2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_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_ElecEwSw_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 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
109 real rswitch_scalar,d_scalar;
110 __m128d dummy_mask,cutoff_mask;
111 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
112 __m128d one = _mm_set1_pd(1.0);
113 __m128d two = _mm_set1_pd(2.0);
119 jindex = nlist->jindex;
121 shiftidx = nlist->shift;
123 shiftvec = fr->shift_vec[0];
124 fshift = fr->fshift[0];
125 facel = _mm_set1_pd(fr->epsfac);
126 charge = mdatoms->chargeA;
128 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
129 ewtab = fr->ic->tabq_coul_FDV0;
130 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
131 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
133 /* Setup water-specific parameters */
134 inr = nlist->iinr[0];
135 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
136 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
137 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
139 jq0 = _mm_set1_pd(charge[inr+0]);
140 jq1 = _mm_set1_pd(charge[inr+1]);
141 jq2 = _mm_set1_pd(charge[inr+2]);
142 qq00 = _mm_mul_pd(iq0,jq0);
143 qq01 = _mm_mul_pd(iq0,jq1);
144 qq02 = _mm_mul_pd(iq0,jq2);
145 qq10 = _mm_mul_pd(iq1,jq0);
146 qq11 = _mm_mul_pd(iq1,jq1);
147 qq12 = _mm_mul_pd(iq1,jq2);
148 qq20 = _mm_mul_pd(iq2,jq0);
149 qq21 = _mm_mul_pd(iq2,jq1);
150 qq22 = _mm_mul_pd(iq2,jq2);
152 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
153 rcutoff_scalar = fr->rcoulomb;
154 rcutoff = _mm_set1_pd(rcutoff_scalar);
155 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
157 rswitch_scalar = fr->rcoulomb_switch;
158 rswitch = _mm_set1_pd(rswitch_scalar);
159 /* Setup switch parameters */
160 d_scalar = rcutoff_scalar-rswitch_scalar;
161 d = _mm_set1_pd(d_scalar);
162 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
163 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
164 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
165 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
166 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
167 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
169 /* Avoid stupid compiler warnings */
177 /* Start outer loop over neighborlists */
178 for(iidx=0; iidx<nri; iidx++)
180 /* Load shift vector for this list */
181 i_shift_offset = DIM*shiftidx[iidx];
183 /* Load limits for loop over neighbors */
184 j_index_start = jindex[iidx];
185 j_index_end = jindex[iidx+1];
187 /* Get outer coordinate index */
189 i_coord_offset = DIM*inr;
191 /* Load i particle coords and add shift vector */
192 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
193 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
195 fix0 = _mm_setzero_pd();
196 fiy0 = _mm_setzero_pd();
197 fiz0 = _mm_setzero_pd();
198 fix1 = _mm_setzero_pd();
199 fiy1 = _mm_setzero_pd();
200 fiz1 = _mm_setzero_pd();
201 fix2 = _mm_setzero_pd();
202 fiy2 = _mm_setzero_pd();
203 fiz2 = _mm_setzero_pd();
205 /* Reset potential sums */
206 velecsum = _mm_setzero_pd();
208 /* Start inner kernel loop */
209 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
212 /* Get j neighbor index, and coordinate index */
215 j_coord_offsetA = DIM*jnrA;
216 j_coord_offsetB = DIM*jnrB;
218 /* load j atom coordinates */
219 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
220 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
222 /* Calculate displacement vector */
223 dx00 = _mm_sub_pd(ix0,jx0);
224 dy00 = _mm_sub_pd(iy0,jy0);
225 dz00 = _mm_sub_pd(iz0,jz0);
226 dx01 = _mm_sub_pd(ix0,jx1);
227 dy01 = _mm_sub_pd(iy0,jy1);
228 dz01 = _mm_sub_pd(iz0,jz1);
229 dx02 = _mm_sub_pd(ix0,jx2);
230 dy02 = _mm_sub_pd(iy0,jy2);
231 dz02 = _mm_sub_pd(iz0,jz2);
232 dx10 = _mm_sub_pd(ix1,jx0);
233 dy10 = _mm_sub_pd(iy1,jy0);
234 dz10 = _mm_sub_pd(iz1,jz0);
235 dx11 = _mm_sub_pd(ix1,jx1);
236 dy11 = _mm_sub_pd(iy1,jy1);
237 dz11 = _mm_sub_pd(iz1,jz1);
238 dx12 = _mm_sub_pd(ix1,jx2);
239 dy12 = _mm_sub_pd(iy1,jy2);
240 dz12 = _mm_sub_pd(iz1,jz2);
241 dx20 = _mm_sub_pd(ix2,jx0);
242 dy20 = _mm_sub_pd(iy2,jy0);
243 dz20 = _mm_sub_pd(iz2,jz0);
244 dx21 = _mm_sub_pd(ix2,jx1);
245 dy21 = _mm_sub_pd(iy2,jy1);
246 dz21 = _mm_sub_pd(iz2,jz1);
247 dx22 = _mm_sub_pd(ix2,jx2);
248 dy22 = _mm_sub_pd(iy2,jy2);
249 dz22 = _mm_sub_pd(iz2,jz2);
251 /* Calculate squared distance and things based on it */
252 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
253 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
254 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
255 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
256 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
257 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
258 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
259 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
260 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
262 rinv00 = gmx_mm_invsqrt_pd(rsq00);
263 rinv01 = gmx_mm_invsqrt_pd(rsq01);
264 rinv02 = gmx_mm_invsqrt_pd(rsq02);
265 rinv10 = gmx_mm_invsqrt_pd(rsq10);
266 rinv11 = gmx_mm_invsqrt_pd(rsq11);
267 rinv12 = gmx_mm_invsqrt_pd(rsq12);
268 rinv20 = gmx_mm_invsqrt_pd(rsq20);
269 rinv21 = gmx_mm_invsqrt_pd(rsq21);
270 rinv22 = gmx_mm_invsqrt_pd(rsq22);
272 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
273 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
274 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
275 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
276 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
277 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
278 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
279 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
280 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
282 fjx0 = _mm_setzero_pd();
283 fjy0 = _mm_setzero_pd();
284 fjz0 = _mm_setzero_pd();
285 fjx1 = _mm_setzero_pd();
286 fjy1 = _mm_setzero_pd();
287 fjz1 = _mm_setzero_pd();
288 fjx2 = _mm_setzero_pd();
289 fjy2 = _mm_setzero_pd();
290 fjz2 = _mm_setzero_pd();
292 /**************************
293 * CALCULATE INTERACTIONS *
294 **************************/
296 if (gmx_mm_any_lt(rsq00,rcutoff2))
299 r00 = _mm_mul_pd(rsq00,rinv00);
301 /* EWALD ELECTROSTATICS */
303 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
304 ewrt = _mm_mul_pd(r00,ewtabscale);
305 ewitab = _mm_cvttpd_epi32(ewrt);
307 eweps = _mm_frcz_pd(ewrt);
309 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
311 twoeweps = _mm_add_pd(eweps,eweps);
312 ewitab = _mm_slli_epi32(ewitab,2);
313 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
314 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
315 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
316 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
317 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
318 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
319 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
320 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
321 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
322 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
324 d = _mm_sub_pd(r00,rswitch);
325 d = _mm_max_pd(d,_mm_setzero_pd());
326 d2 = _mm_mul_pd(d,d);
327 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
329 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
331 /* Evaluate switch function */
332 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
333 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
334 velec = _mm_mul_pd(velec,sw);
335 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
337 /* Update potential sum for this i atom from the interaction with this j atom. */
338 velec = _mm_and_pd(velec,cutoff_mask);
339 velecsum = _mm_add_pd(velecsum,velec);
343 fscal = _mm_and_pd(fscal,cutoff_mask);
345 /* Update vectorial force */
346 fix0 = _mm_macc_pd(dx00,fscal,fix0);
347 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
348 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
350 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
351 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
352 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
356 /**************************
357 * CALCULATE INTERACTIONS *
358 **************************/
360 if (gmx_mm_any_lt(rsq01,rcutoff2))
363 r01 = _mm_mul_pd(rsq01,rinv01);
365 /* EWALD ELECTROSTATICS */
367 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
368 ewrt = _mm_mul_pd(r01,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(qq01,_mm_sub_pd(rinv01,velec));
386 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
388 d = _mm_sub_pd(r01,rswitch);
389 d = _mm_max_pd(d,_mm_setzero_pd());
390 d2 = _mm_mul_pd(d,d);
391 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
393 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
395 /* Evaluate switch function */
396 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
397 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
398 velec = _mm_mul_pd(velec,sw);
399 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
401 /* Update potential sum for this i atom from the interaction with this j atom. */
402 velec = _mm_and_pd(velec,cutoff_mask);
403 velecsum = _mm_add_pd(velecsum,velec);
407 fscal = _mm_and_pd(fscal,cutoff_mask);
409 /* Update vectorial force */
410 fix0 = _mm_macc_pd(dx01,fscal,fix0);
411 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
412 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
414 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
415 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
416 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
420 /**************************
421 * CALCULATE INTERACTIONS *
422 **************************/
424 if (gmx_mm_any_lt(rsq02,rcutoff2))
427 r02 = _mm_mul_pd(rsq02,rinv02);
429 /* EWALD ELECTROSTATICS */
431 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
432 ewrt = _mm_mul_pd(r02,ewtabscale);
433 ewitab = _mm_cvttpd_epi32(ewrt);
435 eweps = _mm_frcz_pd(ewrt);
437 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
439 twoeweps = _mm_add_pd(eweps,eweps);
440 ewitab = _mm_slli_epi32(ewitab,2);
441 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
442 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
443 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
444 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
445 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
446 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
447 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
448 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
449 velec = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
450 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
452 d = _mm_sub_pd(r02,rswitch);
453 d = _mm_max_pd(d,_mm_setzero_pd());
454 d2 = _mm_mul_pd(d,d);
455 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
457 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
459 /* Evaluate switch function */
460 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
461 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
462 velec = _mm_mul_pd(velec,sw);
463 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
465 /* Update potential sum for this i atom from the interaction with this j atom. */
466 velec = _mm_and_pd(velec,cutoff_mask);
467 velecsum = _mm_add_pd(velecsum,velec);
471 fscal = _mm_and_pd(fscal,cutoff_mask);
473 /* Update vectorial force */
474 fix0 = _mm_macc_pd(dx02,fscal,fix0);
475 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
476 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
478 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
479 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
480 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
484 /**************************
485 * CALCULATE INTERACTIONS *
486 **************************/
488 if (gmx_mm_any_lt(rsq10,rcutoff2))
491 r10 = _mm_mul_pd(rsq10,rinv10);
493 /* EWALD ELECTROSTATICS */
495 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
496 ewrt = _mm_mul_pd(r10,ewtabscale);
497 ewitab = _mm_cvttpd_epi32(ewrt);
499 eweps = _mm_frcz_pd(ewrt);
501 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
503 twoeweps = _mm_add_pd(eweps,eweps);
504 ewitab = _mm_slli_epi32(ewitab,2);
505 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
506 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
507 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
508 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
509 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
510 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
511 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
512 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
513 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
514 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
516 d = _mm_sub_pd(r10,rswitch);
517 d = _mm_max_pd(d,_mm_setzero_pd());
518 d2 = _mm_mul_pd(d,d);
519 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
521 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
523 /* Evaluate switch function */
524 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
525 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
526 velec = _mm_mul_pd(velec,sw);
527 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
529 /* Update potential sum for this i atom from the interaction with this j atom. */
530 velec = _mm_and_pd(velec,cutoff_mask);
531 velecsum = _mm_add_pd(velecsum,velec);
535 fscal = _mm_and_pd(fscal,cutoff_mask);
537 /* Update vectorial force */
538 fix1 = _mm_macc_pd(dx10,fscal,fix1);
539 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
540 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
542 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
543 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
544 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
548 /**************************
549 * CALCULATE INTERACTIONS *
550 **************************/
552 if (gmx_mm_any_lt(rsq11,rcutoff2))
555 r11 = _mm_mul_pd(rsq11,rinv11);
557 /* EWALD ELECTROSTATICS */
559 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
560 ewrt = _mm_mul_pd(r11,ewtabscale);
561 ewitab = _mm_cvttpd_epi32(ewrt);
563 eweps = _mm_frcz_pd(ewrt);
565 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
567 twoeweps = _mm_add_pd(eweps,eweps);
568 ewitab = _mm_slli_epi32(ewitab,2);
569 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
570 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
571 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
572 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
573 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
574 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
575 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
576 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
577 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
578 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
580 d = _mm_sub_pd(r11,rswitch);
581 d = _mm_max_pd(d,_mm_setzero_pd());
582 d2 = _mm_mul_pd(d,d);
583 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
585 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
587 /* Evaluate switch function */
588 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
589 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
590 velec = _mm_mul_pd(velec,sw);
591 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
593 /* Update potential sum for this i atom from the interaction with this j atom. */
594 velec = _mm_and_pd(velec,cutoff_mask);
595 velecsum = _mm_add_pd(velecsum,velec);
599 fscal = _mm_and_pd(fscal,cutoff_mask);
601 /* Update vectorial force */
602 fix1 = _mm_macc_pd(dx11,fscal,fix1);
603 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
604 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
606 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
607 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
608 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
612 /**************************
613 * CALCULATE INTERACTIONS *
614 **************************/
616 if (gmx_mm_any_lt(rsq12,rcutoff2))
619 r12 = _mm_mul_pd(rsq12,rinv12);
621 /* EWALD ELECTROSTATICS */
623 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
624 ewrt = _mm_mul_pd(r12,ewtabscale);
625 ewitab = _mm_cvttpd_epi32(ewrt);
627 eweps = _mm_frcz_pd(ewrt);
629 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
631 twoeweps = _mm_add_pd(eweps,eweps);
632 ewitab = _mm_slli_epi32(ewitab,2);
633 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
634 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
635 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
636 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
637 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
638 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
639 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
640 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
641 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
642 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
644 d = _mm_sub_pd(r12,rswitch);
645 d = _mm_max_pd(d,_mm_setzero_pd());
646 d2 = _mm_mul_pd(d,d);
647 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
649 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
651 /* Evaluate switch function */
652 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
653 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
654 velec = _mm_mul_pd(velec,sw);
655 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
657 /* Update potential sum for this i atom from the interaction with this j atom. */
658 velec = _mm_and_pd(velec,cutoff_mask);
659 velecsum = _mm_add_pd(velecsum,velec);
663 fscal = _mm_and_pd(fscal,cutoff_mask);
665 /* Update vectorial force */
666 fix1 = _mm_macc_pd(dx12,fscal,fix1);
667 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
668 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
670 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
671 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
672 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
676 /**************************
677 * CALCULATE INTERACTIONS *
678 **************************/
680 if (gmx_mm_any_lt(rsq20,rcutoff2))
683 r20 = _mm_mul_pd(rsq20,rinv20);
685 /* EWALD ELECTROSTATICS */
687 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
688 ewrt = _mm_mul_pd(r20,ewtabscale);
689 ewitab = _mm_cvttpd_epi32(ewrt);
691 eweps = _mm_frcz_pd(ewrt);
693 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
695 twoeweps = _mm_add_pd(eweps,eweps);
696 ewitab = _mm_slli_epi32(ewitab,2);
697 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
698 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
699 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
700 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
701 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
702 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
703 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
704 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
705 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
706 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
708 d = _mm_sub_pd(r20,rswitch);
709 d = _mm_max_pd(d,_mm_setzero_pd());
710 d2 = _mm_mul_pd(d,d);
711 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
713 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
715 /* Evaluate switch function */
716 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
717 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
718 velec = _mm_mul_pd(velec,sw);
719 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
721 /* Update potential sum for this i atom from the interaction with this j atom. */
722 velec = _mm_and_pd(velec,cutoff_mask);
723 velecsum = _mm_add_pd(velecsum,velec);
727 fscal = _mm_and_pd(fscal,cutoff_mask);
729 /* Update vectorial force */
730 fix2 = _mm_macc_pd(dx20,fscal,fix2);
731 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
732 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
734 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
735 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
736 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
740 /**************************
741 * CALCULATE INTERACTIONS *
742 **************************/
744 if (gmx_mm_any_lt(rsq21,rcutoff2))
747 r21 = _mm_mul_pd(rsq21,rinv21);
749 /* EWALD ELECTROSTATICS */
751 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
752 ewrt = _mm_mul_pd(r21,ewtabscale);
753 ewitab = _mm_cvttpd_epi32(ewrt);
755 eweps = _mm_frcz_pd(ewrt);
757 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
759 twoeweps = _mm_add_pd(eweps,eweps);
760 ewitab = _mm_slli_epi32(ewitab,2);
761 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
762 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
763 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
764 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
765 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
766 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
767 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
768 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
769 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
770 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
772 d = _mm_sub_pd(r21,rswitch);
773 d = _mm_max_pd(d,_mm_setzero_pd());
774 d2 = _mm_mul_pd(d,d);
775 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
777 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
779 /* Evaluate switch function */
780 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
781 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
782 velec = _mm_mul_pd(velec,sw);
783 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
785 /* Update potential sum for this i atom from the interaction with this j atom. */
786 velec = _mm_and_pd(velec,cutoff_mask);
787 velecsum = _mm_add_pd(velecsum,velec);
791 fscal = _mm_and_pd(fscal,cutoff_mask);
793 /* Update vectorial force */
794 fix2 = _mm_macc_pd(dx21,fscal,fix2);
795 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
796 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
798 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
799 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
800 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
804 /**************************
805 * CALCULATE INTERACTIONS *
806 **************************/
808 if (gmx_mm_any_lt(rsq22,rcutoff2))
811 r22 = _mm_mul_pd(rsq22,rinv22);
813 /* EWALD ELECTROSTATICS */
815 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
816 ewrt = _mm_mul_pd(r22,ewtabscale);
817 ewitab = _mm_cvttpd_epi32(ewrt);
819 eweps = _mm_frcz_pd(ewrt);
821 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
823 twoeweps = _mm_add_pd(eweps,eweps);
824 ewitab = _mm_slli_epi32(ewitab,2);
825 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
826 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
827 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
828 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
829 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
830 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
831 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
832 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
833 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
834 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
836 d = _mm_sub_pd(r22,rswitch);
837 d = _mm_max_pd(d,_mm_setzero_pd());
838 d2 = _mm_mul_pd(d,d);
839 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
841 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
843 /* Evaluate switch function */
844 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
845 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
846 velec = _mm_mul_pd(velec,sw);
847 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
849 /* Update potential sum for this i atom from the interaction with this j atom. */
850 velec = _mm_and_pd(velec,cutoff_mask);
851 velecsum = _mm_add_pd(velecsum,velec);
855 fscal = _mm_and_pd(fscal,cutoff_mask);
857 /* Update vectorial force */
858 fix2 = _mm_macc_pd(dx22,fscal,fix2);
859 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
860 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
862 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
863 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
864 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
868 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
870 /* Inner loop uses 612 flops */
877 j_coord_offsetA = DIM*jnrA;
879 /* load j atom coordinates */
880 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
881 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
883 /* Calculate displacement vector */
884 dx00 = _mm_sub_pd(ix0,jx0);
885 dy00 = _mm_sub_pd(iy0,jy0);
886 dz00 = _mm_sub_pd(iz0,jz0);
887 dx01 = _mm_sub_pd(ix0,jx1);
888 dy01 = _mm_sub_pd(iy0,jy1);
889 dz01 = _mm_sub_pd(iz0,jz1);
890 dx02 = _mm_sub_pd(ix0,jx2);
891 dy02 = _mm_sub_pd(iy0,jy2);
892 dz02 = _mm_sub_pd(iz0,jz2);
893 dx10 = _mm_sub_pd(ix1,jx0);
894 dy10 = _mm_sub_pd(iy1,jy0);
895 dz10 = _mm_sub_pd(iz1,jz0);
896 dx11 = _mm_sub_pd(ix1,jx1);
897 dy11 = _mm_sub_pd(iy1,jy1);
898 dz11 = _mm_sub_pd(iz1,jz1);
899 dx12 = _mm_sub_pd(ix1,jx2);
900 dy12 = _mm_sub_pd(iy1,jy2);
901 dz12 = _mm_sub_pd(iz1,jz2);
902 dx20 = _mm_sub_pd(ix2,jx0);
903 dy20 = _mm_sub_pd(iy2,jy0);
904 dz20 = _mm_sub_pd(iz2,jz0);
905 dx21 = _mm_sub_pd(ix2,jx1);
906 dy21 = _mm_sub_pd(iy2,jy1);
907 dz21 = _mm_sub_pd(iz2,jz1);
908 dx22 = _mm_sub_pd(ix2,jx2);
909 dy22 = _mm_sub_pd(iy2,jy2);
910 dz22 = _mm_sub_pd(iz2,jz2);
912 /* Calculate squared distance and things based on it */
913 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
914 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
915 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
916 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
917 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
918 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
919 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
920 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
921 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
923 rinv00 = gmx_mm_invsqrt_pd(rsq00);
924 rinv01 = gmx_mm_invsqrt_pd(rsq01);
925 rinv02 = gmx_mm_invsqrt_pd(rsq02);
926 rinv10 = gmx_mm_invsqrt_pd(rsq10);
927 rinv11 = gmx_mm_invsqrt_pd(rsq11);
928 rinv12 = gmx_mm_invsqrt_pd(rsq12);
929 rinv20 = gmx_mm_invsqrt_pd(rsq20);
930 rinv21 = gmx_mm_invsqrt_pd(rsq21);
931 rinv22 = gmx_mm_invsqrt_pd(rsq22);
933 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
934 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
935 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
936 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
937 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
938 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
939 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
940 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
941 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
943 fjx0 = _mm_setzero_pd();
944 fjy0 = _mm_setzero_pd();
945 fjz0 = _mm_setzero_pd();
946 fjx1 = _mm_setzero_pd();
947 fjy1 = _mm_setzero_pd();
948 fjz1 = _mm_setzero_pd();
949 fjx2 = _mm_setzero_pd();
950 fjy2 = _mm_setzero_pd();
951 fjz2 = _mm_setzero_pd();
953 /**************************
954 * CALCULATE INTERACTIONS *
955 **************************/
957 if (gmx_mm_any_lt(rsq00,rcutoff2))
960 r00 = _mm_mul_pd(rsq00,rinv00);
962 /* EWALD ELECTROSTATICS */
964 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
965 ewrt = _mm_mul_pd(r00,ewtabscale);
966 ewitab = _mm_cvttpd_epi32(ewrt);
968 eweps = _mm_frcz_pd(ewrt);
970 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
972 twoeweps = _mm_add_pd(eweps,eweps);
973 ewitab = _mm_slli_epi32(ewitab,2);
974 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
975 ewtabD = _mm_setzero_pd();
976 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
977 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
978 ewtabFn = _mm_setzero_pd();
979 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
980 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
981 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
982 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
983 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
985 d = _mm_sub_pd(r00,rswitch);
986 d = _mm_max_pd(d,_mm_setzero_pd());
987 d2 = _mm_mul_pd(d,d);
988 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
990 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
992 /* Evaluate switch function */
993 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
994 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
995 velec = _mm_mul_pd(velec,sw);
996 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
998 /* Update potential sum for this i atom from the interaction with this j atom. */
999 velec = _mm_and_pd(velec,cutoff_mask);
1000 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1001 velecsum = _mm_add_pd(velecsum,velec);
1005 fscal = _mm_and_pd(fscal,cutoff_mask);
1007 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1009 /* Update vectorial force */
1010 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1011 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1012 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1014 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1015 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1016 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1020 /**************************
1021 * CALCULATE INTERACTIONS *
1022 **************************/
1024 if (gmx_mm_any_lt(rsq01,rcutoff2))
1027 r01 = _mm_mul_pd(rsq01,rinv01);
1029 /* EWALD ELECTROSTATICS */
1031 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1032 ewrt = _mm_mul_pd(r01,ewtabscale);
1033 ewitab = _mm_cvttpd_epi32(ewrt);
1035 eweps = _mm_frcz_pd(ewrt);
1037 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1039 twoeweps = _mm_add_pd(eweps,eweps);
1040 ewitab = _mm_slli_epi32(ewitab,2);
1041 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1042 ewtabD = _mm_setzero_pd();
1043 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1044 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1045 ewtabFn = _mm_setzero_pd();
1046 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1047 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1048 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1049 velec = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
1050 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1052 d = _mm_sub_pd(r01,rswitch);
1053 d = _mm_max_pd(d,_mm_setzero_pd());
1054 d2 = _mm_mul_pd(d,d);
1055 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1057 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1059 /* Evaluate switch function */
1060 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1061 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
1062 velec = _mm_mul_pd(velec,sw);
1063 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
1065 /* Update potential sum for this i atom from the interaction with this j atom. */
1066 velec = _mm_and_pd(velec,cutoff_mask);
1067 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1068 velecsum = _mm_add_pd(velecsum,velec);
1072 fscal = _mm_and_pd(fscal,cutoff_mask);
1074 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1076 /* Update vectorial force */
1077 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1078 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1079 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1081 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1082 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1083 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1087 /**************************
1088 * CALCULATE INTERACTIONS *
1089 **************************/
1091 if (gmx_mm_any_lt(rsq02,rcutoff2))
1094 r02 = _mm_mul_pd(rsq02,rinv02);
1096 /* EWALD ELECTROSTATICS */
1098 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1099 ewrt = _mm_mul_pd(r02,ewtabscale);
1100 ewitab = _mm_cvttpd_epi32(ewrt);
1102 eweps = _mm_frcz_pd(ewrt);
1104 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1106 twoeweps = _mm_add_pd(eweps,eweps);
1107 ewitab = _mm_slli_epi32(ewitab,2);
1108 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1109 ewtabD = _mm_setzero_pd();
1110 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1111 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1112 ewtabFn = _mm_setzero_pd();
1113 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1114 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1115 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1116 velec = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
1117 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1119 d = _mm_sub_pd(r02,rswitch);
1120 d = _mm_max_pd(d,_mm_setzero_pd());
1121 d2 = _mm_mul_pd(d,d);
1122 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1124 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1126 /* Evaluate switch function */
1127 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1128 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
1129 velec = _mm_mul_pd(velec,sw);
1130 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1132 /* Update potential sum for this i atom from the interaction with this j atom. */
1133 velec = _mm_and_pd(velec,cutoff_mask);
1134 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1135 velecsum = _mm_add_pd(velecsum,velec);
1139 fscal = _mm_and_pd(fscal,cutoff_mask);
1141 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1143 /* Update vectorial force */
1144 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1145 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1146 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1148 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1149 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1150 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1154 /**************************
1155 * CALCULATE INTERACTIONS *
1156 **************************/
1158 if (gmx_mm_any_lt(rsq10,rcutoff2))
1161 r10 = _mm_mul_pd(rsq10,rinv10);
1163 /* EWALD ELECTROSTATICS */
1165 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1166 ewrt = _mm_mul_pd(r10,ewtabscale);
1167 ewitab = _mm_cvttpd_epi32(ewrt);
1169 eweps = _mm_frcz_pd(ewrt);
1171 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1173 twoeweps = _mm_add_pd(eweps,eweps);
1174 ewitab = _mm_slli_epi32(ewitab,2);
1175 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1176 ewtabD = _mm_setzero_pd();
1177 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1178 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1179 ewtabFn = _mm_setzero_pd();
1180 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1181 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1182 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1183 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1184 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1186 d = _mm_sub_pd(r10,rswitch);
1187 d = _mm_max_pd(d,_mm_setzero_pd());
1188 d2 = _mm_mul_pd(d,d);
1189 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1191 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1193 /* Evaluate switch function */
1194 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1195 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1196 velec = _mm_mul_pd(velec,sw);
1197 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1199 /* Update potential sum for this i atom from the interaction with this j atom. */
1200 velec = _mm_and_pd(velec,cutoff_mask);
1201 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1202 velecsum = _mm_add_pd(velecsum,velec);
1206 fscal = _mm_and_pd(fscal,cutoff_mask);
1208 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1210 /* Update vectorial force */
1211 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1212 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1213 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1215 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1216 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1217 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1221 /**************************
1222 * CALCULATE INTERACTIONS *
1223 **************************/
1225 if (gmx_mm_any_lt(rsq11,rcutoff2))
1228 r11 = _mm_mul_pd(rsq11,rinv11);
1230 /* EWALD ELECTROSTATICS */
1232 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1233 ewrt = _mm_mul_pd(r11,ewtabscale);
1234 ewitab = _mm_cvttpd_epi32(ewrt);
1236 eweps = _mm_frcz_pd(ewrt);
1238 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1240 twoeweps = _mm_add_pd(eweps,eweps);
1241 ewitab = _mm_slli_epi32(ewitab,2);
1242 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1243 ewtabD = _mm_setzero_pd();
1244 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1245 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1246 ewtabFn = _mm_setzero_pd();
1247 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1248 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1249 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1250 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
1251 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1253 d = _mm_sub_pd(r11,rswitch);
1254 d = _mm_max_pd(d,_mm_setzero_pd());
1255 d2 = _mm_mul_pd(d,d);
1256 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1258 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1260 /* Evaluate switch function */
1261 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1262 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
1263 velec = _mm_mul_pd(velec,sw);
1264 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1266 /* Update potential sum for this i atom from the interaction with this j atom. */
1267 velec = _mm_and_pd(velec,cutoff_mask);
1268 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1269 velecsum = _mm_add_pd(velecsum,velec);
1273 fscal = _mm_and_pd(fscal,cutoff_mask);
1275 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1277 /* Update vectorial force */
1278 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1279 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1280 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1282 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1283 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1284 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1288 /**************************
1289 * CALCULATE INTERACTIONS *
1290 **************************/
1292 if (gmx_mm_any_lt(rsq12,rcutoff2))
1295 r12 = _mm_mul_pd(rsq12,rinv12);
1297 /* EWALD ELECTROSTATICS */
1299 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1300 ewrt = _mm_mul_pd(r12,ewtabscale);
1301 ewitab = _mm_cvttpd_epi32(ewrt);
1303 eweps = _mm_frcz_pd(ewrt);
1305 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1307 twoeweps = _mm_add_pd(eweps,eweps);
1308 ewitab = _mm_slli_epi32(ewitab,2);
1309 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1310 ewtabD = _mm_setzero_pd();
1311 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1312 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1313 ewtabFn = _mm_setzero_pd();
1314 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1315 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1316 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1317 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1318 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1320 d = _mm_sub_pd(r12,rswitch);
1321 d = _mm_max_pd(d,_mm_setzero_pd());
1322 d2 = _mm_mul_pd(d,d);
1323 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1325 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1327 /* Evaluate switch function */
1328 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1329 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
1330 velec = _mm_mul_pd(velec,sw);
1331 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1333 /* Update potential sum for this i atom from the interaction with this j atom. */
1334 velec = _mm_and_pd(velec,cutoff_mask);
1335 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1336 velecsum = _mm_add_pd(velecsum,velec);
1340 fscal = _mm_and_pd(fscal,cutoff_mask);
1342 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1344 /* Update vectorial force */
1345 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1346 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1347 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1349 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1350 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1351 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1355 /**************************
1356 * CALCULATE INTERACTIONS *
1357 **************************/
1359 if (gmx_mm_any_lt(rsq20,rcutoff2))
1362 r20 = _mm_mul_pd(rsq20,rinv20);
1364 /* EWALD ELECTROSTATICS */
1366 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1367 ewrt = _mm_mul_pd(r20,ewtabscale);
1368 ewitab = _mm_cvttpd_epi32(ewrt);
1370 eweps = _mm_frcz_pd(ewrt);
1372 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1374 twoeweps = _mm_add_pd(eweps,eweps);
1375 ewitab = _mm_slli_epi32(ewitab,2);
1376 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1377 ewtabD = _mm_setzero_pd();
1378 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1379 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1380 ewtabFn = _mm_setzero_pd();
1381 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1382 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1383 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1384 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1385 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1387 d = _mm_sub_pd(r20,rswitch);
1388 d = _mm_max_pd(d,_mm_setzero_pd());
1389 d2 = _mm_mul_pd(d,d);
1390 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1392 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1394 /* Evaluate switch function */
1395 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1396 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1397 velec = _mm_mul_pd(velec,sw);
1398 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1400 /* Update potential sum for this i atom from the interaction with this j atom. */
1401 velec = _mm_and_pd(velec,cutoff_mask);
1402 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1403 velecsum = _mm_add_pd(velecsum,velec);
1407 fscal = _mm_and_pd(fscal,cutoff_mask);
1409 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1411 /* Update vectorial force */
1412 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1413 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1414 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1416 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1417 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1418 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1422 /**************************
1423 * CALCULATE INTERACTIONS *
1424 **************************/
1426 if (gmx_mm_any_lt(rsq21,rcutoff2))
1429 r21 = _mm_mul_pd(rsq21,rinv21);
1431 /* EWALD ELECTROSTATICS */
1433 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1434 ewrt = _mm_mul_pd(r21,ewtabscale);
1435 ewitab = _mm_cvttpd_epi32(ewrt);
1437 eweps = _mm_frcz_pd(ewrt);
1439 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1441 twoeweps = _mm_add_pd(eweps,eweps);
1442 ewitab = _mm_slli_epi32(ewitab,2);
1443 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1444 ewtabD = _mm_setzero_pd();
1445 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1446 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1447 ewtabFn = _mm_setzero_pd();
1448 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1449 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1450 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1451 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1452 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1454 d = _mm_sub_pd(r21,rswitch);
1455 d = _mm_max_pd(d,_mm_setzero_pd());
1456 d2 = _mm_mul_pd(d,d);
1457 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1459 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1461 /* Evaluate switch function */
1462 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1463 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
1464 velec = _mm_mul_pd(velec,sw);
1465 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1467 /* Update potential sum for this i atom from the interaction with this j atom. */
1468 velec = _mm_and_pd(velec,cutoff_mask);
1469 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1470 velecsum = _mm_add_pd(velecsum,velec);
1474 fscal = _mm_and_pd(fscal,cutoff_mask);
1476 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1478 /* Update vectorial force */
1479 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1480 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1481 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1483 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1484 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1485 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1489 /**************************
1490 * CALCULATE INTERACTIONS *
1491 **************************/
1493 if (gmx_mm_any_lt(rsq22,rcutoff2))
1496 r22 = _mm_mul_pd(rsq22,rinv22);
1498 /* EWALD ELECTROSTATICS */
1500 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1501 ewrt = _mm_mul_pd(r22,ewtabscale);
1502 ewitab = _mm_cvttpd_epi32(ewrt);
1504 eweps = _mm_frcz_pd(ewrt);
1506 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1508 twoeweps = _mm_add_pd(eweps,eweps);
1509 ewitab = _mm_slli_epi32(ewitab,2);
1510 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1511 ewtabD = _mm_setzero_pd();
1512 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1513 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1514 ewtabFn = _mm_setzero_pd();
1515 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1516 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1517 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1518 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1519 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1521 d = _mm_sub_pd(r22,rswitch);
1522 d = _mm_max_pd(d,_mm_setzero_pd());
1523 d2 = _mm_mul_pd(d,d);
1524 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1526 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1528 /* Evaluate switch function */
1529 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1530 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
1531 velec = _mm_mul_pd(velec,sw);
1532 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1534 /* Update potential sum for this i atom from the interaction with this j atom. */
1535 velec = _mm_and_pd(velec,cutoff_mask);
1536 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1537 velecsum = _mm_add_pd(velecsum,velec);
1541 fscal = _mm_and_pd(fscal,cutoff_mask);
1543 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1545 /* Update vectorial force */
1546 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1547 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1548 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1550 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1551 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1552 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1556 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1558 /* Inner loop uses 612 flops */
1561 /* End of innermost loop */
1563 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1564 f+i_coord_offset,fshift+i_shift_offset);
1567 /* Update potential energies */
1568 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1570 /* Increment number of inner iterations */
1571 inneriter += j_index_end - j_index_start;
1573 /* Outer loop uses 19 flops */
1576 /* Increment number of outer iterations */
1579 /* Update outer/inner flops */
1581 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*612);
1584 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_avx_128_fma_double
1585 * Electrostatics interaction: Ewald
1586 * VdW interaction: None
1587 * Geometry: Water3-Water3
1588 * Calculate force/pot: Force
1591 nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_avx_128_fma_double
1592 (t_nblist * gmx_restrict nlist,
1593 rvec * gmx_restrict xx,
1594 rvec * gmx_restrict ff,
1595 t_forcerec * gmx_restrict fr,
1596 t_mdatoms * gmx_restrict mdatoms,
1597 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1598 t_nrnb * gmx_restrict nrnb)
1600 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1601 * just 0 for non-waters.
1602 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1603 * jnr indices corresponding to data put in the four positions in the SIMD register.
1605 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1606 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1608 int j_coord_offsetA,j_coord_offsetB;
1609 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1610 real rcutoff_scalar;
1611 real *shiftvec,*fshift,*x,*f;
1612 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1614 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1616 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1618 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1619 int vdwjidx0A,vdwjidx0B;
1620 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1621 int vdwjidx1A,vdwjidx1B;
1622 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1623 int vdwjidx2A,vdwjidx2B;
1624 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1625 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1626 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1627 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1628 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1629 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1630 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1631 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1632 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1633 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1634 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1637 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1639 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1640 real rswitch_scalar,d_scalar;
1641 __m128d dummy_mask,cutoff_mask;
1642 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1643 __m128d one = _mm_set1_pd(1.0);
1644 __m128d two = _mm_set1_pd(2.0);
1650 jindex = nlist->jindex;
1652 shiftidx = nlist->shift;
1654 shiftvec = fr->shift_vec[0];
1655 fshift = fr->fshift[0];
1656 facel = _mm_set1_pd(fr->epsfac);
1657 charge = mdatoms->chargeA;
1659 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1660 ewtab = fr->ic->tabq_coul_FDV0;
1661 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1662 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1664 /* Setup water-specific parameters */
1665 inr = nlist->iinr[0];
1666 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1667 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1668 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1670 jq0 = _mm_set1_pd(charge[inr+0]);
1671 jq1 = _mm_set1_pd(charge[inr+1]);
1672 jq2 = _mm_set1_pd(charge[inr+2]);
1673 qq00 = _mm_mul_pd(iq0,jq0);
1674 qq01 = _mm_mul_pd(iq0,jq1);
1675 qq02 = _mm_mul_pd(iq0,jq2);
1676 qq10 = _mm_mul_pd(iq1,jq0);
1677 qq11 = _mm_mul_pd(iq1,jq1);
1678 qq12 = _mm_mul_pd(iq1,jq2);
1679 qq20 = _mm_mul_pd(iq2,jq0);
1680 qq21 = _mm_mul_pd(iq2,jq1);
1681 qq22 = _mm_mul_pd(iq2,jq2);
1683 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1684 rcutoff_scalar = fr->rcoulomb;
1685 rcutoff = _mm_set1_pd(rcutoff_scalar);
1686 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
1688 rswitch_scalar = fr->rcoulomb_switch;
1689 rswitch = _mm_set1_pd(rswitch_scalar);
1690 /* Setup switch parameters */
1691 d_scalar = rcutoff_scalar-rswitch_scalar;
1692 d = _mm_set1_pd(d_scalar);
1693 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
1694 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1695 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1696 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
1697 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1698 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1700 /* Avoid stupid compiler warnings */
1702 j_coord_offsetA = 0;
1703 j_coord_offsetB = 0;
1708 /* Start outer loop over neighborlists */
1709 for(iidx=0; iidx<nri; iidx++)
1711 /* Load shift vector for this list */
1712 i_shift_offset = DIM*shiftidx[iidx];
1714 /* Load limits for loop over neighbors */
1715 j_index_start = jindex[iidx];
1716 j_index_end = jindex[iidx+1];
1718 /* Get outer coordinate index */
1720 i_coord_offset = DIM*inr;
1722 /* Load i particle coords and add shift vector */
1723 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1724 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1726 fix0 = _mm_setzero_pd();
1727 fiy0 = _mm_setzero_pd();
1728 fiz0 = _mm_setzero_pd();
1729 fix1 = _mm_setzero_pd();
1730 fiy1 = _mm_setzero_pd();
1731 fiz1 = _mm_setzero_pd();
1732 fix2 = _mm_setzero_pd();
1733 fiy2 = _mm_setzero_pd();
1734 fiz2 = _mm_setzero_pd();
1736 /* Start inner kernel loop */
1737 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1740 /* Get j neighbor index, and coordinate index */
1742 jnrB = jjnr[jidx+1];
1743 j_coord_offsetA = DIM*jnrA;
1744 j_coord_offsetB = DIM*jnrB;
1746 /* load j atom coordinates */
1747 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1748 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1750 /* Calculate displacement vector */
1751 dx00 = _mm_sub_pd(ix0,jx0);
1752 dy00 = _mm_sub_pd(iy0,jy0);
1753 dz00 = _mm_sub_pd(iz0,jz0);
1754 dx01 = _mm_sub_pd(ix0,jx1);
1755 dy01 = _mm_sub_pd(iy0,jy1);
1756 dz01 = _mm_sub_pd(iz0,jz1);
1757 dx02 = _mm_sub_pd(ix0,jx2);
1758 dy02 = _mm_sub_pd(iy0,jy2);
1759 dz02 = _mm_sub_pd(iz0,jz2);
1760 dx10 = _mm_sub_pd(ix1,jx0);
1761 dy10 = _mm_sub_pd(iy1,jy0);
1762 dz10 = _mm_sub_pd(iz1,jz0);
1763 dx11 = _mm_sub_pd(ix1,jx1);
1764 dy11 = _mm_sub_pd(iy1,jy1);
1765 dz11 = _mm_sub_pd(iz1,jz1);
1766 dx12 = _mm_sub_pd(ix1,jx2);
1767 dy12 = _mm_sub_pd(iy1,jy2);
1768 dz12 = _mm_sub_pd(iz1,jz2);
1769 dx20 = _mm_sub_pd(ix2,jx0);
1770 dy20 = _mm_sub_pd(iy2,jy0);
1771 dz20 = _mm_sub_pd(iz2,jz0);
1772 dx21 = _mm_sub_pd(ix2,jx1);
1773 dy21 = _mm_sub_pd(iy2,jy1);
1774 dz21 = _mm_sub_pd(iz2,jz1);
1775 dx22 = _mm_sub_pd(ix2,jx2);
1776 dy22 = _mm_sub_pd(iy2,jy2);
1777 dz22 = _mm_sub_pd(iz2,jz2);
1779 /* Calculate squared distance and things based on it */
1780 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1781 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1782 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1783 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1784 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1785 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1786 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1787 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1788 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1790 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1791 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1792 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1793 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1794 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1795 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1796 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1797 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1798 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1800 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1801 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
1802 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
1803 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1804 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1805 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1806 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1807 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1808 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1810 fjx0 = _mm_setzero_pd();
1811 fjy0 = _mm_setzero_pd();
1812 fjz0 = _mm_setzero_pd();
1813 fjx1 = _mm_setzero_pd();
1814 fjy1 = _mm_setzero_pd();
1815 fjz1 = _mm_setzero_pd();
1816 fjx2 = _mm_setzero_pd();
1817 fjy2 = _mm_setzero_pd();
1818 fjz2 = _mm_setzero_pd();
1820 /**************************
1821 * CALCULATE INTERACTIONS *
1822 **************************/
1824 if (gmx_mm_any_lt(rsq00,rcutoff2))
1827 r00 = _mm_mul_pd(rsq00,rinv00);
1829 /* EWALD ELECTROSTATICS */
1831 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1832 ewrt = _mm_mul_pd(r00,ewtabscale);
1833 ewitab = _mm_cvttpd_epi32(ewrt);
1835 eweps = _mm_frcz_pd(ewrt);
1837 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1839 twoeweps = _mm_add_pd(eweps,eweps);
1840 ewitab = _mm_slli_epi32(ewitab,2);
1841 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1842 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1843 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1844 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1845 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1846 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1847 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1848 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1849 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1850 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1852 d = _mm_sub_pd(r00,rswitch);
1853 d = _mm_max_pd(d,_mm_setzero_pd());
1854 d2 = _mm_mul_pd(d,d);
1855 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1857 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1859 /* Evaluate switch function */
1860 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1861 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1862 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1866 fscal = _mm_and_pd(fscal,cutoff_mask);
1868 /* Update vectorial force */
1869 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1870 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1871 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1873 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1874 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1875 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1879 /**************************
1880 * CALCULATE INTERACTIONS *
1881 **************************/
1883 if (gmx_mm_any_lt(rsq01,rcutoff2))
1886 r01 = _mm_mul_pd(rsq01,rinv01);
1888 /* EWALD ELECTROSTATICS */
1890 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1891 ewrt = _mm_mul_pd(r01,ewtabscale);
1892 ewitab = _mm_cvttpd_epi32(ewrt);
1894 eweps = _mm_frcz_pd(ewrt);
1896 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1898 twoeweps = _mm_add_pd(eweps,eweps);
1899 ewitab = _mm_slli_epi32(ewitab,2);
1900 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1901 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1902 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1903 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1904 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1905 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1906 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1907 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1908 velec = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
1909 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1911 d = _mm_sub_pd(r01,rswitch);
1912 d = _mm_max_pd(d,_mm_setzero_pd());
1913 d2 = _mm_mul_pd(d,d);
1914 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1916 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1918 /* Evaluate switch function */
1919 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1920 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
1921 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
1925 fscal = _mm_and_pd(fscal,cutoff_mask);
1927 /* Update vectorial force */
1928 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1929 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1930 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1932 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1933 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1934 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1938 /**************************
1939 * CALCULATE INTERACTIONS *
1940 **************************/
1942 if (gmx_mm_any_lt(rsq02,rcutoff2))
1945 r02 = _mm_mul_pd(rsq02,rinv02);
1947 /* EWALD ELECTROSTATICS */
1949 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1950 ewrt = _mm_mul_pd(r02,ewtabscale);
1951 ewitab = _mm_cvttpd_epi32(ewrt);
1953 eweps = _mm_frcz_pd(ewrt);
1955 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1957 twoeweps = _mm_add_pd(eweps,eweps);
1958 ewitab = _mm_slli_epi32(ewitab,2);
1959 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1960 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1961 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1962 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1963 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1964 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1965 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1966 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1967 velec = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
1968 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1970 d = _mm_sub_pd(r02,rswitch);
1971 d = _mm_max_pd(d,_mm_setzero_pd());
1972 d2 = _mm_mul_pd(d,d);
1973 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1975 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1977 /* Evaluate switch function */
1978 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1979 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
1980 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
1984 fscal = _mm_and_pd(fscal,cutoff_mask);
1986 /* Update vectorial force */
1987 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1988 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1989 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1991 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1992 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1993 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1997 /**************************
1998 * CALCULATE INTERACTIONS *
1999 **************************/
2001 if (gmx_mm_any_lt(rsq10,rcutoff2))
2004 r10 = _mm_mul_pd(rsq10,rinv10);
2006 /* EWALD ELECTROSTATICS */
2008 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2009 ewrt = _mm_mul_pd(r10,ewtabscale);
2010 ewitab = _mm_cvttpd_epi32(ewrt);
2012 eweps = _mm_frcz_pd(ewrt);
2014 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2016 twoeweps = _mm_add_pd(eweps,eweps);
2017 ewitab = _mm_slli_epi32(ewitab,2);
2018 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2019 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2020 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2021 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2022 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2023 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2024 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2025 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2026 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
2027 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2029 d = _mm_sub_pd(r10,rswitch);
2030 d = _mm_max_pd(d,_mm_setzero_pd());
2031 d2 = _mm_mul_pd(d,d);
2032 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2034 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2036 /* Evaluate switch function */
2037 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2038 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
2039 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
2043 fscal = _mm_and_pd(fscal,cutoff_mask);
2045 /* Update vectorial force */
2046 fix1 = _mm_macc_pd(dx10,fscal,fix1);
2047 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
2048 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
2050 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
2051 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
2052 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
2056 /**************************
2057 * CALCULATE INTERACTIONS *
2058 **************************/
2060 if (gmx_mm_any_lt(rsq11,rcutoff2))
2063 r11 = _mm_mul_pd(rsq11,rinv11);
2065 /* EWALD ELECTROSTATICS */
2067 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2068 ewrt = _mm_mul_pd(r11,ewtabscale);
2069 ewitab = _mm_cvttpd_epi32(ewrt);
2071 eweps = _mm_frcz_pd(ewrt);
2073 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2075 twoeweps = _mm_add_pd(eweps,eweps);
2076 ewitab = _mm_slli_epi32(ewitab,2);
2077 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2078 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2079 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2080 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2081 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2082 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2083 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2084 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2085 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2086 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2088 d = _mm_sub_pd(r11,rswitch);
2089 d = _mm_max_pd(d,_mm_setzero_pd());
2090 d2 = _mm_mul_pd(d,d);
2091 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2093 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2095 /* Evaluate switch function */
2096 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2097 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2098 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2102 fscal = _mm_and_pd(fscal,cutoff_mask);
2104 /* Update vectorial force */
2105 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2106 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2107 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2109 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2110 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2111 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2115 /**************************
2116 * CALCULATE INTERACTIONS *
2117 **************************/
2119 if (gmx_mm_any_lt(rsq12,rcutoff2))
2122 r12 = _mm_mul_pd(rsq12,rinv12);
2124 /* EWALD ELECTROSTATICS */
2126 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2127 ewrt = _mm_mul_pd(r12,ewtabscale);
2128 ewitab = _mm_cvttpd_epi32(ewrt);
2130 eweps = _mm_frcz_pd(ewrt);
2132 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2134 twoeweps = _mm_add_pd(eweps,eweps);
2135 ewitab = _mm_slli_epi32(ewitab,2);
2136 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2137 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2138 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2139 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2140 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2141 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2142 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2143 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2144 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2145 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2147 d = _mm_sub_pd(r12,rswitch);
2148 d = _mm_max_pd(d,_mm_setzero_pd());
2149 d2 = _mm_mul_pd(d,d);
2150 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2152 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2154 /* Evaluate switch function */
2155 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2156 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2157 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2161 fscal = _mm_and_pd(fscal,cutoff_mask);
2163 /* Update vectorial force */
2164 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2165 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2166 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2168 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2169 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2170 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2174 /**************************
2175 * CALCULATE INTERACTIONS *
2176 **************************/
2178 if (gmx_mm_any_lt(rsq20,rcutoff2))
2181 r20 = _mm_mul_pd(rsq20,rinv20);
2183 /* EWALD ELECTROSTATICS */
2185 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2186 ewrt = _mm_mul_pd(r20,ewtabscale);
2187 ewitab = _mm_cvttpd_epi32(ewrt);
2189 eweps = _mm_frcz_pd(ewrt);
2191 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2193 twoeweps = _mm_add_pd(eweps,eweps);
2194 ewitab = _mm_slli_epi32(ewitab,2);
2195 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2196 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2197 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2198 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2199 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2200 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2201 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2202 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2203 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
2204 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2206 d = _mm_sub_pd(r20,rswitch);
2207 d = _mm_max_pd(d,_mm_setzero_pd());
2208 d2 = _mm_mul_pd(d,d);
2209 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2211 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2213 /* Evaluate switch function */
2214 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2215 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
2216 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
2220 fscal = _mm_and_pd(fscal,cutoff_mask);
2222 /* Update vectorial force */
2223 fix2 = _mm_macc_pd(dx20,fscal,fix2);
2224 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
2225 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
2227 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
2228 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
2229 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
2233 /**************************
2234 * CALCULATE INTERACTIONS *
2235 **************************/
2237 if (gmx_mm_any_lt(rsq21,rcutoff2))
2240 r21 = _mm_mul_pd(rsq21,rinv21);
2242 /* EWALD ELECTROSTATICS */
2244 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2245 ewrt = _mm_mul_pd(r21,ewtabscale);
2246 ewitab = _mm_cvttpd_epi32(ewrt);
2248 eweps = _mm_frcz_pd(ewrt);
2250 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2252 twoeweps = _mm_add_pd(eweps,eweps);
2253 ewitab = _mm_slli_epi32(ewitab,2);
2254 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2255 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2256 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2257 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2258 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2259 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2260 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2261 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2262 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2263 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2265 d = _mm_sub_pd(r21,rswitch);
2266 d = _mm_max_pd(d,_mm_setzero_pd());
2267 d2 = _mm_mul_pd(d,d);
2268 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2270 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2272 /* Evaluate switch function */
2273 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2274 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2275 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2279 fscal = _mm_and_pd(fscal,cutoff_mask);
2281 /* Update vectorial force */
2282 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2283 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2284 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2286 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2287 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2288 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2292 /**************************
2293 * CALCULATE INTERACTIONS *
2294 **************************/
2296 if (gmx_mm_any_lt(rsq22,rcutoff2))
2299 r22 = _mm_mul_pd(rsq22,rinv22);
2301 /* EWALD ELECTROSTATICS */
2303 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2304 ewrt = _mm_mul_pd(r22,ewtabscale);
2305 ewitab = _mm_cvttpd_epi32(ewrt);
2307 eweps = _mm_frcz_pd(ewrt);
2309 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2311 twoeweps = _mm_add_pd(eweps,eweps);
2312 ewitab = _mm_slli_epi32(ewitab,2);
2313 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2314 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2315 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2316 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2317 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2318 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2319 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2320 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2321 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2322 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2324 d = _mm_sub_pd(r22,rswitch);
2325 d = _mm_max_pd(d,_mm_setzero_pd());
2326 d2 = _mm_mul_pd(d,d);
2327 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2329 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2331 /* Evaluate switch function */
2332 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2333 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2334 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2338 fscal = _mm_and_pd(fscal,cutoff_mask);
2340 /* Update vectorial force */
2341 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2342 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2343 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2345 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2346 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2347 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2351 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2353 /* Inner loop uses 585 flops */
2356 if(jidx<j_index_end)
2360 j_coord_offsetA = DIM*jnrA;
2362 /* load j atom coordinates */
2363 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2364 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2366 /* Calculate displacement vector */
2367 dx00 = _mm_sub_pd(ix0,jx0);
2368 dy00 = _mm_sub_pd(iy0,jy0);
2369 dz00 = _mm_sub_pd(iz0,jz0);
2370 dx01 = _mm_sub_pd(ix0,jx1);
2371 dy01 = _mm_sub_pd(iy0,jy1);
2372 dz01 = _mm_sub_pd(iz0,jz1);
2373 dx02 = _mm_sub_pd(ix0,jx2);
2374 dy02 = _mm_sub_pd(iy0,jy2);
2375 dz02 = _mm_sub_pd(iz0,jz2);
2376 dx10 = _mm_sub_pd(ix1,jx0);
2377 dy10 = _mm_sub_pd(iy1,jy0);
2378 dz10 = _mm_sub_pd(iz1,jz0);
2379 dx11 = _mm_sub_pd(ix1,jx1);
2380 dy11 = _mm_sub_pd(iy1,jy1);
2381 dz11 = _mm_sub_pd(iz1,jz1);
2382 dx12 = _mm_sub_pd(ix1,jx2);
2383 dy12 = _mm_sub_pd(iy1,jy2);
2384 dz12 = _mm_sub_pd(iz1,jz2);
2385 dx20 = _mm_sub_pd(ix2,jx0);
2386 dy20 = _mm_sub_pd(iy2,jy0);
2387 dz20 = _mm_sub_pd(iz2,jz0);
2388 dx21 = _mm_sub_pd(ix2,jx1);
2389 dy21 = _mm_sub_pd(iy2,jy1);
2390 dz21 = _mm_sub_pd(iz2,jz1);
2391 dx22 = _mm_sub_pd(ix2,jx2);
2392 dy22 = _mm_sub_pd(iy2,jy2);
2393 dz22 = _mm_sub_pd(iz2,jz2);
2395 /* Calculate squared distance and things based on it */
2396 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2397 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2398 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2399 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2400 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2401 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2402 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2403 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2404 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2406 rinv00 = gmx_mm_invsqrt_pd(rsq00);
2407 rinv01 = gmx_mm_invsqrt_pd(rsq01);
2408 rinv02 = gmx_mm_invsqrt_pd(rsq02);
2409 rinv10 = gmx_mm_invsqrt_pd(rsq10);
2410 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2411 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2412 rinv20 = gmx_mm_invsqrt_pd(rsq20);
2413 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2414 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2416 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
2417 rinvsq01 = _mm_mul_pd(rinv01,rinv01);
2418 rinvsq02 = _mm_mul_pd(rinv02,rinv02);
2419 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
2420 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
2421 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
2422 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
2423 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
2424 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
2426 fjx0 = _mm_setzero_pd();
2427 fjy0 = _mm_setzero_pd();
2428 fjz0 = _mm_setzero_pd();
2429 fjx1 = _mm_setzero_pd();
2430 fjy1 = _mm_setzero_pd();
2431 fjz1 = _mm_setzero_pd();
2432 fjx2 = _mm_setzero_pd();
2433 fjy2 = _mm_setzero_pd();
2434 fjz2 = _mm_setzero_pd();
2436 /**************************
2437 * CALCULATE INTERACTIONS *
2438 **************************/
2440 if (gmx_mm_any_lt(rsq00,rcutoff2))
2443 r00 = _mm_mul_pd(rsq00,rinv00);
2445 /* EWALD ELECTROSTATICS */
2447 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2448 ewrt = _mm_mul_pd(r00,ewtabscale);
2449 ewitab = _mm_cvttpd_epi32(ewrt);
2451 eweps = _mm_frcz_pd(ewrt);
2453 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2455 twoeweps = _mm_add_pd(eweps,eweps);
2456 ewitab = _mm_slli_epi32(ewitab,2);
2457 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2458 ewtabD = _mm_setzero_pd();
2459 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2460 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2461 ewtabFn = _mm_setzero_pd();
2462 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2463 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2464 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2465 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
2466 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2468 d = _mm_sub_pd(r00,rswitch);
2469 d = _mm_max_pd(d,_mm_setzero_pd());
2470 d2 = _mm_mul_pd(d,d);
2471 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2473 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2475 /* Evaluate switch function */
2476 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2477 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
2478 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
2482 fscal = _mm_and_pd(fscal,cutoff_mask);
2484 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2486 /* Update vectorial force */
2487 fix0 = _mm_macc_pd(dx00,fscal,fix0);
2488 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
2489 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
2491 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
2492 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
2493 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
2497 /**************************
2498 * CALCULATE INTERACTIONS *
2499 **************************/
2501 if (gmx_mm_any_lt(rsq01,rcutoff2))
2504 r01 = _mm_mul_pd(rsq01,rinv01);
2506 /* EWALD ELECTROSTATICS */
2508 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2509 ewrt = _mm_mul_pd(r01,ewtabscale);
2510 ewitab = _mm_cvttpd_epi32(ewrt);
2512 eweps = _mm_frcz_pd(ewrt);
2514 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2516 twoeweps = _mm_add_pd(eweps,eweps);
2517 ewitab = _mm_slli_epi32(ewitab,2);
2518 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2519 ewtabD = _mm_setzero_pd();
2520 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2521 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2522 ewtabFn = _mm_setzero_pd();
2523 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2524 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2525 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2526 velec = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
2527 felec = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2529 d = _mm_sub_pd(r01,rswitch);
2530 d = _mm_max_pd(d,_mm_setzero_pd());
2531 d2 = _mm_mul_pd(d,d);
2532 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2534 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2536 /* Evaluate switch function */
2537 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2538 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
2539 cutoff_mask = _mm_cmplt_pd(rsq01,rcutoff2);
2543 fscal = _mm_and_pd(fscal,cutoff_mask);
2545 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2547 /* Update vectorial force */
2548 fix0 = _mm_macc_pd(dx01,fscal,fix0);
2549 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
2550 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
2552 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
2553 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
2554 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
2558 /**************************
2559 * CALCULATE INTERACTIONS *
2560 **************************/
2562 if (gmx_mm_any_lt(rsq02,rcutoff2))
2565 r02 = _mm_mul_pd(rsq02,rinv02);
2567 /* EWALD ELECTROSTATICS */
2569 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2570 ewrt = _mm_mul_pd(r02,ewtabscale);
2571 ewitab = _mm_cvttpd_epi32(ewrt);
2573 eweps = _mm_frcz_pd(ewrt);
2575 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2577 twoeweps = _mm_add_pd(eweps,eweps);
2578 ewitab = _mm_slli_epi32(ewitab,2);
2579 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2580 ewtabD = _mm_setzero_pd();
2581 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2582 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2583 ewtabFn = _mm_setzero_pd();
2584 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2585 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2586 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2587 velec = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
2588 felec = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2590 d = _mm_sub_pd(r02,rswitch);
2591 d = _mm_max_pd(d,_mm_setzero_pd());
2592 d2 = _mm_mul_pd(d,d);
2593 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2595 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2597 /* Evaluate switch function */
2598 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2599 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
2600 cutoff_mask = _mm_cmplt_pd(rsq02,rcutoff2);
2604 fscal = _mm_and_pd(fscal,cutoff_mask);
2606 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2608 /* Update vectorial force */
2609 fix0 = _mm_macc_pd(dx02,fscal,fix0);
2610 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
2611 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
2613 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
2614 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
2615 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
2619 /**************************
2620 * CALCULATE INTERACTIONS *
2621 **************************/
2623 if (gmx_mm_any_lt(rsq10,rcutoff2))
2626 r10 = _mm_mul_pd(rsq10,rinv10);
2628 /* EWALD ELECTROSTATICS */
2630 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2631 ewrt = _mm_mul_pd(r10,ewtabscale);
2632 ewitab = _mm_cvttpd_epi32(ewrt);
2634 eweps = _mm_frcz_pd(ewrt);
2636 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2638 twoeweps = _mm_add_pd(eweps,eweps);
2639 ewitab = _mm_slli_epi32(ewitab,2);
2640 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2641 ewtabD = _mm_setzero_pd();
2642 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2643 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2644 ewtabFn = _mm_setzero_pd();
2645 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2646 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2647 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2648 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
2649 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2651 d = _mm_sub_pd(r10,rswitch);
2652 d = _mm_max_pd(d,_mm_setzero_pd());
2653 d2 = _mm_mul_pd(d,d);
2654 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2656 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2658 /* Evaluate switch function */
2659 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2660 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
2661 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
2665 fscal = _mm_and_pd(fscal,cutoff_mask);
2667 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2669 /* Update vectorial force */
2670 fix1 = _mm_macc_pd(dx10,fscal,fix1);
2671 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
2672 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
2674 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
2675 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
2676 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
2680 /**************************
2681 * CALCULATE INTERACTIONS *
2682 **************************/
2684 if (gmx_mm_any_lt(rsq11,rcutoff2))
2687 r11 = _mm_mul_pd(rsq11,rinv11);
2689 /* EWALD ELECTROSTATICS */
2691 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2692 ewrt = _mm_mul_pd(r11,ewtabscale);
2693 ewitab = _mm_cvttpd_epi32(ewrt);
2695 eweps = _mm_frcz_pd(ewrt);
2697 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2699 twoeweps = _mm_add_pd(eweps,eweps);
2700 ewitab = _mm_slli_epi32(ewitab,2);
2701 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2702 ewtabD = _mm_setzero_pd();
2703 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2704 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2705 ewtabFn = _mm_setzero_pd();
2706 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2707 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2708 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2709 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2710 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2712 d = _mm_sub_pd(r11,rswitch);
2713 d = _mm_max_pd(d,_mm_setzero_pd());
2714 d2 = _mm_mul_pd(d,d);
2715 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2717 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2719 /* Evaluate switch function */
2720 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2721 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2722 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2726 fscal = _mm_and_pd(fscal,cutoff_mask);
2728 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2730 /* Update vectorial force */
2731 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2732 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2733 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2735 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2736 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2737 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2741 /**************************
2742 * CALCULATE INTERACTIONS *
2743 **************************/
2745 if (gmx_mm_any_lt(rsq12,rcutoff2))
2748 r12 = _mm_mul_pd(rsq12,rinv12);
2750 /* EWALD ELECTROSTATICS */
2752 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2753 ewrt = _mm_mul_pd(r12,ewtabscale);
2754 ewitab = _mm_cvttpd_epi32(ewrt);
2756 eweps = _mm_frcz_pd(ewrt);
2758 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2760 twoeweps = _mm_add_pd(eweps,eweps);
2761 ewitab = _mm_slli_epi32(ewitab,2);
2762 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2763 ewtabD = _mm_setzero_pd();
2764 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2765 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2766 ewtabFn = _mm_setzero_pd();
2767 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2768 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2769 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2770 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2771 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2773 d = _mm_sub_pd(r12,rswitch);
2774 d = _mm_max_pd(d,_mm_setzero_pd());
2775 d2 = _mm_mul_pd(d,d);
2776 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2778 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2780 /* Evaluate switch function */
2781 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2782 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2783 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2787 fscal = _mm_and_pd(fscal,cutoff_mask);
2789 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2791 /* Update vectorial force */
2792 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2793 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2794 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2796 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2797 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2798 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2802 /**************************
2803 * CALCULATE INTERACTIONS *
2804 **************************/
2806 if (gmx_mm_any_lt(rsq20,rcutoff2))
2809 r20 = _mm_mul_pd(rsq20,rinv20);
2811 /* EWALD ELECTROSTATICS */
2813 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2814 ewrt = _mm_mul_pd(r20,ewtabscale);
2815 ewitab = _mm_cvttpd_epi32(ewrt);
2817 eweps = _mm_frcz_pd(ewrt);
2819 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2821 twoeweps = _mm_add_pd(eweps,eweps);
2822 ewitab = _mm_slli_epi32(ewitab,2);
2823 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2824 ewtabD = _mm_setzero_pd();
2825 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2826 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2827 ewtabFn = _mm_setzero_pd();
2828 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2829 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2830 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2831 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
2832 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2834 d = _mm_sub_pd(r20,rswitch);
2835 d = _mm_max_pd(d,_mm_setzero_pd());
2836 d2 = _mm_mul_pd(d,d);
2837 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2839 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2841 /* Evaluate switch function */
2842 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2843 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
2844 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
2848 fscal = _mm_and_pd(fscal,cutoff_mask);
2850 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2852 /* Update vectorial force */
2853 fix2 = _mm_macc_pd(dx20,fscal,fix2);
2854 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
2855 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
2857 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
2858 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
2859 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
2863 /**************************
2864 * CALCULATE INTERACTIONS *
2865 **************************/
2867 if (gmx_mm_any_lt(rsq21,rcutoff2))
2870 r21 = _mm_mul_pd(rsq21,rinv21);
2872 /* EWALD ELECTROSTATICS */
2874 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2875 ewrt = _mm_mul_pd(r21,ewtabscale);
2876 ewitab = _mm_cvttpd_epi32(ewrt);
2878 eweps = _mm_frcz_pd(ewrt);
2880 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2882 twoeweps = _mm_add_pd(eweps,eweps);
2883 ewitab = _mm_slli_epi32(ewitab,2);
2884 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2885 ewtabD = _mm_setzero_pd();
2886 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2887 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2888 ewtabFn = _mm_setzero_pd();
2889 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2890 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2891 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2892 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2893 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2895 d = _mm_sub_pd(r21,rswitch);
2896 d = _mm_max_pd(d,_mm_setzero_pd());
2897 d2 = _mm_mul_pd(d,d);
2898 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2900 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2902 /* Evaluate switch function */
2903 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2904 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2905 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2909 fscal = _mm_and_pd(fscal,cutoff_mask);
2911 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2913 /* Update vectorial force */
2914 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2915 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2916 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2918 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2919 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2920 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2924 /**************************
2925 * CALCULATE INTERACTIONS *
2926 **************************/
2928 if (gmx_mm_any_lt(rsq22,rcutoff2))
2931 r22 = _mm_mul_pd(rsq22,rinv22);
2933 /* EWALD ELECTROSTATICS */
2935 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2936 ewrt = _mm_mul_pd(r22,ewtabscale);
2937 ewitab = _mm_cvttpd_epi32(ewrt);
2939 eweps = _mm_frcz_pd(ewrt);
2941 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2943 twoeweps = _mm_add_pd(eweps,eweps);
2944 ewitab = _mm_slli_epi32(ewitab,2);
2945 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2946 ewtabD = _mm_setzero_pd();
2947 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2948 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2949 ewtabFn = _mm_setzero_pd();
2950 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2951 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2952 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2953 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2954 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2956 d = _mm_sub_pd(r22,rswitch);
2957 d = _mm_max_pd(d,_mm_setzero_pd());
2958 d2 = _mm_mul_pd(d,d);
2959 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2961 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2963 /* Evaluate switch function */
2964 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2965 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2966 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2970 fscal = _mm_and_pd(fscal,cutoff_mask);
2972 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2974 /* Update vectorial force */
2975 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2976 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2977 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2979 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2980 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2981 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2985 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2987 /* Inner loop uses 585 flops */
2990 /* End of innermost loop */
2992 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2993 f+i_coord_offset,fshift+i_shift_offset);
2995 /* Increment number of inner iterations */
2996 inneriter += j_index_end - j_index_start;
2998 /* Outer loop uses 18 flops */
3001 /* Increment number of outer iterations */
3004 /* Update outer/inner flops */
3006 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*585);