2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW4W4_VF_avx_128_fma_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: None
54 * Geometry: Water4-Water4
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSw_VdwNone_GeomW4W4_VF_avx_128_fma_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
83 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
85 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
86 int vdwjidx1A,vdwjidx1B;
87 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
88 int vdwjidx2A,vdwjidx2B;
89 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
90 int vdwjidx3A,vdwjidx3B;
91 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
92 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
93 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
94 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
95 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
96 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
97 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
98 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
99 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
100 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
101 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
104 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
106 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
107 real rswitch_scalar,d_scalar;
108 __m128d dummy_mask,cutoff_mask;
109 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
110 __m128d one = _mm_set1_pd(1.0);
111 __m128d two = _mm_set1_pd(2.0);
117 jindex = nlist->jindex;
119 shiftidx = nlist->shift;
121 shiftvec = fr->shift_vec[0];
122 fshift = fr->fshift[0];
123 facel = _mm_set1_pd(fr->epsfac);
124 charge = mdatoms->chargeA;
126 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
127 ewtab = fr->ic->tabq_coul_FDV0;
128 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
129 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
134 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
135 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
137 jq1 = _mm_set1_pd(charge[inr+1]);
138 jq2 = _mm_set1_pd(charge[inr+2]);
139 jq3 = _mm_set1_pd(charge[inr+3]);
140 qq11 = _mm_mul_pd(iq1,jq1);
141 qq12 = _mm_mul_pd(iq1,jq2);
142 qq13 = _mm_mul_pd(iq1,jq3);
143 qq21 = _mm_mul_pd(iq2,jq1);
144 qq22 = _mm_mul_pd(iq2,jq2);
145 qq23 = _mm_mul_pd(iq2,jq3);
146 qq31 = _mm_mul_pd(iq3,jq1);
147 qq32 = _mm_mul_pd(iq3,jq2);
148 qq33 = _mm_mul_pd(iq3,jq3);
150 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
151 rcutoff_scalar = fr->rcoulomb;
152 rcutoff = _mm_set1_pd(rcutoff_scalar);
153 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
155 rswitch_scalar = fr->rcoulomb_switch;
156 rswitch = _mm_set1_pd(rswitch_scalar);
157 /* Setup switch parameters */
158 d_scalar = rcutoff_scalar-rswitch_scalar;
159 d = _mm_set1_pd(d_scalar);
160 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
161 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
162 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
163 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
164 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
165 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
167 /* Avoid stupid compiler warnings */
175 /* Start outer loop over neighborlists */
176 for(iidx=0; iidx<nri; iidx++)
178 /* Load shift vector for this list */
179 i_shift_offset = DIM*shiftidx[iidx];
181 /* Load limits for loop over neighbors */
182 j_index_start = jindex[iidx];
183 j_index_end = jindex[iidx+1];
185 /* Get outer coordinate index */
187 i_coord_offset = DIM*inr;
189 /* Load i particle coords and add shift vector */
190 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
191 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
193 fix1 = _mm_setzero_pd();
194 fiy1 = _mm_setzero_pd();
195 fiz1 = _mm_setzero_pd();
196 fix2 = _mm_setzero_pd();
197 fiy2 = _mm_setzero_pd();
198 fiz2 = _mm_setzero_pd();
199 fix3 = _mm_setzero_pd();
200 fiy3 = _mm_setzero_pd();
201 fiz3 = _mm_setzero_pd();
203 /* Reset potential sums */
204 velecsum = _mm_setzero_pd();
206 /* Start inner kernel loop */
207 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
210 /* Get j neighbor index, and coordinate index */
213 j_coord_offsetA = DIM*jnrA;
214 j_coord_offsetB = DIM*jnrB;
216 /* load j atom coordinates */
217 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
218 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
220 /* Calculate displacement vector */
221 dx11 = _mm_sub_pd(ix1,jx1);
222 dy11 = _mm_sub_pd(iy1,jy1);
223 dz11 = _mm_sub_pd(iz1,jz1);
224 dx12 = _mm_sub_pd(ix1,jx2);
225 dy12 = _mm_sub_pd(iy1,jy2);
226 dz12 = _mm_sub_pd(iz1,jz2);
227 dx13 = _mm_sub_pd(ix1,jx3);
228 dy13 = _mm_sub_pd(iy1,jy3);
229 dz13 = _mm_sub_pd(iz1,jz3);
230 dx21 = _mm_sub_pd(ix2,jx1);
231 dy21 = _mm_sub_pd(iy2,jy1);
232 dz21 = _mm_sub_pd(iz2,jz1);
233 dx22 = _mm_sub_pd(ix2,jx2);
234 dy22 = _mm_sub_pd(iy2,jy2);
235 dz22 = _mm_sub_pd(iz2,jz2);
236 dx23 = _mm_sub_pd(ix2,jx3);
237 dy23 = _mm_sub_pd(iy2,jy3);
238 dz23 = _mm_sub_pd(iz2,jz3);
239 dx31 = _mm_sub_pd(ix3,jx1);
240 dy31 = _mm_sub_pd(iy3,jy1);
241 dz31 = _mm_sub_pd(iz3,jz1);
242 dx32 = _mm_sub_pd(ix3,jx2);
243 dy32 = _mm_sub_pd(iy3,jy2);
244 dz32 = _mm_sub_pd(iz3,jz2);
245 dx33 = _mm_sub_pd(ix3,jx3);
246 dy33 = _mm_sub_pd(iy3,jy3);
247 dz33 = _mm_sub_pd(iz3,jz3);
249 /* Calculate squared distance and things based on it */
250 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
251 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
252 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
253 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
254 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
255 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
256 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
257 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
258 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
260 rinv11 = gmx_mm_invsqrt_pd(rsq11);
261 rinv12 = gmx_mm_invsqrt_pd(rsq12);
262 rinv13 = gmx_mm_invsqrt_pd(rsq13);
263 rinv21 = gmx_mm_invsqrt_pd(rsq21);
264 rinv22 = gmx_mm_invsqrt_pd(rsq22);
265 rinv23 = gmx_mm_invsqrt_pd(rsq23);
266 rinv31 = gmx_mm_invsqrt_pd(rsq31);
267 rinv32 = gmx_mm_invsqrt_pd(rsq32);
268 rinv33 = gmx_mm_invsqrt_pd(rsq33);
270 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
271 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
272 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
273 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
274 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
275 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
276 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
277 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
278 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
280 fjx1 = _mm_setzero_pd();
281 fjy1 = _mm_setzero_pd();
282 fjz1 = _mm_setzero_pd();
283 fjx2 = _mm_setzero_pd();
284 fjy2 = _mm_setzero_pd();
285 fjz2 = _mm_setzero_pd();
286 fjx3 = _mm_setzero_pd();
287 fjy3 = _mm_setzero_pd();
288 fjz3 = _mm_setzero_pd();
290 /**************************
291 * CALCULATE INTERACTIONS *
292 **************************/
294 if (gmx_mm_any_lt(rsq11,rcutoff2))
297 r11 = _mm_mul_pd(rsq11,rinv11);
299 /* EWALD ELECTROSTATICS */
301 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
302 ewrt = _mm_mul_pd(r11,ewtabscale);
303 ewitab = _mm_cvttpd_epi32(ewrt);
305 eweps = _mm_frcz_pd(ewrt);
307 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
309 twoeweps = _mm_add_pd(eweps,eweps);
310 ewitab = _mm_slli_epi32(ewitab,2);
311 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
312 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
313 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
314 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
315 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
316 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
317 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
318 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
319 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
320 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
322 d = _mm_sub_pd(r11,rswitch);
323 d = _mm_max_pd(d,_mm_setzero_pd());
324 d2 = _mm_mul_pd(d,d);
325 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
327 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
329 /* Evaluate switch function */
330 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
331 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
332 velec = _mm_mul_pd(velec,sw);
333 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
335 /* Update potential sum for this i atom from the interaction with this j atom. */
336 velec = _mm_and_pd(velec,cutoff_mask);
337 velecsum = _mm_add_pd(velecsum,velec);
341 fscal = _mm_and_pd(fscal,cutoff_mask);
343 /* Update vectorial force */
344 fix1 = _mm_macc_pd(dx11,fscal,fix1);
345 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
346 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
348 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
349 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
350 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
354 /**************************
355 * CALCULATE INTERACTIONS *
356 **************************/
358 if (gmx_mm_any_lt(rsq12,rcutoff2))
361 r12 = _mm_mul_pd(rsq12,rinv12);
363 /* EWALD ELECTROSTATICS */
365 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
366 ewrt = _mm_mul_pd(r12,ewtabscale);
367 ewitab = _mm_cvttpd_epi32(ewrt);
369 eweps = _mm_frcz_pd(ewrt);
371 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
373 twoeweps = _mm_add_pd(eweps,eweps);
374 ewitab = _mm_slli_epi32(ewitab,2);
375 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
376 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
377 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
378 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
379 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
380 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
381 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
382 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
383 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
384 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
386 d = _mm_sub_pd(r12,rswitch);
387 d = _mm_max_pd(d,_mm_setzero_pd());
388 d2 = _mm_mul_pd(d,d);
389 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
391 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
393 /* Evaluate switch function */
394 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
395 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
396 velec = _mm_mul_pd(velec,sw);
397 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
399 /* Update potential sum for this i atom from the interaction with this j atom. */
400 velec = _mm_and_pd(velec,cutoff_mask);
401 velecsum = _mm_add_pd(velecsum,velec);
405 fscal = _mm_and_pd(fscal,cutoff_mask);
407 /* Update vectorial force */
408 fix1 = _mm_macc_pd(dx12,fscal,fix1);
409 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
410 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
412 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
413 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
414 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
418 /**************************
419 * CALCULATE INTERACTIONS *
420 **************************/
422 if (gmx_mm_any_lt(rsq13,rcutoff2))
425 r13 = _mm_mul_pd(rsq13,rinv13);
427 /* EWALD ELECTROSTATICS */
429 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
430 ewrt = _mm_mul_pd(r13,ewtabscale);
431 ewitab = _mm_cvttpd_epi32(ewrt);
433 eweps = _mm_frcz_pd(ewrt);
435 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
437 twoeweps = _mm_add_pd(eweps,eweps);
438 ewitab = _mm_slli_epi32(ewitab,2);
439 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
440 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
441 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
442 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
443 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
444 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
445 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
446 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
447 velec = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
448 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
450 d = _mm_sub_pd(r13,rswitch);
451 d = _mm_max_pd(d,_mm_setzero_pd());
452 d2 = _mm_mul_pd(d,d);
453 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
455 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
457 /* Evaluate switch function */
458 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
459 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv13,_mm_mul_pd(velec,dsw)) );
460 velec = _mm_mul_pd(velec,sw);
461 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
463 /* Update potential sum for this i atom from the interaction with this j atom. */
464 velec = _mm_and_pd(velec,cutoff_mask);
465 velecsum = _mm_add_pd(velecsum,velec);
469 fscal = _mm_and_pd(fscal,cutoff_mask);
471 /* Update vectorial force */
472 fix1 = _mm_macc_pd(dx13,fscal,fix1);
473 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
474 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
476 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
477 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
478 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
482 /**************************
483 * CALCULATE INTERACTIONS *
484 **************************/
486 if (gmx_mm_any_lt(rsq21,rcutoff2))
489 r21 = _mm_mul_pd(rsq21,rinv21);
491 /* EWALD ELECTROSTATICS */
493 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
494 ewrt = _mm_mul_pd(r21,ewtabscale);
495 ewitab = _mm_cvttpd_epi32(ewrt);
497 eweps = _mm_frcz_pd(ewrt);
499 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
501 twoeweps = _mm_add_pd(eweps,eweps);
502 ewitab = _mm_slli_epi32(ewitab,2);
503 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
504 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
505 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
506 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
507 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
508 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
509 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
510 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
511 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
512 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
514 d = _mm_sub_pd(r21,rswitch);
515 d = _mm_max_pd(d,_mm_setzero_pd());
516 d2 = _mm_mul_pd(d,d);
517 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
519 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
521 /* Evaluate switch function */
522 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
523 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
524 velec = _mm_mul_pd(velec,sw);
525 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
527 /* Update potential sum for this i atom from the interaction with this j atom. */
528 velec = _mm_and_pd(velec,cutoff_mask);
529 velecsum = _mm_add_pd(velecsum,velec);
533 fscal = _mm_and_pd(fscal,cutoff_mask);
535 /* Update vectorial force */
536 fix2 = _mm_macc_pd(dx21,fscal,fix2);
537 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
538 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
540 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
541 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
542 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
546 /**************************
547 * CALCULATE INTERACTIONS *
548 **************************/
550 if (gmx_mm_any_lt(rsq22,rcutoff2))
553 r22 = _mm_mul_pd(rsq22,rinv22);
555 /* EWALD ELECTROSTATICS */
557 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
558 ewrt = _mm_mul_pd(r22,ewtabscale);
559 ewitab = _mm_cvttpd_epi32(ewrt);
561 eweps = _mm_frcz_pd(ewrt);
563 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
565 twoeweps = _mm_add_pd(eweps,eweps);
566 ewitab = _mm_slli_epi32(ewitab,2);
567 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
568 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
569 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
570 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
571 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
572 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
573 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
574 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
575 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
576 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
578 d = _mm_sub_pd(r22,rswitch);
579 d = _mm_max_pd(d,_mm_setzero_pd());
580 d2 = _mm_mul_pd(d,d);
581 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
583 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
585 /* Evaluate switch function */
586 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
587 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
588 velec = _mm_mul_pd(velec,sw);
589 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
591 /* Update potential sum for this i atom from the interaction with this j atom. */
592 velec = _mm_and_pd(velec,cutoff_mask);
593 velecsum = _mm_add_pd(velecsum,velec);
597 fscal = _mm_and_pd(fscal,cutoff_mask);
599 /* Update vectorial force */
600 fix2 = _mm_macc_pd(dx22,fscal,fix2);
601 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
602 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
604 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
605 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
606 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
610 /**************************
611 * CALCULATE INTERACTIONS *
612 **************************/
614 if (gmx_mm_any_lt(rsq23,rcutoff2))
617 r23 = _mm_mul_pd(rsq23,rinv23);
619 /* EWALD ELECTROSTATICS */
621 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
622 ewrt = _mm_mul_pd(r23,ewtabscale);
623 ewitab = _mm_cvttpd_epi32(ewrt);
625 eweps = _mm_frcz_pd(ewrt);
627 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
629 twoeweps = _mm_add_pd(eweps,eweps);
630 ewitab = _mm_slli_epi32(ewitab,2);
631 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
632 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
633 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
634 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
635 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
636 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
637 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
638 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
639 velec = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
640 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
642 d = _mm_sub_pd(r23,rswitch);
643 d = _mm_max_pd(d,_mm_setzero_pd());
644 d2 = _mm_mul_pd(d,d);
645 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
647 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
649 /* Evaluate switch function */
650 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
651 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv23,_mm_mul_pd(velec,dsw)) );
652 velec = _mm_mul_pd(velec,sw);
653 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
655 /* Update potential sum for this i atom from the interaction with this j atom. */
656 velec = _mm_and_pd(velec,cutoff_mask);
657 velecsum = _mm_add_pd(velecsum,velec);
661 fscal = _mm_and_pd(fscal,cutoff_mask);
663 /* Update vectorial force */
664 fix2 = _mm_macc_pd(dx23,fscal,fix2);
665 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
666 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
668 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
669 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
670 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
674 /**************************
675 * CALCULATE INTERACTIONS *
676 **************************/
678 if (gmx_mm_any_lt(rsq31,rcutoff2))
681 r31 = _mm_mul_pd(rsq31,rinv31);
683 /* EWALD ELECTROSTATICS */
685 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
686 ewrt = _mm_mul_pd(r31,ewtabscale);
687 ewitab = _mm_cvttpd_epi32(ewrt);
689 eweps = _mm_frcz_pd(ewrt);
691 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
693 twoeweps = _mm_add_pd(eweps,eweps);
694 ewitab = _mm_slli_epi32(ewitab,2);
695 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
696 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
697 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
698 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
699 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
700 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
701 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
702 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
703 velec = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
704 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
706 d = _mm_sub_pd(r31,rswitch);
707 d = _mm_max_pd(d,_mm_setzero_pd());
708 d2 = _mm_mul_pd(d,d);
709 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
711 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
713 /* Evaluate switch function */
714 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
715 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv31,_mm_mul_pd(velec,dsw)) );
716 velec = _mm_mul_pd(velec,sw);
717 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
719 /* Update potential sum for this i atom from the interaction with this j atom. */
720 velec = _mm_and_pd(velec,cutoff_mask);
721 velecsum = _mm_add_pd(velecsum,velec);
725 fscal = _mm_and_pd(fscal,cutoff_mask);
727 /* Update vectorial force */
728 fix3 = _mm_macc_pd(dx31,fscal,fix3);
729 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
730 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
732 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
733 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
734 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
738 /**************************
739 * CALCULATE INTERACTIONS *
740 **************************/
742 if (gmx_mm_any_lt(rsq32,rcutoff2))
745 r32 = _mm_mul_pd(rsq32,rinv32);
747 /* EWALD ELECTROSTATICS */
749 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
750 ewrt = _mm_mul_pd(r32,ewtabscale);
751 ewitab = _mm_cvttpd_epi32(ewrt);
753 eweps = _mm_frcz_pd(ewrt);
755 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
757 twoeweps = _mm_add_pd(eweps,eweps);
758 ewitab = _mm_slli_epi32(ewitab,2);
759 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
760 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
761 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
762 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
763 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
764 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
765 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
766 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
767 velec = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
768 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
770 d = _mm_sub_pd(r32,rswitch);
771 d = _mm_max_pd(d,_mm_setzero_pd());
772 d2 = _mm_mul_pd(d,d);
773 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
775 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
777 /* Evaluate switch function */
778 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
779 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv32,_mm_mul_pd(velec,dsw)) );
780 velec = _mm_mul_pd(velec,sw);
781 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
783 /* Update potential sum for this i atom from the interaction with this j atom. */
784 velec = _mm_and_pd(velec,cutoff_mask);
785 velecsum = _mm_add_pd(velecsum,velec);
789 fscal = _mm_and_pd(fscal,cutoff_mask);
791 /* Update vectorial force */
792 fix3 = _mm_macc_pd(dx32,fscal,fix3);
793 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
794 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
796 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
797 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
798 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
802 /**************************
803 * CALCULATE INTERACTIONS *
804 **************************/
806 if (gmx_mm_any_lt(rsq33,rcutoff2))
809 r33 = _mm_mul_pd(rsq33,rinv33);
811 /* EWALD ELECTROSTATICS */
813 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
814 ewrt = _mm_mul_pd(r33,ewtabscale);
815 ewitab = _mm_cvttpd_epi32(ewrt);
817 eweps = _mm_frcz_pd(ewrt);
819 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
821 twoeweps = _mm_add_pd(eweps,eweps);
822 ewitab = _mm_slli_epi32(ewitab,2);
823 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
824 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
825 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
826 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
827 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
828 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
829 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
830 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
831 velec = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
832 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
834 d = _mm_sub_pd(r33,rswitch);
835 d = _mm_max_pd(d,_mm_setzero_pd());
836 d2 = _mm_mul_pd(d,d);
837 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
839 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
841 /* Evaluate switch function */
842 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
843 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv33,_mm_mul_pd(velec,dsw)) );
844 velec = _mm_mul_pd(velec,sw);
845 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
847 /* Update potential sum for this i atom from the interaction with this j atom. */
848 velec = _mm_and_pd(velec,cutoff_mask);
849 velecsum = _mm_add_pd(velecsum,velec);
853 fscal = _mm_and_pd(fscal,cutoff_mask);
855 /* Update vectorial force */
856 fix3 = _mm_macc_pd(dx33,fscal,fix3);
857 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
858 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
860 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
861 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
862 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
866 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
868 /* Inner loop uses 612 flops */
875 j_coord_offsetA = DIM*jnrA;
877 /* load j atom coordinates */
878 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
879 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
881 /* Calculate displacement vector */
882 dx11 = _mm_sub_pd(ix1,jx1);
883 dy11 = _mm_sub_pd(iy1,jy1);
884 dz11 = _mm_sub_pd(iz1,jz1);
885 dx12 = _mm_sub_pd(ix1,jx2);
886 dy12 = _mm_sub_pd(iy1,jy2);
887 dz12 = _mm_sub_pd(iz1,jz2);
888 dx13 = _mm_sub_pd(ix1,jx3);
889 dy13 = _mm_sub_pd(iy1,jy3);
890 dz13 = _mm_sub_pd(iz1,jz3);
891 dx21 = _mm_sub_pd(ix2,jx1);
892 dy21 = _mm_sub_pd(iy2,jy1);
893 dz21 = _mm_sub_pd(iz2,jz1);
894 dx22 = _mm_sub_pd(ix2,jx2);
895 dy22 = _mm_sub_pd(iy2,jy2);
896 dz22 = _mm_sub_pd(iz2,jz2);
897 dx23 = _mm_sub_pd(ix2,jx3);
898 dy23 = _mm_sub_pd(iy2,jy3);
899 dz23 = _mm_sub_pd(iz2,jz3);
900 dx31 = _mm_sub_pd(ix3,jx1);
901 dy31 = _mm_sub_pd(iy3,jy1);
902 dz31 = _mm_sub_pd(iz3,jz1);
903 dx32 = _mm_sub_pd(ix3,jx2);
904 dy32 = _mm_sub_pd(iy3,jy2);
905 dz32 = _mm_sub_pd(iz3,jz2);
906 dx33 = _mm_sub_pd(ix3,jx3);
907 dy33 = _mm_sub_pd(iy3,jy3);
908 dz33 = _mm_sub_pd(iz3,jz3);
910 /* Calculate squared distance and things based on it */
911 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
912 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
913 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
914 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
915 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
916 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
917 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
918 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
919 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
921 rinv11 = gmx_mm_invsqrt_pd(rsq11);
922 rinv12 = gmx_mm_invsqrt_pd(rsq12);
923 rinv13 = gmx_mm_invsqrt_pd(rsq13);
924 rinv21 = gmx_mm_invsqrt_pd(rsq21);
925 rinv22 = gmx_mm_invsqrt_pd(rsq22);
926 rinv23 = gmx_mm_invsqrt_pd(rsq23);
927 rinv31 = gmx_mm_invsqrt_pd(rsq31);
928 rinv32 = gmx_mm_invsqrt_pd(rsq32);
929 rinv33 = gmx_mm_invsqrt_pd(rsq33);
931 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
932 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
933 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
934 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
935 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
936 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
937 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
938 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
939 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
941 fjx1 = _mm_setzero_pd();
942 fjy1 = _mm_setzero_pd();
943 fjz1 = _mm_setzero_pd();
944 fjx2 = _mm_setzero_pd();
945 fjy2 = _mm_setzero_pd();
946 fjz2 = _mm_setzero_pd();
947 fjx3 = _mm_setzero_pd();
948 fjy3 = _mm_setzero_pd();
949 fjz3 = _mm_setzero_pd();
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 if (gmx_mm_any_lt(rsq11,rcutoff2))
958 r11 = _mm_mul_pd(rsq11,rinv11);
960 /* EWALD ELECTROSTATICS */
962 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
963 ewrt = _mm_mul_pd(r11,ewtabscale);
964 ewitab = _mm_cvttpd_epi32(ewrt);
966 eweps = _mm_frcz_pd(ewrt);
968 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
970 twoeweps = _mm_add_pd(eweps,eweps);
971 ewitab = _mm_slli_epi32(ewitab,2);
972 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
973 ewtabD = _mm_setzero_pd();
974 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
975 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
976 ewtabFn = _mm_setzero_pd();
977 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
978 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
979 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
980 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
981 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
983 d = _mm_sub_pd(r11,rswitch);
984 d = _mm_max_pd(d,_mm_setzero_pd());
985 d2 = _mm_mul_pd(d,d);
986 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
988 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
990 /* Evaluate switch function */
991 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
992 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
993 velec = _mm_mul_pd(velec,sw);
994 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
996 /* Update potential sum for this i atom from the interaction with this j atom. */
997 velec = _mm_and_pd(velec,cutoff_mask);
998 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
999 velecsum = _mm_add_pd(velecsum,velec);
1003 fscal = _mm_and_pd(fscal,cutoff_mask);
1005 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1007 /* Update vectorial force */
1008 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1009 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1010 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1012 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1013 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1014 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1018 /**************************
1019 * CALCULATE INTERACTIONS *
1020 **************************/
1022 if (gmx_mm_any_lt(rsq12,rcutoff2))
1025 r12 = _mm_mul_pd(rsq12,rinv12);
1027 /* EWALD ELECTROSTATICS */
1029 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1030 ewrt = _mm_mul_pd(r12,ewtabscale);
1031 ewitab = _mm_cvttpd_epi32(ewrt);
1033 eweps = _mm_frcz_pd(ewrt);
1035 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1037 twoeweps = _mm_add_pd(eweps,eweps);
1038 ewitab = _mm_slli_epi32(ewitab,2);
1039 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1040 ewtabD = _mm_setzero_pd();
1041 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1042 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1043 ewtabFn = _mm_setzero_pd();
1044 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1045 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1046 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1047 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1048 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1050 d = _mm_sub_pd(r12,rswitch);
1051 d = _mm_max_pd(d,_mm_setzero_pd());
1052 d2 = _mm_mul_pd(d,d);
1053 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1055 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1057 /* Evaluate switch function */
1058 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1059 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
1060 velec = _mm_mul_pd(velec,sw);
1061 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1063 /* Update potential sum for this i atom from the interaction with this j atom. */
1064 velec = _mm_and_pd(velec,cutoff_mask);
1065 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1066 velecsum = _mm_add_pd(velecsum,velec);
1070 fscal = _mm_and_pd(fscal,cutoff_mask);
1072 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1074 /* Update vectorial force */
1075 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1076 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1077 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1079 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1080 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1081 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1085 /**************************
1086 * CALCULATE INTERACTIONS *
1087 **************************/
1089 if (gmx_mm_any_lt(rsq13,rcutoff2))
1092 r13 = _mm_mul_pd(rsq13,rinv13);
1094 /* EWALD ELECTROSTATICS */
1096 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1097 ewrt = _mm_mul_pd(r13,ewtabscale);
1098 ewitab = _mm_cvttpd_epi32(ewrt);
1100 eweps = _mm_frcz_pd(ewrt);
1102 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1104 twoeweps = _mm_add_pd(eweps,eweps);
1105 ewitab = _mm_slli_epi32(ewitab,2);
1106 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1107 ewtabD = _mm_setzero_pd();
1108 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1109 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1110 ewtabFn = _mm_setzero_pd();
1111 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1112 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1113 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1114 velec = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
1115 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1117 d = _mm_sub_pd(r13,rswitch);
1118 d = _mm_max_pd(d,_mm_setzero_pd());
1119 d2 = _mm_mul_pd(d,d);
1120 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1122 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1124 /* Evaluate switch function */
1125 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1126 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv13,_mm_mul_pd(velec,dsw)) );
1127 velec = _mm_mul_pd(velec,sw);
1128 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
1130 /* Update potential sum for this i atom from the interaction with this j atom. */
1131 velec = _mm_and_pd(velec,cutoff_mask);
1132 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1133 velecsum = _mm_add_pd(velecsum,velec);
1137 fscal = _mm_and_pd(fscal,cutoff_mask);
1139 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1141 /* Update vectorial force */
1142 fix1 = _mm_macc_pd(dx13,fscal,fix1);
1143 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
1144 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1146 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1147 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1148 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1152 /**************************
1153 * CALCULATE INTERACTIONS *
1154 **************************/
1156 if (gmx_mm_any_lt(rsq21,rcutoff2))
1159 r21 = _mm_mul_pd(rsq21,rinv21);
1161 /* EWALD ELECTROSTATICS */
1163 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1164 ewrt = _mm_mul_pd(r21,ewtabscale);
1165 ewitab = _mm_cvttpd_epi32(ewrt);
1167 eweps = _mm_frcz_pd(ewrt);
1169 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1171 twoeweps = _mm_add_pd(eweps,eweps);
1172 ewitab = _mm_slli_epi32(ewitab,2);
1173 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1174 ewtabD = _mm_setzero_pd();
1175 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1176 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1177 ewtabFn = _mm_setzero_pd();
1178 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1179 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1180 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1181 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1182 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1184 d = _mm_sub_pd(r21,rswitch);
1185 d = _mm_max_pd(d,_mm_setzero_pd());
1186 d2 = _mm_mul_pd(d,d);
1187 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1189 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1191 /* Evaluate switch function */
1192 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1193 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
1194 velec = _mm_mul_pd(velec,sw);
1195 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
1197 /* Update potential sum for this i atom from the interaction with this j atom. */
1198 velec = _mm_and_pd(velec,cutoff_mask);
1199 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1200 velecsum = _mm_add_pd(velecsum,velec);
1204 fscal = _mm_and_pd(fscal,cutoff_mask);
1206 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1208 /* Update vectorial force */
1209 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1210 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1211 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1213 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1214 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1215 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1219 /**************************
1220 * CALCULATE INTERACTIONS *
1221 **************************/
1223 if (gmx_mm_any_lt(rsq22,rcutoff2))
1226 r22 = _mm_mul_pd(rsq22,rinv22);
1228 /* EWALD ELECTROSTATICS */
1230 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1231 ewrt = _mm_mul_pd(r22,ewtabscale);
1232 ewitab = _mm_cvttpd_epi32(ewrt);
1234 eweps = _mm_frcz_pd(ewrt);
1236 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1238 twoeweps = _mm_add_pd(eweps,eweps);
1239 ewitab = _mm_slli_epi32(ewitab,2);
1240 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1241 ewtabD = _mm_setzero_pd();
1242 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1243 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1244 ewtabFn = _mm_setzero_pd();
1245 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1246 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1247 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1248 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1249 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1251 d = _mm_sub_pd(r22,rswitch);
1252 d = _mm_max_pd(d,_mm_setzero_pd());
1253 d2 = _mm_mul_pd(d,d);
1254 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1256 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1258 /* Evaluate switch function */
1259 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1260 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
1261 velec = _mm_mul_pd(velec,sw);
1262 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
1264 /* Update potential sum for this i atom from the interaction with this j atom. */
1265 velec = _mm_and_pd(velec,cutoff_mask);
1266 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1267 velecsum = _mm_add_pd(velecsum,velec);
1271 fscal = _mm_and_pd(fscal,cutoff_mask);
1273 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1275 /* Update vectorial force */
1276 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1277 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1278 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1280 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1281 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1282 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1286 /**************************
1287 * CALCULATE INTERACTIONS *
1288 **************************/
1290 if (gmx_mm_any_lt(rsq23,rcutoff2))
1293 r23 = _mm_mul_pd(rsq23,rinv23);
1295 /* EWALD ELECTROSTATICS */
1297 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1298 ewrt = _mm_mul_pd(r23,ewtabscale);
1299 ewitab = _mm_cvttpd_epi32(ewrt);
1301 eweps = _mm_frcz_pd(ewrt);
1303 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1305 twoeweps = _mm_add_pd(eweps,eweps);
1306 ewitab = _mm_slli_epi32(ewitab,2);
1307 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1308 ewtabD = _mm_setzero_pd();
1309 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1310 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1311 ewtabFn = _mm_setzero_pd();
1312 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1313 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1314 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1315 velec = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
1316 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1318 d = _mm_sub_pd(r23,rswitch);
1319 d = _mm_max_pd(d,_mm_setzero_pd());
1320 d2 = _mm_mul_pd(d,d);
1321 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1323 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1325 /* Evaluate switch function */
1326 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1327 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv23,_mm_mul_pd(velec,dsw)) );
1328 velec = _mm_mul_pd(velec,sw);
1329 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
1331 /* Update potential sum for this i atom from the interaction with this j atom. */
1332 velec = _mm_and_pd(velec,cutoff_mask);
1333 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1334 velecsum = _mm_add_pd(velecsum,velec);
1338 fscal = _mm_and_pd(fscal,cutoff_mask);
1340 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1342 /* Update vectorial force */
1343 fix2 = _mm_macc_pd(dx23,fscal,fix2);
1344 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
1345 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
1347 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
1348 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
1349 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
1353 /**************************
1354 * CALCULATE INTERACTIONS *
1355 **************************/
1357 if (gmx_mm_any_lt(rsq31,rcutoff2))
1360 r31 = _mm_mul_pd(rsq31,rinv31);
1362 /* EWALD ELECTROSTATICS */
1364 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1365 ewrt = _mm_mul_pd(r31,ewtabscale);
1366 ewitab = _mm_cvttpd_epi32(ewrt);
1368 eweps = _mm_frcz_pd(ewrt);
1370 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1372 twoeweps = _mm_add_pd(eweps,eweps);
1373 ewitab = _mm_slli_epi32(ewitab,2);
1374 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1375 ewtabD = _mm_setzero_pd();
1376 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1377 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1378 ewtabFn = _mm_setzero_pd();
1379 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1380 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1381 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1382 velec = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
1383 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1385 d = _mm_sub_pd(r31,rswitch);
1386 d = _mm_max_pd(d,_mm_setzero_pd());
1387 d2 = _mm_mul_pd(d,d);
1388 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1390 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1392 /* Evaluate switch function */
1393 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1394 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv31,_mm_mul_pd(velec,dsw)) );
1395 velec = _mm_mul_pd(velec,sw);
1396 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
1398 /* Update potential sum for this i atom from the interaction with this j atom. */
1399 velec = _mm_and_pd(velec,cutoff_mask);
1400 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1401 velecsum = _mm_add_pd(velecsum,velec);
1405 fscal = _mm_and_pd(fscal,cutoff_mask);
1407 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1409 /* Update vectorial force */
1410 fix3 = _mm_macc_pd(dx31,fscal,fix3);
1411 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
1412 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
1414 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
1415 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
1416 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
1420 /**************************
1421 * CALCULATE INTERACTIONS *
1422 **************************/
1424 if (gmx_mm_any_lt(rsq32,rcutoff2))
1427 r32 = _mm_mul_pd(rsq32,rinv32);
1429 /* EWALD ELECTROSTATICS */
1431 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1432 ewrt = _mm_mul_pd(r32,ewtabscale);
1433 ewitab = _mm_cvttpd_epi32(ewrt);
1435 eweps = _mm_frcz_pd(ewrt);
1437 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1439 twoeweps = _mm_add_pd(eweps,eweps);
1440 ewitab = _mm_slli_epi32(ewitab,2);
1441 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1442 ewtabD = _mm_setzero_pd();
1443 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1444 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1445 ewtabFn = _mm_setzero_pd();
1446 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1447 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1448 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1449 velec = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
1450 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1452 d = _mm_sub_pd(r32,rswitch);
1453 d = _mm_max_pd(d,_mm_setzero_pd());
1454 d2 = _mm_mul_pd(d,d);
1455 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1457 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1459 /* Evaluate switch function */
1460 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1461 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv32,_mm_mul_pd(velec,dsw)) );
1462 velec = _mm_mul_pd(velec,sw);
1463 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
1465 /* Update potential sum for this i atom from the interaction with this j atom. */
1466 velec = _mm_and_pd(velec,cutoff_mask);
1467 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1468 velecsum = _mm_add_pd(velecsum,velec);
1472 fscal = _mm_and_pd(fscal,cutoff_mask);
1474 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1476 /* Update vectorial force */
1477 fix3 = _mm_macc_pd(dx32,fscal,fix3);
1478 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
1479 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
1481 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
1482 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
1483 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
1487 /**************************
1488 * CALCULATE INTERACTIONS *
1489 **************************/
1491 if (gmx_mm_any_lt(rsq33,rcutoff2))
1494 r33 = _mm_mul_pd(rsq33,rinv33);
1496 /* EWALD ELECTROSTATICS */
1498 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1499 ewrt = _mm_mul_pd(r33,ewtabscale);
1500 ewitab = _mm_cvttpd_epi32(ewrt);
1502 eweps = _mm_frcz_pd(ewrt);
1504 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1506 twoeweps = _mm_add_pd(eweps,eweps);
1507 ewitab = _mm_slli_epi32(ewitab,2);
1508 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1509 ewtabD = _mm_setzero_pd();
1510 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1511 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1512 ewtabFn = _mm_setzero_pd();
1513 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1514 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1515 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1516 velec = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
1517 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1519 d = _mm_sub_pd(r33,rswitch);
1520 d = _mm_max_pd(d,_mm_setzero_pd());
1521 d2 = _mm_mul_pd(d,d);
1522 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1524 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1526 /* Evaluate switch function */
1527 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1528 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv33,_mm_mul_pd(velec,dsw)) );
1529 velec = _mm_mul_pd(velec,sw);
1530 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
1532 /* Update potential sum for this i atom from the interaction with this j atom. */
1533 velec = _mm_and_pd(velec,cutoff_mask);
1534 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1535 velecsum = _mm_add_pd(velecsum,velec);
1539 fscal = _mm_and_pd(fscal,cutoff_mask);
1541 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1543 /* Update vectorial force */
1544 fix3 = _mm_macc_pd(dx33,fscal,fix3);
1545 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
1546 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
1548 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
1549 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
1550 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
1554 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1556 /* Inner loop uses 612 flops */
1559 /* End of innermost loop */
1561 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1562 f+i_coord_offset+DIM,fshift+i_shift_offset);
1565 /* Update potential energies */
1566 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1568 /* Increment number of inner iterations */
1569 inneriter += j_index_end - j_index_start;
1571 /* Outer loop uses 19 flops */
1574 /* Increment number of outer iterations */
1577 /* Update outer/inner flops */
1579 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*612);
1582 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW4W4_F_avx_128_fma_double
1583 * Electrostatics interaction: Ewald
1584 * VdW interaction: None
1585 * Geometry: Water4-Water4
1586 * Calculate force/pot: Force
1589 nb_kernel_ElecEwSw_VdwNone_GeomW4W4_F_avx_128_fma_double
1590 (t_nblist * gmx_restrict nlist,
1591 rvec * gmx_restrict xx,
1592 rvec * gmx_restrict ff,
1593 t_forcerec * gmx_restrict fr,
1594 t_mdatoms * gmx_restrict mdatoms,
1595 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1596 t_nrnb * gmx_restrict nrnb)
1598 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1599 * just 0 for non-waters.
1600 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1601 * jnr indices corresponding to data put in the four positions in the SIMD register.
1603 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1604 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1606 int j_coord_offsetA,j_coord_offsetB;
1607 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1608 real rcutoff_scalar;
1609 real *shiftvec,*fshift,*x,*f;
1610 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1612 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1614 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1616 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1617 int vdwjidx1A,vdwjidx1B;
1618 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1619 int vdwjidx2A,vdwjidx2B;
1620 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1621 int vdwjidx3A,vdwjidx3B;
1622 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1623 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1624 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1625 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1626 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1627 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1628 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1629 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1630 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1631 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1632 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1635 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1637 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1638 real rswitch_scalar,d_scalar;
1639 __m128d dummy_mask,cutoff_mask;
1640 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1641 __m128d one = _mm_set1_pd(1.0);
1642 __m128d two = _mm_set1_pd(2.0);
1648 jindex = nlist->jindex;
1650 shiftidx = nlist->shift;
1652 shiftvec = fr->shift_vec[0];
1653 fshift = fr->fshift[0];
1654 facel = _mm_set1_pd(fr->epsfac);
1655 charge = mdatoms->chargeA;
1657 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
1658 ewtab = fr->ic->tabq_coul_FDV0;
1659 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
1660 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1662 /* Setup water-specific parameters */
1663 inr = nlist->iinr[0];
1664 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1665 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1666 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1668 jq1 = _mm_set1_pd(charge[inr+1]);
1669 jq2 = _mm_set1_pd(charge[inr+2]);
1670 jq3 = _mm_set1_pd(charge[inr+3]);
1671 qq11 = _mm_mul_pd(iq1,jq1);
1672 qq12 = _mm_mul_pd(iq1,jq2);
1673 qq13 = _mm_mul_pd(iq1,jq3);
1674 qq21 = _mm_mul_pd(iq2,jq1);
1675 qq22 = _mm_mul_pd(iq2,jq2);
1676 qq23 = _mm_mul_pd(iq2,jq3);
1677 qq31 = _mm_mul_pd(iq3,jq1);
1678 qq32 = _mm_mul_pd(iq3,jq2);
1679 qq33 = _mm_mul_pd(iq3,jq3);
1681 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1682 rcutoff_scalar = fr->rcoulomb;
1683 rcutoff = _mm_set1_pd(rcutoff_scalar);
1684 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
1686 rswitch_scalar = fr->rcoulomb_switch;
1687 rswitch = _mm_set1_pd(rswitch_scalar);
1688 /* Setup switch parameters */
1689 d_scalar = rcutoff_scalar-rswitch_scalar;
1690 d = _mm_set1_pd(d_scalar);
1691 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
1692 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1693 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1694 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
1695 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1696 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1698 /* Avoid stupid compiler warnings */
1700 j_coord_offsetA = 0;
1701 j_coord_offsetB = 0;
1706 /* Start outer loop over neighborlists */
1707 for(iidx=0; iidx<nri; iidx++)
1709 /* Load shift vector for this list */
1710 i_shift_offset = DIM*shiftidx[iidx];
1712 /* Load limits for loop over neighbors */
1713 j_index_start = jindex[iidx];
1714 j_index_end = jindex[iidx+1];
1716 /* Get outer coordinate index */
1718 i_coord_offset = DIM*inr;
1720 /* Load i particle coords and add shift vector */
1721 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1722 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1724 fix1 = _mm_setzero_pd();
1725 fiy1 = _mm_setzero_pd();
1726 fiz1 = _mm_setzero_pd();
1727 fix2 = _mm_setzero_pd();
1728 fiy2 = _mm_setzero_pd();
1729 fiz2 = _mm_setzero_pd();
1730 fix3 = _mm_setzero_pd();
1731 fiy3 = _mm_setzero_pd();
1732 fiz3 = _mm_setzero_pd();
1734 /* Start inner kernel loop */
1735 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1738 /* Get j neighbor index, and coordinate index */
1740 jnrB = jjnr[jidx+1];
1741 j_coord_offsetA = DIM*jnrA;
1742 j_coord_offsetB = DIM*jnrB;
1744 /* load j atom coordinates */
1745 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1746 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1748 /* Calculate displacement vector */
1749 dx11 = _mm_sub_pd(ix1,jx1);
1750 dy11 = _mm_sub_pd(iy1,jy1);
1751 dz11 = _mm_sub_pd(iz1,jz1);
1752 dx12 = _mm_sub_pd(ix1,jx2);
1753 dy12 = _mm_sub_pd(iy1,jy2);
1754 dz12 = _mm_sub_pd(iz1,jz2);
1755 dx13 = _mm_sub_pd(ix1,jx3);
1756 dy13 = _mm_sub_pd(iy1,jy3);
1757 dz13 = _mm_sub_pd(iz1,jz3);
1758 dx21 = _mm_sub_pd(ix2,jx1);
1759 dy21 = _mm_sub_pd(iy2,jy1);
1760 dz21 = _mm_sub_pd(iz2,jz1);
1761 dx22 = _mm_sub_pd(ix2,jx2);
1762 dy22 = _mm_sub_pd(iy2,jy2);
1763 dz22 = _mm_sub_pd(iz2,jz2);
1764 dx23 = _mm_sub_pd(ix2,jx3);
1765 dy23 = _mm_sub_pd(iy2,jy3);
1766 dz23 = _mm_sub_pd(iz2,jz3);
1767 dx31 = _mm_sub_pd(ix3,jx1);
1768 dy31 = _mm_sub_pd(iy3,jy1);
1769 dz31 = _mm_sub_pd(iz3,jz1);
1770 dx32 = _mm_sub_pd(ix3,jx2);
1771 dy32 = _mm_sub_pd(iy3,jy2);
1772 dz32 = _mm_sub_pd(iz3,jz2);
1773 dx33 = _mm_sub_pd(ix3,jx3);
1774 dy33 = _mm_sub_pd(iy3,jy3);
1775 dz33 = _mm_sub_pd(iz3,jz3);
1777 /* Calculate squared distance and things based on it */
1778 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1779 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1780 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1781 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1782 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1783 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1784 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1785 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1786 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1788 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1789 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1790 rinv13 = gmx_mm_invsqrt_pd(rsq13);
1791 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1792 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1793 rinv23 = gmx_mm_invsqrt_pd(rsq23);
1794 rinv31 = gmx_mm_invsqrt_pd(rsq31);
1795 rinv32 = gmx_mm_invsqrt_pd(rsq32);
1796 rinv33 = gmx_mm_invsqrt_pd(rsq33);
1798 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1799 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1800 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
1801 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1802 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1803 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
1804 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
1805 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
1806 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
1808 fjx1 = _mm_setzero_pd();
1809 fjy1 = _mm_setzero_pd();
1810 fjz1 = _mm_setzero_pd();
1811 fjx2 = _mm_setzero_pd();
1812 fjy2 = _mm_setzero_pd();
1813 fjz2 = _mm_setzero_pd();
1814 fjx3 = _mm_setzero_pd();
1815 fjy3 = _mm_setzero_pd();
1816 fjz3 = _mm_setzero_pd();
1818 /**************************
1819 * CALCULATE INTERACTIONS *
1820 **************************/
1822 if (gmx_mm_any_lt(rsq11,rcutoff2))
1825 r11 = _mm_mul_pd(rsq11,rinv11);
1827 /* EWALD ELECTROSTATICS */
1829 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1830 ewrt = _mm_mul_pd(r11,ewtabscale);
1831 ewitab = _mm_cvttpd_epi32(ewrt);
1833 eweps = _mm_frcz_pd(ewrt);
1835 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1837 twoeweps = _mm_add_pd(eweps,eweps);
1838 ewitab = _mm_slli_epi32(ewitab,2);
1839 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1840 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1841 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1842 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1843 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1844 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1845 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1846 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1847 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
1848 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1850 d = _mm_sub_pd(r11,rswitch);
1851 d = _mm_max_pd(d,_mm_setzero_pd());
1852 d2 = _mm_mul_pd(d,d);
1853 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1855 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1857 /* Evaluate switch function */
1858 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1859 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
1860 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
1864 fscal = _mm_and_pd(fscal,cutoff_mask);
1866 /* Update vectorial force */
1867 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1868 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1869 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1871 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1872 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1873 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1877 /**************************
1878 * CALCULATE INTERACTIONS *
1879 **************************/
1881 if (gmx_mm_any_lt(rsq12,rcutoff2))
1884 r12 = _mm_mul_pd(rsq12,rinv12);
1886 /* EWALD ELECTROSTATICS */
1888 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1889 ewrt = _mm_mul_pd(r12,ewtabscale);
1890 ewitab = _mm_cvttpd_epi32(ewrt);
1892 eweps = _mm_frcz_pd(ewrt);
1894 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1896 twoeweps = _mm_add_pd(eweps,eweps);
1897 ewitab = _mm_slli_epi32(ewitab,2);
1898 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1899 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1900 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1901 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1902 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1903 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1904 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1905 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1906 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1907 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1909 d = _mm_sub_pd(r12,rswitch);
1910 d = _mm_max_pd(d,_mm_setzero_pd());
1911 d2 = _mm_mul_pd(d,d);
1912 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1914 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1916 /* Evaluate switch function */
1917 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1918 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
1919 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
1923 fscal = _mm_and_pd(fscal,cutoff_mask);
1925 /* Update vectorial force */
1926 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1927 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1928 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1930 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1931 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1932 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1936 /**************************
1937 * CALCULATE INTERACTIONS *
1938 **************************/
1940 if (gmx_mm_any_lt(rsq13,rcutoff2))
1943 r13 = _mm_mul_pd(rsq13,rinv13);
1945 /* EWALD ELECTROSTATICS */
1947 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1948 ewrt = _mm_mul_pd(r13,ewtabscale);
1949 ewitab = _mm_cvttpd_epi32(ewrt);
1951 eweps = _mm_frcz_pd(ewrt);
1953 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1955 twoeweps = _mm_add_pd(eweps,eweps);
1956 ewitab = _mm_slli_epi32(ewitab,2);
1957 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1958 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1959 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1960 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1961 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1962 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1963 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1964 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1965 velec = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
1966 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1968 d = _mm_sub_pd(r13,rswitch);
1969 d = _mm_max_pd(d,_mm_setzero_pd());
1970 d2 = _mm_mul_pd(d,d);
1971 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1973 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1975 /* Evaluate switch function */
1976 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1977 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv13,_mm_mul_pd(velec,dsw)) );
1978 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
1982 fscal = _mm_and_pd(fscal,cutoff_mask);
1984 /* Update vectorial force */
1985 fix1 = _mm_macc_pd(dx13,fscal,fix1);
1986 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
1987 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1989 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1990 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1991 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1995 /**************************
1996 * CALCULATE INTERACTIONS *
1997 **************************/
1999 if (gmx_mm_any_lt(rsq21,rcutoff2))
2002 r21 = _mm_mul_pd(rsq21,rinv21);
2004 /* EWALD ELECTROSTATICS */
2006 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2007 ewrt = _mm_mul_pd(r21,ewtabscale);
2008 ewitab = _mm_cvttpd_epi32(ewrt);
2010 eweps = _mm_frcz_pd(ewrt);
2012 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2014 twoeweps = _mm_add_pd(eweps,eweps);
2015 ewitab = _mm_slli_epi32(ewitab,2);
2016 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2017 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2018 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2019 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2020 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2021 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2022 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2023 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2024 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2025 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2027 d = _mm_sub_pd(r21,rswitch);
2028 d = _mm_max_pd(d,_mm_setzero_pd());
2029 d2 = _mm_mul_pd(d,d);
2030 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2032 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2034 /* Evaluate switch function */
2035 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2036 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2037 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2041 fscal = _mm_and_pd(fscal,cutoff_mask);
2043 /* Update vectorial force */
2044 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2045 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2046 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2048 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2049 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2050 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2054 /**************************
2055 * CALCULATE INTERACTIONS *
2056 **************************/
2058 if (gmx_mm_any_lt(rsq22,rcutoff2))
2061 r22 = _mm_mul_pd(rsq22,rinv22);
2063 /* EWALD ELECTROSTATICS */
2065 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2066 ewrt = _mm_mul_pd(r22,ewtabscale);
2067 ewitab = _mm_cvttpd_epi32(ewrt);
2069 eweps = _mm_frcz_pd(ewrt);
2071 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2073 twoeweps = _mm_add_pd(eweps,eweps);
2074 ewitab = _mm_slli_epi32(ewitab,2);
2075 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2076 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2077 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2078 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2079 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2080 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2081 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2082 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2083 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2084 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2086 d = _mm_sub_pd(r22,rswitch);
2087 d = _mm_max_pd(d,_mm_setzero_pd());
2088 d2 = _mm_mul_pd(d,d);
2089 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2091 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2093 /* Evaluate switch function */
2094 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2095 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2096 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2100 fscal = _mm_and_pd(fscal,cutoff_mask);
2102 /* Update vectorial force */
2103 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2104 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2105 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2107 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2108 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2109 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2113 /**************************
2114 * CALCULATE INTERACTIONS *
2115 **************************/
2117 if (gmx_mm_any_lt(rsq23,rcutoff2))
2120 r23 = _mm_mul_pd(rsq23,rinv23);
2122 /* EWALD ELECTROSTATICS */
2124 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2125 ewrt = _mm_mul_pd(r23,ewtabscale);
2126 ewitab = _mm_cvttpd_epi32(ewrt);
2128 eweps = _mm_frcz_pd(ewrt);
2130 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2132 twoeweps = _mm_add_pd(eweps,eweps);
2133 ewitab = _mm_slli_epi32(ewitab,2);
2134 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2135 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2136 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2137 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2138 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2139 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2140 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2141 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2142 velec = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
2143 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2145 d = _mm_sub_pd(r23,rswitch);
2146 d = _mm_max_pd(d,_mm_setzero_pd());
2147 d2 = _mm_mul_pd(d,d);
2148 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2150 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2152 /* Evaluate switch function */
2153 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2154 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv23,_mm_mul_pd(velec,dsw)) );
2155 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
2159 fscal = _mm_and_pd(fscal,cutoff_mask);
2161 /* Update vectorial force */
2162 fix2 = _mm_macc_pd(dx23,fscal,fix2);
2163 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
2164 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
2166 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
2167 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
2168 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
2172 /**************************
2173 * CALCULATE INTERACTIONS *
2174 **************************/
2176 if (gmx_mm_any_lt(rsq31,rcutoff2))
2179 r31 = _mm_mul_pd(rsq31,rinv31);
2181 /* EWALD ELECTROSTATICS */
2183 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2184 ewrt = _mm_mul_pd(r31,ewtabscale);
2185 ewitab = _mm_cvttpd_epi32(ewrt);
2187 eweps = _mm_frcz_pd(ewrt);
2189 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2191 twoeweps = _mm_add_pd(eweps,eweps);
2192 ewitab = _mm_slli_epi32(ewitab,2);
2193 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2194 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2195 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2196 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2197 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2198 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2199 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2200 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2201 velec = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
2202 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2204 d = _mm_sub_pd(r31,rswitch);
2205 d = _mm_max_pd(d,_mm_setzero_pd());
2206 d2 = _mm_mul_pd(d,d);
2207 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2209 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2211 /* Evaluate switch function */
2212 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2213 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv31,_mm_mul_pd(velec,dsw)) );
2214 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
2218 fscal = _mm_and_pd(fscal,cutoff_mask);
2220 /* Update vectorial force */
2221 fix3 = _mm_macc_pd(dx31,fscal,fix3);
2222 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
2223 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
2225 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
2226 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
2227 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
2231 /**************************
2232 * CALCULATE INTERACTIONS *
2233 **************************/
2235 if (gmx_mm_any_lt(rsq32,rcutoff2))
2238 r32 = _mm_mul_pd(rsq32,rinv32);
2240 /* EWALD ELECTROSTATICS */
2242 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2243 ewrt = _mm_mul_pd(r32,ewtabscale);
2244 ewitab = _mm_cvttpd_epi32(ewrt);
2246 eweps = _mm_frcz_pd(ewrt);
2248 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2250 twoeweps = _mm_add_pd(eweps,eweps);
2251 ewitab = _mm_slli_epi32(ewitab,2);
2252 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2253 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2254 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2255 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2256 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2257 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2258 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2259 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2260 velec = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
2261 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2263 d = _mm_sub_pd(r32,rswitch);
2264 d = _mm_max_pd(d,_mm_setzero_pd());
2265 d2 = _mm_mul_pd(d,d);
2266 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2268 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2270 /* Evaluate switch function */
2271 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2272 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv32,_mm_mul_pd(velec,dsw)) );
2273 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
2277 fscal = _mm_and_pd(fscal,cutoff_mask);
2279 /* Update vectorial force */
2280 fix3 = _mm_macc_pd(dx32,fscal,fix3);
2281 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
2282 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
2284 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
2285 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
2286 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
2290 /**************************
2291 * CALCULATE INTERACTIONS *
2292 **************************/
2294 if (gmx_mm_any_lt(rsq33,rcutoff2))
2297 r33 = _mm_mul_pd(rsq33,rinv33);
2299 /* EWALD ELECTROSTATICS */
2301 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2302 ewrt = _mm_mul_pd(r33,ewtabscale);
2303 ewitab = _mm_cvttpd_epi32(ewrt);
2305 eweps = _mm_frcz_pd(ewrt);
2307 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2309 twoeweps = _mm_add_pd(eweps,eweps);
2310 ewitab = _mm_slli_epi32(ewitab,2);
2311 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2312 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2313 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2314 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2315 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2316 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2317 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2318 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2319 velec = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
2320 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2322 d = _mm_sub_pd(r33,rswitch);
2323 d = _mm_max_pd(d,_mm_setzero_pd());
2324 d2 = _mm_mul_pd(d,d);
2325 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2327 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2329 /* Evaluate switch function */
2330 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2331 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv33,_mm_mul_pd(velec,dsw)) );
2332 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
2336 fscal = _mm_and_pd(fscal,cutoff_mask);
2338 /* Update vectorial force */
2339 fix3 = _mm_macc_pd(dx33,fscal,fix3);
2340 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
2341 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
2343 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
2344 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
2345 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
2349 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2351 /* Inner loop uses 585 flops */
2354 if(jidx<j_index_end)
2358 j_coord_offsetA = DIM*jnrA;
2360 /* load j atom coordinates */
2361 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
2362 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
2364 /* Calculate displacement vector */
2365 dx11 = _mm_sub_pd(ix1,jx1);
2366 dy11 = _mm_sub_pd(iy1,jy1);
2367 dz11 = _mm_sub_pd(iz1,jz1);
2368 dx12 = _mm_sub_pd(ix1,jx2);
2369 dy12 = _mm_sub_pd(iy1,jy2);
2370 dz12 = _mm_sub_pd(iz1,jz2);
2371 dx13 = _mm_sub_pd(ix1,jx3);
2372 dy13 = _mm_sub_pd(iy1,jy3);
2373 dz13 = _mm_sub_pd(iz1,jz3);
2374 dx21 = _mm_sub_pd(ix2,jx1);
2375 dy21 = _mm_sub_pd(iy2,jy1);
2376 dz21 = _mm_sub_pd(iz2,jz1);
2377 dx22 = _mm_sub_pd(ix2,jx2);
2378 dy22 = _mm_sub_pd(iy2,jy2);
2379 dz22 = _mm_sub_pd(iz2,jz2);
2380 dx23 = _mm_sub_pd(ix2,jx3);
2381 dy23 = _mm_sub_pd(iy2,jy3);
2382 dz23 = _mm_sub_pd(iz2,jz3);
2383 dx31 = _mm_sub_pd(ix3,jx1);
2384 dy31 = _mm_sub_pd(iy3,jy1);
2385 dz31 = _mm_sub_pd(iz3,jz1);
2386 dx32 = _mm_sub_pd(ix3,jx2);
2387 dy32 = _mm_sub_pd(iy3,jy2);
2388 dz32 = _mm_sub_pd(iz3,jz2);
2389 dx33 = _mm_sub_pd(ix3,jx3);
2390 dy33 = _mm_sub_pd(iy3,jy3);
2391 dz33 = _mm_sub_pd(iz3,jz3);
2393 /* Calculate squared distance and things based on it */
2394 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2395 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2396 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
2397 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2398 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2399 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
2400 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
2401 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
2402 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
2404 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2405 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2406 rinv13 = gmx_mm_invsqrt_pd(rsq13);
2407 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2408 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2409 rinv23 = gmx_mm_invsqrt_pd(rsq23);
2410 rinv31 = gmx_mm_invsqrt_pd(rsq31);
2411 rinv32 = gmx_mm_invsqrt_pd(rsq32);
2412 rinv33 = gmx_mm_invsqrt_pd(rsq33);
2414 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
2415 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
2416 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
2417 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
2418 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
2419 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
2420 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
2421 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
2422 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
2424 fjx1 = _mm_setzero_pd();
2425 fjy1 = _mm_setzero_pd();
2426 fjz1 = _mm_setzero_pd();
2427 fjx2 = _mm_setzero_pd();
2428 fjy2 = _mm_setzero_pd();
2429 fjz2 = _mm_setzero_pd();
2430 fjx3 = _mm_setzero_pd();
2431 fjy3 = _mm_setzero_pd();
2432 fjz3 = _mm_setzero_pd();
2434 /**************************
2435 * CALCULATE INTERACTIONS *
2436 **************************/
2438 if (gmx_mm_any_lt(rsq11,rcutoff2))
2441 r11 = _mm_mul_pd(rsq11,rinv11);
2443 /* EWALD ELECTROSTATICS */
2445 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2446 ewrt = _mm_mul_pd(r11,ewtabscale);
2447 ewitab = _mm_cvttpd_epi32(ewrt);
2449 eweps = _mm_frcz_pd(ewrt);
2451 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2453 twoeweps = _mm_add_pd(eweps,eweps);
2454 ewitab = _mm_slli_epi32(ewitab,2);
2455 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2456 ewtabD = _mm_setzero_pd();
2457 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2458 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2459 ewtabFn = _mm_setzero_pd();
2460 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2461 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2462 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2463 velec = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2464 felec = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2466 d = _mm_sub_pd(r11,rswitch);
2467 d = _mm_max_pd(d,_mm_setzero_pd());
2468 d2 = _mm_mul_pd(d,d);
2469 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2471 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2473 /* Evaluate switch function */
2474 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2475 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2476 cutoff_mask = _mm_cmplt_pd(rsq11,rcutoff2);
2480 fscal = _mm_and_pd(fscal,cutoff_mask);
2482 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2484 /* Update vectorial force */
2485 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2486 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2487 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2489 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2490 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2491 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2495 /**************************
2496 * CALCULATE INTERACTIONS *
2497 **************************/
2499 if (gmx_mm_any_lt(rsq12,rcutoff2))
2502 r12 = _mm_mul_pd(rsq12,rinv12);
2504 /* EWALD ELECTROSTATICS */
2506 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2507 ewrt = _mm_mul_pd(r12,ewtabscale);
2508 ewitab = _mm_cvttpd_epi32(ewrt);
2510 eweps = _mm_frcz_pd(ewrt);
2512 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2514 twoeweps = _mm_add_pd(eweps,eweps);
2515 ewitab = _mm_slli_epi32(ewitab,2);
2516 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2517 ewtabD = _mm_setzero_pd();
2518 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2519 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2520 ewtabFn = _mm_setzero_pd();
2521 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2522 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2523 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2524 velec = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2525 felec = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2527 d = _mm_sub_pd(r12,rswitch);
2528 d = _mm_max_pd(d,_mm_setzero_pd());
2529 d2 = _mm_mul_pd(d,d);
2530 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2532 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2534 /* Evaluate switch function */
2535 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2536 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2537 cutoff_mask = _mm_cmplt_pd(rsq12,rcutoff2);
2541 fscal = _mm_and_pd(fscal,cutoff_mask);
2543 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2545 /* Update vectorial force */
2546 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2547 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2548 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2550 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2551 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2552 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2556 /**************************
2557 * CALCULATE INTERACTIONS *
2558 **************************/
2560 if (gmx_mm_any_lt(rsq13,rcutoff2))
2563 r13 = _mm_mul_pd(rsq13,rinv13);
2565 /* EWALD ELECTROSTATICS */
2567 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2568 ewrt = _mm_mul_pd(r13,ewtabscale);
2569 ewitab = _mm_cvttpd_epi32(ewrt);
2571 eweps = _mm_frcz_pd(ewrt);
2573 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2575 twoeweps = _mm_add_pd(eweps,eweps);
2576 ewitab = _mm_slli_epi32(ewitab,2);
2577 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2578 ewtabD = _mm_setzero_pd();
2579 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2580 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2581 ewtabFn = _mm_setzero_pd();
2582 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2583 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2584 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2585 velec = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
2586 felec = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2588 d = _mm_sub_pd(r13,rswitch);
2589 d = _mm_max_pd(d,_mm_setzero_pd());
2590 d2 = _mm_mul_pd(d,d);
2591 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2593 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2595 /* Evaluate switch function */
2596 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2597 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv13,_mm_mul_pd(velec,dsw)) );
2598 cutoff_mask = _mm_cmplt_pd(rsq13,rcutoff2);
2602 fscal = _mm_and_pd(fscal,cutoff_mask);
2604 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2606 /* Update vectorial force */
2607 fix1 = _mm_macc_pd(dx13,fscal,fix1);
2608 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
2609 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
2611 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
2612 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
2613 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
2617 /**************************
2618 * CALCULATE INTERACTIONS *
2619 **************************/
2621 if (gmx_mm_any_lt(rsq21,rcutoff2))
2624 r21 = _mm_mul_pd(rsq21,rinv21);
2626 /* EWALD ELECTROSTATICS */
2628 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2629 ewrt = _mm_mul_pd(r21,ewtabscale);
2630 ewitab = _mm_cvttpd_epi32(ewrt);
2632 eweps = _mm_frcz_pd(ewrt);
2634 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2636 twoeweps = _mm_add_pd(eweps,eweps);
2637 ewitab = _mm_slli_epi32(ewitab,2);
2638 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2639 ewtabD = _mm_setzero_pd();
2640 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2641 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2642 ewtabFn = _mm_setzero_pd();
2643 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2644 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2645 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2646 velec = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2647 felec = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2649 d = _mm_sub_pd(r21,rswitch);
2650 d = _mm_max_pd(d,_mm_setzero_pd());
2651 d2 = _mm_mul_pd(d,d);
2652 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2654 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2656 /* Evaluate switch function */
2657 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2658 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2659 cutoff_mask = _mm_cmplt_pd(rsq21,rcutoff2);
2663 fscal = _mm_and_pd(fscal,cutoff_mask);
2665 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2667 /* Update vectorial force */
2668 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2669 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2670 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2672 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2673 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2674 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2678 /**************************
2679 * CALCULATE INTERACTIONS *
2680 **************************/
2682 if (gmx_mm_any_lt(rsq22,rcutoff2))
2685 r22 = _mm_mul_pd(rsq22,rinv22);
2687 /* EWALD ELECTROSTATICS */
2689 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2690 ewrt = _mm_mul_pd(r22,ewtabscale);
2691 ewitab = _mm_cvttpd_epi32(ewrt);
2693 eweps = _mm_frcz_pd(ewrt);
2695 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2697 twoeweps = _mm_add_pd(eweps,eweps);
2698 ewitab = _mm_slli_epi32(ewitab,2);
2699 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2700 ewtabD = _mm_setzero_pd();
2701 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2702 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2703 ewtabFn = _mm_setzero_pd();
2704 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2705 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2706 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2707 velec = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2708 felec = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2710 d = _mm_sub_pd(r22,rswitch);
2711 d = _mm_max_pd(d,_mm_setzero_pd());
2712 d2 = _mm_mul_pd(d,d);
2713 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2715 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2717 /* Evaluate switch function */
2718 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2719 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2720 cutoff_mask = _mm_cmplt_pd(rsq22,rcutoff2);
2724 fscal = _mm_and_pd(fscal,cutoff_mask);
2726 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2728 /* Update vectorial force */
2729 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2730 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2731 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2733 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2734 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2735 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2739 /**************************
2740 * CALCULATE INTERACTIONS *
2741 **************************/
2743 if (gmx_mm_any_lt(rsq23,rcutoff2))
2746 r23 = _mm_mul_pd(rsq23,rinv23);
2748 /* EWALD ELECTROSTATICS */
2750 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2751 ewrt = _mm_mul_pd(r23,ewtabscale);
2752 ewitab = _mm_cvttpd_epi32(ewrt);
2754 eweps = _mm_frcz_pd(ewrt);
2756 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2758 twoeweps = _mm_add_pd(eweps,eweps);
2759 ewitab = _mm_slli_epi32(ewitab,2);
2760 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2761 ewtabD = _mm_setzero_pd();
2762 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2763 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2764 ewtabFn = _mm_setzero_pd();
2765 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2766 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2767 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2768 velec = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
2769 felec = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2771 d = _mm_sub_pd(r23,rswitch);
2772 d = _mm_max_pd(d,_mm_setzero_pd());
2773 d2 = _mm_mul_pd(d,d);
2774 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2776 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2778 /* Evaluate switch function */
2779 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2780 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv23,_mm_mul_pd(velec,dsw)) );
2781 cutoff_mask = _mm_cmplt_pd(rsq23,rcutoff2);
2785 fscal = _mm_and_pd(fscal,cutoff_mask);
2787 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2789 /* Update vectorial force */
2790 fix2 = _mm_macc_pd(dx23,fscal,fix2);
2791 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
2792 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
2794 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
2795 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
2796 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
2800 /**************************
2801 * CALCULATE INTERACTIONS *
2802 **************************/
2804 if (gmx_mm_any_lt(rsq31,rcutoff2))
2807 r31 = _mm_mul_pd(rsq31,rinv31);
2809 /* EWALD ELECTROSTATICS */
2811 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2812 ewrt = _mm_mul_pd(r31,ewtabscale);
2813 ewitab = _mm_cvttpd_epi32(ewrt);
2815 eweps = _mm_frcz_pd(ewrt);
2817 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2819 twoeweps = _mm_add_pd(eweps,eweps);
2820 ewitab = _mm_slli_epi32(ewitab,2);
2821 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2822 ewtabD = _mm_setzero_pd();
2823 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2824 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2825 ewtabFn = _mm_setzero_pd();
2826 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2827 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2828 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2829 velec = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
2830 felec = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2832 d = _mm_sub_pd(r31,rswitch);
2833 d = _mm_max_pd(d,_mm_setzero_pd());
2834 d2 = _mm_mul_pd(d,d);
2835 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2837 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2839 /* Evaluate switch function */
2840 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2841 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv31,_mm_mul_pd(velec,dsw)) );
2842 cutoff_mask = _mm_cmplt_pd(rsq31,rcutoff2);
2846 fscal = _mm_and_pd(fscal,cutoff_mask);
2848 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2850 /* Update vectorial force */
2851 fix3 = _mm_macc_pd(dx31,fscal,fix3);
2852 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
2853 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
2855 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
2856 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
2857 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
2861 /**************************
2862 * CALCULATE INTERACTIONS *
2863 **************************/
2865 if (gmx_mm_any_lt(rsq32,rcutoff2))
2868 r32 = _mm_mul_pd(rsq32,rinv32);
2870 /* EWALD ELECTROSTATICS */
2872 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2873 ewrt = _mm_mul_pd(r32,ewtabscale);
2874 ewitab = _mm_cvttpd_epi32(ewrt);
2876 eweps = _mm_frcz_pd(ewrt);
2878 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2880 twoeweps = _mm_add_pd(eweps,eweps);
2881 ewitab = _mm_slli_epi32(ewitab,2);
2882 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2883 ewtabD = _mm_setzero_pd();
2884 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2885 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2886 ewtabFn = _mm_setzero_pd();
2887 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2888 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2889 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2890 velec = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
2891 felec = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2893 d = _mm_sub_pd(r32,rswitch);
2894 d = _mm_max_pd(d,_mm_setzero_pd());
2895 d2 = _mm_mul_pd(d,d);
2896 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2898 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2900 /* Evaluate switch function */
2901 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2902 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv32,_mm_mul_pd(velec,dsw)) );
2903 cutoff_mask = _mm_cmplt_pd(rsq32,rcutoff2);
2907 fscal = _mm_and_pd(fscal,cutoff_mask);
2909 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2911 /* Update vectorial force */
2912 fix3 = _mm_macc_pd(dx32,fscal,fix3);
2913 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
2914 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
2916 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
2917 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
2918 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
2922 /**************************
2923 * CALCULATE INTERACTIONS *
2924 **************************/
2926 if (gmx_mm_any_lt(rsq33,rcutoff2))
2929 r33 = _mm_mul_pd(rsq33,rinv33);
2931 /* EWALD ELECTROSTATICS */
2933 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2934 ewrt = _mm_mul_pd(r33,ewtabscale);
2935 ewitab = _mm_cvttpd_epi32(ewrt);
2937 eweps = _mm_frcz_pd(ewrt);
2939 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2941 twoeweps = _mm_add_pd(eweps,eweps);
2942 ewitab = _mm_slli_epi32(ewitab,2);
2943 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2944 ewtabD = _mm_setzero_pd();
2945 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2946 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2947 ewtabFn = _mm_setzero_pd();
2948 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2949 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
2950 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2951 velec = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
2952 felec = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2954 d = _mm_sub_pd(r33,rswitch);
2955 d = _mm_max_pd(d,_mm_setzero_pd());
2956 d2 = _mm_mul_pd(d,d);
2957 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2959 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2961 /* Evaluate switch function */
2962 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2963 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv33,_mm_mul_pd(velec,dsw)) );
2964 cutoff_mask = _mm_cmplt_pd(rsq33,rcutoff2);
2968 fscal = _mm_and_pd(fscal,cutoff_mask);
2970 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2972 /* Update vectorial force */
2973 fix3 = _mm_macc_pd(dx33,fscal,fix3);
2974 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
2975 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
2977 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
2978 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
2979 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
2983 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2985 /* Inner loop uses 585 flops */
2988 /* End of innermost loop */
2990 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2991 f+i_coord_offset+DIM,fshift+i_shift_offset);
2993 /* Increment number of inner iterations */
2994 inneriter += j_index_end - j_index_start;
2996 /* Outer loop uses 18 flops */
2999 /* Increment number of outer iterations */
3002 /* Update outer/inner flops */
3004 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*585);