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.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_avx_128_fma_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: None
54 * Geometry: Water4-Water4
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_avx_128_fma_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
83 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
85 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
86 int vdwjidx1A,vdwjidx1B;
87 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
88 int vdwjidx2A,vdwjidx2B;
89 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
90 int vdwjidx3A,vdwjidx3B;
91 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
92 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
93 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
94 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
95 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
96 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
97 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
98 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
99 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
100 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
101 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
104 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
106 __m128d dummy_mask,cutoff_mask;
107 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
108 __m128d one = _mm_set1_pd(1.0);
109 __m128d two = _mm_set1_pd(2.0);
115 jindex = nlist->jindex;
117 shiftidx = nlist->shift;
119 shiftvec = fr->shift_vec[0];
120 fshift = fr->fshift[0];
121 facel = _mm_set1_pd(fr->epsfac);
122 charge = mdatoms->chargeA;
124 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
125 ewtab = fr->ic->tabq_coul_FDV0;
126 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
127 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
129 /* Setup water-specific parameters */
130 inr = nlist->iinr[0];
131 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
132 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
133 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
135 jq1 = _mm_set1_pd(charge[inr+1]);
136 jq2 = _mm_set1_pd(charge[inr+2]);
137 jq3 = _mm_set1_pd(charge[inr+3]);
138 qq11 = _mm_mul_pd(iq1,jq1);
139 qq12 = _mm_mul_pd(iq1,jq2);
140 qq13 = _mm_mul_pd(iq1,jq3);
141 qq21 = _mm_mul_pd(iq2,jq1);
142 qq22 = _mm_mul_pd(iq2,jq2);
143 qq23 = _mm_mul_pd(iq2,jq3);
144 qq31 = _mm_mul_pd(iq3,jq1);
145 qq32 = _mm_mul_pd(iq3,jq2);
146 qq33 = _mm_mul_pd(iq3,jq3);
148 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
149 rcutoff_scalar = fr->rcoulomb;
150 rcutoff = _mm_set1_pd(rcutoff_scalar);
151 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
153 /* Avoid stupid compiler warnings */
161 /* Start outer loop over neighborlists */
162 for(iidx=0; iidx<nri; iidx++)
164 /* Load shift vector for this list */
165 i_shift_offset = DIM*shiftidx[iidx];
167 /* Load limits for loop over neighbors */
168 j_index_start = jindex[iidx];
169 j_index_end = jindex[iidx+1];
171 /* Get outer coordinate index */
173 i_coord_offset = DIM*inr;
175 /* Load i particle coords and add shift vector */
176 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
177 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
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();
185 fix3 = _mm_setzero_pd();
186 fiy3 = _mm_setzero_pd();
187 fiz3 = _mm_setzero_pd();
189 /* Reset potential sums */
190 velecsum = _mm_setzero_pd();
192 /* Start inner kernel loop */
193 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
196 /* Get j neighbor index, and coordinate index */
199 j_coord_offsetA = DIM*jnrA;
200 j_coord_offsetB = DIM*jnrB;
202 /* load j atom coordinates */
203 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
204 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
206 /* Calculate displacement vector */
207 dx11 = _mm_sub_pd(ix1,jx1);
208 dy11 = _mm_sub_pd(iy1,jy1);
209 dz11 = _mm_sub_pd(iz1,jz1);
210 dx12 = _mm_sub_pd(ix1,jx2);
211 dy12 = _mm_sub_pd(iy1,jy2);
212 dz12 = _mm_sub_pd(iz1,jz2);
213 dx13 = _mm_sub_pd(ix1,jx3);
214 dy13 = _mm_sub_pd(iy1,jy3);
215 dz13 = _mm_sub_pd(iz1,jz3);
216 dx21 = _mm_sub_pd(ix2,jx1);
217 dy21 = _mm_sub_pd(iy2,jy1);
218 dz21 = _mm_sub_pd(iz2,jz1);
219 dx22 = _mm_sub_pd(ix2,jx2);
220 dy22 = _mm_sub_pd(iy2,jy2);
221 dz22 = _mm_sub_pd(iz2,jz2);
222 dx23 = _mm_sub_pd(ix2,jx3);
223 dy23 = _mm_sub_pd(iy2,jy3);
224 dz23 = _mm_sub_pd(iz2,jz3);
225 dx31 = _mm_sub_pd(ix3,jx1);
226 dy31 = _mm_sub_pd(iy3,jy1);
227 dz31 = _mm_sub_pd(iz3,jz1);
228 dx32 = _mm_sub_pd(ix3,jx2);
229 dy32 = _mm_sub_pd(iy3,jy2);
230 dz32 = _mm_sub_pd(iz3,jz2);
231 dx33 = _mm_sub_pd(ix3,jx3);
232 dy33 = _mm_sub_pd(iy3,jy3);
233 dz33 = _mm_sub_pd(iz3,jz3);
235 /* Calculate squared distance and things based on it */
236 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
237 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
238 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
239 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
240 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
241 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
242 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
243 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
244 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
246 rinv11 = gmx_mm_invsqrt_pd(rsq11);
247 rinv12 = gmx_mm_invsqrt_pd(rsq12);
248 rinv13 = gmx_mm_invsqrt_pd(rsq13);
249 rinv21 = gmx_mm_invsqrt_pd(rsq21);
250 rinv22 = gmx_mm_invsqrt_pd(rsq22);
251 rinv23 = gmx_mm_invsqrt_pd(rsq23);
252 rinv31 = gmx_mm_invsqrt_pd(rsq31);
253 rinv32 = gmx_mm_invsqrt_pd(rsq32);
254 rinv33 = gmx_mm_invsqrt_pd(rsq33);
256 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
257 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
258 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
259 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
260 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
261 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
262 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
263 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
264 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
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();
272 fjx3 = _mm_setzero_pd();
273 fjy3 = _mm_setzero_pd();
274 fjz3 = _mm_setzero_pd();
276 /**************************
277 * CALCULATE INTERACTIONS *
278 **************************/
280 if (gmx_mm_any_lt(rsq11,rcutoff2))
283 r11 = _mm_mul_pd(rsq11,rinv11);
285 /* EWALD ELECTROSTATICS */
287 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
288 ewrt = _mm_mul_pd(r11,ewtabscale);
289 ewitab = _mm_cvttpd_epi32(ewrt);
291 eweps = _mm_frcz_pd(ewrt);
293 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
295 twoeweps = _mm_add_pd(eweps,eweps);
296 ewitab = _mm_slli_epi32(ewitab,2);
297 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
298 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
299 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
300 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
301 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
302 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
303 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
304 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
305 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
306 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
308 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
310 /* Update potential sum for this i atom from the interaction with this j atom. */
311 velec = _mm_and_pd(velec,cutoff_mask);
312 velecsum = _mm_add_pd(velecsum,velec);
316 fscal = _mm_and_pd(fscal,cutoff_mask);
318 /* Update vectorial force */
319 fix1 = _mm_macc_pd(dx11,fscal,fix1);
320 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
321 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
323 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
324 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
325 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
329 /**************************
330 * CALCULATE INTERACTIONS *
331 **************************/
333 if (gmx_mm_any_lt(rsq12,rcutoff2))
336 r12 = _mm_mul_pd(rsq12,rinv12);
338 /* EWALD ELECTROSTATICS */
340 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
341 ewrt = _mm_mul_pd(r12,ewtabscale);
342 ewitab = _mm_cvttpd_epi32(ewrt);
344 eweps = _mm_frcz_pd(ewrt);
346 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
348 twoeweps = _mm_add_pd(eweps,eweps);
349 ewitab = _mm_slli_epi32(ewitab,2);
350 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
351 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
352 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
353 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
354 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
355 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
356 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
357 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
358 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
359 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
361 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
363 /* Update potential sum for this i atom from the interaction with this j atom. */
364 velec = _mm_and_pd(velec,cutoff_mask);
365 velecsum = _mm_add_pd(velecsum,velec);
369 fscal = _mm_and_pd(fscal,cutoff_mask);
371 /* Update vectorial force */
372 fix1 = _mm_macc_pd(dx12,fscal,fix1);
373 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
374 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
376 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
377 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
378 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
382 /**************************
383 * CALCULATE INTERACTIONS *
384 **************************/
386 if (gmx_mm_any_lt(rsq13,rcutoff2))
389 r13 = _mm_mul_pd(rsq13,rinv13);
391 /* EWALD ELECTROSTATICS */
393 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
394 ewrt = _mm_mul_pd(r13,ewtabscale);
395 ewitab = _mm_cvttpd_epi32(ewrt);
397 eweps = _mm_frcz_pd(ewrt);
399 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
401 twoeweps = _mm_add_pd(eweps,eweps);
402 ewitab = _mm_slli_epi32(ewitab,2);
403 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
404 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
405 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
406 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
407 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
408 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
409 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
410 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
411 velec = _mm_mul_pd(qq13,_mm_sub_pd(_mm_sub_pd(rinv13,sh_ewald),velec));
412 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
414 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
416 /* Update potential sum for this i atom from the interaction with this j atom. */
417 velec = _mm_and_pd(velec,cutoff_mask);
418 velecsum = _mm_add_pd(velecsum,velec);
422 fscal = _mm_and_pd(fscal,cutoff_mask);
424 /* Update vectorial force */
425 fix1 = _mm_macc_pd(dx13,fscal,fix1);
426 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
427 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
429 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
430 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
431 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
435 /**************************
436 * CALCULATE INTERACTIONS *
437 **************************/
439 if (gmx_mm_any_lt(rsq21,rcutoff2))
442 r21 = _mm_mul_pd(rsq21,rinv21);
444 /* EWALD ELECTROSTATICS */
446 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
447 ewrt = _mm_mul_pd(r21,ewtabscale);
448 ewitab = _mm_cvttpd_epi32(ewrt);
450 eweps = _mm_frcz_pd(ewrt);
452 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
454 twoeweps = _mm_add_pd(eweps,eweps);
455 ewitab = _mm_slli_epi32(ewitab,2);
456 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
457 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
458 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
459 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
460 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
461 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
462 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
463 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
464 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
465 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
467 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
469 /* Update potential sum for this i atom from the interaction with this j atom. */
470 velec = _mm_and_pd(velec,cutoff_mask);
471 velecsum = _mm_add_pd(velecsum,velec);
475 fscal = _mm_and_pd(fscal,cutoff_mask);
477 /* Update vectorial force */
478 fix2 = _mm_macc_pd(dx21,fscal,fix2);
479 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
480 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
482 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
483 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
484 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
488 /**************************
489 * CALCULATE INTERACTIONS *
490 **************************/
492 if (gmx_mm_any_lt(rsq22,rcutoff2))
495 r22 = _mm_mul_pd(rsq22,rinv22);
497 /* EWALD ELECTROSTATICS */
499 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
500 ewrt = _mm_mul_pd(r22,ewtabscale);
501 ewitab = _mm_cvttpd_epi32(ewrt);
503 eweps = _mm_frcz_pd(ewrt);
505 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
507 twoeweps = _mm_add_pd(eweps,eweps);
508 ewitab = _mm_slli_epi32(ewitab,2);
509 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
510 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
511 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
512 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
513 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
514 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
515 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
516 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
517 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
518 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
520 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velec = _mm_and_pd(velec,cutoff_mask);
524 velecsum = _mm_add_pd(velecsum,velec);
528 fscal = _mm_and_pd(fscal,cutoff_mask);
530 /* Update vectorial force */
531 fix2 = _mm_macc_pd(dx22,fscal,fix2);
532 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
533 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
535 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
536 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
537 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
541 /**************************
542 * CALCULATE INTERACTIONS *
543 **************************/
545 if (gmx_mm_any_lt(rsq23,rcutoff2))
548 r23 = _mm_mul_pd(rsq23,rinv23);
550 /* EWALD ELECTROSTATICS */
552 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
553 ewrt = _mm_mul_pd(r23,ewtabscale);
554 ewitab = _mm_cvttpd_epi32(ewrt);
556 eweps = _mm_frcz_pd(ewrt);
558 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
560 twoeweps = _mm_add_pd(eweps,eweps);
561 ewitab = _mm_slli_epi32(ewitab,2);
562 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
563 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
564 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
565 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
566 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
567 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
568 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
569 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
570 velec = _mm_mul_pd(qq23,_mm_sub_pd(_mm_sub_pd(rinv23,sh_ewald),velec));
571 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
573 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
575 /* Update potential sum for this i atom from the interaction with this j atom. */
576 velec = _mm_and_pd(velec,cutoff_mask);
577 velecsum = _mm_add_pd(velecsum,velec);
581 fscal = _mm_and_pd(fscal,cutoff_mask);
583 /* Update vectorial force */
584 fix2 = _mm_macc_pd(dx23,fscal,fix2);
585 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
586 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
588 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
589 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
590 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
594 /**************************
595 * CALCULATE INTERACTIONS *
596 **************************/
598 if (gmx_mm_any_lt(rsq31,rcutoff2))
601 r31 = _mm_mul_pd(rsq31,rinv31);
603 /* EWALD ELECTROSTATICS */
605 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
606 ewrt = _mm_mul_pd(r31,ewtabscale);
607 ewitab = _mm_cvttpd_epi32(ewrt);
609 eweps = _mm_frcz_pd(ewrt);
611 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
613 twoeweps = _mm_add_pd(eweps,eweps);
614 ewitab = _mm_slli_epi32(ewitab,2);
615 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
616 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
617 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
618 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
619 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
620 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
621 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
622 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
623 velec = _mm_mul_pd(qq31,_mm_sub_pd(_mm_sub_pd(rinv31,sh_ewald),velec));
624 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
626 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
628 /* Update potential sum for this i atom from the interaction with this j atom. */
629 velec = _mm_and_pd(velec,cutoff_mask);
630 velecsum = _mm_add_pd(velecsum,velec);
634 fscal = _mm_and_pd(fscal,cutoff_mask);
636 /* Update vectorial force */
637 fix3 = _mm_macc_pd(dx31,fscal,fix3);
638 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
639 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
641 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
642 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
643 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
647 /**************************
648 * CALCULATE INTERACTIONS *
649 **************************/
651 if (gmx_mm_any_lt(rsq32,rcutoff2))
654 r32 = _mm_mul_pd(rsq32,rinv32);
656 /* EWALD ELECTROSTATICS */
658 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
659 ewrt = _mm_mul_pd(r32,ewtabscale);
660 ewitab = _mm_cvttpd_epi32(ewrt);
662 eweps = _mm_frcz_pd(ewrt);
664 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
666 twoeweps = _mm_add_pd(eweps,eweps);
667 ewitab = _mm_slli_epi32(ewitab,2);
668 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
669 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
670 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
671 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
672 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
673 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
674 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
675 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
676 velec = _mm_mul_pd(qq32,_mm_sub_pd(_mm_sub_pd(rinv32,sh_ewald),velec));
677 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
679 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
681 /* Update potential sum for this i atom from the interaction with this j atom. */
682 velec = _mm_and_pd(velec,cutoff_mask);
683 velecsum = _mm_add_pd(velecsum,velec);
687 fscal = _mm_and_pd(fscal,cutoff_mask);
689 /* Update vectorial force */
690 fix3 = _mm_macc_pd(dx32,fscal,fix3);
691 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
692 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
694 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
695 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
696 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
700 /**************************
701 * CALCULATE INTERACTIONS *
702 **************************/
704 if (gmx_mm_any_lt(rsq33,rcutoff2))
707 r33 = _mm_mul_pd(rsq33,rinv33);
709 /* EWALD ELECTROSTATICS */
711 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
712 ewrt = _mm_mul_pd(r33,ewtabscale);
713 ewitab = _mm_cvttpd_epi32(ewrt);
715 eweps = _mm_frcz_pd(ewrt);
717 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
719 twoeweps = _mm_add_pd(eweps,eweps);
720 ewitab = _mm_slli_epi32(ewitab,2);
721 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
722 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
723 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
724 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
725 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
726 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
727 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
728 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
729 velec = _mm_mul_pd(qq33,_mm_sub_pd(_mm_sub_pd(rinv33,sh_ewald),velec));
730 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
732 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
734 /* Update potential sum for this i atom from the interaction with this j atom. */
735 velec = _mm_and_pd(velec,cutoff_mask);
736 velecsum = _mm_add_pd(velecsum,velec);
740 fscal = _mm_and_pd(fscal,cutoff_mask);
742 /* Update vectorial force */
743 fix3 = _mm_macc_pd(dx33,fscal,fix3);
744 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
745 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
747 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
748 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
749 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
753 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
755 /* Inner loop uses 441 flops */
762 j_coord_offsetA = DIM*jnrA;
764 /* load j atom coordinates */
765 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
766 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
768 /* Calculate displacement vector */
769 dx11 = _mm_sub_pd(ix1,jx1);
770 dy11 = _mm_sub_pd(iy1,jy1);
771 dz11 = _mm_sub_pd(iz1,jz1);
772 dx12 = _mm_sub_pd(ix1,jx2);
773 dy12 = _mm_sub_pd(iy1,jy2);
774 dz12 = _mm_sub_pd(iz1,jz2);
775 dx13 = _mm_sub_pd(ix1,jx3);
776 dy13 = _mm_sub_pd(iy1,jy3);
777 dz13 = _mm_sub_pd(iz1,jz3);
778 dx21 = _mm_sub_pd(ix2,jx1);
779 dy21 = _mm_sub_pd(iy2,jy1);
780 dz21 = _mm_sub_pd(iz2,jz1);
781 dx22 = _mm_sub_pd(ix2,jx2);
782 dy22 = _mm_sub_pd(iy2,jy2);
783 dz22 = _mm_sub_pd(iz2,jz2);
784 dx23 = _mm_sub_pd(ix2,jx3);
785 dy23 = _mm_sub_pd(iy2,jy3);
786 dz23 = _mm_sub_pd(iz2,jz3);
787 dx31 = _mm_sub_pd(ix3,jx1);
788 dy31 = _mm_sub_pd(iy3,jy1);
789 dz31 = _mm_sub_pd(iz3,jz1);
790 dx32 = _mm_sub_pd(ix3,jx2);
791 dy32 = _mm_sub_pd(iy3,jy2);
792 dz32 = _mm_sub_pd(iz3,jz2);
793 dx33 = _mm_sub_pd(ix3,jx3);
794 dy33 = _mm_sub_pd(iy3,jy3);
795 dz33 = _mm_sub_pd(iz3,jz3);
797 /* Calculate squared distance and things based on it */
798 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
799 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
800 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
801 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
802 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
803 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
804 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
805 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
806 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
808 rinv11 = gmx_mm_invsqrt_pd(rsq11);
809 rinv12 = gmx_mm_invsqrt_pd(rsq12);
810 rinv13 = gmx_mm_invsqrt_pd(rsq13);
811 rinv21 = gmx_mm_invsqrt_pd(rsq21);
812 rinv22 = gmx_mm_invsqrt_pd(rsq22);
813 rinv23 = gmx_mm_invsqrt_pd(rsq23);
814 rinv31 = gmx_mm_invsqrt_pd(rsq31);
815 rinv32 = gmx_mm_invsqrt_pd(rsq32);
816 rinv33 = gmx_mm_invsqrt_pd(rsq33);
818 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
819 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
820 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
821 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
822 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
823 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
824 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
825 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
826 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
828 fjx1 = _mm_setzero_pd();
829 fjy1 = _mm_setzero_pd();
830 fjz1 = _mm_setzero_pd();
831 fjx2 = _mm_setzero_pd();
832 fjy2 = _mm_setzero_pd();
833 fjz2 = _mm_setzero_pd();
834 fjx3 = _mm_setzero_pd();
835 fjy3 = _mm_setzero_pd();
836 fjz3 = _mm_setzero_pd();
838 /**************************
839 * CALCULATE INTERACTIONS *
840 **************************/
842 if (gmx_mm_any_lt(rsq11,rcutoff2))
845 r11 = _mm_mul_pd(rsq11,rinv11);
847 /* EWALD ELECTROSTATICS */
849 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
850 ewrt = _mm_mul_pd(r11,ewtabscale);
851 ewitab = _mm_cvttpd_epi32(ewrt);
853 eweps = _mm_frcz_pd(ewrt);
855 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
857 twoeweps = _mm_add_pd(eweps,eweps);
858 ewitab = _mm_slli_epi32(ewitab,2);
859 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
860 ewtabD = _mm_setzero_pd();
861 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
862 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
863 ewtabFn = _mm_setzero_pd();
864 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
865 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
866 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
867 velec = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
868 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
870 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
872 /* Update potential sum for this i atom from the interaction with this j atom. */
873 velec = _mm_and_pd(velec,cutoff_mask);
874 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
875 velecsum = _mm_add_pd(velecsum,velec);
879 fscal = _mm_and_pd(fscal,cutoff_mask);
881 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
883 /* Update vectorial force */
884 fix1 = _mm_macc_pd(dx11,fscal,fix1);
885 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
886 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
888 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
889 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
890 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
894 /**************************
895 * CALCULATE INTERACTIONS *
896 **************************/
898 if (gmx_mm_any_lt(rsq12,rcutoff2))
901 r12 = _mm_mul_pd(rsq12,rinv12);
903 /* EWALD ELECTROSTATICS */
905 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
906 ewrt = _mm_mul_pd(r12,ewtabscale);
907 ewitab = _mm_cvttpd_epi32(ewrt);
909 eweps = _mm_frcz_pd(ewrt);
911 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
913 twoeweps = _mm_add_pd(eweps,eweps);
914 ewitab = _mm_slli_epi32(ewitab,2);
915 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
916 ewtabD = _mm_setzero_pd();
917 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
918 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
919 ewtabFn = _mm_setzero_pd();
920 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
921 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
922 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
923 velec = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
924 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
926 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
928 /* Update potential sum for this i atom from the interaction with this j atom. */
929 velec = _mm_and_pd(velec,cutoff_mask);
930 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
931 velecsum = _mm_add_pd(velecsum,velec);
935 fscal = _mm_and_pd(fscal,cutoff_mask);
937 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
939 /* Update vectorial force */
940 fix1 = _mm_macc_pd(dx12,fscal,fix1);
941 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
942 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
944 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
945 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
946 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
950 /**************************
951 * CALCULATE INTERACTIONS *
952 **************************/
954 if (gmx_mm_any_lt(rsq13,rcutoff2))
957 r13 = _mm_mul_pd(rsq13,rinv13);
959 /* EWALD ELECTROSTATICS */
961 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
962 ewrt = _mm_mul_pd(r13,ewtabscale);
963 ewitab = _mm_cvttpd_epi32(ewrt);
965 eweps = _mm_frcz_pd(ewrt);
967 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
969 twoeweps = _mm_add_pd(eweps,eweps);
970 ewitab = _mm_slli_epi32(ewitab,2);
971 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
972 ewtabD = _mm_setzero_pd();
973 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
974 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
975 ewtabFn = _mm_setzero_pd();
976 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
977 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
978 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
979 velec = _mm_mul_pd(qq13,_mm_sub_pd(_mm_sub_pd(rinv13,sh_ewald),velec));
980 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
982 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
984 /* Update potential sum for this i atom from the interaction with this j atom. */
985 velec = _mm_and_pd(velec,cutoff_mask);
986 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
987 velecsum = _mm_add_pd(velecsum,velec);
991 fscal = _mm_and_pd(fscal,cutoff_mask);
993 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
995 /* Update vectorial force */
996 fix1 = _mm_macc_pd(dx13,fscal,fix1);
997 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
998 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1000 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1001 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1002 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1006 /**************************
1007 * CALCULATE INTERACTIONS *
1008 **************************/
1010 if (gmx_mm_any_lt(rsq21,rcutoff2))
1013 r21 = _mm_mul_pd(rsq21,rinv21);
1015 /* EWALD ELECTROSTATICS */
1017 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1018 ewrt = _mm_mul_pd(r21,ewtabscale);
1019 ewitab = _mm_cvttpd_epi32(ewrt);
1021 eweps = _mm_frcz_pd(ewrt);
1023 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1025 twoeweps = _mm_add_pd(eweps,eweps);
1026 ewitab = _mm_slli_epi32(ewitab,2);
1027 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1028 ewtabD = _mm_setzero_pd();
1029 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1030 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1031 ewtabFn = _mm_setzero_pd();
1032 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1033 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1034 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1035 velec = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1036 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1038 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1040 /* Update potential sum for this i atom from the interaction with this j atom. */
1041 velec = _mm_and_pd(velec,cutoff_mask);
1042 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1043 velecsum = _mm_add_pd(velecsum,velec);
1047 fscal = _mm_and_pd(fscal,cutoff_mask);
1049 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1051 /* Update vectorial force */
1052 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1053 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1054 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1056 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1057 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1058 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1062 /**************************
1063 * CALCULATE INTERACTIONS *
1064 **************************/
1066 if (gmx_mm_any_lt(rsq22,rcutoff2))
1069 r22 = _mm_mul_pd(rsq22,rinv22);
1071 /* EWALD ELECTROSTATICS */
1073 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1074 ewrt = _mm_mul_pd(r22,ewtabscale);
1075 ewitab = _mm_cvttpd_epi32(ewrt);
1077 eweps = _mm_frcz_pd(ewrt);
1079 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1081 twoeweps = _mm_add_pd(eweps,eweps);
1082 ewitab = _mm_slli_epi32(ewitab,2);
1083 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1084 ewtabD = _mm_setzero_pd();
1085 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1086 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1087 ewtabFn = _mm_setzero_pd();
1088 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1089 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1090 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1091 velec = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1092 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1094 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1096 /* Update potential sum for this i atom from the interaction with this j atom. */
1097 velec = _mm_and_pd(velec,cutoff_mask);
1098 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1099 velecsum = _mm_add_pd(velecsum,velec);
1103 fscal = _mm_and_pd(fscal,cutoff_mask);
1105 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1107 /* Update vectorial force */
1108 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1109 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1110 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1112 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1113 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1114 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1118 /**************************
1119 * CALCULATE INTERACTIONS *
1120 **************************/
1122 if (gmx_mm_any_lt(rsq23,rcutoff2))
1125 r23 = _mm_mul_pd(rsq23,rinv23);
1127 /* EWALD ELECTROSTATICS */
1129 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1130 ewrt = _mm_mul_pd(r23,ewtabscale);
1131 ewitab = _mm_cvttpd_epi32(ewrt);
1133 eweps = _mm_frcz_pd(ewrt);
1135 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1137 twoeweps = _mm_add_pd(eweps,eweps);
1138 ewitab = _mm_slli_epi32(ewitab,2);
1139 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1140 ewtabD = _mm_setzero_pd();
1141 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1142 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1143 ewtabFn = _mm_setzero_pd();
1144 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1145 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1146 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1147 velec = _mm_mul_pd(qq23,_mm_sub_pd(_mm_sub_pd(rinv23,sh_ewald),velec));
1148 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1150 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
1152 /* Update potential sum for this i atom from the interaction with this j atom. */
1153 velec = _mm_and_pd(velec,cutoff_mask);
1154 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1155 velecsum = _mm_add_pd(velecsum,velec);
1159 fscal = _mm_and_pd(fscal,cutoff_mask);
1161 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1163 /* Update vectorial force */
1164 fix2 = _mm_macc_pd(dx23,fscal,fix2);
1165 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
1166 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
1168 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
1169 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
1170 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
1174 /**************************
1175 * CALCULATE INTERACTIONS *
1176 **************************/
1178 if (gmx_mm_any_lt(rsq31,rcutoff2))
1181 r31 = _mm_mul_pd(rsq31,rinv31);
1183 /* EWALD ELECTROSTATICS */
1185 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1186 ewrt = _mm_mul_pd(r31,ewtabscale);
1187 ewitab = _mm_cvttpd_epi32(ewrt);
1189 eweps = _mm_frcz_pd(ewrt);
1191 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1193 twoeweps = _mm_add_pd(eweps,eweps);
1194 ewitab = _mm_slli_epi32(ewitab,2);
1195 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1196 ewtabD = _mm_setzero_pd();
1197 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1198 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1199 ewtabFn = _mm_setzero_pd();
1200 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1201 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1202 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1203 velec = _mm_mul_pd(qq31,_mm_sub_pd(_mm_sub_pd(rinv31,sh_ewald),velec));
1204 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1206 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
1208 /* Update potential sum for this i atom from the interaction with this j atom. */
1209 velec = _mm_and_pd(velec,cutoff_mask);
1210 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1211 velecsum = _mm_add_pd(velecsum,velec);
1215 fscal = _mm_and_pd(fscal,cutoff_mask);
1217 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1219 /* Update vectorial force */
1220 fix3 = _mm_macc_pd(dx31,fscal,fix3);
1221 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
1222 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
1224 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
1225 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
1226 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
1230 /**************************
1231 * CALCULATE INTERACTIONS *
1232 **************************/
1234 if (gmx_mm_any_lt(rsq32,rcutoff2))
1237 r32 = _mm_mul_pd(rsq32,rinv32);
1239 /* EWALD ELECTROSTATICS */
1241 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1242 ewrt = _mm_mul_pd(r32,ewtabscale);
1243 ewitab = _mm_cvttpd_epi32(ewrt);
1245 eweps = _mm_frcz_pd(ewrt);
1247 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1249 twoeweps = _mm_add_pd(eweps,eweps);
1250 ewitab = _mm_slli_epi32(ewitab,2);
1251 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1252 ewtabD = _mm_setzero_pd();
1253 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1254 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1255 ewtabFn = _mm_setzero_pd();
1256 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1257 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1258 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1259 velec = _mm_mul_pd(qq32,_mm_sub_pd(_mm_sub_pd(rinv32,sh_ewald),velec));
1260 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1262 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
1264 /* Update potential sum for this i atom from the interaction with this j atom. */
1265 velec = _mm_and_pd(velec,cutoff_mask);
1266 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1267 velecsum = _mm_add_pd(velecsum,velec);
1271 fscal = _mm_and_pd(fscal,cutoff_mask);
1273 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1275 /* Update vectorial force */
1276 fix3 = _mm_macc_pd(dx32,fscal,fix3);
1277 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
1278 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
1280 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
1281 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
1282 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
1286 /**************************
1287 * CALCULATE INTERACTIONS *
1288 **************************/
1290 if (gmx_mm_any_lt(rsq33,rcutoff2))
1293 r33 = _mm_mul_pd(rsq33,rinv33);
1295 /* EWALD ELECTROSTATICS */
1297 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1298 ewrt = _mm_mul_pd(r33,ewtabscale);
1299 ewitab = _mm_cvttpd_epi32(ewrt);
1301 eweps = _mm_frcz_pd(ewrt);
1303 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1305 twoeweps = _mm_add_pd(eweps,eweps);
1306 ewitab = _mm_slli_epi32(ewitab,2);
1307 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1308 ewtabD = _mm_setzero_pd();
1309 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1310 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1311 ewtabFn = _mm_setzero_pd();
1312 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1313 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1314 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1315 velec = _mm_mul_pd(qq33,_mm_sub_pd(_mm_sub_pd(rinv33,sh_ewald),velec));
1316 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1318 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
1320 /* Update potential sum for this i atom from the interaction with this j atom. */
1321 velec = _mm_and_pd(velec,cutoff_mask);
1322 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1323 velecsum = _mm_add_pd(velecsum,velec);
1327 fscal = _mm_and_pd(fscal,cutoff_mask);
1329 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1331 /* Update vectorial force */
1332 fix3 = _mm_macc_pd(dx33,fscal,fix3);
1333 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
1334 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
1336 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
1337 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
1338 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
1342 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1344 /* Inner loop uses 441 flops */
1347 /* End of innermost loop */
1349 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1350 f+i_coord_offset+DIM,fshift+i_shift_offset);
1353 /* Update potential energies */
1354 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1356 /* Increment number of inner iterations */
1357 inneriter += j_index_end - j_index_start;
1359 /* Outer loop uses 19 flops */
1362 /* Increment number of outer iterations */
1365 /* Update outer/inner flops */
1367 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*441);
1370 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_avx_128_fma_double
1371 * Electrostatics interaction: Ewald
1372 * VdW interaction: None
1373 * Geometry: Water4-Water4
1374 * Calculate force/pot: Force
1377 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_avx_128_fma_double
1378 (t_nblist * gmx_restrict nlist,
1379 rvec * gmx_restrict xx,
1380 rvec * gmx_restrict ff,
1381 t_forcerec * gmx_restrict fr,
1382 t_mdatoms * gmx_restrict mdatoms,
1383 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1384 t_nrnb * gmx_restrict nrnb)
1386 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1387 * just 0 for non-waters.
1388 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1389 * jnr indices corresponding to data put in the four positions in the SIMD register.
1391 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1392 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1394 int j_coord_offsetA,j_coord_offsetB;
1395 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1396 real rcutoff_scalar;
1397 real *shiftvec,*fshift,*x,*f;
1398 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1400 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1402 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1404 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1405 int vdwjidx1A,vdwjidx1B;
1406 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1407 int vdwjidx2A,vdwjidx2B;
1408 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1409 int vdwjidx3A,vdwjidx3B;
1410 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1411 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1412 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1413 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1414 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1415 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1416 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1417 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1418 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1419 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1420 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1423 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1425 __m128d dummy_mask,cutoff_mask;
1426 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1427 __m128d one = _mm_set1_pd(1.0);
1428 __m128d two = _mm_set1_pd(2.0);
1434 jindex = nlist->jindex;
1436 shiftidx = nlist->shift;
1438 shiftvec = fr->shift_vec[0];
1439 fshift = fr->fshift[0];
1440 facel = _mm_set1_pd(fr->epsfac);
1441 charge = mdatoms->chargeA;
1443 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1444 ewtab = fr->ic->tabq_coul_F;
1445 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1446 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1448 /* Setup water-specific parameters */
1449 inr = nlist->iinr[0];
1450 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1451 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1452 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1454 jq1 = _mm_set1_pd(charge[inr+1]);
1455 jq2 = _mm_set1_pd(charge[inr+2]);
1456 jq3 = _mm_set1_pd(charge[inr+3]);
1457 qq11 = _mm_mul_pd(iq1,jq1);
1458 qq12 = _mm_mul_pd(iq1,jq2);
1459 qq13 = _mm_mul_pd(iq1,jq3);
1460 qq21 = _mm_mul_pd(iq2,jq1);
1461 qq22 = _mm_mul_pd(iq2,jq2);
1462 qq23 = _mm_mul_pd(iq2,jq3);
1463 qq31 = _mm_mul_pd(iq3,jq1);
1464 qq32 = _mm_mul_pd(iq3,jq2);
1465 qq33 = _mm_mul_pd(iq3,jq3);
1467 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1468 rcutoff_scalar = fr->rcoulomb;
1469 rcutoff = _mm_set1_pd(rcutoff_scalar);
1470 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
1472 /* Avoid stupid compiler warnings */
1474 j_coord_offsetA = 0;
1475 j_coord_offsetB = 0;
1480 /* Start outer loop over neighborlists */
1481 for(iidx=0; iidx<nri; iidx++)
1483 /* Load shift vector for this list */
1484 i_shift_offset = DIM*shiftidx[iidx];
1486 /* Load limits for loop over neighbors */
1487 j_index_start = jindex[iidx];
1488 j_index_end = jindex[iidx+1];
1490 /* Get outer coordinate index */
1492 i_coord_offset = DIM*inr;
1494 /* Load i particle coords and add shift vector */
1495 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1496 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1498 fix1 = _mm_setzero_pd();
1499 fiy1 = _mm_setzero_pd();
1500 fiz1 = _mm_setzero_pd();
1501 fix2 = _mm_setzero_pd();
1502 fiy2 = _mm_setzero_pd();
1503 fiz2 = _mm_setzero_pd();
1504 fix3 = _mm_setzero_pd();
1505 fiy3 = _mm_setzero_pd();
1506 fiz3 = _mm_setzero_pd();
1508 /* Start inner kernel loop */
1509 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1512 /* Get j neighbor index, and coordinate index */
1514 jnrB = jjnr[jidx+1];
1515 j_coord_offsetA = DIM*jnrA;
1516 j_coord_offsetB = DIM*jnrB;
1518 /* load j atom coordinates */
1519 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1520 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1522 /* Calculate displacement vector */
1523 dx11 = _mm_sub_pd(ix1,jx1);
1524 dy11 = _mm_sub_pd(iy1,jy1);
1525 dz11 = _mm_sub_pd(iz1,jz1);
1526 dx12 = _mm_sub_pd(ix1,jx2);
1527 dy12 = _mm_sub_pd(iy1,jy2);
1528 dz12 = _mm_sub_pd(iz1,jz2);
1529 dx13 = _mm_sub_pd(ix1,jx3);
1530 dy13 = _mm_sub_pd(iy1,jy3);
1531 dz13 = _mm_sub_pd(iz1,jz3);
1532 dx21 = _mm_sub_pd(ix2,jx1);
1533 dy21 = _mm_sub_pd(iy2,jy1);
1534 dz21 = _mm_sub_pd(iz2,jz1);
1535 dx22 = _mm_sub_pd(ix2,jx2);
1536 dy22 = _mm_sub_pd(iy2,jy2);
1537 dz22 = _mm_sub_pd(iz2,jz2);
1538 dx23 = _mm_sub_pd(ix2,jx3);
1539 dy23 = _mm_sub_pd(iy2,jy3);
1540 dz23 = _mm_sub_pd(iz2,jz3);
1541 dx31 = _mm_sub_pd(ix3,jx1);
1542 dy31 = _mm_sub_pd(iy3,jy1);
1543 dz31 = _mm_sub_pd(iz3,jz1);
1544 dx32 = _mm_sub_pd(ix3,jx2);
1545 dy32 = _mm_sub_pd(iy3,jy2);
1546 dz32 = _mm_sub_pd(iz3,jz2);
1547 dx33 = _mm_sub_pd(ix3,jx3);
1548 dy33 = _mm_sub_pd(iy3,jy3);
1549 dz33 = _mm_sub_pd(iz3,jz3);
1551 /* Calculate squared distance and things based on it */
1552 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1553 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1554 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1555 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1556 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1557 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1558 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1559 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1560 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1562 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1563 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1564 rinv13 = gmx_mm_invsqrt_pd(rsq13);
1565 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1566 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1567 rinv23 = gmx_mm_invsqrt_pd(rsq23);
1568 rinv31 = gmx_mm_invsqrt_pd(rsq31);
1569 rinv32 = gmx_mm_invsqrt_pd(rsq32);
1570 rinv33 = gmx_mm_invsqrt_pd(rsq33);
1572 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1573 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1574 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
1575 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1576 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1577 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
1578 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
1579 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
1580 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
1582 fjx1 = _mm_setzero_pd();
1583 fjy1 = _mm_setzero_pd();
1584 fjz1 = _mm_setzero_pd();
1585 fjx2 = _mm_setzero_pd();
1586 fjy2 = _mm_setzero_pd();
1587 fjz2 = _mm_setzero_pd();
1588 fjx3 = _mm_setzero_pd();
1589 fjy3 = _mm_setzero_pd();
1590 fjz3 = _mm_setzero_pd();
1592 /**************************
1593 * CALCULATE INTERACTIONS *
1594 **************************/
1596 if (gmx_mm_any_lt(rsq11,rcutoff2))
1599 r11 = _mm_mul_pd(rsq11,rinv11);
1601 /* EWALD ELECTROSTATICS */
1603 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1604 ewrt = _mm_mul_pd(r11,ewtabscale);
1605 ewitab = _mm_cvttpd_epi32(ewrt);
1607 eweps = _mm_frcz_pd(ewrt);
1609 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1611 twoeweps = _mm_add_pd(eweps,eweps);
1612 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1614 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1615 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1617 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1621 fscal = _mm_and_pd(fscal,cutoff_mask);
1623 /* Update vectorial force */
1624 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1625 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1626 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1628 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1629 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1630 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1634 /**************************
1635 * CALCULATE INTERACTIONS *
1636 **************************/
1638 if (gmx_mm_any_lt(rsq12,rcutoff2))
1641 r12 = _mm_mul_pd(rsq12,rinv12);
1643 /* EWALD ELECTROSTATICS */
1645 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1646 ewrt = _mm_mul_pd(r12,ewtabscale);
1647 ewitab = _mm_cvttpd_epi32(ewrt);
1649 eweps = _mm_frcz_pd(ewrt);
1651 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1653 twoeweps = _mm_add_pd(eweps,eweps);
1654 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1656 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1657 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1659 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1663 fscal = _mm_and_pd(fscal,cutoff_mask);
1665 /* Update vectorial force */
1666 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1667 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1668 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1670 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1671 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1672 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1676 /**************************
1677 * CALCULATE INTERACTIONS *
1678 **************************/
1680 if (gmx_mm_any_lt(rsq13,rcutoff2))
1683 r13 = _mm_mul_pd(rsq13,rinv13);
1685 /* EWALD ELECTROSTATICS */
1687 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1688 ewrt = _mm_mul_pd(r13,ewtabscale);
1689 ewitab = _mm_cvttpd_epi32(ewrt);
1691 eweps = _mm_frcz_pd(ewrt);
1693 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1695 twoeweps = _mm_add_pd(eweps,eweps);
1696 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1698 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1699 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1701 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
1705 fscal = _mm_and_pd(fscal,cutoff_mask);
1707 /* Update vectorial force */
1708 fix1 = _mm_macc_pd(dx13,fscal,fix1);
1709 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
1710 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1712 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1713 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1714 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1718 /**************************
1719 * CALCULATE INTERACTIONS *
1720 **************************/
1722 if (gmx_mm_any_lt(rsq21,rcutoff2))
1725 r21 = _mm_mul_pd(rsq21,rinv21);
1727 /* EWALD ELECTROSTATICS */
1729 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1730 ewrt = _mm_mul_pd(r21,ewtabscale);
1731 ewitab = _mm_cvttpd_epi32(ewrt);
1733 eweps = _mm_frcz_pd(ewrt);
1735 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1737 twoeweps = _mm_add_pd(eweps,eweps);
1738 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1740 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1741 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1743 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1747 fscal = _mm_and_pd(fscal,cutoff_mask);
1749 /* Update vectorial force */
1750 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1751 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1752 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1754 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1755 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1756 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1760 /**************************
1761 * CALCULATE INTERACTIONS *
1762 **************************/
1764 if (gmx_mm_any_lt(rsq22,rcutoff2))
1767 r22 = _mm_mul_pd(rsq22,rinv22);
1769 /* EWALD ELECTROSTATICS */
1771 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1772 ewrt = _mm_mul_pd(r22,ewtabscale);
1773 ewitab = _mm_cvttpd_epi32(ewrt);
1775 eweps = _mm_frcz_pd(ewrt);
1777 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1779 twoeweps = _mm_add_pd(eweps,eweps);
1780 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1782 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1783 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1785 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1789 fscal = _mm_and_pd(fscal,cutoff_mask);
1791 /* Update vectorial force */
1792 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1793 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1794 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1796 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1797 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1798 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1802 /**************************
1803 * CALCULATE INTERACTIONS *
1804 **************************/
1806 if (gmx_mm_any_lt(rsq23,rcutoff2))
1809 r23 = _mm_mul_pd(rsq23,rinv23);
1811 /* EWALD ELECTROSTATICS */
1813 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1814 ewrt = _mm_mul_pd(r23,ewtabscale);
1815 ewitab = _mm_cvttpd_epi32(ewrt);
1817 eweps = _mm_frcz_pd(ewrt);
1819 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1821 twoeweps = _mm_add_pd(eweps,eweps);
1822 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1824 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1825 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1827 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
1831 fscal = _mm_and_pd(fscal,cutoff_mask);
1833 /* Update vectorial force */
1834 fix2 = _mm_macc_pd(dx23,fscal,fix2);
1835 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
1836 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
1838 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
1839 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
1840 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
1844 /**************************
1845 * CALCULATE INTERACTIONS *
1846 **************************/
1848 if (gmx_mm_any_lt(rsq31,rcutoff2))
1851 r31 = _mm_mul_pd(rsq31,rinv31);
1853 /* EWALD ELECTROSTATICS */
1855 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1856 ewrt = _mm_mul_pd(r31,ewtabscale);
1857 ewitab = _mm_cvttpd_epi32(ewrt);
1859 eweps = _mm_frcz_pd(ewrt);
1861 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1863 twoeweps = _mm_add_pd(eweps,eweps);
1864 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1866 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1867 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1869 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
1873 fscal = _mm_and_pd(fscal,cutoff_mask);
1875 /* Update vectorial force */
1876 fix3 = _mm_macc_pd(dx31,fscal,fix3);
1877 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
1878 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
1880 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
1881 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
1882 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
1886 /**************************
1887 * CALCULATE INTERACTIONS *
1888 **************************/
1890 if (gmx_mm_any_lt(rsq32,rcutoff2))
1893 r32 = _mm_mul_pd(rsq32,rinv32);
1895 /* EWALD ELECTROSTATICS */
1897 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1898 ewrt = _mm_mul_pd(r32,ewtabscale);
1899 ewitab = _mm_cvttpd_epi32(ewrt);
1901 eweps = _mm_frcz_pd(ewrt);
1903 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1905 twoeweps = _mm_add_pd(eweps,eweps);
1906 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1908 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1909 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1911 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
1915 fscal = _mm_and_pd(fscal,cutoff_mask);
1917 /* Update vectorial force */
1918 fix3 = _mm_macc_pd(dx32,fscal,fix3);
1919 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
1920 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
1922 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
1923 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
1924 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
1928 /**************************
1929 * CALCULATE INTERACTIONS *
1930 **************************/
1932 if (gmx_mm_any_lt(rsq33,rcutoff2))
1935 r33 = _mm_mul_pd(rsq33,rinv33);
1937 /* EWALD ELECTROSTATICS */
1939 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1940 ewrt = _mm_mul_pd(r33,ewtabscale);
1941 ewitab = _mm_cvttpd_epi32(ewrt);
1943 eweps = _mm_frcz_pd(ewrt);
1945 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1947 twoeweps = _mm_add_pd(eweps,eweps);
1948 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1950 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1951 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1953 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
1957 fscal = _mm_and_pd(fscal,cutoff_mask);
1959 /* Update vectorial force */
1960 fix3 = _mm_macc_pd(dx33,fscal,fix3);
1961 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
1962 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
1964 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
1965 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
1966 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
1970 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1972 /* Inner loop uses 378 flops */
1975 if(jidx<j_index_end)
1979 j_coord_offsetA = DIM*jnrA;
1981 /* load j atom coordinates */
1982 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
1983 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1985 /* Calculate displacement vector */
1986 dx11 = _mm_sub_pd(ix1,jx1);
1987 dy11 = _mm_sub_pd(iy1,jy1);
1988 dz11 = _mm_sub_pd(iz1,jz1);
1989 dx12 = _mm_sub_pd(ix1,jx2);
1990 dy12 = _mm_sub_pd(iy1,jy2);
1991 dz12 = _mm_sub_pd(iz1,jz2);
1992 dx13 = _mm_sub_pd(ix1,jx3);
1993 dy13 = _mm_sub_pd(iy1,jy3);
1994 dz13 = _mm_sub_pd(iz1,jz3);
1995 dx21 = _mm_sub_pd(ix2,jx1);
1996 dy21 = _mm_sub_pd(iy2,jy1);
1997 dz21 = _mm_sub_pd(iz2,jz1);
1998 dx22 = _mm_sub_pd(ix2,jx2);
1999 dy22 = _mm_sub_pd(iy2,jy2);
2000 dz22 = _mm_sub_pd(iz2,jz2);
2001 dx23 = _mm_sub_pd(ix2,jx3);
2002 dy23 = _mm_sub_pd(iy2,jy3);
2003 dz23 = _mm_sub_pd(iz2,jz3);
2004 dx31 = _mm_sub_pd(ix3,jx1);
2005 dy31 = _mm_sub_pd(iy3,jy1);
2006 dz31 = _mm_sub_pd(iz3,jz1);
2007 dx32 = _mm_sub_pd(ix3,jx2);
2008 dy32 = _mm_sub_pd(iy3,jy2);
2009 dz32 = _mm_sub_pd(iz3,jz2);
2010 dx33 = _mm_sub_pd(ix3,jx3);
2011 dy33 = _mm_sub_pd(iy3,jy3);
2012 dz33 = _mm_sub_pd(iz3,jz3);
2014 /* Calculate squared distance and things based on it */
2015 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2016 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2017 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
2018 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2019 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2020 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
2021 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
2022 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
2023 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
2025 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2026 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2027 rinv13 = gmx_mm_invsqrt_pd(rsq13);
2028 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2029 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2030 rinv23 = gmx_mm_invsqrt_pd(rsq23);
2031 rinv31 = gmx_mm_invsqrt_pd(rsq31);
2032 rinv32 = gmx_mm_invsqrt_pd(rsq32);
2033 rinv33 = gmx_mm_invsqrt_pd(rsq33);
2035 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
2036 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
2037 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
2038 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
2039 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
2040 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
2041 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
2042 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
2043 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
2045 fjx1 = _mm_setzero_pd();
2046 fjy1 = _mm_setzero_pd();
2047 fjz1 = _mm_setzero_pd();
2048 fjx2 = _mm_setzero_pd();
2049 fjy2 = _mm_setzero_pd();
2050 fjz2 = _mm_setzero_pd();
2051 fjx3 = _mm_setzero_pd();
2052 fjy3 = _mm_setzero_pd();
2053 fjz3 = _mm_setzero_pd();
2055 /**************************
2056 * CALCULATE INTERACTIONS *
2057 **************************/
2059 if (gmx_mm_any_lt(rsq11,rcutoff2))
2062 r11 = _mm_mul_pd(rsq11,rinv11);
2064 /* EWALD ELECTROSTATICS */
2066 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2067 ewrt = _mm_mul_pd(r11,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(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2079 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2083 fscal = _mm_and_pd(fscal,cutoff_mask);
2085 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2087 /* Update vectorial force */
2088 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2089 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2090 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2092 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2093 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2094 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2098 /**************************
2099 * CALCULATE INTERACTIONS *
2100 **************************/
2102 if (gmx_mm_any_lt(rsq12,rcutoff2))
2105 r12 = _mm_mul_pd(rsq12,rinv12);
2107 /* EWALD ELECTROSTATICS */
2109 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2110 ewrt = _mm_mul_pd(r12,ewtabscale);
2111 ewitab = _mm_cvttpd_epi32(ewrt);
2113 eweps = _mm_frcz_pd(ewrt);
2115 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2117 twoeweps = _mm_add_pd(eweps,eweps);
2118 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2119 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2120 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2122 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2126 fscal = _mm_and_pd(fscal,cutoff_mask);
2128 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2130 /* Update vectorial force */
2131 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2132 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2133 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2135 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2136 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2137 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2141 /**************************
2142 * CALCULATE INTERACTIONS *
2143 **************************/
2145 if (gmx_mm_any_lt(rsq13,rcutoff2))
2148 r13 = _mm_mul_pd(rsq13,rinv13);
2150 /* EWALD ELECTROSTATICS */
2152 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2153 ewrt = _mm_mul_pd(r13,ewtabscale);
2154 ewitab = _mm_cvttpd_epi32(ewrt);
2156 eweps = _mm_frcz_pd(ewrt);
2158 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2160 twoeweps = _mm_add_pd(eweps,eweps);
2161 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2162 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2163 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2165 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
2169 fscal = _mm_and_pd(fscal,cutoff_mask);
2171 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2173 /* Update vectorial force */
2174 fix1 = _mm_macc_pd(dx13,fscal,fix1);
2175 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
2176 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
2178 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
2179 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
2180 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
2184 /**************************
2185 * CALCULATE INTERACTIONS *
2186 **************************/
2188 if (gmx_mm_any_lt(rsq21,rcutoff2))
2191 r21 = _mm_mul_pd(rsq21,rinv21);
2193 /* EWALD ELECTROSTATICS */
2195 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2196 ewrt = _mm_mul_pd(r21,ewtabscale);
2197 ewitab = _mm_cvttpd_epi32(ewrt);
2199 eweps = _mm_frcz_pd(ewrt);
2201 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2203 twoeweps = _mm_add_pd(eweps,eweps);
2204 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2205 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2206 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2208 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2212 fscal = _mm_and_pd(fscal,cutoff_mask);
2214 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2216 /* Update vectorial force */
2217 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2218 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2219 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2221 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2222 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2223 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2227 /**************************
2228 * CALCULATE INTERACTIONS *
2229 **************************/
2231 if (gmx_mm_any_lt(rsq22,rcutoff2))
2234 r22 = _mm_mul_pd(rsq22,rinv22);
2236 /* EWALD ELECTROSTATICS */
2238 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2239 ewrt = _mm_mul_pd(r22,ewtabscale);
2240 ewitab = _mm_cvttpd_epi32(ewrt);
2242 eweps = _mm_frcz_pd(ewrt);
2244 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2246 twoeweps = _mm_add_pd(eweps,eweps);
2247 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2248 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2249 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2251 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2255 fscal = _mm_and_pd(fscal,cutoff_mask);
2257 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2259 /* Update vectorial force */
2260 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2261 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2262 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2264 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2265 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2266 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2270 /**************************
2271 * CALCULATE INTERACTIONS *
2272 **************************/
2274 if (gmx_mm_any_lt(rsq23,rcutoff2))
2277 r23 = _mm_mul_pd(rsq23,rinv23);
2279 /* EWALD ELECTROSTATICS */
2281 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2282 ewrt = _mm_mul_pd(r23,ewtabscale);
2283 ewitab = _mm_cvttpd_epi32(ewrt);
2285 eweps = _mm_frcz_pd(ewrt);
2287 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2289 twoeweps = _mm_add_pd(eweps,eweps);
2290 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2291 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2292 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2294 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
2298 fscal = _mm_and_pd(fscal,cutoff_mask);
2300 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2302 /* Update vectorial force */
2303 fix2 = _mm_macc_pd(dx23,fscal,fix2);
2304 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
2305 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
2307 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
2308 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
2309 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
2313 /**************************
2314 * CALCULATE INTERACTIONS *
2315 **************************/
2317 if (gmx_mm_any_lt(rsq31,rcutoff2))
2320 r31 = _mm_mul_pd(rsq31,rinv31);
2322 /* EWALD ELECTROSTATICS */
2324 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2325 ewrt = _mm_mul_pd(r31,ewtabscale);
2326 ewitab = _mm_cvttpd_epi32(ewrt);
2328 eweps = _mm_frcz_pd(ewrt);
2330 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2332 twoeweps = _mm_add_pd(eweps,eweps);
2333 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2334 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2335 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2337 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
2341 fscal = _mm_and_pd(fscal,cutoff_mask);
2343 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2345 /* Update vectorial force */
2346 fix3 = _mm_macc_pd(dx31,fscal,fix3);
2347 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
2348 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
2350 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
2351 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
2352 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
2356 /**************************
2357 * CALCULATE INTERACTIONS *
2358 **************************/
2360 if (gmx_mm_any_lt(rsq32,rcutoff2))
2363 r32 = _mm_mul_pd(rsq32,rinv32);
2365 /* EWALD ELECTROSTATICS */
2367 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2368 ewrt = _mm_mul_pd(r32,ewtabscale);
2369 ewitab = _mm_cvttpd_epi32(ewrt);
2371 eweps = _mm_frcz_pd(ewrt);
2373 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2375 twoeweps = _mm_add_pd(eweps,eweps);
2376 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2377 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2378 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2380 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
2384 fscal = _mm_and_pd(fscal,cutoff_mask);
2386 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2388 /* Update vectorial force */
2389 fix3 = _mm_macc_pd(dx32,fscal,fix3);
2390 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
2391 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
2393 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
2394 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
2395 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
2399 /**************************
2400 * CALCULATE INTERACTIONS *
2401 **************************/
2403 if (gmx_mm_any_lt(rsq33,rcutoff2))
2406 r33 = _mm_mul_pd(rsq33,rinv33);
2408 /* EWALD ELECTROSTATICS */
2410 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2411 ewrt = _mm_mul_pd(r33,ewtabscale);
2412 ewitab = _mm_cvttpd_epi32(ewrt);
2414 eweps = _mm_frcz_pd(ewrt);
2416 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2418 twoeweps = _mm_add_pd(eweps,eweps);
2419 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2420 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2421 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2423 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
2427 fscal = _mm_and_pd(fscal,cutoff_mask);
2429 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2431 /* Update vectorial force */
2432 fix3 = _mm_macc_pd(dx33,fscal,fix3);
2433 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
2434 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
2436 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
2437 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
2438 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
2442 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2444 /* Inner loop uses 378 flops */
2447 /* End of innermost loop */
2449 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2450 f+i_coord_offset+DIM,fshift+i_shift_offset);
2452 /* Increment number of inner iterations */
2453 inneriter += j_index_end - j_index_start;
2455 /* Outer loop uses 18 flops */
2458 /* Increment number of outer iterations */
2461 /* Update outer/inner flops */
2463 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*378);