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_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
50 #include "kernelutil_x86_avx_128_fma_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW4W4_VF_avx_128_fma_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: None
56 * Geometry: Water4-Water4
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSw_VdwNone_GeomW4W4_VF_avx_128_fma_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
91 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
92 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
93 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
94 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
95 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
96 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
97 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
98 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
99 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
100 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
101 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
102 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
103 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
104 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
105 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
106 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
109 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
110 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
112 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
113 real rswitch_scalar,d_scalar;
114 __m128 dummy_mask,cutoff_mask;
115 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
116 __m128 one = _mm_set1_ps(1.0);
117 __m128 two = _mm_set1_ps(2.0);
123 jindex = nlist->jindex;
125 shiftidx = nlist->shift;
127 shiftvec = fr->shift_vec[0];
128 fshift = fr->fshift[0];
129 facel = _mm_set1_ps(fr->epsfac);
130 charge = mdatoms->chargeA;
132 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
133 beta = _mm_set1_ps(fr->ic->ewaldcoeff_q);
134 beta2 = _mm_mul_ps(beta,beta);
135 beta3 = _mm_mul_ps(beta,beta2);
136 ewtab = fr->ic->tabq_coul_FDV0;
137 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
138 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
140 /* Setup water-specific parameters */
141 inr = nlist->iinr[0];
142 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
143 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
144 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
146 jq1 = _mm_set1_ps(charge[inr+1]);
147 jq2 = _mm_set1_ps(charge[inr+2]);
148 jq3 = _mm_set1_ps(charge[inr+3]);
149 qq11 = _mm_mul_ps(iq1,jq1);
150 qq12 = _mm_mul_ps(iq1,jq2);
151 qq13 = _mm_mul_ps(iq1,jq3);
152 qq21 = _mm_mul_ps(iq2,jq1);
153 qq22 = _mm_mul_ps(iq2,jq2);
154 qq23 = _mm_mul_ps(iq2,jq3);
155 qq31 = _mm_mul_ps(iq3,jq1);
156 qq32 = _mm_mul_ps(iq3,jq2);
157 qq33 = _mm_mul_ps(iq3,jq3);
159 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
160 rcutoff_scalar = fr->rcoulomb;
161 rcutoff = _mm_set1_ps(rcutoff_scalar);
162 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
164 rswitch_scalar = fr->rcoulomb_switch;
165 rswitch = _mm_set1_ps(rswitch_scalar);
166 /* Setup switch parameters */
167 d_scalar = rcutoff_scalar-rswitch_scalar;
168 d = _mm_set1_ps(d_scalar);
169 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
170 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
171 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
172 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
173 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
174 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
176 /* Avoid stupid compiler warnings */
177 jnrA = jnrB = jnrC = jnrD = 0;
186 for(iidx=0;iidx<4*DIM;iidx++)
191 /* Start outer loop over neighborlists */
192 for(iidx=0; iidx<nri; iidx++)
194 /* Load shift vector for this list */
195 i_shift_offset = DIM*shiftidx[iidx];
197 /* Load limits for loop over neighbors */
198 j_index_start = jindex[iidx];
199 j_index_end = jindex[iidx+1];
201 /* Get outer coordinate index */
203 i_coord_offset = DIM*inr;
205 /* Load i particle coords and add shift vector */
206 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
207 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
209 fix1 = _mm_setzero_ps();
210 fiy1 = _mm_setzero_ps();
211 fiz1 = _mm_setzero_ps();
212 fix2 = _mm_setzero_ps();
213 fiy2 = _mm_setzero_ps();
214 fiz2 = _mm_setzero_ps();
215 fix3 = _mm_setzero_ps();
216 fiy3 = _mm_setzero_ps();
217 fiz3 = _mm_setzero_ps();
219 /* Reset potential sums */
220 velecsum = _mm_setzero_ps();
222 /* Start inner kernel loop */
223 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
226 /* Get j neighbor index, and coordinate index */
231 j_coord_offsetA = DIM*jnrA;
232 j_coord_offsetB = DIM*jnrB;
233 j_coord_offsetC = DIM*jnrC;
234 j_coord_offsetD = DIM*jnrD;
236 /* load j atom coordinates */
237 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
238 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
239 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
241 /* Calculate displacement vector */
242 dx11 = _mm_sub_ps(ix1,jx1);
243 dy11 = _mm_sub_ps(iy1,jy1);
244 dz11 = _mm_sub_ps(iz1,jz1);
245 dx12 = _mm_sub_ps(ix1,jx2);
246 dy12 = _mm_sub_ps(iy1,jy2);
247 dz12 = _mm_sub_ps(iz1,jz2);
248 dx13 = _mm_sub_ps(ix1,jx3);
249 dy13 = _mm_sub_ps(iy1,jy3);
250 dz13 = _mm_sub_ps(iz1,jz3);
251 dx21 = _mm_sub_ps(ix2,jx1);
252 dy21 = _mm_sub_ps(iy2,jy1);
253 dz21 = _mm_sub_ps(iz2,jz1);
254 dx22 = _mm_sub_ps(ix2,jx2);
255 dy22 = _mm_sub_ps(iy2,jy2);
256 dz22 = _mm_sub_ps(iz2,jz2);
257 dx23 = _mm_sub_ps(ix2,jx3);
258 dy23 = _mm_sub_ps(iy2,jy3);
259 dz23 = _mm_sub_ps(iz2,jz3);
260 dx31 = _mm_sub_ps(ix3,jx1);
261 dy31 = _mm_sub_ps(iy3,jy1);
262 dz31 = _mm_sub_ps(iz3,jz1);
263 dx32 = _mm_sub_ps(ix3,jx2);
264 dy32 = _mm_sub_ps(iy3,jy2);
265 dz32 = _mm_sub_ps(iz3,jz2);
266 dx33 = _mm_sub_ps(ix3,jx3);
267 dy33 = _mm_sub_ps(iy3,jy3);
268 dz33 = _mm_sub_ps(iz3,jz3);
270 /* Calculate squared distance and things based on it */
271 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
272 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
273 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
274 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
275 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
276 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
277 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
278 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
279 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
281 rinv11 = gmx_mm_invsqrt_ps(rsq11);
282 rinv12 = gmx_mm_invsqrt_ps(rsq12);
283 rinv13 = gmx_mm_invsqrt_ps(rsq13);
284 rinv21 = gmx_mm_invsqrt_ps(rsq21);
285 rinv22 = gmx_mm_invsqrt_ps(rsq22);
286 rinv23 = gmx_mm_invsqrt_ps(rsq23);
287 rinv31 = gmx_mm_invsqrt_ps(rsq31);
288 rinv32 = gmx_mm_invsqrt_ps(rsq32);
289 rinv33 = gmx_mm_invsqrt_ps(rsq33);
291 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
292 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
293 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
294 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
295 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
296 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
297 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
298 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
299 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
301 fjx1 = _mm_setzero_ps();
302 fjy1 = _mm_setzero_ps();
303 fjz1 = _mm_setzero_ps();
304 fjx2 = _mm_setzero_ps();
305 fjy2 = _mm_setzero_ps();
306 fjz2 = _mm_setzero_ps();
307 fjx3 = _mm_setzero_ps();
308 fjy3 = _mm_setzero_ps();
309 fjz3 = _mm_setzero_ps();
311 /**************************
312 * CALCULATE INTERACTIONS *
313 **************************/
315 if (gmx_mm_any_lt(rsq11,rcutoff2))
318 r11 = _mm_mul_ps(rsq11,rinv11);
320 /* EWALD ELECTROSTATICS */
322 /* Analytical PME correction */
323 zeta2 = _mm_mul_ps(beta2,rsq11);
324 rinv3 = _mm_mul_ps(rinvsq11,rinv11);
325 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
326 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
327 felec = _mm_mul_ps(qq11,felec);
328 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
329 velec = _mm_nmacc_ps(pmecorrV,beta,rinv11);
330 velec = _mm_mul_ps(qq11,velec);
332 d = _mm_sub_ps(r11,rswitch);
333 d = _mm_max_ps(d,_mm_setzero_ps());
334 d2 = _mm_mul_ps(d,d);
335 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
337 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
339 /* Evaluate switch function */
340 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
341 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
342 velec = _mm_mul_ps(velec,sw);
343 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
345 /* Update potential sum for this i atom from the interaction with this j atom. */
346 velec = _mm_and_ps(velec,cutoff_mask);
347 velecsum = _mm_add_ps(velecsum,velec);
351 fscal = _mm_and_ps(fscal,cutoff_mask);
353 /* Update vectorial force */
354 fix1 = _mm_macc_ps(dx11,fscal,fix1);
355 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
356 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
358 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
359 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
360 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
364 /**************************
365 * CALCULATE INTERACTIONS *
366 **************************/
368 if (gmx_mm_any_lt(rsq12,rcutoff2))
371 r12 = _mm_mul_ps(rsq12,rinv12);
373 /* EWALD ELECTROSTATICS */
375 /* Analytical PME correction */
376 zeta2 = _mm_mul_ps(beta2,rsq12);
377 rinv3 = _mm_mul_ps(rinvsq12,rinv12);
378 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
379 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
380 felec = _mm_mul_ps(qq12,felec);
381 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
382 velec = _mm_nmacc_ps(pmecorrV,beta,rinv12);
383 velec = _mm_mul_ps(qq12,velec);
385 d = _mm_sub_ps(r12,rswitch);
386 d = _mm_max_ps(d,_mm_setzero_ps());
387 d2 = _mm_mul_ps(d,d);
388 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
390 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
392 /* Evaluate switch function */
393 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
394 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
395 velec = _mm_mul_ps(velec,sw);
396 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
398 /* Update potential sum for this i atom from the interaction with this j atom. */
399 velec = _mm_and_ps(velec,cutoff_mask);
400 velecsum = _mm_add_ps(velecsum,velec);
404 fscal = _mm_and_ps(fscal,cutoff_mask);
406 /* Update vectorial force */
407 fix1 = _mm_macc_ps(dx12,fscal,fix1);
408 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
409 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
411 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
412 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
413 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
417 /**************************
418 * CALCULATE INTERACTIONS *
419 **************************/
421 if (gmx_mm_any_lt(rsq13,rcutoff2))
424 r13 = _mm_mul_ps(rsq13,rinv13);
426 /* EWALD ELECTROSTATICS */
428 /* Analytical PME correction */
429 zeta2 = _mm_mul_ps(beta2,rsq13);
430 rinv3 = _mm_mul_ps(rinvsq13,rinv13);
431 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
432 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
433 felec = _mm_mul_ps(qq13,felec);
434 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
435 velec = _mm_nmacc_ps(pmecorrV,beta,rinv13);
436 velec = _mm_mul_ps(qq13,velec);
438 d = _mm_sub_ps(r13,rswitch);
439 d = _mm_max_ps(d,_mm_setzero_ps());
440 d2 = _mm_mul_ps(d,d);
441 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
443 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
445 /* Evaluate switch function */
446 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
447 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
448 velec = _mm_mul_ps(velec,sw);
449 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
451 /* Update potential sum for this i atom from the interaction with this j atom. */
452 velec = _mm_and_ps(velec,cutoff_mask);
453 velecsum = _mm_add_ps(velecsum,velec);
457 fscal = _mm_and_ps(fscal,cutoff_mask);
459 /* Update vectorial force */
460 fix1 = _mm_macc_ps(dx13,fscal,fix1);
461 fiy1 = _mm_macc_ps(dy13,fscal,fiy1);
462 fiz1 = _mm_macc_ps(dz13,fscal,fiz1);
464 fjx3 = _mm_macc_ps(dx13,fscal,fjx3);
465 fjy3 = _mm_macc_ps(dy13,fscal,fjy3);
466 fjz3 = _mm_macc_ps(dz13,fscal,fjz3);
470 /**************************
471 * CALCULATE INTERACTIONS *
472 **************************/
474 if (gmx_mm_any_lt(rsq21,rcutoff2))
477 r21 = _mm_mul_ps(rsq21,rinv21);
479 /* EWALD ELECTROSTATICS */
481 /* Analytical PME correction */
482 zeta2 = _mm_mul_ps(beta2,rsq21);
483 rinv3 = _mm_mul_ps(rinvsq21,rinv21);
484 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
485 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
486 felec = _mm_mul_ps(qq21,felec);
487 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
488 velec = _mm_nmacc_ps(pmecorrV,beta,rinv21);
489 velec = _mm_mul_ps(qq21,velec);
491 d = _mm_sub_ps(r21,rswitch);
492 d = _mm_max_ps(d,_mm_setzero_ps());
493 d2 = _mm_mul_ps(d,d);
494 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
496 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
498 /* Evaluate switch function */
499 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
500 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
501 velec = _mm_mul_ps(velec,sw);
502 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
504 /* Update potential sum for this i atom from the interaction with this j atom. */
505 velec = _mm_and_ps(velec,cutoff_mask);
506 velecsum = _mm_add_ps(velecsum,velec);
510 fscal = _mm_and_ps(fscal,cutoff_mask);
512 /* Update vectorial force */
513 fix2 = _mm_macc_ps(dx21,fscal,fix2);
514 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
515 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
517 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
518 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
519 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
523 /**************************
524 * CALCULATE INTERACTIONS *
525 **************************/
527 if (gmx_mm_any_lt(rsq22,rcutoff2))
530 r22 = _mm_mul_ps(rsq22,rinv22);
532 /* EWALD ELECTROSTATICS */
534 /* Analytical PME correction */
535 zeta2 = _mm_mul_ps(beta2,rsq22);
536 rinv3 = _mm_mul_ps(rinvsq22,rinv22);
537 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
538 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
539 felec = _mm_mul_ps(qq22,felec);
540 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
541 velec = _mm_nmacc_ps(pmecorrV,beta,rinv22);
542 velec = _mm_mul_ps(qq22,velec);
544 d = _mm_sub_ps(r22,rswitch);
545 d = _mm_max_ps(d,_mm_setzero_ps());
546 d2 = _mm_mul_ps(d,d);
547 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
549 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
551 /* Evaluate switch function */
552 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
553 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
554 velec = _mm_mul_ps(velec,sw);
555 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
557 /* Update potential sum for this i atom from the interaction with this j atom. */
558 velec = _mm_and_ps(velec,cutoff_mask);
559 velecsum = _mm_add_ps(velecsum,velec);
563 fscal = _mm_and_ps(fscal,cutoff_mask);
565 /* Update vectorial force */
566 fix2 = _mm_macc_ps(dx22,fscal,fix2);
567 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
568 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
570 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
571 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
572 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
576 /**************************
577 * CALCULATE INTERACTIONS *
578 **************************/
580 if (gmx_mm_any_lt(rsq23,rcutoff2))
583 r23 = _mm_mul_ps(rsq23,rinv23);
585 /* EWALD ELECTROSTATICS */
587 /* Analytical PME correction */
588 zeta2 = _mm_mul_ps(beta2,rsq23);
589 rinv3 = _mm_mul_ps(rinvsq23,rinv23);
590 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
591 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
592 felec = _mm_mul_ps(qq23,felec);
593 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
594 velec = _mm_nmacc_ps(pmecorrV,beta,rinv23);
595 velec = _mm_mul_ps(qq23,velec);
597 d = _mm_sub_ps(r23,rswitch);
598 d = _mm_max_ps(d,_mm_setzero_ps());
599 d2 = _mm_mul_ps(d,d);
600 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
602 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
604 /* Evaluate switch function */
605 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
606 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
607 velec = _mm_mul_ps(velec,sw);
608 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
610 /* Update potential sum for this i atom from the interaction with this j atom. */
611 velec = _mm_and_ps(velec,cutoff_mask);
612 velecsum = _mm_add_ps(velecsum,velec);
616 fscal = _mm_and_ps(fscal,cutoff_mask);
618 /* Update vectorial force */
619 fix2 = _mm_macc_ps(dx23,fscal,fix2);
620 fiy2 = _mm_macc_ps(dy23,fscal,fiy2);
621 fiz2 = _mm_macc_ps(dz23,fscal,fiz2);
623 fjx3 = _mm_macc_ps(dx23,fscal,fjx3);
624 fjy3 = _mm_macc_ps(dy23,fscal,fjy3);
625 fjz3 = _mm_macc_ps(dz23,fscal,fjz3);
629 /**************************
630 * CALCULATE INTERACTIONS *
631 **************************/
633 if (gmx_mm_any_lt(rsq31,rcutoff2))
636 r31 = _mm_mul_ps(rsq31,rinv31);
638 /* EWALD ELECTROSTATICS */
640 /* Analytical PME correction */
641 zeta2 = _mm_mul_ps(beta2,rsq31);
642 rinv3 = _mm_mul_ps(rinvsq31,rinv31);
643 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
644 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
645 felec = _mm_mul_ps(qq31,felec);
646 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
647 velec = _mm_nmacc_ps(pmecorrV,beta,rinv31);
648 velec = _mm_mul_ps(qq31,velec);
650 d = _mm_sub_ps(r31,rswitch);
651 d = _mm_max_ps(d,_mm_setzero_ps());
652 d2 = _mm_mul_ps(d,d);
653 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
655 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
657 /* Evaluate switch function */
658 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
659 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
660 velec = _mm_mul_ps(velec,sw);
661 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
663 /* Update potential sum for this i atom from the interaction with this j atom. */
664 velec = _mm_and_ps(velec,cutoff_mask);
665 velecsum = _mm_add_ps(velecsum,velec);
669 fscal = _mm_and_ps(fscal,cutoff_mask);
671 /* Update vectorial force */
672 fix3 = _mm_macc_ps(dx31,fscal,fix3);
673 fiy3 = _mm_macc_ps(dy31,fscal,fiy3);
674 fiz3 = _mm_macc_ps(dz31,fscal,fiz3);
676 fjx1 = _mm_macc_ps(dx31,fscal,fjx1);
677 fjy1 = _mm_macc_ps(dy31,fscal,fjy1);
678 fjz1 = _mm_macc_ps(dz31,fscal,fjz1);
682 /**************************
683 * CALCULATE INTERACTIONS *
684 **************************/
686 if (gmx_mm_any_lt(rsq32,rcutoff2))
689 r32 = _mm_mul_ps(rsq32,rinv32);
691 /* EWALD ELECTROSTATICS */
693 /* Analytical PME correction */
694 zeta2 = _mm_mul_ps(beta2,rsq32);
695 rinv3 = _mm_mul_ps(rinvsq32,rinv32);
696 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
697 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
698 felec = _mm_mul_ps(qq32,felec);
699 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
700 velec = _mm_nmacc_ps(pmecorrV,beta,rinv32);
701 velec = _mm_mul_ps(qq32,velec);
703 d = _mm_sub_ps(r32,rswitch);
704 d = _mm_max_ps(d,_mm_setzero_ps());
705 d2 = _mm_mul_ps(d,d);
706 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
708 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
710 /* Evaluate switch function */
711 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
712 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
713 velec = _mm_mul_ps(velec,sw);
714 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
716 /* Update potential sum for this i atom from the interaction with this j atom. */
717 velec = _mm_and_ps(velec,cutoff_mask);
718 velecsum = _mm_add_ps(velecsum,velec);
722 fscal = _mm_and_ps(fscal,cutoff_mask);
724 /* Update vectorial force */
725 fix3 = _mm_macc_ps(dx32,fscal,fix3);
726 fiy3 = _mm_macc_ps(dy32,fscal,fiy3);
727 fiz3 = _mm_macc_ps(dz32,fscal,fiz3);
729 fjx2 = _mm_macc_ps(dx32,fscal,fjx2);
730 fjy2 = _mm_macc_ps(dy32,fscal,fjy2);
731 fjz2 = _mm_macc_ps(dz32,fscal,fjz2);
735 /**************************
736 * CALCULATE INTERACTIONS *
737 **************************/
739 if (gmx_mm_any_lt(rsq33,rcutoff2))
742 r33 = _mm_mul_ps(rsq33,rinv33);
744 /* EWALD ELECTROSTATICS */
746 /* Analytical PME correction */
747 zeta2 = _mm_mul_ps(beta2,rsq33);
748 rinv3 = _mm_mul_ps(rinvsq33,rinv33);
749 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
750 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
751 felec = _mm_mul_ps(qq33,felec);
752 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
753 velec = _mm_nmacc_ps(pmecorrV,beta,rinv33);
754 velec = _mm_mul_ps(qq33,velec);
756 d = _mm_sub_ps(r33,rswitch);
757 d = _mm_max_ps(d,_mm_setzero_ps());
758 d2 = _mm_mul_ps(d,d);
759 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
761 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
763 /* Evaluate switch function */
764 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
765 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
766 velec = _mm_mul_ps(velec,sw);
767 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
769 /* Update potential sum for this i atom from the interaction with this j atom. */
770 velec = _mm_and_ps(velec,cutoff_mask);
771 velecsum = _mm_add_ps(velecsum,velec);
775 fscal = _mm_and_ps(fscal,cutoff_mask);
777 /* Update vectorial force */
778 fix3 = _mm_macc_ps(dx33,fscal,fix3);
779 fiy3 = _mm_macc_ps(dy33,fscal,fiy3);
780 fiz3 = _mm_macc_ps(dz33,fscal,fiz3);
782 fjx3 = _mm_macc_ps(dx33,fscal,fjx3);
783 fjy3 = _mm_macc_ps(dy33,fscal,fjy3);
784 fjz3 = _mm_macc_ps(dz33,fscal,fjz3);
788 fjptrA = f+j_coord_offsetA;
789 fjptrB = f+j_coord_offsetB;
790 fjptrC = f+j_coord_offsetC;
791 fjptrD = f+j_coord_offsetD;
793 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
794 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
796 /* Inner loop uses 477 flops */
802 /* Get j neighbor index, and coordinate index */
803 jnrlistA = jjnr[jidx];
804 jnrlistB = jjnr[jidx+1];
805 jnrlistC = jjnr[jidx+2];
806 jnrlistD = jjnr[jidx+3];
807 /* Sign of each element will be negative for non-real atoms.
808 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
809 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
811 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
812 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
813 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
814 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
815 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
816 j_coord_offsetA = DIM*jnrA;
817 j_coord_offsetB = DIM*jnrB;
818 j_coord_offsetC = DIM*jnrC;
819 j_coord_offsetD = DIM*jnrD;
821 /* load j atom coordinates */
822 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
823 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
824 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
826 /* Calculate displacement vector */
827 dx11 = _mm_sub_ps(ix1,jx1);
828 dy11 = _mm_sub_ps(iy1,jy1);
829 dz11 = _mm_sub_ps(iz1,jz1);
830 dx12 = _mm_sub_ps(ix1,jx2);
831 dy12 = _mm_sub_ps(iy1,jy2);
832 dz12 = _mm_sub_ps(iz1,jz2);
833 dx13 = _mm_sub_ps(ix1,jx3);
834 dy13 = _mm_sub_ps(iy1,jy3);
835 dz13 = _mm_sub_ps(iz1,jz3);
836 dx21 = _mm_sub_ps(ix2,jx1);
837 dy21 = _mm_sub_ps(iy2,jy1);
838 dz21 = _mm_sub_ps(iz2,jz1);
839 dx22 = _mm_sub_ps(ix2,jx2);
840 dy22 = _mm_sub_ps(iy2,jy2);
841 dz22 = _mm_sub_ps(iz2,jz2);
842 dx23 = _mm_sub_ps(ix2,jx3);
843 dy23 = _mm_sub_ps(iy2,jy3);
844 dz23 = _mm_sub_ps(iz2,jz3);
845 dx31 = _mm_sub_ps(ix3,jx1);
846 dy31 = _mm_sub_ps(iy3,jy1);
847 dz31 = _mm_sub_ps(iz3,jz1);
848 dx32 = _mm_sub_ps(ix3,jx2);
849 dy32 = _mm_sub_ps(iy3,jy2);
850 dz32 = _mm_sub_ps(iz3,jz2);
851 dx33 = _mm_sub_ps(ix3,jx3);
852 dy33 = _mm_sub_ps(iy3,jy3);
853 dz33 = _mm_sub_ps(iz3,jz3);
855 /* Calculate squared distance and things based on it */
856 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
857 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
858 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
859 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
860 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
861 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
862 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
863 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
864 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
866 rinv11 = gmx_mm_invsqrt_ps(rsq11);
867 rinv12 = gmx_mm_invsqrt_ps(rsq12);
868 rinv13 = gmx_mm_invsqrt_ps(rsq13);
869 rinv21 = gmx_mm_invsqrt_ps(rsq21);
870 rinv22 = gmx_mm_invsqrt_ps(rsq22);
871 rinv23 = gmx_mm_invsqrt_ps(rsq23);
872 rinv31 = gmx_mm_invsqrt_ps(rsq31);
873 rinv32 = gmx_mm_invsqrt_ps(rsq32);
874 rinv33 = gmx_mm_invsqrt_ps(rsq33);
876 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
877 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
878 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
879 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
880 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
881 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
882 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
883 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
884 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
886 fjx1 = _mm_setzero_ps();
887 fjy1 = _mm_setzero_ps();
888 fjz1 = _mm_setzero_ps();
889 fjx2 = _mm_setzero_ps();
890 fjy2 = _mm_setzero_ps();
891 fjz2 = _mm_setzero_ps();
892 fjx3 = _mm_setzero_ps();
893 fjy3 = _mm_setzero_ps();
894 fjz3 = _mm_setzero_ps();
896 /**************************
897 * CALCULATE INTERACTIONS *
898 **************************/
900 if (gmx_mm_any_lt(rsq11,rcutoff2))
903 r11 = _mm_mul_ps(rsq11,rinv11);
904 r11 = _mm_andnot_ps(dummy_mask,r11);
906 /* EWALD ELECTROSTATICS */
908 /* Analytical PME correction */
909 zeta2 = _mm_mul_ps(beta2,rsq11);
910 rinv3 = _mm_mul_ps(rinvsq11,rinv11);
911 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
912 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
913 felec = _mm_mul_ps(qq11,felec);
914 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
915 velec = _mm_nmacc_ps(pmecorrV,beta,rinv11);
916 velec = _mm_mul_ps(qq11,velec);
918 d = _mm_sub_ps(r11,rswitch);
919 d = _mm_max_ps(d,_mm_setzero_ps());
920 d2 = _mm_mul_ps(d,d);
921 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
923 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
925 /* Evaluate switch function */
926 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
927 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
928 velec = _mm_mul_ps(velec,sw);
929 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
931 /* Update potential sum for this i atom from the interaction with this j atom. */
932 velec = _mm_and_ps(velec,cutoff_mask);
933 velec = _mm_andnot_ps(dummy_mask,velec);
934 velecsum = _mm_add_ps(velecsum,velec);
938 fscal = _mm_and_ps(fscal,cutoff_mask);
940 fscal = _mm_andnot_ps(dummy_mask,fscal);
942 /* Update vectorial force */
943 fix1 = _mm_macc_ps(dx11,fscal,fix1);
944 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
945 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
947 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
948 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
949 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
953 /**************************
954 * CALCULATE INTERACTIONS *
955 **************************/
957 if (gmx_mm_any_lt(rsq12,rcutoff2))
960 r12 = _mm_mul_ps(rsq12,rinv12);
961 r12 = _mm_andnot_ps(dummy_mask,r12);
963 /* EWALD ELECTROSTATICS */
965 /* Analytical PME correction */
966 zeta2 = _mm_mul_ps(beta2,rsq12);
967 rinv3 = _mm_mul_ps(rinvsq12,rinv12);
968 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
969 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
970 felec = _mm_mul_ps(qq12,felec);
971 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
972 velec = _mm_nmacc_ps(pmecorrV,beta,rinv12);
973 velec = _mm_mul_ps(qq12,velec);
975 d = _mm_sub_ps(r12,rswitch);
976 d = _mm_max_ps(d,_mm_setzero_ps());
977 d2 = _mm_mul_ps(d,d);
978 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
980 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
982 /* Evaluate switch function */
983 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
984 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
985 velec = _mm_mul_ps(velec,sw);
986 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
988 /* Update potential sum for this i atom from the interaction with this j atom. */
989 velec = _mm_and_ps(velec,cutoff_mask);
990 velec = _mm_andnot_ps(dummy_mask,velec);
991 velecsum = _mm_add_ps(velecsum,velec);
995 fscal = _mm_and_ps(fscal,cutoff_mask);
997 fscal = _mm_andnot_ps(dummy_mask,fscal);
999 /* Update vectorial force */
1000 fix1 = _mm_macc_ps(dx12,fscal,fix1);
1001 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
1002 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
1004 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
1005 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
1006 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
1010 /**************************
1011 * CALCULATE INTERACTIONS *
1012 **************************/
1014 if (gmx_mm_any_lt(rsq13,rcutoff2))
1017 r13 = _mm_mul_ps(rsq13,rinv13);
1018 r13 = _mm_andnot_ps(dummy_mask,r13);
1020 /* EWALD ELECTROSTATICS */
1022 /* Analytical PME correction */
1023 zeta2 = _mm_mul_ps(beta2,rsq13);
1024 rinv3 = _mm_mul_ps(rinvsq13,rinv13);
1025 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1026 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1027 felec = _mm_mul_ps(qq13,felec);
1028 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1029 velec = _mm_nmacc_ps(pmecorrV,beta,rinv13);
1030 velec = _mm_mul_ps(qq13,velec);
1032 d = _mm_sub_ps(r13,rswitch);
1033 d = _mm_max_ps(d,_mm_setzero_ps());
1034 d2 = _mm_mul_ps(d,d);
1035 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1037 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1039 /* Evaluate switch function */
1040 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1041 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
1042 velec = _mm_mul_ps(velec,sw);
1043 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
1045 /* Update potential sum for this i atom from the interaction with this j atom. */
1046 velec = _mm_and_ps(velec,cutoff_mask);
1047 velec = _mm_andnot_ps(dummy_mask,velec);
1048 velecsum = _mm_add_ps(velecsum,velec);
1052 fscal = _mm_and_ps(fscal,cutoff_mask);
1054 fscal = _mm_andnot_ps(dummy_mask,fscal);
1056 /* Update vectorial force */
1057 fix1 = _mm_macc_ps(dx13,fscal,fix1);
1058 fiy1 = _mm_macc_ps(dy13,fscal,fiy1);
1059 fiz1 = _mm_macc_ps(dz13,fscal,fiz1);
1061 fjx3 = _mm_macc_ps(dx13,fscal,fjx3);
1062 fjy3 = _mm_macc_ps(dy13,fscal,fjy3);
1063 fjz3 = _mm_macc_ps(dz13,fscal,fjz3);
1067 /**************************
1068 * CALCULATE INTERACTIONS *
1069 **************************/
1071 if (gmx_mm_any_lt(rsq21,rcutoff2))
1074 r21 = _mm_mul_ps(rsq21,rinv21);
1075 r21 = _mm_andnot_ps(dummy_mask,r21);
1077 /* EWALD ELECTROSTATICS */
1079 /* Analytical PME correction */
1080 zeta2 = _mm_mul_ps(beta2,rsq21);
1081 rinv3 = _mm_mul_ps(rinvsq21,rinv21);
1082 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1083 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1084 felec = _mm_mul_ps(qq21,felec);
1085 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1086 velec = _mm_nmacc_ps(pmecorrV,beta,rinv21);
1087 velec = _mm_mul_ps(qq21,velec);
1089 d = _mm_sub_ps(r21,rswitch);
1090 d = _mm_max_ps(d,_mm_setzero_ps());
1091 d2 = _mm_mul_ps(d,d);
1092 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1094 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1096 /* Evaluate switch function */
1097 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1098 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
1099 velec = _mm_mul_ps(velec,sw);
1100 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
1102 /* Update potential sum for this i atom from the interaction with this j atom. */
1103 velec = _mm_and_ps(velec,cutoff_mask);
1104 velec = _mm_andnot_ps(dummy_mask,velec);
1105 velecsum = _mm_add_ps(velecsum,velec);
1109 fscal = _mm_and_ps(fscal,cutoff_mask);
1111 fscal = _mm_andnot_ps(dummy_mask,fscal);
1113 /* Update vectorial force */
1114 fix2 = _mm_macc_ps(dx21,fscal,fix2);
1115 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
1116 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
1118 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
1119 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
1120 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
1124 /**************************
1125 * CALCULATE INTERACTIONS *
1126 **************************/
1128 if (gmx_mm_any_lt(rsq22,rcutoff2))
1131 r22 = _mm_mul_ps(rsq22,rinv22);
1132 r22 = _mm_andnot_ps(dummy_mask,r22);
1134 /* EWALD ELECTROSTATICS */
1136 /* Analytical PME correction */
1137 zeta2 = _mm_mul_ps(beta2,rsq22);
1138 rinv3 = _mm_mul_ps(rinvsq22,rinv22);
1139 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1140 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1141 felec = _mm_mul_ps(qq22,felec);
1142 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1143 velec = _mm_nmacc_ps(pmecorrV,beta,rinv22);
1144 velec = _mm_mul_ps(qq22,velec);
1146 d = _mm_sub_ps(r22,rswitch);
1147 d = _mm_max_ps(d,_mm_setzero_ps());
1148 d2 = _mm_mul_ps(d,d);
1149 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1151 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1153 /* Evaluate switch function */
1154 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1155 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
1156 velec = _mm_mul_ps(velec,sw);
1157 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
1159 /* Update potential sum for this i atom from the interaction with this j atom. */
1160 velec = _mm_and_ps(velec,cutoff_mask);
1161 velec = _mm_andnot_ps(dummy_mask,velec);
1162 velecsum = _mm_add_ps(velecsum,velec);
1166 fscal = _mm_and_ps(fscal,cutoff_mask);
1168 fscal = _mm_andnot_ps(dummy_mask,fscal);
1170 /* Update vectorial force */
1171 fix2 = _mm_macc_ps(dx22,fscal,fix2);
1172 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
1173 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
1175 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
1176 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
1177 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
1181 /**************************
1182 * CALCULATE INTERACTIONS *
1183 **************************/
1185 if (gmx_mm_any_lt(rsq23,rcutoff2))
1188 r23 = _mm_mul_ps(rsq23,rinv23);
1189 r23 = _mm_andnot_ps(dummy_mask,r23);
1191 /* EWALD ELECTROSTATICS */
1193 /* Analytical PME correction */
1194 zeta2 = _mm_mul_ps(beta2,rsq23);
1195 rinv3 = _mm_mul_ps(rinvsq23,rinv23);
1196 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1197 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1198 felec = _mm_mul_ps(qq23,felec);
1199 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1200 velec = _mm_nmacc_ps(pmecorrV,beta,rinv23);
1201 velec = _mm_mul_ps(qq23,velec);
1203 d = _mm_sub_ps(r23,rswitch);
1204 d = _mm_max_ps(d,_mm_setzero_ps());
1205 d2 = _mm_mul_ps(d,d);
1206 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1208 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1210 /* Evaluate switch function */
1211 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1212 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
1213 velec = _mm_mul_ps(velec,sw);
1214 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
1216 /* Update potential sum for this i atom from the interaction with this j atom. */
1217 velec = _mm_and_ps(velec,cutoff_mask);
1218 velec = _mm_andnot_ps(dummy_mask,velec);
1219 velecsum = _mm_add_ps(velecsum,velec);
1223 fscal = _mm_and_ps(fscal,cutoff_mask);
1225 fscal = _mm_andnot_ps(dummy_mask,fscal);
1227 /* Update vectorial force */
1228 fix2 = _mm_macc_ps(dx23,fscal,fix2);
1229 fiy2 = _mm_macc_ps(dy23,fscal,fiy2);
1230 fiz2 = _mm_macc_ps(dz23,fscal,fiz2);
1232 fjx3 = _mm_macc_ps(dx23,fscal,fjx3);
1233 fjy3 = _mm_macc_ps(dy23,fscal,fjy3);
1234 fjz3 = _mm_macc_ps(dz23,fscal,fjz3);
1238 /**************************
1239 * CALCULATE INTERACTIONS *
1240 **************************/
1242 if (gmx_mm_any_lt(rsq31,rcutoff2))
1245 r31 = _mm_mul_ps(rsq31,rinv31);
1246 r31 = _mm_andnot_ps(dummy_mask,r31);
1248 /* EWALD ELECTROSTATICS */
1250 /* Analytical PME correction */
1251 zeta2 = _mm_mul_ps(beta2,rsq31);
1252 rinv3 = _mm_mul_ps(rinvsq31,rinv31);
1253 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1254 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1255 felec = _mm_mul_ps(qq31,felec);
1256 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1257 velec = _mm_nmacc_ps(pmecorrV,beta,rinv31);
1258 velec = _mm_mul_ps(qq31,velec);
1260 d = _mm_sub_ps(r31,rswitch);
1261 d = _mm_max_ps(d,_mm_setzero_ps());
1262 d2 = _mm_mul_ps(d,d);
1263 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1265 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1267 /* Evaluate switch function */
1268 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1269 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
1270 velec = _mm_mul_ps(velec,sw);
1271 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
1273 /* Update potential sum for this i atom from the interaction with this j atom. */
1274 velec = _mm_and_ps(velec,cutoff_mask);
1275 velec = _mm_andnot_ps(dummy_mask,velec);
1276 velecsum = _mm_add_ps(velecsum,velec);
1280 fscal = _mm_and_ps(fscal,cutoff_mask);
1282 fscal = _mm_andnot_ps(dummy_mask,fscal);
1284 /* Update vectorial force */
1285 fix3 = _mm_macc_ps(dx31,fscal,fix3);
1286 fiy3 = _mm_macc_ps(dy31,fscal,fiy3);
1287 fiz3 = _mm_macc_ps(dz31,fscal,fiz3);
1289 fjx1 = _mm_macc_ps(dx31,fscal,fjx1);
1290 fjy1 = _mm_macc_ps(dy31,fscal,fjy1);
1291 fjz1 = _mm_macc_ps(dz31,fscal,fjz1);
1295 /**************************
1296 * CALCULATE INTERACTIONS *
1297 **************************/
1299 if (gmx_mm_any_lt(rsq32,rcutoff2))
1302 r32 = _mm_mul_ps(rsq32,rinv32);
1303 r32 = _mm_andnot_ps(dummy_mask,r32);
1305 /* EWALD ELECTROSTATICS */
1307 /* Analytical PME correction */
1308 zeta2 = _mm_mul_ps(beta2,rsq32);
1309 rinv3 = _mm_mul_ps(rinvsq32,rinv32);
1310 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1311 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1312 felec = _mm_mul_ps(qq32,felec);
1313 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1314 velec = _mm_nmacc_ps(pmecorrV,beta,rinv32);
1315 velec = _mm_mul_ps(qq32,velec);
1317 d = _mm_sub_ps(r32,rswitch);
1318 d = _mm_max_ps(d,_mm_setzero_ps());
1319 d2 = _mm_mul_ps(d,d);
1320 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1322 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1324 /* Evaluate switch function */
1325 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1326 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
1327 velec = _mm_mul_ps(velec,sw);
1328 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
1330 /* Update potential sum for this i atom from the interaction with this j atom. */
1331 velec = _mm_and_ps(velec,cutoff_mask);
1332 velec = _mm_andnot_ps(dummy_mask,velec);
1333 velecsum = _mm_add_ps(velecsum,velec);
1337 fscal = _mm_and_ps(fscal,cutoff_mask);
1339 fscal = _mm_andnot_ps(dummy_mask,fscal);
1341 /* Update vectorial force */
1342 fix3 = _mm_macc_ps(dx32,fscal,fix3);
1343 fiy3 = _mm_macc_ps(dy32,fscal,fiy3);
1344 fiz3 = _mm_macc_ps(dz32,fscal,fiz3);
1346 fjx2 = _mm_macc_ps(dx32,fscal,fjx2);
1347 fjy2 = _mm_macc_ps(dy32,fscal,fjy2);
1348 fjz2 = _mm_macc_ps(dz32,fscal,fjz2);
1352 /**************************
1353 * CALCULATE INTERACTIONS *
1354 **************************/
1356 if (gmx_mm_any_lt(rsq33,rcutoff2))
1359 r33 = _mm_mul_ps(rsq33,rinv33);
1360 r33 = _mm_andnot_ps(dummy_mask,r33);
1362 /* EWALD ELECTROSTATICS */
1364 /* Analytical PME correction */
1365 zeta2 = _mm_mul_ps(beta2,rsq33);
1366 rinv3 = _mm_mul_ps(rinvsq33,rinv33);
1367 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1368 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1369 felec = _mm_mul_ps(qq33,felec);
1370 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1371 velec = _mm_nmacc_ps(pmecorrV,beta,rinv33);
1372 velec = _mm_mul_ps(qq33,velec);
1374 d = _mm_sub_ps(r33,rswitch);
1375 d = _mm_max_ps(d,_mm_setzero_ps());
1376 d2 = _mm_mul_ps(d,d);
1377 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1379 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1381 /* Evaluate switch function */
1382 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1383 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
1384 velec = _mm_mul_ps(velec,sw);
1385 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
1387 /* Update potential sum for this i atom from the interaction with this j atom. */
1388 velec = _mm_and_ps(velec,cutoff_mask);
1389 velec = _mm_andnot_ps(dummy_mask,velec);
1390 velecsum = _mm_add_ps(velecsum,velec);
1394 fscal = _mm_and_ps(fscal,cutoff_mask);
1396 fscal = _mm_andnot_ps(dummy_mask,fscal);
1398 /* Update vectorial force */
1399 fix3 = _mm_macc_ps(dx33,fscal,fix3);
1400 fiy3 = _mm_macc_ps(dy33,fscal,fiy3);
1401 fiz3 = _mm_macc_ps(dz33,fscal,fiz3);
1403 fjx3 = _mm_macc_ps(dx33,fscal,fjx3);
1404 fjy3 = _mm_macc_ps(dy33,fscal,fjy3);
1405 fjz3 = _mm_macc_ps(dz33,fscal,fjz3);
1409 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1410 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1411 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1412 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1414 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1415 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1417 /* Inner loop uses 486 flops */
1420 /* End of innermost loop */
1422 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1423 f+i_coord_offset+DIM,fshift+i_shift_offset);
1426 /* Update potential energies */
1427 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1429 /* Increment number of inner iterations */
1430 inneriter += j_index_end - j_index_start;
1432 /* Outer loop uses 19 flops */
1435 /* Increment number of outer iterations */
1438 /* Update outer/inner flops */
1440 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*486);
1443 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW4W4_F_avx_128_fma_single
1444 * Electrostatics interaction: Ewald
1445 * VdW interaction: None
1446 * Geometry: Water4-Water4
1447 * Calculate force/pot: Force
1450 nb_kernel_ElecEwSw_VdwNone_GeomW4W4_F_avx_128_fma_single
1451 (t_nblist * gmx_restrict nlist,
1452 rvec * gmx_restrict xx,
1453 rvec * gmx_restrict ff,
1454 t_forcerec * gmx_restrict fr,
1455 t_mdatoms * gmx_restrict mdatoms,
1456 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1457 t_nrnb * gmx_restrict nrnb)
1459 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1460 * just 0 for non-waters.
1461 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1462 * jnr indices corresponding to data put in the four positions in the SIMD register.
1464 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1465 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1466 int jnrA,jnrB,jnrC,jnrD;
1467 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1468 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1469 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1470 real rcutoff_scalar;
1471 real *shiftvec,*fshift,*x,*f;
1472 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1473 real scratch[4*DIM];
1474 __m128 fscal,rcutoff,rcutoff2,jidxall;
1476 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1478 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1480 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1481 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1482 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1483 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1484 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1485 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1486 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1487 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1488 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1489 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1490 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1491 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1492 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1493 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1494 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1495 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1496 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1499 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1500 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1502 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1503 real rswitch_scalar,d_scalar;
1504 __m128 dummy_mask,cutoff_mask;
1505 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1506 __m128 one = _mm_set1_ps(1.0);
1507 __m128 two = _mm_set1_ps(2.0);
1513 jindex = nlist->jindex;
1515 shiftidx = nlist->shift;
1517 shiftvec = fr->shift_vec[0];
1518 fshift = fr->fshift[0];
1519 facel = _mm_set1_ps(fr->epsfac);
1520 charge = mdatoms->chargeA;
1522 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
1523 beta = _mm_set1_ps(fr->ic->ewaldcoeff_q);
1524 beta2 = _mm_mul_ps(beta,beta);
1525 beta3 = _mm_mul_ps(beta,beta2);
1526 ewtab = fr->ic->tabq_coul_FDV0;
1527 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
1528 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1530 /* Setup water-specific parameters */
1531 inr = nlist->iinr[0];
1532 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1533 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1534 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1536 jq1 = _mm_set1_ps(charge[inr+1]);
1537 jq2 = _mm_set1_ps(charge[inr+2]);
1538 jq3 = _mm_set1_ps(charge[inr+3]);
1539 qq11 = _mm_mul_ps(iq1,jq1);
1540 qq12 = _mm_mul_ps(iq1,jq2);
1541 qq13 = _mm_mul_ps(iq1,jq3);
1542 qq21 = _mm_mul_ps(iq2,jq1);
1543 qq22 = _mm_mul_ps(iq2,jq2);
1544 qq23 = _mm_mul_ps(iq2,jq3);
1545 qq31 = _mm_mul_ps(iq3,jq1);
1546 qq32 = _mm_mul_ps(iq3,jq2);
1547 qq33 = _mm_mul_ps(iq3,jq3);
1549 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1550 rcutoff_scalar = fr->rcoulomb;
1551 rcutoff = _mm_set1_ps(rcutoff_scalar);
1552 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
1554 rswitch_scalar = fr->rcoulomb_switch;
1555 rswitch = _mm_set1_ps(rswitch_scalar);
1556 /* Setup switch parameters */
1557 d_scalar = rcutoff_scalar-rswitch_scalar;
1558 d = _mm_set1_ps(d_scalar);
1559 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
1560 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1561 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1562 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
1563 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1564 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1566 /* Avoid stupid compiler warnings */
1567 jnrA = jnrB = jnrC = jnrD = 0;
1568 j_coord_offsetA = 0;
1569 j_coord_offsetB = 0;
1570 j_coord_offsetC = 0;
1571 j_coord_offsetD = 0;
1576 for(iidx=0;iidx<4*DIM;iidx++)
1578 scratch[iidx] = 0.0;
1581 /* Start outer loop over neighborlists */
1582 for(iidx=0; iidx<nri; iidx++)
1584 /* Load shift vector for this list */
1585 i_shift_offset = DIM*shiftidx[iidx];
1587 /* Load limits for loop over neighbors */
1588 j_index_start = jindex[iidx];
1589 j_index_end = jindex[iidx+1];
1591 /* Get outer coordinate index */
1593 i_coord_offset = DIM*inr;
1595 /* Load i particle coords and add shift vector */
1596 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1597 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1599 fix1 = _mm_setzero_ps();
1600 fiy1 = _mm_setzero_ps();
1601 fiz1 = _mm_setzero_ps();
1602 fix2 = _mm_setzero_ps();
1603 fiy2 = _mm_setzero_ps();
1604 fiz2 = _mm_setzero_ps();
1605 fix3 = _mm_setzero_ps();
1606 fiy3 = _mm_setzero_ps();
1607 fiz3 = _mm_setzero_ps();
1609 /* Start inner kernel loop */
1610 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1613 /* Get j neighbor index, and coordinate index */
1615 jnrB = jjnr[jidx+1];
1616 jnrC = jjnr[jidx+2];
1617 jnrD = jjnr[jidx+3];
1618 j_coord_offsetA = DIM*jnrA;
1619 j_coord_offsetB = DIM*jnrB;
1620 j_coord_offsetC = DIM*jnrC;
1621 j_coord_offsetD = DIM*jnrD;
1623 /* load j atom coordinates */
1624 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1625 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1626 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1628 /* Calculate displacement vector */
1629 dx11 = _mm_sub_ps(ix1,jx1);
1630 dy11 = _mm_sub_ps(iy1,jy1);
1631 dz11 = _mm_sub_ps(iz1,jz1);
1632 dx12 = _mm_sub_ps(ix1,jx2);
1633 dy12 = _mm_sub_ps(iy1,jy2);
1634 dz12 = _mm_sub_ps(iz1,jz2);
1635 dx13 = _mm_sub_ps(ix1,jx3);
1636 dy13 = _mm_sub_ps(iy1,jy3);
1637 dz13 = _mm_sub_ps(iz1,jz3);
1638 dx21 = _mm_sub_ps(ix2,jx1);
1639 dy21 = _mm_sub_ps(iy2,jy1);
1640 dz21 = _mm_sub_ps(iz2,jz1);
1641 dx22 = _mm_sub_ps(ix2,jx2);
1642 dy22 = _mm_sub_ps(iy2,jy2);
1643 dz22 = _mm_sub_ps(iz2,jz2);
1644 dx23 = _mm_sub_ps(ix2,jx3);
1645 dy23 = _mm_sub_ps(iy2,jy3);
1646 dz23 = _mm_sub_ps(iz2,jz3);
1647 dx31 = _mm_sub_ps(ix3,jx1);
1648 dy31 = _mm_sub_ps(iy3,jy1);
1649 dz31 = _mm_sub_ps(iz3,jz1);
1650 dx32 = _mm_sub_ps(ix3,jx2);
1651 dy32 = _mm_sub_ps(iy3,jy2);
1652 dz32 = _mm_sub_ps(iz3,jz2);
1653 dx33 = _mm_sub_ps(ix3,jx3);
1654 dy33 = _mm_sub_ps(iy3,jy3);
1655 dz33 = _mm_sub_ps(iz3,jz3);
1657 /* Calculate squared distance and things based on it */
1658 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1659 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1660 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1661 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1662 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1663 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1664 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1665 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1666 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1668 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1669 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1670 rinv13 = gmx_mm_invsqrt_ps(rsq13);
1671 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1672 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1673 rinv23 = gmx_mm_invsqrt_ps(rsq23);
1674 rinv31 = gmx_mm_invsqrt_ps(rsq31);
1675 rinv32 = gmx_mm_invsqrt_ps(rsq32);
1676 rinv33 = gmx_mm_invsqrt_ps(rsq33);
1678 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1679 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1680 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
1681 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1682 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1683 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
1684 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
1685 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
1686 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
1688 fjx1 = _mm_setzero_ps();
1689 fjy1 = _mm_setzero_ps();
1690 fjz1 = _mm_setzero_ps();
1691 fjx2 = _mm_setzero_ps();
1692 fjy2 = _mm_setzero_ps();
1693 fjz2 = _mm_setzero_ps();
1694 fjx3 = _mm_setzero_ps();
1695 fjy3 = _mm_setzero_ps();
1696 fjz3 = _mm_setzero_ps();
1698 /**************************
1699 * CALCULATE INTERACTIONS *
1700 **************************/
1702 if (gmx_mm_any_lt(rsq11,rcutoff2))
1705 r11 = _mm_mul_ps(rsq11,rinv11);
1707 /* EWALD ELECTROSTATICS */
1709 /* Analytical PME correction */
1710 zeta2 = _mm_mul_ps(beta2,rsq11);
1711 rinv3 = _mm_mul_ps(rinvsq11,rinv11);
1712 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1713 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1714 felec = _mm_mul_ps(qq11,felec);
1715 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1716 velec = _mm_nmacc_ps(pmecorrV,beta,rinv11);
1717 velec = _mm_mul_ps(qq11,velec);
1719 d = _mm_sub_ps(r11,rswitch);
1720 d = _mm_max_ps(d,_mm_setzero_ps());
1721 d2 = _mm_mul_ps(d,d);
1722 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1724 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1726 /* Evaluate switch function */
1727 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1728 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
1729 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
1733 fscal = _mm_and_ps(fscal,cutoff_mask);
1735 /* Update vectorial force */
1736 fix1 = _mm_macc_ps(dx11,fscal,fix1);
1737 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
1738 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
1740 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
1741 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
1742 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
1746 /**************************
1747 * CALCULATE INTERACTIONS *
1748 **************************/
1750 if (gmx_mm_any_lt(rsq12,rcutoff2))
1753 r12 = _mm_mul_ps(rsq12,rinv12);
1755 /* EWALD ELECTROSTATICS */
1757 /* Analytical PME correction */
1758 zeta2 = _mm_mul_ps(beta2,rsq12);
1759 rinv3 = _mm_mul_ps(rinvsq12,rinv12);
1760 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1761 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1762 felec = _mm_mul_ps(qq12,felec);
1763 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1764 velec = _mm_nmacc_ps(pmecorrV,beta,rinv12);
1765 velec = _mm_mul_ps(qq12,velec);
1767 d = _mm_sub_ps(r12,rswitch);
1768 d = _mm_max_ps(d,_mm_setzero_ps());
1769 d2 = _mm_mul_ps(d,d);
1770 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1772 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1774 /* Evaluate switch function */
1775 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1776 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
1777 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
1781 fscal = _mm_and_ps(fscal,cutoff_mask);
1783 /* Update vectorial force */
1784 fix1 = _mm_macc_ps(dx12,fscal,fix1);
1785 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
1786 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
1788 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
1789 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
1790 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
1794 /**************************
1795 * CALCULATE INTERACTIONS *
1796 **************************/
1798 if (gmx_mm_any_lt(rsq13,rcutoff2))
1801 r13 = _mm_mul_ps(rsq13,rinv13);
1803 /* EWALD ELECTROSTATICS */
1805 /* Analytical PME correction */
1806 zeta2 = _mm_mul_ps(beta2,rsq13);
1807 rinv3 = _mm_mul_ps(rinvsq13,rinv13);
1808 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1809 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1810 felec = _mm_mul_ps(qq13,felec);
1811 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1812 velec = _mm_nmacc_ps(pmecorrV,beta,rinv13);
1813 velec = _mm_mul_ps(qq13,velec);
1815 d = _mm_sub_ps(r13,rswitch);
1816 d = _mm_max_ps(d,_mm_setzero_ps());
1817 d2 = _mm_mul_ps(d,d);
1818 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1820 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1822 /* Evaluate switch function */
1823 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1824 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
1825 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
1829 fscal = _mm_and_ps(fscal,cutoff_mask);
1831 /* Update vectorial force */
1832 fix1 = _mm_macc_ps(dx13,fscal,fix1);
1833 fiy1 = _mm_macc_ps(dy13,fscal,fiy1);
1834 fiz1 = _mm_macc_ps(dz13,fscal,fiz1);
1836 fjx3 = _mm_macc_ps(dx13,fscal,fjx3);
1837 fjy3 = _mm_macc_ps(dy13,fscal,fjy3);
1838 fjz3 = _mm_macc_ps(dz13,fscal,fjz3);
1842 /**************************
1843 * CALCULATE INTERACTIONS *
1844 **************************/
1846 if (gmx_mm_any_lt(rsq21,rcutoff2))
1849 r21 = _mm_mul_ps(rsq21,rinv21);
1851 /* EWALD ELECTROSTATICS */
1853 /* Analytical PME correction */
1854 zeta2 = _mm_mul_ps(beta2,rsq21);
1855 rinv3 = _mm_mul_ps(rinvsq21,rinv21);
1856 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1857 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1858 felec = _mm_mul_ps(qq21,felec);
1859 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1860 velec = _mm_nmacc_ps(pmecorrV,beta,rinv21);
1861 velec = _mm_mul_ps(qq21,velec);
1863 d = _mm_sub_ps(r21,rswitch);
1864 d = _mm_max_ps(d,_mm_setzero_ps());
1865 d2 = _mm_mul_ps(d,d);
1866 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1868 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1870 /* Evaluate switch function */
1871 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1872 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
1873 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
1877 fscal = _mm_and_ps(fscal,cutoff_mask);
1879 /* Update vectorial force */
1880 fix2 = _mm_macc_ps(dx21,fscal,fix2);
1881 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
1882 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
1884 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
1885 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
1886 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
1890 /**************************
1891 * CALCULATE INTERACTIONS *
1892 **************************/
1894 if (gmx_mm_any_lt(rsq22,rcutoff2))
1897 r22 = _mm_mul_ps(rsq22,rinv22);
1899 /* EWALD ELECTROSTATICS */
1901 /* Analytical PME correction */
1902 zeta2 = _mm_mul_ps(beta2,rsq22);
1903 rinv3 = _mm_mul_ps(rinvsq22,rinv22);
1904 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1905 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1906 felec = _mm_mul_ps(qq22,felec);
1907 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1908 velec = _mm_nmacc_ps(pmecorrV,beta,rinv22);
1909 velec = _mm_mul_ps(qq22,velec);
1911 d = _mm_sub_ps(r22,rswitch);
1912 d = _mm_max_ps(d,_mm_setzero_ps());
1913 d2 = _mm_mul_ps(d,d);
1914 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1916 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1918 /* Evaluate switch function */
1919 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1920 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
1921 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
1925 fscal = _mm_and_ps(fscal,cutoff_mask);
1927 /* Update vectorial force */
1928 fix2 = _mm_macc_ps(dx22,fscal,fix2);
1929 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
1930 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
1932 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
1933 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
1934 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
1938 /**************************
1939 * CALCULATE INTERACTIONS *
1940 **************************/
1942 if (gmx_mm_any_lt(rsq23,rcutoff2))
1945 r23 = _mm_mul_ps(rsq23,rinv23);
1947 /* EWALD ELECTROSTATICS */
1949 /* Analytical PME correction */
1950 zeta2 = _mm_mul_ps(beta2,rsq23);
1951 rinv3 = _mm_mul_ps(rinvsq23,rinv23);
1952 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1953 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1954 felec = _mm_mul_ps(qq23,felec);
1955 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
1956 velec = _mm_nmacc_ps(pmecorrV,beta,rinv23);
1957 velec = _mm_mul_ps(qq23,velec);
1959 d = _mm_sub_ps(r23,rswitch);
1960 d = _mm_max_ps(d,_mm_setzero_ps());
1961 d2 = _mm_mul_ps(d,d);
1962 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1964 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1966 /* Evaluate switch function */
1967 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1968 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
1969 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
1973 fscal = _mm_and_ps(fscal,cutoff_mask);
1975 /* Update vectorial force */
1976 fix2 = _mm_macc_ps(dx23,fscal,fix2);
1977 fiy2 = _mm_macc_ps(dy23,fscal,fiy2);
1978 fiz2 = _mm_macc_ps(dz23,fscal,fiz2);
1980 fjx3 = _mm_macc_ps(dx23,fscal,fjx3);
1981 fjy3 = _mm_macc_ps(dy23,fscal,fjy3);
1982 fjz3 = _mm_macc_ps(dz23,fscal,fjz3);
1986 /**************************
1987 * CALCULATE INTERACTIONS *
1988 **************************/
1990 if (gmx_mm_any_lt(rsq31,rcutoff2))
1993 r31 = _mm_mul_ps(rsq31,rinv31);
1995 /* EWALD ELECTROSTATICS */
1997 /* Analytical PME correction */
1998 zeta2 = _mm_mul_ps(beta2,rsq31);
1999 rinv3 = _mm_mul_ps(rinvsq31,rinv31);
2000 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2001 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2002 felec = _mm_mul_ps(qq31,felec);
2003 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2004 velec = _mm_nmacc_ps(pmecorrV,beta,rinv31);
2005 velec = _mm_mul_ps(qq31,velec);
2007 d = _mm_sub_ps(r31,rswitch);
2008 d = _mm_max_ps(d,_mm_setzero_ps());
2009 d2 = _mm_mul_ps(d,d);
2010 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2012 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2014 /* Evaluate switch function */
2015 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2016 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
2017 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
2021 fscal = _mm_and_ps(fscal,cutoff_mask);
2023 /* Update vectorial force */
2024 fix3 = _mm_macc_ps(dx31,fscal,fix3);
2025 fiy3 = _mm_macc_ps(dy31,fscal,fiy3);
2026 fiz3 = _mm_macc_ps(dz31,fscal,fiz3);
2028 fjx1 = _mm_macc_ps(dx31,fscal,fjx1);
2029 fjy1 = _mm_macc_ps(dy31,fscal,fjy1);
2030 fjz1 = _mm_macc_ps(dz31,fscal,fjz1);
2034 /**************************
2035 * CALCULATE INTERACTIONS *
2036 **************************/
2038 if (gmx_mm_any_lt(rsq32,rcutoff2))
2041 r32 = _mm_mul_ps(rsq32,rinv32);
2043 /* EWALD ELECTROSTATICS */
2045 /* Analytical PME correction */
2046 zeta2 = _mm_mul_ps(beta2,rsq32);
2047 rinv3 = _mm_mul_ps(rinvsq32,rinv32);
2048 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2049 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2050 felec = _mm_mul_ps(qq32,felec);
2051 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2052 velec = _mm_nmacc_ps(pmecorrV,beta,rinv32);
2053 velec = _mm_mul_ps(qq32,velec);
2055 d = _mm_sub_ps(r32,rswitch);
2056 d = _mm_max_ps(d,_mm_setzero_ps());
2057 d2 = _mm_mul_ps(d,d);
2058 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2060 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2062 /* Evaluate switch function */
2063 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2064 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
2065 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
2069 fscal = _mm_and_ps(fscal,cutoff_mask);
2071 /* Update vectorial force */
2072 fix3 = _mm_macc_ps(dx32,fscal,fix3);
2073 fiy3 = _mm_macc_ps(dy32,fscal,fiy3);
2074 fiz3 = _mm_macc_ps(dz32,fscal,fiz3);
2076 fjx2 = _mm_macc_ps(dx32,fscal,fjx2);
2077 fjy2 = _mm_macc_ps(dy32,fscal,fjy2);
2078 fjz2 = _mm_macc_ps(dz32,fscal,fjz2);
2082 /**************************
2083 * CALCULATE INTERACTIONS *
2084 **************************/
2086 if (gmx_mm_any_lt(rsq33,rcutoff2))
2089 r33 = _mm_mul_ps(rsq33,rinv33);
2091 /* EWALD ELECTROSTATICS */
2093 /* Analytical PME correction */
2094 zeta2 = _mm_mul_ps(beta2,rsq33);
2095 rinv3 = _mm_mul_ps(rinvsq33,rinv33);
2096 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2097 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2098 felec = _mm_mul_ps(qq33,felec);
2099 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2100 velec = _mm_nmacc_ps(pmecorrV,beta,rinv33);
2101 velec = _mm_mul_ps(qq33,velec);
2103 d = _mm_sub_ps(r33,rswitch);
2104 d = _mm_max_ps(d,_mm_setzero_ps());
2105 d2 = _mm_mul_ps(d,d);
2106 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2108 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2110 /* Evaluate switch function */
2111 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2112 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
2113 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
2117 fscal = _mm_and_ps(fscal,cutoff_mask);
2119 /* Update vectorial force */
2120 fix3 = _mm_macc_ps(dx33,fscal,fix3);
2121 fiy3 = _mm_macc_ps(dy33,fscal,fiy3);
2122 fiz3 = _mm_macc_ps(dz33,fscal,fiz3);
2124 fjx3 = _mm_macc_ps(dx33,fscal,fjx3);
2125 fjy3 = _mm_macc_ps(dy33,fscal,fjy3);
2126 fjz3 = _mm_macc_ps(dz33,fscal,fjz3);
2130 fjptrA = f+j_coord_offsetA;
2131 fjptrB = f+j_coord_offsetB;
2132 fjptrC = f+j_coord_offsetC;
2133 fjptrD = f+j_coord_offsetD;
2135 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2136 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2138 /* Inner loop uses 450 flops */
2141 if(jidx<j_index_end)
2144 /* Get j neighbor index, and coordinate index */
2145 jnrlistA = jjnr[jidx];
2146 jnrlistB = jjnr[jidx+1];
2147 jnrlistC = jjnr[jidx+2];
2148 jnrlistD = jjnr[jidx+3];
2149 /* Sign of each element will be negative for non-real atoms.
2150 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2151 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2153 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2154 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
2155 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
2156 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
2157 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
2158 j_coord_offsetA = DIM*jnrA;
2159 j_coord_offsetB = DIM*jnrB;
2160 j_coord_offsetC = DIM*jnrC;
2161 j_coord_offsetD = DIM*jnrD;
2163 /* load j atom coordinates */
2164 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
2165 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
2166 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
2168 /* Calculate displacement vector */
2169 dx11 = _mm_sub_ps(ix1,jx1);
2170 dy11 = _mm_sub_ps(iy1,jy1);
2171 dz11 = _mm_sub_ps(iz1,jz1);
2172 dx12 = _mm_sub_ps(ix1,jx2);
2173 dy12 = _mm_sub_ps(iy1,jy2);
2174 dz12 = _mm_sub_ps(iz1,jz2);
2175 dx13 = _mm_sub_ps(ix1,jx3);
2176 dy13 = _mm_sub_ps(iy1,jy3);
2177 dz13 = _mm_sub_ps(iz1,jz3);
2178 dx21 = _mm_sub_ps(ix2,jx1);
2179 dy21 = _mm_sub_ps(iy2,jy1);
2180 dz21 = _mm_sub_ps(iz2,jz1);
2181 dx22 = _mm_sub_ps(ix2,jx2);
2182 dy22 = _mm_sub_ps(iy2,jy2);
2183 dz22 = _mm_sub_ps(iz2,jz2);
2184 dx23 = _mm_sub_ps(ix2,jx3);
2185 dy23 = _mm_sub_ps(iy2,jy3);
2186 dz23 = _mm_sub_ps(iz2,jz3);
2187 dx31 = _mm_sub_ps(ix3,jx1);
2188 dy31 = _mm_sub_ps(iy3,jy1);
2189 dz31 = _mm_sub_ps(iz3,jz1);
2190 dx32 = _mm_sub_ps(ix3,jx2);
2191 dy32 = _mm_sub_ps(iy3,jy2);
2192 dz32 = _mm_sub_ps(iz3,jz2);
2193 dx33 = _mm_sub_ps(ix3,jx3);
2194 dy33 = _mm_sub_ps(iy3,jy3);
2195 dz33 = _mm_sub_ps(iz3,jz3);
2197 /* Calculate squared distance and things based on it */
2198 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2199 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2200 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
2201 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2202 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2203 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
2204 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
2205 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
2206 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2208 rinv11 = gmx_mm_invsqrt_ps(rsq11);
2209 rinv12 = gmx_mm_invsqrt_ps(rsq12);
2210 rinv13 = gmx_mm_invsqrt_ps(rsq13);
2211 rinv21 = gmx_mm_invsqrt_ps(rsq21);
2212 rinv22 = gmx_mm_invsqrt_ps(rsq22);
2213 rinv23 = gmx_mm_invsqrt_ps(rsq23);
2214 rinv31 = gmx_mm_invsqrt_ps(rsq31);
2215 rinv32 = gmx_mm_invsqrt_ps(rsq32);
2216 rinv33 = gmx_mm_invsqrt_ps(rsq33);
2218 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
2219 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
2220 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
2221 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
2222 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
2223 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
2224 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
2225 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
2226 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
2228 fjx1 = _mm_setzero_ps();
2229 fjy1 = _mm_setzero_ps();
2230 fjz1 = _mm_setzero_ps();
2231 fjx2 = _mm_setzero_ps();
2232 fjy2 = _mm_setzero_ps();
2233 fjz2 = _mm_setzero_ps();
2234 fjx3 = _mm_setzero_ps();
2235 fjy3 = _mm_setzero_ps();
2236 fjz3 = _mm_setzero_ps();
2238 /**************************
2239 * CALCULATE INTERACTIONS *
2240 **************************/
2242 if (gmx_mm_any_lt(rsq11,rcutoff2))
2245 r11 = _mm_mul_ps(rsq11,rinv11);
2246 r11 = _mm_andnot_ps(dummy_mask,r11);
2248 /* EWALD ELECTROSTATICS */
2250 /* Analytical PME correction */
2251 zeta2 = _mm_mul_ps(beta2,rsq11);
2252 rinv3 = _mm_mul_ps(rinvsq11,rinv11);
2253 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2254 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2255 felec = _mm_mul_ps(qq11,felec);
2256 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2257 velec = _mm_nmacc_ps(pmecorrV,beta,rinv11);
2258 velec = _mm_mul_ps(qq11,velec);
2260 d = _mm_sub_ps(r11,rswitch);
2261 d = _mm_max_ps(d,_mm_setzero_ps());
2262 d2 = _mm_mul_ps(d,d);
2263 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2265 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2267 /* Evaluate switch function */
2268 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2269 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
2270 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
2274 fscal = _mm_and_ps(fscal,cutoff_mask);
2276 fscal = _mm_andnot_ps(dummy_mask,fscal);
2278 /* Update vectorial force */
2279 fix1 = _mm_macc_ps(dx11,fscal,fix1);
2280 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
2281 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
2283 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
2284 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
2285 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
2289 /**************************
2290 * CALCULATE INTERACTIONS *
2291 **************************/
2293 if (gmx_mm_any_lt(rsq12,rcutoff2))
2296 r12 = _mm_mul_ps(rsq12,rinv12);
2297 r12 = _mm_andnot_ps(dummy_mask,r12);
2299 /* EWALD ELECTROSTATICS */
2301 /* Analytical PME correction */
2302 zeta2 = _mm_mul_ps(beta2,rsq12);
2303 rinv3 = _mm_mul_ps(rinvsq12,rinv12);
2304 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2305 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2306 felec = _mm_mul_ps(qq12,felec);
2307 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2308 velec = _mm_nmacc_ps(pmecorrV,beta,rinv12);
2309 velec = _mm_mul_ps(qq12,velec);
2311 d = _mm_sub_ps(r12,rswitch);
2312 d = _mm_max_ps(d,_mm_setzero_ps());
2313 d2 = _mm_mul_ps(d,d);
2314 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2316 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2318 /* Evaluate switch function */
2319 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2320 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
2321 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
2325 fscal = _mm_and_ps(fscal,cutoff_mask);
2327 fscal = _mm_andnot_ps(dummy_mask,fscal);
2329 /* Update vectorial force */
2330 fix1 = _mm_macc_ps(dx12,fscal,fix1);
2331 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
2332 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
2334 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
2335 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
2336 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
2340 /**************************
2341 * CALCULATE INTERACTIONS *
2342 **************************/
2344 if (gmx_mm_any_lt(rsq13,rcutoff2))
2347 r13 = _mm_mul_ps(rsq13,rinv13);
2348 r13 = _mm_andnot_ps(dummy_mask,r13);
2350 /* EWALD ELECTROSTATICS */
2352 /* Analytical PME correction */
2353 zeta2 = _mm_mul_ps(beta2,rsq13);
2354 rinv3 = _mm_mul_ps(rinvsq13,rinv13);
2355 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2356 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2357 felec = _mm_mul_ps(qq13,felec);
2358 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2359 velec = _mm_nmacc_ps(pmecorrV,beta,rinv13);
2360 velec = _mm_mul_ps(qq13,velec);
2362 d = _mm_sub_ps(r13,rswitch);
2363 d = _mm_max_ps(d,_mm_setzero_ps());
2364 d2 = _mm_mul_ps(d,d);
2365 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2367 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2369 /* Evaluate switch function */
2370 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2371 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv13,_mm_mul_ps(velec,dsw)) );
2372 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
2376 fscal = _mm_and_ps(fscal,cutoff_mask);
2378 fscal = _mm_andnot_ps(dummy_mask,fscal);
2380 /* Update vectorial force */
2381 fix1 = _mm_macc_ps(dx13,fscal,fix1);
2382 fiy1 = _mm_macc_ps(dy13,fscal,fiy1);
2383 fiz1 = _mm_macc_ps(dz13,fscal,fiz1);
2385 fjx3 = _mm_macc_ps(dx13,fscal,fjx3);
2386 fjy3 = _mm_macc_ps(dy13,fscal,fjy3);
2387 fjz3 = _mm_macc_ps(dz13,fscal,fjz3);
2391 /**************************
2392 * CALCULATE INTERACTIONS *
2393 **************************/
2395 if (gmx_mm_any_lt(rsq21,rcutoff2))
2398 r21 = _mm_mul_ps(rsq21,rinv21);
2399 r21 = _mm_andnot_ps(dummy_mask,r21);
2401 /* EWALD ELECTROSTATICS */
2403 /* Analytical PME correction */
2404 zeta2 = _mm_mul_ps(beta2,rsq21);
2405 rinv3 = _mm_mul_ps(rinvsq21,rinv21);
2406 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2407 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2408 felec = _mm_mul_ps(qq21,felec);
2409 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2410 velec = _mm_nmacc_ps(pmecorrV,beta,rinv21);
2411 velec = _mm_mul_ps(qq21,velec);
2413 d = _mm_sub_ps(r21,rswitch);
2414 d = _mm_max_ps(d,_mm_setzero_ps());
2415 d2 = _mm_mul_ps(d,d);
2416 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2418 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2420 /* Evaluate switch function */
2421 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2422 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
2423 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
2427 fscal = _mm_and_ps(fscal,cutoff_mask);
2429 fscal = _mm_andnot_ps(dummy_mask,fscal);
2431 /* Update vectorial force */
2432 fix2 = _mm_macc_ps(dx21,fscal,fix2);
2433 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
2434 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
2436 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
2437 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
2438 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
2442 /**************************
2443 * CALCULATE INTERACTIONS *
2444 **************************/
2446 if (gmx_mm_any_lt(rsq22,rcutoff2))
2449 r22 = _mm_mul_ps(rsq22,rinv22);
2450 r22 = _mm_andnot_ps(dummy_mask,r22);
2452 /* EWALD ELECTROSTATICS */
2454 /* Analytical PME correction */
2455 zeta2 = _mm_mul_ps(beta2,rsq22);
2456 rinv3 = _mm_mul_ps(rinvsq22,rinv22);
2457 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2458 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2459 felec = _mm_mul_ps(qq22,felec);
2460 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2461 velec = _mm_nmacc_ps(pmecorrV,beta,rinv22);
2462 velec = _mm_mul_ps(qq22,velec);
2464 d = _mm_sub_ps(r22,rswitch);
2465 d = _mm_max_ps(d,_mm_setzero_ps());
2466 d2 = _mm_mul_ps(d,d);
2467 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2469 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2471 /* Evaluate switch function */
2472 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2473 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
2474 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
2478 fscal = _mm_and_ps(fscal,cutoff_mask);
2480 fscal = _mm_andnot_ps(dummy_mask,fscal);
2482 /* Update vectorial force */
2483 fix2 = _mm_macc_ps(dx22,fscal,fix2);
2484 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
2485 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
2487 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
2488 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
2489 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
2493 /**************************
2494 * CALCULATE INTERACTIONS *
2495 **************************/
2497 if (gmx_mm_any_lt(rsq23,rcutoff2))
2500 r23 = _mm_mul_ps(rsq23,rinv23);
2501 r23 = _mm_andnot_ps(dummy_mask,r23);
2503 /* EWALD ELECTROSTATICS */
2505 /* Analytical PME correction */
2506 zeta2 = _mm_mul_ps(beta2,rsq23);
2507 rinv3 = _mm_mul_ps(rinvsq23,rinv23);
2508 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2509 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2510 felec = _mm_mul_ps(qq23,felec);
2511 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2512 velec = _mm_nmacc_ps(pmecorrV,beta,rinv23);
2513 velec = _mm_mul_ps(qq23,velec);
2515 d = _mm_sub_ps(r23,rswitch);
2516 d = _mm_max_ps(d,_mm_setzero_ps());
2517 d2 = _mm_mul_ps(d,d);
2518 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2520 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2522 /* Evaluate switch function */
2523 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2524 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv23,_mm_mul_ps(velec,dsw)) );
2525 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
2529 fscal = _mm_and_ps(fscal,cutoff_mask);
2531 fscal = _mm_andnot_ps(dummy_mask,fscal);
2533 /* Update vectorial force */
2534 fix2 = _mm_macc_ps(dx23,fscal,fix2);
2535 fiy2 = _mm_macc_ps(dy23,fscal,fiy2);
2536 fiz2 = _mm_macc_ps(dz23,fscal,fiz2);
2538 fjx3 = _mm_macc_ps(dx23,fscal,fjx3);
2539 fjy3 = _mm_macc_ps(dy23,fscal,fjy3);
2540 fjz3 = _mm_macc_ps(dz23,fscal,fjz3);
2544 /**************************
2545 * CALCULATE INTERACTIONS *
2546 **************************/
2548 if (gmx_mm_any_lt(rsq31,rcutoff2))
2551 r31 = _mm_mul_ps(rsq31,rinv31);
2552 r31 = _mm_andnot_ps(dummy_mask,r31);
2554 /* EWALD ELECTROSTATICS */
2556 /* Analytical PME correction */
2557 zeta2 = _mm_mul_ps(beta2,rsq31);
2558 rinv3 = _mm_mul_ps(rinvsq31,rinv31);
2559 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2560 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2561 felec = _mm_mul_ps(qq31,felec);
2562 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2563 velec = _mm_nmacc_ps(pmecorrV,beta,rinv31);
2564 velec = _mm_mul_ps(qq31,velec);
2566 d = _mm_sub_ps(r31,rswitch);
2567 d = _mm_max_ps(d,_mm_setzero_ps());
2568 d2 = _mm_mul_ps(d,d);
2569 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2571 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2573 /* Evaluate switch function */
2574 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2575 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv31,_mm_mul_ps(velec,dsw)) );
2576 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
2580 fscal = _mm_and_ps(fscal,cutoff_mask);
2582 fscal = _mm_andnot_ps(dummy_mask,fscal);
2584 /* Update vectorial force */
2585 fix3 = _mm_macc_ps(dx31,fscal,fix3);
2586 fiy3 = _mm_macc_ps(dy31,fscal,fiy3);
2587 fiz3 = _mm_macc_ps(dz31,fscal,fiz3);
2589 fjx1 = _mm_macc_ps(dx31,fscal,fjx1);
2590 fjy1 = _mm_macc_ps(dy31,fscal,fjy1);
2591 fjz1 = _mm_macc_ps(dz31,fscal,fjz1);
2595 /**************************
2596 * CALCULATE INTERACTIONS *
2597 **************************/
2599 if (gmx_mm_any_lt(rsq32,rcutoff2))
2602 r32 = _mm_mul_ps(rsq32,rinv32);
2603 r32 = _mm_andnot_ps(dummy_mask,r32);
2605 /* EWALD ELECTROSTATICS */
2607 /* Analytical PME correction */
2608 zeta2 = _mm_mul_ps(beta2,rsq32);
2609 rinv3 = _mm_mul_ps(rinvsq32,rinv32);
2610 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2611 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2612 felec = _mm_mul_ps(qq32,felec);
2613 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2614 velec = _mm_nmacc_ps(pmecorrV,beta,rinv32);
2615 velec = _mm_mul_ps(qq32,velec);
2617 d = _mm_sub_ps(r32,rswitch);
2618 d = _mm_max_ps(d,_mm_setzero_ps());
2619 d2 = _mm_mul_ps(d,d);
2620 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2622 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2624 /* Evaluate switch function */
2625 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2626 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv32,_mm_mul_ps(velec,dsw)) );
2627 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
2631 fscal = _mm_and_ps(fscal,cutoff_mask);
2633 fscal = _mm_andnot_ps(dummy_mask,fscal);
2635 /* Update vectorial force */
2636 fix3 = _mm_macc_ps(dx32,fscal,fix3);
2637 fiy3 = _mm_macc_ps(dy32,fscal,fiy3);
2638 fiz3 = _mm_macc_ps(dz32,fscal,fiz3);
2640 fjx2 = _mm_macc_ps(dx32,fscal,fjx2);
2641 fjy2 = _mm_macc_ps(dy32,fscal,fjy2);
2642 fjz2 = _mm_macc_ps(dz32,fscal,fjz2);
2646 /**************************
2647 * CALCULATE INTERACTIONS *
2648 **************************/
2650 if (gmx_mm_any_lt(rsq33,rcutoff2))
2653 r33 = _mm_mul_ps(rsq33,rinv33);
2654 r33 = _mm_andnot_ps(dummy_mask,r33);
2656 /* EWALD ELECTROSTATICS */
2658 /* Analytical PME correction */
2659 zeta2 = _mm_mul_ps(beta2,rsq33);
2660 rinv3 = _mm_mul_ps(rinvsq33,rinv33);
2661 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
2662 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
2663 felec = _mm_mul_ps(qq33,felec);
2664 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
2665 velec = _mm_nmacc_ps(pmecorrV,beta,rinv33);
2666 velec = _mm_mul_ps(qq33,velec);
2668 d = _mm_sub_ps(r33,rswitch);
2669 d = _mm_max_ps(d,_mm_setzero_ps());
2670 d2 = _mm_mul_ps(d,d);
2671 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
2673 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
2675 /* Evaluate switch function */
2676 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2677 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv33,_mm_mul_ps(velec,dsw)) );
2678 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
2682 fscal = _mm_and_ps(fscal,cutoff_mask);
2684 fscal = _mm_andnot_ps(dummy_mask,fscal);
2686 /* Update vectorial force */
2687 fix3 = _mm_macc_ps(dx33,fscal,fix3);
2688 fiy3 = _mm_macc_ps(dy33,fscal,fiy3);
2689 fiz3 = _mm_macc_ps(dz33,fscal,fiz3);
2691 fjx3 = _mm_macc_ps(dx33,fscal,fjx3);
2692 fjy3 = _mm_macc_ps(dy33,fscal,fjy3);
2693 fjz3 = _mm_macc_ps(dz33,fscal,fjz3);
2697 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2698 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2699 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2700 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2702 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2703 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2705 /* Inner loop uses 459 flops */
2708 /* End of innermost loop */
2710 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2711 f+i_coord_offset+DIM,fshift+i_shift_offset);
2713 /* Increment number of inner iterations */
2714 inneriter += j_index_end - j_index_start;
2716 /* Outer loop uses 18 flops */
2719 /* Increment number of outer iterations */
2722 /* Update outer/inner flops */
2724 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*459);