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 sse4_1_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_sse4_1_single.h"
50 #include "kernelutil_x86_sse4_1_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_VF_sse4_1_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LennardJones
56 * Geometry: Water4-Water4
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_VF_sse4_1_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 SSE, 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 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
93 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
94 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
95 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
96 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
97 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
98 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
99 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
100 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
101 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
102 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
103 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
104 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
105 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
106 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
107 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
108 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
109 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
110 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
111 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
114 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
117 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
118 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
120 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
122 __m128 dummy_mask,cutoff_mask;
123 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
124 __m128 one = _mm_set1_ps(1.0);
125 __m128 two = _mm_set1_ps(2.0);
131 jindex = nlist->jindex;
133 shiftidx = nlist->shift;
135 shiftvec = fr->shift_vec[0];
136 fshift = fr->fshift[0];
137 facel = _mm_set1_ps(fr->epsfac);
138 charge = mdatoms->chargeA;
139 nvdwtype = fr->ntype;
141 vdwtype = mdatoms->typeA;
143 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
144 ewtab = fr->ic->tabq_coul_FDV0;
145 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
146 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
148 /* Setup water-specific parameters */
149 inr = nlist->iinr[0];
150 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
151 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
152 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
153 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
155 jq1 = _mm_set1_ps(charge[inr+1]);
156 jq2 = _mm_set1_ps(charge[inr+2]);
157 jq3 = _mm_set1_ps(charge[inr+3]);
158 vdwjidx0A = 2*vdwtype[inr+0];
159 c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
160 c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
161 qq11 = _mm_mul_ps(iq1,jq1);
162 qq12 = _mm_mul_ps(iq1,jq2);
163 qq13 = _mm_mul_ps(iq1,jq3);
164 qq21 = _mm_mul_ps(iq2,jq1);
165 qq22 = _mm_mul_ps(iq2,jq2);
166 qq23 = _mm_mul_ps(iq2,jq3);
167 qq31 = _mm_mul_ps(iq3,jq1);
168 qq32 = _mm_mul_ps(iq3,jq2);
169 qq33 = _mm_mul_ps(iq3,jq3);
171 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
172 rcutoff_scalar = fr->rcoulomb;
173 rcutoff = _mm_set1_ps(rcutoff_scalar);
174 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
176 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
177 rvdw = _mm_set1_ps(fr->rvdw);
179 /* Avoid stupid compiler warnings */
180 jnrA = jnrB = jnrC = jnrD = 0;
189 for(iidx=0;iidx<4*DIM;iidx++)
194 /* Start outer loop over neighborlists */
195 for(iidx=0; iidx<nri; iidx++)
197 /* Load shift vector for this list */
198 i_shift_offset = DIM*shiftidx[iidx];
200 /* Load limits for loop over neighbors */
201 j_index_start = jindex[iidx];
202 j_index_end = jindex[iidx+1];
204 /* Get outer coordinate index */
206 i_coord_offset = DIM*inr;
208 /* Load i particle coords and add shift vector */
209 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
210 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
212 fix0 = _mm_setzero_ps();
213 fiy0 = _mm_setzero_ps();
214 fiz0 = _mm_setzero_ps();
215 fix1 = _mm_setzero_ps();
216 fiy1 = _mm_setzero_ps();
217 fiz1 = _mm_setzero_ps();
218 fix2 = _mm_setzero_ps();
219 fiy2 = _mm_setzero_ps();
220 fiz2 = _mm_setzero_ps();
221 fix3 = _mm_setzero_ps();
222 fiy3 = _mm_setzero_ps();
223 fiz3 = _mm_setzero_ps();
225 /* Reset potential sums */
226 velecsum = _mm_setzero_ps();
227 vvdwsum = _mm_setzero_ps();
229 /* Start inner kernel loop */
230 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
233 /* Get j neighbor index, and coordinate index */
238 j_coord_offsetA = DIM*jnrA;
239 j_coord_offsetB = DIM*jnrB;
240 j_coord_offsetC = DIM*jnrC;
241 j_coord_offsetD = DIM*jnrD;
243 /* load j atom coordinates */
244 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
245 x+j_coord_offsetC,x+j_coord_offsetD,
246 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
247 &jy2,&jz2,&jx3,&jy3,&jz3);
249 /* Calculate displacement vector */
250 dx00 = _mm_sub_ps(ix0,jx0);
251 dy00 = _mm_sub_ps(iy0,jy0);
252 dz00 = _mm_sub_ps(iz0,jz0);
253 dx11 = _mm_sub_ps(ix1,jx1);
254 dy11 = _mm_sub_ps(iy1,jy1);
255 dz11 = _mm_sub_ps(iz1,jz1);
256 dx12 = _mm_sub_ps(ix1,jx2);
257 dy12 = _mm_sub_ps(iy1,jy2);
258 dz12 = _mm_sub_ps(iz1,jz2);
259 dx13 = _mm_sub_ps(ix1,jx3);
260 dy13 = _mm_sub_ps(iy1,jy3);
261 dz13 = _mm_sub_ps(iz1,jz3);
262 dx21 = _mm_sub_ps(ix2,jx1);
263 dy21 = _mm_sub_ps(iy2,jy1);
264 dz21 = _mm_sub_ps(iz2,jz1);
265 dx22 = _mm_sub_ps(ix2,jx2);
266 dy22 = _mm_sub_ps(iy2,jy2);
267 dz22 = _mm_sub_ps(iz2,jz2);
268 dx23 = _mm_sub_ps(ix2,jx3);
269 dy23 = _mm_sub_ps(iy2,jy3);
270 dz23 = _mm_sub_ps(iz2,jz3);
271 dx31 = _mm_sub_ps(ix3,jx1);
272 dy31 = _mm_sub_ps(iy3,jy1);
273 dz31 = _mm_sub_ps(iz3,jz1);
274 dx32 = _mm_sub_ps(ix3,jx2);
275 dy32 = _mm_sub_ps(iy3,jy2);
276 dz32 = _mm_sub_ps(iz3,jz2);
277 dx33 = _mm_sub_ps(ix3,jx3);
278 dy33 = _mm_sub_ps(iy3,jy3);
279 dz33 = _mm_sub_ps(iz3,jz3);
281 /* Calculate squared distance and things based on it */
282 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
283 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
284 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
285 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
286 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
287 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
288 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
289 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
290 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
291 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
293 rinv11 = gmx_mm_invsqrt_ps(rsq11);
294 rinv12 = gmx_mm_invsqrt_ps(rsq12);
295 rinv13 = gmx_mm_invsqrt_ps(rsq13);
296 rinv21 = gmx_mm_invsqrt_ps(rsq21);
297 rinv22 = gmx_mm_invsqrt_ps(rsq22);
298 rinv23 = gmx_mm_invsqrt_ps(rsq23);
299 rinv31 = gmx_mm_invsqrt_ps(rsq31);
300 rinv32 = gmx_mm_invsqrt_ps(rsq32);
301 rinv33 = gmx_mm_invsqrt_ps(rsq33);
303 rinvsq00 = gmx_mm_inv_ps(rsq00);
304 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
305 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
306 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
307 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
308 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
309 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
310 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
311 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
312 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
314 fjx0 = _mm_setzero_ps();
315 fjy0 = _mm_setzero_ps();
316 fjz0 = _mm_setzero_ps();
317 fjx1 = _mm_setzero_ps();
318 fjy1 = _mm_setzero_ps();
319 fjz1 = _mm_setzero_ps();
320 fjx2 = _mm_setzero_ps();
321 fjy2 = _mm_setzero_ps();
322 fjz2 = _mm_setzero_ps();
323 fjx3 = _mm_setzero_ps();
324 fjy3 = _mm_setzero_ps();
325 fjz3 = _mm_setzero_ps();
327 /**************************
328 * CALCULATE INTERACTIONS *
329 **************************/
331 if (gmx_mm_any_lt(rsq00,rcutoff2))
334 /* LENNARD-JONES DISPERSION/REPULSION */
336 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
337 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
338 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
339 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
340 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
341 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
343 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
345 /* Update potential sum for this i atom from the interaction with this j atom. */
346 vvdw = _mm_and_ps(vvdw,cutoff_mask);
347 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
351 fscal = _mm_and_ps(fscal,cutoff_mask);
353 /* Calculate temporary vectorial force */
354 tx = _mm_mul_ps(fscal,dx00);
355 ty = _mm_mul_ps(fscal,dy00);
356 tz = _mm_mul_ps(fscal,dz00);
358 /* Update vectorial force */
359 fix0 = _mm_add_ps(fix0,tx);
360 fiy0 = _mm_add_ps(fiy0,ty);
361 fiz0 = _mm_add_ps(fiz0,tz);
363 fjx0 = _mm_add_ps(fjx0,tx);
364 fjy0 = _mm_add_ps(fjy0,ty);
365 fjz0 = _mm_add_ps(fjz0,tz);
369 /**************************
370 * CALCULATE INTERACTIONS *
371 **************************/
373 if (gmx_mm_any_lt(rsq11,rcutoff2))
376 r11 = _mm_mul_ps(rsq11,rinv11);
378 /* EWALD ELECTROSTATICS */
380 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
381 ewrt = _mm_mul_ps(r11,ewtabscale);
382 ewitab = _mm_cvttps_epi32(ewrt);
383 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
384 ewitab = _mm_slli_epi32(ewitab,2);
385 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
386 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
387 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
388 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
389 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
390 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
391 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
392 velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec));
393 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
395 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
397 /* Update potential sum for this i atom from the interaction with this j atom. */
398 velec = _mm_and_ps(velec,cutoff_mask);
399 velecsum = _mm_add_ps(velecsum,velec);
403 fscal = _mm_and_ps(fscal,cutoff_mask);
405 /* Calculate temporary vectorial force */
406 tx = _mm_mul_ps(fscal,dx11);
407 ty = _mm_mul_ps(fscal,dy11);
408 tz = _mm_mul_ps(fscal,dz11);
410 /* Update vectorial force */
411 fix1 = _mm_add_ps(fix1,tx);
412 fiy1 = _mm_add_ps(fiy1,ty);
413 fiz1 = _mm_add_ps(fiz1,tz);
415 fjx1 = _mm_add_ps(fjx1,tx);
416 fjy1 = _mm_add_ps(fjy1,ty);
417 fjz1 = _mm_add_ps(fjz1,tz);
421 /**************************
422 * CALCULATE INTERACTIONS *
423 **************************/
425 if (gmx_mm_any_lt(rsq12,rcutoff2))
428 r12 = _mm_mul_ps(rsq12,rinv12);
430 /* EWALD ELECTROSTATICS */
432 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
433 ewrt = _mm_mul_ps(r12,ewtabscale);
434 ewitab = _mm_cvttps_epi32(ewrt);
435 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
436 ewitab = _mm_slli_epi32(ewitab,2);
437 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
438 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
439 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
440 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
441 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
442 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
443 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
444 velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec));
445 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
447 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
449 /* Update potential sum for this i atom from the interaction with this j atom. */
450 velec = _mm_and_ps(velec,cutoff_mask);
451 velecsum = _mm_add_ps(velecsum,velec);
455 fscal = _mm_and_ps(fscal,cutoff_mask);
457 /* Calculate temporary vectorial force */
458 tx = _mm_mul_ps(fscal,dx12);
459 ty = _mm_mul_ps(fscal,dy12);
460 tz = _mm_mul_ps(fscal,dz12);
462 /* Update vectorial force */
463 fix1 = _mm_add_ps(fix1,tx);
464 fiy1 = _mm_add_ps(fiy1,ty);
465 fiz1 = _mm_add_ps(fiz1,tz);
467 fjx2 = _mm_add_ps(fjx2,tx);
468 fjy2 = _mm_add_ps(fjy2,ty);
469 fjz2 = _mm_add_ps(fjz2,tz);
473 /**************************
474 * CALCULATE INTERACTIONS *
475 **************************/
477 if (gmx_mm_any_lt(rsq13,rcutoff2))
480 r13 = _mm_mul_ps(rsq13,rinv13);
482 /* EWALD ELECTROSTATICS */
484 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
485 ewrt = _mm_mul_ps(r13,ewtabscale);
486 ewitab = _mm_cvttps_epi32(ewrt);
487 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
488 ewitab = _mm_slli_epi32(ewitab,2);
489 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
490 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
491 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
492 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
493 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
494 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
495 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
496 velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_sub_ps(rinv13,sh_ewald),velec));
497 felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
499 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
501 /* Update potential sum for this i atom from the interaction with this j atom. */
502 velec = _mm_and_ps(velec,cutoff_mask);
503 velecsum = _mm_add_ps(velecsum,velec);
507 fscal = _mm_and_ps(fscal,cutoff_mask);
509 /* Calculate temporary vectorial force */
510 tx = _mm_mul_ps(fscal,dx13);
511 ty = _mm_mul_ps(fscal,dy13);
512 tz = _mm_mul_ps(fscal,dz13);
514 /* Update vectorial force */
515 fix1 = _mm_add_ps(fix1,tx);
516 fiy1 = _mm_add_ps(fiy1,ty);
517 fiz1 = _mm_add_ps(fiz1,tz);
519 fjx3 = _mm_add_ps(fjx3,tx);
520 fjy3 = _mm_add_ps(fjy3,ty);
521 fjz3 = _mm_add_ps(fjz3,tz);
525 /**************************
526 * CALCULATE INTERACTIONS *
527 **************************/
529 if (gmx_mm_any_lt(rsq21,rcutoff2))
532 r21 = _mm_mul_ps(rsq21,rinv21);
534 /* EWALD ELECTROSTATICS */
536 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
537 ewrt = _mm_mul_ps(r21,ewtabscale);
538 ewitab = _mm_cvttps_epi32(ewrt);
539 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
540 ewitab = _mm_slli_epi32(ewitab,2);
541 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
542 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
543 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
544 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
545 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
546 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
547 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
548 velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec));
549 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
551 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
553 /* Update potential sum for this i atom from the interaction with this j atom. */
554 velec = _mm_and_ps(velec,cutoff_mask);
555 velecsum = _mm_add_ps(velecsum,velec);
559 fscal = _mm_and_ps(fscal,cutoff_mask);
561 /* Calculate temporary vectorial force */
562 tx = _mm_mul_ps(fscal,dx21);
563 ty = _mm_mul_ps(fscal,dy21);
564 tz = _mm_mul_ps(fscal,dz21);
566 /* Update vectorial force */
567 fix2 = _mm_add_ps(fix2,tx);
568 fiy2 = _mm_add_ps(fiy2,ty);
569 fiz2 = _mm_add_ps(fiz2,tz);
571 fjx1 = _mm_add_ps(fjx1,tx);
572 fjy1 = _mm_add_ps(fjy1,ty);
573 fjz1 = _mm_add_ps(fjz1,tz);
577 /**************************
578 * CALCULATE INTERACTIONS *
579 **************************/
581 if (gmx_mm_any_lt(rsq22,rcutoff2))
584 r22 = _mm_mul_ps(rsq22,rinv22);
586 /* EWALD ELECTROSTATICS */
588 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
589 ewrt = _mm_mul_ps(r22,ewtabscale);
590 ewitab = _mm_cvttps_epi32(ewrt);
591 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
592 ewitab = _mm_slli_epi32(ewitab,2);
593 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
594 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
595 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
596 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
597 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
598 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
599 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
600 velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec));
601 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
603 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
605 /* Update potential sum for this i atom from the interaction with this j atom. */
606 velec = _mm_and_ps(velec,cutoff_mask);
607 velecsum = _mm_add_ps(velecsum,velec);
611 fscal = _mm_and_ps(fscal,cutoff_mask);
613 /* Calculate temporary vectorial force */
614 tx = _mm_mul_ps(fscal,dx22);
615 ty = _mm_mul_ps(fscal,dy22);
616 tz = _mm_mul_ps(fscal,dz22);
618 /* Update vectorial force */
619 fix2 = _mm_add_ps(fix2,tx);
620 fiy2 = _mm_add_ps(fiy2,ty);
621 fiz2 = _mm_add_ps(fiz2,tz);
623 fjx2 = _mm_add_ps(fjx2,tx);
624 fjy2 = _mm_add_ps(fjy2,ty);
625 fjz2 = _mm_add_ps(fjz2,tz);
629 /**************************
630 * CALCULATE INTERACTIONS *
631 **************************/
633 if (gmx_mm_any_lt(rsq23,rcutoff2))
636 r23 = _mm_mul_ps(rsq23,rinv23);
638 /* EWALD ELECTROSTATICS */
640 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
641 ewrt = _mm_mul_ps(r23,ewtabscale);
642 ewitab = _mm_cvttps_epi32(ewrt);
643 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
644 ewitab = _mm_slli_epi32(ewitab,2);
645 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
646 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
647 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
648 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
649 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
650 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
651 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
652 velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_sub_ps(rinv23,sh_ewald),velec));
653 felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
655 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
657 /* Update potential sum for this i atom from the interaction with this j atom. */
658 velec = _mm_and_ps(velec,cutoff_mask);
659 velecsum = _mm_add_ps(velecsum,velec);
663 fscal = _mm_and_ps(fscal,cutoff_mask);
665 /* Calculate temporary vectorial force */
666 tx = _mm_mul_ps(fscal,dx23);
667 ty = _mm_mul_ps(fscal,dy23);
668 tz = _mm_mul_ps(fscal,dz23);
670 /* Update vectorial force */
671 fix2 = _mm_add_ps(fix2,tx);
672 fiy2 = _mm_add_ps(fiy2,ty);
673 fiz2 = _mm_add_ps(fiz2,tz);
675 fjx3 = _mm_add_ps(fjx3,tx);
676 fjy3 = _mm_add_ps(fjy3,ty);
677 fjz3 = _mm_add_ps(fjz3,tz);
681 /**************************
682 * CALCULATE INTERACTIONS *
683 **************************/
685 if (gmx_mm_any_lt(rsq31,rcutoff2))
688 r31 = _mm_mul_ps(rsq31,rinv31);
690 /* EWALD ELECTROSTATICS */
692 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
693 ewrt = _mm_mul_ps(r31,ewtabscale);
694 ewitab = _mm_cvttps_epi32(ewrt);
695 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
696 ewitab = _mm_slli_epi32(ewitab,2);
697 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
698 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
699 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
700 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
701 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
702 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
703 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
704 velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_sub_ps(rinv31,sh_ewald),velec));
705 felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
707 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
709 /* Update potential sum for this i atom from the interaction with this j atom. */
710 velec = _mm_and_ps(velec,cutoff_mask);
711 velecsum = _mm_add_ps(velecsum,velec);
715 fscal = _mm_and_ps(fscal,cutoff_mask);
717 /* Calculate temporary vectorial force */
718 tx = _mm_mul_ps(fscal,dx31);
719 ty = _mm_mul_ps(fscal,dy31);
720 tz = _mm_mul_ps(fscal,dz31);
722 /* Update vectorial force */
723 fix3 = _mm_add_ps(fix3,tx);
724 fiy3 = _mm_add_ps(fiy3,ty);
725 fiz3 = _mm_add_ps(fiz3,tz);
727 fjx1 = _mm_add_ps(fjx1,tx);
728 fjy1 = _mm_add_ps(fjy1,ty);
729 fjz1 = _mm_add_ps(fjz1,tz);
733 /**************************
734 * CALCULATE INTERACTIONS *
735 **************************/
737 if (gmx_mm_any_lt(rsq32,rcutoff2))
740 r32 = _mm_mul_ps(rsq32,rinv32);
742 /* EWALD ELECTROSTATICS */
744 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
745 ewrt = _mm_mul_ps(r32,ewtabscale);
746 ewitab = _mm_cvttps_epi32(ewrt);
747 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
748 ewitab = _mm_slli_epi32(ewitab,2);
749 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
750 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
751 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
752 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
753 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
754 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
755 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
756 velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_sub_ps(rinv32,sh_ewald),velec));
757 felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
759 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
761 /* Update potential sum for this i atom from the interaction with this j atom. */
762 velec = _mm_and_ps(velec,cutoff_mask);
763 velecsum = _mm_add_ps(velecsum,velec);
767 fscal = _mm_and_ps(fscal,cutoff_mask);
769 /* Calculate temporary vectorial force */
770 tx = _mm_mul_ps(fscal,dx32);
771 ty = _mm_mul_ps(fscal,dy32);
772 tz = _mm_mul_ps(fscal,dz32);
774 /* Update vectorial force */
775 fix3 = _mm_add_ps(fix3,tx);
776 fiy3 = _mm_add_ps(fiy3,ty);
777 fiz3 = _mm_add_ps(fiz3,tz);
779 fjx2 = _mm_add_ps(fjx2,tx);
780 fjy2 = _mm_add_ps(fjy2,ty);
781 fjz2 = _mm_add_ps(fjz2,tz);
785 /**************************
786 * CALCULATE INTERACTIONS *
787 **************************/
789 if (gmx_mm_any_lt(rsq33,rcutoff2))
792 r33 = _mm_mul_ps(rsq33,rinv33);
794 /* EWALD ELECTROSTATICS */
796 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
797 ewrt = _mm_mul_ps(r33,ewtabscale);
798 ewitab = _mm_cvttps_epi32(ewrt);
799 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
800 ewitab = _mm_slli_epi32(ewitab,2);
801 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
802 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
803 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
804 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
805 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
806 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
807 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
808 velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_sub_ps(rinv33,sh_ewald),velec));
809 felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
811 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
813 /* Update potential sum for this i atom from the interaction with this j atom. */
814 velec = _mm_and_ps(velec,cutoff_mask);
815 velecsum = _mm_add_ps(velecsum,velec);
819 fscal = _mm_and_ps(fscal,cutoff_mask);
821 /* Calculate temporary vectorial force */
822 tx = _mm_mul_ps(fscal,dx33);
823 ty = _mm_mul_ps(fscal,dy33);
824 tz = _mm_mul_ps(fscal,dz33);
826 /* Update vectorial force */
827 fix3 = _mm_add_ps(fix3,tx);
828 fiy3 = _mm_add_ps(fiy3,ty);
829 fiz3 = _mm_add_ps(fiz3,tz);
831 fjx3 = _mm_add_ps(fjx3,tx);
832 fjy3 = _mm_add_ps(fjy3,ty);
833 fjz3 = _mm_add_ps(fjz3,tz);
837 fjptrA = f+j_coord_offsetA;
838 fjptrB = f+j_coord_offsetB;
839 fjptrC = f+j_coord_offsetC;
840 fjptrD = f+j_coord_offsetD;
842 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
843 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
844 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
846 /* Inner loop uses 458 flops */
852 /* Get j neighbor index, and coordinate index */
853 jnrlistA = jjnr[jidx];
854 jnrlistB = jjnr[jidx+1];
855 jnrlistC = jjnr[jidx+2];
856 jnrlistD = jjnr[jidx+3];
857 /* Sign of each element will be negative for non-real atoms.
858 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
859 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
861 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
862 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
863 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
864 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
865 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
866 j_coord_offsetA = DIM*jnrA;
867 j_coord_offsetB = DIM*jnrB;
868 j_coord_offsetC = DIM*jnrC;
869 j_coord_offsetD = DIM*jnrD;
871 /* load j atom coordinates */
872 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
873 x+j_coord_offsetC,x+j_coord_offsetD,
874 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
875 &jy2,&jz2,&jx3,&jy3,&jz3);
877 /* Calculate displacement vector */
878 dx00 = _mm_sub_ps(ix0,jx0);
879 dy00 = _mm_sub_ps(iy0,jy0);
880 dz00 = _mm_sub_ps(iz0,jz0);
881 dx11 = _mm_sub_ps(ix1,jx1);
882 dy11 = _mm_sub_ps(iy1,jy1);
883 dz11 = _mm_sub_ps(iz1,jz1);
884 dx12 = _mm_sub_ps(ix1,jx2);
885 dy12 = _mm_sub_ps(iy1,jy2);
886 dz12 = _mm_sub_ps(iz1,jz2);
887 dx13 = _mm_sub_ps(ix1,jx3);
888 dy13 = _mm_sub_ps(iy1,jy3);
889 dz13 = _mm_sub_ps(iz1,jz3);
890 dx21 = _mm_sub_ps(ix2,jx1);
891 dy21 = _mm_sub_ps(iy2,jy1);
892 dz21 = _mm_sub_ps(iz2,jz1);
893 dx22 = _mm_sub_ps(ix2,jx2);
894 dy22 = _mm_sub_ps(iy2,jy2);
895 dz22 = _mm_sub_ps(iz2,jz2);
896 dx23 = _mm_sub_ps(ix2,jx3);
897 dy23 = _mm_sub_ps(iy2,jy3);
898 dz23 = _mm_sub_ps(iz2,jz3);
899 dx31 = _mm_sub_ps(ix3,jx1);
900 dy31 = _mm_sub_ps(iy3,jy1);
901 dz31 = _mm_sub_ps(iz3,jz1);
902 dx32 = _mm_sub_ps(ix3,jx2);
903 dy32 = _mm_sub_ps(iy3,jy2);
904 dz32 = _mm_sub_ps(iz3,jz2);
905 dx33 = _mm_sub_ps(ix3,jx3);
906 dy33 = _mm_sub_ps(iy3,jy3);
907 dz33 = _mm_sub_ps(iz3,jz3);
909 /* Calculate squared distance and things based on it */
910 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
911 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
912 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
913 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
914 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
915 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
916 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
917 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
918 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
919 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
921 rinv11 = gmx_mm_invsqrt_ps(rsq11);
922 rinv12 = gmx_mm_invsqrt_ps(rsq12);
923 rinv13 = gmx_mm_invsqrt_ps(rsq13);
924 rinv21 = gmx_mm_invsqrt_ps(rsq21);
925 rinv22 = gmx_mm_invsqrt_ps(rsq22);
926 rinv23 = gmx_mm_invsqrt_ps(rsq23);
927 rinv31 = gmx_mm_invsqrt_ps(rsq31);
928 rinv32 = gmx_mm_invsqrt_ps(rsq32);
929 rinv33 = gmx_mm_invsqrt_ps(rsq33);
931 rinvsq00 = gmx_mm_inv_ps(rsq00);
932 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
933 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
934 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
935 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
936 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
937 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
938 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
939 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
940 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
942 fjx0 = _mm_setzero_ps();
943 fjy0 = _mm_setzero_ps();
944 fjz0 = _mm_setzero_ps();
945 fjx1 = _mm_setzero_ps();
946 fjy1 = _mm_setzero_ps();
947 fjz1 = _mm_setzero_ps();
948 fjx2 = _mm_setzero_ps();
949 fjy2 = _mm_setzero_ps();
950 fjz2 = _mm_setzero_ps();
951 fjx3 = _mm_setzero_ps();
952 fjy3 = _mm_setzero_ps();
953 fjz3 = _mm_setzero_ps();
955 /**************************
956 * CALCULATE INTERACTIONS *
957 **************************/
959 if (gmx_mm_any_lt(rsq00,rcutoff2))
962 /* LENNARD-JONES DISPERSION/REPULSION */
964 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
965 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
966 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
967 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
968 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
969 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
971 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
973 /* Update potential sum for this i atom from the interaction with this j atom. */
974 vvdw = _mm_and_ps(vvdw,cutoff_mask);
975 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
976 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
980 fscal = _mm_and_ps(fscal,cutoff_mask);
982 fscal = _mm_andnot_ps(dummy_mask,fscal);
984 /* Calculate temporary vectorial force */
985 tx = _mm_mul_ps(fscal,dx00);
986 ty = _mm_mul_ps(fscal,dy00);
987 tz = _mm_mul_ps(fscal,dz00);
989 /* Update vectorial force */
990 fix0 = _mm_add_ps(fix0,tx);
991 fiy0 = _mm_add_ps(fiy0,ty);
992 fiz0 = _mm_add_ps(fiz0,tz);
994 fjx0 = _mm_add_ps(fjx0,tx);
995 fjy0 = _mm_add_ps(fjy0,ty);
996 fjz0 = _mm_add_ps(fjz0,tz);
1000 /**************************
1001 * CALCULATE INTERACTIONS *
1002 **************************/
1004 if (gmx_mm_any_lt(rsq11,rcutoff2))
1007 r11 = _mm_mul_ps(rsq11,rinv11);
1008 r11 = _mm_andnot_ps(dummy_mask,r11);
1010 /* EWALD ELECTROSTATICS */
1012 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1013 ewrt = _mm_mul_ps(r11,ewtabscale);
1014 ewitab = _mm_cvttps_epi32(ewrt);
1015 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1016 ewitab = _mm_slli_epi32(ewitab,2);
1017 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1018 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1019 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1020 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1021 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1022 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1023 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1024 velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec));
1025 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1027 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
1029 /* Update potential sum for this i atom from the interaction with this j atom. */
1030 velec = _mm_and_ps(velec,cutoff_mask);
1031 velec = _mm_andnot_ps(dummy_mask,velec);
1032 velecsum = _mm_add_ps(velecsum,velec);
1036 fscal = _mm_and_ps(fscal,cutoff_mask);
1038 fscal = _mm_andnot_ps(dummy_mask,fscal);
1040 /* Calculate temporary vectorial force */
1041 tx = _mm_mul_ps(fscal,dx11);
1042 ty = _mm_mul_ps(fscal,dy11);
1043 tz = _mm_mul_ps(fscal,dz11);
1045 /* Update vectorial force */
1046 fix1 = _mm_add_ps(fix1,tx);
1047 fiy1 = _mm_add_ps(fiy1,ty);
1048 fiz1 = _mm_add_ps(fiz1,tz);
1050 fjx1 = _mm_add_ps(fjx1,tx);
1051 fjy1 = _mm_add_ps(fjy1,ty);
1052 fjz1 = _mm_add_ps(fjz1,tz);
1056 /**************************
1057 * CALCULATE INTERACTIONS *
1058 **************************/
1060 if (gmx_mm_any_lt(rsq12,rcutoff2))
1063 r12 = _mm_mul_ps(rsq12,rinv12);
1064 r12 = _mm_andnot_ps(dummy_mask,r12);
1066 /* EWALD ELECTROSTATICS */
1068 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1069 ewrt = _mm_mul_ps(r12,ewtabscale);
1070 ewitab = _mm_cvttps_epi32(ewrt);
1071 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1072 ewitab = _mm_slli_epi32(ewitab,2);
1073 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1074 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1075 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1076 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1077 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1078 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1079 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1080 velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec));
1081 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1083 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
1085 /* Update potential sum for this i atom from the interaction with this j atom. */
1086 velec = _mm_and_ps(velec,cutoff_mask);
1087 velec = _mm_andnot_ps(dummy_mask,velec);
1088 velecsum = _mm_add_ps(velecsum,velec);
1092 fscal = _mm_and_ps(fscal,cutoff_mask);
1094 fscal = _mm_andnot_ps(dummy_mask,fscal);
1096 /* Calculate temporary vectorial force */
1097 tx = _mm_mul_ps(fscal,dx12);
1098 ty = _mm_mul_ps(fscal,dy12);
1099 tz = _mm_mul_ps(fscal,dz12);
1101 /* Update vectorial force */
1102 fix1 = _mm_add_ps(fix1,tx);
1103 fiy1 = _mm_add_ps(fiy1,ty);
1104 fiz1 = _mm_add_ps(fiz1,tz);
1106 fjx2 = _mm_add_ps(fjx2,tx);
1107 fjy2 = _mm_add_ps(fjy2,ty);
1108 fjz2 = _mm_add_ps(fjz2,tz);
1112 /**************************
1113 * CALCULATE INTERACTIONS *
1114 **************************/
1116 if (gmx_mm_any_lt(rsq13,rcutoff2))
1119 r13 = _mm_mul_ps(rsq13,rinv13);
1120 r13 = _mm_andnot_ps(dummy_mask,r13);
1122 /* EWALD ELECTROSTATICS */
1124 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1125 ewrt = _mm_mul_ps(r13,ewtabscale);
1126 ewitab = _mm_cvttps_epi32(ewrt);
1127 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1128 ewitab = _mm_slli_epi32(ewitab,2);
1129 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1130 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1131 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1132 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1133 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1134 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1135 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1136 velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_sub_ps(rinv13,sh_ewald),velec));
1137 felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
1139 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
1141 /* Update potential sum for this i atom from the interaction with this j atom. */
1142 velec = _mm_and_ps(velec,cutoff_mask);
1143 velec = _mm_andnot_ps(dummy_mask,velec);
1144 velecsum = _mm_add_ps(velecsum,velec);
1148 fscal = _mm_and_ps(fscal,cutoff_mask);
1150 fscal = _mm_andnot_ps(dummy_mask,fscal);
1152 /* Calculate temporary vectorial force */
1153 tx = _mm_mul_ps(fscal,dx13);
1154 ty = _mm_mul_ps(fscal,dy13);
1155 tz = _mm_mul_ps(fscal,dz13);
1157 /* Update vectorial force */
1158 fix1 = _mm_add_ps(fix1,tx);
1159 fiy1 = _mm_add_ps(fiy1,ty);
1160 fiz1 = _mm_add_ps(fiz1,tz);
1162 fjx3 = _mm_add_ps(fjx3,tx);
1163 fjy3 = _mm_add_ps(fjy3,ty);
1164 fjz3 = _mm_add_ps(fjz3,tz);
1168 /**************************
1169 * CALCULATE INTERACTIONS *
1170 **************************/
1172 if (gmx_mm_any_lt(rsq21,rcutoff2))
1175 r21 = _mm_mul_ps(rsq21,rinv21);
1176 r21 = _mm_andnot_ps(dummy_mask,r21);
1178 /* EWALD ELECTROSTATICS */
1180 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1181 ewrt = _mm_mul_ps(r21,ewtabscale);
1182 ewitab = _mm_cvttps_epi32(ewrt);
1183 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1184 ewitab = _mm_slli_epi32(ewitab,2);
1185 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1186 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1187 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1188 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1189 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1190 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1191 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1192 velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec));
1193 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1195 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
1197 /* Update potential sum for this i atom from the interaction with this j atom. */
1198 velec = _mm_and_ps(velec,cutoff_mask);
1199 velec = _mm_andnot_ps(dummy_mask,velec);
1200 velecsum = _mm_add_ps(velecsum,velec);
1204 fscal = _mm_and_ps(fscal,cutoff_mask);
1206 fscal = _mm_andnot_ps(dummy_mask,fscal);
1208 /* Calculate temporary vectorial force */
1209 tx = _mm_mul_ps(fscal,dx21);
1210 ty = _mm_mul_ps(fscal,dy21);
1211 tz = _mm_mul_ps(fscal,dz21);
1213 /* Update vectorial force */
1214 fix2 = _mm_add_ps(fix2,tx);
1215 fiy2 = _mm_add_ps(fiy2,ty);
1216 fiz2 = _mm_add_ps(fiz2,tz);
1218 fjx1 = _mm_add_ps(fjx1,tx);
1219 fjy1 = _mm_add_ps(fjy1,ty);
1220 fjz1 = _mm_add_ps(fjz1,tz);
1224 /**************************
1225 * CALCULATE INTERACTIONS *
1226 **************************/
1228 if (gmx_mm_any_lt(rsq22,rcutoff2))
1231 r22 = _mm_mul_ps(rsq22,rinv22);
1232 r22 = _mm_andnot_ps(dummy_mask,r22);
1234 /* EWALD ELECTROSTATICS */
1236 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1237 ewrt = _mm_mul_ps(r22,ewtabscale);
1238 ewitab = _mm_cvttps_epi32(ewrt);
1239 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1240 ewitab = _mm_slli_epi32(ewitab,2);
1241 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1242 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1243 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1244 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1245 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1246 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1247 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1248 velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec));
1249 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1251 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
1253 /* Update potential sum for this i atom from the interaction with this j atom. */
1254 velec = _mm_and_ps(velec,cutoff_mask);
1255 velec = _mm_andnot_ps(dummy_mask,velec);
1256 velecsum = _mm_add_ps(velecsum,velec);
1260 fscal = _mm_and_ps(fscal,cutoff_mask);
1262 fscal = _mm_andnot_ps(dummy_mask,fscal);
1264 /* Calculate temporary vectorial force */
1265 tx = _mm_mul_ps(fscal,dx22);
1266 ty = _mm_mul_ps(fscal,dy22);
1267 tz = _mm_mul_ps(fscal,dz22);
1269 /* Update vectorial force */
1270 fix2 = _mm_add_ps(fix2,tx);
1271 fiy2 = _mm_add_ps(fiy2,ty);
1272 fiz2 = _mm_add_ps(fiz2,tz);
1274 fjx2 = _mm_add_ps(fjx2,tx);
1275 fjy2 = _mm_add_ps(fjy2,ty);
1276 fjz2 = _mm_add_ps(fjz2,tz);
1280 /**************************
1281 * CALCULATE INTERACTIONS *
1282 **************************/
1284 if (gmx_mm_any_lt(rsq23,rcutoff2))
1287 r23 = _mm_mul_ps(rsq23,rinv23);
1288 r23 = _mm_andnot_ps(dummy_mask,r23);
1290 /* EWALD ELECTROSTATICS */
1292 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1293 ewrt = _mm_mul_ps(r23,ewtabscale);
1294 ewitab = _mm_cvttps_epi32(ewrt);
1295 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1296 ewitab = _mm_slli_epi32(ewitab,2);
1297 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1298 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1299 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1300 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1301 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1302 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1303 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1304 velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_sub_ps(rinv23,sh_ewald),velec));
1305 felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
1307 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
1309 /* Update potential sum for this i atom from the interaction with this j atom. */
1310 velec = _mm_and_ps(velec,cutoff_mask);
1311 velec = _mm_andnot_ps(dummy_mask,velec);
1312 velecsum = _mm_add_ps(velecsum,velec);
1316 fscal = _mm_and_ps(fscal,cutoff_mask);
1318 fscal = _mm_andnot_ps(dummy_mask,fscal);
1320 /* Calculate temporary vectorial force */
1321 tx = _mm_mul_ps(fscal,dx23);
1322 ty = _mm_mul_ps(fscal,dy23);
1323 tz = _mm_mul_ps(fscal,dz23);
1325 /* Update vectorial force */
1326 fix2 = _mm_add_ps(fix2,tx);
1327 fiy2 = _mm_add_ps(fiy2,ty);
1328 fiz2 = _mm_add_ps(fiz2,tz);
1330 fjx3 = _mm_add_ps(fjx3,tx);
1331 fjy3 = _mm_add_ps(fjy3,ty);
1332 fjz3 = _mm_add_ps(fjz3,tz);
1336 /**************************
1337 * CALCULATE INTERACTIONS *
1338 **************************/
1340 if (gmx_mm_any_lt(rsq31,rcutoff2))
1343 r31 = _mm_mul_ps(rsq31,rinv31);
1344 r31 = _mm_andnot_ps(dummy_mask,r31);
1346 /* EWALD ELECTROSTATICS */
1348 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1349 ewrt = _mm_mul_ps(r31,ewtabscale);
1350 ewitab = _mm_cvttps_epi32(ewrt);
1351 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1352 ewitab = _mm_slli_epi32(ewitab,2);
1353 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1354 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1355 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1356 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1357 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1358 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1359 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1360 velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_sub_ps(rinv31,sh_ewald),velec));
1361 felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
1363 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
1365 /* Update potential sum for this i atom from the interaction with this j atom. */
1366 velec = _mm_and_ps(velec,cutoff_mask);
1367 velec = _mm_andnot_ps(dummy_mask,velec);
1368 velecsum = _mm_add_ps(velecsum,velec);
1372 fscal = _mm_and_ps(fscal,cutoff_mask);
1374 fscal = _mm_andnot_ps(dummy_mask,fscal);
1376 /* Calculate temporary vectorial force */
1377 tx = _mm_mul_ps(fscal,dx31);
1378 ty = _mm_mul_ps(fscal,dy31);
1379 tz = _mm_mul_ps(fscal,dz31);
1381 /* Update vectorial force */
1382 fix3 = _mm_add_ps(fix3,tx);
1383 fiy3 = _mm_add_ps(fiy3,ty);
1384 fiz3 = _mm_add_ps(fiz3,tz);
1386 fjx1 = _mm_add_ps(fjx1,tx);
1387 fjy1 = _mm_add_ps(fjy1,ty);
1388 fjz1 = _mm_add_ps(fjz1,tz);
1392 /**************************
1393 * CALCULATE INTERACTIONS *
1394 **************************/
1396 if (gmx_mm_any_lt(rsq32,rcutoff2))
1399 r32 = _mm_mul_ps(rsq32,rinv32);
1400 r32 = _mm_andnot_ps(dummy_mask,r32);
1402 /* EWALD ELECTROSTATICS */
1404 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1405 ewrt = _mm_mul_ps(r32,ewtabscale);
1406 ewitab = _mm_cvttps_epi32(ewrt);
1407 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1408 ewitab = _mm_slli_epi32(ewitab,2);
1409 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1410 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1411 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1412 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1413 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1414 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1415 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1416 velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_sub_ps(rinv32,sh_ewald),velec));
1417 felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
1419 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
1421 /* Update potential sum for this i atom from the interaction with this j atom. */
1422 velec = _mm_and_ps(velec,cutoff_mask);
1423 velec = _mm_andnot_ps(dummy_mask,velec);
1424 velecsum = _mm_add_ps(velecsum,velec);
1428 fscal = _mm_and_ps(fscal,cutoff_mask);
1430 fscal = _mm_andnot_ps(dummy_mask,fscal);
1432 /* Calculate temporary vectorial force */
1433 tx = _mm_mul_ps(fscal,dx32);
1434 ty = _mm_mul_ps(fscal,dy32);
1435 tz = _mm_mul_ps(fscal,dz32);
1437 /* Update vectorial force */
1438 fix3 = _mm_add_ps(fix3,tx);
1439 fiy3 = _mm_add_ps(fiy3,ty);
1440 fiz3 = _mm_add_ps(fiz3,tz);
1442 fjx2 = _mm_add_ps(fjx2,tx);
1443 fjy2 = _mm_add_ps(fjy2,ty);
1444 fjz2 = _mm_add_ps(fjz2,tz);
1448 /**************************
1449 * CALCULATE INTERACTIONS *
1450 **************************/
1452 if (gmx_mm_any_lt(rsq33,rcutoff2))
1455 r33 = _mm_mul_ps(rsq33,rinv33);
1456 r33 = _mm_andnot_ps(dummy_mask,r33);
1458 /* EWALD ELECTROSTATICS */
1460 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1461 ewrt = _mm_mul_ps(r33,ewtabscale);
1462 ewitab = _mm_cvttps_epi32(ewrt);
1463 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1464 ewitab = _mm_slli_epi32(ewitab,2);
1465 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1466 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1467 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1468 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1469 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1470 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1471 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1472 velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_sub_ps(rinv33,sh_ewald),velec));
1473 felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
1475 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
1477 /* Update potential sum for this i atom from the interaction with this j atom. */
1478 velec = _mm_and_ps(velec,cutoff_mask);
1479 velec = _mm_andnot_ps(dummy_mask,velec);
1480 velecsum = _mm_add_ps(velecsum,velec);
1484 fscal = _mm_and_ps(fscal,cutoff_mask);
1486 fscal = _mm_andnot_ps(dummy_mask,fscal);
1488 /* Calculate temporary vectorial force */
1489 tx = _mm_mul_ps(fscal,dx33);
1490 ty = _mm_mul_ps(fscal,dy33);
1491 tz = _mm_mul_ps(fscal,dz33);
1493 /* Update vectorial force */
1494 fix3 = _mm_add_ps(fix3,tx);
1495 fiy3 = _mm_add_ps(fiy3,ty);
1496 fiz3 = _mm_add_ps(fiz3,tz);
1498 fjx3 = _mm_add_ps(fjx3,tx);
1499 fjy3 = _mm_add_ps(fjy3,ty);
1500 fjz3 = _mm_add_ps(fjz3,tz);
1504 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1505 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1506 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1507 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1509 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1510 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1511 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1513 /* Inner loop uses 467 flops */
1516 /* End of innermost loop */
1518 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1519 f+i_coord_offset,fshift+i_shift_offset);
1522 /* Update potential energies */
1523 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1524 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1526 /* Increment number of inner iterations */
1527 inneriter += j_index_end - j_index_start;
1529 /* Outer loop uses 26 flops */
1532 /* Increment number of outer iterations */
1535 /* Update outer/inner flops */
1537 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*467);
1540 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_F_sse4_1_single
1541 * Electrostatics interaction: Ewald
1542 * VdW interaction: LennardJones
1543 * Geometry: Water4-Water4
1544 * Calculate force/pot: Force
1547 nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_F_sse4_1_single
1548 (t_nblist * gmx_restrict nlist,
1549 rvec * gmx_restrict xx,
1550 rvec * gmx_restrict ff,
1551 t_forcerec * gmx_restrict fr,
1552 t_mdatoms * gmx_restrict mdatoms,
1553 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1554 t_nrnb * gmx_restrict nrnb)
1556 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1557 * just 0 for non-waters.
1558 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1559 * jnr indices corresponding to data put in the four positions in the SIMD register.
1561 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1562 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1563 int jnrA,jnrB,jnrC,jnrD;
1564 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1565 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1566 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1567 real rcutoff_scalar;
1568 real *shiftvec,*fshift,*x,*f;
1569 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1570 real scratch[4*DIM];
1571 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1573 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1575 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1577 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1579 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1580 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1581 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1582 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1583 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1584 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1585 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1586 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1587 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1588 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1589 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1590 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1591 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1592 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1593 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1594 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1595 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1596 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1597 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1598 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1601 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1604 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
1605 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
1607 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1609 __m128 dummy_mask,cutoff_mask;
1610 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1611 __m128 one = _mm_set1_ps(1.0);
1612 __m128 two = _mm_set1_ps(2.0);
1618 jindex = nlist->jindex;
1620 shiftidx = nlist->shift;
1622 shiftvec = fr->shift_vec[0];
1623 fshift = fr->fshift[0];
1624 facel = _mm_set1_ps(fr->epsfac);
1625 charge = mdatoms->chargeA;
1626 nvdwtype = fr->ntype;
1627 vdwparam = fr->nbfp;
1628 vdwtype = mdatoms->typeA;
1630 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
1631 ewtab = fr->ic->tabq_coul_F;
1632 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
1633 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1635 /* Setup water-specific parameters */
1636 inr = nlist->iinr[0];
1637 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1638 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1639 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1640 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1642 jq1 = _mm_set1_ps(charge[inr+1]);
1643 jq2 = _mm_set1_ps(charge[inr+2]);
1644 jq3 = _mm_set1_ps(charge[inr+3]);
1645 vdwjidx0A = 2*vdwtype[inr+0];
1646 c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1647 c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1648 qq11 = _mm_mul_ps(iq1,jq1);
1649 qq12 = _mm_mul_ps(iq1,jq2);
1650 qq13 = _mm_mul_ps(iq1,jq3);
1651 qq21 = _mm_mul_ps(iq2,jq1);
1652 qq22 = _mm_mul_ps(iq2,jq2);
1653 qq23 = _mm_mul_ps(iq2,jq3);
1654 qq31 = _mm_mul_ps(iq3,jq1);
1655 qq32 = _mm_mul_ps(iq3,jq2);
1656 qq33 = _mm_mul_ps(iq3,jq3);
1658 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1659 rcutoff_scalar = fr->rcoulomb;
1660 rcutoff = _mm_set1_ps(rcutoff_scalar);
1661 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
1663 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
1664 rvdw = _mm_set1_ps(fr->rvdw);
1666 /* Avoid stupid compiler warnings */
1667 jnrA = jnrB = jnrC = jnrD = 0;
1668 j_coord_offsetA = 0;
1669 j_coord_offsetB = 0;
1670 j_coord_offsetC = 0;
1671 j_coord_offsetD = 0;
1676 for(iidx=0;iidx<4*DIM;iidx++)
1678 scratch[iidx] = 0.0;
1681 /* Start outer loop over neighborlists */
1682 for(iidx=0; iidx<nri; iidx++)
1684 /* Load shift vector for this list */
1685 i_shift_offset = DIM*shiftidx[iidx];
1687 /* Load limits for loop over neighbors */
1688 j_index_start = jindex[iidx];
1689 j_index_end = jindex[iidx+1];
1691 /* Get outer coordinate index */
1693 i_coord_offset = DIM*inr;
1695 /* Load i particle coords and add shift vector */
1696 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1697 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1699 fix0 = _mm_setzero_ps();
1700 fiy0 = _mm_setzero_ps();
1701 fiz0 = _mm_setzero_ps();
1702 fix1 = _mm_setzero_ps();
1703 fiy1 = _mm_setzero_ps();
1704 fiz1 = _mm_setzero_ps();
1705 fix2 = _mm_setzero_ps();
1706 fiy2 = _mm_setzero_ps();
1707 fiz2 = _mm_setzero_ps();
1708 fix3 = _mm_setzero_ps();
1709 fiy3 = _mm_setzero_ps();
1710 fiz3 = _mm_setzero_ps();
1712 /* Start inner kernel loop */
1713 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1716 /* Get j neighbor index, and coordinate index */
1718 jnrB = jjnr[jidx+1];
1719 jnrC = jjnr[jidx+2];
1720 jnrD = jjnr[jidx+3];
1721 j_coord_offsetA = DIM*jnrA;
1722 j_coord_offsetB = DIM*jnrB;
1723 j_coord_offsetC = DIM*jnrC;
1724 j_coord_offsetD = DIM*jnrD;
1726 /* load j atom coordinates */
1727 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1728 x+j_coord_offsetC,x+j_coord_offsetD,
1729 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1730 &jy2,&jz2,&jx3,&jy3,&jz3);
1732 /* Calculate displacement vector */
1733 dx00 = _mm_sub_ps(ix0,jx0);
1734 dy00 = _mm_sub_ps(iy0,jy0);
1735 dz00 = _mm_sub_ps(iz0,jz0);
1736 dx11 = _mm_sub_ps(ix1,jx1);
1737 dy11 = _mm_sub_ps(iy1,jy1);
1738 dz11 = _mm_sub_ps(iz1,jz1);
1739 dx12 = _mm_sub_ps(ix1,jx2);
1740 dy12 = _mm_sub_ps(iy1,jy2);
1741 dz12 = _mm_sub_ps(iz1,jz2);
1742 dx13 = _mm_sub_ps(ix1,jx3);
1743 dy13 = _mm_sub_ps(iy1,jy3);
1744 dz13 = _mm_sub_ps(iz1,jz3);
1745 dx21 = _mm_sub_ps(ix2,jx1);
1746 dy21 = _mm_sub_ps(iy2,jy1);
1747 dz21 = _mm_sub_ps(iz2,jz1);
1748 dx22 = _mm_sub_ps(ix2,jx2);
1749 dy22 = _mm_sub_ps(iy2,jy2);
1750 dz22 = _mm_sub_ps(iz2,jz2);
1751 dx23 = _mm_sub_ps(ix2,jx3);
1752 dy23 = _mm_sub_ps(iy2,jy3);
1753 dz23 = _mm_sub_ps(iz2,jz3);
1754 dx31 = _mm_sub_ps(ix3,jx1);
1755 dy31 = _mm_sub_ps(iy3,jy1);
1756 dz31 = _mm_sub_ps(iz3,jz1);
1757 dx32 = _mm_sub_ps(ix3,jx2);
1758 dy32 = _mm_sub_ps(iy3,jy2);
1759 dz32 = _mm_sub_ps(iz3,jz2);
1760 dx33 = _mm_sub_ps(ix3,jx3);
1761 dy33 = _mm_sub_ps(iy3,jy3);
1762 dz33 = _mm_sub_ps(iz3,jz3);
1764 /* Calculate squared distance and things based on it */
1765 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1766 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1767 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1768 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1769 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1770 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1771 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1772 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1773 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1774 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1776 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1777 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1778 rinv13 = gmx_mm_invsqrt_ps(rsq13);
1779 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1780 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1781 rinv23 = gmx_mm_invsqrt_ps(rsq23);
1782 rinv31 = gmx_mm_invsqrt_ps(rsq31);
1783 rinv32 = gmx_mm_invsqrt_ps(rsq32);
1784 rinv33 = gmx_mm_invsqrt_ps(rsq33);
1786 rinvsq00 = gmx_mm_inv_ps(rsq00);
1787 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1788 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1789 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
1790 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1791 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1792 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
1793 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
1794 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
1795 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
1797 fjx0 = _mm_setzero_ps();
1798 fjy0 = _mm_setzero_ps();
1799 fjz0 = _mm_setzero_ps();
1800 fjx1 = _mm_setzero_ps();
1801 fjy1 = _mm_setzero_ps();
1802 fjz1 = _mm_setzero_ps();
1803 fjx2 = _mm_setzero_ps();
1804 fjy2 = _mm_setzero_ps();
1805 fjz2 = _mm_setzero_ps();
1806 fjx3 = _mm_setzero_ps();
1807 fjy3 = _mm_setzero_ps();
1808 fjz3 = _mm_setzero_ps();
1810 /**************************
1811 * CALCULATE INTERACTIONS *
1812 **************************/
1814 if (gmx_mm_any_lt(rsq00,rcutoff2))
1817 /* LENNARD-JONES DISPERSION/REPULSION */
1819 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1820 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1822 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1826 fscal = _mm_and_ps(fscal,cutoff_mask);
1828 /* Calculate temporary vectorial force */
1829 tx = _mm_mul_ps(fscal,dx00);
1830 ty = _mm_mul_ps(fscal,dy00);
1831 tz = _mm_mul_ps(fscal,dz00);
1833 /* Update vectorial force */
1834 fix0 = _mm_add_ps(fix0,tx);
1835 fiy0 = _mm_add_ps(fiy0,ty);
1836 fiz0 = _mm_add_ps(fiz0,tz);
1838 fjx0 = _mm_add_ps(fjx0,tx);
1839 fjy0 = _mm_add_ps(fjy0,ty);
1840 fjz0 = _mm_add_ps(fjz0,tz);
1844 /**************************
1845 * CALCULATE INTERACTIONS *
1846 **************************/
1848 if (gmx_mm_any_lt(rsq11,rcutoff2))
1851 r11 = _mm_mul_ps(rsq11,rinv11);
1853 /* EWALD ELECTROSTATICS */
1855 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1856 ewrt = _mm_mul_ps(r11,ewtabscale);
1857 ewitab = _mm_cvttps_epi32(ewrt);
1858 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1859 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1860 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1862 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1863 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1865 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
1869 fscal = _mm_and_ps(fscal,cutoff_mask);
1871 /* Calculate temporary vectorial force */
1872 tx = _mm_mul_ps(fscal,dx11);
1873 ty = _mm_mul_ps(fscal,dy11);
1874 tz = _mm_mul_ps(fscal,dz11);
1876 /* Update vectorial force */
1877 fix1 = _mm_add_ps(fix1,tx);
1878 fiy1 = _mm_add_ps(fiy1,ty);
1879 fiz1 = _mm_add_ps(fiz1,tz);
1881 fjx1 = _mm_add_ps(fjx1,tx);
1882 fjy1 = _mm_add_ps(fjy1,ty);
1883 fjz1 = _mm_add_ps(fjz1,tz);
1887 /**************************
1888 * CALCULATE INTERACTIONS *
1889 **************************/
1891 if (gmx_mm_any_lt(rsq12,rcutoff2))
1894 r12 = _mm_mul_ps(rsq12,rinv12);
1896 /* EWALD ELECTROSTATICS */
1898 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1899 ewrt = _mm_mul_ps(r12,ewtabscale);
1900 ewitab = _mm_cvttps_epi32(ewrt);
1901 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1902 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1903 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1905 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1906 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1908 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
1912 fscal = _mm_and_ps(fscal,cutoff_mask);
1914 /* Calculate temporary vectorial force */
1915 tx = _mm_mul_ps(fscal,dx12);
1916 ty = _mm_mul_ps(fscal,dy12);
1917 tz = _mm_mul_ps(fscal,dz12);
1919 /* Update vectorial force */
1920 fix1 = _mm_add_ps(fix1,tx);
1921 fiy1 = _mm_add_ps(fiy1,ty);
1922 fiz1 = _mm_add_ps(fiz1,tz);
1924 fjx2 = _mm_add_ps(fjx2,tx);
1925 fjy2 = _mm_add_ps(fjy2,ty);
1926 fjz2 = _mm_add_ps(fjz2,tz);
1930 /**************************
1931 * CALCULATE INTERACTIONS *
1932 **************************/
1934 if (gmx_mm_any_lt(rsq13,rcutoff2))
1937 r13 = _mm_mul_ps(rsq13,rinv13);
1939 /* EWALD ELECTROSTATICS */
1941 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1942 ewrt = _mm_mul_ps(r13,ewtabscale);
1943 ewitab = _mm_cvttps_epi32(ewrt);
1944 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1945 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1946 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1948 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1949 felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
1951 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
1955 fscal = _mm_and_ps(fscal,cutoff_mask);
1957 /* Calculate temporary vectorial force */
1958 tx = _mm_mul_ps(fscal,dx13);
1959 ty = _mm_mul_ps(fscal,dy13);
1960 tz = _mm_mul_ps(fscal,dz13);
1962 /* Update vectorial force */
1963 fix1 = _mm_add_ps(fix1,tx);
1964 fiy1 = _mm_add_ps(fiy1,ty);
1965 fiz1 = _mm_add_ps(fiz1,tz);
1967 fjx3 = _mm_add_ps(fjx3,tx);
1968 fjy3 = _mm_add_ps(fjy3,ty);
1969 fjz3 = _mm_add_ps(fjz3,tz);
1973 /**************************
1974 * CALCULATE INTERACTIONS *
1975 **************************/
1977 if (gmx_mm_any_lt(rsq21,rcutoff2))
1980 r21 = _mm_mul_ps(rsq21,rinv21);
1982 /* EWALD ELECTROSTATICS */
1984 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1985 ewrt = _mm_mul_ps(r21,ewtabscale);
1986 ewitab = _mm_cvttps_epi32(ewrt);
1987 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1988 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1989 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1991 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1992 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1994 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
1998 fscal = _mm_and_ps(fscal,cutoff_mask);
2000 /* Calculate temporary vectorial force */
2001 tx = _mm_mul_ps(fscal,dx21);
2002 ty = _mm_mul_ps(fscal,dy21);
2003 tz = _mm_mul_ps(fscal,dz21);
2005 /* Update vectorial force */
2006 fix2 = _mm_add_ps(fix2,tx);
2007 fiy2 = _mm_add_ps(fiy2,ty);
2008 fiz2 = _mm_add_ps(fiz2,tz);
2010 fjx1 = _mm_add_ps(fjx1,tx);
2011 fjy1 = _mm_add_ps(fjy1,ty);
2012 fjz1 = _mm_add_ps(fjz1,tz);
2016 /**************************
2017 * CALCULATE INTERACTIONS *
2018 **************************/
2020 if (gmx_mm_any_lt(rsq22,rcutoff2))
2023 r22 = _mm_mul_ps(rsq22,rinv22);
2025 /* EWALD ELECTROSTATICS */
2027 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2028 ewrt = _mm_mul_ps(r22,ewtabscale);
2029 ewitab = _mm_cvttps_epi32(ewrt);
2030 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2031 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2032 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2034 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2035 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2037 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
2041 fscal = _mm_and_ps(fscal,cutoff_mask);
2043 /* Calculate temporary vectorial force */
2044 tx = _mm_mul_ps(fscal,dx22);
2045 ty = _mm_mul_ps(fscal,dy22);
2046 tz = _mm_mul_ps(fscal,dz22);
2048 /* Update vectorial force */
2049 fix2 = _mm_add_ps(fix2,tx);
2050 fiy2 = _mm_add_ps(fiy2,ty);
2051 fiz2 = _mm_add_ps(fiz2,tz);
2053 fjx2 = _mm_add_ps(fjx2,tx);
2054 fjy2 = _mm_add_ps(fjy2,ty);
2055 fjz2 = _mm_add_ps(fjz2,tz);
2059 /**************************
2060 * CALCULATE INTERACTIONS *
2061 **************************/
2063 if (gmx_mm_any_lt(rsq23,rcutoff2))
2066 r23 = _mm_mul_ps(rsq23,rinv23);
2068 /* EWALD ELECTROSTATICS */
2070 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2071 ewrt = _mm_mul_ps(r23,ewtabscale);
2072 ewitab = _mm_cvttps_epi32(ewrt);
2073 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2074 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2075 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2077 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2078 felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
2080 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
2084 fscal = _mm_and_ps(fscal,cutoff_mask);
2086 /* Calculate temporary vectorial force */
2087 tx = _mm_mul_ps(fscal,dx23);
2088 ty = _mm_mul_ps(fscal,dy23);
2089 tz = _mm_mul_ps(fscal,dz23);
2091 /* Update vectorial force */
2092 fix2 = _mm_add_ps(fix2,tx);
2093 fiy2 = _mm_add_ps(fiy2,ty);
2094 fiz2 = _mm_add_ps(fiz2,tz);
2096 fjx3 = _mm_add_ps(fjx3,tx);
2097 fjy3 = _mm_add_ps(fjy3,ty);
2098 fjz3 = _mm_add_ps(fjz3,tz);
2102 /**************************
2103 * CALCULATE INTERACTIONS *
2104 **************************/
2106 if (gmx_mm_any_lt(rsq31,rcutoff2))
2109 r31 = _mm_mul_ps(rsq31,rinv31);
2111 /* EWALD ELECTROSTATICS */
2113 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2114 ewrt = _mm_mul_ps(r31,ewtabscale);
2115 ewitab = _mm_cvttps_epi32(ewrt);
2116 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2117 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2118 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2120 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2121 felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
2123 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
2127 fscal = _mm_and_ps(fscal,cutoff_mask);
2129 /* Calculate temporary vectorial force */
2130 tx = _mm_mul_ps(fscal,dx31);
2131 ty = _mm_mul_ps(fscal,dy31);
2132 tz = _mm_mul_ps(fscal,dz31);
2134 /* Update vectorial force */
2135 fix3 = _mm_add_ps(fix3,tx);
2136 fiy3 = _mm_add_ps(fiy3,ty);
2137 fiz3 = _mm_add_ps(fiz3,tz);
2139 fjx1 = _mm_add_ps(fjx1,tx);
2140 fjy1 = _mm_add_ps(fjy1,ty);
2141 fjz1 = _mm_add_ps(fjz1,tz);
2145 /**************************
2146 * CALCULATE INTERACTIONS *
2147 **************************/
2149 if (gmx_mm_any_lt(rsq32,rcutoff2))
2152 r32 = _mm_mul_ps(rsq32,rinv32);
2154 /* EWALD ELECTROSTATICS */
2156 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2157 ewrt = _mm_mul_ps(r32,ewtabscale);
2158 ewitab = _mm_cvttps_epi32(ewrt);
2159 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2160 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2161 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2163 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2164 felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
2166 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
2170 fscal = _mm_and_ps(fscal,cutoff_mask);
2172 /* Calculate temporary vectorial force */
2173 tx = _mm_mul_ps(fscal,dx32);
2174 ty = _mm_mul_ps(fscal,dy32);
2175 tz = _mm_mul_ps(fscal,dz32);
2177 /* Update vectorial force */
2178 fix3 = _mm_add_ps(fix3,tx);
2179 fiy3 = _mm_add_ps(fiy3,ty);
2180 fiz3 = _mm_add_ps(fiz3,tz);
2182 fjx2 = _mm_add_ps(fjx2,tx);
2183 fjy2 = _mm_add_ps(fjy2,ty);
2184 fjz2 = _mm_add_ps(fjz2,tz);
2188 /**************************
2189 * CALCULATE INTERACTIONS *
2190 **************************/
2192 if (gmx_mm_any_lt(rsq33,rcutoff2))
2195 r33 = _mm_mul_ps(rsq33,rinv33);
2197 /* EWALD ELECTROSTATICS */
2199 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2200 ewrt = _mm_mul_ps(r33,ewtabscale);
2201 ewitab = _mm_cvttps_epi32(ewrt);
2202 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2203 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2204 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2206 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2207 felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
2209 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
2213 fscal = _mm_and_ps(fscal,cutoff_mask);
2215 /* Calculate temporary vectorial force */
2216 tx = _mm_mul_ps(fscal,dx33);
2217 ty = _mm_mul_ps(fscal,dy33);
2218 tz = _mm_mul_ps(fscal,dz33);
2220 /* Update vectorial force */
2221 fix3 = _mm_add_ps(fix3,tx);
2222 fiy3 = _mm_add_ps(fiy3,ty);
2223 fiz3 = _mm_add_ps(fiz3,tz);
2225 fjx3 = _mm_add_ps(fjx3,tx);
2226 fjy3 = _mm_add_ps(fjy3,ty);
2227 fjz3 = _mm_add_ps(fjz3,tz);
2231 fjptrA = f+j_coord_offsetA;
2232 fjptrB = f+j_coord_offsetB;
2233 fjptrC = f+j_coord_offsetC;
2234 fjptrD = f+j_coord_offsetD;
2236 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2237 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2238 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2240 /* Inner loop uses 384 flops */
2243 if(jidx<j_index_end)
2246 /* Get j neighbor index, and coordinate index */
2247 jnrlistA = jjnr[jidx];
2248 jnrlistB = jjnr[jidx+1];
2249 jnrlistC = jjnr[jidx+2];
2250 jnrlistD = jjnr[jidx+3];
2251 /* Sign of each element will be negative for non-real atoms.
2252 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2253 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2255 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2256 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
2257 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
2258 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
2259 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
2260 j_coord_offsetA = DIM*jnrA;
2261 j_coord_offsetB = DIM*jnrB;
2262 j_coord_offsetC = DIM*jnrC;
2263 j_coord_offsetD = DIM*jnrD;
2265 /* load j atom coordinates */
2266 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2267 x+j_coord_offsetC,x+j_coord_offsetD,
2268 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2269 &jy2,&jz2,&jx3,&jy3,&jz3);
2271 /* Calculate displacement vector */
2272 dx00 = _mm_sub_ps(ix0,jx0);
2273 dy00 = _mm_sub_ps(iy0,jy0);
2274 dz00 = _mm_sub_ps(iz0,jz0);
2275 dx11 = _mm_sub_ps(ix1,jx1);
2276 dy11 = _mm_sub_ps(iy1,jy1);
2277 dz11 = _mm_sub_ps(iz1,jz1);
2278 dx12 = _mm_sub_ps(ix1,jx2);
2279 dy12 = _mm_sub_ps(iy1,jy2);
2280 dz12 = _mm_sub_ps(iz1,jz2);
2281 dx13 = _mm_sub_ps(ix1,jx3);
2282 dy13 = _mm_sub_ps(iy1,jy3);
2283 dz13 = _mm_sub_ps(iz1,jz3);
2284 dx21 = _mm_sub_ps(ix2,jx1);
2285 dy21 = _mm_sub_ps(iy2,jy1);
2286 dz21 = _mm_sub_ps(iz2,jz1);
2287 dx22 = _mm_sub_ps(ix2,jx2);
2288 dy22 = _mm_sub_ps(iy2,jy2);
2289 dz22 = _mm_sub_ps(iz2,jz2);
2290 dx23 = _mm_sub_ps(ix2,jx3);
2291 dy23 = _mm_sub_ps(iy2,jy3);
2292 dz23 = _mm_sub_ps(iz2,jz3);
2293 dx31 = _mm_sub_ps(ix3,jx1);
2294 dy31 = _mm_sub_ps(iy3,jy1);
2295 dz31 = _mm_sub_ps(iz3,jz1);
2296 dx32 = _mm_sub_ps(ix3,jx2);
2297 dy32 = _mm_sub_ps(iy3,jy2);
2298 dz32 = _mm_sub_ps(iz3,jz2);
2299 dx33 = _mm_sub_ps(ix3,jx3);
2300 dy33 = _mm_sub_ps(iy3,jy3);
2301 dz33 = _mm_sub_ps(iz3,jz3);
2303 /* Calculate squared distance and things based on it */
2304 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
2305 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2306 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2307 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
2308 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2309 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2310 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
2311 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
2312 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
2313 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2315 rinv11 = gmx_mm_invsqrt_ps(rsq11);
2316 rinv12 = gmx_mm_invsqrt_ps(rsq12);
2317 rinv13 = gmx_mm_invsqrt_ps(rsq13);
2318 rinv21 = gmx_mm_invsqrt_ps(rsq21);
2319 rinv22 = gmx_mm_invsqrt_ps(rsq22);
2320 rinv23 = gmx_mm_invsqrt_ps(rsq23);
2321 rinv31 = gmx_mm_invsqrt_ps(rsq31);
2322 rinv32 = gmx_mm_invsqrt_ps(rsq32);
2323 rinv33 = gmx_mm_invsqrt_ps(rsq33);
2325 rinvsq00 = gmx_mm_inv_ps(rsq00);
2326 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
2327 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
2328 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
2329 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
2330 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
2331 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
2332 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
2333 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
2334 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
2336 fjx0 = _mm_setzero_ps();
2337 fjy0 = _mm_setzero_ps();
2338 fjz0 = _mm_setzero_ps();
2339 fjx1 = _mm_setzero_ps();
2340 fjy1 = _mm_setzero_ps();
2341 fjz1 = _mm_setzero_ps();
2342 fjx2 = _mm_setzero_ps();
2343 fjy2 = _mm_setzero_ps();
2344 fjz2 = _mm_setzero_ps();
2345 fjx3 = _mm_setzero_ps();
2346 fjy3 = _mm_setzero_ps();
2347 fjz3 = _mm_setzero_ps();
2349 /**************************
2350 * CALCULATE INTERACTIONS *
2351 **************************/
2353 if (gmx_mm_any_lt(rsq00,rcutoff2))
2356 /* LENNARD-JONES DISPERSION/REPULSION */
2358 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2359 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
2361 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
2365 fscal = _mm_and_ps(fscal,cutoff_mask);
2367 fscal = _mm_andnot_ps(dummy_mask,fscal);
2369 /* Calculate temporary vectorial force */
2370 tx = _mm_mul_ps(fscal,dx00);
2371 ty = _mm_mul_ps(fscal,dy00);
2372 tz = _mm_mul_ps(fscal,dz00);
2374 /* Update vectorial force */
2375 fix0 = _mm_add_ps(fix0,tx);
2376 fiy0 = _mm_add_ps(fiy0,ty);
2377 fiz0 = _mm_add_ps(fiz0,tz);
2379 fjx0 = _mm_add_ps(fjx0,tx);
2380 fjy0 = _mm_add_ps(fjy0,ty);
2381 fjz0 = _mm_add_ps(fjz0,tz);
2385 /**************************
2386 * CALCULATE INTERACTIONS *
2387 **************************/
2389 if (gmx_mm_any_lt(rsq11,rcutoff2))
2392 r11 = _mm_mul_ps(rsq11,rinv11);
2393 r11 = _mm_andnot_ps(dummy_mask,r11);
2395 /* EWALD ELECTROSTATICS */
2397 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2398 ewrt = _mm_mul_ps(r11,ewtabscale);
2399 ewitab = _mm_cvttps_epi32(ewrt);
2400 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2401 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2402 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2404 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2405 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
2407 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
2411 fscal = _mm_and_ps(fscal,cutoff_mask);
2413 fscal = _mm_andnot_ps(dummy_mask,fscal);
2415 /* Calculate temporary vectorial force */
2416 tx = _mm_mul_ps(fscal,dx11);
2417 ty = _mm_mul_ps(fscal,dy11);
2418 tz = _mm_mul_ps(fscal,dz11);
2420 /* Update vectorial force */
2421 fix1 = _mm_add_ps(fix1,tx);
2422 fiy1 = _mm_add_ps(fiy1,ty);
2423 fiz1 = _mm_add_ps(fiz1,tz);
2425 fjx1 = _mm_add_ps(fjx1,tx);
2426 fjy1 = _mm_add_ps(fjy1,ty);
2427 fjz1 = _mm_add_ps(fjz1,tz);
2431 /**************************
2432 * CALCULATE INTERACTIONS *
2433 **************************/
2435 if (gmx_mm_any_lt(rsq12,rcutoff2))
2438 r12 = _mm_mul_ps(rsq12,rinv12);
2439 r12 = _mm_andnot_ps(dummy_mask,r12);
2441 /* EWALD ELECTROSTATICS */
2443 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2444 ewrt = _mm_mul_ps(r12,ewtabscale);
2445 ewitab = _mm_cvttps_epi32(ewrt);
2446 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2447 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2448 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2450 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2451 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
2453 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
2457 fscal = _mm_and_ps(fscal,cutoff_mask);
2459 fscal = _mm_andnot_ps(dummy_mask,fscal);
2461 /* Calculate temporary vectorial force */
2462 tx = _mm_mul_ps(fscal,dx12);
2463 ty = _mm_mul_ps(fscal,dy12);
2464 tz = _mm_mul_ps(fscal,dz12);
2466 /* Update vectorial force */
2467 fix1 = _mm_add_ps(fix1,tx);
2468 fiy1 = _mm_add_ps(fiy1,ty);
2469 fiz1 = _mm_add_ps(fiz1,tz);
2471 fjx2 = _mm_add_ps(fjx2,tx);
2472 fjy2 = _mm_add_ps(fjy2,ty);
2473 fjz2 = _mm_add_ps(fjz2,tz);
2477 /**************************
2478 * CALCULATE INTERACTIONS *
2479 **************************/
2481 if (gmx_mm_any_lt(rsq13,rcutoff2))
2484 r13 = _mm_mul_ps(rsq13,rinv13);
2485 r13 = _mm_andnot_ps(dummy_mask,r13);
2487 /* EWALD ELECTROSTATICS */
2489 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2490 ewrt = _mm_mul_ps(r13,ewtabscale);
2491 ewitab = _mm_cvttps_epi32(ewrt);
2492 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2493 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2494 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2496 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2497 felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
2499 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
2503 fscal = _mm_and_ps(fscal,cutoff_mask);
2505 fscal = _mm_andnot_ps(dummy_mask,fscal);
2507 /* Calculate temporary vectorial force */
2508 tx = _mm_mul_ps(fscal,dx13);
2509 ty = _mm_mul_ps(fscal,dy13);
2510 tz = _mm_mul_ps(fscal,dz13);
2512 /* Update vectorial force */
2513 fix1 = _mm_add_ps(fix1,tx);
2514 fiy1 = _mm_add_ps(fiy1,ty);
2515 fiz1 = _mm_add_ps(fiz1,tz);
2517 fjx3 = _mm_add_ps(fjx3,tx);
2518 fjy3 = _mm_add_ps(fjy3,ty);
2519 fjz3 = _mm_add_ps(fjz3,tz);
2523 /**************************
2524 * CALCULATE INTERACTIONS *
2525 **************************/
2527 if (gmx_mm_any_lt(rsq21,rcutoff2))
2530 r21 = _mm_mul_ps(rsq21,rinv21);
2531 r21 = _mm_andnot_ps(dummy_mask,r21);
2533 /* EWALD ELECTROSTATICS */
2535 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2536 ewrt = _mm_mul_ps(r21,ewtabscale);
2537 ewitab = _mm_cvttps_epi32(ewrt);
2538 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2539 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2540 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2542 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2543 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2545 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
2549 fscal = _mm_and_ps(fscal,cutoff_mask);
2551 fscal = _mm_andnot_ps(dummy_mask,fscal);
2553 /* Calculate temporary vectorial force */
2554 tx = _mm_mul_ps(fscal,dx21);
2555 ty = _mm_mul_ps(fscal,dy21);
2556 tz = _mm_mul_ps(fscal,dz21);
2558 /* Update vectorial force */
2559 fix2 = _mm_add_ps(fix2,tx);
2560 fiy2 = _mm_add_ps(fiy2,ty);
2561 fiz2 = _mm_add_ps(fiz2,tz);
2563 fjx1 = _mm_add_ps(fjx1,tx);
2564 fjy1 = _mm_add_ps(fjy1,ty);
2565 fjz1 = _mm_add_ps(fjz1,tz);
2569 /**************************
2570 * CALCULATE INTERACTIONS *
2571 **************************/
2573 if (gmx_mm_any_lt(rsq22,rcutoff2))
2576 r22 = _mm_mul_ps(rsq22,rinv22);
2577 r22 = _mm_andnot_ps(dummy_mask,r22);
2579 /* EWALD ELECTROSTATICS */
2581 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2582 ewrt = _mm_mul_ps(r22,ewtabscale);
2583 ewitab = _mm_cvttps_epi32(ewrt);
2584 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2585 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2586 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2588 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2589 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2591 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
2595 fscal = _mm_and_ps(fscal,cutoff_mask);
2597 fscal = _mm_andnot_ps(dummy_mask,fscal);
2599 /* Calculate temporary vectorial force */
2600 tx = _mm_mul_ps(fscal,dx22);
2601 ty = _mm_mul_ps(fscal,dy22);
2602 tz = _mm_mul_ps(fscal,dz22);
2604 /* Update vectorial force */
2605 fix2 = _mm_add_ps(fix2,tx);
2606 fiy2 = _mm_add_ps(fiy2,ty);
2607 fiz2 = _mm_add_ps(fiz2,tz);
2609 fjx2 = _mm_add_ps(fjx2,tx);
2610 fjy2 = _mm_add_ps(fjy2,ty);
2611 fjz2 = _mm_add_ps(fjz2,tz);
2615 /**************************
2616 * CALCULATE INTERACTIONS *
2617 **************************/
2619 if (gmx_mm_any_lt(rsq23,rcutoff2))
2622 r23 = _mm_mul_ps(rsq23,rinv23);
2623 r23 = _mm_andnot_ps(dummy_mask,r23);
2625 /* EWALD ELECTROSTATICS */
2627 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2628 ewrt = _mm_mul_ps(r23,ewtabscale);
2629 ewitab = _mm_cvttps_epi32(ewrt);
2630 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2631 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2632 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2634 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2635 felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
2637 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
2641 fscal = _mm_and_ps(fscal,cutoff_mask);
2643 fscal = _mm_andnot_ps(dummy_mask,fscal);
2645 /* Calculate temporary vectorial force */
2646 tx = _mm_mul_ps(fscal,dx23);
2647 ty = _mm_mul_ps(fscal,dy23);
2648 tz = _mm_mul_ps(fscal,dz23);
2650 /* Update vectorial force */
2651 fix2 = _mm_add_ps(fix2,tx);
2652 fiy2 = _mm_add_ps(fiy2,ty);
2653 fiz2 = _mm_add_ps(fiz2,tz);
2655 fjx3 = _mm_add_ps(fjx3,tx);
2656 fjy3 = _mm_add_ps(fjy3,ty);
2657 fjz3 = _mm_add_ps(fjz3,tz);
2661 /**************************
2662 * CALCULATE INTERACTIONS *
2663 **************************/
2665 if (gmx_mm_any_lt(rsq31,rcutoff2))
2668 r31 = _mm_mul_ps(rsq31,rinv31);
2669 r31 = _mm_andnot_ps(dummy_mask,r31);
2671 /* EWALD ELECTROSTATICS */
2673 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2674 ewrt = _mm_mul_ps(r31,ewtabscale);
2675 ewitab = _mm_cvttps_epi32(ewrt);
2676 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2677 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2678 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2680 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2681 felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
2683 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
2687 fscal = _mm_and_ps(fscal,cutoff_mask);
2689 fscal = _mm_andnot_ps(dummy_mask,fscal);
2691 /* Calculate temporary vectorial force */
2692 tx = _mm_mul_ps(fscal,dx31);
2693 ty = _mm_mul_ps(fscal,dy31);
2694 tz = _mm_mul_ps(fscal,dz31);
2696 /* Update vectorial force */
2697 fix3 = _mm_add_ps(fix3,tx);
2698 fiy3 = _mm_add_ps(fiy3,ty);
2699 fiz3 = _mm_add_ps(fiz3,tz);
2701 fjx1 = _mm_add_ps(fjx1,tx);
2702 fjy1 = _mm_add_ps(fjy1,ty);
2703 fjz1 = _mm_add_ps(fjz1,tz);
2707 /**************************
2708 * CALCULATE INTERACTIONS *
2709 **************************/
2711 if (gmx_mm_any_lt(rsq32,rcutoff2))
2714 r32 = _mm_mul_ps(rsq32,rinv32);
2715 r32 = _mm_andnot_ps(dummy_mask,r32);
2717 /* EWALD ELECTROSTATICS */
2719 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2720 ewrt = _mm_mul_ps(r32,ewtabscale);
2721 ewitab = _mm_cvttps_epi32(ewrt);
2722 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2723 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2724 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2726 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2727 felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
2729 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
2733 fscal = _mm_and_ps(fscal,cutoff_mask);
2735 fscal = _mm_andnot_ps(dummy_mask,fscal);
2737 /* Calculate temporary vectorial force */
2738 tx = _mm_mul_ps(fscal,dx32);
2739 ty = _mm_mul_ps(fscal,dy32);
2740 tz = _mm_mul_ps(fscal,dz32);
2742 /* Update vectorial force */
2743 fix3 = _mm_add_ps(fix3,tx);
2744 fiy3 = _mm_add_ps(fiy3,ty);
2745 fiz3 = _mm_add_ps(fiz3,tz);
2747 fjx2 = _mm_add_ps(fjx2,tx);
2748 fjy2 = _mm_add_ps(fjy2,ty);
2749 fjz2 = _mm_add_ps(fjz2,tz);
2753 /**************************
2754 * CALCULATE INTERACTIONS *
2755 **************************/
2757 if (gmx_mm_any_lt(rsq33,rcutoff2))
2760 r33 = _mm_mul_ps(rsq33,rinv33);
2761 r33 = _mm_andnot_ps(dummy_mask,r33);
2763 /* EWALD ELECTROSTATICS */
2765 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2766 ewrt = _mm_mul_ps(r33,ewtabscale);
2767 ewitab = _mm_cvttps_epi32(ewrt);
2768 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2769 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2770 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2772 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2773 felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
2775 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
2779 fscal = _mm_and_ps(fscal,cutoff_mask);
2781 fscal = _mm_andnot_ps(dummy_mask,fscal);
2783 /* Calculate temporary vectorial force */
2784 tx = _mm_mul_ps(fscal,dx33);
2785 ty = _mm_mul_ps(fscal,dy33);
2786 tz = _mm_mul_ps(fscal,dz33);
2788 /* Update vectorial force */
2789 fix3 = _mm_add_ps(fix3,tx);
2790 fiy3 = _mm_add_ps(fiy3,ty);
2791 fiz3 = _mm_add_ps(fiz3,tz);
2793 fjx3 = _mm_add_ps(fjx3,tx);
2794 fjy3 = _mm_add_ps(fjy3,ty);
2795 fjz3 = _mm_add_ps(fjz3,tz);
2799 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2800 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2801 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2802 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2804 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2805 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2806 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2808 /* Inner loop uses 393 flops */
2811 /* End of innermost loop */
2813 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2814 f+i_coord_offset,fshift+i_shift_offset);
2816 /* Increment number of inner iterations */
2817 inneriter += j_index_end - j_index_start;
2819 /* Outer loop uses 24 flops */
2822 /* Increment number of outer iterations */
2825 /* Update outer/inner flops */
2827 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*393);