2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_VF_avx_128_fma_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LennardJones
56 * Geometry: Water4-Water4
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_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;
89 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
90 int vdwjidx0A,vdwjidx0B;
91 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 int vdwjidx1A,vdwjidx1B;
93 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
94 int vdwjidx2A,vdwjidx2B;
95 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
96 int vdwjidx3A,vdwjidx3B;
97 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
98 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
99 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
100 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
101 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
102 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
103 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
104 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
105 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
106 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
107 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
108 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
111 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
114 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
115 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
117 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
119 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
120 real rswitch_scalar,d_scalar;
121 __m128d dummy_mask,cutoff_mask;
122 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
123 __m128d one = _mm_set1_pd(1.0);
124 __m128d two = _mm_set1_pd(2.0);
130 jindex = nlist->jindex;
132 shiftidx = nlist->shift;
134 shiftvec = fr->shift_vec[0];
135 fshift = fr->fshift[0];
136 facel = _mm_set1_pd(fr->epsfac);
137 charge = mdatoms->chargeA;
138 nvdwtype = fr->ntype;
140 vdwtype = mdatoms->typeA;
142 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
143 ewtab = fr->ic->tabq_coul_FDV0;
144 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
145 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
147 /* Setup water-specific parameters */
148 inr = nlist->iinr[0];
149 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
150 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
151 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
152 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
154 jq1 = _mm_set1_pd(charge[inr+1]);
155 jq2 = _mm_set1_pd(charge[inr+2]);
156 jq3 = _mm_set1_pd(charge[inr+3]);
157 vdwjidx0A = 2*vdwtype[inr+0];
158 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
159 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
160 qq11 = _mm_mul_pd(iq1,jq1);
161 qq12 = _mm_mul_pd(iq1,jq2);
162 qq13 = _mm_mul_pd(iq1,jq3);
163 qq21 = _mm_mul_pd(iq2,jq1);
164 qq22 = _mm_mul_pd(iq2,jq2);
165 qq23 = _mm_mul_pd(iq2,jq3);
166 qq31 = _mm_mul_pd(iq3,jq1);
167 qq32 = _mm_mul_pd(iq3,jq2);
168 qq33 = _mm_mul_pd(iq3,jq3);
170 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
171 rcutoff_scalar = fr->rcoulomb;
172 rcutoff = _mm_set1_pd(rcutoff_scalar);
173 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
175 rswitch_scalar = fr->rcoulomb_switch;
176 rswitch = _mm_set1_pd(rswitch_scalar);
177 /* Setup switch parameters */
178 d_scalar = rcutoff_scalar-rswitch_scalar;
179 d = _mm_set1_pd(d_scalar);
180 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
181 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
182 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
183 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
184 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
185 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
187 /* Avoid stupid compiler warnings */
195 /* Start outer loop over neighborlists */
196 for(iidx=0; iidx<nri; iidx++)
198 /* Load shift vector for this list */
199 i_shift_offset = DIM*shiftidx[iidx];
201 /* Load limits for loop over neighbors */
202 j_index_start = jindex[iidx];
203 j_index_end = jindex[iidx+1];
205 /* Get outer coordinate index */
207 i_coord_offset = DIM*inr;
209 /* Load i particle coords and add shift vector */
210 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
211 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
213 fix0 = _mm_setzero_pd();
214 fiy0 = _mm_setzero_pd();
215 fiz0 = _mm_setzero_pd();
216 fix1 = _mm_setzero_pd();
217 fiy1 = _mm_setzero_pd();
218 fiz1 = _mm_setzero_pd();
219 fix2 = _mm_setzero_pd();
220 fiy2 = _mm_setzero_pd();
221 fiz2 = _mm_setzero_pd();
222 fix3 = _mm_setzero_pd();
223 fiy3 = _mm_setzero_pd();
224 fiz3 = _mm_setzero_pd();
226 /* Reset potential sums */
227 velecsum = _mm_setzero_pd();
228 vvdwsum = _mm_setzero_pd();
230 /* Start inner kernel loop */
231 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
234 /* Get j neighbor index, and coordinate index */
237 j_coord_offsetA = DIM*jnrA;
238 j_coord_offsetB = DIM*jnrB;
240 /* load j atom coordinates */
241 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
242 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
243 &jy2,&jz2,&jx3,&jy3,&jz3);
245 /* Calculate displacement vector */
246 dx00 = _mm_sub_pd(ix0,jx0);
247 dy00 = _mm_sub_pd(iy0,jy0);
248 dz00 = _mm_sub_pd(iz0,jz0);
249 dx11 = _mm_sub_pd(ix1,jx1);
250 dy11 = _mm_sub_pd(iy1,jy1);
251 dz11 = _mm_sub_pd(iz1,jz1);
252 dx12 = _mm_sub_pd(ix1,jx2);
253 dy12 = _mm_sub_pd(iy1,jy2);
254 dz12 = _mm_sub_pd(iz1,jz2);
255 dx13 = _mm_sub_pd(ix1,jx3);
256 dy13 = _mm_sub_pd(iy1,jy3);
257 dz13 = _mm_sub_pd(iz1,jz3);
258 dx21 = _mm_sub_pd(ix2,jx1);
259 dy21 = _mm_sub_pd(iy2,jy1);
260 dz21 = _mm_sub_pd(iz2,jz1);
261 dx22 = _mm_sub_pd(ix2,jx2);
262 dy22 = _mm_sub_pd(iy2,jy2);
263 dz22 = _mm_sub_pd(iz2,jz2);
264 dx23 = _mm_sub_pd(ix2,jx3);
265 dy23 = _mm_sub_pd(iy2,jy3);
266 dz23 = _mm_sub_pd(iz2,jz3);
267 dx31 = _mm_sub_pd(ix3,jx1);
268 dy31 = _mm_sub_pd(iy3,jy1);
269 dz31 = _mm_sub_pd(iz3,jz1);
270 dx32 = _mm_sub_pd(ix3,jx2);
271 dy32 = _mm_sub_pd(iy3,jy2);
272 dz32 = _mm_sub_pd(iz3,jz2);
273 dx33 = _mm_sub_pd(ix3,jx3);
274 dy33 = _mm_sub_pd(iy3,jy3);
275 dz33 = _mm_sub_pd(iz3,jz3);
277 /* Calculate squared distance and things based on it */
278 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
279 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
280 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
281 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
282 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
283 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
284 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
285 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
286 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
287 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
289 rinv00 = gmx_mm_invsqrt_pd(rsq00);
290 rinv11 = gmx_mm_invsqrt_pd(rsq11);
291 rinv12 = gmx_mm_invsqrt_pd(rsq12);
292 rinv13 = gmx_mm_invsqrt_pd(rsq13);
293 rinv21 = gmx_mm_invsqrt_pd(rsq21);
294 rinv22 = gmx_mm_invsqrt_pd(rsq22);
295 rinv23 = gmx_mm_invsqrt_pd(rsq23);
296 rinv31 = gmx_mm_invsqrt_pd(rsq31);
297 rinv32 = gmx_mm_invsqrt_pd(rsq32);
298 rinv33 = gmx_mm_invsqrt_pd(rsq33);
300 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
301 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
302 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
303 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
304 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
305 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
306 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
307 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
308 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
309 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
311 fjx0 = _mm_setzero_pd();
312 fjy0 = _mm_setzero_pd();
313 fjz0 = _mm_setzero_pd();
314 fjx1 = _mm_setzero_pd();
315 fjy1 = _mm_setzero_pd();
316 fjz1 = _mm_setzero_pd();
317 fjx2 = _mm_setzero_pd();
318 fjy2 = _mm_setzero_pd();
319 fjz2 = _mm_setzero_pd();
320 fjx3 = _mm_setzero_pd();
321 fjy3 = _mm_setzero_pd();
322 fjz3 = _mm_setzero_pd();
324 /**************************
325 * CALCULATE INTERACTIONS *
326 **************************/
328 if (gmx_mm_any_lt(rsq00,rcutoff2))
331 r00 = _mm_mul_pd(rsq00,rinv00);
333 /* LENNARD-JONES DISPERSION/REPULSION */
335 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
336 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
337 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
338 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
339 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
341 d = _mm_sub_pd(r00,rswitch);
342 d = _mm_max_pd(d,_mm_setzero_pd());
343 d2 = _mm_mul_pd(d,d);
344 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
346 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
348 /* Evaluate switch function */
349 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
350 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
351 vvdw = _mm_mul_pd(vvdw,sw);
352 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
354 /* Update potential sum for this i atom from the interaction with this j atom. */
355 vvdw = _mm_and_pd(vvdw,cutoff_mask);
356 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
360 fscal = _mm_and_pd(fscal,cutoff_mask);
362 /* Update vectorial force */
363 fix0 = _mm_macc_pd(dx00,fscal,fix0);
364 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
365 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
367 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
368 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
369 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
373 /**************************
374 * CALCULATE INTERACTIONS *
375 **************************/
377 if (gmx_mm_any_lt(rsq11,rcutoff2))
380 r11 = _mm_mul_pd(rsq11,rinv11);
382 /* EWALD ELECTROSTATICS */
384 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
385 ewrt = _mm_mul_pd(r11,ewtabscale);
386 ewitab = _mm_cvttpd_epi32(ewrt);
388 eweps = _mm_frcz_pd(ewrt);
390 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
392 twoeweps = _mm_add_pd(eweps,eweps);
393 ewitab = _mm_slli_epi32(ewitab,2);
394 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
395 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
396 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
397 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
398 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
399 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
400 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
401 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
402 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
403 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
405 d = _mm_sub_pd(r11,rswitch);
406 d = _mm_max_pd(d,_mm_setzero_pd());
407 d2 = _mm_mul_pd(d,d);
408 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
410 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
412 /* Evaluate switch function */
413 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
414 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
415 velec = _mm_mul_pd(velec,sw);
416 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
418 /* Update potential sum for this i atom from the interaction with this j atom. */
419 velec = _mm_and_pd(velec,cutoff_mask);
420 velecsum = _mm_add_pd(velecsum,velec);
424 fscal = _mm_and_pd(fscal,cutoff_mask);
426 /* Update vectorial force */
427 fix1 = _mm_macc_pd(dx11,fscal,fix1);
428 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
429 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
431 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
432 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
433 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
437 /**************************
438 * CALCULATE INTERACTIONS *
439 **************************/
441 if (gmx_mm_any_lt(rsq12,rcutoff2))
444 r12 = _mm_mul_pd(rsq12,rinv12);
446 /* EWALD ELECTROSTATICS */
448 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
449 ewrt = _mm_mul_pd(r12,ewtabscale);
450 ewitab = _mm_cvttpd_epi32(ewrt);
452 eweps = _mm_frcz_pd(ewrt);
454 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
456 twoeweps = _mm_add_pd(eweps,eweps);
457 ewitab = _mm_slli_epi32(ewitab,2);
458 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
459 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
460 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
461 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
462 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
463 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
464 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
465 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
466 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
467 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
469 d = _mm_sub_pd(r12,rswitch);
470 d = _mm_max_pd(d,_mm_setzero_pd());
471 d2 = _mm_mul_pd(d,d);
472 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
474 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
476 /* Evaluate switch function */
477 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
478 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
479 velec = _mm_mul_pd(velec,sw);
480 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
482 /* Update potential sum for this i atom from the interaction with this j atom. */
483 velec = _mm_and_pd(velec,cutoff_mask);
484 velecsum = _mm_add_pd(velecsum,velec);
488 fscal = _mm_and_pd(fscal,cutoff_mask);
490 /* Update vectorial force */
491 fix1 = _mm_macc_pd(dx12,fscal,fix1);
492 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
493 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
495 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
496 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
497 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
501 /**************************
502 * CALCULATE INTERACTIONS *
503 **************************/
505 if (gmx_mm_any_lt(rsq13,rcutoff2))
508 r13 = _mm_mul_pd(rsq13,rinv13);
510 /* EWALD ELECTROSTATICS */
512 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
513 ewrt = _mm_mul_pd(r13,ewtabscale);
514 ewitab = _mm_cvttpd_epi32(ewrt);
516 eweps = _mm_frcz_pd(ewrt);
518 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
520 twoeweps = _mm_add_pd(eweps,eweps);
521 ewitab = _mm_slli_epi32(ewitab,2);
522 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
523 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
524 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
525 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
526 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
527 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
528 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
529 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
530 velec = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
531 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
533 d = _mm_sub_pd(r13,rswitch);
534 d = _mm_max_pd(d,_mm_setzero_pd());
535 d2 = _mm_mul_pd(d,d);
536 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
538 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
540 /* Evaluate switch function */
541 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
542 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv13,_mm_mul_pd(velec,dsw)) );
543 velec = _mm_mul_pd(velec,sw);
544 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
546 /* Update potential sum for this i atom from the interaction with this j atom. */
547 velec = _mm_and_pd(velec,cutoff_mask);
548 velecsum = _mm_add_pd(velecsum,velec);
552 fscal = _mm_and_pd(fscal,cutoff_mask);
554 /* Update vectorial force */
555 fix1 = _mm_macc_pd(dx13,fscal,fix1);
556 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
557 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
559 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
560 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
561 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
565 /**************************
566 * CALCULATE INTERACTIONS *
567 **************************/
569 if (gmx_mm_any_lt(rsq21,rcutoff2))
572 r21 = _mm_mul_pd(rsq21,rinv21);
574 /* EWALD ELECTROSTATICS */
576 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
577 ewrt = _mm_mul_pd(r21,ewtabscale);
578 ewitab = _mm_cvttpd_epi32(ewrt);
580 eweps = _mm_frcz_pd(ewrt);
582 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
584 twoeweps = _mm_add_pd(eweps,eweps);
585 ewitab = _mm_slli_epi32(ewitab,2);
586 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
587 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
588 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
589 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
590 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
591 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
592 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
593 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
594 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
595 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
597 d = _mm_sub_pd(r21,rswitch);
598 d = _mm_max_pd(d,_mm_setzero_pd());
599 d2 = _mm_mul_pd(d,d);
600 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
602 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
604 /* Evaluate switch function */
605 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
606 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
607 velec = _mm_mul_pd(velec,sw);
608 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
610 /* Update potential sum for this i atom from the interaction with this j atom. */
611 velec = _mm_and_pd(velec,cutoff_mask);
612 velecsum = _mm_add_pd(velecsum,velec);
616 fscal = _mm_and_pd(fscal,cutoff_mask);
618 /* Update vectorial force */
619 fix2 = _mm_macc_pd(dx21,fscal,fix2);
620 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
621 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
623 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
624 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
625 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
629 /**************************
630 * CALCULATE INTERACTIONS *
631 **************************/
633 if (gmx_mm_any_lt(rsq22,rcutoff2))
636 r22 = _mm_mul_pd(rsq22,rinv22);
638 /* EWALD ELECTROSTATICS */
640 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
641 ewrt = _mm_mul_pd(r22,ewtabscale);
642 ewitab = _mm_cvttpd_epi32(ewrt);
644 eweps = _mm_frcz_pd(ewrt);
646 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
648 twoeweps = _mm_add_pd(eweps,eweps);
649 ewitab = _mm_slli_epi32(ewitab,2);
650 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
651 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
652 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
653 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
654 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
655 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
656 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
657 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
658 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
659 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
661 d = _mm_sub_pd(r22,rswitch);
662 d = _mm_max_pd(d,_mm_setzero_pd());
663 d2 = _mm_mul_pd(d,d);
664 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
666 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
668 /* Evaluate switch function */
669 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
670 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
671 velec = _mm_mul_pd(velec,sw);
672 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
674 /* Update potential sum for this i atom from the interaction with this j atom. */
675 velec = _mm_and_pd(velec,cutoff_mask);
676 velecsum = _mm_add_pd(velecsum,velec);
680 fscal = _mm_and_pd(fscal,cutoff_mask);
682 /* Update vectorial force */
683 fix2 = _mm_macc_pd(dx22,fscal,fix2);
684 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
685 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
687 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
688 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
689 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
693 /**************************
694 * CALCULATE INTERACTIONS *
695 **************************/
697 if (gmx_mm_any_lt(rsq23,rcutoff2))
700 r23 = _mm_mul_pd(rsq23,rinv23);
702 /* EWALD ELECTROSTATICS */
704 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
705 ewrt = _mm_mul_pd(r23,ewtabscale);
706 ewitab = _mm_cvttpd_epi32(ewrt);
708 eweps = _mm_frcz_pd(ewrt);
710 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
712 twoeweps = _mm_add_pd(eweps,eweps);
713 ewitab = _mm_slli_epi32(ewitab,2);
714 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
715 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
716 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
717 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
718 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
719 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
720 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
721 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
722 velec = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
723 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
725 d = _mm_sub_pd(r23,rswitch);
726 d = _mm_max_pd(d,_mm_setzero_pd());
727 d2 = _mm_mul_pd(d,d);
728 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
730 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
732 /* Evaluate switch function */
733 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
734 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv23,_mm_mul_pd(velec,dsw)) );
735 velec = _mm_mul_pd(velec,sw);
736 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
738 /* Update potential sum for this i atom from the interaction with this j atom. */
739 velec = _mm_and_pd(velec,cutoff_mask);
740 velecsum = _mm_add_pd(velecsum,velec);
744 fscal = _mm_and_pd(fscal,cutoff_mask);
746 /* Update vectorial force */
747 fix2 = _mm_macc_pd(dx23,fscal,fix2);
748 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
749 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
751 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
752 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
753 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
757 /**************************
758 * CALCULATE INTERACTIONS *
759 **************************/
761 if (gmx_mm_any_lt(rsq31,rcutoff2))
764 r31 = _mm_mul_pd(rsq31,rinv31);
766 /* EWALD ELECTROSTATICS */
768 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
769 ewrt = _mm_mul_pd(r31,ewtabscale);
770 ewitab = _mm_cvttpd_epi32(ewrt);
772 eweps = _mm_frcz_pd(ewrt);
774 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
776 twoeweps = _mm_add_pd(eweps,eweps);
777 ewitab = _mm_slli_epi32(ewitab,2);
778 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
779 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
780 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
781 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
782 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
783 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
784 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
785 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
786 velec = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
787 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
789 d = _mm_sub_pd(r31,rswitch);
790 d = _mm_max_pd(d,_mm_setzero_pd());
791 d2 = _mm_mul_pd(d,d);
792 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
794 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
796 /* Evaluate switch function */
797 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
798 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv31,_mm_mul_pd(velec,dsw)) );
799 velec = _mm_mul_pd(velec,sw);
800 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
802 /* Update potential sum for this i atom from the interaction with this j atom. */
803 velec = _mm_and_pd(velec,cutoff_mask);
804 velecsum = _mm_add_pd(velecsum,velec);
808 fscal = _mm_and_pd(fscal,cutoff_mask);
810 /* Update vectorial force */
811 fix3 = _mm_macc_pd(dx31,fscal,fix3);
812 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
813 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
815 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
816 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
817 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
821 /**************************
822 * CALCULATE INTERACTIONS *
823 **************************/
825 if (gmx_mm_any_lt(rsq32,rcutoff2))
828 r32 = _mm_mul_pd(rsq32,rinv32);
830 /* EWALD ELECTROSTATICS */
832 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
833 ewrt = _mm_mul_pd(r32,ewtabscale);
834 ewitab = _mm_cvttpd_epi32(ewrt);
836 eweps = _mm_frcz_pd(ewrt);
838 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
840 twoeweps = _mm_add_pd(eweps,eweps);
841 ewitab = _mm_slli_epi32(ewitab,2);
842 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
843 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
844 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
845 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
846 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
847 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
848 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
849 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
850 velec = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
851 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
853 d = _mm_sub_pd(r32,rswitch);
854 d = _mm_max_pd(d,_mm_setzero_pd());
855 d2 = _mm_mul_pd(d,d);
856 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
858 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
860 /* Evaluate switch function */
861 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
862 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv32,_mm_mul_pd(velec,dsw)) );
863 velec = _mm_mul_pd(velec,sw);
864 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
866 /* Update potential sum for this i atom from the interaction with this j atom. */
867 velec = _mm_and_pd(velec,cutoff_mask);
868 velecsum = _mm_add_pd(velecsum,velec);
872 fscal = _mm_and_pd(fscal,cutoff_mask);
874 /* Update vectorial force */
875 fix3 = _mm_macc_pd(dx32,fscal,fix3);
876 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
877 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
879 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
880 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
881 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
885 /**************************
886 * CALCULATE INTERACTIONS *
887 **************************/
889 if (gmx_mm_any_lt(rsq33,rcutoff2))
892 r33 = _mm_mul_pd(rsq33,rinv33);
894 /* EWALD ELECTROSTATICS */
896 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
897 ewrt = _mm_mul_pd(r33,ewtabscale);
898 ewitab = _mm_cvttpd_epi32(ewrt);
900 eweps = _mm_frcz_pd(ewrt);
902 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
904 twoeweps = _mm_add_pd(eweps,eweps);
905 ewitab = _mm_slli_epi32(ewitab,2);
906 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
907 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
908 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
909 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
910 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
911 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
912 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
913 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
914 velec = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
915 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
917 d = _mm_sub_pd(r33,rswitch);
918 d = _mm_max_pd(d,_mm_setzero_pd());
919 d2 = _mm_mul_pd(d,d);
920 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
922 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
924 /* Evaluate switch function */
925 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
926 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv33,_mm_mul_pd(velec,dsw)) );
927 velec = _mm_mul_pd(velec,sw);
928 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
930 /* Update potential sum for this i atom from the interaction with this j atom. */
931 velec = _mm_and_pd(velec,cutoff_mask);
932 velecsum = _mm_add_pd(velecsum,velec);
936 fscal = _mm_and_pd(fscal,cutoff_mask);
938 /* Update vectorial force */
939 fix3 = _mm_macc_pd(dx33,fscal,fix3);
940 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
941 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
943 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
944 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
945 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
949 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
951 /* Inner loop uses 677 flops */
958 j_coord_offsetA = DIM*jnrA;
960 /* load j atom coordinates */
961 gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
962 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
963 &jy2,&jz2,&jx3,&jy3,&jz3);
965 /* Calculate displacement vector */
966 dx00 = _mm_sub_pd(ix0,jx0);
967 dy00 = _mm_sub_pd(iy0,jy0);
968 dz00 = _mm_sub_pd(iz0,jz0);
969 dx11 = _mm_sub_pd(ix1,jx1);
970 dy11 = _mm_sub_pd(iy1,jy1);
971 dz11 = _mm_sub_pd(iz1,jz1);
972 dx12 = _mm_sub_pd(ix1,jx2);
973 dy12 = _mm_sub_pd(iy1,jy2);
974 dz12 = _mm_sub_pd(iz1,jz2);
975 dx13 = _mm_sub_pd(ix1,jx3);
976 dy13 = _mm_sub_pd(iy1,jy3);
977 dz13 = _mm_sub_pd(iz1,jz3);
978 dx21 = _mm_sub_pd(ix2,jx1);
979 dy21 = _mm_sub_pd(iy2,jy1);
980 dz21 = _mm_sub_pd(iz2,jz1);
981 dx22 = _mm_sub_pd(ix2,jx2);
982 dy22 = _mm_sub_pd(iy2,jy2);
983 dz22 = _mm_sub_pd(iz2,jz2);
984 dx23 = _mm_sub_pd(ix2,jx3);
985 dy23 = _mm_sub_pd(iy2,jy3);
986 dz23 = _mm_sub_pd(iz2,jz3);
987 dx31 = _mm_sub_pd(ix3,jx1);
988 dy31 = _mm_sub_pd(iy3,jy1);
989 dz31 = _mm_sub_pd(iz3,jz1);
990 dx32 = _mm_sub_pd(ix3,jx2);
991 dy32 = _mm_sub_pd(iy3,jy2);
992 dz32 = _mm_sub_pd(iz3,jz2);
993 dx33 = _mm_sub_pd(ix3,jx3);
994 dy33 = _mm_sub_pd(iy3,jy3);
995 dz33 = _mm_sub_pd(iz3,jz3);
997 /* Calculate squared distance and things based on it */
998 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
999 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1000 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1001 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1002 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1003 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1004 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1005 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1006 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1007 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1009 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1010 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1011 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1012 rinv13 = gmx_mm_invsqrt_pd(rsq13);
1013 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1014 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1015 rinv23 = gmx_mm_invsqrt_pd(rsq23);
1016 rinv31 = gmx_mm_invsqrt_pd(rsq31);
1017 rinv32 = gmx_mm_invsqrt_pd(rsq32);
1018 rinv33 = gmx_mm_invsqrt_pd(rsq33);
1020 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1021 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1022 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1023 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
1024 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1025 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1026 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
1027 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
1028 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
1029 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
1031 fjx0 = _mm_setzero_pd();
1032 fjy0 = _mm_setzero_pd();
1033 fjz0 = _mm_setzero_pd();
1034 fjx1 = _mm_setzero_pd();
1035 fjy1 = _mm_setzero_pd();
1036 fjz1 = _mm_setzero_pd();
1037 fjx2 = _mm_setzero_pd();
1038 fjy2 = _mm_setzero_pd();
1039 fjz2 = _mm_setzero_pd();
1040 fjx3 = _mm_setzero_pd();
1041 fjy3 = _mm_setzero_pd();
1042 fjz3 = _mm_setzero_pd();
1044 /**************************
1045 * CALCULATE INTERACTIONS *
1046 **************************/
1048 if (gmx_mm_any_lt(rsq00,rcutoff2))
1051 r00 = _mm_mul_pd(rsq00,rinv00);
1053 /* LENNARD-JONES DISPERSION/REPULSION */
1055 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1056 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
1057 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1058 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1059 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1061 d = _mm_sub_pd(r00,rswitch);
1062 d = _mm_max_pd(d,_mm_setzero_pd());
1063 d2 = _mm_mul_pd(d,d);
1064 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1066 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1068 /* Evaluate switch function */
1069 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1070 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1071 vvdw = _mm_mul_pd(vvdw,sw);
1072 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1074 /* Update potential sum for this i atom from the interaction with this j atom. */
1075 vvdw = _mm_and_pd(vvdw,cutoff_mask);
1076 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
1077 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
1081 fscal = _mm_and_pd(fscal,cutoff_mask);
1083 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1085 /* Update vectorial force */
1086 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1087 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1088 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1090 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1091 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1092 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1096 /**************************
1097 * CALCULATE INTERACTIONS *
1098 **************************/
1100 if (gmx_mm_any_lt(rsq11,rcutoff2))
1103 r11 = _mm_mul_pd(rsq11,rinv11);
1105 /* EWALD ELECTROSTATICS */
1107 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1108 ewrt = _mm_mul_pd(r11,ewtabscale);
1109 ewitab = _mm_cvttpd_epi32(ewrt);
1111 eweps = _mm_frcz_pd(ewrt);
1113 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1115 twoeweps = _mm_add_pd(eweps,eweps);
1116 ewitab = _mm_slli_epi32(ewitab,2);
1117 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1118 ewtabD = _mm_setzero_pd();
1119 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1120 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1121 ewtabFn = _mm_setzero_pd();
1122 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1123 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1124 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1125 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
1126 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1128 d = _mm_sub_pd(r11,rswitch);
1129 d = _mm_max_pd(d,_mm_setzero_pd());
1130 d2 = _mm_mul_pd(d,d);
1131 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1133 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1135 /* Evaluate switch function */
1136 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1137 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
1138 velec = _mm_mul_pd(velec,sw);
1139 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1141 /* Update potential sum for this i atom from the interaction with this j atom. */
1142 velec = _mm_and_pd(velec,cutoff_mask);
1143 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1144 velecsum = _mm_add_pd(velecsum,velec);
1148 fscal = _mm_and_pd(fscal,cutoff_mask);
1150 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1152 /* Update vectorial force */
1153 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1154 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1155 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1157 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1158 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1159 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1163 /**************************
1164 * CALCULATE INTERACTIONS *
1165 **************************/
1167 if (gmx_mm_any_lt(rsq12,rcutoff2))
1170 r12 = _mm_mul_pd(rsq12,rinv12);
1172 /* EWALD ELECTROSTATICS */
1174 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1175 ewrt = _mm_mul_pd(r12,ewtabscale);
1176 ewitab = _mm_cvttpd_epi32(ewrt);
1178 eweps = _mm_frcz_pd(ewrt);
1180 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1182 twoeweps = _mm_add_pd(eweps,eweps);
1183 ewitab = _mm_slli_epi32(ewitab,2);
1184 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1185 ewtabD = _mm_setzero_pd();
1186 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1187 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1188 ewtabFn = _mm_setzero_pd();
1189 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1190 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1191 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1192 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1193 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1195 d = _mm_sub_pd(r12,rswitch);
1196 d = _mm_max_pd(d,_mm_setzero_pd());
1197 d2 = _mm_mul_pd(d,d);
1198 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1200 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1202 /* Evaluate switch function */
1203 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1204 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
1205 velec = _mm_mul_pd(velec,sw);
1206 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1208 /* Update potential sum for this i atom from the interaction with this j atom. */
1209 velec = _mm_and_pd(velec,cutoff_mask);
1210 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1211 velecsum = _mm_add_pd(velecsum,velec);
1215 fscal = _mm_and_pd(fscal,cutoff_mask);
1217 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1219 /* Update vectorial force */
1220 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1221 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1222 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1224 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1225 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1226 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1230 /**************************
1231 * CALCULATE INTERACTIONS *
1232 **************************/
1234 if (gmx_mm_any_lt(rsq13,rcutoff2))
1237 r13 = _mm_mul_pd(rsq13,rinv13);
1239 /* EWALD ELECTROSTATICS */
1241 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1242 ewrt = _mm_mul_pd(r13,ewtabscale);
1243 ewitab = _mm_cvttpd_epi32(ewrt);
1245 eweps = _mm_frcz_pd(ewrt);
1247 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1249 twoeweps = _mm_add_pd(eweps,eweps);
1250 ewitab = _mm_slli_epi32(ewitab,2);
1251 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1252 ewtabD = _mm_setzero_pd();
1253 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1254 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1255 ewtabFn = _mm_setzero_pd();
1256 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1257 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1258 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1259 velec = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
1260 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1262 d = _mm_sub_pd(r13,rswitch);
1263 d = _mm_max_pd(d,_mm_setzero_pd());
1264 d2 = _mm_mul_pd(d,d);
1265 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1267 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1269 /* Evaluate switch function */
1270 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1271 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv13,_mm_mul_pd(velec,dsw)) );
1272 velec = _mm_mul_pd(velec,sw);
1273 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
1275 /* Update potential sum for this i atom from the interaction with this j atom. */
1276 velec = _mm_and_pd(velec,cutoff_mask);
1277 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1278 velecsum = _mm_add_pd(velecsum,velec);
1282 fscal = _mm_and_pd(fscal,cutoff_mask);
1284 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1286 /* Update vectorial force */
1287 fix1 = _mm_macc_pd(dx13,fscal,fix1);
1288 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
1289 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1291 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1292 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1293 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1297 /**************************
1298 * CALCULATE INTERACTIONS *
1299 **************************/
1301 if (gmx_mm_any_lt(rsq21,rcutoff2))
1304 r21 = _mm_mul_pd(rsq21,rinv21);
1306 /* EWALD ELECTROSTATICS */
1308 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1309 ewrt = _mm_mul_pd(r21,ewtabscale);
1310 ewitab = _mm_cvttpd_epi32(ewrt);
1312 eweps = _mm_frcz_pd(ewrt);
1314 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1316 twoeweps = _mm_add_pd(eweps,eweps);
1317 ewitab = _mm_slli_epi32(ewitab,2);
1318 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1319 ewtabD = _mm_setzero_pd();
1320 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1321 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1322 ewtabFn = _mm_setzero_pd();
1323 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1324 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1325 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1326 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1327 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1329 d = _mm_sub_pd(r21,rswitch);
1330 d = _mm_max_pd(d,_mm_setzero_pd());
1331 d2 = _mm_mul_pd(d,d);
1332 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1334 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1336 /* Evaluate switch function */
1337 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1338 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
1339 velec = _mm_mul_pd(velec,sw);
1340 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1342 /* Update potential sum for this i atom from the interaction with this j atom. */
1343 velec = _mm_and_pd(velec,cutoff_mask);
1344 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1345 velecsum = _mm_add_pd(velecsum,velec);
1349 fscal = _mm_and_pd(fscal,cutoff_mask);
1351 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1353 /* Update vectorial force */
1354 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1355 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1356 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1358 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1359 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1360 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1364 /**************************
1365 * CALCULATE INTERACTIONS *
1366 **************************/
1368 if (gmx_mm_any_lt(rsq22,rcutoff2))
1371 r22 = _mm_mul_pd(rsq22,rinv22);
1373 /* EWALD ELECTROSTATICS */
1375 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1376 ewrt = _mm_mul_pd(r22,ewtabscale);
1377 ewitab = _mm_cvttpd_epi32(ewrt);
1379 eweps = _mm_frcz_pd(ewrt);
1381 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1383 twoeweps = _mm_add_pd(eweps,eweps);
1384 ewitab = _mm_slli_epi32(ewitab,2);
1385 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1386 ewtabD = _mm_setzero_pd();
1387 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1388 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1389 ewtabFn = _mm_setzero_pd();
1390 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1391 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1392 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1393 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1394 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1396 d = _mm_sub_pd(r22,rswitch);
1397 d = _mm_max_pd(d,_mm_setzero_pd());
1398 d2 = _mm_mul_pd(d,d);
1399 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1401 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1403 /* Evaluate switch function */
1404 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1405 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
1406 velec = _mm_mul_pd(velec,sw);
1407 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1409 /* Update potential sum for this i atom from the interaction with this j atom. */
1410 velec = _mm_and_pd(velec,cutoff_mask);
1411 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1412 velecsum = _mm_add_pd(velecsum,velec);
1416 fscal = _mm_and_pd(fscal,cutoff_mask);
1418 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1420 /* Update vectorial force */
1421 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1422 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1423 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1425 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1426 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1427 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1431 /**************************
1432 * CALCULATE INTERACTIONS *
1433 **************************/
1435 if (gmx_mm_any_lt(rsq23,rcutoff2))
1438 r23 = _mm_mul_pd(rsq23,rinv23);
1440 /* EWALD ELECTROSTATICS */
1442 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1443 ewrt = _mm_mul_pd(r23,ewtabscale);
1444 ewitab = _mm_cvttpd_epi32(ewrt);
1446 eweps = _mm_frcz_pd(ewrt);
1448 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1450 twoeweps = _mm_add_pd(eweps,eweps);
1451 ewitab = _mm_slli_epi32(ewitab,2);
1452 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1453 ewtabD = _mm_setzero_pd();
1454 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1455 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1456 ewtabFn = _mm_setzero_pd();
1457 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1458 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1459 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1460 velec = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
1461 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1463 d = _mm_sub_pd(r23,rswitch);
1464 d = _mm_max_pd(d,_mm_setzero_pd());
1465 d2 = _mm_mul_pd(d,d);
1466 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1468 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1470 /* Evaluate switch function */
1471 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1472 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv23,_mm_mul_pd(velec,dsw)) );
1473 velec = _mm_mul_pd(velec,sw);
1474 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
1476 /* Update potential sum for this i atom from the interaction with this j atom. */
1477 velec = _mm_and_pd(velec,cutoff_mask);
1478 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1479 velecsum = _mm_add_pd(velecsum,velec);
1483 fscal = _mm_and_pd(fscal,cutoff_mask);
1485 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1487 /* Update vectorial force */
1488 fix2 = _mm_macc_pd(dx23,fscal,fix2);
1489 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
1490 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
1492 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
1493 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
1494 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
1498 /**************************
1499 * CALCULATE INTERACTIONS *
1500 **************************/
1502 if (gmx_mm_any_lt(rsq31,rcutoff2))
1505 r31 = _mm_mul_pd(rsq31,rinv31);
1507 /* EWALD ELECTROSTATICS */
1509 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1510 ewrt = _mm_mul_pd(r31,ewtabscale);
1511 ewitab = _mm_cvttpd_epi32(ewrt);
1513 eweps = _mm_frcz_pd(ewrt);
1515 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1517 twoeweps = _mm_add_pd(eweps,eweps);
1518 ewitab = _mm_slli_epi32(ewitab,2);
1519 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1520 ewtabD = _mm_setzero_pd();
1521 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1522 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1523 ewtabFn = _mm_setzero_pd();
1524 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1525 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1526 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1527 velec = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
1528 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1530 d = _mm_sub_pd(r31,rswitch);
1531 d = _mm_max_pd(d,_mm_setzero_pd());
1532 d2 = _mm_mul_pd(d,d);
1533 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1535 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1537 /* Evaluate switch function */
1538 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1539 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv31,_mm_mul_pd(velec,dsw)) );
1540 velec = _mm_mul_pd(velec,sw);
1541 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
1543 /* Update potential sum for this i atom from the interaction with this j atom. */
1544 velec = _mm_and_pd(velec,cutoff_mask);
1545 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1546 velecsum = _mm_add_pd(velecsum,velec);
1550 fscal = _mm_and_pd(fscal,cutoff_mask);
1552 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1554 /* Update vectorial force */
1555 fix3 = _mm_macc_pd(dx31,fscal,fix3);
1556 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
1557 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
1559 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
1560 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
1561 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
1565 /**************************
1566 * CALCULATE INTERACTIONS *
1567 **************************/
1569 if (gmx_mm_any_lt(rsq32,rcutoff2))
1572 r32 = _mm_mul_pd(rsq32,rinv32);
1574 /* EWALD ELECTROSTATICS */
1576 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1577 ewrt = _mm_mul_pd(r32,ewtabscale);
1578 ewitab = _mm_cvttpd_epi32(ewrt);
1580 eweps = _mm_frcz_pd(ewrt);
1582 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1584 twoeweps = _mm_add_pd(eweps,eweps);
1585 ewitab = _mm_slli_epi32(ewitab,2);
1586 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1587 ewtabD = _mm_setzero_pd();
1588 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1589 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1590 ewtabFn = _mm_setzero_pd();
1591 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1592 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1593 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1594 velec = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
1595 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1597 d = _mm_sub_pd(r32,rswitch);
1598 d = _mm_max_pd(d,_mm_setzero_pd());
1599 d2 = _mm_mul_pd(d,d);
1600 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1602 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1604 /* Evaluate switch function */
1605 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1606 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv32,_mm_mul_pd(velec,dsw)) );
1607 velec = _mm_mul_pd(velec,sw);
1608 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
1610 /* Update potential sum for this i atom from the interaction with this j atom. */
1611 velec = _mm_and_pd(velec,cutoff_mask);
1612 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1613 velecsum = _mm_add_pd(velecsum,velec);
1617 fscal = _mm_and_pd(fscal,cutoff_mask);
1619 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1621 /* Update vectorial force */
1622 fix3 = _mm_macc_pd(dx32,fscal,fix3);
1623 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
1624 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
1626 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
1627 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
1628 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
1632 /**************************
1633 * CALCULATE INTERACTIONS *
1634 **************************/
1636 if (gmx_mm_any_lt(rsq33,rcutoff2))
1639 r33 = _mm_mul_pd(rsq33,rinv33);
1641 /* EWALD ELECTROSTATICS */
1643 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1644 ewrt = _mm_mul_pd(r33,ewtabscale);
1645 ewitab = _mm_cvttpd_epi32(ewrt);
1647 eweps = _mm_frcz_pd(ewrt);
1649 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1651 twoeweps = _mm_add_pd(eweps,eweps);
1652 ewitab = _mm_slli_epi32(ewitab,2);
1653 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1654 ewtabD = _mm_setzero_pd();
1655 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1656 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1657 ewtabFn = _mm_setzero_pd();
1658 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1659 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1660 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1661 velec = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
1662 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1664 d = _mm_sub_pd(r33,rswitch);
1665 d = _mm_max_pd(d,_mm_setzero_pd());
1666 d2 = _mm_mul_pd(d,d);
1667 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1669 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1671 /* Evaluate switch function */
1672 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1673 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv33,_mm_mul_pd(velec,dsw)) );
1674 velec = _mm_mul_pd(velec,sw);
1675 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
1677 /* Update potential sum for this i atom from the interaction with this j atom. */
1678 velec = _mm_and_pd(velec,cutoff_mask);
1679 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1680 velecsum = _mm_add_pd(velecsum,velec);
1684 fscal = _mm_and_pd(fscal,cutoff_mask);
1686 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1688 /* Update vectorial force */
1689 fix3 = _mm_macc_pd(dx33,fscal,fix3);
1690 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
1691 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
1693 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
1694 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
1695 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
1699 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1701 /* Inner loop uses 677 flops */
1704 /* End of innermost loop */
1706 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1707 f+i_coord_offset,fshift+i_shift_offset);
1710 /* Update potential energies */
1711 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1712 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1714 /* Increment number of inner iterations */
1715 inneriter += j_index_end - j_index_start;
1717 /* Outer loop uses 26 flops */
1720 /* Increment number of outer iterations */
1723 /* Update outer/inner flops */
1725 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*677);
1728 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_F_avx_128_fma_double
1729 * Electrostatics interaction: Ewald
1730 * VdW interaction: LennardJones
1731 * Geometry: Water4-Water4
1732 * Calculate force/pot: Force
1735 nb_kernel_ElecEwSw_VdwLJSw_GeomW4W4_F_avx_128_fma_double
1736 (t_nblist * gmx_restrict nlist,
1737 rvec * gmx_restrict xx,
1738 rvec * gmx_restrict ff,
1739 t_forcerec * gmx_restrict fr,
1740 t_mdatoms * gmx_restrict mdatoms,
1741 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1742 t_nrnb * gmx_restrict nrnb)
1744 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1745 * just 0 for non-waters.
1746 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1747 * jnr indices corresponding to data put in the four positions in the SIMD register.
1749 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1750 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1752 int j_coord_offsetA,j_coord_offsetB;
1753 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1754 real rcutoff_scalar;
1755 real *shiftvec,*fshift,*x,*f;
1756 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1758 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1760 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1762 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1764 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1765 int vdwjidx0A,vdwjidx0B;
1766 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1767 int vdwjidx1A,vdwjidx1B;
1768 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1769 int vdwjidx2A,vdwjidx2B;
1770 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1771 int vdwjidx3A,vdwjidx3B;
1772 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1773 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1774 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1775 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1776 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1777 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1778 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1779 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1780 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1781 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1782 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1783 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1786 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1789 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
1790 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
1792 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1794 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1795 real rswitch_scalar,d_scalar;
1796 __m128d dummy_mask,cutoff_mask;
1797 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1798 __m128d one = _mm_set1_pd(1.0);
1799 __m128d two = _mm_set1_pd(2.0);
1805 jindex = nlist->jindex;
1807 shiftidx = nlist->shift;
1809 shiftvec = fr->shift_vec[0];
1810 fshift = fr->fshift[0];
1811 facel = _mm_set1_pd(fr->epsfac);
1812 charge = mdatoms->chargeA;
1813 nvdwtype = fr->ntype;
1814 vdwparam = fr->nbfp;
1815 vdwtype = mdatoms->typeA;
1817 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1818 ewtab = fr->ic->tabq_coul_FDV0;
1819 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1820 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1822 /* Setup water-specific parameters */
1823 inr = nlist->iinr[0];
1824 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1825 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1826 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1827 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1829 jq1 = _mm_set1_pd(charge[inr+1]);
1830 jq2 = _mm_set1_pd(charge[inr+2]);
1831 jq3 = _mm_set1_pd(charge[inr+3]);
1832 vdwjidx0A = 2*vdwtype[inr+0];
1833 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1834 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1835 qq11 = _mm_mul_pd(iq1,jq1);
1836 qq12 = _mm_mul_pd(iq1,jq2);
1837 qq13 = _mm_mul_pd(iq1,jq3);
1838 qq21 = _mm_mul_pd(iq2,jq1);
1839 qq22 = _mm_mul_pd(iq2,jq2);
1840 qq23 = _mm_mul_pd(iq2,jq3);
1841 qq31 = _mm_mul_pd(iq3,jq1);
1842 qq32 = _mm_mul_pd(iq3,jq2);
1843 qq33 = _mm_mul_pd(iq3,jq3);
1845 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1846 rcutoff_scalar = fr->rcoulomb;
1847 rcutoff = _mm_set1_pd(rcutoff_scalar);
1848 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
1850 rswitch_scalar = fr->rcoulomb_switch;
1851 rswitch = _mm_set1_pd(rswitch_scalar);
1852 /* Setup switch parameters */
1853 d_scalar = rcutoff_scalar-rswitch_scalar;
1854 d = _mm_set1_pd(d_scalar);
1855 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
1856 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1857 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1858 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
1859 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1860 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1862 /* Avoid stupid compiler warnings */
1864 j_coord_offsetA = 0;
1865 j_coord_offsetB = 0;
1870 /* Start outer loop over neighborlists */
1871 for(iidx=0; iidx<nri; iidx++)
1873 /* Load shift vector for this list */
1874 i_shift_offset = DIM*shiftidx[iidx];
1876 /* Load limits for loop over neighbors */
1877 j_index_start = jindex[iidx];
1878 j_index_end = jindex[iidx+1];
1880 /* Get outer coordinate index */
1882 i_coord_offset = DIM*inr;
1884 /* Load i particle coords and add shift vector */
1885 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1886 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1888 fix0 = _mm_setzero_pd();
1889 fiy0 = _mm_setzero_pd();
1890 fiz0 = _mm_setzero_pd();
1891 fix1 = _mm_setzero_pd();
1892 fiy1 = _mm_setzero_pd();
1893 fiz1 = _mm_setzero_pd();
1894 fix2 = _mm_setzero_pd();
1895 fiy2 = _mm_setzero_pd();
1896 fiz2 = _mm_setzero_pd();
1897 fix3 = _mm_setzero_pd();
1898 fiy3 = _mm_setzero_pd();
1899 fiz3 = _mm_setzero_pd();
1901 /* Start inner kernel loop */
1902 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1905 /* Get j neighbor index, and coordinate index */
1907 jnrB = jjnr[jidx+1];
1908 j_coord_offsetA = DIM*jnrA;
1909 j_coord_offsetB = DIM*jnrB;
1911 /* load j atom coordinates */
1912 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1913 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1914 &jy2,&jz2,&jx3,&jy3,&jz3);
1916 /* Calculate displacement vector */
1917 dx00 = _mm_sub_pd(ix0,jx0);
1918 dy00 = _mm_sub_pd(iy0,jy0);
1919 dz00 = _mm_sub_pd(iz0,jz0);
1920 dx11 = _mm_sub_pd(ix1,jx1);
1921 dy11 = _mm_sub_pd(iy1,jy1);
1922 dz11 = _mm_sub_pd(iz1,jz1);
1923 dx12 = _mm_sub_pd(ix1,jx2);
1924 dy12 = _mm_sub_pd(iy1,jy2);
1925 dz12 = _mm_sub_pd(iz1,jz2);
1926 dx13 = _mm_sub_pd(ix1,jx3);
1927 dy13 = _mm_sub_pd(iy1,jy3);
1928 dz13 = _mm_sub_pd(iz1,jz3);
1929 dx21 = _mm_sub_pd(ix2,jx1);
1930 dy21 = _mm_sub_pd(iy2,jy1);
1931 dz21 = _mm_sub_pd(iz2,jz1);
1932 dx22 = _mm_sub_pd(ix2,jx2);
1933 dy22 = _mm_sub_pd(iy2,jy2);
1934 dz22 = _mm_sub_pd(iz2,jz2);
1935 dx23 = _mm_sub_pd(ix2,jx3);
1936 dy23 = _mm_sub_pd(iy2,jy3);
1937 dz23 = _mm_sub_pd(iz2,jz3);
1938 dx31 = _mm_sub_pd(ix3,jx1);
1939 dy31 = _mm_sub_pd(iy3,jy1);
1940 dz31 = _mm_sub_pd(iz3,jz1);
1941 dx32 = _mm_sub_pd(ix3,jx2);
1942 dy32 = _mm_sub_pd(iy3,jy2);
1943 dz32 = _mm_sub_pd(iz3,jz2);
1944 dx33 = _mm_sub_pd(ix3,jx3);
1945 dy33 = _mm_sub_pd(iy3,jy3);
1946 dz33 = _mm_sub_pd(iz3,jz3);
1948 /* Calculate squared distance and things based on it */
1949 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1950 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1951 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1952 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1953 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1954 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1955 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1956 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1957 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1958 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1960 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1961 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1962 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1963 rinv13 = gmx_mm_invsqrt_pd(rsq13);
1964 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1965 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1966 rinv23 = gmx_mm_invsqrt_pd(rsq23);
1967 rinv31 = gmx_mm_invsqrt_pd(rsq31);
1968 rinv32 = gmx_mm_invsqrt_pd(rsq32);
1969 rinv33 = gmx_mm_invsqrt_pd(rsq33);
1971 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1972 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1973 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1974 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
1975 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1976 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1977 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
1978 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
1979 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
1980 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
1982 fjx0 = _mm_setzero_pd();
1983 fjy0 = _mm_setzero_pd();
1984 fjz0 = _mm_setzero_pd();
1985 fjx1 = _mm_setzero_pd();
1986 fjy1 = _mm_setzero_pd();
1987 fjz1 = _mm_setzero_pd();
1988 fjx2 = _mm_setzero_pd();
1989 fjy2 = _mm_setzero_pd();
1990 fjz2 = _mm_setzero_pd();
1991 fjx3 = _mm_setzero_pd();
1992 fjy3 = _mm_setzero_pd();
1993 fjz3 = _mm_setzero_pd();
1995 /**************************
1996 * CALCULATE INTERACTIONS *
1997 **************************/
1999 if (gmx_mm_any_lt(rsq00,rcutoff2))
2002 r00 = _mm_mul_pd(rsq00,rinv00);
2004 /* LENNARD-JONES DISPERSION/REPULSION */
2006 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2007 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
2008 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
2009 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
2010 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
2012 d = _mm_sub_pd(r00,rswitch);
2013 d = _mm_max_pd(d,_mm_setzero_pd());
2014 d2 = _mm_mul_pd(d,d);
2015 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2017 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2019 /* Evaluate switch function */
2020 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2021 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
2022 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
2026 fscal = _mm_and_pd(fscal,cutoff_mask);
2028 /* Update vectorial force */
2029 fix0 = _mm_macc_pd(dx00,fscal,fix0);
2030 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
2031 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
2033 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
2034 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
2035 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
2039 /**************************
2040 * CALCULATE INTERACTIONS *
2041 **************************/
2043 if (gmx_mm_any_lt(rsq11,rcutoff2))
2046 r11 = _mm_mul_pd(rsq11,rinv11);
2048 /* EWALD ELECTROSTATICS */
2050 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2051 ewrt = _mm_mul_pd(r11,ewtabscale);
2052 ewitab = _mm_cvttpd_epi32(ewrt);
2054 eweps = _mm_frcz_pd(ewrt);
2056 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2058 twoeweps = _mm_add_pd(eweps,eweps);
2059 ewitab = _mm_slli_epi32(ewitab,2);
2060 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2061 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2062 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2063 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2064 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2065 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2066 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2067 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2068 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2069 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2071 d = _mm_sub_pd(r11,rswitch);
2072 d = _mm_max_pd(d,_mm_setzero_pd());
2073 d2 = _mm_mul_pd(d,d);
2074 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2076 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2078 /* Evaluate switch function */
2079 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2080 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2081 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2085 fscal = _mm_and_pd(fscal,cutoff_mask);
2087 /* Update vectorial force */
2088 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2089 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2090 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2092 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2093 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2094 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2098 /**************************
2099 * CALCULATE INTERACTIONS *
2100 **************************/
2102 if (gmx_mm_any_lt(rsq12,rcutoff2))
2105 r12 = _mm_mul_pd(rsq12,rinv12);
2107 /* EWALD ELECTROSTATICS */
2109 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2110 ewrt = _mm_mul_pd(r12,ewtabscale);
2111 ewitab = _mm_cvttpd_epi32(ewrt);
2113 eweps = _mm_frcz_pd(ewrt);
2115 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2117 twoeweps = _mm_add_pd(eweps,eweps);
2118 ewitab = _mm_slli_epi32(ewitab,2);
2119 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2120 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2121 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2122 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2123 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2124 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2125 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2126 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2127 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2128 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2130 d = _mm_sub_pd(r12,rswitch);
2131 d = _mm_max_pd(d,_mm_setzero_pd());
2132 d2 = _mm_mul_pd(d,d);
2133 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2135 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2137 /* Evaluate switch function */
2138 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2139 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2140 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2144 fscal = _mm_and_pd(fscal,cutoff_mask);
2146 /* Update vectorial force */
2147 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2148 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2149 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2151 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2152 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2153 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2157 /**************************
2158 * CALCULATE INTERACTIONS *
2159 **************************/
2161 if (gmx_mm_any_lt(rsq13,rcutoff2))
2164 r13 = _mm_mul_pd(rsq13,rinv13);
2166 /* EWALD ELECTROSTATICS */
2168 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2169 ewrt = _mm_mul_pd(r13,ewtabscale);
2170 ewitab = _mm_cvttpd_epi32(ewrt);
2172 eweps = _mm_frcz_pd(ewrt);
2174 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2176 twoeweps = _mm_add_pd(eweps,eweps);
2177 ewitab = _mm_slli_epi32(ewitab,2);
2178 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2179 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2180 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2181 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2182 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2183 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2184 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2185 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2186 velec = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
2187 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2189 d = _mm_sub_pd(r13,rswitch);
2190 d = _mm_max_pd(d,_mm_setzero_pd());
2191 d2 = _mm_mul_pd(d,d);
2192 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2194 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2196 /* Evaluate switch function */
2197 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2198 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv13,_mm_mul_pd(velec,dsw)) );
2199 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
2203 fscal = _mm_and_pd(fscal,cutoff_mask);
2205 /* Update vectorial force */
2206 fix1 = _mm_macc_pd(dx13,fscal,fix1);
2207 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
2208 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
2210 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
2211 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
2212 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
2216 /**************************
2217 * CALCULATE INTERACTIONS *
2218 **************************/
2220 if (gmx_mm_any_lt(rsq21,rcutoff2))
2223 r21 = _mm_mul_pd(rsq21,rinv21);
2225 /* EWALD ELECTROSTATICS */
2227 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2228 ewrt = _mm_mul_pd(r21,ewtabscale);
2229 ewitab = _mm_cvttpd_epi32(ewrt);
2231 eweps = _mm_frcz_pd(ewrt);
2233 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2235 twoeweps = _mm_add_pd(eweps,eweps);
2236 ewitab = _mm_slli_epi32(ewitab,2);
2237 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2238 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2239 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2240 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2241 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2242 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2243 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2244 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2245 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2246 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2248 d = _mm_sub_pd(r21,rswitch);
2249 d = _mm_max_pd(d,_mm_setzero_pd());
2250 d2 = _mm_mul_pd(d,d);
2251 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2253 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2255 /* Evaluate switch function */
2256 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2257 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2258 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2262 fscal = _mm_and_pd(fscal,cutoff_mask);
2264 /* Update vectorial force */
2265 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2266 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2267 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2269 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2270 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2271 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2275 /**************************
2276 * CALCULATE INTERACTIONS *
2277 **************************/
2279 if (gmx_mm_any_lt(rsq22,rcutoff2))
2282 r22 = _mm_mul_pd(rsq22,rinv22);
2284 /* EWALD ELECTROSTATICS */
2286 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2287 ewrt = _mm_mul_pd(r22,ewtabscale);
2288 ewitab = _mm_cvttpd_epi32(ewrt);
2290 eweps = _mm_frcz_pd(ewrt);
2292 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2294 twoeweps = _mm_add_pd(eweps,eweps);
2295 ewitab = _mm_slli_epi32(ewitab,2);
2296 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2297 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2298 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2299 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2300 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2301 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2302 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2303 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2304 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2305 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2307 d = _mm_sub_pd(r22,rswitch);
2308 d = _mm_max_pd(d,_mm_setzero_pd());
2309 d2 = _mm_mul_pd(d,d);
2310 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2312 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2314 /* Evaluate switch function */
2315 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2316 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2317 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2321 fscal = _mm_and_pd(fscal,cutoff_mask);
2323 /* Update vectorial force */
2324 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2325 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2326 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2328 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2329 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2330 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2334 /**************************
2335 * CALCULATE INTERACTIONS *
2336 **************************/
2338 if (gmx_mm_any_lt(rsq23,rcutoff2))
2341 r23 = _mm_mul_pd(rsq23,rinv23);
2343 /* EWALD ELECTROSTATICS */
2345 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2346 ewrt = _mm_mul_pd(r23,ewtabscale);
2347 ewitab = _mm_cvttpd_epi32(ewrt);
2349 eweps = _mm_frcz_pd(ewrt);
2351 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2353 twoeweps = _mm_add_pd(eweps,eweps);
2354 ewitab = _mm_slli_epi32(ewitab,2);
2355 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2356 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2357 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2358 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2359 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2360 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2361 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2362 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2363 velec = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
2364 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2366 d = _mm_sub_pd(r23,rswitch);
2367 d = _mm_max_pd(d,_mm_setzero_pd());
2368 d2 = _mm_mul_pd(d,d);
2369 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2371 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2373 /* Evaluate switch function */
2374 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2375 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv23,_mm_mul_pd(velec,dsw)) );
2376 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
2380 fscal = _mm_and_pd(fscal,cutoff_mask);
2382 /* Update vectorial force */
2383 fix2 = _mm_macc_pd(dx23,fscal,fix2);
2384 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
2385 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
2387 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
2388 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
2389 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
2393 /**************************
2394 * CALCULATE INTERACTIONS *
2395 **************************/
2397 if (gmx_mm_any_lt(rsq31,rcutoff2))
2400 r31 = _mm_mul_pd(rsq31,rinv31);
2402 /* EWALD ELECTROSTATICS */
2404 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2405 ewrt = _mm_mul_pd(r31,ewtabscale);
2406 ewitab = _mm_cvttpd_epi32(ewrt);
2408 eweps = _mm_frcz_pd(ewrt);
2410 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2412 twoeweps = _mm_add_pd(eweps,eweps);
2413 ewitab = _mm_slli_epi32(ewitab,2);
2414 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2415 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2416 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2417 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2418 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2419 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2420 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2421 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2422 velec = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
2423 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2425 d = _mm_sub_pd(r31,rswitch);
2426 d = _mm_max_pd(d,_mm_setzero_pd());
2427 d2 = _mm_mul_pd(d,d);
2428 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2430 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2432 /* Evaluate switch function */
2433 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2434 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv31,_mm_mul_pd(velec,dsw)) );
2435 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
2439 fscal = _mm_and_pd(fscal,cutoff_mask);
2441 /* Update vectorial force */
2442 fix3 = _mm_macc_pd(dx31,fscal,fix3);
2443 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
2444 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
2446 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
2447 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
2448 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
2452 /**************************
2453 * CALCULATE INTERACTIONS *
2454 **************************/
2456 if (gmx_mm_any_lt(rsq32,rcutoff2))
2459 r32 = _mm_mul_pd(rsq32,rinv32);
2461 /* EWALD ELECTROSTATICS */
2463 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2464 ewrt = _mm_mul_pd(r32,ewtabscale);
2465 ewitab = _mm_cvttpd_epi32(ewrt);
2467 eweps = _mm_frcz_pd(ewrt);
2469 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2471 twoeweps = _mm_add_pd(eweps,eweps);
2472 ewitab = _mm_slli_epi32(ewitab,2);
2473 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2474 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2475 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2476 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2477 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2478 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2479 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2480 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2481 velec = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
2482 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2484 d = _mm_sub_pd(r32,rswitch);
2485 d = _mm_max_pd(d,_mm_setzero_pd());
2486 d2 = _mm_mul_pd(d,d);
2487 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2489 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2491 /* Evaluate switch function */
2492 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2493 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv32,_mm_mul_pd(velec,dsw)) );
2494 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
2498 fscal = _mm_and_pd(fscal,cutoff_mask);
2500 /* Update vectorial force */
2501 fix3 = _mm_macc_pd(dx32,fscal,fix3);
2502 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
2503 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
2505 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
2506 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
2507 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
2511 /**************************
2512 * CALCULATE INTERACTIONS *
2513 **************************/
2515 if (gmx_mm_any_lt(rsq33,rcutoff2))
2518 r33 = _mm_mul_pd(rsq33,rinv33);
2520 /* EWALD ELECTROSTATICS */
2522 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2523 ewrt = _mm_mul_pd(r33,ewtabscale);
2524 ewitab = _mm_cvttpd_epi32(ewrt);
2526 eweps = _mm_frcz_pd(ewrt);
2528 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2530 twoeweps = _mm_add_pd(eweps,eweps);
2531 ewitab = _mm_slli_epi32(ewitab,2);
2532 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2533 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2534 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2535 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2536 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2537 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2538 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2539 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2540 velec = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
2541 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2543 d = _mm_sub_pd(r33,rswitch);
2544 d = _mm_max_pd(d,_mm_setzero_pd());
2545 d2 = _mm_mul_pd(d,d);
2546 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2548 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2550 /* Evaluate switch function */
2551 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2552 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv33,_mm_mul_pd(velec,dsw)) );
2553 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
2557 fscal = _mm_and_pd(fscal,cutoff_mask);
2559 /* Update vectorial force */
2560 fix3 = _mm_macc_pd(dx33,fscal,fix3);
2561 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
2562 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
2564 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
2565 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
2566 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
2570 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2572 /* Inner loop uses 647 flops */
2575 if(jidx<j_index_end)
2579 j_coord_offsetA = DIM*jnrA;
2581 /* load j atom coordinates */
2582 gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2583 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2584 &jy2,&jz2,&jx3,&jy3,&jz3);
2586 /* Calculate displacement vector */
2587 dx00 = _mm_sub_pd(ix0,jx0);
2588 dy00 = _mm_sub_pd(iy0,jy0);
2589 dz00 = _mm_sub_pd(iz0,jz0);
2590 dx11 = _mm_sub_pd(ix1,jx1);
2591 dy11 = _mm_sub_pd(iy1,jy1);
2592 dz11 = _mm_sub_pd(iz1,jz1);
2593 dx12 = _mm_sub_pd(ix1,jx2);
2594 dy12 = _mm_sub_pd(iy1,jy2);
2595 dz12 = _mm_sub_pd(iz1,jz2);
2596 dx13 = _mm_sub_pd(ix1,jx3);
2597 dy13 = _mm_sub_pd(iy1,jy3);
2598 dz13 = _mm_sub_pd(iz1,jz3);
2599 dx21 = _mm_sub_pd(ix2,jx1);
2600 dy21 = _mm_sub_pd(iy2,jy1);
2601 dz21 = _mm_sub_pd(iz2,jz1);
2602 dx22 = _mm_sub_pd(ix2,jx2);
2603 dy22 = _mm_sub_pd(iy2,jy2);
2604 dz22 = _mm_sub_pd(iz2,jz2);
2605 dx23 = _mm_sub_pd(ix2,jx3);
2606 dy23 = _mm_sub_pd(iy2,jy3);
2607 dz23 = _mm_sub_pd(iz2,jz3);
2608 dx31 = _mm_sub_pd(ix3,jx1);
2609 dy31 = _mm_sub_pd(iy3,jy1);
2610 dz31 = _mm_sub_pd(iz3,jz1);
2611 dx32 = _mm_sub_pd(ix3,jx2);
2612 dy32 = _mm_sub_pd(iy3,jy2);
2613 dz32 = _mm_sub_pd(iz3,jz2);
2614 dx33 = _mm_sub_pd(ix3,jx3);
2615 dy33 = _mm_sub_pd(iy3,jy3);
2616 dz33 = _mm_sub_pd(iz3,jz3);
2618 /* Calculate squared distance and things based on it */
2619 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2620 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2621 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2622 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
2623 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2624 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2625 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
2626 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
2627 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
2628 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
2630 rinv00 = gmx_mm_invsqrt_pd(rsq00);
2631 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2632 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2633 rinv13 = gmx_mm_invsqrt_pd(rsq13);
2634 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2635 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2636 rinv23 = gmx_mm_invsqrt_pd(rsq23);
2637 rinv31 = gmx_mm_invsqrt_pd(rsq31);
2638 rinv32 = gmx_mm_invsqrt_pd(rsq32);
2639 rinv33 = gmx_mm_invsqrt_pd(rsq33);
2641 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
2642 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
2643 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
2644 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
2645 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
2646 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
2647 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
2648 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
2649 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
2650 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
2652 fjx0 = _mm_setzero_pd();
2653 fjy0 = _mm_setzero_pd();
2654 fjz0 = _mm_setzero_pd();
2655 fjx1 = _mm_setzero_pd();
2656 fjy1 = _mm_setzero_pd();
2657 fjz1 = _mm_setzero_pd();
2658 fjx2 = _mm_setzero_pd();
2659 fjy2 = _mm_setzero_pd();
2660 fjz2 = _mm_setzero_pd();
2661 fjx3 = _mm_setzero_pd();
2662 fjy3 = _mm_setzero_pd();
2663 fjz3 = _mm_setzero_pd();
2665 /**************************
2666 * CALCULATE INTERACTIONS *
2667 **************************/
2669 if (gmx_mm_any_lt(rsq00,rcutoff2))
2672 r00 = _mm_mul_pd(rsq00,rinv00);
2674 /* LENNARD-JONES DISPERSION/REPULSION */
2676 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2677 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
2678 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
2679 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
2680 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
2682 d = _mm_sub_pd(r00,rswitch);
2683 d = _mm_max_pd(d,_mm_setzero_pd());
2684 d2 = _mm_mul_pd(d,d);
2685 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2687 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2689 /* Evaluate switch function */
2690 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2691 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
2692 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
2696 fscal = _mm_and_pd(fscal,cutoff_mask);
2698 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2700 /* Update vectorial force */
2701 fix0 = _mm_macc_pd(dx00,fscal,fix0);
2702 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
2703 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
2705 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
2706 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
2707 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
2711 /**************************
2712 * CALCULATE INTERACTIONS *
2713 **************************/
2715 if (gmx_mm_any_lt(rsq11,rcutoff2))
2718 r11 = _mm_mul_pd(rsq11,rinv11);
2720 /* EWALD ELECTROSTATICS */
2722 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2723 ewrt = _mm_mul_pd(r11,ewtabscale);
2724 ewitab = _mm_cvttpd_epi32(ewrt);
2726 eweps = _mm_frcz_pd(ewrt);
2728 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2730 twoeweps = _mm_add_pd(eweps,eweps);
2731 ewitab = _mm_slli_epi32(ewitab,2);
2732 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2733 ewtabD = _mm_setzero_pd();
2734 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2735 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2736 ewtabFn = _mm_setzero_pd();
2737 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2738 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2739 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2740 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2741 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2743 d = _mm_sub_pd(r11,rswitch);
2744 d = _mm_max_pd(d,_mm_setzero_pd());
2745 d2 = _mm_mul_pd(d,d);
2746 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2748 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2750 /* Evaluate switch function */
2751 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2752 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2753 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2757 fscal = _mm_and_pd(fscal,cutoff_mask);
2759 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2761 /* Update vectorial force */
2762 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2763 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2764 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2766 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2767 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2768 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2772 /**************************
2773 * CALCULATE INTERACTIONS *
2774 **************************/
2776 if (gmx_mm_any_lt(rsq12,rcutoff2))
2779 r12 = _mm_mul_pd(rsq12,rinv12);
2781 /* EWALD ELECTROSTATICS */
2783 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2784 ewrt = _mm_mul_pd(r12,ewtabscale);
2785 ewitab = _mm_cvttpd_epi32(ewrt);
2787 eweps = _mm_frcz_pd(ewrt);
2789 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2791 twoeweps = _mm_add_pd(eweps,eweps);
2792 ewitab = _mm_slli_epi32(ewitab,2);
2793 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2794 ewtabD = _mm_setzero_pd();
2795 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2796 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2797 ewtabFn = _mm_setzero_pd();
2798 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2799 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2800 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2801 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2802 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2804 d = _mm_sub_pd(r12,rswitch);
2805 d = _mm_max_pd(d,_mm_setzero_pd());
2806 d2 = _mm_mul_pd(d,d);
2807 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2809 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2811 /* Evaluate switch function */
2812 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2813 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2814 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2818 fscal = _mm_and_pd(fscal,cutoff_mask);
2820 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2822 /* Update vectorial force */
2823 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2824 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2825 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2827 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2828 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2829 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2833 /**************************
2834 * CALCULATE INTERACTIONS *
2835 **************************/
2837 if (gmx_mm_any_lt(rsq13,rcutoff2))
2840 r13 = _mm_mul_pd(rsq13,rinv13);
2842 /* EWALD ELECTROSTATICS */
2844 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2845 ewrt = _mm_mul_pd(r13,ewtabscale);
2846 ewitab = _mm_cvttpd_epi32(ewrt);
2848 eweps = _mm_frcz_pd(ewrt);
2850 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2852 twoeweps = _mm_add_pd(eweps,eweps);
2853 ewitab = _mm_slli_epi32(ewitab,2);
2854 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2855 ewtabD = _mm_setzero_pd();
2856 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2857 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2858 ewtabFn = _mm_setzero_pd();
2859 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2860 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2861 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2862 velec = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
2863 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2865 d = _mm_sub_pd(r13,rswitch);
2866 d = _mm_max_pd(d,_mm_setzero_pd());
2867 d2 = _mm_mul_pd(d,d);
2868 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2870 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2872 /* Evaluate switch function */
2873 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2874 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv13,_mm_mul_pd(velec,dsw)) );
2875 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
2879 fscal = _mm_and_pd(fscal,cutoff_mask);
2881 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2883 /* Update vectorial force */
2884 fix1 = _mm_macc_pd(dx13,fscal,fix1);
2885 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
2886 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
2888 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
2889 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
2890 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
2894 /**************************
2895 * CALCULATE INTERACTIONS *
2896 **************************/
2898 if (gmx_mm_any_lt(rsq21,rcutoff2))
2901 r21 = _mm_mul_pd(rsq21,rinv21);
2903 /* EWALD ELECTROSTATICS */
2905 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2906 ewrt = _mm_mul_pd(r21,ewtabscale);
2907 ewitab = _mm_cvttpd_epi32(ewrt);
2909 eweps = _mm_frcz_pd(ewrt);
2911 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2913 twoeweps = _mm_add_pd(eweps,eweps);
2914 ewitab = _mm_slli_epi32(ewitab,2);
2915 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2916 ewtabD = _mm_setzero_pd();
2917 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2918 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2919 ewtabFn = _mm_setzero_pd();
2920 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2921 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2922 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2923 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2924 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2926 d = _mm_sub_pd(r21,rswitch);
2927 d = _mm_max_pd(d,_mm_setzero_pd());
2928 d2 = _mm_mul_pd(d,d);
2929 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2931 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2933 /* Evaluate switch function */
2934 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2935 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2936 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2940 fscal = _mm_and_pd(fscal,cutoff_mask);
2942 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2944 /* Update vectorial force */
2945 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2946 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2947 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2949 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2950 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2951 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2955 /**************************
2956 * CALCULATE INTERACTIONS *
2957 **************************/
2959 if (gmx_mm_any_lt(rsq22,rcutoff2))
2962 r22 = _mm_mul_pd(rsq22,rinv22);
2964 /* EWALD ELECTROSTATICS */
2966 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2967 ewrt = _mm_mul_pd(r22,ewtabscale);
2968 ewitab = _mm_cvttpd_epi32(ewrt);
2970 eweps = _mm_frcz_pd(ewrt);
2972 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2974 twoeweps = _mm_add_pd(eweps,eweps);
2975 ewitab = _mm_slli_epi32(ewitab,2);
2976 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2977 ewtabD = _mm_setzero_pd();
2978 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2979 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2980 ewtabFn = _mm_setzero_pd();
2981 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2982 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2983 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2984 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2985 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2987 d = _mm_sub_pd(r22,rswitch);
2988 d = _mm_max_pd(d,_mm_setzero_pd());
2989 d2 = _mm_mul_pd(d,d);
2990 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2992 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2994 /* Evaluate switch function */
2995 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2996 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2997 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
3001 fscal = _mm_and_pd(fscal,cutoff_mask);
3003 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
3005 /* Update vectorial force */
3006 fix2 = _mm_macc_pd(dx22,fscal,fix2);
3007 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
3008 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
3010 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
3011 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
3012 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
3016 /**************************
3017 * CALCULATE INTERACTIONS *
3018 **************************/
3020 if (gmx_mm_any_lt(rsq23,rcutoff2))
3023 r23 = _mm_mul_pd(rsq23,rinv23);
3025 /* EWALD ELECTROSTATICS */
3027 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
3028 ewrt = _mm_mul_pd(r23,ewtabscale);
3029 ewitab = _mm_cvttpd_epi32(ewrt);
3031 eweps = _mm_frcz_pd(ewrt);
3033 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
3035 twoeweps = _mm_add_pd(eweps,eweps);
3036 ewitab = _mm_slli_epi32(ewitab,2);
3037 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
3038 ewtabD = _mm_setzero_pd();
3039 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
3040 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
3041 ewtabFn = _mm_setzero_pd();
3042 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
3043 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
3044 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
3045 velec = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
3046 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
3048 d = _mm_sub_pd(r23,rswitch);
3049 d = _mm_max_pd(d,_mm_setzero_pd());
3050 d2 = _mm_mul_pd(d,d);
3051 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
3053 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
3055 /* Evaluate switch function */
3056 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
3057 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv23,_mm_mul_pd(velec,dsw)) );
3058 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
3062 fscal = _mm_and_pd(fscal,cutoff_mask);
3064 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
3066 /* Update vectorial force */
3067 fix2 = _mm_macc_pd(dx23,fscal,fix2);
3068 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
3069 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
3071 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
3072 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
3073 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
3077 /**************************
3078 * CALCULATE INTERACTIONS *
3079 **************************/
3081 if (gmx_mm_any_lt(rsq31,rcutoff2))
3084 r31 = _mm_mul_pd(rsq31,rinv31);
3086 /* EWALD ELECTROSTATICS */
3088 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
3089 ewrt = _mm_mul_pd(r31,ewtabscale);
3090 ewitab = _mm_cvttpd_epi32(ewrt);
3092 eweps = _mm_frcz_pd(ewrt);
3094 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
3096 twoeweps = _mm_add_pd(eweps,eweps);
3097 ewitab = _mm_slli_epi32(ewitab,2);
3098 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
3099 ewtabD = _mm_setzero_pd();
3100 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
3101 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
3102 ewtabFn = _mm_setzero_pd();
3103 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
3104 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
3105 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
3106 velec = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
3107 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
3109 d = _mm_sub_pd(r31,rswitch);
3110 d = _mm_max_pd(d,_mm_setzero_pd());
3111 d2 = _mm_mul_pd(d,d);
3112 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
3114 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
3116 /* Evaluate switch function */
3117 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
3118 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv31,_mm_mul_pd(velec,dsw)) );
3119 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
3123 fscal = _mm_and_pd(fscal,cutoff_mask);
3125 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
3127 /* Update vectorial force */
3128 fix3 = _mm_macc_pd(dx31,fscal,fix3);
3129 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
3130 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
3132 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
3133 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
3134 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
3138 /**************************
3139 * CALCULATE INTERACTIONS *
3140 **************************/
3142 if (gmx_mm_any_lt(rsq32,rcutoff2))
3145 r32 = _mm_mul_pd(rsq32,rinv32);
3147 /* EWALD ELECTROSTATICS */
3149 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
3150 ewrt = _mm_mul_pd(r32,ewtabscale);
3151 ewitab = _mm_cvttpd_epi32(ewrt);
3153 eweps = _mm_frcz_pd(ewrt);
3155 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
3157 twoeweps = _mm_add_pd(eweps,eweps);
3158 ewitab = _mm_slli_epi32(ewitab,2);
3159 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
3160 ewtabD = _mm_setzero_pd();
3161 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
3162 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
3163 ewtabFn = _mm_setzero_pd();
3164 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
3165 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
3166 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
3167 velec = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
3168 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
3170 d = _mm_sub_pd(r32,rswitch);
3171 d = _mm_max_pd(d,_mm_setzero_pd());
3172 d2 = _mm_mul_pd(d,d);
3173 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
3175 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
3177 /* Evaluate switch function */
3178 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
3179 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv32,_mm_mul_pd(velec,dsw)) );
3180 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
3184 fscal = _mm_and_pd(fscal,cutoff_mask);
3186 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
3188 /* Update vectorial force */
3189 fix3 = _mm_macc_pd(dx32,fscal,fix3);
3190 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
3191 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
3193 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
3194 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
3195 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
3199 /**************************
3200 * CALCULATE INTERACTIONS *
3201 **************************/
3203 if (gmx_mm_any_lt(rsq33,rcutoff2))
3206 r33 = _mm_mul_pd(rsq33,rinv33);
3208 /* EWALD ELECTROSTATICS */
3210 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
3211 ewrt = _mm_mul_pd(r33,ewtabscale);
3212 ewitab = _mm_cvttpd_epi32(ewrt);
3214 eweps = _mm_frcz_pd(ewrt);
3216 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
3218 twoeweps = _mm_add_pd(eweps,eweps);
3219 ewitab = _mm_slli_epi32(ewitab,2);
3220 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
3221 ewtabD = _mm_setzero_pd();
3222 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
3223 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
3224 ewtabFn = _mm_setzero_pd();
3225 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
3226 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
3227 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
3228 velec = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
3229 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
3231 d = _mm_sub_pd(r33,rswitch);
3232 d = _mm_max_pd(d,_mm_setzero_pd());
3233 d2 = _mm_mul_pd(d,d);
3234 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
3236 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
3238 /* Evaluate switch function */
3239 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
3240 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv33,_mm_mul_pd(velec,dsw)) );
3241 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
3245 fscal = _mm_and_pd(fscal,cutoff_mask);
3247 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
3249 /* Update vectorial force */
3250 fix3 = _mm_macc_pd(dx33,fscal,fix3);
3251 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
3252 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
3254 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
3255 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
3256 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
3260 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
3262 /* Inner loop uses 647 flops */
3265 /* End of innermost loop */
3267 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
3268 f+i_coord_offset,fshift+i_shift_offset);
3270 /* Increment number of inner iterations */
3271 inneriter += j_index_end - j_index_start;
3273 /* Outer loop uses 24 flops */
3276 /* Increment number of outer iterations */
3279 /* Update outer/inner flops */
3281 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*647);