2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
50 #include "kernelutil_x86_avx_128_fma_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_avx_128_fma_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: None
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_avx_128_fma_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
96 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
100 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
102 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
103 real rswitch_scalar,d_scalar;
104 __m128 dummy_mask,cutoff_mask;
105 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
106 __m128 one = _mm_set1_ps(1.0);
107 __m128 two = _mm_set1_ps(2.0);
113 jindex = nlist->jindex;
115 shiftidx = nlist->shift;
117 shiftvec = fr->shift_vec[0];
118 fshift = fr->fshift[0];
119 facel = _mm_set1_ps(fr->epsfac);
120 charge = mdatoms->chargeA;
122 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
123 beta = _mm_set1_ps(fr->ic->ewaldcoeff_q);
124 beta2 = _mm_mul_ps(beta,beta);
125 beta3 = _mm_mul_ps(beta,beta2);
126 ewtab = fr->ic->tabq_coul_FDV0;
127 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
128 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
130 /* Setup water-specific parameters */
131 inr = nlist->iinr[0];
132 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
133 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
134 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
136 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
137 rcutoff_scalar = fr->rcoulomb;
138 rcutoff = _mm_set1_ps(rcutoff_scalar);
139 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
141 rswitch_scalar = fr->rcoulomb_switch;
142 rswitch = _mm_set1_ps(rswitch_scalar);
143 /* Setup switch parameters */
144 d_scalar = rcutoff_scalar-rswitch_scalar;
145 d = _mm_set1_ps(d_scalar);
146 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
147 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
148 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
149 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
150 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
151 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
153 /* Avoid stupid compiler warnings */
154 jnrA = jnrB = jnrC = jnrD = 0;
163 for(iidx=0;iidx<4*DIM;iidx++)
168 /* Start outer loop over neighborlists */
169 for(iidx=0; iidx<nri; iidx++)
171 /* Load shift vector for this list */
172 i_shift_offset = DIM*shiftidx[iidx];
174 /* Load limits for loop over neighbors */
175 j_index_start = jindex[iidx];
176 j_index_end = jindex[iidx+1];
178 /* Get outer coordinate index */
180 i_coord_offset = DIM*inr;
182 /* Load i particle coords and add shift vector */
183 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
184 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
186 fix1 = _mm_setzero_ps();
187 fiy1 = _mm_setzero_ps();
188 fiz1 = _mm_setzero_ps();
189 fix2 = _mm_setzero_ps();
190 fiy2 = _mm_setzero_ps();
191 fiz2 = _mm_setzero_ps();
192 fix3 = _mm_setzero_ps();
193 fiy3 = _mm_setzero_ps();
194 fiz3 = _mm_setzero_ps();
196 /* Reset potential sums */
197 velecsum = _mm_setzero_ps();
199 /* Start inner kernel loop */
200 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
203 /* Get j neighbor index, and coordinate index */
208 j_coord_offsetA = DIM*jnrA;
209 j_coord_offsetB = DIM*jnrB;
210 j_coord_offsetC = DIM*jnrC;
211 j_coord_offsetD = DIM*jnrD;
213 /* load j atom coordinates */
214 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
215 x+j_coord_offsetC,x+j_coord_offsetD,
218 /* Calculate displacement vector */
219 dx10 = _mm_sub_ps(ix1,jx0);
220 dy10 = _mm_sub_ps(iy1,jy0);
221 dz10 = _mm_sub_ps(iz1,jz0);
222 dx20 = _mm_sub_ps(ix2,jx0);
223 dy20 = _mm_sub_ps(iy2,jy0);
224 dz20 = _mm_sub_ps(iz2,jz0);
225 dx30 = _mm_sub_ps(ix3,jx0);
226 dy30 = _mm_sub_ps(iy3,jy0);
227 dz30 = _mm_sub_ps(iz3,jz0);
229 /* Calculate squared distance and things based on it */
230 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
231 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
232 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
234 rinv10 = gmx_mm_invsqrt_ps(rsq10);
235 rinv20 = gmx_mm_invsqrt_ps(rsq20);
236 rinv30 = gmx_mm_invsqrt_ps(rsq30);
238 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
239 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
240 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
242 /* Load parameters for j particles */
243 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
244 charge+jnrC+0,charge+jnrD+0);
246 fjx0 = _mm_setzero_ps();
247 fjy0 = _mm_setzero_ps();
248 fjz0 = _mm_setzero_ps();
250 /**************************
251 * CALCULATE INTERACTIONS *
252 **************************/
254 if (gmx_mm_any_lt(rsq10,rcutoff2))
257 r10 = _mm_mul_ps(rsq10,rinv10);
259 /* Compute parameters for interactions between i and j atoms */
260 qq10 = _mm_mul_ps(iq1,jq0);
262 /* EWALD ELECTROSTATICS */
264 /* Analytical PME correction */
265 zeta2 = _mm_mul_ps(beta2,rsq10);
266 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
267 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
268 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
269 felec = _mm_mul_ps(qq10,felec);
270 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
271 velec = _mm_nmacc_ps(pmecorrV,beta,rinv10);
272 velec = _mm_mul_ps(qq10,velec);
274 d = _mm_sub_ps(r10,rswitch);
275 d = _mm_max_ps(d,_mm_setzero_ps());
276 d2 = _mm_mul_ps(d,d);
277 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
279 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
281 /* Evaluate switch function */
282 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
283 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
284 velec = _mm_mul_ps(velec,sw);
285 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
287 /* Update potential sum for this i atom from the interaction with this j atom. */
288 velec = _mm_and_ps(velec,cutoff_mask);
289 velecsum = _mm_add_ps(velecsum,velec);
293 fscal = _mm_and_ps(fscal,cutoff_mask);
295 /* Update vectorial force */
296 fix1 = _mm_macc_ps(dx10,fscal,fix1);
297 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
298 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
300 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
301 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
302 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
306 /**************************
307 * CALCULATE INTERACTIONS *
308 **************************/
310 if (gmx_mm_any_lt(rsq20,rcutoff2))
313 r20 = _mm_mul_ps(rsq20,rinv20);
315 /* Compute parameters for interactions between i and j atoms */
316 qq20 = _mm_mul_ps(iq2,jq0);
318 /* EWALD ELECTROSTATICS */
320 /* Analytical PME correction */
321 zeta2 = _mm_mul_ps(beta2,rsq20);
322 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
323 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
324 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
325 felec = _mm_mul_ps(qq20,felec);
326 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
327 velec = _mm_nmacc_ps(pmecorrV,beta,rinv20);
328 velec = _mm_mul_ps(qq20,velec);
330 d = _mm_sub_ps(r20,rswitch);
331 d = _mm_max_ps(d,_mm_setzero_ps());
332 d2 = _mm_mul_ps(d,d);
333 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
335 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
337 /* Evaluate switch function */
338 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
339 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
340 velec = _mm_mul_ps(velec,sw);
341 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
343 /* Update potential sum for this i atom from the interaction with this j atom. */
344 velec = _mm_and_ps(velec,cutoff_mask);
345 velecsum = _mm_add_ps(velecsum,velec);
349 fscal = _mm_and_ps(fscal,cutoff_mask);
351 /* Update vectorial force */
352 fix2 = _mm_macc_ps(dx20,fscal,fix2);
353 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
354 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
356 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
357 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
358 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
362 /**************************
363 * CALCULATE INTERACTIONS *
364 **************************/
366 if (gmx_mm_any_lt(rsq30,rcutoff2))
369 r30 = _mm_mul_ps(rsq30,rinv30);
371 /* Compute parameters for interactions between i and j atoms */
372 qq30 = _mm_mul_ps(iq3,jq0);
374 /* EWALD ELECTROSTATICS */
376 /* Analytical PME correction */
377 zeta2 = _mm_mul_ps(beta2,rsq30);
378 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
379 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
380 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
381 felec = _mm_mul_ps(qq30,felec);
382 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
383 velec = _mm_nmacc_ps(pmecorrV,beta,rinv30);
384 velec = _mm_mul_ps(qq30,velec);
386 d = _mm_sub_ps(r30,rswitch);
387 d = _mm_max_ps(d,_mm_setzero_ps());
388 d2 = _mm_mul_ps(d,d);
389 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
391 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
393 /* Evaluate switch function */
394 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
395 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
396 velec = _mm_mul_ps(velec,sw);
397 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
399 /* Update potential sum for this i atom from the interaction with this j atom. */
400 velec = _mm_and_ps(velec,cutoff_mask);
401 velecsum = _mm_add_ps(velecsum,velec);
405 fscal = _mm_and_ps(fscal,cutoff_mask);
407 /* Update vectorial force */
408 fix3 = _mm_macc_ps(dx30,fscal,fix3);
409 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
410 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
412 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
413 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
414 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
418 fjptrA = f+j_coord_offsetA;
419 fjptrB = f+j_coord_offsetB;
420 fjptrC = f+j_coord_offsetC;
421 fjptrD = f+j_coord_offsetD;
423 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
425 /* Inner loop uses 159 flops */
431 /* Get j neighbor index, and coordinate index */
432 jnrlistA = jjnr[jidx];
433 jnrlistB = jjnr[jidx+1];
434 jnrlistC = jjnr[jidx+2];
435 jnrlistD = jjnr[jidx+3];
436 /* Sign of each element will be negative for non-real atoms.
437 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
438 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
440 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
441 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
442 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
443 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
444 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
445 j_coord_offsetA = DIM*jnrA;
446 j_coord_offsetB = DIM*jnrB;
447 j_coord_offsetC = DIM*jnrC;
448 j_coord_offsetD = DIM*jnrD;
450 /* load j atom coordinates */
451 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
452 x+j_coord_offsetC,x+j_coord_offsetD,
455 /* Calculate displacement vector */
456 dx10 = _mm_sub_ps(ix1,jx0);
457 dy10 = _mm_sub_ps(iy1,jy0);
458 dz10 = _mm_sub_ps(iz1,jz0);
459 dx20 = _mm_sub_ps(ix2,jx0);
460 dy20 = _mm_sub_ps(iy2,jy0);
461 dz20 = _mm_sub_ps(iz2,jz0);
462 dx30 = _mm_sub_ps(ix3,jx0);
463 dy30 = _mm_sub_ps(iy3,jy0);
464 dz30 = _mm_sub_ps(iz3,jz0);
466 /* Calculate squared distance and things based on it */
467 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
468 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
469 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
471 rinv10 = gmx_mm_invsqrt_ps(rsq10);
472 rinv20 = gmx_mm_invsqrt_ps(rsq20);
473 rinv30 = gmx_mm_invsqrt_ps(rsq30);
475 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
476 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
477 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
479 /* Load parameters for j particles */
480 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
481 charge+jnrC+0,charge+jnrD+0);
483 fjx0 = _mm_setzero_ps();
484 fjy0 = _mm_setzero_ps();
485 fjz0 = _mm_setzero_ps();
487 /**************************
488 * CALCULATE INTERACTIONS *
489 **************************/
491 if (gmx_mm_any_lt(rsq10,rcutoff2))
494 r10 = _mm_mul_ps(rsq10,rinv10);
495 r10 = _mm_andnot_ps(dummy_mask,r10);
497 /* Compute parameters for interactions between i and j atoms */
498 qq10 = _mm_mul_ps(iq1,jq0);
500 /* EWALD ELECTROSTATICS */
502 /* Analytical PME correction */
503 zeta2 = _mm_mul_ps(beta2,rsq10);
504 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
505 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
506 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
507 felec = _mm_mul_ps(qq10,felec);
508 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
509 velec = _mm_nmacc_ps(pmecorrV,beta,rinv10);
510 velec = _mm_mul_ps(qq10,velec);
512 d = _mm_sub_ps(r10,rswitch);
513 d = _mm_max_ps(d,_mm_setzero_ps());
514 d2 = _mm_mul_ps(d,d);
515 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
517 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
519 /* Evaluate switch function */
520 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
521 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
522 velec = _mm_mul_ps(velec,sw);
523 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
525 /* Update potential sum for this i atom from the interaction with this j atom. */
526 velec = _mm_and_ps(velec,cutoff_mask);
527 velec = _mm_andnot_ps(dummy_mask,velec);
528 velecsum = _mm_add_ps(velecsum,velec);
532 fscal = _mm_and_ps(fscal,cutoff_mask);
534 fscal = _mm_andnot_ps(dummy_mask,fscal);
536 /* Update vectorial force */
537 fix1 = _mm_macc_ps(dx10,fscal,fix1);
538 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
539 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
541 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
542 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
543 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
547 /**************************
548 * CALCULATE INTERACTIONS *
549 **************************/
551 if (gmx_mm_any_lt(rsq20,rcutoff2))
554 r20 = _mm_mul_ps(rsq20,rinv20);
555 r20 = _mm_andnot_ps(dummy_mask,r20);
557 /* Compute parameters for interactions between i and j atoms */
558 qq20 = _mm_mul_ps(iq2,jq0);
560 /* EWALD ELECTROSTATICS */
562 /* Analytical PME correction */
563 zeta2 = _mm_mul_ps(beta2,rsq20);
564 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
565 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
566 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
567 felec = _mm_mul_ps(qq20,felec);
568 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
569 velec = _mm_nmacc_ps(pmecorrV,beta,rinv20);
570 velec = _mm_mul_ps(qq20,velec);
572 d = _mm_sub_ps(r20,rswitch);
573 d = _mm_max_ps(d,_mm_setzero_ps());
574 d2 = _mm_mul_ps(d,d);
575 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
577 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
579 /* Evaluate switch function */
580 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
581 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
582 velec = _mm_mul_ps(velec,sw);
583 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
585 /* Update potential sum for this i atom from the interaction with this j atom. */
586 velec = _mm_and_ps(velec,cutoff_mask);
587 velec = _mm_andnot_ps(dummy_mask,velec);
588 velecsum = _mm_add_ps(velecsum,velec);
592 fscal = _mm_and_ps(fscal,cutoff_mask);
594 fscal = _mm_andnot_ps(dummy_mask,fscal);
596 /* Update vectorial force */
597 fix2 = _mm_macc_ps(dx20,fscal,fix2);
598 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
599 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
601 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
602 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
603 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
607 /**************************
608 * CALCULATE INTERACTIONS *
609 **************************/
611 if (gmx_mm_any_lt(rsq30,rcutoff2))
614 r30 = _mm_mul_ps(rsq30,rinv30);
615 r30 = _mm_andnot_ps(dummy_mask,r30);
617 /* Compute parameters for interactions between i and j atoms */
618 qq30 = _mm_mul_ps(iq3,jq0);
620 /* EWALD ELECTROSTATICS */
622 /* Analytical PME correction */
623 zeta2 = _mm_mul_ps(beta2,rsq30);
624 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
625 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
626 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
627 felec = _mm_mul_ps(qq30,felec);
628 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
629 velec = _mm_nmacc_ps(pmecorrV,beta,rinv30);
630 velec = _mm_mul_ps(qq30,velec);
632 d = _mm_sub_ps(r30,rswitch);
633 d = _mm_max_ps(d,_mm_setzero_ps());
634 d2 = _mm_mul_ps(d,d);
635 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
637 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
639 /* Evaluate switch function */
640 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
641 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
642 velec = _mm_mul_ps(velec,sw);
643 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
645 /* Update potential sum for this i atom from the interaction with this j atom. */
646 velec = _mm_and_ps(velec,cutoff_mask);
647 velec = _mm_andnot_ps(dummy_mask,velec);
648 velecsum = _mm_add_ps(velecsum,velec);
652 fscal = _mm_and_ps(fscal,cutoff_mask);
654 fscal = _mm_andnot_ps(dummy_mask,fscal);
656 /* Update vectorial force */
657 fix3 = _mm_macc_ps(dx30,fscal,fix3);
658 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
659 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
661 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
662 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
663 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
667 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
668 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
669 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
670 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
672 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
674 /* Inner loop uses 162 flops */
677 /* End of innermost loop */
679 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
680 f+i_coord_offset+DIM,fshift+i_shift_offset);
683 /* Update potential energies */
684 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
686 /* Increment number of inner iterations */
687 inneriter += j_index_end - j_index_start;
689 /* Outer loop uses 19 flops */
692 /* Increment number of outer iterations */
695 /* Update outer/inner flops */
697 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*162);
700 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_avx_128_fma_single
701 * Electrostatics interaction: Ewald
702 * VdW interaction: None
703 * Geometry: Water4-Particle
704 * Calculate force/pot: Force
707 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_avx_128_fma_single
708 (t_nblist * gmx_restrict nlist,
709 rvec * gmx_restrict xx,
710 rvec * gmx_restrict ff,
711 t_forcerec * gmx_restrict fr,
712 t_mdatoms * gmx_restrict mdatoms,
713 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
714 t_nrnb * gmx_restrict nrnb)
716 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
717 * just 0 for non-waters.
718 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
719 * jnr indices corresponding to data put in the four positions in the SIMD register.
721 int i_shift_offset,i_coord_offset,outeriter,inneriter;
722 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
723 int jnrA,jnrB,jnrC,jnrD;
724 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
725 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
726 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
728 real *shiftvec,*fshift,*x,*f;
729 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
731 __m128 fscal,rcutoff,rcutoff2,jidxall;
733 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
735 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
737 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
738 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
739 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
740 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
741 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
742 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
743 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
746 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
747 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
749 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
750 real rswitch_scalar,d_scalar;
751 __m128 dummy_mask,cutoff_mask;
752 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
753 __m128 one = _mm_set1_ps(1.0);
754 __m128 two = _mm_set1_ps(2.0);
760 jindex = nlist->jindex;
762 shiftidx = nlist->shift;
764 shiftvec = fr->shift_vec[0];
765 fshift = fr->fshift[0];
766 facel = _mm_set1_ps(fr->epsfac);
767 charge = mdatoms->chargeA;
769 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
770 beta = _mm_set1_ps(fr->ic->ewaldcoeff_q);
771 beta2 = _mm_mul_ps(beta,beta);
772 beta3 = _mm_mul_ps(beta,beta2);
773 ewtab = fr->ic->tabq_coul_FDV0;
774 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
775 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
777 /* Setup water-specific parameters */
778 inr = nlist->iinr[0];
779 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
780 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
781 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
783 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
784 rcutoff_scalar = fr->rcoulomb;
785 rcutoff = _mm_set1_ps(rcutoff_scalar);
786 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
788 rswitch_scalar = fr->rcoulomb_switch;
789 rswitch = _mm_set1_ps(rswitch_scalar);
790 /* Setup switch parameters */
791 d_scalar = rcutoff_scalar-rswitch_scalar;
792 d = _mm_set1_ps(d_scalar);
793 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
794 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
795 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
796 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
797 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
798 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
800 /* Avoid stupid compiler warnings */
801 jnrA = jnrB = jnrC = jnrD = 0;
810 for(iidx=0;iidx<4*DIM;iidx++)
815 /* Start outer loop over neighborlists */
816 for(iidx=0; iidx<nri; iidx++)
818 /* Load shift vector for this list */
819 i_shift_offset = DIM*shiftidx[iidx];
821 /* Load limits for loop over neighbors */
822 j_index_start = jindex[iidx];
823 j_index_end = jindex[iidx+1];
825 /* Get outer coordinate index */
827 i_coord_offset = DIM*inr;
829 /* Load i particle coords and add shift vector */
830 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
831 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
833 fix1 = _mm_setzero_ps();
834 fiy1 = _mm_setzero_ps();
835 fiz1 = _mm_setzero_ps();
836 fix2 = _mm_setzero_ps();
837 fiy2 = _mm_setzero_ps();
838 fiz2 = _mm_setzero_ps();
839 fix3 = _mm_setzero_ps();
840 fiy3 = _mm_setzero_ps();
841 fiz3 = _mm_setzero_ps();
843 /* Start inner kernel loop */
844 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
847 /* Get j neighbor index, and coordinate index */
852 j_coord_offsetA = DIM*jnrA;
853 j_coord_offsetB = DIM*jnrB;
854 j_coord_offsetC = DIM*jnrC;
855 j_coord_offsetD = DIM*jnrD;
857 /* load j atom coordinates */
858 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
859 x+j_coord_offsetC,x+j_coord_offsetD,
862 /* Calculate displacement vector */
863 dx10 = _mm_sub_ps(ix1,jx0);
864 dy10 = _mm_sub_ps(iy1,jy0);
865 dz10 = _mm_sub_ps(iz1,jz0);
866 dx20 = _mm_sub_ps(ix2,jx0);
867 dy20 = _mm_sub_ps(iy2,jy0);
868 dz20 = _mm_sub_ps(iz2,jz0);
869 dx30 = _mm_sub_ps(ix3,jx0);
870 dy30 = _mm_sub_ps(iy3,jy0);
871 dz30 = _mm_sub_ps(iz3,jz0);
873 /* Calculate squared distance and things based on it */
874 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
875 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
876 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
878 rinv10 = gmx_mm_invsqrt_ps(rsq10);
879 rinv20 = gmx_mm_invsqrt_ps(rsq20);
880 rinv30 = gmx_mm_invsqrt_ps(rsq30);
882 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
883 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
884 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
886 /* Load parameters for j particles */
887 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
888 charge+jnrC+0,charge+jnrD+0);
890 fjx0 = _mm_setzero_ps();
891 fjy0 = _mm_setzero_ps();
892 fjz0 = _mm_setzero_ps();
894 /**************************
895 * CALCULATE INTERACTIONS *
896 **************************/
898 if (gmx_mm_any_lt(rsq10,rcutoff2))
901 r10 = _mm_mul_ps(rsq10,rinv10);
903 /* Compute parameters for interactions between i and j atoms */
904 qq10 = _mm_mul_ps(iq1,jq0);
906 /* EWALD ELECTROSTATICS */
908 /* Analytical PME correction */
909 zeta2 = _mm_mul_ps(beta2,rsq10);
910 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
911 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
912 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
913 felec = _mm_mul_ps(qq10,felec);
914 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
915 velec = _mm_nmacc_ps(pmecorrV,beta,rinv10);
916 velec = _mm_mul_ps(qq10,velec);
918 d = _mm_sub_ps(r10,rswitch);
919 d = _mm_max_ps(d,_mm_setzero_ps());
920 d2 = _mm_mul_ps(d,d);
921 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
923 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
925 /* Evaluate switch function */
926 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
927 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
928 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
932 fscal = _mm_and_ps(fscal,cutoff_mask);
934 /* Update vectorial force */
935 fix1 = _mm_macc_ps(dx10,fscal,fix1);
936 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
937 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
939 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
940 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
941 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
945 /**************************
946 * CALCULATE INTERACTIONS *
947 **************************/
949 if (gmx_mm_any_lt(rsq20,rcutoff2))
952 r20 = _mm_mul_ps(rsq20,rinv20);
954 /* Compute parameters for interactions between i and j atoms */
955 qq20 = _mm_mul_ps(iq2,jq0);
957 /* EWALD ELECTROSTATICS */
959 /* Analytical PME correction */
960 zeta2 = _mm_mul_ps(beta2,rsq20);
961 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
962 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
963 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
964 felec = _mm_mul_ps(qq20,felec);
965 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
966 velec = _mm_nmacc_ps(pmecorrV,beta,rinv20);
967 velec = _mm_mul_ps(qq20,velec);
969 d = _mm_sub_ps(r20,rswitch);
970 d = _mm_max_ps(d,_mm_setzero_ps());
971 d2 = _mm_mul_ps(d,d);
972 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
974 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
976 /* Evaluate switch function */
977 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
978 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
979 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
983 fscal = _mm_and_ps(fscal,cutoff_mask);
985 /* Update vectorial force */
986 fix2 = _mm_macc_ps(dx20,fscal,fix2);
987 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
988 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
990 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
991 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
992 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
996 /**************************
997 * CALCULATE INTERACTIONS *
998 **************************/
1000 if (gmx_mm_any_lt(rsq30,rcutoff2))
1003 r30 = _mm_mul_ps(rsq30,rinv30);
1005 /* Compute parameters for interactions between i and j atoms */
1006 qq30 = _mm_mul_ps(iq3,jq0);
1008 /* EWALD ELECTROSTATICS */
1010 /* Analytical PME correction */
1011 zeta2 = _mm_mul_ps(beta2,rsq30);
1012 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
1013 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1014 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1015 felec = _mm_mul_ps(qq30,felec);
1016 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1017 velec = _mm_nmacc_ps(pmecorrV,beta,rinv30);
1018 velec = _mm_mul_ps(qq30,velec);
1020 d = _mm_sub_ps(r30,rswitch);
1021 d = _mm_max_ps(d,_mm_setzero_ps());
1022 d2 = _mm_mul_ps(d,d);
1023 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1025 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1027 /* Evaluate switch function */
1028 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1029 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1030 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1034 fscal = _mm_and_ps(fscal,cutoff_mask);
1036 /* Update vectorial force */
1037 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1038 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1039 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1041 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1042 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1043 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1047 fjptrA = f+j_coord_offsetA;
1048 fjptrB = f+j_coord_offsetB;
1049 fjptrC = f+j_coord_offsetC;
1050 fjptrD = f+j_coord_offsetD;
1052 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1054 /* Inner loop uses 150 flops */
1057 if(jidx<j_index_end)
1060 /* Get j neighbor index, and coordinate index */
1061 jnrlistA = jjnr[jidx];
1062 jnrlistB = jjnr[jidx+1];
1063 jnrlistC = jjnr[jidx+2];
1064 jnrlistD = jjnr[jidx+3];
1065 /* Sign of each element will be negative for non-real atoms.
1066 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1067 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1069 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1070 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1071 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1072 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1073 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1074 j_coord_offsetA = DIM*jnrA;
1075 j_coord_offsetB = DIM*jnrB;
1076 j_coord_offsetC = DIM*jnrC;
1077 j_coord_offsetD = DIM*jnrD;
1079 /* load j atom coordinates */
1080 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1081 x+j_coord_offsetC,x+j_coord_offsetD,
1084 /* Calculate displacement vector */
1085 dx10 = _mm_sub_ps(ix1,jx0);
1086 dy10 = _mm_sub_ps(iy1,jy0);
1087 dz10 = _mm_sub_ps(iz1,jz0);
1088 dx20 = _mm_sub_ps(ix2,jx0);
1089 dy20 = _mm_sub_ps(iy2,jy0);
1090 dz20 = _mm_sub_ps(iz2,jz0);
1091 dx30 = _mm_sub_ps(ix3,jx0);
1092 dy30 = _mm_sub_ps(iy3,jy0);
1093 dz30 = _mm_sub_ps(iz3,jz0);
1095 /* Calculate squared distance and things based on it */
1096 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1097 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1098 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1100 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1101 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1102 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1104 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1105 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1106 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1108 /* Load parameters for j particles */
1109 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1110 charge+jnrC+0,charge+jnrD+0);
1112 fjx0 = _mm_setzero_ps();
1113 fjy0 = _mm_setzero_ps();
1114 fjz0 = _mm_setzero_ps();
1116 /**************************
1117 * CALCULATE INTERACTIONS *
1118 **************************/
1120 if (gmx_mm_any_lt(rsq10,rcutoff2))
1123 r10 = _mm_mul_ps(rsq10,rinv10);
1124 r10 = _mm_andnot_ps(dummy_mask,r10);
1126 /* Compute parameters for interactions between i and j atoms */
1127 qq10 = _mm_mul_ps(iq1,jq0);
1129 /* EWALD ELECTROSTATICS */
1131 /* Analytical PME correction */
1132 zeta2 = _mm_mul_ps(beta2,rsq10);
1133 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
1134 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1135 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1136 felec = _mm_mul_ps(qq10,felec);
1137 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1138 velec = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1139 velec = _mm_mul_ps(qq10,velec);
1141 d = _mm_sub_ps(r10,rswitch);
1142 d = _mm_max_ps(d,_mm_setzero_ps());
1143 d2 = _mm_mul_ps(d,d);
1144 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1146 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1148 /* Evaluate switch function */
1149 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1150 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1151 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1155 fscal = _mm_and_ps(fscal,cutoff_mask);
1157 fscal = _mm_andnot_ps(dummy_mask,fscal);
1159 /* Update vectorial force */
1160 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1161 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1162 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1164 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1165 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1166 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1170 /**************************
1171 * CALCULATE INTERACTIONS *
1172 **************************/
1174 if (gmx_mm_any_lt(rsq20,rcutoff2))
1177 r20 = _mm_mul_ps(rsq20,rinv20);
1178 r20 = _mm_andnot_ps(dummy_mask,r20);
1180 /* Compute parameters for interactions between i and j atoms */
1181 qq20 = _mm_mul_ps(iq2,jq0);
1183 /* EWALD ELECTROSTATICS */
1185 /* Analytical PME correction */
1186 zeta2 = _mm_mul_ps(beta2,rsq20);
1187 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
1188 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1189 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1190 felec = _mm_mul_ps(qq20,felec);
1191 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1192 velec = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1193 velec = _mm_mul_ps(qq20,velec);
1195 d = _mm_sub_ps(r20,rswitch);
1196 d = _mm_max_ps(d,_mm_setzero_ps());
1197 d2 = _mm_mul_ps(d,d);
1198 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1200 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1202 /* Evaluate switch function */
1203 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1204 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1205 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1209 fscal = _mm_and_ps(fscal,cutoff_mask);
1211 fscal = _mm_andnot_ps(dummy_mask,fscal);
1213 /* Update vectorial force */
1214 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1215 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1216 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1218 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1219 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1220 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1224 /**************************
1225 * CALCULATE INTERACTIONS *
1226 **************************/
1228 if (gmx_mm_any_lt(rsq30,rcutoff2))
1231 r30 = _mm_mul_ps(rsq30,rinv30);
1232 r30 = _mm_andnot_ps(dummy_mask,r30);
1234 /* Compute parameters for interactions between i and j atoms */
1235 qq30 = _mm_mul_ps(iq3,jq0);
1237 /* EWALD ELECTROSTATICS */
1239 /* Analytical PME correction */
1240 zeta2 = _mm_mul_ps(beta2,rsq30);
1241 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
1242 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1243 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1244 felec = _mm_mul_ps(qq30,felec);
1245 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1246 velec = _mm_nmacc_ps(pmecorrV,beta,rinv30);
1247 velec = _mm_mul_ps(qq30,velec);
1249 d = _mm_sub_ps(r30,rswitch);
1250 d = _mm_max_ps(d,_mm_setzero_ps());
1251 d2 = _mm_mul_ps(d,d);
1252 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1254 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1256 /* Evaluate switch function */
1257 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1258 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1259 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1263 fscal = _mm_and_ps(fscal,cutoff_mask);
1265 fscal = _mm_andnot_ps(dummy_mask,fscal);
1267 /* Update vectorial force */
1268 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1269 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1270 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1272 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1273 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1274 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1278 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1279 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1280 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1281 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1283 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1285 /* Inner loop uses 153 flops */
1288 /* End of innermost loop */
1290 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1291 f+i_coord_offset+DIM,fshift+i_shift_offset);
1293 /* Increment number of inner iterations */
1294 inneriter += j_index_end - j_index_start;
1296 /* Outer loop uses 18 flops */
1299 /* Increment number of outer iterations */
1302 /* Update outer/inner flops */
1304 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*153);