2 * Note: this file was generated by the Gromacs sse2_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_sse2_double.h"
34 #include "kernelutil_x86_sse2_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_sse2_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Water3-Water3
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_sse2_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 int vdwjidx1A,vdwjidx1B;
75 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
76 int vdwjidx2A,vdwjidx2B;
77 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
78 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
79 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
80 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
81 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
82 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
83 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
84 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
85 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
86 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
87 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
90 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
92 __m128d dummy_mask,cutoff_mask;
93 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
94 __m128d one = _mm_set1_pd(1.0);
95 __m128d two = _mm_set1_pd(2.0);
101 jindex = nlist->jindex;
103 shiftidx = nlist->shift;
105 shiftvec = fr->shift_vec[0];
106 fshift = fr->fshift[0];
107 facel = _mm_set1_pd(fr->epsfac);
108 charge = mdatoms->chargeA;
110 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
111 ewtab = fr->ic->tabq_coul_FDV0;
112 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
113 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
115 /* Setup water-specific parameters */
116 inr = nlist->iinr[0];
117 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
118 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
119 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
121 jq0 = _mm_set1_pd(charge[inr+0]);
122 jq1 = _mm_set1_pd(charge[inr+1]);
123 jq2 = _mm_set1_pd(charge[inr+2]);
124 qq00 = _mm_mul_pd(iq0,jq0);
125 qq01 = _mm_mul_pd(iq0,jq1);
126 qq02 = _mm_mul_pd(iq0,jq2);
127 qq10 = _mm_mul_pd(iq1,jq0);
128 qq11 = _mm_mul_pd(iq1,jq1);
129 qq12 = _mm_mul_pd(iq1,jq2);
130 qq20 = _mm_mul_pd(iq2,jq0);
131 qq21 = _mm_mul_pd(iq2,jq1);
132 qq22 = _mm_mul_pd(iq2,jq2);
134 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
135 rcutoff_scalar = fr->rcoulomb;
136 rcutoff = _mm_set1_pd(rcutoff_scalar);
137 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
139 /* Avoid stupid compiler warnings */
147 /* Start outer loop over neighborlists */
148 for(iidx=0; iidx<nri; iidx++)
150 /* Load shift vector for this list */
151 i_shift_offset = DIM*shiftidx[iidx];
153 /* Load limits for loop over neighbors */
154 j_index_start = jindex[iidx];
155 j_index_end = jindex[iidx+1];
157 /* Get outer coordinate index */
159 i_coord_offset = DIM*inr;
161 /* Load i particle coords and add shift vector */
162 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
163 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
165 fix0 = _mm_setzero_pd();
166 fiy0 = _mm_setzero_pd();
167 fiz0 = _mm_setzero_pd();
168 fix1 = _mm_setzero_pd();
169 fiy1 = _mm_setzero_pd();
170 fiz1 = _mm_setzero_pd();
171 fix2 = _mm_setzero_pd();
172 fiy2 = _mm_setzero_pd();
173 fiz2 = _mm_setzero_pd();
175 /* Reset potential sums */
176 velecsum = _mm_setzero_pd();
178 /* Start inner kernel loop */
179 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
182 /* Get j neighbor index, and coordinate index */
185 j_coord_offsetA = DIM*jnrA;
186 j_coord_offsetB = DIM*jnrB;
188 /* load j atom coordinates */
189 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
190 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
192 /* Calculate displacement vector */
193 dx00 = _mm_sub_pd(ix0,jx0);
194 dy00 = _mm_sub_pd(iy0,jy0);
195 dz00 = _mm_sub_pd(iz0,jz0);
196 dx01 = _mm_sub_pd(ix0,jx1);
197 dy01 = _mm_sub_pd(iy0,jy1);
198 dz01 = _mm_sub_pd(iz0,jz1);
199 dx02 = _mm_sub_pd(ix0,jx2);
200 dy02 = _mm_sub_pd(iy0,jy2);
201 dz02 = _mm_sub_pd(iz0,jz2);
202 dx10 = _mm_sub_pd(ix1,jx0);
203 dy10 = _mm_sub_pd(iy1,jy0);
204 dz10 = _mm_sub_pd(iz1,jz0);
205 dx11 = _mm_sub_pd(ix1,jx1);
206 dy11 = _mm_sub_pd(iy1,jy1);
207 dz11 = _mm_sub_pd(iz1,jz1);
208 dx12 = _mm_sub_pd(ix1,jx2);
209 dy12 = _mm_sub_pd(iy1,jy2);
210 dz12 = _mm_sub_pd(iz1,jz2);
211 dx20 = _mm_sub_pd(ix2,jx0);
212 dy20 = _mm_sub_pd(iy2,jy0);
213 dz20 = _mm_sub_pd(iz2,jz0);
214 dx21 = _mm_sub_pd(ix2,jx1);
215 dy21 = _mm_sub_pd(iy2,jy1);
216 dz21 = _mm_sub_pd(iz2,jz1);
217 dx22 = _mm_sub_pd(ix2,jx2);
218 dy22 = _mm_sub_pd(iy2,jy2);
219 dz22 = _mm_sub_pd(iz2,jz2);
221 /* Calculate squared distance and things based on it */
222 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
223 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
224 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
225 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
226 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
227 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
228 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
229 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
230 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
232 rinv00 = gmx_mm_invsqrt_pd(rsq00);
233 rinv01 = gmx_mm_invsqrt_pd(rsq01);
234 rinv02 = gmx_mm_invsqrt_pd(rsq02);
235 rinv10 = gmx_mm_invsqrt_pd(rsq10);
236 rinv11 = gmx_mm_invsqrt_pd(rsq11);
237 rinv12 = gmx_mm_invsqrt_pd(rsq12);
238 rinv20 = gmx_mm_invsqrt_pd(rsq20);
239 rinv21 = gmx_mm_invsqrt_pd(rsq21);
240 rinv22 = gmx_mm_invsqrt_pd(rsq22);
242 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
243 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
244 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
245 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
246 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
247 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
248 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
249 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
250 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
252 fjx0 = _mm_setzero_pd();
253 fjy0 = _mm_setzero_pd();
254 fjz0 = _mm_setzero_pd();
255 fjx1 = _mm_setzero_pd();
256 fjy1 = _mm_setzero_pd();
257 fjz1 = _mm_setzero_pd();
258 fjx2 = _mm_setzero_pd();
259 fjy2 = _mm_setzero_pd();
260 fjz2 = _mm_setzero_pd();
262 /**************************
263 * CALCULATE INTERACTIONS *
264 **************************/
266 if (gmx_mm_any_lt(rsq00,rcutoff2))
269 r00 = _mm_mul_pd(rsq00,rinv00);
271 /* EWALD ELECTROSTATICS */
273 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
274 ewrt = _mm_mul_pd(r00,ewtabscale);
275 ewitab = _mm_cvttpd_epi32(ewrt);
276 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
277 ewitab = _mm_slli_epi32(ewitab,2);
278 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
279 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
280 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
281 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
282 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
283 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
284 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
285 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
286 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
287 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
289 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
291 /* Update potential sum for this i atom from the interaction with this j atom. */
292 velec = _mm_and_pd(velec,cutoff_mask);
293 velecsum = _mm_add_pd(velecsum,velec);
297 fscal = _mm_and_pd(fscal,cutoff_mask);
299 /* Calculate temporary vectorial force */
300 tx = _mm_mul_pd(fscal,dx00);
301 ty = _mm_mul_pd(fscal,dy00);
302 tz = _mm_mul_pd(fscal,dz00);
304 /* Update vectorial force */
305 fix0 = _mm_add_pd(fix0,tx);
306 fiy0 = _mm_add_pd(fiy0,ty);
307 fiz0 = _mm_add_pd(fiz0,tz);
309 fjx0 = _mm_add_pd(fjx0,tx);
310 fjy0 = _mm_add_pd(fjy0,ty);
311 fjz0 = _mm_add_pd(fjz0,tz);
315 /**************************
316 * CALCULATE INTERACTIONS *
317 **************************/
319 if (gmx_mm_any_lt(rsq01,rcutoff2))
322 r01 = _mm_mul_pd(rsq01,rinv01);
324 /* EWALD ELECTROSTATICS */
326 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
327 ewrt = _mm_mul_pd(r01,ewtabscale);
328 ewitab = _mm_cvttpd_epi32(ewrt);
329 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
330 ewitab = _mm_slli_epi32(ewitab,2);
331 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
332 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
333 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
334 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
335 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
336 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
337 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
338 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
339 velec = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
340 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
342 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
344 /* Update potential sum for this i atom from the interaction with this j atom. */
345 velec = _mm_and_pd(velec,cutoff_mask);
346 velecsum = _mm_add_pd(velecsum,velec);
350 fscal = _mm_and_pd(fscal,cutoff_mask);
352 /* Calculate temporary vectorial force */
353 tx = _mm_mul_pd(fscal,dx01);
354 ty = _mm_mul_pd(fscal,dy01);
355 tz = _mm_mul_pd(fscal,dz01);
357 /* Update vectorial force */
358 fix0 = _mm_add_pd(fix0,tx);
359 fiy0 = _mm_add_pd(fiy0,ty);
360 fiz0 = _mm_add_pd(fiz0,tz);
362 fjx1 = _mm_add_pd(fjx1,tx);
363 fjy1 = _mm_add_pd(fjy1,ty);
364 fjz1 = _mm_add_pd(fjz1,tz);
368 /**************************
369 * CALCULATE INTERACTIONS *
370 **************************/
372 if (gmx_mm_any_lt(rsq02,rcutoff2))
375 r02 = _mm_mul_pd(rsq02,rinv02);
377 /* EWALD ELECTROSTATICS */
379 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
380 ewrt = _mm_mul_pd(r02,ewtabscale);
381 ewitab = _mm_cvttpd_epi32(ewrt);
382 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
383 ewitab = _mm_slli_epi32(ewitab,2);
384 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
385 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
386 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
387 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
388 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
389 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
390 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
391 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
392 velec = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
393 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
395 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
397 /* Update potential sum for this i atom from the interaction with this j atom. */
398 velec = _mm_and_pd(velec,cutoff_mask);
399 velecsum = _mm_add_pd(velecsum,velec);
403 fscal = _mm_and_pd(fscal,cutoff_mask);
405 /* Calculate temporary vectorial force */
406 tx = _mm_mul_pd(fscal,dx02);
407 ty = _mm_mul_pd(fscal,dy02);
408 tz = _mm_mul_pd(fscal,dz02);
410 /* Update vectorial force */
411 fix0 = _mm_add_pd(fix0,tx);
412 fiy0 = _mm_add_pd(fiy0,ty);
413 fiz0 = _mm_add_pd(fiz0,tz);
415 fjx2 = _mm_add_pd(fjx2,tx);
416 fjy2 = _mm_add_pd(fjy2,ty);
417 fjz2 = _mm_add_pd(fjz2,tz);
421 /**************************
422 * CALCULATE INTERACTIONS *
423 **************************/
425 if (gmx_mm_any_lt(rsq10,rcutoff2))
428 r10 = _mm_mul_pd(rsq10,rinv10);
430 /* EWALD ELECTROSTATICS */
432 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
433 ewrt = _mm_mul_pd(r10,ewtabscale);
434 ewitab = _mm_cvttpd_epi32(ewrt);
435 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
436 ewitab = _mm_slli_epi32(ewitab,2);
437 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
438 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
439 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
440 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
441 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
442 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
443 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
444 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
445 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
446 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
448 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
450 /* Update potential sum for this i atom from the interaction with this j atom. */
451 velec = _mm_and_pd(velec,cutoff_mask);
452 velecsum = _mm_add_pd(velecsum,velec);
456 fscal = _mm_and_pd(fscal,cutoff_mask);
458 /* Calculate temporary vectorial force */
459 tx = _mm_mul_pd(fscal,dx10);
460 ty = _mm_mul_pd(fscal,dy10);
461 tz = _mm_mul_pd(fscal,dz10);
463 /* Update vectorial force */
464 fix1 = _mm_add_pd(fix1,tx);
465 fiy1 = _mm_add_pd(fiy1,ty);
466 fiz1 = _mm_add_pd(fiz1,tz);
468 fjx0 = _mm_add_pd(fjx0,tx);
469 fjy0 = _mm_add_pd(fjy0,ty);
470 fjz0 = _mm_add_pd(fjz0,tz);
474 /**************************
475 * CALCULATE INTERACTIONS *
476 **************************/
478 if (gmx_mm_any_lt(rsq11,rcutoff2))
481 r11 = _mm_mul_pd(rsq11,rinv11);
483 /* EWALD ELECTROSTATICS */
485 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
486 ewrt = _mm_mul_pd(r11,ewtabscale);
487 ewitab = _mm_cvttpd_epi32(ewrt);
488 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
489 ewitab = _mm_slli_epi32(ewitab,2);
490 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
491 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
492 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
493 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
494 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
495 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
496 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
497 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
498 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
499 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
501 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
503 /* Update potential sum for this i atom from the interaction with this j atom. */
504 velec = _mm_and_pd(velec,cutoff_mask);
505 velecsum = _mm_add_pd(velecsum,velec);
509 fscal = _mm_and_pd(fscal,cutoff_mask);
511 /* Calculate temporary vectorial force */
512 tx = _mm_mul_pd(fscal,dx11);
513 ty = _mm_mul_pd(fscal,dy11);
514 tz = _mm_mul_pd(fscal,dz11);
516 /* Update vectorial force */
517 fix1 = _mm_add_pd(fix1,tx);
518 fiy1 = _mm_add_pd(fiy1,ty);
519 fiz1 = _mm_add_pd(fiz1,tz);
521 fjx1 = _mm_add_pd(fjx1,tx);
522 fjy1 = _mm_add_pd(fjy1,ty);
523 fjz1 = _mm_add_pd(fjz1,tz);
527 /**************************
528 * CALCULATE INTERACTIONS *
529 **************************/
531 if (gmx_mm_any_lt(rsq12,rcutoff2))
534 r12 = _mm_mul_pd(rsq12,rinv12);
536 /* EWALD ELECTROSTATICS */
538 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
539 ewrt = _mm_mul_pd(r12,ewtabscale);
540 ewitab = _mm_cvttpd_epi32(ewrt);
541 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
542 ewitab = _mm_slli_epi32(ewitab,2);
543 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
544 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
545 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
546 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
547 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
548 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
549 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
550 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
551 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
552 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
554 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
556 /* Update potential sum for this i atom from the interaction with this j atom. */
557 velec = _mm_and_pd(velec,cutoff_mask);
558 velecsum = _mm_add_pd(velecsum,velec);
562 fscal = _mm_and_pd(fscal,cutoff_mask);
564 /* Calculate temporary vectorial force */
565 tx = _mm_mul_pd(fscal,dx12);
566 ty = _mm_mul_pd(fscal,dy12);
567 tz = _mm_mul_pd(fscal,dz12);
569 /* Update vectorial force */
570 fix1 = _mm_add_pd(fix1,tx);
571 fiy1 = _mm_add_pd(fiy1,ty);
572 fiz1 = _mm_add_pd(fiz1,tz);
574 fjx2 = _mm_add_pd(fjx2,tx);
575 fjy2 = _mm_add_pd(fjy2,ty);
576 fjz2 = _mm_add_pd(fjz2,tz);
580 /**************************
581 * CALCULATE INTERACTIONS *
582 **************************/
584 if (gmx_mm_any_lt(rsq20,rcutoff2))
587 r20 = _mm_mul_pd(rsq20,rinv20);
589 /* EWALD ELECTROSTATICS */
591 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
592 ewrt = _mm_mul_pd(r20,ewtabscale);
593 ewitab = _mm_cvttpd_epi32(ewrt);
594 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
595 ewitab = _mm_slli_epi32(ewitab,2);
596 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
597 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
598 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
599 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
600 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
601 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
602 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
603 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
604 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
605 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
607 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
609 /* Update potential sum for this i atom from the interaction with this j atom. */
610 velec = _mm_and_pd(velec,cutoff_mask);
611 velecsum = _mm_add_pd(velecsum,velec);
615 fscal = _mm_and_pd(fscal,cutoff_mask);
617 /* Calculate temporary vectorial force */
618 tx = _mm_mul_pd(fscal,dx20);
619 ty = _mm_mul_pd(fscal,dy20);
620 tz = _mm_mul_pd(fscal,dz20);
622 /* Update vectorial force */
623 fix2 = _mm_add_pd(fix2,tx);
624 fiy2 = _mm_add_pd(fiy2,ty);
625 fiz2 = _mm_add_pd(fiz2,tz);
627 fjx0 = _mm_add_pd(fjx0,tx);
628 fjy0 = _mm_add_pd(fjy0,ty);
629 fjz0 = _mm_add_pd(fjz0,tz);
633 /**************************
634 * CALCULATE INTERACTIONS *
635 **************************/
637 if (gmx_mm_any_lt(rsq21,rcutoff2))
640 r21 = _mm_mul_pd(rsq21,rinv21);
642 /* EWALD ELECTROSTATICS */
644 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
645 ewrt = _mm_mul_pd(r21,ewtabscale);
646 ewitab = _mm_cvttpd_epi32(ewrt);
647 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
648 ewitab = _mm_slli_epi32(ewitab,2);
649 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
650 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
651 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
652 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
653 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
654 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
655 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
656 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
657 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
658 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
660 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
662 /* Update potential sum for this i atom from the interaction with this j atom. */
663 velec = _mm_and_pd(velec,cutoff_mask);
664 velecsum = _mm_add_pd(velecsum,velec);
668 fscal = _mm_and_pd(fscal,cutoff_mask);
670 /* Calculate temporary vectorial force */
671 tx = _mm_mul_pd(fscal,dx21);
672 ty = _mm_mul_pd(fscal,dy21);
673 tz = _mm_mul_pd(fscal,dz21);
675 /* Update vectorial force */
676 fix2 = _mm_add_pd(fix2,tx);
677 fiy2 = _mm_add_pd(fiy2,ty);
678 fiz2 = _mm_add_pd(fiz2,tz);
680 fjx1 = _mm_add_pd(fjx1,tx);
681 fjy1 = _mm_add_pd(fjy1,ty);
682 fjz1 = _mm_add_pd(fjz1,tz);
686 /**************************
687 * CALCULATE INTERACTIONS *
688 **************************/
690 if (gmx_mm_any_lt(rsq22,rcutoff2))
693 r22 = _mm_mul_pd(rsq22,rinv22);
695 /* EWALD ELECTROSTATICS */
697 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
698 ewrt = _mm_mul_pd(r22,ewtabscale);
699 ewitab = _mm_cvttpd_epi32(ewrt);
700 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
701 ewitab = _mm_slli_epi32(ewitab,2);
702 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
703 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
704 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
705 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
706 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
707 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
708 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
709 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
710 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
711 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
713 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
715 /* Update potential sum for this i atom from the interaction with this j atom. */
716 velec = _mm_and_pd(velec,cutoff_mask);
717 velecsum = _mm_add_pd(velecsum,velec);
721 fscal = _mm_and_pd(fscal,cutoff_mask);
723 /* Calculate temporary vectorial force */
724 tx = _mm_mul_pd(fscal,dx22);
725 ty = _mm_mul_pd(fscal,dy22);
726 tz = _mm_mul_pd(fscal,dz22);
728 /* Update vectorial force */
729 fix2 = _mm_add_pd(fix2,tx);
730 fiy2 = _mm_add_pd(fiy2,ty);
731 fiz2 = _mm_add_pd(fiz2,tz);
733 fjx2 = _mm_add_pd(fjx2,tx);
734 fjy2 = _mm_add_pd(fjy2,ty);
735 fjz2 = _mm_add_pd(fjz2,tz);
739 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
741 /* Inner loop uses 414 flops */
748 j_coord_offsetA = DIM*jnrA;
750 /* load j atom coordinates */
751 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
752 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
754 /* Calculate displacement vector */
755 dx00 = _mm_sub_pd(ix0,jx0);
756 dy00 = _mm_sub_pd(iy0,jy0);
757 dz00 = _mm_sub_pd(iz0,jz0);
758 dx01 = _mm_sub_pd(ix0,jx1);
759 dy01 = _mm_sub_pd(iy0,jy1);
760 dz01 = _mm_sub_pd(iz0,jz1);
761 dx02 = _mm_sub_pd(ix0,jx2);
762 dy02 = _mm_sub_pd(iy0,jy2);
763 dz02 = _mm_sub_pd(iz0,jz2);
764 dx10 = _mm_sub_pd(ix1,jx0);
765 dy10 = _mm_sub_pd(iy1,jy0);
766 dz10 = _mm_sub_pd(iz1,jz0);
767 dx11 = _mm_sub_pd(ix1,jx1);
768 dy11 = _mm_sub_pd(iy1,jy1);
769 dz11 = _mm_sub_pd(iz1,jz1);
770 dx12 = _mm_sub_pd(ix1,jx2);
771 dy12 = _mm_sub_pd(iy1,jy2);
772 dz12 = _mm_sub_pd(iz1,jz2);
773 dx20 = _mm_sub_pd(ix2,jx0);
774 dy20 = _mm_sub_pd(iy2,jy0);
775 dz20 = _mm_sub_pd(iz2,jz0);
776 dx21 = _mm_sub_pd(ix2,jx1);
777 dy21 = _mm_sub_pd(iy2,jy1);
778 dz21 = _mm_sub_pd(iz2,jz1);
779 dx22 = _mm_sub_pd(ix2,jx2);
780 dy22 = _mm_sub_pd(iy2,jy2);
781 dz22 = _mm_sub_pd(iz2,jz2);
783 /* Calculate squared distance and things based on it */
784 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
785 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
786 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
787 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
788 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
789 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
790 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
791 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
792 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
794 rinv00 = gmx_mm_invsqrt_pd(rsq00);
795 rinv01 = gmx_mm_invsqrt_pd(rsq01);
796 rinv02 = gmx_mm_invsqrt_pd(rsq02);
797 rinv10 = gmx_mm_invsqrt_pd(rsq10);
798 rinv11 = gmx_mm_invsqrt_pd(rsq11);
799 rinv12 = gmx_mm_invsqrt_pd(rsq12);
800 rinv20 = gmx_mm_invsqrt_pd(rsq20);
801 rinv21 = gmx_mm_invsqrt_pd(rsq21);
802 rinv22 = gmx_mm_invsqrt_pd(rsq22);
804 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
805 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
806 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
807 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
808 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
809 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
810 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
811 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
812 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
814 fjx0 = _mm_setzero_pd();
815 fjy0 = _mm_setzero_pd();
816 fjz0 = _mm_setzero_pd();
817 fjx1 = _mm_setzero_pd();
818 fjy1 = _mm_setzero_pd();
819 fjz1 = _mm_setzero_pd();
820 fjx2 = _mm_setzero_pd();
821 fjy2 = _mm_setzero_pd();
822 fjz2 = _mm_setzero_pd();
824 /**************************
825 * CALCULATE INTERACTIONS *
826 **************************/
828 if (gmx_mm_any_lt(rsq00,rcutoff2))
831 r00 = _mm_mul_pd(rsq00,rinv00);
833 /* EWALD ELECTROSTATICS */
835 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
836 ewrt = _mm_mul_pd(r00,ewtabscale);
837 ewitab = _mm_cvttpd_epi32(ewrt);
838 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
839 ewitab = _mm_slli_epi32(ewitab,2);
840 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
841 ewtabD = _mm_setzero_pd();
842 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
843 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
844 ewtabFn = _mm_setzero_pd();
845 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
846 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
847 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
848 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
849 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
851 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
853 /* Update potential sum for this i atom from the interaction with this j atom. */
854 velec = _mm_and_pd(velec,cutoff_mask);
855 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
856 velecsum = _mm_add_pd(velecsum,velec);
860 fscal = _mm_and_pd(fscal,cutoff_mask);
862 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
864 /* Calculate temporary vectorial force */
865 tx = _mm_mul_pd(fscal,dx00);
866 ty = _mm_mul_pd(fscal,dy00);
867 tz = _mm_mul_pd(fscal,dz00);
869 /* Update vectorial force */
870 fix0 = _mm_add_pd(fix0,tx);
871 fiy0 = _mm_add_pd(fiy0,ty);
872 fiz0 = _mm_add_pd(fiz0,tz);
874 fjx0 = _mm_add_pd(fjx0,tx);
875 fjy0 = _mm_add_pd(fjy0,ty);
876 fjz0 = _mm_add_pd(fjz0,tz);
880 /**************************
881 * CALCULATE INTERACTIONS *
882 **************************/
884 if (gmx_mm_any_lt(rsq01,rcutoff2))
887 r01 = _mm_mul_pd(rsq01,rinv01);
889 /* EWALD ELECTROSTATICS */
891 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
892 ewrt = _mm_mul_pd(r01,ewtabscale);
893 ewitab = _mm_cvttpd_epi32(ewrt);
894 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
895 ewitab = _mm_slli_epi32(ewitab,2);
896 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
897 ewtabD = _mm_setzero_pd();
898 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
899 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
900 ewtabFn = _mm_setzero_pd();
901 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
902 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
903 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
904 velec = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
905 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
907 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
909 /* Update potential sum for this i atom from the interaction with this j atom. */
910 velec = _mm_and_pd(velec,cutoff_mask);
911 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
912 velecsum = _mm_add_pd(velecsum,velec);
916 fscal = _mm_and_pd(fscal,cutoff_mask);
918 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
920 /* Calculate temporary vectorial force */
921 tx = _mm_mul_pd(fscal,dx01);
922 ty = _mm_mul_pd(fscal,dy01);
923 tz = _mm_mul_pd(fscal,dz01);
925 /* Update vectorial force */
926 fix0 = _mm_add_pd(fix0,tx);
927 fiy0 = _mm_add_pd(fiy0,ty);
928 fiz0 = _mm_add_pd(fiz0,tz);
930 fjx1 = _mm_add_pd(fjx1,tx);
931 fjy1 = _mm_add_pd(fjy1,ty);
932 fjz1 = _mm_add_pd(fjz1,tz);
936 /**************************
937 * CALCULATE INTERACTIONS *
938 **************************/
940 if (gmx_mm_any_lt(rsq02,rcutoff2))
943 r02 = _mm_mul_pd(rsq02,rinv02);
945 /* EWALD ELECTROSTATICS */
947 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
948 ewrt = _mm_mul_pd(r02,ewtabscale);
949 ewitab = _mm_cvttpd_epi32(ewrt);
950 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
951 ewitab = _mm_slli_epi32(ewitab,2);
952 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
953 ewtabD = _mm_setzero_pd();
954 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
955 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
956 ewtabFn = _mm_setzero_pd();
957 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
958 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
959 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
960 velec = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
961 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
963 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
965 /* Update potential sum for this i atom from the interaction with this j atom. */
966 velec = _mm_and_pd(velec,cutoff_mask);
967 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
968 velecsum = _mm_add_pd(velecsum,velec);
972 fscal = _mm_and_pd(fscal,cutoff_mask);
974 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
976 /* Calculate temporary vectorial force */
977 tx = _mm_mul_pd(fscal,dx02);
978 ty = _mm_mul_pd(fscal,dy02);
979 tz = _mm_mul_pd(fscal,dz02);
981 /* Update vectorial force */
982 fix0 = _mm_add_pd(fix0,tx);
983 fiy0 = _mm_add_pd(fiy0,ty);
984 fiz0 = _mm_add_pd(fiz0,tz);
986 fjx2 = _mm_add_pd(fjx2,tx);
987 fjy2 = _mm_add_pd(fjy2,ty);
988 fjz2 = _mm_add_pd(fjz2,tz);
992 /**************************
993 * CALCULATE INTERACTIONS *
994 **************************/
996 if (gmx_mm_any_lt(rsq10,rcutoff2))
999 r10 = _mm_mul_pd(rsq10,rinv10);
1001 /* EWALD ELECTROSTATICS */
1003 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1004 ewrt = _mm_mul_pd(r10,ewtabscale);
1005 ewitab = _mm_cvttpd_epi32(ewrt);
1006 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1007 ewitab = _mm_slli_epi32(ewitab,2);
1008 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1009 ewtabD = _mm_setzero_pd();
1010 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1011 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1012 ewtabFn = _mm_setzero_pd();
1013 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1014 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1015 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1016 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
1017 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1019 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1021 /* Update potential sum for this i atom from the interaction with this j atom. */
1022 velec = _mm_and_pd(velec,cutoff_mask);
1023 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1024 velecsum = _mm_add_pd(velecsum,velec);
1028 fscal = _mm_and_pd(fscal,cutoff_mask);
1030 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1032 /* Calculate temporary vectorial force */
1033 tx = _mm_mul_pd(fscal,dx10);
1034 ty = _mm_mul_pd(fscal,dy10);
1035 tz = _mm_mul_pd(fscal,dz10);
1037 /* Update vectorial force */
1038 fix1 = _mm_add_pd(fix1,tx);
1039 fiy1 = _mm_add_pd(fiy1,ty);
1040 fiz1 = _mm_add_pd(fiz1,tz);
1042 fjx0 = _mm_add_pd(fjx0,tx);
1043 fjy0 = _mm_add_pd(fjy0,ty);
1044 fjz0 = _mm_add_pd(fjz0,tz);
1048 /**************************
1049 * CALCULATE INTERACTIONS *
1050 **************************/
1052 if (gmx_mm_any_lt(rsq11,rcutoff2))
1055 r11 = _mm_mul_pd(rsq11,rinv11);
1057 /* EWALD ELECTROSTATICS */
1059 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060 ewrt = _mm_mul_pd(r11,ewtabscale);
1061 ewitab = _mm_cvttpd_epi32(ewrt);
1062 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1063 ewitab = _mm_slli_epi32(ewitab,2);
1064 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1065 ewtabD = _mm_setzero_pd();
1066 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1067 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1068 ewtabFn = _mm_setzero_pd();
1069 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1070 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1071 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1072 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
1073 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1075 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1077 /* Update potential sum for this i atom from the interaction with this j atom. */
1078 velec = _mm_and_pd(velec,cutoff_mask);
1079 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1080 velecsum = _mm_add_pd(velecsum,velec);
1084 fscal = _mm_and_pd(fscal,cutoff_mask);
1086 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1088 /* Calculate temporary vectorial force */
1089 tx = _mm_mul_pd(fscal,dx11);
1090 ty = _mm_mul_pd(fscal,dy11);
1091 tz = _mm_mul_pd(fscal,dz11);
1093 /* Update vectorial force */
1094 fix1 = _mm_add_pd(fix1,tx);
1095 fiy1 = _mm_add_pd(fiy1,ty);
1096 fiz1 = _mm_add_pd(fiz1,tz);
1098 fjx1 = _mm_add_pd(fjx1,tx);
1099 fjy1 = _mm_add_pd(fjy1,ty);
1100 fjz1 = _mm_add_pd(fjz1,tz);
1104 /**************************
1105 * CALCULATE INTERACTIONS *
1106 **************************/
1108 if (gmx_mm_any_lt(rsq12,rcutoff2))
1111 r12 = _mm_mul_pd(rsq12,rinv12);
1113 /* EWALD ELECTROSTATICS */
1115 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116 ewrt = _mm_mul_pd(r12,ewtabscale);
1117 ewitab = _mm_cvttpd_epi32(ewrt);
1118 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1119 ewitab = _mm_slli_epi32(ewitab,2);
1120 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1121 ewtabD = _mm_setzero_pd();
1122 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1123 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1124 ewtabFn = _mm_setzero_pd();
1125 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1126 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1127 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1128 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
1129 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1131 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1133 /* Update potential sum for this i atom from the interaction with this j atom. */
1134 velec = _mm_and_pd(velec,cutoff_mask);
1135 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1136 velecsum = _mm_add_pd(velecsum,velec);
1140 fscal = _mm_and_pd(fscal,cutoff_mask);
1142 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1144 /* Calculate temporary vectorial force */
1145 tx = _mm_mul_pd(fscal,dx12);
1146 ty = _mm_mul_pd(fscal,dy12);
1147 tz = _mm_mul_pd(fscal,dz12);
1149 /* Update vectorial force */
1150 fix1 = _mm_add_pd(fix1,tx);
1151 fiy1 = _mm_add_pd(fiy1,ty);
1152 fiz1 = _mm_add_pd(fiz1,tz);
1154 fjx2 = _mm_add_pd(fjx2,tx);
1155 fjy2 = _mm_add_pd(fjy2,ty);
1156 fjz2 = _mm_add_pd(fjz2,tz);
1160 /**************************
1161 * CALCULATE INTERACTIONS *
1162 **************************/
1164 if (gmx_mm_any_lt(rsq20,rcutoff2))
1167 r20 = _mm_mul_pd(rsq20,rinv20);
1169 /* EWALD ELECTROSTATICS */
1171 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1172 ewrt = _mm_mul_pd(r20,ewtabscale);
1173 ewitab = _mm_cvttpd_epi32(ewrt);
1174 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1175 ewitab = _mm_slli_epi32(ewitab,2);
1176 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1177 ewtabD = _mm_setzero_pd();
1178 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1179 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1180 ewtabFn = _mm_setzero_pd();
1181 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1182 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1183 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1184 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
1185 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1187 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1189 /* Update potential sum for this i atom from the interaction with this j atom. */
1190 velec = _mm_and_pd(velec,cutoff_mask);
1191 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1192 velecsum = _mm_add_pd(velecsum,velec);
1196 fscal = _mm_and_pd(fscal,cutoff_mask);
1198 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1200 /* Calculate temporary vectorial force */
1201 tx = _mm_mul_pd(fscal,dx20);
1202 ty = _mm_mul_pd(fscal,dy20);
1203 tz = _mm_mul_pd(fscal,dz20);
1205 /* Update vectorial force */
1206 fix2 = _mm_add_pd(fix2,tx);
1207 fiy2 = _mm_add_pd(fiy2,ty);
1208 fiz2 = _mm_add_pd(fiz2,tz);
1210 fjx0 = _mm_add_pd(fjx0,tx);
1211 fjy0 = _mm_add_pd(fjy0,ty);
1212 fjz0 = _mm_add_pd(fjz0,tz);
1216 /**************************
1217 * CALCULATE INTERACTIONS *
1218 **************************/
1220 if (gmx_mm_any_lt(rsq21,rcutoff2))
1223 r21 = _mm_mul_pd(rsq21,rinv21);
1225 /* EWALD ELECTROSTATICS */
1227 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228 ewrt = _mm_mul_pd(r21,ewtabscale);
1229 ewitab = _mm_cvttpd_epi32(ewrt);
1230 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1231 ewitab = _mm_slli_epi32(ewitab,2);
1232 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1233 ewtabD = _mm_setzero_pd();
1234 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1235 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1236 ewtabFn = _mm_setzero_pd();
1237 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1238 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1239 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1240 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1241 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1243 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1245 /* Update potential sum for this i atom from the interaction with this j atom. */
1246 velec = _mm_and_pd(velec,cutoff_mask);
1247 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1248 velecsum = _mm_add_pd(velecsum,velec);
1252 fscal = _mm_and_pd(fscal,cutoff_mask);
1254 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1256 /* Calculate temporary vectorial force */
1257 tx = _mm_mul_pd(fscal,dx21);
1258 ty = _mm_mul_pd(fscal,dy21);
1259 tz = _mm_mul_pd(fscal,dz21);
1261 /* Update vectorial force */
1262 fix2 = _mm_add_pd(fix2,tx);
1263 fiy2 = _mm_add_pd(fiy2,ty);
1264 fiz2 = _mm_add_pd(fiz2,tz);
1266 fjx1 = _mm_add_pd(fjx1,tx);
1267 fjy1 = _mm_add_pd(fjy1,ty);
1268 fjz1 = _mm_add_pd(fjz1,tz);
1272 /**************************
1273 * CALCULATE INTERACTIONS *
1274 **************************/
1276 if (gmx_mm_any_lt(rsq22,rcutoff2))
1279 r22 = _mm_mul_pd(rsq22,rinv22);
1281 /* EWALD ELECTROSTATICS */
1283 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1284 ewrt = _mm_mul_pd(r22,ewtabscale);
1285 ewitab = _mm_cvttpd_epi32(ewrt);
1286 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1287 ewitab = _mm_slli_epi32(ewitab,2);
1288 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1289 ewtabD = _mm_setzero_pd();
1290 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1291 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1292 ewtabFn = _mm_setzero_pd();
1293 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1294 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1295 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1296 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1297 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1299 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1301 /* Update potential sum for this i atom from the interaction with this j atom. */
1302 velec = _mm_and_pd(velec,cutoff_mask);
1303 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1304 velecsum = _mm_add_pd(velecsum,velec);
1308 fscal = _mm_and_pd(fscal,cutoff_mask);
1310 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1312 /* Calculate temporary vectorial force */
1313 tx = _mm_mul_pd(fscal,dx22);
1314 ty = _mm_mul_pd(fscal,dy22);
1315 tz = _mm_mul_pd(fscal,dz22);
1317 /* Update vectorial force */
1318 fix2 = _mm_add_pd(fix2,tx);
1319 fiy2 = _mm_add_pd(fiy2,ty);
1320 fiz2 = _mm_add_pd(fiz2,tz);
1322 fjx2 = _mm_add_pd(fjx2,tx);
1323 fjy2 = _mm_add_pd(fjy2,ty);
1324 fjz2 = _mm_add_pd(fjz2,tz);
1328 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1330 /* Inner loop uses 414 flops */
1333 /* End of innermost loop */
1335 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1336 f+i_coord_offset,fshift+i_shift_offset);
1339 /* Update potential energies */
1340 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1342 /* Increment number of inner iterations */
1343 inneriter += j_index_end - j_index_start;
1345 /* Outer loop uses 19 flops */
1348 /* Increment number of outer iterations */
1351 /* Update outer/inner flops */
1353 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*414);
1356 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_sse2_double
1357 * Electrostatics interaction: Ewald
1358 * VdW interaction: None
1359 * Geometry: Water3-Water3
1360 * Calculate force/pot: Force
1363 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_sse2_double
1364 (t_nblist * gmx_restrict nlist,
1365 rvec * gmx_restrict xx,
1366 rvec * gmx_restrict ff,
1367 t_forcerec * gmx_restrict fr,
1368 t_mdatoms * gmx_restrict mdatoms,
1369 nb_kernel_data_t * gmx_restrict kernel_data,
1370 t_nrnb * gmx_restrict nrnb)
1372 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1373 * just 0 for non-waters.
1374 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1375 * jnr indices corresponding to data put in the four positions in the SIMD register.
1377 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1378 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1380 int j_coord_offsetA,j_coord_offsetB;
1381 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1382 real rcutoff_scalar;
1383 real *shiftvec,*fshift,*x,*f;
1384 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1386 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1388 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1390 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1391 int vdwjidx0A,vdwjidx0B;
1392 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1393 int vdwjidx1A,vdwjidx1B;
1394 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1395 int vdwjidx2A,vdwjidx2B;
1396 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1397 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1398 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1399 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1400 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1401 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1402 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1403 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1404 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1405 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1406 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1409 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1411 __m128d dummy_mask,cutoff_mask;
1412 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1413 __m128d one = _mm_set1_pd(1.0);
1414 __m128d two = _mm_set1_pd(2.0);
1420 jindex = nlist->jindex;
1422 shiftidx = nlist->shift;
1424 shiftvec = fr->shift_vec[0];
1425 fshift = fr->fshift[0];
1426 facel = _mm_set1_pd(fr->epsfac);
1427 charge = mdatoms->chargeA;
1429 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1430 ewtab = fr->ic->tabq_coul_F;
1431 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1432 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1434 /* Setup water-specific parameters */
1435 inr = nlist->iinr[0];
1436 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1437 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1438 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1440 jq0 = _mm_set1_pd(charge[inr+0]);
1441 jq1 = _mm_set1_pd(charge[inr+1]);
1442 jq2 = _mm_set1_pd(charge[inr+2]);
1443 qq00 = _mm_mul_pd(iq0,jq0);
1444 qq01 = _mm_mul_pd(iq0,jq1);
1445 qq02 = _mm_mul_pd(iq0,jq2);
1446 qq10 = _mm_mul_pd(iq1,jq0);
1447 qq11 = _mm_mul_pd(iq1,jq1);
1448 qq12 = _mm_mul_pd(iq1,jq2);
1449 qq20 = _mm_mul_pd(iq2,jq0);
1450 qq21 = _mm_mul_pd(iq2,jq1);
1451 qq22 = _mm_mul_pd(iq2,jq2);
1453 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1454 rcutoff_scalar = fr->rcoulomb;
1455 rcutoff = _mm_set1_pd(rcutoff_scalar);
1456 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
1458 /* Avoid stupid compiler warnings */
1460 j_coord_offsetA = 0;
1461 j_coord_offsetB = 0;
1466 /* Start outer loop over neighborlists */
1467 for(iidx=0; iidx<nri; iidx++)
1469 /* Load shift vector for this list */
1470 i_shift_offset = DIM*shiftidx[iidx];
1472 /* Load limits for loop over neighbors */
1473 j_index_start = jindex[iidx];
1474 j_index_end = jindex[iidx+1];
1476 /* Get outer coordinate index */
1478 i_coord_offset = DIM*inr;
1480 /* Load i particle coords and add shift vector */
1481 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1482 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1484 fix0 = _mm_setzero_pd();
1485 fiy0 = _mm_setzero_pd();
1486 fiz0 = _mm_setzero_pd();
1487 fix1 = _mm_setzero_pd();
1488 fiy1 = _mm_setzero_pd();
1489 fiz1 = _mm_setzero_pd();
1490 fix2 = _mm_setzero_pd();
1491 fiy2 = _mm_setzero_pd();
1492 fiz2 = _mm_setzero_pd();
1494 /* Start inner kernel loop */
1495 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1498 /* Get j neighbor index, and coordinate index */
1500 jnrB = jjnr[jidx+1];
1501 j_coord_offsetA = DIM*jnrA;
1502 j_coord_offsetB = DIM*jnrB;
1504 /* load j atom coordinates */
1505 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1506 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1508 /* Calculate displacement vector */
1509 dx00 = _mm_sub_pd(ix0,jx0);
1510 dy00 = _mm_sub_pd(iy0,jy0);
1511 dz00 = _mm_sub_pd(iz0,jz0);
1512 dx01 = _mm_sub_pd(ix0,jx1);
1513 dy01 = _mm_sub_pd(iy0,jy1);
1514 dz01 = _mm_sub_pd(iz0,jz1);
1515 dx02 = _mm_sub_pd(ix0,jx2);
1516 dy02 = _mm_sub_pd(iy0,jy2);
1517 dz02 = _mm_sub_pd(iz0,jz2);
1518 dx10 = _mm_sub_pd(ix1,jx0);
1519 dy10 = _mm_sub_pd(iy1,jy0);
1520 dz10 = _mm_sub_pd(iz1,jz0);
1521 dx11 = _mm_sub_pd(ix1,jx1);
1522 dy11 = _mm_sub_pd(iy1,jy1);
1523 dz11 = _mm_sub_pd(iz1,jz1);
1524 dx12 = _mm_sub_pd(ix1,jx2);
1525 dy12 = _mm_sub_pd(iy1,jy2);
1526 dz12 = _mm_sub_pd(iz1,jz2);
1527 dx20 = _mm_sub_pd(ix2,jx0);
1528 dy20 = _mm_sub_pd(iy2,jy0);
1529 dz20 = _mm_sub_pd(iz2,jz0);
1530 dx21 = _mm_sub_pd(ix2,jx1);
1531 dy21 = _mm_sub_pd(iy2,jy1);
1532 dz21 = _mm_sub_pd(iz2,jz1);
1533 dx22 = _mm_sub_pd(ix2,jx2);
1534 dy22 = _mm_sub_pd(iy2,jy2);
1535 dz22 = _mm_sub_pd(iz2,jz2);
1537 /* Calculate squared distance and things based on it */
1538 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1539 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1540 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1541 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1542 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1543 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1544 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1545 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1546 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1548 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1549 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1550 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1551 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1552 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1553 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1554 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1555 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1556 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1558 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1559 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
1560 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
1561 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1562 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1563 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1564 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1565 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1566 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1568 fjx0 = _mm_setzero_pd();
1569 fjy0 = _mm_setzero_pd();
1570 fjz0 = _mm_setzero_pd();
1571 fjx1 = _mm_setzero_pd();
1572 fjy1 = _mm_setzero_pd();
1573 fjz1 = _mm_setzero_pd();
1574 fjx2 = _mm_setzero_pd();
1575 fjy2 = _mm_setzero_pd();
1576 fjz2 = _mm_setzero_pd();
1578 /**************************
1579 * CALCULATE INTERACTIONS *
1580 **************************/
1582 if (gmx_mm_any_lt(rsq00,rcutoff2))
1585 r00 = _mm_mul_pd(rsq00,rinv00);
1587 /* EWALD ELECTROSTATICS */
1589 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1590 ewrt = _mm_mul_pd(r00,ewtabscale);
1591 ewitab = _mm_cvttpd_epi32(ewrt);
1592 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1593 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1595 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1596 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1598 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1602 fscal = _mm_and_pd(fscal,cutoff_mask);
1604 /* Calculate temporary vectorial force */
1605 tx = _mm_mul_pd(fscal,dx00);
1606 ty = _mm_mul_pd(fscal,dy00);
1607 tz = _mm_mul_pd(fscal,dz00);
1609 /* Update vectorial force */
1610 fix0 = _mm_add_pd(fix0,tx);
1611 fiy0 = _mm_add_pd(fiy0,ty);
1612 fiz0 = _mm_add_pd(fiz0,tz);
1614 fjx0 = _mm_add_pd(fjx0,tx);
1615 fjy0 = _mm_add_pd(fjy0,ty);
1616 fjz0 = _mm_add_pd(fjz0,tz);
1620 /**************************
1621 * CALCULATE INTERACTIONS *
1622 **************************/
1624 if (gmx_mm_any_lt(rsq01,rcutoff2))
1627 r01 = _mm_mul_pd(rsq01,rinv01);
1629 /* EWALD ELECTROSTATICS */
1631 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1632 ewrt = _mm_mul_pd(r01,ewtabscale);
1633 ewitab = _mm_cvttpd_epi32(ewrt);
1634 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1635 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1637 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1638 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1640 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
1644 fscal = _mm_and_pd(fscal,cutoff_mask);
1646 /* Calculate temporary vectorial force */
1647 tx = _mm_mul_pd(fscal,dx01);
1648 ty = _mm_mul_pd(fscal,dy01);
1649 tz = _mm_mul_pd(fscal,dz01);
1651 /* Update vectorial force */
1652 fix0 = _mm_add_pd(fix0,tx);
1653 fiy0 = _mm_add_pd(fiy0,ty);
1654 fiz0 = _mm_add_pd(fiz0,tz);
1656 fjx1 = _mm_add_pd(fjx1,tx);
1657 fjy1 = _mm_add_pd(fjy1,ty);
1658 fjz1 = _mm_add_pd(fjz1,tz);
1662 /**************************
1663 * CALCULATE INTERACTIONS *
1664 **************************/
1666 if (gmx_mm_any_lt(rsq02,rcutoff2))
1669 r02 = _mm_mul_pd(rsq02,rinv02);
1671 /* EWALD ELECTROSTATICS */
1673 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1674 ewrt = _mm_mul_pd(r02,ewtabscale);
1675 ewitab = _mm_cvttpd_epi32(ewrt);
1676 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1677 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1679 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1680 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1682 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1686 fscal = _mm_and_pd(fscal,cutoff_mask);
1688 /* Calculate temporary vectorial force */
1689 tx = _mm_mul_pd(fscal,dx02);
1690 ty = _mm_mul_pd(fscal,dy02);
1691 tz = _mm_mul_pd(fscal,dz02);
1693 /* Update vectorial force */
1694 fix0 = _mm_add_pd(fix0,tx);
1695 fiy0 = _mm_add_pd(fiy0,ty);
1696 fiz0 = _mm_add_pd(fiz0,tz);
1698 fjx2 = _mm_add_pd(fjx2,tx);
1699 fjy2 = _mm_add_pd(fjy2,ty);
1700 fjz2 = _mm_add_pd(fjz2,tz);
1704 /**************************
1705 * CALCULATE INTERACTIONS *
1706 **************************/
1708 if (gmx_mm_any_lt(rsq10,rcutoff2))
1711 r10 = _mm_mul_pd(rsq10,rinv10);
1713 /* EWALD ELECTROSTATICS */
1715 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1716 ewrt = _mm_mul_pd(r10,ewtabscale);
1717 ewitab = _mm_cvttpd_epi32(ewrt);
1718 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1719 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1721 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1722 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1724 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1728 fscal = _mm_and_pd(fscal,cutoff_mask);
1730 /* Calculate temporary vectorial force */
1731 tx = _mm_mul_pd(fscal,dx10);
1732 ty = _mm_mul_pd(fscal,dy10);
1733 tz = _mm_mul_pd(fscal,dz10);
1735 /* Update vectorial force */
1736 fix1 = _mm_add_pd(fix1,tx);
1737 fiy1 = _mm_add_pd(fiy1,ty);
1738 fiz1 = _mm_add_pd(fiz1,tz);
1740 fjx0 = _mm_add_pd(fjx0,tx);
1741 fjy0 = _mm_add_pd(fjy0,ty);
1742 fjz0 = _mm_add_pd(fjz0,tz);
1746 /**************************
1747 * CALCULATE INTERACTIONS *
1748 **************************/
1750 if (gmx_mm_any_lt(rsq11,rcutoff2))
1753 r11 = _mm_mul_pd(rsq11,rinv11);
1755 /* EWALD ELECTROSTATICS */
1757 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1758 ewrt = _mm_mul_pd(r11,ewtabscale);
1759 ewitab = _mm_cvttpd_epi32(ewrt);
1760 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1761 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1763 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1764 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1766 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1770 fscal = _mm_and_pd(fscal,cutoff_mask);
1772 /* Calculate temporary vectorial force */
1773 tx = _mm_mul_pd(fscal,dx11);
1774 ty = _mm_mul_pd(fscal,dy11);
1775 tz = _mm_mul_pd(fscal,dz11);
1777 /* Update vectorial force */
1778 fix1 = _mm_add_pd(fix1,tx);
1779 fiy1 = _mm_add_pd(fiy1,ty);
1780 fiz1 = _mm_add_pd(fiz1,tz);
1782 fjx1 = _mm_add_pd(fjx1,tx);
1783 fjy1 = _mm_add_pd(fjy1,ty);
1784 fjz1 = _mm_add_pd(fjz1,tz);
1788 /**************************
1789 * CALCULATE INTERACTIONS *
1790 **************************/
1792 if (gmx_mm_any_lt(rsq12,rcutoff2))
1795 r12 = _mm_mul_pd(rsq12,rinv12);
1797 /* EWALD ELECTROSTATICS */
1799 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1800 ewrt = _mm_mul_pd(r12,ewtabscale);
1801 ewitab = _mm_cvttpd_epi32(ewrt);
1802 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1803 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1805 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1806 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1808 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1812 fscal = _mm_and_pd(fscal,cutoff_mask);
1814 /* Calculate temporary vectorial force */
1815 tx = _mm_mul_pd(fscal,dx12);
1816 ty = _mm_mul_pd(fscal,dy12);
1817 tz = _mm_mul_pd(fscal,dz12);
1819 /* Update vectorial force */
1820 fix1 = _mm_add_pd(fix1,tx);
1821 fiy1 = _mm_add_pd(fiy1,ty);
1822 fiz1 = _mm_add_pd(fiz1,tz);
1824 fjx2 = _mm_add_pd(fjx2,tx);
1825 fjy2 = _mm_add_pd(fjy2,ty);
1826 fjz2 = _mm_add_pd(fjz2,tz);
1830 /**************************
1831 * CALCULATE INTERACTIONS *
1832 **************************/
1834 if (gmx_mm_any_lt(rsq20,rcutoff2))
1837 r20 = _mm_mul_pd(rsq20,rinv20);
1839 /* EWALD ELECTROSTATICS */
1841 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1842 ewrt = _mm_mul_pd(r20,ewtabscale);
1843 ewitab = _mm_cvttpd_epi32(ewrt);
1844 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1845 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1847 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1848 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1850 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1854 fscal = _mm_and_pd(fscal,cutoff_mask);
1856 /* Calculate temporary vectorial force */
1857 tx = _mm_mul_pd(fscal,dx20);
1858 ty = _mm_mul_pd(fscal,dy20);
1859 tz = _mm_mul_pd(fscal,dz20);
1861 /* Update vectorial force */
1862 fix2 = _mm_add_pd(fix2,tx);
1863 fiy2 = _mm_add_pd(fiy2,ty);
1864 fiz2 = _mm_add_pd(fiz2,tz);
1866 fjx0 = _mm_add_pd(fjx0,tx);
1867 fjy0 = _mm_add_pd(fjy0,ty);
1868 fjz0 = _mm_add_pd(fjz0,tz);
1872 /**************************
1873 * CALCULATE INTERACTIONS *
1874 **************************/
1876 if (gmx_mm_any_lt(rsq21,rcutoff2))
1879 r21 = _mm_mul_pd(rsq21,rinv21);
1881 /* EWALD ELECTROSTATICS */
1883 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1884 ewrt = _mm_mul_pd(r21,ewtabscale);
1885 ewitab = _mm_cvttpd_epi32(ewrt);
1886 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1887 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1889 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1890 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1892 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1896 fscal = _mm_and_pd(fscal,cutoff_mask);
1898 /* Calculate temporary vectorial force */
1899 tx = _mm_mul_pd(fscal,dx21);
1900 ty = _mm_mul_pd(fscal,dy21);
1901 tz = _mm_mul_pd(fscal,dz21);
1903 /* Update vectorial force */
1904 fix2 = _mm_add_pd(fix2,tx);
1905 fiy2 = _mm_add_pd(fiy2,ty);
1906 fiz2 = _mm_add_pd(fiz2,tz);
1908 fjx1 = _mm_add_pd(fjx1,tx);
1909 fjy1 = _mm_add_pd(fjy1,ty);
1910 fjz1 = _mm_add_pd(fjz1,tz);
1914 /**************************
1915 * CALCULATE INTERACTIONS *
1916 **************************/
1918 if (gmx_mm_any_lt(rsq22,rcutoff2))
1921 r22 = _mm_mul_pd(rsq22,rinv22);
1923 /* EWALD ELECTROSTATICS */
1925 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1926 ewrt = _mm_mul_pd(r22,ewtabscale);
1927 ewitab = _mm_cvttpd_epi32(ewrt);
1928 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1929 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1931 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1932 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1934 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1938 fscal = _mm_and_pd(fscal,cutoff_mask);
1940 /* Calculate temporary vectorial force */
1941 tx = _mm_mul_pd(fscal,dx22);
1942 ty = _mm_mul_pd(fscal,dy22);
1943 tz = _mm_mul_pd(fscal,dz22);
1945 /* Update vectorial force */
1946 fix2 = _mm_add_pd(fix2,tx);
1947 fiy2 = _mm_add_pd(fiy2,ty);
1948 fiz2 = _mm_add_pd(fiz2,tz);
1950 fjx2 = _mm_add_pd(fjx2,tx);
1951 fjy2 = _mm_add_pd(fjy2,ty);
1952 fjz2 = _mm_add_pd(fjz2,tz);
1956 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1958 /* Inner loop uses 351 flops */
1961 if(jidx<j_index_end)
1965 j_coord_offsetA = DIM*jnrA;
1967 /* load j atom coordinates */
1968 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1969 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1971 /* Calculate displacement vector */
1972 dx00 = _mm_sub_pd(ix0,jx0);
1973 dy00 = _mm_sub_pd(iy0,jy0);
1974 dz00 = _mm_sub_pd(iz0,jz0);
1975 dx01 = _mm_sub_pd(ix0,jx1);
1976 dy01 = _mm_sub_pd(iy0,jy1);
1977 dz01 = _mm_sub_pd(iz0,jz1);
1978 dx02 = _mm_sub_pd(ix0,jx2);
1979 dy02 = _mm_sub_pd(iy0,jy2);
1980 dz02 = _mm_sub_pd(iz0,jz2);
1981 dx10 = _mm_sub_pd(ix1,jx0);
1982 dy10 = _mm_sub_pd(iy1,jy0);
1983 dz10 = _mm_sub_pd(iz1,jz0);
1984 dx11 = _mm_sub_pd(ix1,jx1);
1985 dy11 = _mm_sub_pd(iy1,jy1);
1986 dz11 = _mm_sub_pd(iz1,jz1);
1987 dx12 = _mm_sub_pd(ix1,jx2);
1988 dy12 = _mm_sub_pd(iy1,jy2);
1989 dz12 = _mm_sub_pd(iz1,jz2);
1990 dx20 = _mm_sub_pd(ix2,jx0);
1991 dy20 = _mm_sub_pd(iy2,jy0);
1992 dz20 = _mm_sub_pd(iz2,jz0);
1993 dx21 = _mm_sub_pd(ix2,jx1);
1994 dy21 = _mm_sub_pd(iy2,jy1);
1995 dz21 = _mm_sub_pd(iz2,jz1);
1996 dx22 = _mm_sub_pd(ix2,jx2);
1997 dy22 = _mm_sub_pd(iy2,jy2);
1998 dz22 = _mm_sub_pd(iz2,jz2);
2000 /* Calculate squared distance and things based on it */
2001 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2002 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2003 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2004 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2005 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2006 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2007 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2008 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2009 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2011 rinv00 = gmx_mm_invsqrt_pd(rsq00);
2012 rinv01 = gmx_mm_invsqrt_pd(rsq01);
2013 rinv02 = gmx_mm_invsqrt_pd(rsq02);
2014 rinv10 = gmx_mm_invsqrt_pd(rsq10);
2015 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2016 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2017 rinv20 = gmx_mm_invsqrt_pd(rsq20);
2018 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2019 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2021 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
2022 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
2023 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
2024 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
2025 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
2026 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
2027 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
2028 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
2029 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
2031 fjx0 = _mm_setzero_pd();
2032 fjy0 = _mm_setzero_pd();
2033 fjz0 = _mm_setzero_pd();
2034 fjx1 = _mm_setzero_pd();
2035 fjy1 = _mm_setzero_pd();
2036 fjz1 = _mm_setzero_pd();
2037 fjx2 = _mm_setzero_pd();
2038 fjy2 = _mm_setzero_pd();
2039 fjz2 = _mm_setzero_pd();
2041 /**************************
2042 * CALCULATE INTERACTIONS *
2043 **************************/
2045 if (gmx_mm_any_lt(rsq00,rcutoff2))
2048 r00 = _mm_mul_pd(rsq00,rinv00);
2050 /* EWALD ELECTROSTATICS */
2052 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2053 ewrt = _mm_mul_pd(r00,ewtabscale);
2054 ewitab = _mm_cvttpd_epi32(ewrt);
2055 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2056 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2057 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2058 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2060 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
2064 fscal = _mm_and_pd(fscal,cutoff_mask);
2066 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2068 /* Calculate temporary vectorial force */
2069 tx = _mm_mul_pd(fscal,dx00);
2070 ty = _mm_mul_pd(fscal,dy00);
2071 tz = _mm_mul_pd(fscal,dz00);
2073 /* Update vectorial force */
2074 fix0 = _mm_add_pd(fix0,tx);
2075 fiy0 = _mm_add_pd(fiy0,ty);
2076 fiz0 = _mm_add_pd(fiz0,tz);
2078 fjx0 = _mm_add_pd(fjx0,tx);
2079 fjy0 = _mm_add_pd(fjy0,ty);
2080 fjz0 = _mm_add_pd(fjz0,tz);
2084 /**************************
2085 * CALCULATE INTERACTIONS *
2086 **************************/
2088 if (gmx_mm_any_lt(rsq01,rcutoff2))
2091 r01 = _mm_mul_pd(rsq01,rinv01);
2093 /* EWALD ELECTROSTATICS */
2095 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2096 ewrt = _mm_mul_pd(r01,ewtabscale);
2097 ewitab = _mm_cvttpd_epi32(ewrt);
2098 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2099 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2100 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2101 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2103 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
2107 fscal = _mm_and_pd(fscal,cutoff_mask);
2109 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2111 /* Calculate temporary vectorial force */
2112 tx = _mm_mul_pd(fscal,dx01);
2113 ty = _mm_mul_pd(fscal,dy01);
2114 tz = _mm_mul_pd(fscal,dz01);
2116 /* Update vectorial force */
2117 fix0 = _mm_add_pd(fix0,tx);
2118 fiy0 = _mm_add_pd(fiy0,ty);
2119 fiz0 = _mm_add_pd(fiz0,tz);
2121 fjx1 = _mm_add_pd(fjx1,tx);
2122 fjy1 = _mm_add_pd(fjy1,ty);
2123 fjz1 = _mm_add_pd(fjz1,tz);
2127 /**************************
2128 * CALCULATE INTERACTIONS *
2129 **************************/
2131 if (gmx_mm_any_lt(rsq02,rcutoff2))
2134 r02 = _mm_mul_pd(rsq02,rinv02);
2136 /* EWALD ELECTROSTATICS */
2138 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2139 ewrt = _mm_mul_pd(r02,ewtabscale);
2140 ewitab = _mm_cvttpd_epi32(ewrt);
2141 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2142 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2143 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2144 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2146 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
2150 fscal = _mm_and_pd(fscal,cutoff_mask);
2152 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2154 /* Calculate temporary vectorial force */
2155 tx = _mm_mul_pd(fscal,dx02);
2156 ty = _mm_mul_pd(fscal,dy02);
2157 tz = _mm_mul_pd(fscal,dz02);
2159 /* Update vectorial force */
2160 fix0 = _mm_add_pd(fix0,tx);
2161 fiy0 = _mm_add_pd(fiy0,ty);
2162 fiz0 = _mm_add_pd(fiz0,tz);
2164 fjx2 = _mm_add_pd(fjx2,tx);
2165 fjy2 = _mm_add_pd(fjy2,ty);
2166 fjz2 = _mm_add_pd(fjz2,tz);
2170 /**************************
2171 * CALCULATE INTERACTIONS *
2172 **************************/
2174 if (gmx_mm_any_lt(rsq10,rcutoff2))
2177 r10 = _mm_mul_pd(rsq10,rinv10);
2179 /* EWALD ELECTROSTATICS */
2181 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2182 ewrt = _mm_mul_pd(r10,ewtabscale);
2183 ewitab = _mm_cvttpd_epi32(ewrt);
2184 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2185 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2186 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2187 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2189 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
2193 fscal = _mm_and_pd(fscal,cutoff_mask);
2195 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2197 /* Calculate temporary vectorial force */
2198 tx = _mm_mul_pd(fscal,dx10);
2199 ty = _mm_mul_pd(fscal,dy10);
2200 tz = _mm_mul_pd(fscal,dz10);
2202 /* Update vectorial force */
2203 fix1 = _mm_add_pd(fix1,tx);
2204 fiy1 = _mm_add_pd(fiy1,ty);
2205 fiz1 = _mm_add_pd(fiz1,tz);
2207 fjx0 = _mm_add_pd(fjx0,tx);
2208 fjy0 = _mm_add_pd(fjy0,ty);
2209 fjz0 = _mm_add_pd(fjz0,tz);
2213 /**************************
2214 * CALCULATE INTERACTIONS *
2215 **************************/
2217 if (gmx_mm_any_lt(rsq11,rcutoff2))
2220 r11 = _mm_mul_pd(rsq11,rinv11);
2222 /* EWALD ELECTROSTATICS */
2224 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2225 ewrt = _mm_mul_pd(r11,ewtabscale);
2226 ewitab = _mm_cvttpd_epi32(ewrt);
2227 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2228 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2229 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2230 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2232 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2236 fscal = _mm_and_pd(fscal,cutoff_mask);
2238 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2240 /* Calculate temporary vectorial force */
2241 tx = _mm_mul_pd(fscal,dx11);
2242 ty = _mm_mul_pd(fscal,dy11);
2243 tz = _mm_mul_pd(fscal,dz11);
2245 /* Update vectorial force */
2246 fix1 = _mm_add_pd(fix1,tx);
2247 fiy1 = _mm_add_pd(fiy1,ty);
2248 fiz1 = _mm_add_pd(fiz1,tz);
2250 fjx1 = _mm_add_pd(fjx1,tx);
2251 fjy1 = _mm_add_pd(fjy1,ty);
2252 fjz1 = _mm_add_pd(fjz1,tz);
2256 /**************************
2257 * CALCULATE INTERACTIONS *
2258 **************************/
2260 if (gmx_mm_any_lt(rsq12,rcutoff2))
2263 r12 = _mm_mul_pd(rsq12,rinv12);
2265 /* EWALD ELECTROSTATICS */
2267 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2268 ewrt = _mm_mul_pd(r12,ewtabscale);
2269 ewitab = _mm_cvttpd_epi32(ewrt);
2270 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2271 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2272 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2273 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2275 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2279 fscal = _mm_and_pd(fscal,cutoff_mask);
2281 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2283 /* Calculate temporary vectorial force */
2284 tx = _mm_mul_pd(fscal,dx12);
2285 ty = _mm_mul_pd(fscal,dy12);
2286 tz = _mm_mul_pd(fscal,dz12);
2288 /* Update vectorial force */
2289 fix1 = _mm_add_pd(fix1,tx);
2290 fiy1 = _mm_add_pd(fiy1,ty);
2291 fiz1 = _mm_add_pd(fiz1,tz);
2293 fjx2 = _mm_add_pd(fjx2,tx);
2294 fjy2 = _mm_add_pd(fjy2,ty);
2295 fjz2 = _mm_add_pd(fjz2,tz);
2299 /**************************
2300 * CALCULATE INTERACTIONS *
2301 **************************/
2303 if (gmx_mm_any_lt(rsq20,rcutoff2))
2306 r20 = _mm_mul_pd(rsq20,rinv20);
2308 /* EWALD ELECTROSTATICS */
2310 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2311 ewrt = _mm_mul_pd(r20,ewtabscale);
2312 ewitab = _mm_cvttpd_epi32(ewrt);
2313 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2314 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2315 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2316 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2318 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
2322 fscal = _mm_and_pd(fscal,cutoff_mask);
2324 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2326 /* Calculate temporary vectorial force */
2327 tx = _mm_mul_pd(fscal,dx20);
2328 ty = _mm_mul_pd(fscal,dy20);
2329 tz = _mm_mul_pd(fscal,dz20);
2331 /* Update vectorial force */
2332 fix2 = _mm_add_pd(fix2,tx);
2333 fiy2 = _mm_add_pd(fiy2,ty);
2334 fiz2 = _mm_add_pd(fiz2,tz);
2336 fjx0 = _mm_add_pd(fjx0,tx);
2337 fjy0 = _mm_add_pd(fjy0,ty);
2338 fjz0 = _mm_add_pd(fjz0,tz);
2342 /**************************
2343 * CALCULATE INTERACTIONS *
2344 **************************/
2346 if (gmx_mm_any_lt(rsq21,rcutoff2))
2349 r21 = _mm_mul_pd(rsq21,rinv21);
2351 /* EWALD ELECTROSTATICS */
2353 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2354 ewrt = _mm_mul_pd(r21,ewtabscale);
2355 ewitab = _mm_cvttpd_epi32(ewrt);
2356 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2357 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2358 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2359 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2361 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2365 fscal = _mm_and_pd(fscal,cutoff_mask);
2367 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2369 /* Calculate temporary vectorial force */
2370 tx = _mm_mul_pd(fscal,dx21);
2371 ty = _mm_mul_pd(fscal,dy21);
2372 tz = _mm_mul_pd(fscal,dz21);
2374 /* Update vectorial force */
2375 fix2 = _mm_add_pd(fix2,tx);
2376 fiy2 = _mm_add_pd(fiy2,ty);
2377 fiz2 = _mm_add_pd(fiz2,tz);
2379 fjx1 = _mm_add_pd(fjx1,tx);
2380 fjy1 = _mm_add_pd(fjy1,ty);
2381 fjz1 = _mm_add_pd(fjz1,tz);
2385 /**************************
2386 * CALCULATE INTERACTIONS *
2387 **************************/
2389 if (gmx_mm_any_lt(rsq22,rcutoff2))
2392 r22 = _mm_mul_pd(rsq22,rinv22);
2394 /* EWALD ELECTROSTATICS */
2396 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2397 ewrt = _mm_mul_pd(r22,ewtabscale);
2398 ewitab = _mm_cvttpd_epi32(ewrt);
2399 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2400 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2401 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
2402 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2404 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2408 fscal = _mm_and_pd(fscal,cutoff_mask);
2410 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2412 /* Calculate temporary vectorial force */
2413 tx = _mm_mul_pd(fscal,dx22);
2414 ty = _mm_mul_pd(fscal,dy22);
2415 tz = _mm_mul_pd(fscal,dz22);
2417 /* Update vectorial force */
2418 fix2 = _mm_add_pd(fix2,tx);
2419 fiy2 = _mm_add_pd(fiy2,ty);
2420 fiz2 = _mm_add_pd(fiz2,tz);
2422 fjx2 = _mm_add_pd(fjx2,tx);
2423 fjy2 = _mm_add_pd(fjy2,ty);
2424 fjz2 = _mm_add_pd(fjz2,tz);
2428 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2430 /* Inner loop uses 351 flops */
2433 /* End of innermost loop */
2435 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2436 f+i_coord_offset,fshift+i_shift_offset);
2438 /* Increment number of inner iterations */
2439 inneriter += j_index_end - j_index_start;
2441 /* Outer loop uses 18 flops */
2444 /* Increment number of outer iterations */
2447 /* Update outer/inner flops */
2449 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*351);