2 * Note: this file was generated by the Gromacs avx_128_fma_double 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_avx_128_fma_double.h"
34 #include "kernelutil_x86_avx_128_fma_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
80 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
83 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
84 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
86 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
88 __m128d dummy_mask,cutoff_mask;
89 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
90 __m128d one = _mm_set1_pd(1.0);
91 __m128d two = _mm_set1_pd(2.0);
97 jindex = nlist->jindex;
99 shiftidx = nlist->shift;
101 shiftvec = fr->shift_vec[0];
102 fshift = fr->fshift[0];
103 facel = _mm_set1_pd(fr->epsfac);
104 charge = mdatoms->chargeA;
105 nvdwtype = fr->ntype;
107 vdwtype = mdatoms->typeA;
109 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
110 ewtab = fr->ic->tabq_coul_FDV0;
111 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
112 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
114 /* Setup water-specific parameters */
115 inr = nlist->iinr[0];
116 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
117 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
118 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
119 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
121 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
122 rcutoff_scalar = fr->rcoulomb;
123 rcutoff = _mm_set1_pd(rcutoff_scalar);
124 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
126 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
127 rvdw = _mm_set1_pd(fr->rvdw);
129 /* Avoid stupid compiler warnings */
137 /* Start outer loop over neighborlists */
138 for(iidx=0; iidx<nri; iidx++)
140 /* Load shift vector for this list */
141 i_shift_offset = DIM*shiftidx[iidx];
143 /* Load limits for loop over neighbors */
144 j_index_start = jindex[iidx];
145 j_index_end = jindex[iidx+1];
147 /* Get outer coordinate index */
149 i_coord_offset = DIM*inr;
151 /* Load i particle coords and add shift vector */
152 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
153 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
155 fix0 = _mm_setzero_pd();
156 fiy0 = _mm_setzero_pd();
157 fiz0 = _mm_setzero_pd();
158 fix1 = _mm_setzero_pd();
159 fiy1 = _mm_setzero_pd();
160 fiz1 = _mm_setzero_pd();
161 fix2 = _mm_setzero_pd();
162 fiy2 = _mm_setzero_pd();
163 fiz2 = _mm_setzero_pd();
165 /* Reset potential sums */
166 velecsum = _mm_setzero_pd();
167 vvdwsum = _mm_setzero_pd();
169 /* Start inner kernel loop */
170 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
173 /* Get j neighbor index, and coordinate index */
176 j_coord_offsetA = DIM*jnrA;
177 j_coord_offsetB = DIM*jnrB;
179 /* load j atom coordinates */
180 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
183 /* Calculate displacement vector */
184 dx00 = _mm_sub_pd(ix0,jx0);
185 dy00 = _mm_sub_pd(iy0,jy0);
186 dz00 = _mm_sub_pd(iz0,jz0);
187 dx10 = _mm_sub_pd(ix1,jx0);
188 dy10 = _mm_sub_pd(iy1,jy0);
189 dz10 = _mm_sub_pd(iz1,jz0);
190 dx20 = _mm_sub_pd(ix2,jx0);
191 dy20 = _mm_sub_pd(iy2,jy0);
192 dz20 = _mm_sub_pd(iz2,jz0);
194 /* Calculate squared distance and things based on it */
195 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
196 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
197 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
199 rinv00 = gmx_mm_invsqrt_pd(rsq00);
200 rinv10 = gmx_mm_invsqrt_pd(rsq10);
201 rinv20 = gmx_mm_invsqrt_pd(rsq20);
203 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
204 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
205 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
207 /* Load parameters for j particles */
208 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
209 vdwjidx0A = 2*vdwtype[jnrA+0];
210 vdwjidx0B = 2*vdwtype[jnrB+0];
212 fjx0 = _mm_setzero_pd();
213 fjy0 = _mm_setzero_pd();
214 fjz0 = _mm_setzero_pd();
216 /**************************
217 * CALCULATE INTERACTIONS *
218 **************************/
220 if (gmx_mm_any_lt(rsq00,rcutoff2))
223 r00 = _mm_mul_pd(rsq00,rinv00);
225 /* Compute parameters for interactions between i and j atoms */
226 qq00 = _mm_mul_pd(iq0,jq0);
227 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
228 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
230 /* EWALD ELECTROSTATICS */
232 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
233 ewrt = _mm_mul_pd(r00,ewtabscale);
234 ewitab = _mm_cvttpd_epi32(ewrt);
236 eweps = _mm_frcz_pd(ewrt);
238 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
240 twoeweps = _mm_add_pd(eweps,eweps);
241 ewitab = _mm_slli_epi32(ewitab,2);
242 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
243 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
244 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
245 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
246 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
247 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
248 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
249 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
250 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
251 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
253 /* LENNARD-JONES DISPERSION/REPULSION */
255 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
256 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
257 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
258 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
259 _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
260 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
262 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
264 /* Update potential sum for this i atom from the interaction with this j atom. */
265 velec = _mm_and_pd(velec,cutoff_mask);
266 velecsum = _mm_add_pd(velecsum,velec);
267 vvdw = _mm_and_pd(vvdw,cutoff_mask);
268 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
270 fscal = _mm_add_pd(felec,fvdw);
272 fscal = _mm_and_pd(fscal,cutoff_mask);
274 /* Update vectorial force */
275 fix0 = _mm_macc_pd(dx00,fscal,fix0);
276 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
277 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
279 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
280 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
281 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
285 /**************************
286 * CALCULATE INTERACTIONS *
287 **************************/
289 if (gmx_mm_any_lt(rsq10,rcutoff2))
292 r10 = _mm_mul_pd(rsq10,rinv10);
294 /* Compute parameters for interactions between i and j atoms */
295 qq10 = _mm_mul_pd(iq1,jq0);
297 /* EWALD ELECTROSTATICS */
299 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
300 ewrt = _mm_mul_pd(r10,ewtabscale);
301 ewitab = _mm_cvttpd_epi32(ewrt);
303 eweps = _mm_frcz_pd(ewrt);
305 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
307 twoeweps = _mm_add_pd(eweps,eweps);
308 ewitab = _mm_slli_epi32(ewitab,2);
309 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
310 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
311 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
312 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
313 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
314 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
315 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
316 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
317 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
318 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
320 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
322 /* Update potential sum for this i atom from the interaction with this j atom. */
323 velec = _mm_and_pd(velec,cutoff_mask);
324 velecsum = _mm_add_pd(velecsum,velec);
328 fscal = _mm_and_pd(fscal,cutoff_mask);
330 /* Update vectorial force */
331 fix1 = _mm_macc_pd(dx10,fscal,fix1);
332 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
333 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
335 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
336 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
337 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
341 /**************************
342 * CALCULATE INTERACTIONS *
343 **************************/
345 if (gmx_mm_any_lt(rsq20,rcutoff2))
348 r20 = _mm_mul_pd(rsq20,rinv20);
350 /* Compute parameters for interactions between i and j atoms */
351 qq20 = _mm_mul_pd(iq2,jq0);
353 /* EWALD ELECTROSTATICS */
355 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
356 ewrt = _mm_mul_pd(r20,ewtabscale);
357 ewitab = _mm_cvttpd_epi32(ewrt);
359 eweps = _mm_frcz_pd(ewrt);
361 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
363 twoeweps = _mm_add_pd(eweps,eweps);
364 ewitab = _mm_slli_epi32(ewitab,2);
365 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
366 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
367 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
368 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
369 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
370 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
371 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
372 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
373 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
374 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
376 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
378 /* Update potential sum for this i atom from the interaction with this j atom. */
379 velec = _mm_and_pd(velec,cutoff_mask);
380 velecsum = _mm_add_pd(velecsum,velec);
384 fscal = _mm_and_pd(fscal,cutoff_mask);
386 /* Update vectorial force */
387 fix2 = _mm_macc_pd(dx20,fscal,fix2);
388 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
389 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
391 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
392 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
393 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
397 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
399 /* Inner loop uses 168 flops */
406 j_coord_offsetA = DIM*jnrA;
408 /* load j atom coordinates */
409 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
412 /* Calculate displacement vector */
413 dx00 = _mm_sub_pd(ix0,jx0);
414 dy00 = _mm_sub_pd(iy0,jy0);
415 dz00 = _mm_sub_pd(iz0,jz0);
416 dx10 = _mm_sub_pd(ix1,jx0);
417 dy10 = _mm_sub_pd(iy1,jy0);
418 dz10 = _mm_sub_pd(iz1,jz0);
419 dx20 = _mm_sub_pd(ix2,jx0);
420 dy20 = _mm_sub_pd(iy2,jy0);
421 dz20 = _mm_sub_pd(iz2,jz0);
423 /* Calculate squared distance and things based on it */
424 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
425 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
426 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
428 rinv00 = gmx_mm_invsqrt_pd(rsq00);
429 rinv10 = gmx_mm_invsqrt_pd(rsq10);
430 rinv20 = gmx_mm_invsqrt_pd(rsq20);
432 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
433 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
434 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
436 /* Load parameters for j particles */
437 jq0 = _mm_load_sd(charge+jnrA+0);
438 vdwjidx0A = 2*vdwtype[jnrA+0];
440 fjx0 = _mm_setzero_pd();
441 fjy0 = _mm_setzero_pd();
442 fjz0 = _mm_setzero_pd();
444 /**************************
445 * CALCULATE INTERACTIONS *
446 **************************/
448 if (gmx_mm_any_lt(rsq00,rcutoff2))
451 r00 = _mm_mul_pd(rsq00,rinv00);
453 /* Compute parameters for interactions between i and j atoms */
454 qq00 = _mm_mul_pd(iq0,jq0);
455 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
457 /* EWALD ELECTROSTATICS */
459 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
460 ewrt = _mm_mul_pd(r00,ewtabscale);
461 ewitab = _mm_cvttpd_epi32(ewrt);
463 eweps = _mm_frcz_pd(ewrt);
465 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
467 twoeweps = _mm_add_pd(eweps,eweps);
468 ewitab = _mm_slli_epi32(ewitab,2);
469 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
470 ewtabD = _mm_setzero_pd();
471 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
472 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
473 ewtabFn = _mm_setzero_pd();
474 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
475 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
476 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
477 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
478 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
480 /* LENNARD-JONES DISPERSION/REPULSION */
482 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
483 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
484 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
485 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
486 _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
487 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
489 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
491 /* Update potential sum for this i atom from the interaction with this j atom. */
492 velec = _mm_and_pd(velec,cutoff_mask);
493 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
494 velecsum = _mm_add_pd(velecsum,velec);
495 vvdw = _mm_and_pd(vvdw,cutoff_mask);
496 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
497 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
499 fscal = _mm_add_pd(felec,fvdw);
501 fscal = _mm_and_pd(fscal,cutoff_mask);
503 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
505 /* Update vectorial force */
506 fix0 = _mm_macc_pd(dx00,fscal,fix0);
507 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
508 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
510 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
511 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
512 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
516 /**************************
517 * CALCULATE INTERACTIONS *
518 **************************/
520 if (gmx_mm_any_lt(rsq10,rcutoff2))
523 r10 = _mm_mul_pd(rsq10,rinv10);
525 /* Compute parameters for interactions between i and j atoms */
526 qq10 = _mm_mul_pd(iq1,jq0);
528 /* EWALD ELECTROSTATICS */
530 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
531 ewrt = _mm_mul_pd(r10,ewtabscale);
532 ewitab = _mm_cvttpd_epi32(ewrt);
534 eweps = _mm_frcz_pd(ewrt);
536 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
538 twoeweps = _mm_add_pd(eweps,eweps);
539 ewitab = _mm_slli_epi32(ewitab,2);
540 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
541 ewtabD = _mm_setzero_pd();
542 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
543 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
544 ewtabFn = _mm_setzero_pd();
545 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
546 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
547 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
548 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
549 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
551 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
553 /* Update potential sum for this i atom from the interaction with this j atom. */
554 velec = _mm_and_pd(velec,cutoff_mask);
555 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
556 velecsum = _mm_add_pd(velecsum,velec);
560 fscal = _mm_and_pd(fscal,cutoff_mask);
562 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
564 /* Update vectorial force */
565 fix1 = _mm_macc_pd(dx10,fscal,fix1);
566 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
567 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
569 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
570 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
571 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
575 /**************************
576 * CALCULATE INTERACTIONS *
577 **************************/
579 if (gmx_mm_any_lt(rsq20,rcutoff2))
582 r20 = _mm_mul_pd(rsq20,rinv20);
584 /* Compute parameters for interactions between i and j atoms */
585 qq20 = _mm_mul_pd(iq2,jq0);
587 /* EWALD ELECTROSTATICS */
589 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
590 ewrt = _mm_mul_pd(r20,ewtabscale);
591 ewitab = _mm_cvttpd_epi32(ewrt);
593 eweps = _mm_frcz_pd(ewrt);
595 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
597 twoeweps = _mm_add_pd(eweps,eweps);
598 ewitab = _mm_slli_epi32(ewitab,2);
599 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
600 ewtabD = _mm_setzero_pd();
601 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
602 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
603 ewtabFn = _mm_setzero_pd();
604 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
605 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
606 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
607 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
608 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
610 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
612 /* Update potential sum for this i atom from the interaction with this j atom. */
613 velec = _mm_and_pd(velec,cutoff_mask);
614 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
615 velecsum = _mm_add_pd(velecsum,velec);
619 fscal = _mm_and_pd(fscal,cutoff_mask);
621 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
623 /* Update vectorial force */
624 fix2 = _mm_macc_pd(dx20,fscal,fix2);
625 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
626 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
628 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
629 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
630 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
634 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
636 /* Inner loop uses 168 flops */
639 /* End of innermost loop */
641 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
642 f+i_coord_offset,fshift+i_shift_offset);
645 /* Update potential energies */
646 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
647 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
649 /* Increment number of inner iterations */
650 inneriter += j_index_end - j_index_start;
652 /* Outer loop uses 20 flops */
655 /* Increment number of outer iterations */
658 /* Update outer/inner flops */
660 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*168);
663 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
664 * Electrostatics interaction: Ewald
665 * VdW interaction: LennardJones
666 * Geometry: Water3-Particle
667 * Calculate force/pot: Force
670 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
671 (t_nblist * gmx_restrict nlist,
672 rvec * gmx_restrict xx,
673 rvec * gmx_restrict ff,
674 t_forcerec * gmx_restrict fr,
675 t_mdatoms * gmx_restrict mdatoms,
676 nb_kernel_data_t * gmx_restrict kernel_data,
677 t_nrnb * gmx_restrict nrnb)
679 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
680 * just 0 for non-waters.
681 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
682 * jnr indices corresponding to data put in the four positions in the SIMD register.
684 int i_shift_offset,i_coord_offset,outeriter,inneriter;
685 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
687 int j_coord_offsetA,j_coord_offsetB;
688 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
690 real *shiftvec,*fshift,*x,*f;
691 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
693 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
695 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
697 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
698 int vdwjidx0A,vdwjidx0B;
699 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
700 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
701 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
702 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
703 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
706 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
709 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
710 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
712 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
714 __m128d dummy_mask,cutoff_mask;
715 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
716 __m128d one = _mm_set1_pd(1.0);
717 __m128d two = _mm_set1_pd(2.0);
723 jindex = nlist->jindex;
725 shiftidx = nlist->shift;
727 shiftvec = fr->shift_vec[0];
728 fshift = fr->fshift[0];
729 facel = _mm_set1_pd(fr->epsfac);
730 charge = mdatoms->chargeA;
731 nvdwtype = fr->ntype;
733 vdwtype = mdatoms->typeA;
735 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
736 ewtab = fr->ic->tabq_coul_F;
737 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
738 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
740 /* Setup water-specific parameters */
741 inr = nlist->iinr[0];
742 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
743 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
744 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
745 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
747 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
748 rcutoff_scalar = fr->rcoulomb;
749 rcutoff = _mm_set1_pd(rcutoff_scalar);
750 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
752 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
753 rvdw = _mm_set1_pd(fr->rvdw);
755 /* Avoid stupid compiler warnings */
763 /* Start outer loop over neighborlists */
764 for(iidx=0; iidx<nri; iidx++)
766 /* Load shift vector for this list */
767 i_shift_offset = DIM*shiftidx[iidx];
769 /* Load limits for loop over neighbors */
770 j_index_start = jindex[iidx];
771 j_index_end = jindex[iidx+1];
773 /* Get outer coordinate index */
775 i_coord_offset = DIM*inr;
777 /* Load i particle coords and add shift vector */
778 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
779 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
781 fix0 = _mm_setzero_pd();
782 fiy0 = _mm_setzero_pd();
783 fiz0 = _mm_setzero_pd();
784 fix1 = _mm_setzero_pd();
785 fiy1 = _mm_setzero_pd();
786 fiz1 = _mm_setzero_pd();
787 fix2 = _mm_setzero_pd();
788 fiy2 = _mm_setzero_pd();
789 fiz2 = _mm_setzero_pd();
791 /* Start inner kernel loop */
792 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
795 /* Get j neighbor index, and coordinate index */
798 j_coord_offsetA = DIM*jnrA;
799 j_coord_offsetB = DIM*jnrB;
801 /* load j atom coordinates */
802 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
805 /* Calculate displacement vector */
806 dx00 = _mm_sub_pd(ix0,jx0);
807 dy00 = _mm_sub_pd(iy0,jy0);
808 dz00 = _mm_sub_pd(iz0,jz0);
809 dx10 = _mm_sub_pd(ix1,jx0);
810 dy10 = _mm_sub_pd(iy1,jy0);
811 dz10 = _mm_sub_pd(iz1,jz0);
812 dx20 = _mm_sub_pd(ix2,jx0);
813 dy20 = _mm_sub_pd(iy2,jy0);
814 dz20 = _mm_sub_pd(iz2,jz0);
816 /* Calculate squared distance and things based on it */
817 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
818 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
819 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
821 rinv00 = gmx_mm_invsqrt_pd(rsq00);
822 rinv10 = gmx_mm_invsqrt_pd(rsq10);
823 rinv20 = gmx_mm_invsqrt_pd(rsq20);
825 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
826 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
827 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
829 /* Load parameters for j particles */
830 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
831 vdwjidx0A = 2*vdwtype[jnrA+0];
832 vdwjidx0B = 2*vdwtype[jnrB+0];
834 fjx0 = _mm_setzero_pd();
835 fjy0 = _mm_setzero_pd();
836 fjz0 = _mm_setzero_pd();
838 /**************************
839 * CALCULATE INTERACTIONS *
840 **************************/
842 if (gmx_mm_any_lt(rsq00,rcutoff2))
845 r00 = _mm_mul_pd(rsq00,rinv00);
847 /* Compute parameters for interactions between i and j atoms */
848 qq00 = _mm_mul_pd(iq0,jq0);
849 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
850 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
852 /* EWALD ELECTROSTATICS */
854 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
855 ewrt = _mm_mul_pd(r00,ewtabscale);
856 ewitab = _mm_cvttpd_epi32(ewrt);
858 eweps = _mm_frcz_pd(ewrt);
860 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
862 twoeweps = _mm_add_pd(eweps,eweps);
863 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
865 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
866 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
868 /* LENNARD-JONES DISPERSION/REPULSION */
870 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
871 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
873 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
875 fscal = _mm_add_pd(felec,fvdw);
877 fscal = _mm_and_pd(fscal,cutoff_mask);
879 /* Update vectorial force */
880 fix0 = _mm_macc_pd(dx00,fscal,fix0);
881 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
882 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
884 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
885 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
886 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
890 /**************************
891 * CALCULATE INTERACTIONS *
892 **************************/
894 if (gmx_mm_any_lt(rsq10,rcutoff2))
897 r10 = _mm_mul_pd(rsq10,rinv10);
899 /* Compute parameters for interactions between i and j atoms */
900 qq10 = _mm_mul_pd(iq1,jq0);
902 /* EWALD ELECTROSTATICS */
904 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
905 ewrt = _mm_mul_pd(r10,ewtabscale);
906 ewitab = _mm_cvttpd_epi32(ewrt);
908 eweps = _mm_frcz_pd(ewrt);
910 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
912 twoeweps = _mm_add_pd(eweps,eweps);
913 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
915 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
916 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
918 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
922 fscal = _mm_and_pd(fscal,cutoff_mask);
924 /* Update vectorial force */
925 fix1 = _mm_macc_pd(dx10,fscal,fix1);
926 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
927 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
929 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
930 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
931 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
935 /**************************
936 * CALCULATE INTERACTIONS *
937 **************************/
939 if (gmx_mm_any_lt(rsq20,rcutoff2))
942 r20 = _mm_mul_pd(rsq20,rinv20);
944 /* Compute parameters for interactions between i and j atoms */
945 qq20 = _mm_mul_pd(iq2,jq0);
947 /* EWALD ELECTROSTATICS */
949 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
950 ewrt = _mm_mul_pd(r20,ewtabscale);
951 ewitab = _mm_cvttpd_epi32(ewrt);
953 eweps = _mm_frcz_pd(ewrt);
955 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
957 twoeweps = _mm_add_pd(eweps,eweps);
958 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
960 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
961 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
963 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
967 fscal = _mm_and_pd(fscal,cutoff_mask);
969 /* Update vectorial force */
970 fix2 = _mm_macc_pd(dx20,fscal,fix2);
971 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
972 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
974 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
975 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
976 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
980 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
982 /* Inner loop uses 136 flops */
989 j_coord_offsetA = DIM*jnrA;
991 /* load j atom coordinates */
992 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
995 /* Calculate displacement vector */
996 dx00 = _mm_sub_pd(ix0,jx0);
997 dy00 = _mm_sub_pd(iy0,jy0);
998 dz00 = _mm_sub_pd(iz0,jz0);
999 dx10 = _mm_sub_pd(ix1,jx0);
1000 dy10 = _mm_sub_pd(iy1,jy0);
1001 dz10 = _mm_sub_pd(iz1,jz0);
1002 dx20 = _mm_sub_pd(ix2,jx0);
1003 dy20 = _mm_sub_pd(iy2,jy0);
1004 dz20 = _mm_sub_pd(iz2,jz0);
1006 /* Calculate squared distance and things based on it */
1007 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1008 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1009 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1011 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1012 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1013 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1015 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1016 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1017 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1019 /* Load parameters for j particles */
1020 jq0 = _mm_load_sd(charge+jnrA+0);
1021 vdwjidx0A = 2*vdwtype[jnrA+0];
1023 fjx0 = _mm_setzero_pd();
1024 fjy0 = _mm_setzero_pd();
1025 fjz0 = _mm_setzero_pd();
1027 /**************************
1028 * CALCULATE INTERACTIONS *
1029 **************************/
1031 if (gmx_mm_any_lt(rsq00,rcutoff2))
1034 r00 = _mm_mul_pd(rsq00,rinv00);
1036 /* Compute parameters for interactions between i and j atoms */
1037 qq00 = _mm_mul_pd(iq0,jq0);
1038 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1040 /* EWALD ELECTROSTATICS */
1042 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1043 ewrt = _mm_mul_pd(r00,ewtabscale);
1044 ewitab = _mm_cvttpd_epi32(ewrt);
1046 eweps = _mm_frcz_pd(ewrt);
1048 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1050 twoeweps = _mm_add_pd(eweps,eweps);
1051 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1052 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1053 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1055 /* LENNARD-JONES DISPERSION/REPULSION */
1057 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1058 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1060 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1062 fscal = _mm_add_pd(felec,fvdw);
1064 fscal = _mm_and_pd(fscal,cutoff_mask);
1066 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1068 /* Update vectorial force */
1069 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1070 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1071 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1073 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1074 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1075 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1079 /**************************
1080 * CALCULATE INTERACTIONS *
1081 **************************/
1083 if (gmx_mm_any_lt(rsq10,rcutoff2))
1086 r10 = _mm_mul_pd(rsq10,rinv10);
1088 /* Compute parameters for interactions between i and j atoms */
1089 qq10 = _mm_mul_pd(iq1,jq0);
1091 /* EWALD ELECTROSTATICS */
1093 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1094 ewrt = _mm_mul_pd(r10,ewtabscale);
1095 ewitab = _mm_cvttpd_epi32(ewrt);
1097 eweps = _mm_frcz_pd(ewrt);
1099 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1101 twoeweps = _mm_add_pd(eweps,eweps);
1102 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1103 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1104 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1106 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1110 fscal = _mm_and_pd(fscal,cutoff_mask);
1112 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1114 /* Update vectorial force */
1115 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1116 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1117 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1119 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1120 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1121 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1125 /**************************
1126 * CALCULATE INTERACTIONS *
1127 **************************/
1129 if (gmx_mm_any_lt(rsq20,rcutoff2))
1132 r20 = _mm_mul_pd(rsq20,rinv20);
1134 /* Compute parameters for interactions between i and j atoms */
1135 qq20 = _mm_mul_pd(iq2,jq0);
1137 /* EWALD ELECTROSTATICS */
1139 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1140 ewrt = _mm_mul_pd(r20,ewtabscale);
1141 ewitab = _mm_cvttpd_epi32(ewrt);
1143 eweps = _mm_frcz_pd(ewrt);
1145 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1147 twoeweps = _mm_add_pd(eweps,eweps);
1148 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1149 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1150 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1152 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1156 fscal = _mm_and_pd(fscal,cutoff_mask);
1158 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1160 /* Update vectorial force */
1161 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1162 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1163 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1165 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1166 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1167 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1171 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1173 /* Inner loop uses 136 flops */
1176 /* End of innermost loop */
1178 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1179 f+i_coord_offset,fshift+i_shift_offset);
1181 /* Increment number of inner iterations */
1182 inneriter += j_index_end - j_index_start;
1184 /* Outer loop uses 18 flops */
1187 /* Increment number of outer iterations */
1190 /* Update outer/inner flops */
1192 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*136);