
,&iy1
,&iz1
,&ix2
,&iy2
,&iz2
,&ix3
,&iy3
,&iz3
);
585 fix1
= _mm_setzero_ps();
586 fiy1
= _mm_setzero_ps();
587 fiz1
= _mm_setzero_ps();
588 fix2
= _mm_setzero_ps();
589 fiy2
= _mm_setzero_ps();
590 fiz2
= _mm_setzero_ps();
591 fix3
= _mm_setzero_ps();
592 fiy3
= _mm_setzero_ps();
593 fiz3
= _mm_setzero_ps();
595 /* Start inner kernel loop */
596 for(jidx
=j_index_start
; jidx
<j_index_end
&& jjnr
[jidx
+3]>=0; jidx
+=4)
599 /* Get j neighbor index, and coordinate index */
604 j_coord_offsetA
= DIM
*jnrA
;
605 j_coord_offsetB
= DIM
*jnrB
;
606 j_coord_offsetC
= DIM
*jnrC
;
607 j_coord_offsetD
= DIM
*jnrD
;
609 /* load j atom coordinates */
610 gmx_mm_load_1rvec_4ptr_swizzle_ps(x
+j_coord_offsetA
,x
+j_coord_offsetB
,
611 x
+j_coord_offsetC
,x
+j_coord_offsetD
,
614 /* Calculate displacement vector */
615 dx10
= _mm_sub_ps(ix1
,jx0
);
616 dy10
= _mm_sub_ps(iy1
,jy0
);
617 dz10
= _mm_sub_ps(iz1
,jz0
);
618 dx20
= _mm_sub_ps(ix2
,jx0
);
619 dy20
= _mm_sub_ps(iy2
,jy0
);
620 dz20
= _mm_sub_ps(iz2
,jz0
);
621 dx30
= _mm_sub_ps(ix3
,jx0
);
622 dy30
= _mm_sub_ps(iy3
,jy0
);
623 dz30
= _mm_sub_ps(iz3
,jz0
);
625 /* Calculate squared distance and things based on it */
626 rsq10
= gmx_mm_calc_rsq_ps(dx10
,dy10
,dz10
);
627 rsq20
= gmx_mm_calc_rsq_ps(dx20
,dy20
,dz20
);
628 rsq30
= gmx_mm_calc_rsq_ps(dx30
,dy30
,dz30
);
630 rinv10
= avx128fma_invsqrt_f(rsq10
);
631 rinv20
= avx128fma_invsqrt_f(rsq20
);
632 rinv30
= avx128fma_invsqrt_f(rsq30
);
634 rinvsq10
= _mm_mul_ps(rinv10
,rinv10
);
635 rinvsq20
= _mm_mul_ps(rinv20
,rinv20
);
636 rinvsq30
= _mm_mul_ps(rinv30
,rinv30
);
638 /* Load parameters for j particles */
639 jq0
= gmx_mm_load_4real_swizzle_ps(charge
+jnrA
+0,charge
+jnrB
+0,
640 charge
+jnrC
+0,charge
+jnrD
+0);
642 fjx0
= _mm_setzero_ps();
643 fjy0
= _mm_setzero_ps();
644 fjz0
= _mm_setzero_ps();
646 /**************************
647 * CALCULATE INTERACTIONS *
648 **************************/
650 /* Compute parameters for interactions between i and j atoms */
651 qq10
= _mm_mul_ps(iq1
,jq0
);
653 /* REACTION-FIELD ELECTROSTATICS */
654 felec
= _mm_mul_ps(qq10
,_mm_msub_ps(rinv10
,rinvsq10
,krf2
));
658 /* Update vectorial force */
659 fix1
= _mm_macc_ps(dx10
,fscal
,fix1
);
660 fiy1
= _mm_macc_ps(dy10
,fscal
,fiy1
);
661 fiz1
= _mm_macc_ps(dz10
,fscal
,fiz1
);
663 fjx0
= _mm_macc_ps(dx10
,fscal
,fjx0
);
664 fjy0
= _mm_macc_ps(dy10
,fscal
,fjy0
);
665 fjz0
= _mm_macc_ps(dz10
,fscal
,fjz0
);
667 /**************************
668 * CALCULATE INTERACTIONS *
669 **************************/
671 /* Compute parameters for interactions between i and j atoms */
672 qq20
= _mm_mul_ps(iq2
,jq0
);
674 /* REACTION-FIELD ELECTROSTATICS */
675 felec
= _mm_mul_ps(qq20
,_mm_msub_ps(rinv20
,rinvsq20
,krf2
));
679 /* Update vectorial force */
680 fix2
= _mm_macc_ps(dx20
,fscal
,fix2
);
681 fiy2
= _mm_macc_ps(dy20
,fscal
,fiy2
);
682 fiz2
= _mm_macc_ps(dz20
,fscal
,fiz2
);
684 fjx0
= _mm_macc_ps(dx20
,fscal
,fjx0
);
685 fjy0
= _mm_macc_ps(dy20
,fscal
,fjy0
);
686 fjz0
= _mm_macc_ps(dz20
,fscal
,fjz0
);
688 /**************************
689 * CALCULATE INTERACTIONS *
690 **************************/
692 /* Compute parameters for interactions between i and j atoms */
693 qq30
= _mm_mul_ps(iq3
,jq0
);
695 /* REACTION-FIELD ELECTROSTATICS */
696 felec
= _mm_mul_ps(qq30
,_mm_msub_ps(rinv30
,rinvsq30
,krf2
));
700 /* Update vectorial force */
701 fix3
= _mm_macc_ps(dx30
,fscal
,fix3
);
702 fiy3
= _mm_macc_ps(dy30
,fscal
,fiy3
);
703 fiz3
= _mm_macc_ps(dz30
,fscal
,fiz3
);
705 fjx0
= _mm_macc_ps(dx30
,fscal
,fjx0
);
706 fjy0
= _mm_macc_ps(dy30
,fscal
,fjy0
);
707 fjz0
= _mm_macc_ps(dz30
,fscal
,fjz0
);
709 fjptrA
= f
+j_coord_offsetA
;
710 fjptrB
= f
+j_coord_offsetB
;
711 fjptrC
= f
+j_coord_offsetC
;
712 fjptrD
= f
+j_coord_offsetD
;
714 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA
,fjptrB
,fjptrC
,fjptrD
,fjx0
,fjy0
,fjz0
);
716 /* Inner loop uses 90 flops */
722 /* Get j neighbor index, and coordinate index */
723 jnrlistA
= jjnr
[jidx
];
724 jnrlistB
= jjnr
[jidx
+1];
725 jnrlistC
= jjnr
[jidx
+2];
726 jnrlistD
= jjnr
[jidx
+3];
727 /* Sign of each element will be negative for non-real atoms.
728 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
729 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
731 dummy_mask
= gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i
*)(jjnr
+jidx
)),_mm_setzero_si128()));
732 jnrA
= (jnrlistA
>=0) ? jnrlistA
: 0;
733 jnrB
= (jnrlistB
>=0) ? jnrlistB
: 0;
734 jnrC
= (jnrlistC
>=0) ? jnrlistC
: 0;
735 jnrD
= (jnrlistD
>=0) ? jnrlistD
: 0;
736 j_coord_offsetA
= DIM
*jnrA
;
737 j_coord_offsetB
= DIM
*jnrB
;
738 j_coord_offsetC
= DIM
*jnrC
;
739 j_coord_offsetD
= DIM
*jnrD
;
741 /* load j atom coordinates */
742 gmx_mm_load_1rvec_4ptr_swizzle_ps(x
+j_coord_offsetA
,x
+j_coord_offsetB
,
743 x
+j_coord_offsetC
,x
+j_coord_offsetD
,
746 /* Calculate displacement vector */
747 dx10
= _mm_sub_ps(ix1
,jx0
);
748 dy10
= _mm_sub_ps(iy1
,jy0
);
749 dz10
= _mm_sub_ps(iz1
,jz0
);
750 dx20
= _mm_sub_ps(ix2
,jx0
);
751 dy20
= _mm_sub_ps(iy2
,jy0
);
752 dz20
= _mm_sub_ps(iz2
,jz0
);
753 dx30
= _mm_sub_ps(ix3
,jx0
);
754 dy30
= _mm_sub_ps(iy3
,jy0
);
755 dz30
= _mm_sub_ps(iz3
,jz0
);
757 /* Calculate squared distance and things based on it */
758 rsq10
= gmx_mm_calc_rsq_ps(dx10
,dy10
,dz10
);
759 rsq20
= gmx_mm_calc_rsq_ps(dx20
,dy20
,dz20
);
760 rsq30
= gmx_mm_calc_rsq_ps(dx30
,dy30
,dz30
);
762 rinv10
= avx128fma_invsqrt_f(rsq10
);
763 rinv20
= avx128fma_invsqrt_f(rsq20
);
764 rinv30
= avx128fma_invsqrt_f(rsq30
);
766 rinvsq10
= _mm_mul_ps(rinv10
,rinv10
);
767 rinvsq20
= _mm_mul_ps(rinv20
,rinv20
);
768 rinvsq30
= _mm_mul_ps(rinv30
,rinv30
);
770 /* Load parameters for j particles */
771 jq0
= gmx_mm_load_4real_swizzle_ps(charge
+jnrA
+0,charge
+jnrB
+0,
772 charge
+jnrC
+0,charge
+jnrD
+0);
774 fjx0
= _mm_setzero_ps();
775 fjy0
= _mm_setzero_ps();
776 fjz0
= _mm_setzero_ps();
778 /**************************
779 * CALCULATE INTERACTIONS *
780 **************************/
782 /* Compute parameters for interactions between i and j atoms */
783 qq10
= _mm_mul_ps(iq1
,jq0
);
785 /* REACTION-FIELD ELECTROSTATICS */
786 felec
= _mm_mul_ps(qq10
,_mm_msub_ps(rinv10
,rinvsq10
,krf2
));
790 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
792 /* Update vectorial force */
793 fix1
= _mm_macc_ps(dx10
,fscal
,fix1
);
794 fiy1
= _mm_macc_ps(dy10
,fscal
,fiy1
);
795 fiz1
= _mm_macc_ps(dz10
,fscal
,fiz1
);
797 fjx0
= _mm_macc_ps(dx10
,fscal
,fjx0
);
798 fjy0
= _mm_macc_ps(dy10
,fscal
,fjy0
);
799 fjz0
= _mm_macc_ps(dz10
,fscal
,fjz0
);
801 /**************************
802 * CALCULATE INTERACTIONS *
803 **************************/
805 /* Compute parameters for interactions between i and j atoms */
806 qq20
= _mm_mul_ps(iq2
,jq0
);
808 /* REACTION-FIELD ELECTROSTATICS */
809 felec
= _mm_mul_ps(qq20
,_mm_msub_ps(rinv20
,rinvsq20
,krf2
));
813 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
815 /* Update vectorial force */
816 fix2
= _mm_macc_ps(dx20
,fscal
,fix2
);
817 fiy2
= _mm_macc_ps(dy20
,fscal
,fiy2
);
818 fiz2
= _mm_macc_ps(dz20
,fscal
,fiz2
);
820 fjx0
= _mm_macc_ps(dx20
,fscal
,fjx0
);
821 fjy0
= _mm_macc_ps(dy20
,fscal
,fjy0
);
822 fjz0
= _mm_macc_ps(dz20
,fscal
,fjz0
);
824 /**************************
825 * CALCULATE INTERACTIONS *
826 **************************/
828 /* Compute parameters for interactions between i and j atoms */
829 qq30
= _mm_mul_ps(iq3
,jq0
);
831 /* REACTION-FIELD ELECTROSTATICS */
832 felec
= _mm_mul_ps(qq30
,_mm_msub_ps(rinv30
,rinvsq30
,krf2
));
836 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
838 /* Update vectorial force */
839 fix3
= _mm_macc_ps(dx30
,fscal
,fix3
);
840 fiy3
= _mm_macc_ps(dy30
,fscal
,fiy3
);
841 fiz3
= _mm_macc_ps(dz30
,fscal
,fiz3
);
843 fjx0
= _mm_macc_ps(dx30
,fscal
,fjx0
);
844 fjy0
= _mm_macc_ps(dy30
,fscal
,fjy0
);
845 fjz0
= _mm_macc_ps(dz30
,fscal
,fjz0
);
847 fjptrA
= (jnrlistA
>=0) ? f
+j_coord_offsetA
: scratch
;
848 fjptrB
= (jnrlistB
>=0) ? f
+j_coord_offsetB
: scratch
;
849 fjptrC
= (jnrlistC
>=0) ? f
+j_coord_offsetC
: scratch
;
850 fjptrD
= (jnrlistD
>=0) ? f
+j_coord_offsetD
: scratch
;
852 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA
,fjptrB
,fjptrC
,fjptrD
,fjx0
,fjy0
,fjz0
);
854 /* Inner loop uses 90 flops */
857 /* End of innermost loop */
859 gmx_mm_update_iforce_3atom_swizzle_ps(fix1
,fiy1
,fiz1
,fix2
,fiy2
,fiz2
,fix3
,fiy3
,fiz3
,
860 f
+i_coord_offset
+DIM
,fshift
+i_shift_offset
);
862 /* Increment number of inner iterations */
863 inneriter
+= j_index_end
- j_index_start
;
865 /* Outer loop uses 18 flops */
868 /* Increment number of outer iterations */
871 /* Update outer/inner flops */
873 inc_nrnb(nrnb
,eNR_NBKERNEL_ELEC_W4_F
,outeriter
*18 + inneriter
*90);