2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_VF_avx_128_fma_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Water3
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_VF_avx_128_fma_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 int vdwjidx1A,vdwjidx1B;
91 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
92 int vdwjidx2A,vdwjidx2B;
93 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
94 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
95 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
96 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
97 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
98 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
99 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
100 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
101 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
102 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
103 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
106 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
109 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
110 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
112 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
114 __m128d dummy_mask,cutoff_mask;
115 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
116 __m128d one = _mm_set1_pd(1.0);
117 __m128d two = _mm_set1_pd(2.0);
123 jindex = nlist->jindex;
125 shiftidx = nlist->shift;
127 shiftvec = fr->shift_vec[0];
128 fshift = fr->fshift[0];
129 facel = _mm_set1_pd(fr->epsfac);
130 charge = mdatoms->chargeA;
131 nvdwtype = fr->ntype;
133 vdwtype = mdatoms->typeA;
135 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
136 ewtab = fr->ic->tabq_coul_FDV0;
137 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
138 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
140 /* Setup water-specific parameters */
141 inr = nlist->iinr[0];
142 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
143 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
144 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
145 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
147 jq0 = _mm_set1_pd(charge[inr+0]);
148 jq1 = _mm_set1_pd(charge[inr+1]);
149 jq2 = _mm_set1_pd(charge[inr+2]);
150 vdwjidx0A = 2*vdwtype[inr+0];
151 qq00 = _mm_mul_pd(iq0,jq0);
152 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
153 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
154 qq01 = _mm_mul_pd(iq0,jq1);
155 qq02 = _mm_mul_pd(iq0,jq2);
156 qq10 = _mm_mul_pd(iq1,jq0);
157 qq11 = _mm_mul_pd(iq1,jq1);
158 qq12 = _mm_mul_pd(iq1,jq2);
159 qq20 = _mm_mul_pd(iq2,jq0);
160 qq21 = _mm_mul_pd(iq2,jq1);
161 qq22 = _mm_mul_pd(iq2,jq2);
163 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
164 rcutoff_scalar = fr->rcoulomb;
165 rcutoff = _mm_set1_pd(rcutoff_scalar);
166 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
168 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
169 rvdw = _mm_set1_pd(fr->rvdw);
171 /* Avoid stupid compiler warnings */
179 /* Start outer loop over neighborlists */
180 for(iidx=0; iidx<nri; iidx++)
182 /* Load shift vector for this list */
183 i_shift_offset = DIM*shiftidx[iidx];
185 /* Load limits for loop over neighbors */
186 j_index_start = jindex[iidx];
187 j_index_end = jindex[iidx+1];
189 /* Get outer coordinate index */
191 i_coord_offset = DIM*inr;
193 /* Load i particle coords and add shift vector */
194 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
195 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
197 fix0 = _mm_setzero_pd();
198 fiy0 = _mm_setzero_pd();
199 fiz0 = _mm_setzero_pd();
200 fix1 = _mm_setzero_pd();
201 fiy1 = _mm_setzero_pd();
202 fiz1 = _mm_setzero_pd();
203 fix2 = _mm_setzero_pd();
204 fiy2 = _mm_setzero_pd();
205 fiz2 = _mm_setzero_pd();
207 /* Reset potential sums */
208 velecsum = _mm_setzero_pd();
209 vvdwsum = _mm_setzero_pd();
211 /* Start inner kernel loop */
212 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
215 /* Get j neighbor index, and coordinate index */
218 j_coord_offsetA = DIM*jnrA;
219 j_coord_offsetB = DIM*jnrB;
221 /* load j atom coordinates */
222 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
223 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
225 /* Calculate displacement vector */
226 dx00 = _mm_sub_pd(ix0,jx0);
227 dy00 = _mm_sub_pd(iy0,jy0);
228 dz00 = _mm_sub_pd(iz0,jz0);
229 dx01 = _mm_sub_pd(ix0,jx1);
230 dy01 = _mm_sub_pd(iy0,jy1);
231 dz01 = _mm_sub_pd(iz0,jz1);
232 dx02 = _mm_sub_pd(ix0,jx2);
233 dy02 = _mm_sub_pd(iy0,jy2);
234 dz02 = _mm_sub_pd(iz0,jz2);
235 dx10 = _mm_sub_pd(ix1,jx0);
236 dy10 = _mm_sub_pd(iy1,jy0);
237 dz10 = _mm_sub_pd(iz1,jz0);
238 dx11 = _mm_sub_pd(ix1,jx1);
239 dy11 = _mm_sub_pd(iy1,jy1);
240 dz11 = _mm_sub_pd(iz1,jz1);
241 dx12 = _mm_sub_pd(ix1,jx2);
242 dy12 = _mm_sub_pd(iy1,jy2);
243 dz12 = _mm_sub_pd(iz1,jz2);
244 dx20 = _mm_sub_pd(ix2,jx0);
245 dy20 = _mm_sub_pd(iy2,jy0);
246 dz20 = _mm_sub_pd(iz2,jz0);
247 dx21 = _mm_sub_pd(ix2,jx1);
248 dy21 = _mm_sub_pd(iy2,jy1);
249 dz21 = _mm_sub_pd(iz2,jz1);
250 dx22 = _mm_sub_pd(ix2,jx2);
251 dy22 = _mm_sub_pd(iy2,jy2);
252 dz22 = _mm_sub_pd(iz2,jz2);
254 /* Calculate squared distance and things based on it */
255 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
256 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
257 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
258 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
259 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
260 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
261 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
262 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
263 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
265 rinv00 = gmx_mm_invsqrt_pd(rsq00);
266 rinv01 = gmx_mm_invsqrt_pd(rsq01);
267 rinv02 = gmx_mm_invsqrt_pd(rsq02);
268 rinv10 = gmx_mm_invsqrt_pd(rsq10);
269 rinv11 = gmx_mm_invsqrt_pd(rsq11);
270 rinv12 = gmx_mm_invsqrt_pd(rsq12);
271 rinv20 = gmx_mm_invsqrt_pd(rsq20);
272 rinv21 = gmx_mm_invsqrt_pd(rsq21);
273 rinv22 = gmx_mm_invsqrt_pd(rsq22);
275 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
276 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
277 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
278 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
279 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
280 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
281 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
282 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
283 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
285 fjx0 = _mm_setzero_pd();
286 fjy0 = _mm_setzero_pd();
287 fjz0 = _mm_setzero_pd();
288 fjx1 = _mm_setzero_pd();
289 fjy1 = _mm_setzero_pd();
290 fjz1 = _mm_setzero_pd();
291 fjx2 = _mm_setzero_pd();
292 fjy2 = _mm_setzero_pd();
293 fjz2 = _mm_setzero_pd();
295 /**************************
296 * CALCULATE INTERACTIONS *
297 **************************/
299 if (gmx_mm_any_lt(rsq00,rcutoff2))
302 r00 = _mm_mul_pd(rsq00,rinv00);
304 /* EWALD ELECTROSTATICS */
306 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
307 ewrt = _mm_mul_pd(r00,ewtabscale);
308 ewitab = _mm_cvttpd_epi32(ewrt);
310 eweps = _mm_frcz_pd(ewrt);
312 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
314 twoeweps = _mm_add_pd(eweps,eweps);
315 ewitab = _mm_slli_epi32(ewitab,2);
316 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
317 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
318 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
319 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
320 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
321 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
322 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
323 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
324 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
325 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
327 /* LENNARD-JONES DISPERSION/REPULSION */
329 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
330 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
331 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
332 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
333 _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
334 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
336 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
338 /* Update potential sum for this i atom from the interaction with this j atom. */
339 velec = _mm_and_pd(velec,cutoff_mask);
340 velecsum = _mm_add_pd(velecsum,velec);
341 vvdw = _mm_and_pd(vvdw,cutoff_mask);
342 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
344 fscal = _mm_add_pd(felec,fvdw);
346 fscal = _mm_and_pd(fscal,cutoff_mask);
348 /* Update vectorial force */
349 fix0 = _mm_macc_pd(dx00,fscal,fix0);
350 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
351 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
353 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
354 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
355 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
359 /**************************
360 * CALCULATE INTERACTIONS *
361 **************************/
363 if (gmx_mm_any_lt(rsq01,rcutoff2))
366 r01 = _mm_mul_pd(rsq01,rinv01);
368 /* EWALD ELECTROSTATICS */
370 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
371 ewrt = _mm_mul_pd(r01,ewtabscale);
372 ewitab = _mm_cvttpd_epi32(ewrt);
374 eweps = _mm_frcz_pd(ewrt);
376 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
378 twoeweps = _mm_add_pd(eweps,eweps);
379 ewitab = _mm_slli_epi32(ewitab,2);
380 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
381 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
382 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
383 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
384 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
385 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
386 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
387 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
388 velec = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
389 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
391 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
393 /* Update potential sum for this i atom from the interaction with this j atom. */
394 velec = _mm_and_pd(velec,cutoff_mask);
395 velecsum = _mm_add_pd(velecsum,velec);
399 fscal = _mm_and_pd(fscal,cutoff_mask);
401 /* Update vectorial force */
402 fix0 = _mm_macc_pd(dx01,fscal,fix0);
403 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
404 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
406 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
407 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
408 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
412 /**************************
413 * CALCULATE INTERACTIONS *
414 **************************/
416 if (gmx_mm_any_lt(rsq02,rcutoff2))
419 r02 = _mm_mul_pd(rsq02,rinv02);
421 /* EWALD ELECTROSTATICS */
423 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
424 ewrt = _mm_mul_pd(r02,ewtabscale);
425 ewitab = _mm_cvttpd_epi32(ewrt);
427 eweps = _mm_frcz_pd(ewrt);
429 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
431 twoeweps = _mm_add_pd(eweps,eweps);
432 ewitab = _mm_slli_epi32(ewitab,2);
433 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
434 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
435 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
436 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
437 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
438 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
439 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
440 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
441 velec = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
442 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
444 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
446 /* Update potential sum for this i atom from the interaction with this j atom. */
447 velec = _mm_and_pd(velec,cutoff_mask);
448 velecsum = _mm_add_pd(velecsum,velec);
452 fscal = _mm_and_pd(fscal,cutoff_mask);
454 /* Update vectorial force */
455 fix0 = _mm_macc_pd(dx02,fscal,fix0);
456 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
457 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
459 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
460 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
461 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
465 /**************************
466 * CALCULATE INTERACTIONS *
467 **************************/
469 if (gmx_mm_any_lt(rsq10,rcutoff2))
472 r10 = _mm_mul_pd(rsq10,rinv10);
474 /* EWALD ELECTROSTATICS */
476 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
477 ewrt = _mm_mul_pd(r10,ewtabscale);
478 ewitab = _mm_cvttpd_epi32(ewrt);
480 eweps = _mm_frcz_pd(ewrt);
482 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
484 twoeweps = _mm_add_pd(eweps,eweps);
485 ewitab = _mm_slli_epi32(ewitab,2);
486 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
487 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
488 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
489 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
490 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
491 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
492 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
493 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
494 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
495 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
497 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
499 /* Update potential sum for this i atom from the interaction with this j atom. */
500 velec = _mm_and_pd(velec,cutoff_mask);
501 velecsum = _mm_add_pd(velecsum,velec);
505 fscal = _mm_and_pd(fscal,cutoff_mask);
507 /* Update vectorial force */
508 fix1 = _mm_macc_pd(dx10,fscal,fix1);
509 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
510 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
512 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
513 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
514 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
518 /**************************
519 * CALCULATE INTERACTIONS *
520 **************************/
522 if (gmx_mm_any_lt(rsq11,rcutoff2))
525 r11 = _mm_mul_pd(rsq11,rinv11);
527 /* EWALD ELECTROSTATICS */
529 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
530 ewrt = _mm_mul_pd(r11,ewtabscale);
531 ewitab = _mm_cvttpd_epi32(ewrt);
533 eweps = _mm_frcz_pd(ewrt);
535 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
537 twoeweps = _mm_add_pd(eweps,eweps);
538 ewitab = _mm_slli_epi32(ewitab,2);
539 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
540 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
541 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
542 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
543 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
544 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
545 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
546 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
547 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
548 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
550 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
552 /* Update potential sum for this i atom from the interaction with this j atom. */
553 velec = _mm_and_pd(velec,cutoff_mask);
554 velecsum = _mm_add_pd(velecsum,velec);
558 fscal = _mm_and_pd(fscal,cutoff_mask);
560 /* Update vectorial force */
561 fix1 = _mm_macc_pd(dx11,fscal,fix1);
562 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
563 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
565 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
566 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
567 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
571 /**************************
572 * CALCULATE INTERACTIONS *
573 **************************/
575 if (gmx_mm_any_lt(rsq12,rcutoff2))
578 r12 = _mm_mul_pd(rsq12,rinv12);
580 /* EWALD ELECTROSTATICS */
582 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
583 ewrt = _mm_mul_pd(r12,ewtabscale);
584 ewitab = _mm_cvttpd_epi32(ewrt);
586 eweps = _mm_frcz_pd(ewrt);
588 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
590 twoeweps = _mm_add_pd(eweps,eweps);
591 ewitab = _mm_slli_epi32(ewitab,2);
592 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
593 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
594 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
595 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
596 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
597 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
598 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
599 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
600 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
601 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
603 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
605 /* Update potential sum for this i atom from the interaction with this j atom. */
606 velec = _mm_and_pd(velec,cutoff_mask);
607 velecsum = _mm_add_pd(velecsum,velec);
611 fscal = _mm_and_pd(fscal,cutoff_mask);
613 /* Update vectorial force */
614 fix1 = _mm_macc_pd(dx12,fscal,fix1);
615 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
616 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
618 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
619 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
620 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
624 /**************************
625 * CALCULATE INTERACTIONS *
626 **************************/
628 if (gmx_mm_any_lt(rsq20,rcutoff2))
631 r20 = _mm_mul_pd(rsq20,rinv20);
633 /* EWALD ELECTROSTATICS */
635 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
636 ewrt = _mm_mul_pd(r20,ewtabscale);
637 ewitab = _mm_cvttpd_epi32(ewrt);
639 eweps = _mm_frcz_pd(ewrt);
641 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
643 twoeweps = _mm_add_pd(eweps,eweps);
644 ewitab = _mm_slli_epi32(ewitab,2);
645 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
646 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
647 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
648 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
649 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
650 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
651 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
652 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
653 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
654 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
656 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
658 /* Update potential sum for this i atom from the interaction with this j atom. */
659 velec = _mm_and_pd(velec,cutoff_mask);
660 velecsum = _mm_add_pd(velecsum,velec);
664 fscal = _mm_and_pd(fscal,cutoff_mask);
666 /* Update vectorial force */
667 fix2 = _mm_macc_pd(dx20,fscal,fix2);
668 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
669 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
671 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
672 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
673 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
677 /**************************
678 * CALCULATE INTERACTIONS *
679 **************************/
681 if (gmx_mm_any_lt(rsq21,rcutoff2))
684 r21 = _mm_mul_pd(rsq21,rinv21);
686 /* EWALD ELECTROSTATICS */
688 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
689 ewrt = _mm_mul_pd(r21,ewtabscale);
690 ewitab = _mm_cvttpd_epi32(ewrt);
692 eweps = _mm_frcz_pd(ewrt);
694 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
696 twoeweps = _mm_add_pd(eweps,eweps);
697 ewitab = _mm_slli_epi32(ewitab,2);
698 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
699 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
700 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
701 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
702 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
703 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
704 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
705 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
706 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
707 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
709 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
711 /* Update potential sum for this i atom from the interaction with this j atom. */
712 velec = _mm_and_pd(velec,cutoff_mask);
713 velecsum = _mm_add_pd(velecsum,velec);
717 fscal = _mm_and_pd(fscal,cutoff_mask);
719 /* Update vectorial force */
720 fix2 = _mm_macc_pd(dx21,fscal,fix2);
721 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
722 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
724 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
725 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
726 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
730 /**************************
731 * CALCULATE INTERACTIONS *
732 **************************/
734 if (gmx_mm_any_lt(rsq22,rcutoff2))
737 r22 = _mm_mul_pd(rsq22,rinv22);
739 /* EWALD ELECTROSTATICS */
741 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
742 ewrt = _mm_mul_pd(r22,ewtabscale);
743 ewitab = _mm_cvttpd_epi32(ewrt);
745 eweps = _mm_frcz_pd(ewrt);
747 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
749 twoeweps = _mm_add_pd(eweps,eweps);
750 ewitab = _mm_slli_epi32(ewitab,2);
751 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
752 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
753 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
754 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
755 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
756 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
757 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
758 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
759 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
760 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
762 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
764 /* Update potential sum for this i atom from the interaction with this j atom. */
765 velec = _mm_and_pd(velec,cutoff_mask);
766 velecsum = _mm_add_pd(velecsum,velec);
770 fscal = _mm_and_pd(fscal,cutoff_mask);
772 /* Update vectorial force */
773 fix2 = _mm_macc_pd(dx22,fscal,fix2);
774 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
775 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
777 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
778 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
779 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
783 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
785 /* Inner loop uses 459 flops */
792 j_coord_offsetA = DIM*jnrA;
794 /* load j atom coordinates */
795 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
796 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
798 /* Calculate displacement vector */
799 dx00 = _mm_sub_pd(ix0,jx0);
800 dy00 = _mm_sub_pd(iy0,jy0);
801 dz00 = _mm_sub_pd(iz0,jz0);
802 dx01 = _mm_sub_pd(ix0,jx1);
803 dy01 = _mm_sub_pd(iy0,jy1);
804 dz01 = _mm_sub_pd(iz0,jz1);
805 dx02 = _mm_sub_pd(ix0,jx2);
806 dy02 = _mm_sub_pd(iy0,jy2);
807 dz02 = _mm_sub_pd(iz0,jz2);
808 dx10 = _mm_sub_pd(ix1,jx0);
809 dy10 = _mm_sub_pd(iy1,jy0);
810 dz10 = _mm_sub_pd(iz1,jz0);
811 dx11 = _mm_sub_pd(ix1,jx1);
812 dy11 = _mm_sub_pd(iy1,jy1);
813 dz11 = _mm_sub_pd(iz1,jz1);
814 dx12 = _mm_sub_pd(ix1,jx2);
815 dy12 = _mm_sub_pd(iy1,jy2);
816 dz12 = _mm_sub_pd(iz1,jz2);
817 dx20 = _mm_sub_pd(ix2,jx0);
818 dy20 = _mm_sub_pd(iy2,jy0);
819 dz20 = _mm_sub_pd(iz2,jz0);
820 dx21 = _mm_sub_pd(ix2,jx1);
821 dy21 = _mm_sub_pd(iy2,jy1);
822 dz21 = _mm_sub_pd(iz2,jz1);
823 dx22 = _mm_sub_pd(ix2,jx2);
824 dy22 = _mm_sub_pd(iy2,jy2);
825 dz22 = _mm_sub_pd(iz2,jz2);
827 /* Calculate squared distance and things based on it */
828 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
829 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
830 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
831 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
832 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
833 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
834 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
835 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
836 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
838 rinv00 = gmx_mm_invsqrt_pd(rsq00);
839 rinv01 = gmx_mm_invsqrt_pd(rsq01);
840 rinv02 = gmx_mm_invsqrt_pd(rsq02);
841 rinv10 = gmx_mm_invsqrt_pd(rsq10);
842 rinv11 = gmx_mm_invsqrt_pd(rsq11);
843 rinv12 = gmx_mm_invsqrt_pd(rsq12);
844 rinv20 = gmx_mm_invsqrt_pd(rsq20);
845 rinv21 = gmx_mm_invsqrt_pd(rsq21);
846 rinv22 = gmx_mm_invsqrt_pd(rsq22);
848 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
849 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
850 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
851 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
852 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
853 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
854 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
855 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
856 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
858 fjx0 = _mm_setzero_pd();
859 fjy0 = _mm_setzero_pd();
860 fjz0 = _mm_setzero_pd();
861 fjx1 = _mm_setzero_pd();
862 fjy1 = _mm_setzero_pd();
863 fjz1 = _mm_setzero_pd();
864 fjx2 = _mm_setzero_pd();
865 fjy2 = _mm_setzero_pd();
866 fjz2 = _mm_setzero_pd();
868 /**************************
869 * CALCULATE INTERACTIONS *
870 **************************/
872 if (gmx_mm_any_lt(rsq00,rcutoff2))
875 r00 = _mm_mul_pd(rsq00,rinv00);
877 /* EWALD ELECTROSTATICS */
879 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
880 ewrt = _mm_mul_pd(r00,ewtabscale);
881 ewitab = _mm_cvttpd_epi32(ewrt);
883 eweps = _mm_frcz_pd(ewrt);
885 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
887 twoeweps = _mm_add_pd(eweps,eweps);
888 ewitab = _mm_slli_epi32(ewitab,2);
889 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
890 ewtabD = _mm_setzero_pd();
891 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
892 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
893 ewtabFn = _mm_setzero_pd();
894 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
895 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
896 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
897 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
898 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
900 /* LENNARD-JONES DISPERSION/REPULSION */
902 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
903 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
904 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
905 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
906 _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
907 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
909 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
911 /* Update potential sum for this i atom from the interaction with this j atom. */
912 velec = _mm_and_pd(velec,cutoff_mask);
913 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
914 velecsum = _mm_add_pd(velecsum,velec);
915 vvdw = _mm_and_pd(vvdw,cutoff_mask);
916 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
917 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
919 fscal = _mm_add_pd(felec,fvdw);
921 fscal = _mm_and_pd(fscal,cutoff_mask);
923 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
925 /* Update vectorial force */
926 fix0 = _mm_macc_pd(dx00,fscal,fix0);
927 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
928 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
930 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
931 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
932 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
936 /**************************
937 * CALCULATE INTERACTIONS *
938 **************************/
940 if (gmx_mm_any_lt(rsq01,rcutoff2))
943 r01 = _mm_mul_pd(rsq01,rinv01);
945 /* EWALD ELECTROSTATICS */
947 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
948 ewrt = _mm_mul_pd(r01,ewtabscale);
949 ewitab = _mm_cvttpd_epi32(ewrt);
951 eweps = _mm_frcz_pd(ewrt);
953 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
955 twoeweps = _mm_add_pd(eweps,eweps);
956 ewitab = _mm_slli_epi32(ewitab,2);
957 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
958 ewtabD = _mm_setzero_pd();
959 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
960 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
961 ewtabFn = _mm_setzero_pd();
962 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
963 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
964 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
965 velec = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
966 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
968 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
970 /* Update potential sum for this i atom from the interaction with this j atom. */
971 velec = _mm_and_pd(velec,cutoff_mask);
972 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
973 velecsum = _mm_add_pd(velecsum,velec);
977 fscal = _mm_and_pd(fscal,cutoff_mask);
979 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
981 /* Update vectorial force */
982 fix0 = _mm_macc_pd(dx01,fscal,fix0);
983 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
984 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
986 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
987 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
988 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
992 /**************************
993 * CALCULATE INTERACTIONS *
994 **************************/
996 if (gmx_mm_any_lt(rsq02,rcutoff2))
999 r02 = _mm_mul_pd(rsq02,rinv02);
1001 /* EWALD ELECTROSTATICS */
1003 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1004 ewrt = _mm_mul_pd(r02,ewtabscale);
1005 ewitab = _mm_cvttpd_epi32(ewrt);
1007 eweps = _mm_frcz_pd(ewrt);
1009 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1011 twoeweps = _mm_add_pd(eweps,eweps);
1012 ewitab = _mm_slli_epi32(ewitab,2);
1013 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1014 ewtabD = _mm_setzero_pd();
1015 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1016 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1017 ewtabFn = _mm_setzero_pd();
1018 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1019 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1020 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1021 velec = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
1022 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1024 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1026 /* Update potential sum for this i atom from the interaction with this j atom. */
1027 velec = _mm_and_pd(velec,cutoff_mask);
1028 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1029 velecsum = _mm_add_pd(velecsum,velec);
1033 fscal = _mm_and_pd(fscal,cutoff_mask);
1035 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1037 /* Update vectorial force */
1038 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1039 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1040 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1042 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1043 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1044 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1048 /**************************
1049 * CALCULATE INTERACTIONS *
1050 **************************/
1052 if (gmx_mm_any_lt(rsq10,rcutoff2))
1055 r10 = _mm_mul_pd(rsq10,rinv10);
1057 /* EWALD ELECTROSTATICS */
1059 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060 ewrt = _mm_mul_pd(r10,ewtabscale);
1061 ewitab = _mm_cvttpd_epi32(ewrt);
1063 eweps = _mm_frcz_pd(ewrt);
1065 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1067 twoeweps = _mm_add_pd(eweps,eweps);
1068 ewitab = _mm_slli_epi32(ewitab,2);
1069 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1070 ewtabD = _mm_setzero_pd();
1071 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1072 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1073 ewtabFn = _mm_setzero_pd();
1074 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1075 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1076 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1077 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
1078 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1080 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1082 /* Update potential sum for this i atom from the interaction with this j atom. */
1083 velec = _mm_and_pd(velec,cutoff_mask);
1084 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1085 velecsum = _mm_add_pd(velecsum,velec);
1089 fscal = _mm_and_pd(fscal,cutoff_mask);
1091 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1093 /* Update vectorial force */
1094 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1095 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1096 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1098 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1099 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1100 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1104 /**************************
1105 * CALCULATE INTERACTIONS *
1106 **************************/
1108 if (gmx_mm_any_lt(rsq11,rcutoff2))
1111 r11 = _mm_mul_pd(rsq11,rinv11);
1113 /* EWALD ELECTROSTATICS */
1115 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116 ewrt = _mm_mul_pd(r11,ewtabscale);
1117 ewitab = _mm_cvttpd_epi32(ewrt);
1119 eweps = _mm_frcz_pd(ewrt);
1121 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1123 twoeweps = _mm_add_pd(eweps,eweps);
1124 ewitab = _mm_slli_epi32(ewitab,2);
1125 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1126 ewtabD = _mm_setzero_pd();
1127 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1128 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1129 ewtabFn = _mm_setzero_pd();
1130 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1131 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1132 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1133 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
1134 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1136 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1138 /* Update potential sum for this i atom from the interaction with this j atom. */
1139 velec = _mm_and_pd(velec,cutoff_mask);
1140 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1141 velecsum = _mm_add_pd(velecsum,velec);
1145 fscal = _mm_and_pd(fscal,cutoff_mask);
1147 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1149 /* Update vectorial force */
1150 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1151 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1152 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1154 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1155 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1156 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1160 /**************************
1161 * CALCULATE INTERACTIONS *
1162 **************************/
1164 if (gmx_mm_any_lt(rsq12,rcutoff2))
1167 r12 = _mm_mul_pd(rsq12,rinv12);
1169 /* EWALD ELECTROSTATICS */
1171 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1172 ewrt = _mm_mul_pd(r12,ewtabscale);
1173 ewitab = _mm_cvttpd_epi32(ewrt);
1175 eweps = _mm_frcz_pd(ewrt);
1177 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1179 twoeweps = _mm_add_pd(eweps,eweps);
1180 ewitab = _mm_slli_epi32(ewitab,2);
1181 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1182 ewtabD = _mm_setzero_pd();
1183 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1184 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1185 ewtabFn = _mm_setzero_pd();
1186 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1187 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1188 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1189 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
1190 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1192 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1194 /* Update potential sum for this i atom from the interaction with this j atom. */
1195 velec = _mm_and_pd(velec,cutoff_mask);
1196 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1197 velecsum = _mm_add_pd(velecsum,velec);
1201 fscal = _mm_and_pd(fscal,cutoff_mask);
1203 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1205 /* Update vectorial force */
1206 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1207 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1208 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1210 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1211 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1212 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1216 /**************************
1217 * CALCULATE INTERACTIONS *
1218 **************************/
1220 if (gmx_mm_any_lt(rsq20,rcutoff2))
1223 r20 = _mm_mul_pd(rsq20,rinv20);
1225 /* EWALD ELECTROSTATICS */
1227 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228 ewrt = _mm_mul_pd(r20,ewtabscale);
1229 ewitab = _mm_cvttpd_epi32(ewrt);
1231 eweps = _mm_frcz_pd(ewrt);
1233 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1235 twoeweps = _mm_add_pd(eweps,eweps);
1236 ewitab = _mm_slli_epi32(ewitab,2);
1237 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1238 ewtabD = _mm_setzero_pd();
1239 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1240 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1241 ewtabFn = _mm_setzero_pd();
1242 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1243 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1244 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1245 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
1246 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1248 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1250 /* Update potential sum for this i atom from the interaction with this j atom. */
1251 velec = _mm_and_pd(velec,cutoff_mask);
1252 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1253 velecsum = _mm_add_pd(velecsum,velec);
1257 fscal = _mm_and_pd(fscal,cutoff_mask);
1259 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1261 /* Update vectorial force */
1262 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1263 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1264 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1266 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1267 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1268 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1272 /**************************
1273 * CALCULATE INTERACTIONS *
1274 **************************/
1276 if (gmx_mm_any_lt(rsq21,rcutoff2))
1279 r21 = _mm_mul_pd(rsq21,rinv21);
1281 /* EWALD ELECTROSTATICS */
1283 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1284 ewrt = _mm_mul_pd(r21,ewtabscale);
1285 ewitab = _mm_cvttpd_epi32(ewrt);
1287 eweps = _mm_frcz_pd(ewrt);
1289 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1291 twoeweps = _mm_add_pd(eweps,eweps);
1292 ewitab = _mm_slli_epi32(ewitab,2);
1293 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1294 ewtabD = _mm_setzero_pd();
1295 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1296 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1297 ewtabFn = _mm_setzero_pd();
1298 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1299 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1300 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1301 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1302 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1304 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1306 /* Update potential sum for this i atom from the interaction with this j atom. */
1307 velec = _mm_and_pd(velec,cutoff_mask);
1308 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1309 velecsum = _mm_add_pd(velecsum,velec);
1313 fscal = _mm_and_pd(fscal,cutoff_mask);
1315 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1317 /* Update vectorial force */
1318 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1319 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1320 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1322 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1323 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1324 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1328 /**************************
1329 * CALCULATE INTERACTIONS *
1330 **************************/
1332 if (gmx_mm_any_lt(rsq22,rcutoff2))
1335 r22 = _mm_mul_pd(rsq22,rinv22);
1337 /* EWALD ELECTROSTATICS */
1339 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1340 ewrt = _mm_mul_pd(r22,ewtabscale);
1341 ewitab = _mm_cvttpd_epi32(ewrt);
1343 eweps = _mm_frcz_pd(ewrt);
1345 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1347 twoeweps = _mm_add_pd(eweps,eweps);
1348 ewitab = _mm_slli_epi32(ewitab,2);
1349 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1350 ewtabD = _mm_setzero_pd();
1351 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1352 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1353 ewtabFn = _mm_setzero_pd();
1354 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1355 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1356 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1357 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1358 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1360 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1362 /* Update potential sum for this i atom from the interaction with this j atom. */
1363 velec = _mm_and_pd(velec,cutoff_mask);
1364 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1365 velecsum = _mm_add_pd(velecsum,velec);
1369 fscal = _mm_and_pd(fscal,cutoff_mask);
1371 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1373 /* Update vectorial force */
1374 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1375 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1376 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1378 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1379 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1380 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1384 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1386 /* Inner loop uses 459 flops */
1389 /* End of innermost loop */
1391 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1392 f+i_coord_offset,fshift+i_shift_offset);
1395 /* Update potential energies */
1396 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1397 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1399 /* Increment number of inner iterations */
1400 inneriter += j_index_end - j_index_start;
1402 /* Outer loop uses 20 flops */
1405 /* Increment number of outer iterations */
1408 /* Update outer/inner flops */
1410 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*459);
1413 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_F_avx_128_fma_double
1414 * Electrostatics interaction: Ewald
1415 * VdW interaction: LennardJones
1416 * Geometry: Water3-Water3
1417 * Calculate force/pot: Force
1420 nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_F_avx_128_fma_double
1421 (t_nblist * gmx_restrict nlist,
1422 rvec * gmx_restrict xx,
1423 rvec * gmx_restrict ff,
1424 t_forcerec * gmx_restrict fr,
1425 t_mdatoms * gmx_restrict mdatoms,
1426 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1427 t_nrnb * gmx_restrict nrnb)
1429 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1430 * just 0 for non-waters.
1431 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1432 * jnr indices corresponding to data put in the four positions in the SIMD register.
1434 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1435 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1437 int j_coord_offsetA,j_coord_offsetB;
1438 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1439 real rcutoff_scalar;
1440 real *shiftvec,*fshift,*x,*f;
1441 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1443 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1445 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1447 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1448 int vdwjidx0A,vdwjidx0B;
1449 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1450 int vdwjidx1A,vdwjidx1B;
1451 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1452 int vdwjidx2A,vdwjidx2B;
1453 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1454 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1455 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1456 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1457 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1458 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1459 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1460 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1461 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1462 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1463 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1466 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1469 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
1470 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
1472 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1474 __m128d dummy_mask,cutoff_mask;
1475 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1476 __m128d one = _mm_set1_pd(1.0);
1477 __m128d two = _mm_set1_pd(2.0);
1483 jindex = nlist->jindex;
1485 shiftidx = nlist->shift;
1487 shiftvec = fr->shift_vec[0];
1488 fshift = fr->fshift[0];
1489 facel = _mm_set1_pd(fr->epsfac);
1490 charge = mdatoms->chargeA;
1491 nvdwtype = fr->ntype;
1492 vdwparam = fr->nbfp;
1493 vdwtype = mdatoms->typeA;
1495 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1496 ewtab = fr->ic->tabq_coul_F;
1497 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1498 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1500 /* Setup water-specific parameters */
1501 inr = nlist->iinr[0];
1502 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1503 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1504 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1505 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1507 jq0 = _mm_set1_pd(charge[inr+0]);
1508 jq1 = _mm_set1_pd(charge[inr+1]);
1509 jq2 = _mm_set1_pd(charge[inr+2]);
1510 vdwjidx0A = 2*vdwtype[inr+0];
1511 qq00 = _mm_mul_pd(iq0,jq0);
1512 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1513 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1514 qq01 = _mm_mul_pd(iq0,jq1);
1515 qq02 = _mm_mul_pd(iq0,jq2);
1516 qq10 = _mm_mul_pd(iq1,jq0);
1517 qq11 = _mm_mul_pd(iq1,jq1);
1518 qq12 = _mm_mul_pd(iq1,jq2);
1519 qq20 = _mm_mul_pd(iq2,jq0);
1520 qq21 = _mm_mul_pd(iq2,jq1);
1521 qq22 = _mm_mul_pd(iq2,jq2);
1523 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1524 rcutoff_scalar = fr->rcoulomb;
1525 rcutoff = _mm_set1_pd(rcutoff_scalar);
1526 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
1528 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
1529 rvdw = _mm_set1_pd(fr->rvdw);
1531 /* Avoid stupid compiler warnings */
1533 j_coord_offsetA = 0;
1534 j_coord_offsetB = 0;
1539 /* Start outer loop over neighborlists */
1540 for(iidx=0; iidx<nri; iidx++)
1542 /* Load shift vector for this list */
1543 i_shift_offset = DIM*shiftidx[iidx];
1545 /* Load limits for loop over neighbors */
1546 j_index_start = jindex[iidx];
1547 j_index_end = jindex[iidx+1];
1549 /* Get outer coordinate index */
1551 i_coord_offset = DIM*inr;
1553 /* Load i particle coords and add shift vector */
1554 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1555 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1557 fix0 = _mm_setzero_pd();
1558 fiy0 = _mm_setzero_pd();
1559 fiz0 = _mm_setzero_pd();
1560 fix1 = _mm_setzero_pd();
1561 fiy1 = _mm_setzero_pd();
1562 fiz1 = _mm_setzero_pd();
1563 fix2 = _mm_setzero_pd();
1564 fiy2 = _mm_setzero_pd();
1565 fiz2 = _mm_setzero_pd();
1567 /* Start inner kernel loop */
1568 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1571 /* Get j neighbor index, and coordinate index */
1573 jnrB = jjnr[jidx+1];
1574 j_coord_offsetA = DIM*jnrA;
1575 j_coord_offsetB = DIM*jnrB;
1577 /* load j atom coordinates */
1578 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1579 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1581 /* Calculate displacement vector */
1582 dx00 = _mm_sub_pd(ix0,jx0);
1583 dy00 = _mm_sub_pd(iy0,jy0);
1584 dz00 = _mm_sub_pd(iz0,jz0);
1585 dx01 = _mm_sub_pd(ix0,jx1);
1586 dy01 = _mm_sub_pd(iy0,jy1);
1587 dz01 = _mm_sub_pd(iz0,jz1);
1588 dx02 = _mm_sub_pd(ix0,jx2);
1589 dy02 = _mm_sub_pd(iy0,jy2);
1590 dz02 = _mm_sub_pd(iz0,jz2);
1591 dx10 = _mm_sub_pd(ix1,jx0);
1592 dy10 = _mm_sub_pd(iy1,jy0);
1593 dz10 = _mm_sub_pd(iz1,jz0);
1594 dx11 = _mm_sub_pd(ix1,jx1);
1595 dy11 = _mm_sub_pd(iy1,jy1);
1596 dz11 = _mm_sub_pd(iz1,jz1);
1597 dx12 = _mm_sub_pd(ix1,jx2);
1598 dy12 = _mm_sub_pd(iy1,jy2);
1599 dz12 = _mm_sub_pd(iz1,jz2);
1600 dx20 = _mm_sub_pd(ix2,jx0);
1601 dy20 = _mm_sub_pd(iy2,jy0);
1602 dz20 = _mm_sub_pd(iz2,jz0);
1603 dx21 = _mm_sub_pd(ix2,jx1);
1604 dy21 = _mm_sub_pd(iy2,jy1);
1605 dz21 = _mm_sub_pd(iz2,jz1);
1606 dx22 = _mm_sub_pd(ix2,jx2);
1607 dy22 = _mm_sub_pd(iy2,jy2);
1608 dz22 = _mm_sub_pd(iz2,jz2);
1610 /* Calculate squared distance and things based on it */
1611 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1612 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1613 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1614 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1615 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1616 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1617 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1618 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1619 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1621 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1622 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1623 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1624 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1625 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1626 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1627 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1628 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1629 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1631 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1632 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
1633 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
1634 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1635 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1636 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1637 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1638 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1639 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1641 fjx0 = _mm_setzero_pd();
1642 fjy0 = _mm_setzero_pd();
1643 fjz0 = _mm_setzero_pd();
1644 fjx1 = _mm_setzero_pd();
1645 fjy1 = _mm_setzero_pd();
1646 fjz1 = _mm_setzero_pd();
1647 fjx2 = _mm_setzero_pd();
1648 fjy2 = _mm_setzero_pd();
1649 fjz2 = _mm_setzero_pd();
1651 /**************************
1652 * CALCULATE INTERACTIONS *
1653 **************************/
1655 if (gmx_mm_any_lt(rsq00,rcutoff2))
1658 r00 = _mm_mul_pd(rsq00,rinv00);
1660 /* EWALD ELECTROSTATICS */
1662 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1663 ewrt = _mm_mul_pd(r00,ewtabscale);
1664 ewitab = _mm_cvttpd_epi32(ewrt);
1666 eweps = _mm_frcz_pd(ewrt);
1668 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1670 twoeweps = _mm_add_pd(eweps,eweps);
1671 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1673 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1674 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1676 /* LENNARD-JONES DISPERSION/REPULSION */
1678 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1679 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1681 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1683 fscal = _mm_add_pd(felec,fvdw);
1685 fscal = _mm_and_pd(fscal,cutoff_mask);
1687 /* Update vectorial force */
1688 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1689 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1690 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1692 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1693 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1694 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1698 /**************************
1699 * CALCULATE INTERACTIONS *
1700 **************************/
1702 if (gmx_mm_any_lt(rsq01,rcutoff2))
1705 r01 = _mm_mul_pd(rsq01,rinv01);
1707 /* EWALD ELECTROSTATICS */
1709 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1710 ewrt = _mm_mul_pd(r01,ewtabscale);
1711 ewitab = _mm_cvttpd_epi32(ewrt);
1713 eweps = _mm_frcz_pd(ewrt);
1715 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1717 twoeweps = _mm_add_pd(eweps,eweps);
1718 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1720 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1721 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1723 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
1727 fscal = _mm_and_pd(fscal,cutoff_mask);
1729 /* Update vectorial force */
1730 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1731 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1732 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1734 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1735 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1736 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1740 /**************************
1741 * CALCULATE INTERACTIONS *
1742 **************************/
1744 if (gmx_mm_any_lt(rsq02,rcutoff2))
1747 r02 = _mm_mul_pd(rsq02,rinv02);
1749 /* EWALD ELECTROSTATICS */
1751 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1752 ewrt = _mm_mul_pd(r02,ewtabscale);
1753 ewitab = _mm_cvttpd_epi32(ewrt);
1755 eweps = _mm_frcz_pd(ewrt);
1757 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1759 twoeweps = _mm_add_pd(eweps,eweps);
1760 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1762 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1763 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1765 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1769 fscal = _mm_and_pd(fscal,cutoff_mask);
1771 /* Update vectorial force */
1772 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1773 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1774 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1776 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1777 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1778 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1782 /**************************
1783 * CALCULATE INTERACTIONS *
1784 **************************/
1786 if (gmx_mm_any_lt(rsq10,rcutoff2))
1789 r10 = _mm_mul_pd(rsq10,rinv10);
1791 /* EWALD ELECTROSTATICS */
1793 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1794 ewrt = _mm_mul_pd(r10,ewtabscale);
1795 ewitab = _mm_cvttpd_epi32(ewrt);
1797 eweps = _mm_frcz_pd(ewrt);
1799 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1801 twoeweps = _mm_add_pd(eweps,eweps);
1802 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1804 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1805 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1807 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1811 fscal = _mm_and_pd(fscal,cutoff_mask);
1813 /* Update vectorial force */
1814 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1815 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1816 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1818 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1819 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1820 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1824 /**************************
1825 * CALCULATE INTERACTIONS *
1826 **************************/
1828 if (gmx_mm_any_lt(rsq11,rcutoff2))
1831 r11 = _mm_mul_pd(rsq11,rinv11);
1833 /* EWALD ELECTROSTATICS */
1835 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1836 ewrt = _mm_mul_pd(r11,ewtabscale);
1837 ewitab = _mm_cvttpd_epi32(ewrt);
1839 eweps = _mm_frcz_pd(ewrt);
1841 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1843 twoeweps = _mm_add_pd(eweps,eweps);
1844 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1846 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1847 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1849 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1853 fscal = _mm_and_pd(fscal,cutoff_mask);
1855 /* Update vectorial force */
1856 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1857 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1858 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1860 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1861 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1862 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1866 /**************************
1867 * CALCULATE INTERACTIONS *
1868 **************************/
1870 if (gmx_mm_any_lt(rsq12,rcutoff2))
1873 r12 = _mm_mul_pd(rsq12,rinv12);
1875 /* EWALD ELECTROSTATICS */
1877 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1878 ewrt = _mm_mul_pd(r12,ewtabscale);
1879 ewitab = _mm_cvttpd_epi32(ewrt);
1881 eweps = _mm_frcz_pd(ewrt);
1883 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1885 twoeweps = _mm_add_pd(eweps,eweps);
1886 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1888 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1889 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1891 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1895 fscal = _mm_and_pd(fscal,cutoff_mask);
1897 /* Update vectorial force */
1898 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1899 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1900 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1902 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1903 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1904 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1908 /**************************
1909 * CALCULATE INTERACTIONS *
1910 **************************/
1912 if (gmx_mm_any_lt(rsq20,rcutoff2))
1915 r20 = _mm_mul_pd(rsq20,rinv20);
1917 /* EWALD ELECTROSTATICS */
1919 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1920 ewrt = _mm_mul_pd(r20,ewtabscale);
1921 ewitab = _mm_cvttpd_epi32(ewrt);
1923 eweps = _mm_frcz_pd(ewrt);
1925 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1927 twoeweps = _mm_add_pd(eweps,eweps);
1928 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1930 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1931 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1933 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1937 fscal = _mm_and_pd(fscal,cutoff_mask);
1939 /* Update vectorial force */
1940 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1941 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1942 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1944 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1945 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1946 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1950 /**************************
1951 * CALCULATE INTERACTIONS *
1952 **************************/
1954 if (gmx_mm_any_lt(rsq21,rcutoff2))
1957 r21 = _mm_mul_pd(rsq21,rinv21);
1959 /* EWALD ELECTROSTATICS */
1961 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1962 ewrt = _mm_mul_pd(r21,ewtabscale);
1963 ewitab = _mm_cvttpd_epi32(ewrt);
1965 eweps = _mm_frcz_pd(ewrt);
1967 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1969 twoeweps = _mm_add_pd(eweps,eweps);
1970 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1972 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1973 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1975 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1979 fscal = _mm_and_pd(fscal,cutoff_mask);
1981 /* Update vectorial force */
1982 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1983 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1984 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1986 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1987 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1988 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1992 /**************************
1993 * CALCULATE INTERACTIONS *
1994 **************************/
1996 if (gmx_mm_any_lt(rsq22,rcutoff2))
1999 r22 = _mm_mul_pd(rsq22,rinv22);
2001 /* EWALD ELECTROSTATICS */
2003 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2004 ewrt = _mm_mul_pd(r22,ewtabscale);
2005 ewitab = _mm_cvttpd_epi32(ewrt);
2007 eweps = _mm_frcz_pd(ewrt);
2009 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2011 twoeweps = _mm_add_pd(eweps,eweps);
2012 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2014 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2015 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2017 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2021 fscal = _mm_and_pd(fscal,cutoff_mask);
2023 /* Update vectorial force */
2024 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2025 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2026 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2028 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2029 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2030 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2034 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2036 /* Inner loop uses 385 flops */
2039 if(jidx<j_index_end)
2043 j_coord_offsetA = DIM*jnrA;
2045 /* load j atom coordinates */
2046 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2047 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2049 /* Calculate displacement vector */
2050 dx00 = _mm_sub_pd(ix0,jx0);
2051 dy00 = _mm_sub_pd(iy0,jy0);
2052 dz00 = _mm_sub_pd(iz0,jz0);
2053 dx01 = _mm_sub_pd(ix0,jx1);
2054 dy01 = _mm_sub_pd(iy0,jy1);
2055 dz01 = _mm_sub_pd(iz0,jz1);
2056 dx02 = _mm_sub_pd(ix0,jx2);
2057 dy02 = _mm_sub_pd(iy0,jy2);
2058 dz02 = _mm_sub_pd(iz0,jz2);
2059 dx10 = _mm_sub_pd(ix1,jx0);
2060 dy10 = _mm_sub_pd(iy1,jy0);
2061 dz10 = _mm_sub_pd(iz1,jz0);
2062 dx11 = _mm_sub_pd(ix1,jx1);
2063 dy11 = _mm_sub_pd(iy1,jy1);
2064 dz11 = _mm_sub_pd(iz1,jz1);
2065 dx12 = _mm_sub_pd(ix1,jx2);
2066 dy12 = _mm_sub_pd(iy1,jy2);
2067 dz12 = _mm_sub_pd(iz1,jz2);
2068 dx20 = _mm_sub_pd(ix2,jx0);
2069 dy20 = _mm_sub_pd(iy2,jy0);
2070 dz20 = _mm_sub_pd(iz2,jz0);
2071 dx21 = _mm_sub_pd(ix2,jx1);
2072 dy21 = _mm_sub_pd(iy2,jy1);
2073 dz21 = _mm_sub_pd(iz2,jz1);
2074 dx22 = _mm_sub_pd(ix2,jx2);
2075 dy22 = _mm_sub_pd(iy2,jy2);
2076 dz22 = _mm_sub_pd(iz2,jz2);
2078 /* Calculate squared distance and things based on it */
2079 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2080 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2081 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2082 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2083 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2084 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2085 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2086 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2087 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2089 rinv00 = gmx_mm_invsqrt_pd(rsq00);
2090 rinv01 = gmx_mm_invsqrt_pd(rsq01);
2091 rinv02 = gmx_mm_invsqrt_pd(rsq02);
2092 rinv10 = gmx_mm_invsqrt_pd(rsq10);
2093 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2094 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2095 rinv20 = gmx_mm_invsqrt_pd(rsq20);
2096 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2097 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2099 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
2100 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
2101 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
2102 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
2103 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
2104 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
2105 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
2106 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
2107 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
2109 fjx0 = _mm_setzero_pd();
2110 fjy0 = _mm_setzero_pd();
2111 fjz0 = _mm_setzero_pd();
2112 fjx1 = _mm_setzero_pd();
2113 fjy1 = _mm_setzero_pd();
2114 fjz1 = _mm_setzero_pd();
2115 fjx2 = _mm_setzero_pd();
2116 fjy2 = _mm_setzero_pd();
2117 fjz2 = _mm_setzero_pd();
2119 /**************************
2120 * CALCULATE INTERACTIONS *
2121 **************************/
2123 if (gmx_mm_any_lt(rsq00,rcutoff2))
2126 r00 = _mm_mul_pd(rsq00,rinv00);
2128 /* EWALD ELECTROSTATICS */
2130 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2131 ewrt = _mm_mul_pd(r00,ewtabscale);
2132 ewitab = _mm_cvttpd_epi32(ewrt);
2134 eweps = _mm_frcz_pd(ewrt);
2136 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2138 twoeweps = _mm_add_pd(eweps,eweps);
2139 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2140 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2141 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2143 /* LENNARD-JONES DISPERSION/REPULSION */
2145 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2146 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
2148 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
2150 fscal = _mm_add_pd(felec,fvdw);
2152 fscal = _mm_and_pd(fscal,cutoff_mask);
2154 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2156 /* Update vectorial force */
2157 fix0 = _mm_macc_pd(dx00,fscal,fix0);
2158 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
2159 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
2161 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
2162 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
2163 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
2167 /**************************
2168 * CALCULATE INTERACTIONS *
2169 **************************/
2171 if (gmx_mm_any_lt(rsq01,rcutoff2))
2174 r01 = _mm_mul_pd(rsq01,rinv01);
2176 /* EWALD ELECTROSTATICS */
2178 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2179 ewrt = _mm_mul_pd(r01,ewtabscale);
2180 ewitab = _mm_cvttpd_epi32(ewrt);
2182 eweps = _mm_frcz_pd(ewrt);
2184 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2186 twoeweps = _mm_add_pd(eweps,eweps);
2187 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2188 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2189 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2191 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
2195 fscal = _mm_and_pd(fscal,cutoff_mask);
2197 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2199 /* Update vectorial force */
2200 fix0 = _mm_macc_pd(dx01,fscal,fix0);
2201 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
2202 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
2204 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
2205 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
2206 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
2210 /**************************
2211 * CALCULATE INTERACTIONS *
2212 **************************/
2214 if (gmx_mm_any_lt(rsq02,rcutoff2))
2217 r02 = _mm_mul_pd(rsq02,rinv02);
2219 /* EWALD ELECTROSTATICS */
2221 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2222 ewrt = _mm_mul_pd(r02,ewtabscale);
2223 ewitab = _mm_cvttpd_epi32(ewrt);
2225 eweps = _mm_frcz_pd(ewrt);
2227 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2229 twoeweps = _mm_add_pd(eweps,eweps);
2230 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2231 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2232 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2234 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
2238 fscal = _mm_and_pd(fscal,cutoff_mask);
2240 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2242 /* Update vectorial force */
2243 fix0 = _mm_macc_pd(dx02,fscal,fix0);
2244 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
2245 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
2247 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
2248 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
2249 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
2253 /**************************
2254 * CALCULATE INTERACTIONS *
2255 **************************/
2257 if (gmx_mm_any_lt(rsq10,rcutoff2))
2260 r10 = _mm_mul_pd(rsq10,rinv10);
2262 /* EWALD ELECTROSTATICS */
2264 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2265 ewrt = _mm_mul_pd(r10,ewtabscale);
2266 ewitab = _mm_cvttpd_epi32(ewrt);
2268 eweps = _mm_frcz_pd(ewrt);
2270 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2272 twoeweps = _mm_add_pd(eweps,eweps);
2273 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2274 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2275 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2277 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
2281 fscal = _mm_and_pd(fscal,cutoff_mask);
2283 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2285 /* Update vectorial force */
2286 fix1 = _mm_macc_pd(dx10,fscal,fix1);
2287 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
2288 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
2290 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
2291 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
2292 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
2296 /**************************
2297 * CALCULATE INTERACTIONS *
2298 **************************/
2300 if (gmx_mm_any_lt(rsq11,rcutoff2))
2303 r11 = _mm_mul_pd(rsq11,rinv11);
2305 /* EWALD ELECTROSTATICS */
2307 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2308 ewrt = _mm_mul_pd(r11,ewtabscale);
2309 ewitab = _mm_cvttpd_epi32(ewrt);
2311 eweps = _mm_frcz_pd(ewrt);
2313 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2315 twoeweps = _mm_add_pd(eweps,eweps);
2316 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2317 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2318 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2320 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2324 fscal = _mm_and_pd(fscal,cutoff_mask);
2326 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2328 /* Update vectorial force */
2329 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2330 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2331 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2333 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2334 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2335 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2339 /**************************
2340 * CALCULATE INTERACTIONS *
2341 **************************/
2343 if (gmx_mm_any_lt(rsq12,rcutoff2))
2346 r12 = _mm_mul_pd(rsq12,rinv12);
2348 /* EWALD ELECTROSTATICS */
2350 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2351 ewrt = _mm_mul_pd(r12,ewtabscale);
2352 ewitab = _mm_cvttpd_epi32(ewrt);
2354 eweps = _mm_frcz_pd(ewrt);
2356 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2358 twoeweps = _mm_add_pd(eweps,eweps);
2359 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2360 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2361 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2363 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2367 fscal = _mm_and_pd(fscal,cutoff_mask);
2369 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2371 /* Update vectorial force */
2372 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2373 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2374 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2376 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2377 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2378 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2382 /**************************
2383 * CALCULATE INTERACTIONS *
2384 **************************/
2386 if (gmx_mm_any_lt(rsq20,rcutoff2))
2389 r20 = _mm_mul_pd(rsq20,rinv20);
2391 /* EWALD ELECTROSTATICS */
2393 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2394 ewrt = _mm_mul_pd(r20,ewtabscale);
2395 ewitab = _mm_cvttpd_epi32(ewrt);
2397 eweps = _mm_frcz_pd(ewrt);
2399 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2401 twoeweps = _mm_add_pd(eweps,eweps);
2402 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2403 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2404 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2406 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
2410 fscal = _mm_and_pd(fscal,cutoff_mask);
2412 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2414 /* Update vectorial force */
2415 fix2 = _mm_macc_pd(dx20,fscal,fix2);
2416 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
2417 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
2419 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
2420 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
2421 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
2425 /**************************
2426 * CALCULATE INTERACTIONS *
2427 **************************/
2429 if (gmx_mm_any_lt(rsq21,rcutoff2))
2432 r21 = _mm_mul_pd(rsq21,rinv21);
2434 /* EWALD ELECTROSTATICS */
2436 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2437 ewrt = _mm_mul_pd(r21,ewtabscale);
2438 ewitab = _mm_cvttpd_epi32(ewrt);
2440 eweps = _mm_frcz_pd(ewrt);
2442 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2444 twoeweps = _mm_add_pd(eweps,eweps);
2445 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2446 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2447 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2449 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2453 fscal = _mm_and_pd(fscal,cutoff_mask);
2455 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2457 /* Update vectorial force */
2458 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2459 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2460 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2462 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2463 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2464 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2468 /**************************
2469 * CALCULATE INTERACTIONS *
2470 **************************/
2472 if (gmx_mm_any_lt(rsq22,rcutoff2))
2475 r22 = _mm_mul_pd(rsq22,rinv22);
2477 /* EWALD ELECTROSTATICS */
2479 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2480 ewrt = _mm_mul_pd(r22,ewtabscale);
2481 ewitab = _mm_cvttpd_epi32(ewrt);
2483 eweps = _mm_frcz_pd(ewrt);
2485 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2487 twoeweps = _mm_add_pd(eweps,eweps);
2488 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2489 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2490 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2492 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2496 fscal = _mm_and_pd(fscal,cutoff_mask);
2498 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2500 /* Update vectorial force */
2501 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2502 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2503 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2505 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2506 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2507 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2511 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2513 /* Inner loop uses 385 flops */
2516 /* End of innermost loop */
2518 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2519 f+i_coord_offset,fshift+i_shift_offset);
2521 /* Increment number of inner iterations */
2522 inneriter += j_index_end - j_index_start;
2524 /* Outer loop uses 18 flops */
2527 /* Increment number of outer iterations */
2530 /* Update outer/inner flops */
2532 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*385);