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 sse2_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_sse2_single.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_sse2_single
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: None
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_sse2_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 SSE, 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 tx
,ty
,tz
,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 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
123 rcutoff_scalar
= fr
->ic
->rcoulomb
;
124 rcutoff
= _mm_set1_ps(rcutoff_scalar
);
125 rcutoff2
= _mm_mul_ps(rcutoff
,rcutoff
);
127 /* Avoid stupid compiler warnings */
128 jnrA
= jnrB
= jnrC
= jnrD
= 0;
137 for(iidx
=0;iidx
<4*DIM
;iidx
++)
142 /* Start outer loop over neighborlists */
143 for(iidx
=0; iidx
<nri
; iidx
++)
145 /* Load shift vector for this list */
146 i_shift_offset
= DIM
*shiftidx
[iidx
];
148 /* Load limits for loop over neighbors */
149 j_index_start
= jindex
[iidx
];
150 j_index_end
= jindex
[iidx
+1];
152 /* Get outer coordinate index */
154 i_coord_offset
= DIM
*inr
;
156 /* Load i particle coords and add shift vector */
157 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec
+i_shift_offset
,x
+i_coord_offset
+DIM
,
158 &ix1
,&iy1
,&iz1
,&ix2
,&iy2
,&iz2
,&ix3
,&iy3
,&iz3
);
160 fix1
= _mm_setzero_ps();
161 fiy1
= _mm_setzero_ps();
162 fiz1
= _mm_setzero_ps();
163 fix2
= _mm_setzero_ps();
164 fiy2
= _mm_setzero_ps();
165 fiz2
= _mm_setzero_ps();
166 fix3
= _mm_setzero_ps();
167 fiy3
= _mm_setzero_ps();
168 fiz3
= _mm_setzero_ps();
170 /* Reset potential sums */
171 velecsum
= _mm_setzero_ps();
173 /* Start inner kernel loop */
174 for(jidx
=j_index_start
; jidx
<j_index_end
&& jjnr
[jidx
+3]>=0; jidx
+=4)
177 /* Get j neighbor index, and coordinate index */
182 j_coord_offsetA
= DIM
*jnrA
;
183 j_coord_offsetB
= DIM
*jnrB
;
184 j_coord_offsetC
= DIM
*jnrC
;
185 j_coord_offsetD
= DIM
*jnrD
;
187 /* load j atom coordinates */
188 gmx_mm_load_1rvec_4ptr_swizzle_ps(x
+j_coord_offsetA
,x
+j_coord_offsetB
,
189 x
+j_coord_offsetC
,x
+j_coord_offsetD
,
192 /* Calculate displacement vector */
193 dx10
= _mm_sub_ps(ix1
,jx0
);
194 dy10
= _mm_sub_ps(iy1
,jy0
);
195 dz10
= _mm_sub_ps(iz1
,jz0
);
196 dx20
= _mm_sub_ps(ix2
,jx0
);
197 dy20
= _mm_sub_ps(iy2
,jy0
);
198 dz20
= _mm_sub_ps(iz2
,jz0
);
199 dx30
= _mm_sub_ps(ix3
,jx0
);
200 dy30
= _mm_sub_ps(iy3
,jy0
);
201 dz30
= _mm_sub_ps(iz3
,jz0
);
203 /* Calculate squared distance and things based on it */
204 rsq10
= gmx_mm_calc_rsq_ps(dx10
,dy10
,dz10
);
205 rsq20
= gmx_mm_calc_rsq_ps(dx20
,dy20
,dz20
);
206 rsq30
= gmx_mm_calc_rsq_ps(dx30
,dy30
,dz30
);
208 rinv10
= sse2_invsqrt_f(rsq10
);
209 rinv20
= sse2_invsqrt_f(rsq20
);
210 rinv30
= sse2_invsqrt_f(rsq30
);
212 rinvsq10
= _mm_mul_ps(rinv10
,rinv10
);
213 rinvsq20
= _mm_mul_ps(rinv20
,rinv20
);
214 rinvsq30
= _mm_mul_ps(rinv30
,rinv30
);
216 /* Load parameters for j particles */
217 jq0
= gmx_mm_load_4real_swizzle_ps(charge
+jnrA
+0,charge
+jnrB
+0,
218 charge
+jnrC
+0,charge
+jnrD
+0);
220 fjx0
= _mm_setzero_ps();
221 fjy0
= _mm_setzero_ps();
222 fjz0
= _mm_setzero_ps();
224 /**************************
225 * CALCULATE INTERACTIONS *
226 **************************/
228 if (gmx_mm_any_lt(rsq10
,rcutoff2
))
231 /* Compute parameters for interactions between i and j atoms */
232 qq10
= _mm_mul_ps(iq1
,jq0
);
234 /* REACTION-FIELD ELECTROSTATICS */
235 velec
= _mm_mul_ps(qq10
,_mm_sub_ps(_mm_add_ps(rinv10
,_mm_mul_ps(krf
,rsq10
)),crf
));
236 felec
= _mm_mul_ps(qq10
,_mm_sub_ps(_mm_mul_ps(rinv10
,rinvsq10
),krf2
));
238 cutoff_mask
= _mm_cmplt_ps(rsq10
,rcutoff2
);
240 /* Update potential sum for this i atom from the interaction with this j atom. */
241 velec
= _mm_and_ps(velec
,cutoff_mask
);
242 velecsum
= _mm_add_ps(velecsum
,velec
);
246 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
248 /* Calculate temporary vectorial force */
249 tx
= _mm_mul_ps(fscal
,dx10
);
250 ty
= _mm_mul_ps(fscal
,dy10
);
251 tz
= _mm_mul_ps(fscal
,dz10
);
253 /* Update vectorial force */
254 fix1
= _mm_add_ps(fix1
,tx
);
255 fiy1
= _mm_add_ps(fiy1
,ty
);
256 fiz1
= _mm_add_ps(fiz1
,tz
);
258 fjx0
= _mm_add_ps(fjx0
,tx
);
259 fjy0
= _mm_add_ps(fjy0
,ty
);
260 fjz0
= _mm_add_ps(fjz0
,tz
);
264 /**************************
265 * CALCULATE INTERACTIONS *
266 **************************/
268 if (gmx_mm_any_lt(rsq20
,rcutoff2
))
271 /* Compute parameters for interactions between i and j atoms */
272 qq20
= _mm_mul_ps(iq2
,jq0
);
274 /* REACTION-FIELD ELECTROSTATICS */
275 velec
= _mm_mul_ps(qq20
,_mm_sub_ps(_mm_add_ps(rinv20
,_mm_mul_ps(krf
,rsq20
)),crf
));
276 felec
= _mm_mul_ps(qq20
,_mm_sub_ps(_mm_mul_ps(rinv20
,rinvsq20
),krf2
));
278 cutoff_mask
= _mm_cmplt_ps(rsq20
,rcutoff2
);
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velec
= _mm_and_ps(velec
,cutoff_mask
);
282 velecsum
= _mm_add_ps(velecsum
,velec
);
286 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
288 /* Calculate temporary vectorial force */
289 tx
= _mm_mul_ps(fscal
,dx20
);
290 ty
= _mm_mul_ps(fscal
,dy20
);
291 tz
= _mm_mul_ps(fscal
,dz20
);
293 /* Update vectorial force */
294 fix2
= _mm_add_ps(fix2
,tx
);
295 fiy2
= _mm_add_ps(fiy2
,ty
);
296 fiz2
= _mm_add_ps(fiz2
,tz
);
298 fjx0
= _mm_add_ps(fjx0
,tx
);
299 fjy0
= _mm_add_ps(fjy0
,ty
);
300 fjz0
= _mm_add_ps(fjz0
,tz
);
304 /**************************
305 * CALCULATE INTERACTIONS *
306 **************************/
308 if (gmx_mm_any_lt(rsq30
,rcutoff2
))
311 /* Compute parameters for interactions between i and j atoms */
312 qq30
= _mm_mul_ps(iq3
,jq0
);
314 /* REACTION-FIELD ELECTROSTATICS */
315 velec
= _mm_mul_ps(qq30
,_mm_sub_ps(_mm_add_ps(rinv30
,_mm_mul_ps(krf
,rsq30
)),crf
));
316 felec
= _mm_mul_ps(qq30
,_mm_sub_ps(_mm_mul_ps(rinv30
,rinvsq30
),krf2
));
318 cutoff_mask
= _mm_cmplt_ps(rsq30
,rcutoff2
);
320 /* Update potential sum for this i atom from the interaction with this j atom. */
321 velec
= _mm_and_ps(velec
,cutoff_mask
);
322 velecsum
= _mm_add_ps(velecsum
,velec
);
326 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
328 /* Calculate temporary vectorial force */
329 tx
= _mm_mul_ps(fscal
,dx30
);
330 ty
= _mm_mul_ps(fscal
,dy30
);
331 tz
= _mm_mul_ps(fscal
,dz30
);
333 /* Update vectorial force */
334 fix3
= _mm_add_ps(fix3
,tx
);
335 fiy3
= _mm_add_ps(fiy3
,ty
);
336 fiz3
= _mm_add_ps(fiz3
,tz
);
338 fjx0
= _mm_add_ps(fjx0
,tx
);
339 fjy0
= _mm_add_ps(fjy0
,ty
);
340 fjz0
= _mm_add_ps(fjz0
,tz
);
344 fjptrA
= f
+j_coord_offsetA
;
345 fjptrB
= f
+j_coord_offsetB
;
346 fjptrC
= f
+j_coord_offsetC
;
347 fjptrD
= f
+j_coord_offsetD
;
349 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA
,fjptrB
,fjptrC
,fjptrD
,fjx0
,fjy0
,fjz0
);
351 /* Inner loop uses 108 flops */
357 /* Get j neighbor index, and coordinate index */
358 jnrlistA
= jjnr
[jidx
];
359 jnrlistB
= jjnr
[jidx
+1];
360 jnrlistC
= jjnr
[jidx
+2];
361 jnrlistD
= jjnr
[jidx
+3];
362 /* Sign of each element will be negative for non-real atoms.
363 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
364 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
366 dummy_mask
= gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i
*)(jjnr
+jidx
)),_mm_setzero_si128()));
367 jnrA
= (jnrlistA
>=0) ? jnrlistA
: 0;
368 jnrB
= (jnrlistB
>=0) ? jnrlistB
: 0;
369 jnrC
= (jnrlistC
>=0) ? jnrlistC
: 0;
370 jnrD
= (jnrlistD
>=0) ? jnrlistD
: 0;
371 j_coord_offsetA
= DIM
*jnrA
;
372 j_coord_offsetB
= DIM
*jnrB
;
373 j_coord_offsetC
= DIM
*jnrC
;
374 j_coord_offsetD
= DIM
*jnrD
;
376 /* load j atom coordinates */
377 gmx_mm_load_1rvec_4ptr_swizzle_ps(x
+j_coord_offsetA
,x
+j_coord_offsetB
,
378 x
+j_coord_offsetC
,x
+j_coord_offsetD
,
381 /* Calculate displacement vector */
382 dx10
= _mm_sub_ps(ix1
,jx0
);
383 dy10
= _mm_sub_ps(iy1
,jy0
);
384 dz10
= _mm_sub_ps(iz1
,jz0
);
385 dx20
= _mm_sub_ps(ix2
,jx0
);
386 dy20
= _mm_sub_ps(iy2
,jy0
);
387 dz20
= _mm_sub_ps(iz2
,jz0
);
388 dx30
= _mm_sub_ps(ix3
,jx0
);
389 dy30
= _mm_sub_ps(iy3
,jy0
);
390 dz30
= _mm_sub_ps(iz3
,jz0
);
392 /* Calculate squared distance and things based on it */
393 rsq10
= gmx_mm_calc_rsq_ps(dx10
,dy10
,dz10
);
394 rsq20
= gmx_mm_calc_rsq_ps(dx20
,dy20
,dz20
);
395 rsq30
= gmx_mm_calc_rsq_ps(dx30
,dy30
,dz30
);
397 rinv10
= sse2_invsqrt_f(rsq10
);
398 rinv20
= sse2_invsqrt_f(rsq20
);
399 rinv30
= sse2_invsqrt_f(rsq30
);
401 rinvsq10
= _mm_mul_ps(rinv10
,rinv10
);
402 rinvsq20
= _mm_mul_ps(rinv20
,rinv20
);
403 rinvsq30
= _mm_mul_ps(rinv30
,rinv30
);
405 /* Load parameters for j particles */
406 jq0
= gmx_mm_load_4real_swizzle_ps(charge
+jnrA
+0,charge
+jnrB
+0,
407 charge
+jnrC
+0,charge
+jnrD
+0);
409 fjx0
= _mm_setzero_ps();
410 fjy0
= _mm_setzero_ps();
411 fjz0
= _mm_setzero_ps();
413 /**************************
414 * CALCULATE INTERACTIONS *
415 **************************/
417 if (gmx_mm_any_lt(rsq10
,rcutoff2
))
420 /* Compute parameters for interactions between i and j atoms */
421 qq10
= _mm_mul_ps(iq1
,jq0
);
423 /* REACTION-FIELD ELECTROSTATICS */
424 velec
= _mm_mul_ps(qq10
,_mm_sub_ps(_mm_add_ps(rinv10
,_mm_mul_ps(krf
,rsq10
)),crf
));
425 felec
= _mm_mul_ps(qq10
,_mm_sub_ps(_mm_mul_ps(rinv10
,rinvsq10
),krf2
));
427 cutoff_mask
= _mm_cmplt_ps(rsq10
,rcutoff2
);
429 /* Update potential sum for this i atom from the interaction with this j atom. */
430 velec
= _mm_and_ps(velec
,cutoff_mask
);
431 velec
= _mm_andnot_ps(dummy_mask
,velec
);
432 velecsum
= _mm_add_ps(velecsum
,velec
);
436 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
438 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
440 /* Calculate temporary vectorial force */
441 tx
= _mm_mul_ps(fscal
,dx10
);
442 ty
= _mm_mul_ps(fscal
,dy10
);
443 tz
= _mm_mul_ps(fscal
,dz10
);
445 /* Update vectorial force */
446 fix1
= _mm_add_ps(fix1
,tx
);
447 fiy1
= _mm_add_ps(fiy1
,ty
);
448 fiz1
= _mm_add_ps(fiz1
,tz
);
450 fjx0
= _mm_add_ps(fjx0
,tx
);
451 fjy0
= _mm_add_ps(fjy0
,ty
);
452 fjz0
= _mm_add_ps(fjz0
,tz
);
456 /**************************
457 * CALCULATE INTERACTIONS *
458 **************************/
460 if (gmx_mm_any_lt(rsq20
,rcutoff2
))
463 /* Compute parameters for interactions between i and j atoms */
464 qq20
= _mm_mul_ps(iq2
,jq0
);
466 /* REACTION-FIELD ELECTROSTATICS */
467 velec
= _mm_mul_ps(qq20
,_mm_sub_ps(_mm_add_ps(rinv20
,_mm_mul_ps(krf
,rsq20
)),crf
));
468 felec
= _mm_mul_ps(qq20
,_mm_sub_ps(_mm_mul_ps(rinv20
,rinvsq20
),krf2
));
470 cutoff_mask
= _mm_cmplt_ps(rsq20
,rcutoff2
);
472 /* Update potential sum for this i atom from the interaction with this j atom. */
473 velec
= _mm_and_ps(velec
,cutoff_mask
);
474 velec
= _mm_andnot_ps(dummy_mask
,velec
);
475 velecsum
= _mm_add_ps(velecsum
,velec
);
479 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
481 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
483 /* Calculate temporary vectorial force */
484 tx
= _mm_mul_ps(fscal
,dx20
);
485 ty
= _mm_mul_ps(fscal
,dy20
);
486 tz
= _mm_mul_ps(fscal
,dz20
);
488 /* Update vectorial force */
489 fix2
= _mm_add_ps(fix2
,tx
);
490 fiy2
= _mm_add_ps(fiy2
,ty
);
491 fiz2
= _mm_add_ps(fiz2
,tz
);
493 fjx0
= _mm_add_ps(fjx0
,tx
);
494 fjy0
= _mm_add_ps(fjy0
,ty
);
495 fjz0
= _mm_add_ps(fjz0
,tz
);
499 /**************************
500 * CALCULATE INTERACTIONS *
501 **************************/
503 if (gmx_mm_any_lt(rsq30
,rcutoff2
))
506 /* Compute parameters for interactions between i and j atoms */
507 qq30
= _mm_mul_ps(iq3
,jq0
);
509 /* REACTION-FIELD ELECTROSTATICS */
510 velec
= _mm_mul_ps(qq30
,_mm_sub_ps(_mm_add_ps(rinv30
,_mm_mul_ps(krf
,rsq30
)),crf
));
511 felec
= _mm_mul_ps(qq30
,_mm_sub_ps(_mm_mul_ps(rinv30
,rinvsq30
),krf2
));
513 cutoff_mask
= _mm_cmplt_ps(rsq30
,rcutoff2
);
515 /* Update potential sum for this i atom from the interaction with this j atom. */
516 velec
= _mm_and_ps(velec
,cutoff_mask
);
517 velec
= _mm_andnot_ps(dummy_mask
,velec
);
518 velecsum
= _mm_add_ps(velecsum
,velec
);
522 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
524 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
526 /* Calculate temporary vectorial force */
527 tx
= _mm_mul_ps(fscal
,dx30
);
528 ty
= _mm_mul_ps(fscal
,dy30
);
529 tz
= _mm_mul_ps(fscal
,dz30
);
531 /* Update vectorial force */
532 fix3
= _mm_add_ps(fix3
,tx
);
533 fiy3
= _mm_add_ps(fiy3
,ty
);
534 fiz3
= _mm_add_ps(fiz3
,tz
);
536 fjx0
= _mm_add_ps(fjx0
,tx
);
537 fjy0
= _mm_add_ps(fjy0
,ty
);
538 fjz0
= _mm_add_ps(fjz0
,tz
);
542 fjptrA
= (jnrlistA
>=0) ? f
+j_coord_offsetA
: scratch
;
543 fjptrB
= (jnrlistB
>=0) ? f
+j_coord_offsetB
: scratch
;
544 fjptrC
= (jnrlistC
>=0) ? f
+j_coord_offsetC
: scratch
;
545 fjptrD
= (jnrlistD
>=0) ? f
+j_coord_offsetD
: scratch
;
547 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA
,fjptrB
,fjptrC
,fjptrD
,fjx0
,fjy0
,fjz0
);
549 /* Inner loop uses 108 flops */
552 /* End of innermost loop */
554 gmx_mm_update_iforce_3atom_swizzle_ps(fix1
,fiy1
,fiz1
,fix2
,fiy2
,fiz2
,fix3
,fiy3
,fiz3
,
555 f
+i_coord_offset
+DIM
,fshift
+i_shift_offset
);
558 /* Update potential energies */
559 gmx_mm_update_1pot_ps(velecsum
,kernel_data
->energygrp_elec
+ggid
);
561 /* Increment number of inner iterations */
562 inneriter
+= j_index_end
- j_index_start
;
564 /* Outer loop uses 19 flops */
567 /* Increment number of outer iterations */
570 /* Update outer/inner flops */
572 inc_nrnb(nrnb
,eNR_NBKERNEL_ELEC_W4_VF
,outeriter
*19 + inneriter
*108);
575 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_sse2_single
576 * Electrostatics interaction: ReactionField
577 * VdW interaction: None
578 * Geometry: Water4-Particle
579 * Calculate force/pot: Force
582 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_sse2_single
583 (t_nblist
* gmx_restrict nlist
,
584 rvec
* gmx_restrict xx
,
585 rvec
* gmx_restrict ff
,
586 struct t_forcerec
* gmx_restrict fr
,
587 t_mdatoms
* gmx_restrict mdatoms
,
588 nb_kernel_data_t gmx_unused
* gmx_restrict kernel_data
,
589 t_nrnb
* gmx_restrict nrnb
)
591 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
592 * just 0 for non-waters.
593 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
594 * jnr indices corresponding to data put in the four positions in the SIMD register.
596 int i_shift_offset
,i_coord_offset
,outeriter
,inneriter
;
597 int j_index_start
,j_index_end
,jidx
,nri
,inr
,ggid
,iidx
;
598 int jnrA
,jnrB
,jnrC
,jnrD
;
599 int jnrlistA
,jnrlistB
,jnrlistC
,jnrlistD
;
600 int j_coord_offsetA
,j_coord_offsetB
,j_coord_offsetC
,j_coord_offsetD
;
601 int *iinr
,*jindex
,*jjnr
,*shiftidx
,*gid
;
603 real
*shiftvec
,*fshift
,*x
,*f
;
604 real
*fjptrA
,*fjptrB
,*fjptrC
,*fjptrD
;
606 __m128 tx
,ty
,tz
,fscal
,rcutoff
,rcutoff2
,jidxall
;
608 __m128 ix1
,iy1
,iz1
,fix1
,fiy1
,fiz1
,iq1
,isai1
;
610 __m128 ix2
,iy2
,iz2
,fix2
,fiy2
,fiz2
,iq2
,isai2
;
612 __m128 ix3
,iy3
,iz3
,fix3
,fiy3
,fiz3
,iq3
,isai3
;
613 int vdwjidx0A
,vdwjidx0B
,vdwjidx0C
,vdwjidx0D
;
614 __m128 jx0
,jy0
,jz0
,fjx0
,fjy0
,fjz0
,jq0
,isaj0
;
615 __m128 dx10
,dy10
,dz10
,rsq10
,rinv10
,rinvsq10
,r10
,qq10
,c6_10
,c12_10
;
616 __m128 dx20
,dy20
,dz20
,rsq20
,rinv20
,rinvsq20
,r20
,qq20
,c6_20
,c12_20
;
617 __m128 dx30
,dy30
,dz30
,rsq30
,rinv30
,rinvsq30
,r30
,qq30
,c6_30
,c12_30
;
618 __m128 velec
,felec
,velecsum
,facel
,crf
,krf
,krf2
;
620 __m128 dummy_mask
,cutoff_mask
;
621 __m128 signbit
= _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
622 __m128 one
= _mm_set1_ps(1.0);
623 __m128 two
= _mm_set1_ps(2.0);
629 jindex
= nlist
->jindex
;
631 shiftidx
= nlist
->shift
;
633 shiftvec
= fr
->shift_vec
[0];
634 fshift
= fr
->fshift
[0];
635 facel
= _mm_set1_ps(fr
->ic
->epsfac
);
636 charge
= mdatoms
->chargeA
;
637 krf
= _mm_set1_ps(fr
->ic
->k_rf
);
638 krf2
= _mm_set1_ps(fr
->ic
->k_rf
*2.0);
639 crf
= _mm_set1_ps(fr
->ic
->c_rf
);
641 /* Setup water-specific parameters */
642 inr
= nlist
->iinr
[0];
643 iq1
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+1]));
644 iq2
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+2]));
645 iq3
= _mm_mul_ps(facel
,_mm_set1_ps(charge
[inr
+3]));
647 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
648 rcutoff_scalar
= fr
->ic
->rcoulomb
;
649 rcutoff
= _mm_set1_ps(rcutoff_scalar
);
650 rcutoff2
= _mm_mul_ps(rcutoff
,rcutoff
);
652 /* Avoid stupid compiler warnings */
653 jnrA
= jnrB
= jnrC
= jnrD
= 0;
662 for(iidx
=0;iidx
<4*DIM
;iidx
++)
667 /* Start outer loop over neighborlists */
668 for(iidx
=0; iidx
<nri
; iidx
++)
670 /* Load shift vector for this list */
671 i_shift_offset
= DIM
*shiftidx
[iidx
];
673 /* Load limits for loop over neighbors */
674 j_index_start
= jindex
[iidx
];
675 j_index_end
= jindex
[iidx
+1];
677 /* Get outer coordinate index */
679 i_coord_offset
= DIM
*inr
;
681 /* Load i particle coords and add shift vector */
682 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec
+i_shift_offset
,x
+i_coord_offset
+DIM
,
683 &ix1
,&iy1
,&iz1
,&ix2
,&iy2
,&iz2
,&ix3
,&iy3
,&iz3
);
685 fix1
= _mm_setzero_ps();
686 fiy1
= _mm_setzero_ps();
687 fiz1
= _mm_setzero_ps();
688 fix2
= _mm_setzero_ps();
689 fiy2
= _mm_setzero_ps();
690 fiz2
= _mm_setzero_ps();
691 fix3
= _mm_setzero_ps();
692 fiy3
= _mm_setzero_ps();
693 fiz3
= _mm_setzero_ps();
695 /* Start inner kernel loop */
696 for(jidx
=j_index_start
; jidx
<j_index_end
&& jjnr
[jidx
+3]>=0; jidx
+=4)
699 /* Get j neighbor index, and coordinate index */
704 j_coord_offsetA
= DIM
*jnrA
;
705 j_coord_offsetB
= DIM
*jnrB
;
706 j_coord_offsetC
= DIM
*jnrC
;
707 j_coord_offsetD
= DIM
*jnrD
;
709 /* load j atom coordinates */
710 gmx_mm_load_1rvec_4ptr_swizzle_ps(x
+j_coord_offsetA
,x
+j_coord_offsetB
,
711 x
+j_coord_offsetC
,x
+j_coord_offsetD
,
714 /* Calculate displacement vector */
715 dx10
= _mm_sub_ps(ix1
,jx0
);
716 dy10
= _mm_sub_ps(iy1
,jy0
);
717 dz10
= _mm_sub_ps(iz1
,jz0
);
718 dx20
= _mm_sub_ps(ix2
,jx0
);
719 dy20
= _mm_sub_ps(iy2
,jy0
);
720 dz20
= _mm_sub_ps(iz2
,jz0
);
721 dx30
= _mm_sub_ps(ix3
,jx0
);
722 dy30
= _mm_sub_ps(iy3
,jy0
);
723 dz30
= _mm_sub_ps(iz3
,jz0
);
725 /* Calculate squared distance and things based on it */
726 rsq10
= gmx_mm_calc_rsq_ps(dx10
,dy10
,dz10
);
727 rsq20
= gmx_mm_calc_rsq_ps(dx20
,dy20
,dz20
);
728 rsq30
= gmx_mm_calc_rsq_ps(dx30
,dy30
,dz30
);
730 rinv10
= sse2_invsqrt_f(rsq10
);
731 rinv20
= sse2_invsqrt_f(rsq20
);
732 rinv30
= sse2_invsqrt_f(rsq30
);
734 rinvsq10
= _mm_mul_ps(rinv10
,rinv10
);
735 rinvsq20
= _mm_mul_ps(rinv20
,rinv20
);
736 rinvsq30
= _mm_mul_ps(rinv30
,rinv30
);
738 /* Load parameters for j particles */
739 jq0
= gmx_mm_load_4real_swizzle_ps(charge
+jnrA
+0,charge
+jnrB
+0,
740 charge
+jnrC
+0,charge
+jnrD
+0);
742 fjx0
= _mm_setzero_ps();
743 fjy0
= _mm_setzero_ps();
744 fjz0
= _mm_setzero_ps();
746 /**************************
747 * CALCULATE INTERACTIONS *
748 **************************/
750 if (gmx_mm_any_lt(rsq10
,rcutoff2
))
753 /* Compute parameters for interactions between i and j atoms */
754 qq10
= _mm_mul_ps(iq1
,jq0
);
756 /* REACTION-FIELD ELECTROSTATICS */
757 felec
= _mm_mul_ps(qq10
,_mm_sub_ps(_mm_mul_ps(rinv10
,rinvsq10
),krf2
));
759 cutoff_mask
= _mm_cmplt_ps(rsq10
,rcutoff2
);
763 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
765 /* Calculate temporary vectorial force */
766 tx
= _mm_mul_ps(fscal
,dx10
);
767 ty
= _mm_mul_ps(fscal
,dy10
);
768 tz
= _mm_mul_ps(fscal
,dz10
);
770 /* Update vectorial force */
771 fix1
= _mm_add_ps(fix1
,tx
);
772 fiy1
= _mm_add_ps(fiy1
,ty
);
773 fiz1
= _mm_add_ps(fiz1
,tz
);
775 fjx0
= _mm_add_ps(fjx0
,tx
);
776 fjy0
= _mm_add_ps(fjy0
,ty
);
777 fjz0
= _mm_add_ps(fjz0
,tz
);
781 /**************************
782 * CALCULATE INTERACTIONS *
783 **************************/
785 if (gmx_mm_any_lt(rsq20
,rcutoff2
))
788 /* Compute parameters for interactions between i and j atoms */
789 qq20
= _mm_mul_ps(iq2
,jq0
);
791 /* REACTION-FIELD ELECTROSTATICS */
792 felec
= _mm_mul_ps(qq20
,_mm_sub_ps(_mm_mul_ps(rinv20
,rinvsq20
),krf2
));
794 cutoff_mask
= _mm_cmplt_ps(rsq20
,rcutoff2
);
798 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
800 /* Calculate temporary vectorial force */
801 tx
= _mm_mul_ps(fscal
,dx20
);
802 ty
= _mm_mul_ps(fscal
,dy20
);
803 tz
= _mm_mul_ps(fscal
,dz20
);
805 /* Update vectorial force */
806 fix2
= _mm_add_ps(fix2
,tx
);
807 fiy2
= _mm_add_ps(fiy2
,ty
);
808 fiz2
= _mm_add_ps(fiz2
,tz
);
810 fjx0
= _mm_add_ps(fjx0
,tx
);
811 fjy0
= _mm_add_ps(fjy0
,ty
);
812 fjz0
= _mm_add_ps(fjz0
,tz
);
816 /**************************
817 * CALCULATE INTERACTIONS *
818 **************************/
820 if (gmx_mm_any_lt(rsq30
,rcutoff2
))
823 /* Compute parameters for interactions between i and j atoms */
824 qq30
= _mm_mul_ps(iq3
,jq0
);
826 /* REACTION-FIELD ELECTROSTATICS */
827 felec
= _mm_mul_ps(qq30
,_mm_sub_ps(_mm_mul_ps(rinv30
,rinvsq30
),krf2
));
829 cutoff_mask
= _mm_cmplt_ps(rsq30
,rcutoff2
);
833 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
835 /* Calculate temporary vectorial force */
836 tx
= _mm_mul_ps(fscal
,dx30
);
837 ty
= _mm_mul_ps(fscal
,dy30
);
838 tz
= _mm_mul_ps(fscal
,dz30
);
840 /* Update vectorial force */
841 fix3
= _mm_add_ps(fix3
,tx
);
842 fiy3
= _mm_add_ps(fiy3
,ty
);
843 fiz3
= _mm_add_ps(fiz3
,tz
);
845 fjx0
= _mm_add_ps(fjx0
,tx
);
846 fjy0
= _mm_add_ps(fjy0
,ty
);
847 fjz0
= _mm_add_ps(fjz0
,tz
);
851 fjptrA
= f
+j_coord_offsetA
;
852 fjptrB
= f
+j_coord_offsetB
;
853 fjptrC
= f
+j_coord_offsetC
;
854 fjptrD
= f
+j_coord_offsetD
;
856 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA
,fjptrB
,fjptrC
,fjptrD
,fjx0
,fjy0
,fjz0
);
858 /* Inner loop uses 90 flops */
864 /* Get j neighbor index, and coordinate index */
865 jnrlistA
= jjnr
[jidx
];
866 jnrlistB
= jjnr
[jidx
+1];
867 jnrlistC
= jjnr
[jidx
+2];
868 jnrlistD
= jjnr
[jidx
+3];
869 /* Sign of each element will be negative for non-real atoms.
870 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
871 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
873 dummy_mask
= gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i
*)(jjnr
+jidx
)),_mm_setzero_si128()));
874 jnrA
= (jnrlistA
>=0) ? jnrlistA
: 0;
875 jnrB
= (jnrlistB
>=0) ? jnrlistB
: 0;
876 jnrC
= (jnrlistC
>=0) ? jnrlistC
: 0;
877 jnrD
= (jnrlistD
>=0) ? jnrlistD
: 0;
878 j_coord_offsetA
= DIM
*jnrA
;
879 j_coord_offsetB
= DIM
*jnrB
;
880 j_coord_offsetC
= DIM
*jnrC
;
881 j_coord_offsetD
= DIM
*jnrD
;
883 /* load j atom coordinates */
884 gmx_mm_load_1rvec_4ptr_swizzle_ps(x
+j_coord_offsetA
,x
+j_coord_offsetB
,
885 x
+j_coord_offsetC
,x
+j_coord_offsetD
,
888 /* Calculate displacement vector */
889 dx10
= _mm_sub_ps(ix1
,jx0
);
890 dy10
= _mm_sub_ps(iy1
,jy0
);
891 dz10
= _mm_sub_ps(iz1
,jz0
);
892 dx20
= _mm_sub_ps(ix2
,jx0
);
893 dy20
= _mm_sub_ps(iy2
,jy0
);
894 dz20
= _mm_sub_ps(iz2
,jz0
);
895 dx30
= _mm_sub_ps(ix3
,jx0
);
896 dy30
= _mm_sub_ps(iy3
,jy0
);
897 dz30
= _mm_sub_ps(iz3
,jz0
);
899 /* Calculate squared distance and things based on it */
900 rsq10
= gmx_mm_calc_rsq_ps(dx10
,dy10
,dz10
);
901 rsq20
= gmx_mm_calc_rsq_ps(dx20
,dy20
,dz20
);
902 rsq30
= gmx_mm_calc_rsq_ps(dx30
,dy30
,dz30
);
904 rinv10
= sse2_invsqrt_f(rsq10
);
905 rinv20
= sse2_invsqrt_f(rsq20
);
906 rinv30
= sse2_invsqrt_f(rsq30
);
908 rinvsq10
= _mm_mul_ps(rinv10
,rinv10
);
909 rinvsq20
= _mm_mul_ps(rinv20
,rinv20
);
910 rinvsq30
= _mm_mul_ps(rinv30
,rinv30
);
912 /* Load parameters for j particles */
913 jq0
= gmx_mm_load_4real_swizzle_ps(charge
+jnrA
+0,charge
+jnrB
+0,
914 charge
+jnrC
+0,charge
+jnrD
+0);
916 fjx0
= _mm_setzero_ps();
917 fjy0
= _mm_setzero_ps();
918 fjz0
= _mm_setzero_ps();
920 /**************************
921 * CALCULATE INTERACTIONS *
922 **************************/
924 if (gmx_mm_any_lt(rsq10
,rcutoff2
))
927 /* Compute parameters for interactions between i and j atoms */
928 qq10
= _mm_mul_ps(iq1
,jq0
);
930 /* REACTION-FIELD ELECTROSTATICS */
931 felec
= _mm_mul_ps(qq10
,_mm_sub_ps(_mm_mul_ps(rinv10
,rinvsq10
),krf2
));
933 cutoff_mask
= _mm_cmplt_ps(rsq10
,rcutoff2
);
937 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
939 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
941 /* Calculate temporary vectorial force */
942 tx
= _mm_mul_ps(fscal
,dx10
);
943 ty
= _mm_mul_ps(fscal
,dy10
);
944 tz
= _mm_mul_ps(fscal
,dz10
);
946 /* Update vectorial force */
947 fix1
= _mm_add_ps(fix1
,tx
);
948 fiy1
= _mm_add_ps(fiy1
,ty
);
949 fiz1
= _mm_add_ps(fiz1
,tz
);
951 fjx0
= _mm_add_ps(fjx0
,tx
);
952 fjy0
= _mm_add_ps(fjy0
,ty
);
953 fjz0
= _mm_add_ps(fjz0
,tz
);
957 /**************************
958 * CALCULATE INTERACTIONS *
959 **************************/
961 if (gmx_mm_any_lt(rsq20
,rcutoff2
))
964 /* Compute parameters for interactions between i and j atoms */
965 qq20
= _mm_mul_ps(iq2
,jq0
);
967 /* REACTION-FIELD ELECTROSTATICS */
968 felec
= _mm_mul_ps(qq20
,_mm_sub_ps(_mm_mul_ps(rinv20
,rinvsq20
),krf2
));
970 cutoff_mask
= _mm_cmplt_ps(rsq20
,rcutoff2
);
974 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
976 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
978 /* Calculate temporary vectorial force */
979 tx
= _mm_mul_ps(fscal
,dx20
);
980 ty
= _mm_mul_ps(fscal
,dy20
);
981 tz
= _mm_mul_ps(fscal
,dz20
);
983 /* Update vectorial force */
984 fix2
= _mm_add_ps(fix2
,tx
);
985 fiy2
= _mm_add_ps(fiy2
,ty
);
986 fiz2
= _mm_add_ps(fiz2
,tz
);
988 fjx0
= _mm_add_ps(fjx0
,tx
);
989 fjy0
= _mm_add_ps(fjy0
,ty
);
990 fjz0
= _mm_add_ps(fjz0
,tz
);
994 /**************************
995 * CALCULATE INTERACTIONS *
996 **************************/
998 if (gmx_mm_any_lt(rsq30
,rcutoff2
))
1001 /* Compute parameters for interactions between i and j atoms */
1002 qq30
= _mm_mul_ps(iq3
,jq0
);
1004 /* REACTION-FIELD ELECTROSTATICS */
1005 felec
= _mm_mul_ps(qq30
,_mm_sub_ps(_mm_mul_ps(rinv30
,rinvsq30
),krf2
));
1007 cutoff_mask
= _mm_cmplt_ps(rsq30
,rcutoff2
);
1011 fscal
= _mm_and_ps(fscal
,cutoff_mask
);
1013 fscal
= _mm_andnot_ps(dummy_mask
,fscal
);
1015 /* Calculate temporary vectorial force */
1016 tx
= _mm_mul_ps(fscal
,dx30
);
1017 ty
= _mm_mul_ps(fscal
,dy30
);
1018 tz
= _mm_mul_ps(fscal
,dz30
);
1020 /* Update vectorial force */
1021 fix3
= _mm_add_ps(fix3
,tx
);
1022 fiy3
= _mm_add_ps(fiy3
,ty
);
1023 fiz3
= _mm_add_ps(fiz3
,tz
);
1025 fjx0
= _mm_add_ps(fjx0
,tx
);
1026 fjy0
= _mm_add_ps(fjy0
,ty
);
1027 fjz0
= _mm_add_ps(fjz0
,tz
);
1031 fjptrA
= (jnrlistA
>=0) ? f
+j_coord_offsetA
: scratch
;
1032 fjptrB
= (jnrlistB
>=0) ? f
+j_coord_offsetB
: scratch
;
1033 fjptrC
= (jnrlistC
>=0) ? f
+j_coord_offsetC
: scratch
;
1034 fjptrD
= (jnrlistD
>=0) ? f
+j_coord_offsetD
: scratch
;
1036 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA
,fjptrB
,fjptrC
,fjptrD
,fjx0
,fjy0
,fjz0
);
1038 /* Inner loop uses 90 flops */
1041 /* End of innermost loop */
1043 gmx_mm_update_iforce_3atom_swizzle_ps(fix1
,fiy1
,fiz1
,fix2
,fiy2
,fiz2
,fix3
,fiy3
,fiz3
,
1044 f
+i_coord_offset
+DIM
,fshift
+i_shift_offset
);
1046 /* Increment number of inner iterations */
1047 inneriter
+= j_index_end
- j_index_start
;
1049 /* Outer loop uses 18 flops */
1052 /* Increment number of outer iterations */
1055 /* Update outer/inner flops */
1057 inc_nrnb(nrnb
,eNR_NBKERNEL_ELEC_W4_F
,outeriter
*18 + inneriter
*90);