2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.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_ElecEw_VdwNone_GeomW3W3_VF_avx_128_fma_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: None
56 * Geometry: Water3-Water3
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEw_VdwNone_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 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
108 __m128d dummy_mask,cutoff_mask;
109 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
110 __m128d one = _mm_set1_pd(1.0);
111 __m128d two = _mm_set1_pd(2.0);
117 jindex = nlist->jindex;
119 shiftidx = nlist->shift;
121 shiftvec = fr->shift_vec[0];
122 fshift = fr->fshift[0];
123 facel = _mm_set1_pd(fr->epsfac);
124 charge = mdatoms->chargeA;
126 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
127 ewtab = fr->ic->tabq_coul_FDV0;
128 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
129 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
134 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
135 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
137 jq0 = _mm_set1_pd(charge[inr+0]);
138 jq1 = _mm_set1_pd(charge[inr+1]);
139 jq2 = _mm_set1_pd(charge[inr+2]);
140 qq00 = _mm_mul_pd(iq0,jq0);
141 qq01 = _mm_mul_pd(iq0,jq1);
142 qq02 = _mm_mul_pd(iq0,jq2);
143 qq10 = _mm_mul_pd(iq1,jq0);
144 qq11 = _mm_mul_pd(iq1,jq1);
145 qq12 = _mm_mul_pd(iq1,jq2);
146 qq20 = _mm_mul_pd(iq2,jq0);
147 qq21 = _mm_mul_pd(iq2,jq1);
148 qq22 = _mm_mul_pd(iq2,jq2);
150 /* Avoid stupid compiler warnings */
158 /* Start outer loop over neighborlists */
159 for(iidx=0; iidx<nri; iidx++)
161 /* Load shift vector for this list */
162 i_shift_offset = DIM*shiftidx[iidx];
164 /* Load limits for loop over neighbors */
165 j_index_start = jindex[iidx];
166 j_index_end = jindex[iidx+1];
168 /* Get outer coordinate index */
170 i_coord_offset = DIM*inr;
172 /* Load i particle coords and add shift vector */
173 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
174 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
176 fix0 = _mm_setzero_pd();
177 fiy0 = _mm_setzero_pd();
178 fiz0 = _mm_setzero_pd();
179 fix1 = _mm_setzero_pd();
180 fiy1 = _mm_setzero_pd();
181 fiz1 = _mm_setzero_pd();
182 fix2 = _mm_setzero_pd();
183 fiy2 = _mm_setzero_pd();
184 fiz2 = _mm_setzero_pd();
186 /* Reset potential sums */
187 velecsum = _mm_setzero_pd();
189 /* Start inner kernel loop */
190 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
193 /* Get j neighbor index, and coordinate index */
196 j_coord_offsetA = DIM*jnrA;
197 j_coord_offsetB = DIM*jnrB;
199 /* load j atom coordinates */
200 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
201 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
203 /* Calculate displacement vector */
204 dx00 = _mm_sub_pd(ix0,jx0);
205 dy00 = _mm_sub_pd(iy0,jy0);
206 dz00 = _mm_sub_pd(iz0,jz0);
207 dx01 = _mm_sub_pd(ix0,jx1);
208 dy01 = _mm_sub_pd(iy0,jy1);
209 dz01 = _mm_sub_pd(iz0,jz1);
210 dx02 = _mm_sub_pd(ix0,jx2);
211 dy02 = _mm_sub_pd(iy0,jy2);
212 dz02 = _mm_sub_pd(iz0,jz2);
213 dx10 = _mm_sub_pd(ix1,jx0);
214 dy10 = _mm_sub_pd(iy1,jy0);
215 dz10 = _mm_sub_pd(iz1,jz0);
216 dx11 = _mm_sub_pd(ix1,jx1);
217 dy11 = _mm_sub_pd(iy1,jy1);
218 dz11 = _mm_sub_pd(iz1,jz1);
219 dx12 = _mm_sub_pd(ix1,jx2);
220 dy12 = _mm_sub_pd(iy1,jy2);
221 dz12 = _mm_sub_pd(iz1,jz2);
222 dx20 = _mm_sub_pd(ix2,jx0);
223 dy20 = _mm_sub_pd(iy2,jy0);
224 dz20 = _mm_sub_pd(iz2,jz0);
225 dx21 = _mm_sub_pd(ix2,jx1);
226 dy21 = _mm_sub_pd(iy2,jy1);
227 dz21 = _mm_sub_pd(iz2,jz1);
228 dx22 = _mm_sub_pd(ix2,jx2);
229 dy22 = _mm_sub_pd(iy2,jy2);
230 dz22 = _mm_sub_pd(iz2,jz2);
232 /* Calculate squared distance and things based on it */
233 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
234 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
235 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
236 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
237 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
238 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
239 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
240 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
241 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
243 rinv00 = gmx_mm_invsqrt_pd(rsq00);
244 rinv01 = gmx_mm_invsqrt_pd(rsq01);
245 rinv02 = gmx_mm_invsqrt_pd(rsq02);
246 rinv10 = gmx_mm_invsqrt_pd(rsq10);
247 rinv11 = gmx_mm_invsqrt_pd(rsq11);
248 rinv12 = gmx_mm_invsqrt_pd(rsq12);
249 rinv20 = gmx_mm_invsqrt_pd(rsq20);
250 rinv21 = gmx_mm_invsqrt_pd(rsq21);
251 rinv22 = gmx_mm_invsqrt_pd(rsq22);
253 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
254 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
255 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
256 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
257 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
258 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
259 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
260 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
261 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
263 fjx0 = _mm_setzero_pd();
264 fjy0 = _mm_setzero_pd();
265 fjz0 = _mm_setzero_pd();
266 fjx1 = _mm_setzero_pd();
267 fjy1 = _mm_setzero_pd();
268 fjz1 = _mm_setzero_pd();
269 fjx2 = _mm_setzero_pd();
270 fjy2 = _mm_setzero_pd();
271 fjz2 = _mm_setzero_pd();
273 /**************************
274 * CALCULATE INTERACTIONS *
275 **************************/
277 r00 = _mm_mul_pd(rsq00,rinv00);
279 /* EWALD ELECTROSTATICS */
281 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
282 ewrt = _mm_mul_pd(r00,ewtabscale);
283 ewitab = _mm_cvttpd_epi32(ewrt);
285 eweps = _mm_frcz_pd(ewrt);
287 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
289 twoeweps = _mm_add_pd(eweps,eweps);
290 ewitab = _mm_slli_epi32(ewitab,2);
291 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
292 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
293 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
294 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
295 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
296 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
297 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
298 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
299 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
300 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
302 /* Update potential sum for this i atom from the interaction with this j atom. */
303 velecsum = _mm_add_pd(velecsum,velec);
307 /* Update vectorial force */
308 fix0 = _mm_macc_pd(dx00,fscal,fix0);
309 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
310 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
312 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
313 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
314 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
316 /**************************
317 * CALCULATE INTERACTIONS *
318 **************************/
320 r01 = _mm_mul_pd(rsq01,rinv01);
322 /* EWALD ELECTROSTATICS */
324 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
325 ewrt = _mm_mul_pd(r01,ewtabscale);
326 ewitab = _mm_cvttpd_epi32(ewrt);
328 eweps = _mm_frcz_pd(ewrt);
330 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
332 twoeweps = _mm_add_pd(eweps,eweps);
333 ewitab = _mm_slli_epi32(ewitab,2);
334 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
335 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
336 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
337 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
338 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
339 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
340 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
341 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
342 velec = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
343 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
345 /* Update potential sum for this i atom from the interaction with this j atom. */
346 velecsum = _mm_add_pd(velecsum,velec);
350 /* Update vectorial force */
351 fix0 = _mm_macc_pd(dx01,fscal,fix0);
352 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
353 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
355 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
356 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
357 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
359 /**************************
360 * CALCULATE INTERACTIONS *
361 **************************/
363 r02 = _mm_mul_pd(rsq02,rinv02);
365 /* EWALD ELECTROSTATICS */
367 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
368 ewrt = _mm_mul_pd(r02,ewtabscale);
369 ewitab = _mm_cvttpd_epi32(ewrt);
371 eweps = _mm_frcz_pd(ewrt);
373 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
375 twoeweps = _mm_add_pd(eweps,eweps);
376 ewitab = _mm_slli_epi32(ewitab,2);
377 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
378 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
379 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
380 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
381 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
382 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
383 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
384 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
385 velec = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
386 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
388 /* Update potential sum for this i atom from the interaction with this j atom. */
389 velecsum = _mm_add_pd(velecsum,velec);
393 /* Update vectorial force */
394 fix0 = _mm_macc_pd(dx02,fscal,fix0);
395 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
396 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
398 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
399 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
400 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
402 /**************************
403 * CALCULATE INTERACTIONS *
404 **************************/
406 r10 = _mm_mul_pd(rsq10,rinv10);
408 /* EWALD ELECTROSTATICS */
410 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
411 ewrt = _mm_mul_pd(r10,ewtabscale);
412 ewitab = _mm_cvttpd_epi32(ewrt);
414 eweps = _mm_frcz_pd(ewrt);
416 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
418 twoeweps = _mm_add_pd(eweps,eweps);
419 ewitab = _mm_slli_epi32(ewitab,2);
420 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
421 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
422 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
423 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
424 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
425 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
426 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
427 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
428 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
429 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
431 /* Update potential sum for this i atom from the interaction with this j atom. */
432 velecsum = _mm_add_pd(velecsum,velec);
436 /* Update vectorial force */
437 fix1 = _mm_macc_pd(dx10,fscal,fix1);
438 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
439 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
441 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
442 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
443 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
445 /**************************
446 * CALCULATE INTERACTIONS *
447 **************************/
449 r11 = _mm_mul_pd(rsq11,rinv11);
451 /* EWALD ELECTROSTATICS */
453 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
454 ewrt = _mm_mul_pd(r11,ewtabscale);
455 ewitab = _mm_cvttpd_epi32(ewrt);
457 eweps = _mm_frcz_pd(ewrt);
459 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
461 twoeweps = _mm_add_pd(eweps,eweps);
462 ewitab = _mm_slli_epi32(ewitab,2);
463 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
464 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
465 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
466 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
467 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
468 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
469 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
470 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
471 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
472 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
474 /* Update potential sum for this i atom from the interaction with this j atom. */
475 velecsum = _mm_add_pd(velecsum,velec);
479 /* Update vectorial force */
480 fix1 = _mm_macc_pd(dx11,fscal,fix1);
481 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
482 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
484 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
485 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
486 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
488 /**************************
489 * CALCULATE INTERACTIONS *
490 **************************/
492 r12 = _mm_mul_pd(rsq12,rinv12);
494 /* EWALD ELECTROSTATICS */
496 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
497 ewrt = _mm_mul_pd(r12,ewtabscale);
498 ewitab = _mm_cvttpd_epi32(ewrt);
500 eweps = _mm_frcz_pd(ewrt);
502 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
504 twoeweps = _mm_add_pd(eweps,eweps);
505 ewitab = _mm_slli_epi32(ewitab,2);
506 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
507 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
508 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
509 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
510 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
511 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
512 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
513 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
514 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
515 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
517 /* Update potential sum for this i atom from the interaction with this j atom. */
518 velecsum = _mm_add_pd(velecsum,velec);
522 /* Update vectorial force */
523 fix1 = _mm_macc_pd(dx12,fscal,fix1);
524 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
525 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
527 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
528 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
529 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
531 /**************************
532 * CALCULATE INTERACTIONS *
533 **************************/
535 r20 = _mm_mul_pd(rsq20,rinv20);
537 /* EWALD ELECTROSTATICS */
539 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
540 ewrt = _mm_mul_pd(r20,ewtabscale);
541 ewitab = _mm_cvttpd_epi32(ewrt);
543 eweps = _mm_frcz_pd(ewrt);
545 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
547 twoeweps = _mm_add_pd(eweps,eweps);
548 ewitab = _mm_slli_epi32(ewitab,2);
549 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
550 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
551 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
552 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
553 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
554 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
555 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
556 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
557 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
558 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
560 /* Update potential sum for this i atom from the interaction with this j atom. */
561 velecsum = _mm_add_pd(velecsum,velec);
565 /* Update vectorial force */
566 fix2 = _mm_macc_pd(dx20,fscal,fix2);
567 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
568 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
570 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
571 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
572 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
574 /**************************
575 * CALCULATE INTERACTIONS *
576 **************************/
578 r21 = _mm_mul_pd(rsq21,rinv21);
580 /* EWALD ELECTROSTATICS */
582 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
583 ewrt = _mm_mul_pd(r21,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(qq21,_mm_sub_pd(rinv21,velec));
601 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
603 /* Update potential sum for this i atom from the interaction with this j atom. */
604 velecsum = _mm_add_pd(velecsum,velec);
608 /* Update vectorial force */
609 fix2 = _mm_macc_pd(dx21,fscal,fix2);
610 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
611 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
613 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
614 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
615 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
617 /**************************
618 * CALCULATE INTERACTIONS *
619 **************************/
621 r22 = _mm_mul_pd(rsq22,rinv22);
623 /* EWALD ELECTROSTATICS */
625 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
626 ewrt = _mm_mul_pd(r22,ewtabscale);
627 ewitab = _mm_cvttpd_epi32(ewrt);
629 eweps = _mm_frcz_pd(ewrt);
631 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
633 twoeweps = _mm_add_pd(eweps,eweps);
634 ewitab = _mm_slli_epi32(ewitab,2);
635 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
636 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
637 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
638 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
639 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
640 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
641 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
642 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
643 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
644 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
646 /* Update potential sum for this i atom from the interaction with this j atom. */
647 velecsum = _mm_add_pd(velecsum,velec);
651 /* Update vectorial force */
652 fix2 = _mm_macc_pd(dx22,fscal,fix2);
653 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
654 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
656 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
657 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
658 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
660 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
662 /* Inner loop uses 396 flops */
669 j_coord_offsetA = DIM*jnrA;
671 /* load j atom coordinates */
672 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
673 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
675 /* Calculate displacement vector */
676 dx00 = _mm_sub_pd(ix0,jx0);
677 dy00 = _mm_sub_pd(iy0,jy0);
678 dz00 = _mm_sub_pd(iz0,jz0);
679 dx01 = _mm_sub_pd(ix0,jx1);
680 dy01 = _mm_sub_pd(iy0,jy1);
681 dz01 = _mm_sub_pd(iz0,jz1);
682 dx02 = _mm_sub_pd(ix0,jx2);
683 dy02 = _mm_sub_pd(iy0,jy2);
684 dz02 = _mm_sub_pd(iz0,jz2);
685 dx10 = _mm_sub_pd(ix1,jx0);
686 dy10 = _mm_sub_pd(iy1,jy0);
687 dz10 = _mm_sub_pd(iz1,jz0);
688 dx11 = _mm_sub_pd(ix1,jx1);
689 dy11 = _mm_sub_pd(iy1,jy1);
690 dz11 = _mm_sub_pd(iz1,jz1);
691 dx12 = _mm_sub_pd(ix1,jx2);
692 dy12 = _mm_sub_pd(iy1,jy2);
693 dz12 = _mm_sub_pd(iz1,jz2);
694 dx20 = _mm_sub_pd(ix2,jx0);
695 dy20 = _mm_sub_pd(iy2,jy0);
696 dz20 = _mm_sub_pd(iz2,jz0);
697 dx21 = _mm_sub_pd(ix2,jx1);
698 dy21 = _mm_sub_pd(iy2,jy1);
699 dz21 = _mm_sub_pd(iz2,jz1);
700 dx22 = _mm_sub_pd(ix2,jx2);
701 dy22 = _mm_sub_pd(iy2,jy2);
702 dz22 = _mm_sub_pd(iz2,jz2);
704 /* Calculate squared distance and things based on it */
705 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
706 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
707 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
708 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
709 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
710 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
711 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
712 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
713 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
715 rinv00 = gmx_mm_invsqrt_pd(rsq00);
716 rinv01 = gmx_mm_invsqrt_pd(rsq01);
717 rinv02 = gmx_mm_invsqrt_pd(rsq02);
718 rinv10 = gmx_mm_invsqrt_pd(rsq10);
719 rinv11 = gmx_mm_invsqrt_pd(rsq11);
720 rinv12 = gmx_mm_invsqrt_pd(rsq12);
721 rinv20 = gmx_mm_invsqrt_pd(rsq20);
722 rinv21 = gmx_mm_invsqrt_pd(rsq21);
723 rinv22 = gmx_mm_invsqrt_pd(rsq22);
725 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
726 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
727 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
728 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
729 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
730 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
731 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
732 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
733 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
735 fjx0 = _mm_setzero_pd();
736 fjy0 = _mm_setzero_pd();
737 fjz0 = _mm_setzero_pd();
738 fjx1 = _mm_setzero_pd();
739 fjy1 = _mm_setzero_pd();
740 fjz1 = _mm_setzero_pd();
741 fjx2 = _mm_setzero_pd();
742 fjy2 = _mm_setzero_pd();
743 fjz2 = _mm_setzero_pd();
745 /**************************
746 * CALCULATE INTERACTIONS *
747 **************************/
749 r00 = _mm_mul_pd(rsq00,rinv00);
751 /* EWALD ELECTROSTATICS */
753 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
754 ewrt = _mm_mul_pd(r00,ewtabscale);
755 ewitab = _mm_cvttpd_epi32(ewrt);
757 eweps = _mm_frcz_pd(ewrt);
759 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
761 twoeweps = _mm_add_pd(eweps,eweps);
762 ewitab = _mm_slli_epi32(ewitab,2);
763 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
764 ewtabD = _mm_setzero_pd();
765 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
766 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
767 ewtabFn = _mm_setzero_pd();
768 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
769 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
770 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
771 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
772 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
774 /* Update potential sum for this i atom from the interaction with this j atom. */
775 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
776 velecsum = _mm_add_pd(velecsum,velec);
780 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
782 /* Update vectorial force */
783 fix0 = _mm_macc_pd(dx00,fscal,fix0);
784 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
785 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
787 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
788 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
789 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
791 /**************************
792 * CALCULATE INTERACTIONS *
793 **************************/
795 r01 = _mm_mul_pd(rsq01,rinv01);
797 /* EWALD ELECTROSTATICS */
799 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
800 ewrt = _mm_mul_pd(r01,ewtabscale);
801 ewitab = _mm_cvttpd_epi32(ewrt);
803 eweps = _mm_frcz_pd(ewrt);
805 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
807 twoeweps = _mm_add_pd(eweps,eweps);
808 ewitab = _mm_slli_epi32(ewitab,2);
809 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
810 ewtabD = _mm_setzero_pd();
811 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
812 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
813 ewtabFn = _mm_setzero_pd();
814 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
815 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
816 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
817 velec = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
818 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
820 /* Update potential sum for this i atom from the interaction with this j atom. */
821 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
822 velecsum = _mm_add_pd(velecsum,velec);
826 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
828 /* Update vectorial force */
829 fix0 = _mm_macc_pd(dx01,fscal,fix0);
830 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
831 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
833 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
834 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
835 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
837 /**************************
838 * CALCULATE INTERACTIONS *
839 **************************/
841 r02 = _mm_mul_pd(rsq02,rinv02);
843 /* EWALD ELECTROSTATICS */
845 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
846 ewrt = _mm_mul_pd(r02,ewtabscale);
847 ewitab = _mm_cvttpd_epi32(ewrt);
849 eweps = _mm_frcz_pd(ewrt);
851 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
853 twoeweps = _mm_add_pd(eweps,eweps);
854 ewitab = _mm_slli_epi32(ewitab,2);
855 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
856 ewtabD = _mm_setzero_pd();
857 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
858 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
859 ewtabFn = _mm_setzero_pd();
860 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
861 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
862 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
863 velec = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
864 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
866 /* Update potential sum for this i atom from the interaction with this j atom. */
867 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
868 velecsum = _mm_add_pd(velecsum,velec);
872 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
874 /* Update vectorial force */
875 fix0 = _mm_macc_pd(dx02,fscal,fix0);
876 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
877 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
879 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
880 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
881 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
883 /**************************
884 * CALCULATE INTERACTIONS *
885 **************************/
887 r10 = _mm_mul_pd(rsq10,rinv10);
889 /* EWALD ELECTROSTATICS */
891 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
892 ewrt = _mm_mul_pd(r10,ewtabscale);
893 ewitab = _mm_cvttpd_epi32(ewrt);
895 eweps = _mm_frcz_pd(ewrt);
897 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
899 twoeweps = _mm_add_pd(eweps,eweps);
900 ewitab = _mm_slli_epi32(ewitab,2);
901 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
902 ewtabD = _mm_setzero_pd();
903 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
904 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
905 ewtabFn = _mm_setzero_pd();
906 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
907 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
908 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
909 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
910 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
912 /* Update potential sum for this i atom from the interaction with this j atom. */
913 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
914 velecsum = _mm_add_pd(velecsum,velec);
918 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
920 /* Update vectorial force */
921 fix1 = _mm_macc_pd(dx10,fscal,fix1);
922 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
923 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
925 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
926 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
927 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
929 /**************************
930 * CALCULATE INTERACTIONS *
931 **************************/
933 r11 = _mm_mul_pd(rsq11,rinv11);
935 /* EWALD ELECTROSTATICS */
937 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
938 ewrt = _mm_mul_pd(r11,ewtabscale);
939 ewitab = _mm_cvttpd_epi32(ewrt);
941 eweps = _mm_frcz_pd(ewrt);
943 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
945 twoeweps = _mm_add_pd(eweps,eweps);
946 ewitab = _mm_slli_epi32(ewitab,2);
947 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
948 ewtabD = _mm_setzero_pd();
949 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
950 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
951 ewtabFn = _mm_setzero_pd();
952 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
953 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
954 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
955 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
956 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
958 /* Update potential sum for this i atom from the interaction with this j atom. */
959 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
960 velecsum = _mm_add_pd(velecsum,velec);
964 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
966 /* Update vectorial force */
967 fix1 = _mm_macc_pd(dx11,fscal,fix1);
968 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
969 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
971 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
972 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
973 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
975 /**************************
976 * CALCULATE INTERACTIONS *
977 **************************/
979 r12 = _mm_mul_pd(rsq12,rinv12);
981 /* EWALD ELECTROSTATICS */
983 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
984 ewrt = _mm_mul_pd(r12,ewtabscale);
985 ewitab = _mm_cvttpd_epi32(ewrt);
987 eweps = _mm_frcz_pd(ewrt);
989 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
991 twoeweps = _mm_add_pd(eweps,eweps);
992 ewitab = _mm_slli_epi32(ewitab,2);
993 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
994 ewtabD = _mm_setzero_pd();
995 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
996 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
997 ewtabFn = _mm_setzero_pd();
998 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
999 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1000 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1001 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1002 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1004 /* Update potential sum for this i atom from the interaction with this j atom. */
1005 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1006 velecsum = _mm_add_pd(velecsum,velec);
1010 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1012 /* Update vectorial force */
1013 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1014 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1015 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1017 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1018 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1019 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1021 /**************************
1022 * CALCULATE INTERACTIONS *
1023 **************************/
1025 r20 = _mm_mul_pd(rsq20,rinv20);
1027 /* EWALD ELECTROSTATICS */
1029 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1030 ewrt = _mm_mul_pd(r20,ewtabscale);
1031 ewitab = _mm_cvttpd_epi32(ewrt);
1033 eweps = _mm_frcz_pd(ewrt);
1035 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1037 twoeweps = _mm_add_pd(eweps,eweps);
1038 ewitab = _mm_slli_epi32(ewitab,2);
1039 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1040 ewtabD = _mm_setzero_pd();
1041 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1042 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1043 ewtabFn = _mm_setzero_pd();
1044 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1045 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1046 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1047 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1048 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1050 /* Update potential sum for this i atom from the interaction with this j atom. */
1051 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1052 velecsum = _mm_add_pd(velecsum,velec);
1056 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1058 /* Update vectorial force */
1059 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1060 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1061 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1063 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1064 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1065 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1067 /**************************
1068 * CALCULATE INTERACTIONS *
1069 **************************/
1071 r21 = _mm_mul_pd(rsq21,rinv21);
1073 /* EWALD ELECTROSTATICS */
1075 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1076 ewrt = _mm_mul_pd(r21,ewtabscale);
1077 ewitab = _mm_cvttpd_epi32(ewrt);
1079 eweps = _mm_frcz_pd(ewrt);
1081 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1083 twoeweps = _mm_add_pd(eweps,eweps);
1084 ewitab = _mm_slli_epi32(ewitab,2);
1085 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1086 ewtabD = _mm_setzero_pd();
1087 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1088 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1089 ewtabFn = _mm_setzero_pd();
1090 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1091 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1092 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1093 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1094 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1096 /* Update potential sum for this i atom from the interaction with this j atom. */
1097 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1098 velecsum = _mm_add_pd(velecsum,velec);
1102 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1104 /* Update vectorial force */
1105 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1106 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1107 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1109 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1110 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1111 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1113 /**************************
1114 * CALCULATE INTERACTIONS *
1115 **************************/
1117 r22 = _mm_mul_pd(rsq22,rinv22);
1119 /* EWALD ELECTROSTATICS */
1121 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1122 ewrt = _mm_mul_pd(r22,ewtabscale);
1123 ewitab = _mm_cvttpd_epi32(ewrt);
1125 eweps = _mm_frcz_pd(ewrt);
1127 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1129 twoeweps = _mm_add_pd(eweps,eweps);
1130 ewitab = _mm_slli_epi32(ewitab,2);
1131 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1132 ewtabD = _mm_setzero_pd();
1133 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1134 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1135 ewtabFn = _mm_setzero_pd();
1136 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1137 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1138 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1139 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1140 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1142 /* Update potential sum for this i atom from the interaction with this j atom. */
1143 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1144 velecsum = _mm_add_pd(velecsum,velec);
1148 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1150 /* Update vectorial force */
1151 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1152 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1153 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1155 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1156 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1157 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1159 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1161 /* Inner loop uses 396 flops */
1164 /* End of innermost loop */
1166 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1167 f+i_coord_offset,fshift+i_shift_offset);
1170 /* Update potential energies */
1171 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1173 /* Increment number of inner iterations */
1174 inneriter += j_index_end - j_index_start;
1176 /* Outer loop uses 19 flops */
1179 /* Increment number of outer iterations */
1182 /* Update outer/inner flops */
1184 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*396);
1187 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_128_fma_double
1188 * Electrostatics interaction: Ewald
1189 * VdW interaction: None
1190 * Geometry: Water3-Water3
1191 * Calculate force/pot: Force
1194 nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_128_fma_double
1195 (t_nblist * gmx_restrict nlist,
1196 rvec * gmx_restrict xx,
1197 rvec * gmx_restrict ff,
1198 t_forcerec * gmx_restrict fr,
1199 t_mdatoms * gmx_restrict mdatoms,
1200 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1201 t_nrnb * gmx_restrict nrnb)
1203 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1204 * just 0 for non-waters.
1205 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1206 * jnr indices corresponding to data put in the four positions in the SIMD register.
1208 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1209 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1211 int j_coord_offsetA,j_coord_offsetB;
1212 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1213 real rcutoff_scalar;
1214 real *shiftvec,*fshift,*x,*f;
1215 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1217 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1219 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1221 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1222 int vdwjidx0A,vdwjidx0B;
1223 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1224 int vdwjidx1A,vdwjidx1B;
1225 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1226 int vdwjidx2A,vdwjidx2B;
1227 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1228 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1229 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1230 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1231 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1232 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1233 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1234 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1235 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1236 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1237 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1240 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1242 __m128d dummy_mask,cutoff_mask;
1243 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1244 __m128d one = _mm_set1_pd(1.0);
1245 __m128d two = _mm_set1_pd(2.0);
1251 jindex = nlist->jindex;
1253 shiftidx = nlist->shift;
1255 shiftvec = fr->shift_vec[0];
1256 fshift = fr->fshift[0];
1257 facel = _mm_set1_pd(fr->epsfac);
1258 charge = mdatoms->chargeA;
1260 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1261 ewtab = fr->ic->tabq_coul_F;
1262 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1263 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1265 /* Setup water-specific parameters */
1266 inr = nlist->iinr[0];
1267 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1268 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1269 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1271 jq0 = _mm_set1_pd(charge[inr+0]);
1272 jq1 = _mm_set1_pd(charge[inr+1]);
1273 jq2 = _mm_set1_pd(charge[inr+2]);
1274 qq00 = _mm_mul_pd(iq0,jq0);
1275 qq01 = _mm_mul_pd(iq0,jq1);
1276 qq02 = _mm_mul_pd(iq0,jq2);
1277 qq10 = _mm_mul_pd(iq1,jq0);
1278 qq11 = _mm_mul_pd(iq1,jq1);
1279 qq12 = _mm_mul_pd(iq1,jq2);
1280 qq20 = _mm_mul_pd(iq2,jq0);
1281 qq21 = _mm_mul_pd(iq2,jq1);
1282 qq22 = _mm_mul_pd(iq2,jq2);
1284 /* Avoid stupid compiler warnings */
1286 j_coord_offsetA = 0;
1287 j_coord_offsetB = 0;
1292 /* Start outer loop over neighborlists */
1293 for(iidx=0; iidx<nri; iidx++)
1295 /* Load shift vector for this list */
1296 i_shift_offset = DIM*shiftidx[iidx];
1298 /* Load limits for loop over neighbors */
1299 j_index_start = jindex[iidx];
1300 j_index_end = jindex[iidx+1];
1302 /* Get outer coordinate index */
1304 i_coord_offset = DIM*inr;
1306 /* Load i particle coords and add shift vector */
1307 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1308 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1310 fix0 = _mm_setzero_pd();
1311 fiy0 = _mm_setzero_pd();
1312 fiz0 = _mm_setzero_pd();
1313 fix1 = _mm_setzero_pd();
1314 fiy1 = _mm_setzero_pd();
1315 fiz1 = _mm_setzero_pd();
1316 fix2 = _mm_setzero_pd();
1317 fiy2 = _mm_setzero_pd();
1318 fiz2 = _mm_setzero_pd();
1320 /* Start inner kernel loop */
1321 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1324 /* Get j neighbor index, and coordinate index */
1326 jnrB = jjnr[jidx+1];
1327 j_coord_offsetA = DIM*jnrA;
1328 j_coord_offsetB = DIM*jnrB;
1330 /* load j atom coordinates */
1331 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1332 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1334 /* Calculate displacement vector */
1335 dx00 = _mm_sub_pd(ix0,jx0);
1336 dy00 = _mm_sub_pd(iy0,jy0);
1337 dz00 = _mm_sub_pd(iz0,jz0);
1338 dx01 = _mm_sub_pd(ix0,jx1);
1339 dy01 = _mm_sub_pd(iy0,jy1);
1340 dz01 = _mm_sub_pd(iz0,jz1);
1341 dx02 = _mm_sub_pd(ix0,jx2);
1342 dy02 = _mm_sub_pd(iy0,jy2);
1343 dz02 = _mm_sub_pd(iz0,jz2);
1344 dx10 = _mm_sub_pd(ix1,jx0);
1345 dy10 = _mm_sub_pd(iy1,jy0);
1346 dz10 = _mm_sub_pd(iz1,jz0);
1347 dx11 = _mm_sub_pd(ix1,jx1);
1348 dy11 = _mm_sub_pd(iy1,jy1);
1349 dz11 = _mm_sub_pd(iz1,jz1);
1350 dx12 = _mm_sub_pd(ix1,jx2);
1351 dy12 = _mm_sub_pd(iy1,jy2);
1352 dz12 = _mm_sub_pd(iz1,jz2);
1353 dx20 = _mm_sub_pd(ix2,jx0);
1354 dy20 = _mm_sub_pd(iy2,jy0);
1355 dz20 = _mm_sub_pd(iz2,jz0);
1356 dx21 = _mm_sub_pd(ix2,jx1);
1357 dy21 = _mm_sub_pd(iy2,jy1);
1358 dz21 = _mm_sub_pd(iz2,jz1);
1359 dx22 = _mm_sub_pd(ix2,jx2);
1360 dy22 = _mm_sub_pd(iy2,jy2);
1361 dz22 = _mm_sub_pd(iz2,jz2);
1363 /* Calculate squared distance and things based on it */
1364 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1365 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1366 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1367 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1368 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1369 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1370 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1371 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1372 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1374 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1375 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1376 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1377 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1378 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1379 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1380 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1381 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1382 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1384 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1385 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
1386 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
1387 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1388 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1389 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1390 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1391 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1392 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1394 fjx0 = _mm_setzero_pd();
1395 fjy0 = _mm_setzero_pd();
1396 fjz0 = _mm_setzero_pd();
1397 fjx1 = _mm_setzero_pd();
1398 fjy1 = _mm_setzero_pd();
1399 fjz1 = _mm_setzero_pd();
1400 fjx2 = _mm_setzero_pd();
1401 fjy2 = _mm_setzero_pd();
1402 fjz2 = _mm_setzero_pd();
1404 /**************************
1405 * CALCULATE INTERACTIONS *
1406 **************************/
1408 r00 = _mm_mul_pd(rsq00,rinv00);
1410 /* EWALD ELECTROSTATICS */
1412 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1413 ewrt = _mm_mul_pd(r00,ewtabscale);
1414 ewitab = _mm_cvttpd_epi32(ewrt);
1416 eweps = _mm_frcz_pd(ewrt);
1418 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1420 twoeweps = _mm_add_pd(eweps,eweps);
1421 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1423 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1424 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1428 /* Update vectorial force */
1429 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1430 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1431 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1433 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1434 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1435 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1437 /**************************
1438 * CALCULATE INTERACTIONS *
1439 **************************/
1441 r01 = _mm_mul_pd(rsq01,rinv01);
1443 /* EWALD ELECTROSTATICS */
1445 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1446 ewrt = _mm_mul_pd(r01,ewtabscale);
1447 ewitab = _mm_cvttpd_epi32(ewrt);
1449 eweps = _mm_frcz_pd(ewrt);
1451 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1453 twoeweps = _mm_add_pd(eweps,eweps);
1454 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1456 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1457 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1461 /* Update vectorial force */
1462 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1463 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1464 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1466 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1467 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1468 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1470 /**************************
1471 * CALCULATE INTERACTIONS *
1472 **************************/
1474 r02 = _mm_mul_pd(rsq02,rinv02);
1476 /* EWALD ELECTROSTATICS */
1478 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1479 ewrt = _mm_mul_pd(r02,ewtabscale);
1480 ewitab = _mm_cvttpd_epi32(ewrt);
1482 eweps = _mm_frcz_pd(ewrt);
1484 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1486 twoeweps = _mm_add_pd(eweps,eweps);
1487 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1489 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1490 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1494 /* Update vectorial force */
1495 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1496 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1497 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1499 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1500 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1501 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1503 /**************************
1504 * CALCULATE INTERACTIONS *
1505 **************************/
1507 r10 = _mm_mul_pd(rsq10,rinv10);
1509 /* EWALD ELECTROSTATICS */
1511 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1512 ewrt = _mm_mul_pd(r10,ewtabscale);
1513 ewitab = _mm_cvttpd_epi32(ewrt);
1515 eweps = _mm_frcz_pd(ewrt);
1517 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1519 twoeweps = _mm_add_pd(eweps,eweps);
1520 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1522 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1523 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1527 /* Update vectorial force */
1528 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1529 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1530 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1532 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1533 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1534 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1536 /**************************
1537 * CALCULATE INTERACTIONS *
1538 **************************/
1540 r11 = _mm_mul_pd(rsq11,rinv11);
1542 /* EWALD ELECTROSTATICS */
1544 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1545 ewrt = _mm_mul_pd(r11,ewtabscale);
1546 ewitab = _mm_cvttpd_epi32(ewrt);
1548 eweps = _mm_frcz_pd(ewrt);
1550 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1552 twoeweps = _mm_add_pd(eweps,eweps);
1553 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1555 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1556 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1560 /* Update vectorial force */
1561 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1562 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1563 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1565 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1566 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1567 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1569 /**************************
1570 * CALCULATE INTERACTIONS *
1571 **************************/
1573 r12 = _mm_mul_pd(rsq12,rinv12);
1575 /* EWALD ELECTROSTATICS */
1577 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1578 ewrt = _mm_mul_pd(r12,ewtabscale);
1579 ewitab = _mm_cvttpd_epi32(ewrt);
1581 eweps = _mm_frcz_pd(ewrt);
1583 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1585 twoeweps = _mm_add_pd(eweps,eweps);
1586 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1588 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1589 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1593 /* Update vectorial force */
1594 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1595 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1596 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1598 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1599 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1600 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1602 /**************************
1603 * CALCULATE INTERACTIONS *
1604 **************************/
1606 r20 = _mm_mul_pd(rsq20,rinv20);
1608 /* EWALD ELECTROSTATICS */
1610 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1611 ewrt = _mm_mul_pd(r20,ewtabscale);
1612 ewitab = _mm_cvttpd_epi32(ewrt);
1614 eweps = _mm_frcz_pd(ewrt);
1616 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1618 twoeweps = _mm_add_pd(eweps,eweps);
1619 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1621 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1622 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1626 /* Update vectorial force */
1627 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1628 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1629 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1631 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1632 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1633 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1635 /**************************
1636 * CALCULATE INTERACTIONS *
1637 **************************/
1639 r21 = _mm_mul_pd(rsq21,rinv21);
1641 /* EWALD ELECTROSTATICS */
1643 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1644 ewrt = _mm_mul_pd(r21,ewtabscale);
1645 ewitab = _mm_cvttpd_epi32(ewrt);
1647 eweps = _mm_frcz_pd(ewrt);
1649 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1651 twoeweps = _mm_add_pd(eweps,eweps);
1652 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1654 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1655 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1659 /* Update vectorial force */
1660 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1661 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1662 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1664 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1665 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1666 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1668 /**************************
1669 * CALCULATE INTERACTIONS *
1670 **************************/
1672 r22 = _mm_mul_pd(rsq22,rinv22);
1674 /* EWALD ELECTROSTATICS */
1676 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1677 ewrt = _mm_mul_pd(r22,ewtabscale);
1678 ewitab = _mm_cvttpd_epi32(ewrt);
1680 eweps = _mm_frcz_pd(ewrt);
1682 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1684 twoeweps = _mm_add_pd(eweps,eweps);
1685 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1687 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1688 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1692 /* Update vectorial force */
1693 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1694 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1695 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1697 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1698 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1699 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1701 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1703 /* Inner loop uses 351 flops */
1706 if(jidx<j_index_end)
1710 j_coord_offsetA = DIM*jnrA;
1712 /* load j atom coordinates */
1713 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1714 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1716 /* Calculate displacement vector */
1717 dx00 = _mm_sub_pd(ix0,jx0);
1718 dy00 = _mm_sub_pd(iy0,jy0);
1719 dz00 = _mm_sub_pd(iz0,jz0);
1720 dx01 = _mm_sub_pd(ix0,jx1);
1721 dy01 = _mm_sub_pd(iy0,jy1);
1722 dz01 = _mm_sub_pd(iz0,jz1);
1723 dx02 = _mm_sub_pd(ix0,jx2);
1724 dy02 = _mm_sub_pd(iy0,jy2);
1725 dz02 = _mm_sub_pd(iz0,jz2);
1726 dx10 = _mm_sub_pd(ix1,jx0);
1727 dy10 = _mm_sub_pd(iy1,jy0);
1728 dz10 = _mm_sub_pd(iz1,jz0);
1729 dx11 = _mm_sub_pd(ix1,jx1);
1730 dy11 = _mm_sub_pd(iy1,jy1);
1731 dz11 = _mm_sub_pd(iz1,jz1);
1732 dx12 = _mm_sub_pd(ix1,jx2);
1733 dy12 = _mm_sub_pd(iy1,jy2);
1734 dz12 = _mm_sub_pd(iz1,jz2);
1735 dx20 = _mm_sub_pd(ix2,jx0);
1736 dy20 = _mm_sub_pd(iy2,jy0);
1737 dz20 = _mm_sub_pd(iz2,jz0);
1738 dx21 = _mm_sub_pd(ix2,jx1);
1739 dy21 = _mm_sub_pd(iy2,jy1);
1740 dz21 = _mm_sub_pd(iz2,jz1);
1741 dx22 = _mm_sub_pd(ix2,jx2);
1742 dy22 = _mm_sub_pd(iy2,jy2);
1743 dz22 = _mm_sub_pd(iz2,jz2);
1745 /* Calculate squared distance and things based on it */
1746 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1747 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1748 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1749 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1750 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1751 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1752 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1753 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1754 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1756 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1757 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1758 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1759 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1760 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1761 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1762 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1763 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1764 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1766 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1767 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
1768 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
1769 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1770 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1771 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1772 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1773 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1774 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1776 fjx0 = _mm_setzero_pd();
1777 fjy0 = _mm_setzero_pd();
1778 fjz0 = _mm_setzero_pd();
1779 fjx1 = _mm_setzero_pd();
1780 fjy1 = _mm_setzero_pd();
1781 fjz1 = _mm_setzero_pd();
1782 fjx2 = _mm_setzero_pd();
1783 fjy2 = _mm_setzero_pd();
1784 fjz2 = _mm_setzero_pd();
1786 /**************************
1787 * CALCULATE INTERACTIONS *
1788 **************************/
1790 r00 = _mm_mul_pd(rsq00,rinv00);
1792 /* EWALD ELECTROSTATICS */
1794 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1795 ewrt = _mm_mul_pd(r00,ewtabscale);
1796 ewitab = _mm_cvttpd_epi32(ewrt);
1798 eweps = _mm_frcz_pd(ewrt);
1800 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1802 twoeweps = _mm_add_pd(eweps,eweps);
1803 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1804 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1805 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1809 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1811 /* Update vectorial force */
1812 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1813 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1814 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1816 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1817 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1818 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1820 /**************************
1821 * CALCULATE INTERACTIONS *
1822 **************************/
1824 r01 = _mm_mul_pd(rsq01,rinv01);
1826 /* EWALD ELECTROSTATICS */
1828 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1829 ewrt = _mm_mul_pd(r01,ewtabscale);
1830 ewitab = _mm_cvttpd_epi32(ewrt);
1832 eweps = _mm_frcz_pd(ewrt);
1834 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1836 twoeweps = _mm_add_pd(eweps,eweps);
1837 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1838 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1839 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1843 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1845 /* Update vectorial force */
1846 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1847 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1848 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1850 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1851 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1852 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1854 /**************************
1855 * CALCULATE INTERACTIONS *
1856 **************************/
1858 r02 = _mm_mul_pd(rsq02,rinv02);
1860 /* EWALD ELECTROSTATICS */
1862 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1863 ewrt = _mm_mul_pd(r02,ewtabscale);
1864 ewitab = _mm_cvttpd_epi32(ewrt);
1866 eweps = _mm_frcz_pd(ewrt);
1868 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1870 twoeweps = _mm_add_pd(eweps,eweps);
1871 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1872 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1873 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1877 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1879 /* Update vectorial force */
1880 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1881 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1882 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1884 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1885 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1886 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1888 /**************************
1889 * CALCULATE INTERACTIONS *
1890 **************************/
1892 r10 = _mm_mul_pd(rsq10,rinv10);
1894 /* EWALD ELECTROSTATICS */
1896 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1897 ewrt = _mm_mul_pd(r10,ewtabscale);
1898 ewitab = _mm_cvttpd_epi32(ewrt);
1900 eweps = _mm_frcz_pd(ewrt);
1902 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1904 twoeweps = _mm_add_pd(eweps,eweps);
1905 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1906 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1907 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1911 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1913 /* Update vectorial force */
1914 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1915 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1916 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1918 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1919 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1920 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1922 /**************************
1923 * CALCULATE INTERACTIONS *
1924 **************************/
1926 r11 = _mm_mul_pd(rsq11,rinv11);
1928 /* EWALD ELECTROSTATICS */
1930 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1931 ewrt = _mm_mul_pd(r11,ewtabscale);
1932 ewitab = _mm_cvttpd_epi32(ewrt);
1934 eweps = _mm_frcz_pd(ewrt);
1936 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1938 twoeweps = _mm_add_pd(eweps,eweps);
1939 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1940 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1941 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1945 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1947 /* Update vectorial force */
1948 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1949 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1950 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1952 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1953 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1954 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1956 /**************************
1957 * CALCULATE INTERACTIONS *
1958 **************************/
1960 r12 = _mm_mul_pd(rsq12,rinv12);
1962 /* EWALD ELECTROSTATICS */
1964 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1965 ewrt = _mm_mul_pd(r12,ewtabscale);
1966 ewitab = _mm_cvttpd_epi32(ewrt);
1968 eweps = _mm_frcz_pd(ewrt);
1970 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1972 twoeweps = _mm_add_pd(eweps,eweps);
1973 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1974 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1975 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1979 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1981 /* Update vectorial force */
1982 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1983 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1984 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1986 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1987 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1988 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1990 /**************************
1991 * CALCULATE INTERACTIONS *
1992 **************************/
1994 r20 = _mm_mul_pd(rsq20,rinv20);
1996 /* EWALD ELECTROSTATICS */
1998 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1999 ewrt = _mm_mul_pd(r20,ewtabscale);
2000 ewitab = _mm_cvttpd_epi32(ewrt);
2002 eweps = _mm_frcz_pd(ewrt);
2004 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2006 twoeweps = _mm_add_pd(eweps,eweps);
2007 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2008 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2009 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2013 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2015 /* Update vectorial force */
2016 fix2 = _mm_macc_pd(dx20,fscal,fix2);
2017 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
2018 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
2020 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
2021 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
2022 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
2024 /**************************
2025 * CALCULATE INTERACTIONS *
2026 **************************/
2028 r21 = _mm_mul_pd(rsq21,rinv21);
2030 /* EWALD ELECTROSTATICS */
2032 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2033 ewrt = _mm_mul_pd(r21,ewtabscale);
2034 ewitab = _mm_cvttpd_epi32(ewrt);
2036 eweps = _mm_frcz_pd(ewrt);
2038 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2040 twoeweps = _mm_add_pd(eweps,eweps);
2041 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2042 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2043 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2047 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2049 /* Update vectorial force */
2050 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2051 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2052 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2054 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2055 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2056 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2058 /**************************
2059 * CALCULATE INTERACTIONS *
2060 **************************/
2062 r22 = _mm_mul_pd(rsq22,rinv22);
2064 /* EWALD ELECTROSTATICS */
2066 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2067 ewrt = _mm_mul_pd(r22,ewtabscale);
2068 ewitab = _mm_cvttpd_epi32(ewrt);
2070 eweps = _mm_frcz_pd(ewrt);
2072 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2074 twoeweps = _mm_add_pd(eweps,eweps);
2075 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2076 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2077 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2081 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2083 /* Update vectorial force */
2084 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2085 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2086 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2088 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2089 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2090 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2092 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2094 /* Inner loop uses 351 flops */
2097 /* End of innermost loop */
2099 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2100 f+i_coord_offset,fshift+i_shift_offset);
2102 /* Increment number of inner iterations */
2103 inneriter += j_index_end - j_index_start;
2105 /* Outer loop uses 18 flops */
2108 /* Increment number of outer iterations */
2111 /* Update outer/inner flops */
2113 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*351);