2 * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse4_1_single.h"
34 #include "kernelutil_x86_sse4_1_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_single
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
72 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
76 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
77 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
78 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
79 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
80 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
81 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
82 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
83 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
86 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
89 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
90 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
91 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
92 real rswitch_scalar,d_scalar;
93 __m128 dummy_mask,cutoff_mask;
94 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
95 __m128 one = _mm_set1_ps(1.0);
96 __m128 two = _mm_set1_ps(2.0);
102 jindex = nlist->jindex;
104 shiftidx = nlist->shift;
106 shiftvec = fr->shift_vec[0];
107 fshift = fr->fshift[0];
108 facel = _mm_set1_ps(fr->epsfac);
109 charge = mdatoms->chargeA;
110 krf = _mm_set1_ps(fr->ic->k_rf);
111 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
112 crf = _mm_set1_ps(fr->ic->c_rf);
113 nvdwtype = fr->ntype;
115 vdwtype = mdatoms->typeA;
117 /* Setup water-specific parameters */
118 inr = nlist->iinr[0];
119 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
120 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
121 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
122 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
124 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
125 rcutoff_scalar = fr->rcoulomb;
126 rcutoff = _mm_set1_ps(rcutoff_scalar);
127 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
129 rswitch_scalar = fr->rvdw_switch;
130 rswitch = _mm_set1_ps(rswitch_scalar);
131 /* Setup switch parameters */
132 d_scalar = rcutoff_scalar-rswitch_scalar;
133 d = _mm_set1_ps(d_scalar);
134 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
135 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
136 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
137 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
138 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
139 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
141 /* Avoid stupid compiler warnings */
142 jnrA = jnrB = jnrC = jnrD = 0;
151 for(iidx=0;iidx<4*DIM;iidx++)
156 /* Start outer loop over neighborlists */
157 for(iidx=0; iidx<nri; iidx++)
159 /* Load shift vector for this list */
160 i_shift_offset = DIM*shiftidx[iidx];
162 /* Load limits for loop over neighbors */
163 j_index_start = jindex[iidx];
164 j_index_end = jindex[iidx+1];
166 /* Get outer coordinate index */
168 i_coord_offset = DIM*inr;
170 /* Load i particle coords and add shift vector */
171 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
172 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
174 fix0 = _mm_setzero_ps();
175 fiy0 = _mm_setzero_ps();
176 fiz0 = _mm_setzero_ps();
177 fix1 = _mm_setzero_ps();
178 fiy1 = _mm_setzero_ps();
179 fiz1 = _mm_setzero_ps();
180 fix2 = _mm_setzero_ps();
181 fiy2 = _mm_setzero_ps();
182 fiz2 = _mm_setzero_ps();
183 fix3 = _mm_setzero_ps();
184 fiy3 = _mm_setzero_ps();
185 fiz3 = _mm_setzero_ps();
187 /* Reset potential sums */
188 velecsum = _mm_setzero_ps();
189 vvdwsum = _mm_setzero_ps();
191 /* Start inner kernel loop */
192 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
195 /* Get j neighbor index, and coordinate index */
200 j_coord_offsetA = DIM*jnrA;
201 j_coord_offsetB = DIM*jnrB;
202 j_coord_offsetC = DIM*jnrC;
203 j_coord_offsetD = DIM*jnrD;
205 /* load j atom coordinates */
206 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
207 x+j_coord_offsetC,x+j_coord_offsetD,
210 /* Calculate displacement vector */
211 dx00 = _mm_sub_ps(ix0,jx0);
212 dy00 = _mm_sub_ps(iy0,jy0);
213 dz00 = _mm_sub_ps(iz0,jz0);
214 dx10 = _mm_sub_ps(ix1,jx0);
215 dy10 = _mm_sub_ps(iy1,jy0);
216 dz10 = _mm_sub_ps(iz1,jz0);
217 dx20 = _mm_sub_ps(ix2,jx0);
218 dy20 = _mm_sub_ps(iy2,jy0);
219 dz20 = _mm_sub_ps(iz2,jz0);
220 dx30 = _mm_sub_ps(ix3,jx0);
221 dy30 = _mm_sub_ps(iy3,jy0);
222 dz30 = _mm_sub_ps(iz3,jz0);
224 /* Calculate squared distance and things based on it */
225 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
226 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
227 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
228 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
230 rinv00 = gmx_mm_invsqrt_ps(rsq00);
231 rinv10 = gmx_mm_invsqrt_ps(rsq10);
232 rinv20 = gmx_mm_invsqrt_ps(rsq20);
233 rinv30 = gmx_mm_invsqrt_ps(rsq30);
235 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
236 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
237 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
238 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
240 /* Load parameters for j particles */
241 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
242 charge+jnrC+0,charge+jnrD+0);
243 vdwjidx0A = 2*vdwtype[jnrA+0];
244 vdwjidx0B = 2*vdwtype[jnrB+0];
245 vdwjidx0C = 2*vdwtype[jnrC+0];
246 vdwjidx0D = 2*vdwtype[jnrD+0];
248 /**************************
249 * CALCULATE INTERACTIONS *
250 **************************/
252 if (gmx_mm_any_lt(rsq00,rcutoff2))
255 r00 = _mm_mul_ps(rsq00,rinv00);
257 /* Compute parameters for interactions between i and j atoms */
258 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
259 vdwparam+vdwioffset0+vdwjidx0B,
260 vdwparam+vdwioffset0+vdwjidx0C,
261 vdwparam+vdwioffset0+vdwjidx0D,
264 /* LENNARD-JONES DISPERSION/REPULSION */
266 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
267 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
268 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
269 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
270 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
272 d = _mm_sub_ps(r00,rswitch);
273 d = _mm_max_ps(d,_mm_setzero_ps());
274 d2 = _mm_mul_ps(d,d);
275 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
277 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
279 /* Evaluate switch function */
280 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
281 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
282 vvdw = _mm_mul_ps(vvdw,sw);
283 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
285 /* Update potential sum for this i atom from the interaction with this j atom. */
286 vvdw = _mm_and_ps(vvdw,cutoff_mask);
287 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
291 fscal = _mm_and_ps(fscal,cutoff_mask);
293 /* Calculate temporary vectorial force */
294 tx = _mm_mul_ps(fscal,dx00);
295 ty = _mm_mul_ps(fscal,dy00);
296 tz = _mm_mul_ps(fscal,dz00);
298 /* Update vectorial force */
299 fix0 = _mm_add_ps(fix0,tx);
300 fiy0 = _mm_add_ps(fiy0,ty);
301 fiz0 = _mm_add_ps(fiz0,tz);
303 fjptrA = f+j_coord_offsetA;
304 fjptrB = f+j_coord_offsetB;
305 fjptrC = f+j_coord_offsetC;
306 fjptrD = f+j_coord_offsetD;
307 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
311 /**************************
312 * CALCULATE INTERACTIONS *
313 **************************/
315 if (gmx_mm_any_lt(rsq10,rcutoff2))
318 /* Compute parameters for interactions between i and j atoms */
319 qq10 = _mm_mul_ps(iq1,jq0);
321 /* REACTION-FIELD ELECTROSTATICS */
322 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
323 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
325 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
327 /* Update potential sum for this i atom from the interaction with this j atom. */
328 velec = _mm_and_ps(velec,cutoff_mask);
329 velecsum = _mm_add_ps(velecsum,velec);
333 fscal = _mm_and_ps(fscal,cutoff_mask);
335 /* Calculate temporary vectorial force */
336 tx = _mm_mul_ps(fscal,dx10);
337 ty = _mm_mul_ps(fscal,dy10);
338 tz = _mm_mul_ps(fscal,dz10);
340 /* Update vectorial force */
341 fix1 = _mm_add_ps(fix1,tx);
342 fiy1 = _mm_add_ps(fiy1,ty);
343 fiz1 = _mm_add_ps(fiz1,tz);
345 fjptrA = f+j_coord_offsetA;
346 fjptrB = f+j_coord_offsetB;
347 fjptrC = f+j_coord_offsetC;
348 fjptrD = f+j_coord_offsetD;
349 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
353 /**************************
354 * CALCULATE INTERACTIONS *
355 **************************/
357 if (gmx_mm_any_lt(rsq20,rcutoff2))
360 /* Compute parameters for interactions between i and j atoms */
361 qq20 = _mm_mul_ps(iq2,jq0);
363 /* REACTION-FIELD ELECTROSTATICS */
364 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
365 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
367 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
369 /* Update potential sum for this i atom from the interaction with this j atom. */
370 velec = _mm_and_ps(velec,cutoff_mask);
371 velecsum = _mm_add_ps(velecsum,velec);
375 fscal = _mm_and_ps(fscal,cutoff_mask);
377 /* Calculate temporary vectorial force */
378 tx = _mm_mul_ps(fscal,dx20);
379 ty = _mm_mul_ps(fscal,dy20);
380 tz = _mm_mul_ps(fscal,dz20);
382 /* Update vectorial force */
383 fix2 = _mm_add_ps(fix2,tx);
384 fiy2 = _mm_add_ps(fiy2,ty);
385 fiz2 = _mm_add_ps(fiz2,tz);
387 fjptrA = f+j_coord_offsetA;
388 fjptrB = f+j_coord_offsetB;
389 fjptrC = f+j_coord_offsetC;
390 fjptrD = f+j_coord_offsetD;
391 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
395 /**************************
396 * CALCULATE INTERACTIONS *
397 **************************/
399 if (gmx_mm_any_lt(rsq30,rcutoff2))
402 /* Compute parameters for interactions between i and j atoms */
403 qq30 = _mm_mul_ps(iq3,jq0);
405 /* REACTION-FIELD ELECTROSTATICS */
406 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
407 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
409 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
411 /* Update potential sum for this i atom from the interaction with this j atom. */
412 velec = _mm_and_ps(velec,cutoff_mask);
413 velecsum = _mm_add_ps(velecsum,velec);
417 fscal = _mm_and_ps(fscal,cutoff_mask);
419 /* Calculate temporary vectorial force */
420 tx = _mm_mul_ps(fscal,dx30);
421 ty = _mm_mul_ps(fscal,dy30);
422 tz = _mm_mul_ps(fscal,dz30);
424 /* Update vectorial force */
425 fix3 = _mm_add_ps(fix3,tx);
426 fiy3 = _mm_add_ps(fiy3,ty);
427 fiz3 = _mm_add_ps(fiz3,tz);
429 fjptrA = f+j_coord_offsetA;
430 fjptrB = f+j_coord_offsetB;
431 fjptrC = f+j_coord_offsetC;
432 fjptrD = f+j_coord_offsetD;
433 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
437 /* Inner loop uses 167 flops */
443 /* Get j neighbor index, and coordinate index */
444 jnrlistA = jjnr[jidx];
445 jnrlistB = jjnr[jidx+1];
446 jnrlistC = jjnr[jidx+2];
447 jnrlistD = jjnr[jidx+3];
448 /* Sign of each element will be negative for non-real atoms.
449 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
450 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
452 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
453 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
454 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
455 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
456 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
457 j_coord_offsetA = DIM*jnrA;
458 j_coord_offsetB = DIM*jnrB;
459 j_coord_offsetC = DIM*jnrC;
460 j_coord_offsetD = DIM*jnrD;
462 /* load j atom coordinates */
463 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
464 x+j_coord_offsetC,x+j_coord_offsetD,
467 /* Calculate displacement vector */
468 dx00 = _mm_sub_ps(ix0,jx0);
469 dy00 = _mm_sub_ps(iy0,jy0);
470 dz00 = _mm_sub_ps(iz0,jz0);
471 dx10 = _mm_sub_ps(ix1,jx0);
472 dy10 = _mm_sub_ps(iy1,jy0);
473 dz10 = _mm_sub_ps(iz1,jz0);
474 dx20 = _mm_sub_ps(ix2,jx0);
475 dy20 = _mm_sub_ps(iy2,jy0);
476 dz20 = _mm_sub_ps(iz2,jz0);
477 dx30 = _mm_sub_ps(ix3,jx0);
478 dy30 = _mm_sub_ps(iy3,jy0);
479 dz30 = _mm_sub_ps(iz3,jz0);
481 /* Calculate squared distance and things based on it */
482 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
483 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
484 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
485 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
487 rinv00 = gmx_mm_invsqrt_ps(rsq00);
488 rinv10 = gmx_mm_invsqrt_ps(rsq10);
489 rinv20 = gmx_mm_invsqrt_ps(rsq20);
490 rinv30 = gmx_mm_invsqrt_ps(rsq30);
492 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
493 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
494 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
495 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
497 /* Load parameters for j particles */
498 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
499 charge+jnrC+0,charge+jnrD+0);
500 vdwjidx0A = 2*vdwtype[jnrA+0];
501 vdwjidx0B = 2*vdwtype[jnrB+0];
502 vdwjidx0C = 2*vdwtype[jnrC+0];
503 vdwjidx0D = 2*vdwtype[jnrD+0];
505 /**************************
506 * CALCULATE INTERACTIONS *
507 **************************/
509 if (gmx_mm_any_lt(rsq00,rcutoff2))
512 r00 = _mm_mul_ps(rsq00,rinv00);
513 r00 = _mm_andnot_ps(dummy_mask,r00);
515 /* Compute parameters for interactions between i and j atoms */
516 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
517 vdwparam+vdwioffset0+vdwjidx0B,
518 vdwparam+vdwioffset0+vdwjidx0C,
519 vdwparam+vdwioffset0+vdwjidx0D,
522 /* LENNARD-JONES DISPERSION/REPULSION */
524 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
525 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
526 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
527 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
528 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
530 d = _mm_sub_ps(r00,rswitch);
531 d = _mm_max_ps(d,_mm_setzero_ps());
532 d2 = _mm_mul_ps(d,d);
533 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
535 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
537 /* Evaluate switch function */
538 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
539 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
540 vvdw = _mm_mul_ps(vvdw,sw);
541 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
543 /* Update potential sum for this i atom from the interaction with this j atom. */
544 vvdw = _mm_and_ps(vvdw,cutoff_mask);
545 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
546 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
550 fscal = _mm_and_ps(fscal,cutoff_mask);
552 fscal = _mm_andnot_ps(dummy_mask,fscal);
554 /* Calculate temporary vectorial force */
555 tx = _mm_mul_ps(fscal,dx00);
556 ty = _mm_mul_ps(fscal,dy00);
557 tz = _mm_mul_ps(fscal,dz00);
559 /* Update vectorial force */
560 fix0 = _mm_add_ps(fix0,tx);
561 fiy0 = _mm_add_ps(fiy0,ty);
562 fiz0 = _mm_add_ps(fiz0,tz);
564 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
565 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
566 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
567 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
568 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
572 /**************************
573 * CALCULATE INTERACTIONS *
574 **************************/
576 if (gmx_mm_any_lt(rsq10,rcutoff2))
579 /* Compute parameters for interactions between i and j atoms */
580 qq10 = _mm_mul_ps(iq1,jq0);
582 /* REACTION-FIELD ELECTROSTATICS */
583 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
584 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
586 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
588 /* Update potential sum for this i atom from the interaction with this j atom. */
589 velec = _mm_and_ps(velec,cutoff_mask);
590 velec = _mm_andnot_ps(dummy_mask,velec);
591 velecsum = _mm_add_ps(velecsum,velec);
595 fscal = _mm_and_ps(fscal,cutoff_mask);
597 fscal = _mm_andnot_ps(dummy_mask,fscal);
599 /* Calculate temporary vectorial force */
600 tx = _mm_mul_ps(fscal,dx10);
601 ty = _mm_mul_ps(fscal,dy10);
602 tz = _mm_mul_ps(fscal,dz10);
604 /* Update vectorial force */
605 fix1 = _mm_add_ps(fix1,tx);
606 fiy1 = _mm_add_ps(fiy1,ty);
607 fiz1 = _mm_add_ps(fiz1,tz);
609 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
610 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
611 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
612 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
613 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
617 /**************************
618 * CALCULATE INTERACTIONS *
619 **************************/
621 if (gmx_mm_any_lt(rsq20,rcutoff2))
624 /* Compute parameters for interactions between i and j atoms */
625 qq20 = _mm_mul_ps(iq2,jq0);
627 /* REACTION-FIELD ELECTROSTATICS */
628 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
629 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
631 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
633 /* Update potential sum for this i atom from the interaction with this j atom. */
634 velec = _mm_and_ps(velec,cutoff_mask);
635 velec = _mm_andnot_ps(dummy_mask,velec);
636 velecsum = _mm_add_ps(velecsum,velec);
640 fscal = _mm_and_ps(fscal,cutoff_mask);
642 fscal = _mm_andnot_ps(dummy_mask,fscal);
644 /* Calculate temporary vectorial force */
645 tx = _mm_mul_ps(fscal,dx20);
646 ty = _mm_mul_ps(fscal,dy20);
647 tz = _mm_mul_ps(fscal,dz20);
649 /* Update vectorial force */
650 fix2 = _mm_add_ps(fix2,tx);
651 fiy2 = _mm_add_ps(fiy2,ty);
652 fiz2 = _mm_add_ps(fiz2,tz);
654 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
655 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
656 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
657 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
658 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
662 /**************************
663 * CALCULATE INTERACTIONS *
664 **************************/
666 if (gmx_mm_any_lt(rsq30,rcutoff2))
669 /* Compute parameters for interactions between i and j atoms */
670 qq30 = _mm_mul_ps(iq3,jq0);
672 /* REACTION-FIELD ELECTROSTATICS */
673 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
674 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
676 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
678 /* Update potential sum for this i atom from the interaction with this j atom. */
679 velec = _mm_and_ps(velec,cutoff_mask);
680 velec = _mm_andnot_ps(dummy_mask,velec);
681 velecsum = _mm_add_ps(velecsum,velec);
685 fscal = _mm_and_ps(fscal,cutoff_mask);
687 fscal = _mm_andnot_ps(dummy_mask,fscal);
689 /* Calculate temporary vectorial force */
690 tx = _mm_mul_ps(fscal,dx30);
691 ty = _mm_mul_ps(fscal,dy30);
692 tz = _mm_mul_ps(fscal,dz30);
694 /* Update vectorial force */
695 fix3 = _mm_add_ps(fix3,tx);
696 fiy3 = _mm_add_ps(fiy3,ty);
697 fiz3 = _mm_add_ps(fiz3,tz);
699 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
700 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
701 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
702 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
703 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
707 /* Inner loop uses 168 flops */
710 /* End of innermost loop */
712 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
713 f+i_coord_offset,fshift+i_shift_offset);
716 /* Update potential energies */
717 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
718 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
720 /* Increment number of inner iterations */
721 inneriter += j_index_end - j_index_start;
723 /* Outer loop uses 26 flops */
726 /* Increment number of outer iterations */
729 /* Update outer/inner flops */
731 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
734 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_single
735 * Electrostatics interaction: ReactionField
736 * VdW interaction: LennardJones
737 * Geometry: Water4-Particle
738 * Calculate force/pot: Force
741 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_single
742 (t_nblist * gmx_restrict nlist,
743 rvec * gmx_restrict xx,
744 rvec * gmx_restrict ff,
745 t_forcerec * gmx_restrict fr,
746 t_mdatoms * gmx_restrict mdatoms,
747 nb_kernel_data_t * gmx_restrict kernel_data,
748 t_nrnb * gmx_restrict nrnb)
750 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
751 * just 0 for non-waters.
752 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
753 * jnr indices corresponding to data put in the four positions in the SIMD register.
755 int i_shift_offset,i_coord_offset,outeriter,inneriter;
756 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
757 int jnrA,jnrB,jnrC,jnrD;
758 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
759 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
760 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
762 real *shiftvec,*fshift,*x,*f;
763 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
765 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
767 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
769 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
771 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
773 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
774 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
775 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
776 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
777 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
778 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
779 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
780 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
783 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
786 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
787 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
788 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
789 real rswitch_scalar,d_scalar;
790 __m128 dummy_mask,cutoff_mask;
791 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
792 __m128 one = _mm_set1_ps(1.0);
793 __m128 two = _mm_set1_ps(2.0);
799 jindex = nlist->jindex;
801 shiftidx = nlist->shift;
803 shiftvec = fr->shift_vec[0];
804 fshift = fr->fshift[0];
805 facel = _mm_set1_ps(fr->epsfac);
806 charge = mdatoms->chargeA;
807 krf = _mm_set1_ps(fr->ic->k_rf);
808 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
809 crf = _mm_set1_ps(fr->ic->c_rf);
810 nvdwtype = fr->ntype;
812 vdwtype = mdatoms->typeA;
814 /* Setup water-specific parameters */
815 inr = nlist->iinr[0];
816 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
817 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
818 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
819 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
821 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
822 rcutoff_scalar = fr->rcoulomb;
823 rcutoff = _mm_set1_ps(rcutoff_scalar);
824 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
826 rswitch_scalar = fr->rvdw_switch;
827 rswitch = _mm_set1_ps(rswitch_scalar);
828 /* Setup switch parameters */
829 d_scalar = rcutoff_scalar-rswitch_scalar;
830 d = _mm_set1_ps(d_scalar);
831 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
832 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
833 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
834 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
835 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
836 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
838 /* Avoid stupid compiler warnings */
839 jnrA = jnrB = jnrC = jnrD = 0;
848 for(iidx=0;iidx<4*DIM;iidx++)
853 /* Start outer loop over neighborlists */
854 for(iidx=0; iidx<nri; iidx++)
856 /* Load shift vector for this list */
857 i_shift_offset = DIM*shiftidx[iidx];
859 /* Load limits for loop over neighbors */
860 j_index_start = jindex[iidx];
861 j_index_end = jindex[iidx+1];
863 /* Get outer coordinate index */
865 i_coord_offset = DIM*inr;
867 /* Load i particle coords and add shift vector */
868 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
869 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
871 fix0 = _mm_setzero_ps();
872 fiy0 = _mm_setzero_ps();
873 fiz0 = _mm_setzero_ps();
874 fix1 = _mm_setzero_ps();
875 fiy1 = _mm_setzero_ps();
876 fiz1 = _mm_setzero_ps();
877 fix2 = _mm_setzero_ps();
878 fiy2 = _mm_setzero_ps();
879 fiz2 = _mm_setzero_ps();
880 fix3 = _mm_setzero_ps();
881 fiy3 = _mm_setzero_ps();
882 fiz3 = _mm_setzero_ps();
884 /* Start inner kernel loop */
885 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
888 /* Get j neighbor index, and coordinate index */
893 j_coord_offsetA = DIM*jnrA;
894 j_coord_offsetB = DIM*jnrB;
895 j_coord_offsetC = DIM*jnrC;
896 j_coord_offsetD = DIM*jnrD;
898 /* load j atom coordinates */
899 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
900 x+j_coord_offsetC,x+j_coord_offsetD,
903 /* Calculate displacement vector */
904 dx00 = _mm_sub_ps(ix0,jx0);
905 dy00 = _mm_sub_ps(iy0,jy0);
906 dz00 = _mm_sub_ps(iz0,jz0);
907 dx10 = _mm_sub_ps(ix1,jx0);
908 dy10 = _mm_sub_ps(iy1,jy0);
909 dz10 = _mm_sub_ps(iz1,jz0);
910 dx20 = _mm_sub_ps(ix2,jx0);
911 dy20 = _mm_sub_ps(iy2,jy0);
912 dz20 = _mm_sub_ps(iz2,jz0);
913 dx30 = _mm_sub_ps(ix3,jx0);
914 dy30 = _mm_sub_ps(iy3,jy0);
915 dz30 = _mm_sub_ps(iz3,jz0);
917 /* Calculate squared distance and things based on it */
918 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
919 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
920 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
921 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
923 rinv00 = gmx_mm_invsqrt_ps(rsq00);
924 rinv10 = gmx_mm_invsqrt_ps(rsq10);
925 rinv20 = gmx_mm_invsqrt_ps(rsq20);
926 rinv30 = gmx_mm_invsqrt_ps(rsq30);
928 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
929 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
930 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
931 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
933 /* Load parameters for j particles */
934 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
935 charge+jnrC+0,charge+jnrD+0);
936 vdwjidx0A = 2*vdwtype[jnrA+0];
937 vdwjidx0B = 2*vdwtype[jnrB+0];
938 vdwjidx0C = 2*vdwtype[jnrC+0];
939 vdwjidx0D = 2*vdwtype[jnrD+0];
941 /**************************
942 * CALCULATE INTERACTIONS *
943 **************************/
945 if (gmx_mm_any_lt(rsq00,rcutoff2))
948 r00 = _mm_mul_ps(rsq00,rinv00);
950 /* Compute parameters for interactions between i and j atoms */
951 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
952 vdwparam+vdwioffset0+vdwjidx0B,
953 vdwparam+vdwioffset0+vdwjidx0C,
954 vdwparam+vdwioffset0+vdwjidx0D,
957 /* LENNARD-JONES DISPERSION/REPULSION */
959 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
960 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
961 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
962 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
963 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
965 d = _mm_sub_ps(r00,rswitch);
966 d = _mm_max_ps(d,_mm_setzero_ps());
967 d2 = _mm_mul_ps(d,d);
968 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
970 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
972 /* Evaluate switch function */
973 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
974 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
975 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
979 fscal = _mm_and_ps(fscal,cutoff_mask);
981 /* Calculate temporary vectorial force */
982 tx = _mm_mul_ps(fscal,dx00);
983 ty = _mm_mul_ps(fscal,dy00);
984 tz = _mm_mul_ps(fscal,dz00);
986 /* Update vectorial force */
987 fix0 = _mm_add_ps(fix0,tx);
988 fiy0 = _mm_add_ps(fiy0,ty);
989 fiz0 = _mm_add_ps(fiz0,tz);
991 fjptrA = f+j_coord_offsetA;
992 fjptrB = f+j_coord_offsetB;
993 fjptrC = f+j_coord_offsetC;
994 fjptrD = f+j_coord_offsetD;
995 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
999 /**************************
1000 * CALCULATE INTERACTIONS *
1001 **************************/
1003 if (gmx_mm_any_lt(rsq10,rcutoff2))
1006 /* Compute parameters for interactions between i and j atoms */
1007 qq10 = _mm_mul_ps(iq1,jq0);
1009 /* REACTION-FIELD ELECTROSTATICS */
1010 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1012 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1016 fscal = _mm_and_ps(fscal,cutoff_mask);
1018 /* Calculate temporary vectorial force */
1019 tx = _mm_mul_ps(fscal,dx10);
1020 ty = _mm_mul_ps(fscal,dy10);
1021 tz = _mm_mul_ps(fscal,dz10);
1023 /* Update vectorial force */
1024 fix1 = _mm_add_ps(fix1,tx);
1025 fiy1 = _mm_add_ps(fiy1,ty);
1026 fiz1 = _mm_add_ps(fiz1,tz);
1028 fjptrA = f+j_coord_offsetA;
1029 fjptrB = f+j_coord_offsetB;
1030 fjptrC = f+j_coord_offsetC;
1031 fjptrD = f+j_coord_offsetD;
1032 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1036 /**************************
1037 * CALCULATE INTERACTIONS *
1038 **************************/
1040 if (gmx_mm_any_lt(rsq20,rcutoff2))
1043 /* Compute parameters for interactions between i and j atoms */
1044 qq20 = _mm_mul_ps(iq2,jq0);
1046 /* REACTION-FIELD ELECTROSTATICS */
1047 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1049 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1053 fscal = _mm_and_ps(fscal,cutoff_mask);
1055 /* Calculate temporary vectorial force */
1056 tx = _mm_mul_ps(fscal,dx20);
1057 ty = _mm_mul_ps(fscal,dy20);
1058 tz = _mm_mul_ps(fscal,dz20);
1060 /* Update vectorial force */
1061 fix2 = _mm_add_ps(fix2,tx);
1062 fiy2 = _mm_add_ps(fiy2,ty);
1063 fiz2 = _mm_add_ps(fiz2,tz);
1065 fjptrA = f+j_coord_offsetA;
1066 fjptrB = f+j_coord_offsetB;
1067 fjptrC = f+j_coord_offsetC;
1068 fjptrD = f+j_coord_offsetD;
1069 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1073 /**************************
1074 * CALCULATE INTERACTIONS *
1075 **************************/
1077 if (gmx_mm_any_lt(rsq30,rcutoff2))
1080 /* Compute parameters for interactions between i and j atoms */
1081 qq30 = _mm_mul_ps(iq3,jq0);
1083 /* REACTION-FIELD ELECTROSTATICS */
1084 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1086 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1090 fscal = _mm_and_ps(fscal,cutoff_mask);
1092 /* Calculate temporary vectorial force */
1093 tx = _mm_mul_ps(fscal,dx30);
1094 ty = _mm_mul_ps(fscal,dy30);
1095 tz = _mm_mul_ps(fscal,dz30);
1097 /* Update vectorial force */
1098 fix3 = _mm_add_ps(fix3,tx);
1099 fiy3 = _mm_add_ps(fiy3,ty);
1100 fiz3 = _mm_add_ps(fiz3,tz);
1102 fjptrA = f+j_coord_offsetA;
1103 fjptrB = f+j_coord_offsetB;
1104 fjptrC = f+j_coord_offsetC;
1105 fjptrD = f+j_coord_offsetD;
1106 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1110 /* Inner loop uses 146 flops */
1113 if(jidx<j_index_end)
1116 /* Get j neighbor index, and coordinate index */
1117 jnrlistA = jjnr[jidx];
1118 jnrlistB = jjnr[jidx+1];
1119 jnrlistC = jjnr[jidx+2];
1120 jnrlistD = jjnr[jidx+3];
1121 /* Sign of each element will be negative for non-real atoms.
1122 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1123 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1125 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1126 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1127 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1128 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1129 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1130 j_coord_offsetA = DIM*jnrA;
1131 j_coord_offsetB = DIM*jnrB;
1132 j_coord_offsetC = DIM*jnrC;
1133 j_coord_offsetD = DIM*jnrD;
1135 /* load j atom coordinates */
1136 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1137 x+j_coord_offsetC,x+j_coord_offsetD,
1140 /* Calculate displacement vector */
1141 dx00 = _mm_sub_ps(ix0,jx0);
1142 dy00 = _mm_sub_ps(iy0,jy0);
1143 dz00 = _mm_sub_ps(iz0,jz0);
1144 dx10 = _mm_sub_ps(ix1,jx0);
1145 dy10 = _mm_sub_ps(iy1,jy0);
1146 dz10 = _mm_sub_ps(iz1,jz0);
1147 dx20 = _mm_sub_ps(ix2,jx0);
1148 dy20 = _mm_sub_ps(iy2,jy0);
1149 dz20 = _mm_sub_ps(iz2,jz0);
1150 dx30 = _mm_sub_ps(ix3,jx0);
1151 dy30 = _mm_sub_ps(iy3,jy0);
1152 dz30 = _mm_sub_ps(iz3,jz0);
1154 /* Calculate squared distance and things based on it */
1155 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1156 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1157 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1158 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1160 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1161 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1162 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1163 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1165 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1166 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1167 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1168 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1170 /* Load parameters for j particles */
1171 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1172 charge+jnrC+0,charge+jnrD+0);
1173 vdwjidx0A = 2*vdwtype[jnrA+0];
1174 vdwjidx0B = 2*vdwtype[jnrB+0];
1175 vdwjidx0C = 2*vdwtype[jnrC+0];
1176 vdwjidx0D = 2*vdwtype[jnrD+0];
1178 /**************************
1179 * CALCULATE INTERACTIONS *
1180 **************************/
1182 if (gmx_mm_any_lt(rsq00,rcutoff2))
1185 r00 = _mm_mul_ps(rsq00,rinv00);
1186 r00 = _mm_andnot_ps(dummy_mask,r00);
1188 /* Compute parameters for interactions between i and j atoms */
1189 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1190 vdwparam+vdwioffset0+vdwjidx0B,
1191 vdwparam+vdwioffset0+vdwjidx0C,
1192 vdwparam+vdwioffset0+vdwjidx0D,
1195 /* LENNARD-JONES DISPERSION/REPULSION */
1197 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1198 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
1199 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1200 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1201 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1203 d = _mm_sub_ps(r00,rswitch);
1204 d = _mm_max_ps(d,_mm_setzero_ps());
1205 d2 = _mm_mul_ps(d,d);
1206 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1208 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1210 /* Evaluate switch function */
1211 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1212 fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1213 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1217 fscal = _mm_and_ps(fscal,cutoff_mask);
1219 fscal = _mm_andnot_ps(dummy_mask,fscal);
1221 /* Calculate temporary vectorial force */
1222 tx = _mm_mul_ps(fscal,dx00);
1223 ty = _mm_mul_ps(fscal,dy00);
1224 tz = _mm_mul_ps(fscal,dz00);
1226 /* Update vectorial force */
1227 fix0 = _mm_add_ps(fix0,tx);
1228 fiy0 = _mm_add_ps(fiy0,ty);
1229 fiz0 = _mm_add_ps(fiz0,tz);
1231 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1232 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1233 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1234 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1235 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1239 /**************************
1240 * CALCULATE INTERACTIONS *
1241 **************************/
1243 if (gmx_mm_any_lt(rsq10,rcutoff2))
1246 /* Compute parameters for interactions between i and j atoms */
1247 qq10 = _mm_mul_ps(iq1,jq0);
1249 /* REACTION-FIELD ELECTROSTATICS */
1250 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1252 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1256 fscal = _mm_and_ps(fscal,cutoff_mask);
1258 fscal = _mm_andnot_ps(dummy_mask,fscal);
1260 /* Calculate temporary vectorial force */
1261 tx = _mm_mul_ps(fscal,dx10);
1262 ty = _mm_mul_ps(fscal,dy10);
1263 tz = _mm_mul_ps(fscal,dz10);
1265 /* Update vectorial force */
1266 fix1 = _mm_add_ps(fix1,tx);
1267 fiy1 = _mm_add_ps(fiy1,ty);
1268 fiz1 = _mm_add_ps(fiz1,tz);
1270 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1271 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1272 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1273 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1274 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1278 /**************************
1279 * CALCULATE INTERACTIONS *
1280 **************************/
1282 if (gmx_mm_any_lt(rsq20,rcutoff2))
1285 /* Compute parameters for interactions between i and j atoms */
1286 qq20 = _mm_mul_ps(iq2,jq0);
1288 /* REACTION-FIELD ELECTROSTATICS */
1289 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1291 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1295 fscal = _mm_and_ps(fscal,cutoff_mask);
1297 fscal = _mm_andnot_ps(dummy_mask,fscal);
1299 /* Calculate temporary vectorial force */
1300 tx = _mm_mul_ps(fscal,dx20);
1301 ty = _mm_mul_ps(fscal,dy20);
1302 tz = _mm_mul_ps(fscal,dz20);
1304 /* Update vectorial force */
1305 fix2 = _mm_add_ps(fix2,tx);
1306 fiy2 = _mm_add_ps(fiy2,ty);
1307 fiz2 = _mm_add_ps(fiz2,tz);
1309 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1310 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1311 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1312 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1313 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1317 /**************************
1318 * CALCULATE INTERACTIONS *
1319 **************************/
1321 if (gmx_mm_any_lt(rsq30,rcutoff2))
1324 /* Compute parameters for interactions between i and j atoms */
1325 qq30 = _mm_mul_ps(iq3,jq0);
1327 /* REACTION-FIELD ELECTROSTATICS */
1328 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1330 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1334 fscal = _mm_and_ps(fscal,cutoff_mask);
1336 fscal = _mm_andnot_ps(dummy_mask,fscal);
1338 /* Calculate temporary vectorial force */
1339 tx = _mm_mul_ps(fscal,dx30);
1340 ty = _mm_mul_ps(fscal,dy30);
1341 tz = _mm_mul_ps(fscal,dz30);
1343 /* Update vectorial force */
1344 fix3 = _mm_add_ps(fix3,tx);
1345 fiy3 = _mm_add_ps(fiy3,ty);
1346 fiz3 = _mm_add_ps(fiz3,tz);
1348 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1349 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1350 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1351 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1352 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1356 /* Inner loop uses 147 flops */
1359 /* End of innermost loop */
1361 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1362 f+i_coord_offset,fshift+i_shift_offset);
1364 /* Increment number of inner iterations */
1365 inneriter += j_index_end - j_index_start;
1367 /* Outer loop uses 24 flops */
1370 /* Increment number of outer iterations */
1373 /* Update outer/inner flops */
1375 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);