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 sse4_1_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse4_1_single.h"
48 #include "kernelutil_x86_sse4_1_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: CubicSplineTable
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
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 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
97 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
100 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
104 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
106 __m128i ifour = _mm_set1_epi32(4);
107 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
109 __m128 dummy_mask,cutoff_mask;
110 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
111 __m128 one = _mm_set1_ps(1.0);
112 __m128 two = _mm_set1_ps(2.0);
118 jindex = nlist->jindex;
120 shiftidx = nlist->shift;
122 shiftvec = fr->shift_vec[0];
123 fshift = fr->fshift[0];
124 facel = _mm_set1_ps(fr->epsfac);
125 charge = mdatoms->chargeA;
126 krf = _mm_set1_ps(fr->ic->k_rf);
127 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
128 crf = _mm_set1_ps(fr->ic->c_rf);
129 nvdwtype = fr->ntype;
131 vdwtype = mdatoms->typeA;
133 vftab = kernel_data->table_vdw->data;
134 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
136 /* Setup water-specific parameters */
137 inr = nlist->iinr[0];
138 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
139 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
140 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
141 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
143 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
144 rcutoff_scalar = fr->rcoulomb;
145 rcutoff = _mm_set1_ps(rcutoff_scalar);
146 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
148 /* Avoid stupid compiler warnings */
149 jnrA = jnrB = jnrC = jnrD = 0;
158 for(iidx=0;iidx<4*DIM;iidx++)
163 /* Start outer loop over neighborlists */
164 for(iidx=0; iidx<nri; iidx++)
166 /* Load shift vector for this list */
167 i_shift_offset = DIM*shiftidx[iidx];
169 /* Load limits for loop over neighbors */
170 j_index_start = jindex[iidx];
171 j_index_end = jindex[iidx+1];
173 /* Get outer coordinate index */
175 i_coord_offset = DIM*inr;
177 /* Load i particle coords and add shift vector */
178 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
179 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
181 fix0 = _mm_setzero_ps();
182 fiy0 = _mm_setzero_ps();
183 fiz0 = _mm_setzero_ps();
184 fix1 = _mm_setzero_ps();
185 fiy1 = _mm_setzero_ps();
186 fiz1 = _mm_setzero_ps();
187 fix2 = _mm_setzero_ps();
188 fiy2 = _mm_setzero_ps();
189 fiz2 = _mm_setzero_ps();
190 fix3 = _mm_setzero_ps();
191 fiy3 = _mm_setzero_ps();
192 fiz3 = _mm_setzero_ps();
194 /* Reset potential sums */
195 velecsum = _mm_setzero_ps();
196 vvdwsum = _mm_setzero_ps();
198 /* Start inner kernel loop */
199 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
202 /* Get j neighbor index, and coordinate index */
207 j_coord_offsetA = DIM*jnrA;
208 j_coord_offsetB = DIM*jnrB;
209 j_coord_offsetC = DIM*jnrC;
210 j_coord_offsetD = DIM*jnrD;
212 /* load j atom coordinates */
213 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
214 x+j_coord_offsetC,x+j_coord_offsetD,
217 /* Calculate displacement vector */
218 dx00 = _mm_sub_ps(ix0,jx0);
219 dy00 = _mm_sub_ps(iy0,jy0);
220 dz00 = _mm_sub_ps(iz0,jz0);
221 dx10 = _mm_sub_ps(ix1,jx0);
222 dy10 = _mm_sub_ps(iy1,jy0);
223 dz10 = _mm_sub_ps(iz1,jz0);
224 dx20 = _mm_sub_ps(ix2,jx0);
225 dy20 = _mm_sub_ps(iy2,jy0);
226 dz20 = _mm_sub_ps(iz2,jz0);
227 dx30 = _mm_sub_ps(ix3,jx0);
228 dy30 = _mm_sub_ps(iy3,jy0);
229 dz30 = _mm_sub_ps(iz3,jz0);
231 /* Calculate squared distance and things based on it */
232 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
233 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
234 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
235 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
237 rinv00 = gmx_mm_invsqrt_ps(rsq00);
238 rinv10 = gmx_mm_invsqrt_ps(rsq10);
239 rinv20 = gmx_mm_invsqrt_ps(rsq20);
240 rinv30 = gmx_mm_invsqrt_ps(rsq30);
242 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
243 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
244 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
246 /* Load parameters for j particles */
247 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
248 charge+jnrC+0,charge+jnrD+0);
249 vdwjidx0A = 2*vdwtype[jnrA+0];
250 vdwjidx0B = 2*vdwtype[jnrB+0];
251 vdwjidx0C = 2*vdwtype[jnrC+0];
252 vdwjidx0D = 2*vdwtype[jnrD+0];
254 fjx0 = _mm_setzero_ps();
255 fjy0 = _mm_setzero_ps();
256 fjz0 = _mm_setzero_ps();
258 /**************************
259 * CALCULATE INTERACTIONS *
260 **************************/
262 r00 = _mm_mul_ps(rsq00,rinv00);
264 /* Compute parameters for interactions between i and j atoms */
265 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
266 vdwparam+vdwioffset0+vdwjidx0B,
267 vdwparam+vdwioffset0+vdwjidx0C,
268 vdwparam+vdwioffset0+vdwjidx0D,
271 /* Calculate table index by multiplying r with table scale and truncate to integer */
272 rt = _mm_mul_ps(r00,vftabscale);
273 vfitab = _mm_cvttps_epi32(rt);
274 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
275 vfitab = _mm_slli_epi32(vfitab,3);
277 /* CUBIC SPLINE TABLE DISPERSION */
278 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
279 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
280 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
281 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
282 _MM_TRANSPOSE4_PS(Y,F,G,H);
283 Heps = _mm_mul_ps(vfeps,H);
284 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
285 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
286 vvdw6 = _mm_mul_ps(c6_00,VV);
287 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
288 fvdw6 = _mm_mul_ps(c6_00,FF);
290 /* CUBIC SPLINE TABLE REPULSION */
291 vfitab = _mm_add_epi32(vfitab,ifour);
292 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
293 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
294 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
295 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
296 _MM_TRANSPOSE4_PS(Y,F,G,H);
297 Heps = _mm_mul_ps(vfeps,H);
298 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
299 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
300 vvdw12 = _mm_mul_ps(c12_00,VV);
301 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
302 fvdw12 = _mm_mul_ps(c12_00,FF);
303 vvdw = _mm_add_ps(vvdw12,vvdw6);
304 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
306 /* Update potential sum for this i atom from the interaction with this j atom. */
307 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
311 /* Calculate temporary vectorial force */
312 tx = _mm_mul_ps(fscal,dx00);
313 ty = _mm_mul_ps(fscal,dy00);
314 tz = _mm_mul_ps(fscal,dz00);
316 /* Update vectorial force */
317 fix0 = _mm_add_ps(fix0,tx);
318 fiy0 = _mm_add_ps(fiy0,ty);
319 fiz0 = _mm_add_ps(fiz0,tz);
321 fjx0 = _mm_add_ps(fjx0,tx);
322 fjy0 = _mm_add_ps(fjy0,ty);
323 fjz0 = _mm_add_ps(fjz0,tz);
325 /**************************
326 * CALCULATE INTERACTIONS *
327 **************************/
329 if (gmx_mm_any_lt(rsq10,rcutoff2))
332 /* Compute parameters for interactions between i and j atoms */
333 qq10 = _mm_mul_ps(iq1,jq0);
335 /* REACTION-FIELD ELECTROSTATICS */
336 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
337 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
339 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
341 /* Update potential sum for this i atom from the interaction with this j atom. */
342 velec = _mm_and_ps(velec,cutoff_mask);
343 velecsum = _mm_add_ps(velecsum,velec);
347 fscal = _mm_and_ps(fscal,cutoff_mask);
349 /* Calculate temporary vectorial force */
350 tx = _mm_mul_ps(fscal,dx10);
351 ty = _mm_mul_ps(fscal,dy10);
352 tz = _mm_mul_ps(fscal,dz10);
354 /* Update vectorial force */
355 fix1 = _mm_add_ps(fix1,tx);
356 fiy1 = _mm_add_ps(fiy1,ty);
357 fiz1 = _mm_add_ps(fiz1,tz);
359 fjx0 = _mm_add_ps(fjx0,tx);
360 fjy0 = _mm_add_ps(fjy0,ty);
361 fjz0 = _mm_add_ps(fjz0,tz);
365 /**************************
366 * CALCULATE INTERACTIONS *
367 **************************/
369 if (gmx_mm_any_lt(rsq20,rcutoff2))
372 /* Compute parameters for interactions between i and j atoms */
373 qq20 = _mm_mul_ps(iq2,jq0);
375 /* REACTION-FIELD ELECTROSTATICS */
376 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
377 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
379 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
381 /* Update potential sum for this i atom from the interaction with this j atom. */
382 velec = _mm_and_ps(velec,cutoff_mask);
383 velecsum = _mm_add_ps(velecsum,velec);
387 fscal = _mm_and_ps(fscal,cutoff_mask);
389 /* Calculate temporary vectorial force */
390 tx = _mm_mul_ps(fscal,dx20);
391 ty = _mm_mul_ps(fscal,dy20);
392 tz = _mm_mul_ps(fscal,dz20);
394 /* Update vectorial force */
395 fix2 = _mm_add_ps(fix2,tx);
396 fiy2 = _mm_add_ps(fiy2,ty);
397 fiz2 = _mm_add_ps(fiz2,tz);
399 fjx0 = _mm_add_ps(fjx0,tx);
400 fjy0 = _mm_add_ps(fjy0,ty);
401 fjz0 = _mm_add_ps(fjz0,tz);
405 /**************************
406 * CALCULATE INTERACTIONS *
407 **************************/
409 if (gmx_mm_any_lt(rsq30,rcutoff2))
412 /* Compute parameters for interactions between i and j atoms */
413 qq30 = _mm_mul_ps(iq3,jq0);
415 /* REACTION-FIELD ELECTROSTATICS */
416 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
417 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
419 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
421 /* Update potential sum for this i atom from the interaction with this j atom. */
422 velec = _mm_and_ps(velec,cutoff_mask);
423 velecsum = _mm_add_ps(velecsum,velec);
427 fscal = _mm_and_ps(fscal,cutoff_mask);
429 /* Calculate temporary vectorial force */
430 tx = _mm_mul_ps(fscal,dx30);
431 ty = _mm_mul_ps(fscal,dy30);
432 tz = _mm_mul_ps(fscal,dz30);
434 /* Update vectorial force */
435 fix3 = _mm_add_ps(fix3,tx);
436 fiy3 = _mm_add_ps(fiy3,ty);
437 fiz3 = _mm_add_ps(fiz3,tz);
439 fjx0 = _mm_add_ps(fjx0,tx);
440 fjy0 = _mm_add_ps(fjy0,ty);
441 fjz0 = _mm_add_ps(fjz0,tz);
445 fjptrA = f+j_coord_offsetA;
446 fjptrB = f+j_coord_offsetB;
447 fjptrC = f+j_coord_offsetC;
448 fjptrD = f+j_coord_offsetD;
450 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
452 /* Inner loop uses 164 flops */
458 /* Get j neighbor index, and coordinate index */
459 jnrlistA = jjnr[jidx];
460 jnrlistB = jjnr[jidx+1];
461 jnrlistC = jjnr[jidx+2];
462 jnrlistD = jjnr[jidx+3];
463 /* Sign of each element will be negative for non-real atoms.
464 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
465 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
467 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
468 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
469 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
470 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
471 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
472 j_coord_offsetA = DIM*jnrA;
473 j_coord_offsetB = DIM*jnrB;
474 j_coord_offsetC = DIM*jnrC;
475 j_coord_offsetD = DIM*jnrD;
477 /* load j atom coordinates */
478 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
479 x+j_coord_offsetC,x+j_coord_offsetD,
482 /* Calculate displacement vector */
483 dx00 = _mm_sub_ps(ix0,jx0);
484 dy00 = _mm_sub_ps(iy0,jy0);
485 dz00 = _mm_sub_ps(iz0,jz0);
486 dx10 = _mm_sub_ps(ix1,jx0);
487 dy10 = _mm_sub_ps(iy1,jy0);
488 dz10 = _mm_sub_ps(iz1,jz0);
489 dx20 = _mm_sub_ps(ix2,jx0);
490 dy20 = _mm_sub_ps(iy2,jy0);
491 dz20 = _mm_sub_ps(iz2,jz0);
492 dx30 = _mm_sub_ps(ix3,jx0);
493 dy30 = _mm_sub_ps(iy3,jy0);
494 dz30 = _mm_sub_ps(iz3,jz0);
496 /* Calculate squared distance and things based on it */
497 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
498 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
499 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
500 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
502 rinv00 = gmx_mm_invsqrt_ps(rsq00);
503 rinv10 = gmx_mm_invsqrt_ps(rsq10);
504 rinv20 = gmx_mm_invsqrt_ps(rsq20);
505 rinv30 = gmx_mm_invsqrt_ps(rsq30);
507 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
508 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
509 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
511 /* Load parameters for j particles */
512 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
513 charge+jnrC+0,charge+jnrD+0);
514 vdwjidx0A = 2*vdwtype[jnrA+0];
515 vdwjidx0B = 2*vdwtype[jnrB+0];
516 vdwjidx0C = 2*vdwtype[jnrC+0];
517 vdwjidx0D = 2*vdwtype[jnrD+0];
519 fjx0 = _mm_setzero_ps();
520 fjy0 = _mm_setzero_ps();
521 fjz0 = _mm_setzero_ps();
523 /**************************
524 * CALCULATE INTERACTIONS *
525 **************************/
527 r00 = _mm_mul_ps(rsq00,rinv00);
528 r00 = _mm_andnot_ps(dummy_mask,r00);
530 /* Compute parameters for interactions between i and j atoms */
531 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
532 vdwparam+vdwioffset0+vdwjidx0B,
533 vdwparam+vdwioffset0+vdwjidx0C,
534 vdwparam+vdwioffset0+vdwjidx0D,
537 /* Calculate table index by multiplying r with table scale and truncate to integer */
538 rt = _mm_mul_ps(r00,vftabscale);
539 vfitab = _mm_cvttps_epi32(rt);
540 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
541 vfitab = _mm_slli_epi32(vfitab,3);
543 /* CUBIC SPLINE TABLE DISPERSION */
544 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
545 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
546 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
547 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
548 _MM_TRANSPOSE4_PS(Y,F,G,H);
549 Heps = _mm_mul_ps(vfeps,H);
550 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
551 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
552 vvdw6 = _mm_mul_ps(c6_00,VV);
553 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
554 fvdw6 = _mm_mul_ps(c6_00,FF);
556 /* CUBIC SPLINE TABLE REPULSION */
557 vfitab = _mm_add_epi32(vfitab,ifour);
558 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
559 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
560 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
561 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
562 _MM_TRANSPOSE4_PS(Y,F,G,H);
563 Heps = _mm_mul_ps(vfeps,H);
564 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
565 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
566 vvdw12 = _mm_mul_ps(c12_00,VV);
567 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
568 fvdw12 = _mm_mul_ps(c12_00,FF);
569 vvdw = _mm_add_ps(vvdw12,vvdw6);
570 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
572 /* Update potential sum for this i atom from the interaction with this j atom. */
573 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
574 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
578 fscal = _mm_andnot_ps(dummy_mask,fscal);
580 /* Calculate temporary vectorial force */
581 tx = _mm_mul_ps(fscal,dx00);
582 ty = _mm_mul_ps(fscal,dy00);
583 tz = _mm_mul_ps(fscal,dz00);
585 /* Update vectorial force */
586 fix0 = _mm_add_ps(fix0,tx);
587 fiy0 = _mm_add_ps(fiy0,ty);
588 fiz0 = _mm_add_ps(fiz0,tz);
590 fjx0 = _mm_add_ps(fjx0,tx);
591 fjy0 = _mm_add_ps(fjy0,ty);
592 fjz0 = _mm_add_ps(fjz0,tz);
594 /**************************
595 * CALCULATE INTERACTIONS *
596 **************************/
598 if (gmx_mm_any_lt(rsq10,rcutoff2))
601 /* Compute parameters for interactions between i and j atoms */
602 qq10 = _mm_mul_ps(iq1,jq0);
604 /* REACTION-FIELD ELECTROSTATICS */
605 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
606 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
608 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
610 /* Update potential sum for this i atom from the interaction with this j atom. */
611 velec = _mm_and_ps(velec,cutoff_mask);
612 velec = _mm_andnot_ps(dummy_mask,velec);
613 velecsum = _mm_add_ps(velecsum,velec);
617 fscal = _mm_and_ps(fscal,cutoff_mask);
619 fscal = _mm_andnot_ps(dummy_mask,fscal);
621 /* Calculate temporary vectorial force */
622 tx = _mm_mul_ps(fscal,dx10);
623 ty = _mm_mul_ps(fscal,dy10);
624 tz = _mm_mul_ps(fscal,dz10);
626 /* Update vectorial force */
627 fix1 = _mm_add_ps(fix1,tx);
628 fiy1 = _mm_add_ps(fiy1,ty);
629 fiz1 = _mm_add_ps(fiz1,tz);
631 fjx0 = _mm_add_ps(fjx0,tx);
632 fjy0 = _mm_add_ps(fjy0,ty);
633 fjz0 = _mm_add_ps(fjz0,tz);
637 /**************************
638 * CALCULATE INTERACTIONS *
639 **************************/
641 if (gmx_mm_any_lt(rsq20,rcutoff2))
644 /* Compute parameters for interactions between i and j atoms */
645 qq20 = _mm_mul_ps(iq2,jq0);
647 /* REACTION-FIELD ELECTROSTATICS */
648 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
649 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
651 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
653 /* Update potential sum for this i atom from the interaction with this j atom. */
654 velec = _mm_and_ps(velec,cutoff_mask);
655 velec = _mm_andnot_ps(dummy_mask,velec);
656 velecsum = _mm_add_ps(velecsum,velec);
660 fscal = _mm_and_ps(fscal,cutoff_mask);
662 fscal = _mm_andnot_ps(dummy_mask,fscal);
664 /* Calculate temporary vectorial force */
665 tx = _mm_mul_ps(fscal,dx20);
666 ty = _mm_mul_ps(fscal,dy20);
667 tz = _mm_mul_ps(fscal,dz20);
669 /* Update vectorial force */
670 fix2 = _mm_add_ps(fix2,tx);
671 fiy2 = _mm_add_ps(fiy2,ty);
672 fiz2 = _mm_add_ps(fiz2,tz);
674 fjx0 = _mm_add_ps(fjx0,tx);
675 fjy0 = _mm_add_ps(fjy0,ty);
676 fjz0 = _mm_add_ps(fjz0,tz);
680 /**************************
681 * CALCULATE INTERACTIONS *
682 **************************/
684 if (gmx_mm_any_lt(rsq30,rcutoff2))
687 /* Compute parameters for interactions between i and j atoms */
688 qq30 = _mm_mul_ps(iq3,jq0);
690 /* REACTION-FIELD ELECTROSTATICS */
691 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
692 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
694 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
696 /* Update potential sum for this i atom from the interaction with this j atom. */
697 velec = _mm_and_ps(velec,cutoff_mask);
698 velec = _mm_andnot_ps(dummy_mask,velec);
699 velecsum = _mm_add_ps(velecsum,velec);
703 fscal = _mm_and_ps(fscal,cutoff_mask);
705 fscal = _mm_andnot_ps(dummy_mask,fscal);
707 /* Calculate temporary vectorial force */
708 tx = _mm_mul_ps(fscal,dx30);
709 ty = _mm_mul_ps(fscal,dy30);
710 tz = _mm_mul_ps(fscal,dz30);
712 /* Update vectorial force */
713 fix3 = _mm_add_ps(fix3,tx);
714 fiy3 = _mm_add_ps(fiy3,ty);
715 fiz3 = _mm_add_ps(fiz3,tz);
717 fjx0 = _mm_add_ps(fjx0,tx);
718 fjy0 = _mm_add_ps(fjy0,ty);
719 fjz0 = _mm_add_ps(fjz0,tz);
723 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
724 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
725 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
726 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
728 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
730 /* Inner loop uses 165 flops */
733 /* End of innermost loop */
735 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
736 f+i_coord_offset,fshift+i_shift_offset);
739 /* Update potential energies */
740 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
741 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
743 /* Increment number of inner iterations */
744 inneriter += j_index_end - j_index_start;
746 /* Outer loop uses 26 flops */
749 /* Increment number of outer iterations */
752 /* Update outer/inner flops */
754 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*165);
757 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single
758 * Electrostatics interaction: ReactionField
759 * VdW interaction: CubicSplineTable
760 * Geometry: Water4-Particle
761 * Calculate force/pot: Force
764 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single
765 (t_nblist * gmx_restrict nlist,
766 rvec * gmx_restrict xx,
767 rvec * gmx_restrict ff,
768 t_forcerec * gmx_restrict fr,
769 t_mdatoms * gmx_restrict mdatoms,
770 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
771 t_nrnb * gmx_restrict nrnb)
773 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
774 * just 0 for non-waters.
775 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
776 * jnr indices corresponding to data put in the four positions in the SIMD register.
778 int i_shift_offset,i_coord_offset,outeriter,inneriter;
779 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
780 int jnrA,jnrB,jnrC,jnrD;
781 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
782 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
783 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
785 real *shiftvec,*fshift,*x,*f;
786 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
788 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
790 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
792 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
794 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
796 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
797 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
798 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
799 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
800 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
801 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
802 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
803 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
806 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
809 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
810 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
812 __m128i ifour = _mm_set1_epi32(4);
813 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
815 __m128 dummy_mask,cutoff_mask;
816 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
817 __m128 one = _mm_set1_ps(1.0);
818 __m128 two = _mm_set1_ps(2.0);
824 jindex = nlist->jindex;
826 shiftidx = nlist->shift;
828 shiftvec = fr->shift_vec[0];
829 fshift = fr->fshift[0];
830 facel = _mm_set1_ps(fr->epsfac);
831 charge = mdatoms->chargeA;
832 krf = _mm_set1_ps(fr->ic->k_rf);
833 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
834 crf = _mm_set1_ps(fr->ic->c_rf);
835 nvdwtype = fr->ntype;
837 vdwtype = mdatoms->typeA;
839 vftab = kernel_data->table_vdw->data;
840 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
842 /* Setup water-specific parameters */
843 inr = nlist->iinr[0];
844 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
845 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
846 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
847 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
849 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
850 rcutoff_scalar = fr->rcoulomb;
851 rcutoff = _mm_set1_ps(rcutoff_scalar);
852 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
854 /* Avoid stupid compiler warnings */
855 jnrA = jnrB = jnrC = jnrD = 0;
864 for(iidx=0;iidx<4*DIM;iidx++)
869 /* Start outer loop over neighborlists */
870 for(iidx=0; iidx<nri; iidx++)
872 /* Load shift vector for this list */
873 i_shift_offset = DIM*shiftidx[iidx];
875 /* Load limits for loop over neighbors */
876 j_index_start = jindex[iidx];
877 j_index_end = jindex[iidx+1];
879 /* Get outer coordinate index */
881 i_coord_offset = DIM*inr;
883 /* Load i particle coords and add shift vector */
884 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
885 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
887 fix0 = _mm_setzero_ps();
888 fiy0 = _mm_setzero_ps();
889 fiz0 = _mm_setzero_ps();
890 fix1 = _mm_setzero_ps();
891 fiy1 = _mm_setzero_ps();
892 fiz1 = _mm_setzero_ps();
893 fix2 = _mm_setzero_ps();
894 fiy2 = _mm_setzero_ps();
895 fiz2 = _mm_setzero_ps();
896 fix3 = _mm_setzero_ps();
897 fiy3 = _mm_setzero_ps();
898 fiz3 = _mm_setzero_ps();
900 /* Start inner kernel loop */
901 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
904 /* Get j neighbor index, and coordinate index */
909 j_coord_offsetA = DIM*jnrA;
910 j_coord_offsetB = DIM*jnrB;
911 j_coord_offsetC = DIM*jnrC;
912 j_coord_offsetD = DIM*jnrD;
914 /* load j atom coordinates */
915 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
916 x+j_coord_offsetC,x+j_coord_offsetD,
919 /* Calculate displacement vector */
920 dx00 = _mm_sub_ps(ix0,jx0);
921 dy00 = _mm_sub_ps(iy0,jy0);
922 dz00 = _mm_sub_ps(iz0,jz0);
923 dx10 = _mm_sub_ps(ix1,jx0);
924 dy10 = _mm_sub_ps(iy1,jy0);
925 dz10 = _mm_sub_ps(iz1,jz0);
926 dx20 = _mm_sub_ps(ix2,jx0);
927 dy20 = _mm_sub_ps(iy2,jy0);
928 dz20 = _mm_sub_ps(iz2,jz0);
929 dx30 = _mm_sub_ps(ix3,jx0);
930 dy30 = _mm_sub_ps(iy3,jy0);
931 dz30 = _mm_sub_ps(iz3,jz0);
933 /* Calculate squared distance and things based on it */
934 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
935 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
936 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
937 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
939 rinv00 = gmx_mm_invsqrt_ps(rsq00);
940 rinv10 = gmx_mm_invsqrt_ps(rsq10);
941 rinv20 = gmx_mm_invsqrt_ps(rsq20);
942 rinv30 = gmx_mm_invsqrt_ps(rsq30);
944 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
945 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
946 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
948 /* Load parameters for j particles */
949 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
950 charge+jnrC+0,charge+jnrD+0);
951 vdwjidx0A = 2*vdwtype[jnrA+0];
952 vdwjidx0B = 2*vdwtype[jnrB+0];
953 vdwjidx0C = 2*vdwtype[jnrC+0];
954 vdwjidx0D = 2*vdwtype[jnrD+0];
956 fjx0 = _mm_setzero_ps();
957 fjy0 = _mm_setzero_ps();
958 fjz0 = _mm_setzero_ps();
960 /**************************
961 * CALCULATE INTERACTIONS *
962 **************************/
964 r00 = _mm_mul_ps(rsq00,rinv00);
966 /* Compute parameters for interactions between i and j atoms */
967 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
968 vdwparam+vdwioffset0+vdwjidx0B,
969 vdwparam+vdwioffset0+vdwjidx0C,
970 vdwparam+vdwioffset0+vdwjidx0D,
973 /* Calculate table index by multiplying r with table scale and truncate to integer */
974 rt = _mm_mul_ps(r00,vftabscale);
975 vfitab = _mm_cvttps_epi32(rt);
976 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
977 vfitab = _mm_slli_epi32(vfitab,3);
979 /* CUBIC SPLINE TABLE DISPERSION */
980 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
981 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
982 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
983 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
984 _MM_TRANSPOSE4_PS(Y,F,G,H);
985 Heps = _mm_mul_ps(vfeps,H);
986 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
987 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
988 fvdw6 = _mm_mul_ps(c6_00,FF);
990 /* CUBIC SPLINE TABLE REPULSION */
991 vfitab = _mm_add_epi32(vfitab,ifour);
992 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
993 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
994 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
995 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
996 _MM_TRANSPOSE4_PS(Y,F,G,H);
997 Heps = _mm_mul_ps(vfeps,H);
998 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
999 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1000 fvdw12 = _mm_mul_ps(c12_00,FF);
1001 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1005 /* Calculate temporary vectorial force */
1006 tx = _mm_mul_ps(fscal,dx00);
1007 ty = _mm_mul_ps(fscal,dy00);
1008 tz = _mm_mul_ps(fscal,dz00);
1010 /* Update vectorial force */
1011 fix0 = _mm_add_ps(fix0,tx);
1012 fiy0 = _mm_add_ps(fiy0,ty);
1013 fiz0 = _mm_add_ps(fiz0,tz);
1015 fjx0 = _mm_add_ps(fjx0,tx);
1016 fjy0 = _mm_add_ps(fjy0,ty);
1017 fjz0 = _mm_add_ps(fjz0,tz);
1019 /**************************
1020 * CALCULATE INTERACTIONS *
1021 **************************/
1023 if (gmx_mm_any_lt(rsq10,rcutoff2))
1026 /* Compute parameters for interactions between i and j atoms */
1027 qq10 = _mm_mul_ps(iq1,jq0);
1029 /* REACTION-FIELD ELECTROSTATICS */
1030 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1032 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1036 fscal = _mm_and_ps(fscal,cutoff_mask);
1038 /* Calculate temporary vectorial force */
1039 tx = _mm_mul_ps(fscal,dx10);
1040 ty = _mm_mul_ps(fscal,dy10);
1041 tz = _mm_mul_ps(fscal,dz10);
1043 /* Update vectorial force */
1044 fix1 = _mm_add_ps(fix1,tx);
1045 fiy1 = _mm_add_ps(fiy1,ty);
1046 fiz1 = _mm_add_ps(fiz1,tz);
1048 fjx0 = _mm_add_ps(fjx0,tx);
1049 fjy0 = _mm_add_ps(fjy0,ty);
1050 fjz0 = _mm_add_ps(fjz0,tz);
1054 /**************************
1055 * CALCULATE INTERACTIONS *
1056 **************************/
1058 if (gmx_mm_any_lt(rsq20,rcutoff2))
1061 /* Compute parameters for interactions between i and j atoms */
1062 qq20 = _mm_mul_ps(iq2,jq0);
1064 /* REACTION-FIELD ELECTROSTATICS */
1065 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1067 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1071 fscal = _mm_and_ps(fscal,cutoff_mask);
1073 /* Calculate temporary vectorial force */
1074 tx = _mm_mul_ps(fscal,dx20);
1075 ty = _mm_mul_ps(fscal,dy20);
1076 tz = _mm_mul_ps(fscal,dz20);
1078 /* Update vectorial force */
1079 fix2 = _mm_add_ps(fix2,tx);
1080 fiy2 = _mm_add_ps(fiy2,ty);
1081 fiz2 = _mm_add_ps(fiz2,tz);
1083 fjx0 = _mm_add_ps(fjx0,tx);
1084 fjy0 = _mm_add_ps(fjy0,ty);
1085 fjz0 = _mm_add_ps(fjz0,tz);
1089 /**************************
1090 * CALCULATE INTERACTIONS *
1091 **************************/
1093 if (gmx_mm_any_lt(rsq30,rcutoff2))
1096 /* Compute parameters for interactions between i and j atoms */
1097 qq30 = _mm_mul_ps(iq3,jq0);
1099 /* REACTION-FIELD ELECTROSTATICS */
1100 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1102 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1106 fscal = _mm_and_ps(fscal,cutoff_mask);
1108 /* Calculate temporary vectorial force */
1109 tx = _mm_mul_ps(fscal,dx30);
1110 ty = _mm_mul_ps(fscal,dy30);
1111 tz = _mm_mul_ps(fscal,dz30);
1113 /* Update vectorial force */
1114 fix3 = _mm_add_ps(fix3,tx);
1115 fiy3 = _mm_add_ps(fiy3,ty);
1116 fiz3 = _mm_add_ps(fiz3,tz);
1118 fjx0 = _mm_add_ps(fjx0,tx);
1119 fjy0 = _mm_add_ps(fjy0,ty);
1120 fjz0 = _mm_add_ps(fjz0,tz);
1124 fjptrA = f+j_coord_offsetA;
1125 fjptrB = f+j_coord_offsetB;
1126 fjptrC = f+j_coord_offsetC;
1127 fjptrD = f+j_coord_offsetD;
1129 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1131 /* Inner loop uses 138 flops */
1134 if(jidx<j_index_end)
1137 /* Get j neighbor index, and coordinate index */
1138 jnrlistA = jjnr[jidx];
1139 jnrlistB = jjnr[jidx+1];
1140 jnrlistC = jjnr[jidx+2];
1141 jnrlistD = jjnr[jidx+3];
1142 /* Sign of each element will be negative for non-real atoms.
1143 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1144 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1146 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1147 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1148 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1149 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1150 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1151 j_coord_offsetA = DIM*jnrA;
1152 j_coord_offsetB = DIM*jnrB;
1153 j_coord_offsetC = DIM*jnrC;
1154 j_coord_offsetD = DIM*jnrD;
1156 /* load j atom coordinates */
1157 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1158 x+j_coord_offsetC,x+j_coord_offsetD,
1161 /* Calculate displacement vector */
1162 dx00 = _mm_sub_ps(ix0,jx0);
1163 dy00 = _mm_sub_ps(iy0,jy0);
1164 dz00 = _mm_sub_ps(iz0,jz0);
1165 dx10 = _mm_sub_ps(ix1,jx0);
1166 dy10 = _mm_sub_ps(iy1,jy0);
1167 dz10 = _mm_sub_ps(iz1,jz0);
1168 dx20 = _mm_sub_ps(ix2,jx0);
1169 dy20 = _mm_sub_ps(iy2,jy0);
1170 dz20 = _mm_sub_ps(iz2,jz0);
1171 dx30 = _mm_sub_ps(ix3,jx0);
1172 dy30 = _mm_sub_ps(iy3,jy0);
1173 dz30 = _mm_sub_ps(iz3,jz0);
1175 /* Calculate squared distance and things based on it */
1176 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1177 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1178 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1179 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1181 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1182 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1183 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1184 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1186 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1187 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1188 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1190 /* Load parameters for j particles */
1191 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1192 charge+jnrC+0,charge+jnrD+0);
1193 vdwjidx0A = 2*vdwtype[jnrA+0];
1194 vdwjidx0B = 2*vdwtype[jnrB+0];
1195 vdwjidx0C = 2*vdwtype[jnrC+0];
1196 vdwjidx0D = 2*vdwtype[jnrD+0];
1198 fjx0 = _mm_setzero_ps();
1199 fjy0 = _mm_setzero_ps();
1200 fjz0 = _mm_setzero_ps();
1202 /**************************
1203 * CALCULATE INTERACTIONS *
1204 **************************/
1206 r00 = _mm_mul_ps(rsq00,rinv00);
1207 r00 = _mm_andnot_ps(dummy_mask,r00);
1209 /* Compute parameters for interactions between i and j atoms */
1210 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1211 vdwparam+vdwioffset0+vdwjidx0B,
1212 vdwparam+vdwioffset0+vdwjidx0C,
1213 vdwparam+vdwioffset0+vdwjidx0D,
1216 /* Calculate table index by multiplying r with table scale and truncate to integer */
1217 rt = _mm_mul_ps(r00,vftabscale);
1218 vfitab = _mm_cvttps_epi32(rt);
1219 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1220 vfitab = _mm_slli_epi32(vfitab,3);
1222 /* CUBIC SPLINE TABLE DISPERSION */
1223 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1224 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1225 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1226 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1227 _MM_TRANSPOSE4_PS(Y,F,G,H);
1228 Heps = _mm_mul_ps(vfeps,H);
1229 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1230 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1231 fvdw6 = _mm_mul_ps(c6_00,FF);
1233 /* CUBIC SPLINE TABLE REPULSION */
1234 vfitab = _mm_add_epi32(vfitab,ifour);
1235 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1236 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1237 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1238 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1239 _MM_TRANSPOSE4_PS(Y,F,G,H);
1240 Heps = _mm_mul_ps(vfeps,H);
1241 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1242 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1243 fvdw12 = _mm_mul_ps(c12_00,FF);
1244 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1248 fscal = _mm_andnot_ps(dummy_mask,fscal);
1250 /* Calculate temporary vectorial force */
1251 tx = _mm_mul_ps(fscal,dx00);
1252 ty = _mm_mul_ps(fscal,dy00);
1253 tz = _mm_mul_ps(fscal,dz00);
1255 /* Update vectorial force */
1256 fix0 = _mm_add_ps(fix0,tx);
1257 fiy0 = _mm_add_ps(fiy0,ty);
1258 fiz0 = _mm_add_ps(fiz0,tz);
1260 fjx0 = _mm_add_ps(fjx0,tx);
1261 fjy0 = _mm_add_ps(fjy0,ty);
1262 fjz0 = _mm_add_ps(fjz0,tz);
1264 /**************************
1265 * CALCULATE INTERACTIONS *
1266 **************************/
1268 if (gmx_mm_any_lt(rsq10,rcutoff2))
1271 /* Compute parameters for interactions between i and j atoms */
1272 qq10 = _mm_mul_ps(iq1,jq0);
1274 /* REACTION-FIELD ELECTROSTATICS */
1275 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1277 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1281 fscal = _mm_and_ps(fscal,cutoff_mask);
1283 fscal = _mm_andnot_ps(dummy_mask,fscal);
1285 /* Calculate temporary vectorial force */
1286 tx = _mm_mul_ps(fscal,dx10);
1287 ty = _mm_mul_ps(fscal,dy10);
1288 tz = _mm_mul_ps(fscal,dz10);
1290 /* Update vectorial force */
1291 fix1 = _mm_add_ps(fix1,tx);
1292 fiy1 = _mm_add_ps(fiy1,ty);
1293 fiz1 = _mm_add_ps(fiz1,tz);
1295 fjx0 = _mm_add_ps(fjx0,tx);
1296 fjy0 = _mm_add_ps(fjy0,ty);
1297 fjz0 = _mm_add_ps(fjz0,tz);
1301 /**************************
1302 * CALCULATE INTERACTIONS *
1303 **************************/
1305 if (gmx_mm_any_lt(rsq20,rcutoff2))
1308 /* Compute parameters for interactions between i and j atoms */
1309 qq20 = _mm_mul_ps(iq2,jq0);
1311 /* REACTION-FIELD ELECTROSTATICS */
1312 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1314 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1318 fscal = _mm_and_ps(fscal,cutoff_mask);
1320 fscal = _mm_andnot_ps(dummy_mask,fscal);
1322 /* Calculate temporary vectorial force */
1323 tx = _mm_mul_ps(fscal,dx20);
1324 ty = _mm_mul_ps(fscal,dy20);
1325 tz = _mm_mul_ps(fscal,dz20);
1327 /* Update vectorial force */
1328 fix2 = _mm_add_ps(fix2,tx);
1329 fiy2 = _mm_add_ps(fiy2,ty);
1330 fiz2 = _mm_add_ps(fiz2,tz);
1332 fjx0 = _mm_add_ps(fjx0,tx);
1333 fjy0 = _mm_add_ps(fjy0,ty);
1334 fjz0 = _mm_add_ps(fjz0,tz);
1338 /**************************
1339 * CALCULATE INTERACTIONS *
1340 **************************/
1342 if (gmx_mm_any_lt(rsq30,rcutoff2))
1345 /* Compute parameters for interactions between i and j atoms */
1346 qq30 = _mm_mul_ps(iq3,jq0);
1348 /* REACTION-FIELD ELECTROSTATICS */
1349 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1351 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1355 fscal = _mm_and_ps(fscal,cutoff_mask);
1357 fscal = _mm_andnot_ps(dummy_mask,fscal);
1359 /* Calculate temporary vectorial force */
1360 tx = _mm_mul_ps(fscal,dx30);
1361 ty = _mm_mul_ps(fscal,dy30);
1362 tz = _mm_mul_ps(fscal,dz30);
1364 /* Update vectorial force */
1365 fix3 = _mm_add_ps(fix3,tx);
1366 fiy3 = _mm_add_ps(fiy3,ty);
1367 fiz3 = _mm_add_ps(fiz3,tz);
1369 fjx0 = _mm_add_ps(fjx0,tx);
1370 fjy0 = _mm_add_ps(fjy0,ty);
1371 fjz0 = _mm_add_ps(fjz0,tz);
1375 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1376 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1377 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1378 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1380 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1382 /* Inner loop uses 139 flops */
1385 /* End of innermost loop */
1387 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1388 f+i_coord_offset,fshift+i_shift_offset);
1390 /* Increment number of inner iterations */
1391 inneriter += j_index_end - j_index_start;
1393 /* Outer loop uses 24 flops */
1396 /* Increment number of outer iterations */
1399 /* Update outer/inner flops */
1401 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*139);