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_ElecEw_VdwCSTab_GeomW4P1_VF_sse4_1_single
38 * Electrostatics interaction: Ewald
39 * VdW interaction: CubicSplineTable
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEw_VdwCSTab_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);
92 __m128i ifour = _mm_set1_epi32(4);
93 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
96 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
98 __m128 dummy_mask,cutoff_mask;
99 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
100 __m128 one = _mm_set1_ps(1.0);
101 __m128 two = _mm_set1_ps(2.0);
107 jindex = nlist->jindex;
109 shiftidx = nlist->shift;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm_set1_ps(fr->epsfac);
114 charge = mdatoms->chargeA;
115 nvdwtype = fr->ntype;
117 vdwtype = mdatoms->typeA;
119 vftab = kernel_data->table_vdw->data;
120 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
122 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
123 ewtab = fr->ic->tabq_coul_FDV0;
124 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
125 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
127 /* Setup water-specific parameters */
128 inr = nlist->iinr[0];
129 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
130 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
131 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
132 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
134 /* Avoid stupid compiler warnings */
135 jnrA = jnrB = jnrC = jnrD = 0;
144 for(iidx=0;iidx<4*DIM;iidx++)
149 /* Start outer loop over neighborlists */
150 for(iidx=0; iidx<nri; iidx++)
152 /* Load shift vector for this list */
153 i_shift_offset = DIM*shiftidx[iidx];
155 /* Load limits for loop over neighbors */
156 j_index_start = jindex[iidx];
157 j_index_end = jindex[iidx+1];
159 /* Get outer coordinate index */
161 i_coord_offset = DIM*inr;
163 /* Load i particle coords and add shift vector */
164 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
165 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
167 fix0 = _mm_setzero_ps();
168 fiy0 = _mm_setzero_ps();
169 fiz0 = _mm_setzero_ps();
170 fix1 = _mm_setzero_ps();
171 fiy1 = _mm_setzero_ps();
172 fiz1 = _mm_setzero_ps();
173 fix2 = _mm_setzero_ps();
174 fiy2 = _mm_setzero_ps();
175 fiz2 = _mm_setzero_ps();
176 fix3 = _mm_setzero_ps();
177 fiy3 = _mm_setzero_ps();
178 fiz3 = _mm_setzero_ps();
180 /* Reset potential sums */
181 velecsum = _mm_setzero_ps();
182 vvdwsum = _mm_setzero_ps();
184 /* Start inner kernel loop */
185 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
188 /* Get j neighbor index, and coordinate index */
193 j_coord_offsetA = DIM*jnrA;
194 j_coord_offsetB = DIM*jnrB;
195 j_coord_offsetC = DIM*jnrC;
196 j_coord_offsetD = DIM*jnrD;
198 /* load j atom coordinates */
199 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
200 x+j_coord_offsetC,x+j_coord_offsetD,
203 /* Calculate displacement vector */
204 dx00 = _mm_sub_ps(ix0,jx0);
205 dy00 = _mm_sub_ps(iy0,jy0);
206 dz00 = _mm_sub_ps(iz0,jz0);
207 dx10 = _mm_sub_ps(ix1,jx0);
208 dy10 = _mm_sub_ps(iy1,jy0);
209 dz10 = _mm_sub_ps(iz1,jz0);
210 dx20 = _mm_sub_ps(ix2,jx0);
211 dy20 = _mm_sub_ps(iy2,jy0);
212 dz20 = _mm_sub_ps(iz2,jz0);
213 dx30 = _mm_sub_ps(ix3,jx0);
214 dy30 = _mm_sub_ps(iy3,jy0);
215 dz30 = _mm_sub_ps(iz3,jz0);
217 /* Calculate squared distance and things based on it */
218 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
219 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
220 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
221 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
223 rinv00 = gmx_mm_invsqrt_ps(rsq00);
224 rinv10 = gmx_mm_invsqrt_ps(rsq10);
225 rinv20 = gmx_mm_invsqrt_ps(rsq20);
226 rinv30 = gmx_mm_invsqrt_ps(rsq30);
228 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
229 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
230 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
232 /* Load parameters for j particles */
233 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
234 charge+jnrC+0,charge+jnrD+0);
235 vdwjidx0A = 2*vdwtype[jnrA+0];
236 vdwjidx0B = 2*vdwtype[jnrB+0];
237 vdwjidx0C = 2*vdwtype[jnrC+0];
238 vdwjidx0D = 2*vdwtype[jnrD+0];
240 /**************************
241 * CALCULATE INTERACTIONS *
242 **************************/
244 r00 = _mm_mul_ps(rsq00,rinv00);
246 /* Compute parameters for interactions between i and j atoms */
247 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
248 vdwparam+vdwioffset0+vdwjidx0B,
249 vdwparam+vdwioffset0+vdwjidx0C,
250 vdwparam+vdwioffset0+vdwjidx0D,
253 /* Calculate table index by multiplying r with table scale and truncate to integer */
254 rt = _mm_mul_ps(r00,vftabscale);
255 vfitab = _mm_cvttps_epi32(rt);
256 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
257 vfitab = _mm_slli_epi32(vfitab,3);
259 /* CUBIC SPLINE TABLE DISPERSION */
260 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
261 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
262 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
263 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
264 _MM_TRANSPOSE4_PS(Y,F,G,H);
265 Heps = _mm_mul_ps(vfeps,H);
266 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
267 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
268 vvdw6 = _mm_mul_ps(c6_00,VV);
269 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
270 fvdw6 = _mm_mul_ps(c6_00,FF);
272 /* CUBIC SPLINE TABLE REPULSION */
273 vfitab = _mm_add_epi32(vfitab,ifour);
274 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
275 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
276 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
277 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
278 _MM_TRANSPOSE4_PS(Y,F,G,H);
279 Heps = _mm_mul_ps(vfeps,H);
280 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
281 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
282 vvdw12 = _mm_mul_ps(c12_00,VV);
283 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
284 fvdw12 = _mm_mul_ps(c12_00,FF);
285 vvdw = _mm_add_ps(vvdw12,vvdw6);
286 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
288 /* Update potential sum for this i atom from the interaction with this j atom. */
289 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
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);
309 /**************************
310 * CALCULATE INTERACTIONS *
311 **************************/
313 r10 = _mm_mul_ps(rsq10,rinv10);
315 /* Compute parameters for interactions between i and j atoms */
316 qq10 = _mm_mul_ps(iq1,jq0);
318 /* EWALD ELECTROSTATICS */
320 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
321 ewrt = _mm_mul_ps(r10,ewtabscale);
322 ewitab = _mm_cvttps_epi32(ewrt);
323 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
324 ewitab = _mm_slli_epi32(ewitab,2);
325 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
326 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
327 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
328 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
329 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
330 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
331 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
332 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
333 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
335 /* Update potential sum for this i atom from the interaction with this j atom. */
336 velecsum = _mm_add_ps(velecsum,velec);
340 /* Calculate temporary vectorial force */
341 tx = _mm_mul_ps(fscal,dx10);
342 ty = _mm_mul_ps(fscal,dy10);
343 tz = _mm_mul_ps(fscal,dz10);
345 /* Update vectorial force */
346 fix1 = _mm_add_ps(fix1,tx);
347 fiy1 = _mm_add_ps(fiy1,ty);
348 fiz1 = _mm_add_ps(fiz1,tz);
350 fjptrA = f+j_coord_offsetA;
351 fjptrB = f+j_coord_offsetB;
352 fjptrC = f+j_coord_offsetC;
353 fjptrD = f+j_coord_offsetD;
354 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
356 /**************************
357 * CALCULATE INTERACTIONS *
358 **************************/
360 r20 = _mm_mul_ps(rsq20,rinv20);
362 /* Compute parameters for interactions between i and j atoms */
363 qq20 = _mm_mul_ps(iq2,jq0);
365 /* EWALD ELECTROSTATICS */
367 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
368 ewrt = _mm_mul_ps(r20,ewtabscale);
369 ewitab = _mm_cvttps_epi32(ewrt);
370 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
371 ewitab = _mm_slli_epi32(ewitab,2);
372 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
373 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
374 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
375 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
376 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
377 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
378 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
379 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
380 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
382 /* Update potential sum for this i atom from the interaction with this j atom. */
383 velecsum = _mm_add_ps(velecsum,velec);
387 /* Calculate temporary vectorial force */
388 tx = _mm_mul_ps(fscal,dx20);
389 ty = _mm_mul_ps(fscal,dy20);
390 tz = _mm_mul_ps(fscal,dz20);
392 /* Update vectorial force */
393 fix2 = _mm_add_ps(fix2,tx);
394 fiy2 = _mm_add_ps(fiy2,ty);
395 fiz2 = _mm_add_ps(fiz2,tz);
397 fjptrA = f+j_coord_offsetA;
398 fjptrB = f+j_coord_offsetB;
399 fjptrC = f+j_coord_offsetC;
400 fjptrD = f+j_coord_offsetD;
401 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
403 /**************************
404 * CALCULATE INTERACTIONS *
405 **************************/
407 r30 = _mm_mul_ps(rsq30,rinv30);
409 /* Compute parameters for interactions between i and j atoms */
410 qq30 = _mm_mul_ps(iq3,jq0);
412 /* EWALD ELECTROSTATICS */
414 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
415 ewrt = _mm_mul_ps(r30,ewtabscale);
416 ewitab = _mm_cvttps_epi32(ewrt);
417 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
418 ewitab = _mm_slli_epi32(ewitab,2);
419 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
420 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
421 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
422 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
423 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
424 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
425 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
426 velec = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
427 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
429 /* Update potential sum for this i atom from the interaction with this j atom. */
430 velecsum = _mm_add_ps(velecsum,velec);
434 /* Calculate temporary vectorial force */
435 tx = _mm_mul_ps(fscal,dx30);
436 ty = _mm_mul_ps(fscal,dy30);
437 tz = _mm_mul_ps(fscal,dz30);
439 /* Update vectorial force */
440 fix3 = _mm_add_ps(fix3,tx);
441 fiy3 = _mm_add_ps(fiy3,ty);
442 fiz3 = _mm_add_ps(fiz3,tz);
444 fjptrA = f+j_coord_offsetA;
445 fjptrB = f+j_coord_offsetB;
446 fjptrC = f+j_coord_offsetC;
447 fjptrD = f+j_coord_offsetD;
448 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
450 /* Inner loop uses 179 flops */
456 /* Get j neighbor index, and coordinate index */
457 jnrlistA = jjnr[jidx];
458 jnrlistB = jjnr[jidx+1];
459 jnrlistC = jjnr[jidx+2];
460 jnrlistD = jjnr[jidx+3];
461 /* Sign of each element will be negative for non-real atoms.
462 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
463 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
465 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
466 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
467 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
468 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
469 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
470 j_coord_offsetA = DIM*jnrA;
471 j_coord_offsetB = DIM*jnrB;
472 j_coord_offsetC = DIM*jnrC;
473 j_coord_offsetD = DIM*jnrD;
475 /* load j atom coordinates */
476 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
477 x+j_coord_offsetC,x+j_coord_offsetD,
480 /* Calculate displacement vector */
481 dx00 = _mm_sub_ps(ix0,jx0);
482 dy00 = _mm_sub_ps(iy0,jy0);
483 dz00 = _mm_sub_ps(iz0,jz0);
484 dx10 = _mm_sub_ps(ix1,jx0);
485 dy10 = _mm_sub_ps(iy1,jy0);
486 dz10 = _mm_sub_ps(iz1,jz0);
487 dx20 = _mm_sub_ps(ix2,jx0);
488 dy20 = _mm_sub_ps(iy2,jy0);
489 dz20 = _mm_sub_ps(iz2,jz0);
490 dx30 = _mm_sub_ps(ix3,jx0);
491 dy30 = _mm_sub_ps(iy3,jy0);
492 dz30 = _mm_sub_ps(iz3,jz0);
494 /* Calculate squared distance and things based on it */
495 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
496 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
497 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
498 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
500 rinv00 = gmx_mm_invsqrt_ps(rsq00);
501 rinv10 = gmx_mm_invsqrt_ps(rsq10);
502 rinv20 = gmx_mm_invsqrt_ps(rsq20);
503 rinv30 = gmx_mm_invsqrt_ps(rsq30);
505 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
506 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
507 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
509 /* Load parameters for j particles */
510 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
511 charge+jnrC+0,charge+jnrD+0);
512 vdwjidx0A = 2*vdwtype[jnrA+0];
513 vdwjidx0B = 2*vdwtype[jnrB+0];
514 vdwjidx0C = 2*vdwtype[jnrC+0];
515 vdwjidx0D = 2*vdwtype[jnrD+0];
517 /**************************
518 * CALCULATE INTERACTIONS *
519 **************************/
521 r00 = _mm_mul_ps(rsq00,rinv00);
522 r00 = _mm_andnot_ps(dummy_mask,r00);
524 /* Compute parameters for interactions between i and j atoms */
525 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
526 vdwparam+vdwioffset0+vdwjidx0B,
527 vdwparam+vdwioffset0+vdwjidx0C,
528 vdwparam+vdwioffset0+vdwjidx0D,
531 /* Calculate table index by multiplying r with table scale and truncate to integer */
532 rt = _mm_mul_ps(r00,vftabscale);
533 vfitab = _mm_cvttps_epi32(rt);
534 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
535 vfitab = _mm_slli_epi32(vfitab,3);
537 /* CUBIC SPLINE TABLE DISPERSION */
538 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
539 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
540 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
541 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
542 _MM_TRANSPOSE4_PS(Y,F,G,H);
543 Heps = _mm_mul_ps(vfeps,H);
544 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
545 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
546 vvdw6 = _mm_mul_ps(c6_00,VV);
547 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
548 fvdw6 = _mm_mul_ps(c6_00,FF);
550 /* CUBIC SPLINE TABLE REPULSION */
551 vfitab = _mm_add_epi32(vfitab,ifour);
552 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
553 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
554 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
555 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
556 _MM_TRANSPOSE4_PS(Y,F,G,H);
557 Heps = _mm_mul_ps(vfeps,H);
558 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
559 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
560 vvdw12 = _mm_mul_ps(c12_00,VV);
561 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
562 fvdw12 = _mm_mul_ps(c12_00,FF);
563 vvdw = _mm_add_ps(vvdw12,vvdw6);
564 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
566 /* Update potential sum for this i atom from the interaction with this j atom. */
567 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
568 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
572 fscal = _mm_andnot_ps(dummy_mask,fscal);
574 /* Calculate temporary vectorial force */
575 tx = _mm_mul_ps(fscal,dx00);
576 ty = _mm_mul_ps(fscal,dy00);
577 tz = _mm_mul_ps(fscal,dz00);
579 /* Update vectorial force */
580 fix0 = _mm_add_ps(fix0,tx);
581 fiy0 = _mm_add_ps(fiy0,ty);
582 fiz0 = _mm_add_ps(fiz0,tz);
584 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
585 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
586 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
587 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
588 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
590 /**************************
591 * CALCULATE INTERACTIONS *
592 **************************/
594 r10 = _mm_mul_ps(rsq10,rinv10);
595 r10 = _mm_andnot_ps(dummy_mask,r10);
597 /* Compute parameters for interactions between i and j atoms */
598 qq10 = _mm_mul_ps(iq1,jq0);
600 /* EWALD ELECTROSTATICS */
602 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
603 ewrt = _mm_mul_ps(r10,ewtabscale);
604 ewitab = _mm_cvttps_epi32(ewrt);
605 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
606 ewitab = _mm_slli_epi32(ewitab,2);
607 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
608 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
609 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
610 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
611 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
612 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
613 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
614 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
615 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
617 /* Update potential sum for this i atom from the interaction with this j atom. */
618 velec = _mm_andnot_ps(dummy_mask,velec);
619 velecsum = _mm_add_ps(velecsum,velec);
623 fscal = _mm_andnot_ps(dummy_mask,fscal);
625 /* Calculate temporary vectorial force */
626 tx = _mm_mul_ps(fscal,dx10);
627 ty = _mm_mul_ps(fscal,dy10);
628 tz = _mm_mul_ps(fscal,dz10);
630 /* Update vectorial force */
631 fix1 = _mm_add_ps(fix1,tx);
632 fiy1 = _mm_add_ps(fiy1,ty);
633 fiz1 = _mm_add_ps(fiz1,tz);
635 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
636 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
637 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
638 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
639 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
641 /**************************
642 * CALCULATE INTERACTIONS *
643 **************************/
645 r20 = _mm_mul_ps(rsq20,rinv20);
646 r20 = _mm_andnot_ps(dummy_mask,r20);
648 /* Compute parameters for interactions between i and j atoms */
649 qq20 = _mm_mul_ps(iq2,jq0);
651 /* EWALD ELECTROSTATICS */
653 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
654 ewrt = _mm_mul_ps(r20,ewtabscale);
655 ewitab = _mm_cvttps_epi32(ewrt);
656 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
657 ewitab = _mm_slli_epi32(ewitab,2);
658 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
659 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
660 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
661 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
662 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
663 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
664 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
665 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
666 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
668 /* Update potential sum for this i atom from the interaction with this j atom. */
669 velec = _mm_andnot_ps(dummy_mask,velec);
670 velecsum = _mm_add_ps(velecsum,velec);
674 fscal = _mm_andnot_ps(dummy_mask,fscal);
676 /* Calculate temporary vectorial force */
677 tx = _mm_mul_ps(fscal,dx20);
678 ty = _mm_mul_ps(fscal,dy20);
679 tz = _mm_mul_ps(fscal,dz20);
681 /* Update vectorial force */
682 fix2 = _mm_add_ps(fix2,tx);
683 fiy2 = _mm_add_ps(fiy2,ty);
684 fiz2 = _mm_add_ps(fiz2,tz);
686 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
687 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
688 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
689 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
690 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
692 /**************************
693 * CALCULATE INTERACTIONS *
694 **************************/
696 r30 = _mm_mul_ps(rsq30,rinv30);
697 r30 = _mm_andnot_ps(dummy_mask,r30);
699 /* Compute parameters for interactions between i and j atoms */
700 qq30 = _mm_mul_ps(iq3,jq0);
702 /* EWALD ELECTROSTATICS */
704 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
705 ewrt = _mm_mul_ps(r30,ewtabscale);
706 ewitab = _mm_cvttps_epi32(ewrt);
707 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
708 ewitab = _mm_slli_epi32(ewitab,2);
709 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
710 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
711 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
712 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
713 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
714 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
715 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
716 velec = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
717 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
719 /* Update potential sum for this i atom from the interaction with this j atom. */
720 velec = _mm_andnot_ps(dummy_mask,velec);
721 velecsum = _mm_add_ps(velecsum,velec);
725 fscal = _mm_andnot_ps(dummy_mask,fscal);
727 /* Calculate temporary vectorial force */
728 tx = _mm_mul_ps(fscal,dx30);
729 ty = _mm_mul_ps(fscal,dy30);
730 tz = _mm_mul_ps(fscal,dz30);
732 /* Update vectorial force */
733 fix3 = _mm_add_ps(fix3,tx);
734 fiy3 = _mm_add_ps(fiy3,ty);
735 fiz3 = _mm_add_ps(fiz3,tz);
737 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
738 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
739 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
740 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
741 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
743 /* Inner loop uses 183 flops */
746 /* End of innermost loop */
748 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
749 f+i_coord_offset,fshift+i_shift_offset);
752 /* Update potential energies */
753 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
754 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
756 /* Increment number of inner iterations */
757 inneriter += j_index_end - j_index_start;
759 /* Outer loop uses 26 flops */
762 /* Increment number of outer iterations */
765 /* Update outer/inner flops */
767 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*183);
770 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse4_1_single
771 * Electrostatics interaction: Ewald
772 * VdW interaction: CubicSplineTable
773 * Geometry: Water4-Particle
774 * Calculate force/pot: Force
777 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse4_1_single
778 (t_nblist * gmx_restrict nlist,
779 rvec * gmx_restrict xx,
780 rvec * gmx_restrict ff,
781 t_forcerec * gmx_restrict fr,
782 t_mdatoms * gmx_restrict mdatoms,
783 nb_kernel_data_t * gmx_restrict kernel_data,
784 t_nrnb * gmx_restrict nrnb)
786 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
787 * just 0 for non-waters.
788 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
789 * jnr indices corresponding to data put in the four positions in the SIMD register.
791 int i_shift_offset,i_coord_offset,outeriter,inneriter;
792 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
793 int jnrA,jnrB,jnrC,jnrD;
794 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
795 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
796 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
798 real *shiftvec,*fshift,*x,*f;
799 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
801 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
803 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
805 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
807 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
809 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
810 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
811 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
812 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
813 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
814 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
815 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
816 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
819 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
822 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
823 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
825 __m128i ifour = _mm_set1_epi32(4);
826 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
829 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
831 __m128 dummy_mask,cutoff_mask;
832 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
833 __m128 one = _mm_set1_ps(1.0);
834 __m128 two = _mm_set1_ps(2.0);
840 jindex = nlist->jindex;
842 shiftidx = nlist->shift;
844 shiftvec = fr->shift_vec[0];
845 fshift = fr->fshift[0];
846 facel = _mm_set1_ps(fr->epsfac);
847 charge = mdatoms->chargeA;
848 nvdwtype = fr->ntype;
850 vdwtype = mdatoms->typeA;
852 vftab = kernel_data->table_vdw->data;
853 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
855 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
856 ewtab = fr->ic->tabq_coul_F;
857 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
858 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
860 /* Setup water-specific parameters */
861 inr = nlist->iinr[0];
862 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
863 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
864 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
865 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
867 /* Avoid stupid compiler warnings */
868 jnrA = jnrB = jnrC = jnrD = 0;
877 for(iidx=0;iidx<4*DIM;iidx++)
882 /* Start outer loop over neighborlists */
883 for(iidx=0; iidx<nri; iidx++)
885 /* Load shift vector for this list */
886 i_shift_offset = DIM*shiftidx[iidx];
888 /* Load limits for loop over neighbors */
889 j_index_start = jindex[iidx];
890 j_index_end = jindex[iidx+1];
892 /* Get outer coordinate index */
894 i_coord_offset = DIM*inr;
896 /* Load i particle coords and add shift vector */
897 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
898 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
900 fix0 = _mm_setzero_ps();
901 fiy0 = _mm_setzero_ps();
902 fiz0 = _mm_setzero_ps();
903 fix1 = _mm_setzero_ps();
904 fiy1 = _mm_setzero_ps();
905 fiz1 = _mm_setzero_ps();
906 fix2 = _mm_setzero_ps();
907 fiy2 = _mm_setzero_ps();
908 fiz2 = _mm_setzero_ps();
909 fix3 = _mm_setzero_ps();
910 fiy3 = _mm_setzero_ps();
911 fiz3 = _mm_setzero_ps();
913 /* Start inner kernel loop */
914 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
917 /* Get j neighbor index, and coordinate index */
922 j_coord_offsetA = DIM*jnrA;
923 j_coord_offsetB = DIM*jnrB;
924 j_coord_offsetC = DIM*jnrC;
925 j_coord_offsetD = DIM*jnrD;
927 /* load j atom coordinates */
928 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
929 x+j_coord_offsetC,x+j_coord_offsetD,
932 /* Calculate displacement vector */
933 dx00 = _mm_sub_ps(ix0,jx0);
934 dy00 = _mm_sub_ps(iy0,jy0);
935 dz00 = _mm_sub_ps(iz0,jz0);
936 dx10 = _mm_sub_ps(ix1,jx0);
937 dy10 = _mm_sub_ps(iy1,jy0);
938 dz10 = _mm_sub_ps(iz1,jz0);
939 dx20 = _mm_sub_ps(ix2,jx0);
940 dy20 = _mm_sub_ps(iy2,jy0);
941 dz20 = _mm_sub_ps(iz2,jz0);
942 dx30 = _mm_sub_ps(ix3,jx0);
943 dy30 = _mm_sub_ps(iy3,jy0);
944 dz30 = _mm_sub_ps(iz3,jz0);
946 /* Calculate squared distance and things based on it */
947 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
948 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
949 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
950 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
952 rinv00 = gmx_mm_invsqrt_ps(rsq00);
953 rinv10 = gmx_mm_invsqrt_ps(rsq10);
954 rinv20 = gmx_mm_invsqrt_ps(rsq20);
955 rinv30 = gmx_mm_invsqrt_ps(rsq30);
957 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
958 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
959 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
961 /* Load parameters for j particles */
962 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
963 charge+jnrC+0,charge+jnrD+0);
964 vdwjidx0A = 2*vdwtype[jnrA+0];
965 vdwjidx0B = 2*vdwtype[jnrB+0];
966 vdwjidx0C = 2*vdwtype[jnrC+0];
967 vdwjidx0D = 2*vdwtype[jnrD+0];
969 /**************************
970 * CALCULATE INTERACTIONS *
971 **************************/
973 r00 = _mm_mul_ps(rsq00,rinv00);
975 /* Compute parameters for interactions between i and j atoms */
976 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
977 vdwparam+vdwioffset0+vdwjidx0B,
978 vdwparam+vdwioffset0+vdwjidx0C,
979 vdwparam+vdwioffset0+vdwjidx0D,
982 /* Calculate table index by multiplying r with table scale and truncate to integer */
983 rt = _mm_mul_ps(r00,vftabscale);
984 vfitab = _mm_cvttps_epi32(rt);
985 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
986 vfitab = _mm_slli_epi32(vfitab,3);
988 /* CUBIC SPLINE TABLE DISPERSION */
989 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
990 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
991 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
992 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
993 _MM_TRANSPOSE4_PS(Y,F,G,H);
994 Heps = _mm_mul_ps(vfeps,H);
995 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
996 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
997 fvdw6 = _mm_mul_ps(c6_00,FF);
999 /* CUBIC SPLINE TABLE REPULSION */
1000 vfitab = _mm_add_epi32(vfitab,ifour);
1001 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1002 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1003 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1004 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1005 _MM_TRANSPOSE4_PS(Y,F,G,H);
1006 Heps = _mm_mul_ps(vfeps,H);
1007 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1008 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1009 fvdw12 = _mm_mul_ps(c12_00,FF);
1010 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1014 /* Calculate temporary vectorial force */
1015 tx = _mm_mul_ps(fscal,dx00);
1016 ty = _mm_mul_ps(fscal,dy00);
1017 tz = _mm_mul_ps(fscal,dz00);
1019 /* Update vectorial force */
1020 fix0 = _mm_add_ps(fix0,tx);
1021 fiy0 = _mm_add_ps(fiy0,ty);
1022 fiz0 = _mm_add_ps(fiz0,tz);
1024 fjptrA = f+j_coord_offsetA;
1025 fjptrB = f+j_coord_offsetB;
1026 fjptrC = f+j_coord_offsetC;
1027 fjptrD = f+j_coord_offsetD;
1028 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1030 /**************************
1031 * CALCULATE INTERACTIONS *
1032 **************************/
1034 r10 = _mm_mul_ps(rsq10,rinv10);
1036 /* Compute parameters for interactions between i and j atoms */
1037 qq10 = _mm_mul_ps(iq1,jq0);
1039 /* EWALD ELECTROSTATICS */
1041 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1042 ewrt = _mm_mul_ps(r10,ewtabscale);
1043 ewitab = _mm_cvttps_epi32(ewrt);
1044 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1045 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1046 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1048 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1049 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1053 /* Calculate temporary vectorial force */
1054 tx = _mm_mul_ps(fscal,dx10);
1055 ty = _mm_mul_ps(fscal,dy10);
1056 tz = _mm_mul_ps(fscal,dz10);
1058 /* Update vectorial force */
1059 fix1 = _mm_add_ps(fix1,tx);
1060 fiy1 = _mm_add_ps(fiy1,ty);
1061 fiz1 = _mm_add_ps(fiz1,tz);
1063 fjptrA = f+j_coord_offsetA;
1064 fjptrB = f+j_coord_offsetB;
1065 fjptrC = f+j_coord_offsetC;
1066 fjptrD = f+j_coord_offsetD;
1067 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1069 /**************************
1070 * CALCULATE INTERACTIONS *
1071 **************************/
1073 r20 = _mm_mul_ps(rsq20,rinv20);
1075 /* Compute parameters for interactions between i and j atoms */
1076 qq20 = _mm_mul_ps(iq2,jq0);
1078 /* EWALD ELECTROSTATICS */
1080 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1081 ewrt = _mm_mul_ps(r20,ewtabscale);
1082 ewitab = _mm_cvttps_epi32(ewrt);
1083 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1084 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1085 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1087 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1088 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1092 /* Calculate temporary vectorial force */
1093 tx = _mm_mul_ps(fscal,dx20);
1094 ty = _mm_mul_ps(fscal,dy20);
1095 tz = _mm_mul_ps(fscal,dz20);
1097 /* Update vectorial force */
1098 fix2 = _mm_add_ps(fix2,tx);
1099 fiy2 = _mm_add_ps(fiy2,ty);
1100 fiz2 = _mm_add_ps(fiz2,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);
1108 /**************************
1109 * CALCULATE INTERACTIONS *
1110 **************************/
1112 r30 = _mm_mul_ps(rsq30,rinv30);
1114 /* Compute parameters for interactions between i and j atoms */
1115 qq30 = _mm_mul_ps(iq3,jq0);
1117 /* EWALD ELECTROSTATICS */
1119 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120 ewrt = _mm_mul_ps(r30,ewtabscale);
1121 ewitab = _mm_cvttps_epi32(ewrt);
1122 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1123 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1124 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1126 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1127 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1131 /* Calculate temporary vectorial force */
1132 tx = _mm_mul_ps(fscal,dx30);
1133 ty = _mm_mul_ps(fscal,dy30);
1134 tz = _mm_mul_ps(fscal,dz30);
1136 /* Update vectorial force */
1137 fix3 = _mm_add_ps(fix3,tx);
1138 fiy3 = _mm_add_ps(fiy3,ty);
1139 fiz3 = _mm_add_ps(fiz3,tz);
1141 fjptrA = f+j_coord_offsetA;
1142 fjptrB = f+j_coord_offsetB;
1143 fjptrC = f+j_coord_offsetC;
1144 fjptrD = f+j_coord_offsetD;
1145 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1147 /* Inner loop uses 156 flops */
1150 if(jidx<j_index_end)
1153 /* Get j neighbor index, and coordinate index */
1154 jnrlistA = jjnr[jidx];
1155 jnrlistB = jjnr[jidx+1];
1156 jnrlistC = jjnr[jidx+2];
1157 jnrlistD = jjnr[jidx+3];
1158 /* Sign of each element will be negative for non-real atoms.
1159 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1160 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1162 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1163 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1164 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1165 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1166 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1167 j_coord_offsetA = DIM*jnrA;
1168 j_coord_offsetB = DIM*jnrB;
1169 j_coord_offsetC = DIM*jnrC;
1170 j_coord_offsetD = DIM*jnrD;
1172 /* load j atom coordinates */
1173 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1174 x+j_coord_offsetC,x+j_coord_offsetD,
1177 /* Calculate displacement vector */
1178 dx00 = _mm_sub_ps(ix0,jx0);
1179 dy00 = _mm_sub_ps(iy0,jy0);
1180 dz00 = _mm_sub_ps(iz0,jz0);
1181 dx10 = _mm_sub_ps(ix1,jx0);
1182 dy10 = _mm_sub_ps(iy1,jy0);
1183 dz10 = _mm_sub_ps(iz1,jz0);
1184 dx20 = _mm_sub_ps(ix2,jx0);
1185 dy20 = _mm_sub_ps(iy2,jy0);
1186 dz20 = _mm_sub_ps(iz2,jz0);
1187 dx30 = _mm_sub_ps(ix3,jx0);
1188 dy30 = _mm_sub_ps(iy3,jy0);
1189 dz30 = _mm_sub_ps(iz3,jz0);
1191 /* Calculate squared distance and things based on it */
1192 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1193 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1194 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1195 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1197 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1198 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1199 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1200 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1202 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1203 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1204 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1206 /* Load parameters for j particles */
1207 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1208 charge+jnrC+0,charge+jnrD+0);
1209 vdwjidx0A = 2*vdwtype[jnrA+0];
1210 vdwjidx0B = 2*vdwtype[jnrB+0];
1211 vdwjidx0C = 2*vdwtype[jnrC+0];
1212 vdwjidx0D = 2*vdwtype[jnrD+0];
1214 /**************************
1215 * CALCULATE INTERACTIONS *
1216 **************************/
1218 r00 = _mm_mul_ps(rsq00,rinv00);
1219 r00 = _mm_andnot_ps(dummy_mask,r00);
1221 /* Compute parameters for interactions between i and j atoms */
1222 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1223 vdwparam+vdwioffset0+vdwjidx0B,
1224 vdwparam+vdwioffset0+vdwjidx0C,
1225 vdwparam+vdwioffset0+vdwjidx0D,
1228 /* Calculate table index by multiplying r with table scale and truncate to integer */
1229 rt = _mm_mul_ps(r00,vftabscale);
1230 vfitab = _mm_cvttps_epi32(rt);
1231 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1232 vfitab = _mm_slli_epi32(vfitab,3);
1234 /* CUBIC SPLINE TABLE DISPERSION */
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 fvdw6 = _mm_mul_ps(c6_00,FF);
1245 /* CUBIC SPLINE TABLE REPULSION */
1246 vfitab = _mm_add_epi32(vfitab,ifour);
1247 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1248 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1249 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1250 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1251 _MM_TRANSPOSE4_PS(Y,F,G,H);
1252 Heps = _mm_mul_ps(vfeps,H);
1253 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1254 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1255 fvdw12 = _mm_mul_ps(c12_00,FF);
1256 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1260 fscal = _mm_andnot_ps(dummy_mask,fscal);
1262 /* Calculate temporary vectorial force */
1263 tx = _mm_mul_ps(fscal,dx00);
1264 ty = _mm_mul_ps(fscal,dy00);
1265 tz = _mm_mul_ps(fscal,dz00);
1267 /* Update vectorial force */
1268 fix0 = _mm_add_ps(fix0,tx);
1269 fiy0 = _mm_add_ps(fiy0,ty);
1270 fiz0 = _mm_add_ps(fiz0,tz);
1272 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1273 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1274 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1275 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1276 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1278 /**************************
1279 * CALCULATE INTERACTIONS *
1280 **************************/
1282 r10 = _mm_mul_ps(rsq10,rinv10);
1283 r10 = _mm_andnot_ps(dummy_mask,r10);
1285 /* Compute parameters for interactions between i and j atoms */
1286 qq10 = _mm_mul_ps(iq1,jq0);
1288 /* EWALD ELECTROSTATICS */
1290 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1291 ewrt = _mm_mul_ps(r10,ewtabscale);
1292 ewitab = _mm_cvttps_epi32(ewrt);
1293 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1294 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1295 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1297 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1298 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1302 fscal = _mm_andnot_ps(dummy_mask,fscal);
1304 /* Calculate temporary vectorial force */
1305 tx = _mm_mul_ps(fscal,dx10);
1306 ty = _mm_mul_ps(fscal,dy10);
1307 tz = _mm_mul_ps(fscal,dz10);
1309 /* Update vectorial force */
1310 fix1 = _mm_add_ps(fix1,tx);
1311 fiy1 = _mm_add_ps(fiy1,ty);
1312 fiz1 = _mm_add_ps(fiz1,tz);
1314 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1315 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1316 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1317 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1318 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1320 /**************************
1321 * CALCULATE INTERACTIONS *
1322 **************************/
1324 r20 = _mm_mul_ps(rsq20,rinv20);
1325 r20 = _mm_andnot_ps(dummy_mask,r20);
1327 /* Compute parameters for interactions between i and j atoms */
1328 qq20 = _mm_mul_ps(iq2,jq0);
1330 /* EWALD ELECTROSTATICS */
1332 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1333 ewrt = _mm_mul_ps(r20,ewtabscale);
1334 ewitab = _mm_cvttps_epi32(ewrt);
1335 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1336 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1337 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1339 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1340 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1344 fscal = _mm_andnot_ps(dummy_mask,fscal);
1346 /* Calculate temporary vectorial force */
1347 tx = _mm_mul_ps(fscal,dx20);
1348 ty = _mm_mul_ps(fscal,dy20);
1349 tz = _mm_mul_ps(fscal,dz20);
1351 /* Update vectorial force */
1352 fix2 = _mm_add_ps(fix2,tx);
1353 fiy2 = _mm_add_ps(fiy2,ty);
1354 fiz2 = _mm_add_ps(fiz2,tz);
1356 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1357 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1358 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1359 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1360 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1362 /**************************
1363 * CALCULATE INTERACTIONS *
1364 **************************/
1366 r30 = _mm_mul_ps(rsq30,rinv30);
1367 r30 = _mm_andnot_ps(dummy_mask,r30);
1369 /* Compute parameters for interactions between i and j atoms */
1370 qq30 = _mm_mul_ps(iq3,jq0);
1372 /* EWALD ELECTROSTATICS */
1374 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1375 ewrt = _mm_mul_ps(r30,ewtabscale);
1376 ewitab = _mm_cvttps_epi32(ewrt);
1377 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1378 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1379 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1381 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1382 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1386 fscal = _mm_andnot_ps(dummy_mask,fscal);
1388 /* Calculate temporary vectorial force */
1389 tx = _mm_mul_ps(fscal,dx30);
1390 ty = _mm_mul_ps(fscal,dy30);
1391 tz = _mm_mul_ps(fscal,dz30);
1393 /* Update vectorial force */
1394 fix3 = _mm_add_ps(fix3,tx);
1395 fiy3 = _mm_add_ps(fiy3,ty);
1396 fiz3 = _mm_add_ps(fiz3,tz);
1398 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1399 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1400 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1401 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1402 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1404 /* Inner loop uses 160 flops */
1407 /* End of innermost loop */
1409 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1410 f+i_coord_offset,fshift+i_shift_offset);
1412 /* Increment number of inner iterations */
1413 inneriter += j_index_end - j_index_start;
1415 /* Outer loop uses 24 flops */
1418 /* Increment number of outer iterations */
1421 /* Update outer/inner flops */
1423 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*160);