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.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse4_1_single.h"
48 #include "kernelutil_x86_sse4_1_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_single
52 * Electrostatics interaction: Ewald
53 * VdW interaction: LennardJones
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
89 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
90 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
101 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
103 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
105 __m128 dummy_mask,cutoff_mask;
106 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
107 __m128 one = _mm_set1_ps(1.0);
108 __m128 two = _mm_set1_ps(2.0);
114 jindex = nlist->jindex;
116 shiftidx = nlist->shift;
118 shiftvec = fr->shift_vec[0];
119 fshift = fr->fshift[0];
120 facel = _mm_set1_ps(fr->epsfac);
121 charge = mdatoms->chargeA;
122 nvdwtype = fr->ntype;
124 vdwtype = mdatoms->typeA;
126 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
127 ewtab = fr->ic->tabq_coul_FDV0;
128 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
129 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
134 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
135 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
136 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
138 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
139 rcutoff_scalar = fr->rcoulomb;
140 rcutoff = _mm_set1_ps(rcutoff_scalar);
141 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
143 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
144 rvdw = _mm_set1_ps(fr->rvdw);
146 /* Avoid stupid compiler warnings */
147 jnrA = jnrB = jnrC = jnrD = 0;
156 for(iidx=0;iidx<4*DIM;iidx++)
161 /* Start outer loop over neighborlists */
162 for(iidx=0; iidx<nri; iidx++)
164 /* Load shift vector for this list */
165 i_shift_offset = DIM*shiftidx[iidx];
167 /* Load limits for loop over neighbors */
168 j_index_start = jindex[iidx];
169 j_index_end = jindex[iidx+1];
171 /* Get outer coordinate index */
173 i_coord_offset = DIM*inr;
175 /* Load i particle coords and add shift vector */
176 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
177 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
179 fix0 = _mm_setzero_ps();
180 fiy0 = _mm_setzero_ps();
181 fiz0 = _mm_setzero_ps();
182 fix1 = _mm_setzero_ps();
183 fiy1 = _mm_setzero_ps();
184 fiz1 = _mm_setzero_ps();
185 fix2 = _mm_setzero_ps();
186 fiy2 = _mm_setzero_ps();
187 fiz2 = _mm_setzero_ps();
189 /* Reset potential sums */
190 velecsum = _mm_setzero_ps();
191 vvdwsum = _mm_setzero_ps();
193 /* Start inner kernel loop */
194 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
197 /* Get j neighbor index, and coordinate index */
202 j_coord_offsetA = DIM*jnrA;
203 j_coord_offsetB = DIM*jnrB;
204 j_coord_offsetC = DIM*jnrC;
205 j_coord_offsetD = DIM*jnrD;
207 /* load j atom coordinates */
208 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
209 x+j_coord_offsetC,x+j_coord_offsetD,
212 /* Calculate displacement vector */
213 dx00 = _mm_sub_ps(ix0,jx0);
214 dy00 = _mm_sub_ps(iy0,jy0);
215 dz00 = _mm_sub_ps(iz0,jz0);
216 dx10 = _mm_sub_ps(ix1,jx0);
217 dy10 = _mm_sub_ps(iy1,jy0);
218 dz10 = _mm_sub_ps(iz1,jz0);
219 dx20 = _mm_sub_ps(ix2,jx0);
220 dy20 = _mm_sub_ps(iy2,jy0);
221 dz20 = _mm_sub_ps(iz2,jz0);
223 /* Calculate squared distance and things based on it */
224 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
225 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
226 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
228 rinv00 = gmx_mm_invsqrt_ps(rsq00);
229 rinv10 = gmx_mm_invsqrt_ps(rsq10);
230 rinv20 = gmx_mm_invsqrt_ps(rsq20);
232 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
233 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
234 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
236 /* Load parameters for j particles */
237 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
238 charge+jnrC+0,charge+jnrD+0);
239 vdwjidx0A = 2*vdwtype[jnrA+0];
240 vdwjidx0B = 2*vdwtype[jnrB+0];
241 vdwjidx0C = 2*vdwtype[jnrC+0];
242 vdwjidx0D = 2*vdwtype[jnrD+0];
244 fjx0 = _mm_setzero_ps();
245 fjy0 = _mm_setzero_ps();
246 fjz0 = _mm_setzero_ps();
248 /**************************
249 * CALCULATE INTERACTIONS *
250 **************************/
252 if (gmx_mm_any_lt(rsq00,rcutoff2))
255 r00 = _mm_mul_ps(rsq00,rinv00);
257 /* Compute parameters for interactions between i and j atoms */
258 qq00 = _mm_mul_ps(iq0,jq0);
259 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
260 vdwparam+vdwioffset0+vdwjidx0B,
261 vdwparam+vdwioffset0+vdwjidx0C,
262 vdwparam+vdwioffset0+vdwjidx0D,
265 /* EWALD ELECTROSTATICS */
267 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
268 ewrt = _mm_mul_ps(r00,ewtabscale);
269 ewitab = _mm_cvttps_epi32(ewrt);
270 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
271 ewitab = _mm_slli_epi32(ewitab,2);
272 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
273 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
274 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
275 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
276 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
277 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
278 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
279 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
280 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
282 /* LENNARD-JONES DISPERSION/REPULSION */
284 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
285 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
286 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
287 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) ,
288 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
289 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
291 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
293 /* Update potential sum for this i atom from the interaction with this j atom. */
294 velec = _mm_and_ps(velec,cutoff_mask);
295 velecsum = _mm_add_ps(velecsum,velec);
296 vvdw = _mm_and_ps(vvdw,cutoff_mask);
297 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
299 fscal = _mm_add_ps(felec,fvdw);
301 fscal = _mm_and_ps(fscal,cutoff_mask);
303 /* Calculate temporary vectorial force */
304 tx = _mm_mul_ps(fscal,dx00);
305 ty = _mm_mul_ps(fscal,dy00);
306 tz = _mm_mul_ps(fscal,dz00);
308 /* Update vectorial force */
309 fix0 = _mm_add_ps(fix0,tx);
310 fiy0 = _mm_add_ps(fiy0,ty);
311 fiz0 = _mm_add_ps(fiz0,tz);
313 fjx0 = _mm_add_ps(fjx0,tx);
314 fjy0 = _mm_add_ps(fjy0,ty);
315 fjz0 = _mm_add_ps(fjz0,tz);
319 /**************************
320 * CALCULATE INTERACTIONS *
321 **************************/
323 if (gmx_mm_any_lt(rsq10,rcutoff2))
326 r10 = _mm_mul_ps(rsq10,rinv10);
328 /* Compute parameters for interactions between i and j atoms */
329 qq10 = _mm_mul_ps(iq1,jq0);
331 /* EWALD ELECTROSTATICS */
333 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
334 ewrt = _mm_mul_ps(r10,ewtabscale);
335 ewitab = _mm_cvttps_epi32(ewrt);
336 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
337 ewitab = _mm_slli_epi32(ewitab,2);
338 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
339 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
340 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
341 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
342 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
343 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
344 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
345 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
346 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
348 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
350 /* Update potential sum for this i atom from the interaction with this j atom. */
351 velec = _mm_and_ps(velec,cutoff_mask);
352 velecsum = _mm_add_ps(velecsum,velec);
356 fscal = _mm_and_ps(fscal,cutoff_mask);
358 /* Calculate temporary vectorial force */
359 tx = _mm_mul_ps(fscal,dx10);
360 ty = _mm_mul_ps(fscal,dy10);
361 tz = _mm_mul_ps(fscal,dz10);
363 /* Update vectorial force */
364 fix1 = _mm_add_ps(fix1,tx);
365 fiy1 = _mm_add_ps(fiy1,ty);
366 fiz1 = _mm_add_ps(fiz1,tz);
368 fjx0 = _mm_add_ps(fjx0,tx);
369 fjy0 = _mm_add_ps(fjy0,ty);
370 fjz0 = _mm_add_ps(fjz0,tz);
374 /**************************
375 * CALCULATE INTERACTIONS *
376 **************************/
378 if (gmx_mm_any_lt(rsq20,rcutoff2))
381 r20 = _mm_mul_ps(rsq20,rinv20);
383 /* Compute parameters for interactions between i and j atoms */
384 qq20 = _mm_mul_ps(iq2,jq0);
386 /* EWALD ELECTROSTATICS */
388 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
389 ewrt = _mm_mul_ps(r20,ewtabscale);
390 ewitab = _mm_cvttps_epi32(ewrt);
391 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
392 ewitab = _mm_slli_epi32(ewitab,2);
393 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
394 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
395 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
396 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
397 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
398 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
399 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
400 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
401 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
403 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
405 /* Update potential sum for this i atom from the interaction with this j atom. */
406 velec = _mm_and_ps(velec,cutoff_mask);
407 velecsum = _mm_add_ps(velecsum,velec);
411 fscal = _mm_and_ps(fscal,cutoff_mask);
413 /* Calculate temporary vectorial force */
414 tx = _mm_mul_ps(fscal,dx20);
415 ty = _mm_mul_ps(fscal,dy20);
416 tz = _mm_mul_ps(fscal,dz20);
418 /* Update vectorial force */
419 fix2 = _mm_add_ps(fix2,tx);
420 fiy2 = _mm_add_ps(fiy2,ty);
421 fiz2 = _mm_add_ps(fiz2,tz);
423 fjx0 = _mm_add_ps(fjx0,tx);
424 fjy0 = _mm_add_ps(fjy0,ty);
425 fjz0 = _mm_add_ps(fjz0,tz);
429 fjptrA = f+j_coord_offsetA;
430 fjptrB = f+j_coord_offsetB;
431 fjptrC = f+j_coord_offsetC;
432 fjptrD = f+j_coord_offsetD;
434 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
436 /* Inner loop uses 156 flops */
442 /* Get j neighbor index, and coordinate index */
443 jnrlistA = jjnr[jidx];
444 jnrlistB = jjnr[jidx+1];
445 jnrlistC = jjnr[jidx+2];
446 jnrlistD = jjnr[jidx+3];
447 /* Sign of each element will be negative for non-real atoms.
448 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
449 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
451 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
452 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
453 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
454 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
455 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
456 j_coord_offsetA = DIM*jnrA;
457 j_coord_offsetB = DIM*jnrB;
458 j_coord_offsetC = DIM*jnrC;
459 j_coord_offsetD = DIM*jnrD;
461 /* load j atom coordinates */
462 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
463 x+j_coord_offsetC,x+j_coord_offsetD,
466 /* Calculate displacement vector */
467 dx00 = _mm_sub_ps(ix0,jx0);
468 dy00 = _mm_sub_ps(iy0,jy0);
469 dz00 = _mm_sub_ps(iz0,jz0);
470 dx10 = _mm_sub_ps(ix1,jx0);
471 dy10 = _mm_sub_ps(iy1,jy0);
472 dz10 = _mm_sub_ps(iz1,jz0);
473 dx20 = _mm_sub_ps(ix2,jx0);
474 dy20 = _mm_sub_ps(iy2,jy0);
475 dz20 = _mm_sub_ps(iz2,jz0);
477 /* Calculate squared distance and things based on it */
478 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
479 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
480 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
482 rinv00 = gmx_mm_invsqrt_ps(rsq00);
483 rinv10 = gmx_mm_invsqrt_ps(rsq10);
484 rinv20 = gmx_mm_invsqrt_ps(rsq20);
486 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
487 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
488 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
490 /* Load parameters for j particles */
491 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
492 charge+jnrC+0,charge+jnrD+0);
493 vdwjidx0A = 2*vdwtype[jnrA+0];
494 vdwjidx0B = 2*vdwtype[jnrB+0];
495 vdwjidx0C = 2*vdwtype[jnrC+0];
496 vdwjidx0D = 2*vdwtype[jnrD+0];
498 fjx0 = _mm_setzero_ps();
499 fjy0 = _mm_setzero_ps();
500 fjz0 = _mm_setzero_ps();
502 /**************************
503 * CALCULATE INTERACTIONS *
504 **************************/
506 if (gmx_mm_any_lt(rsq00,rcutoff2))
509 r00 = _mm_mul_ps(rsq00,rinv00);
510 r00 = _mm_andnot_ps(dummy_mask,r00);
512 /* Compute parameters for interactions between i and j atoms */
513 qq00 = _mm_mul_ps(iq0,jq0);
514 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
515 vdwparam+vdwioffset0+vdwjidx0B,
516 vdwparam+vdwioffset0+vdwjidx0C,
517 vdwparam+vdwioffset0+vdwjidx0D,
520 /* EWALD ELECTROSTATICS */
522 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
523 ewrt = _mm_mul_ps(r00,ewtabscale);
524 ewitab = _mm_cvttps_epi32(ewrt);
525 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
526 ewitab = _mm_slli_epi32(ewitab,2);
527 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
528 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
529 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
530 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
531 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
532 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
533 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
534 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
535 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
537 /* LENNARD-JONES DISPERSION/REPULSION */
539 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
540 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
541 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
542 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) ,
543 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
544 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
546 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
548 /* Update potential sum for this i atom from the interaction with this j atom. */
549 velec = _mm_and_ps(velec,cutoff_mask);
550 velec = _mm_andnot_ps(dummy_mask,velec);
551 velecsum = _mm_add_ps(velecsum,velec);
552 vvdw = _mm_and_ps(vvdw,cutoff_mask);
553 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
554 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
556 fscal = _mm_add_ps(felec,fvdw);
558 fscal = _mm_and_ps(fscal,cutoff_mask);
560 fscal = _mm_andnot_ps(dummy_mask,fscal);
562 /* Calculate temporary vectorial force */
563 tx = _mm_mul_ps(fscal,dx00);
564 ty = _mm_mul_ps(fscal,dy00);
565 tz = _mm_mul_ps(fscal,dz00);
567 /* Update vectorial force */
568 fix0 = _mm_add_ps(fix0,tx);
569 fiy0 = _mm_add_ps(fiy0,ty);
570 fiz0 = _mm_add_ps(fiz0,tz);
572 fjx0 = _mm_add_ps(fjx0,tx);
573 fjy0 = _mm_add_ps(fjy0,ty);
574 fjz0 = _mm_add_ps(fjz0,tz);
578 /**************************
579 * CALCULATE INTERACTIONS *
580 **************************/
582 if (gmx_mm_any_lt(rsq10,rcutoff2))
585 r10 = _mm_mul_ps(rsq10,rinv10);
586 r10 = _mm_andnot_ps(dummy_mask,r10);
588 /* Compute parameters for interactions between i and j atoms */
589 qq10 = _mm_mul_ps(iq1,jq0);
591 /* EWALD ELECTROSTATICS */
593 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
594 ewrt = _mm_mul_ps(r10,ewtabscale);
595 ewitab = _mm_cvttps_epi32(ewrt);
596 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
597 ewitab = _mm_slli_epi32(ewitab,2);
598 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
599 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
600 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
601 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
602 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
603 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
604 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
605 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
606 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
608 cutoff_mask = _mm_cmplt_ps(rsq10,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 velec = _mm_andnot_ps(dummy_mask,velec);
613 velecsum = _mm_add_ps(velecsum,velec);
617 fscal = _mm_and_ps(fscal,cutoff_mask);
619 fscal = _mm_andnot_ps(dummy_mask,fscal);
621 /* Calculate temporary vectorial force */
622 tx = _mm_mul_ps(fscal,dx10);
623 ty = _mm_mul_ps(fscal,dy10);
624 tz = _mm_mul_ps(fscal,dz10);
626 /* Update vectorial force */
627 fix1 = _mm_add_ps(fix1,tx);
628 fiy1 = _mm_add_ps(fiy1,ty);
629 fiz1 = _mm_add_ps(fiz1,tz);
631 fjx0 = _mm_add_ps(fjx0,tx);
632 fjy0 = _mm_add_ps(fjy0,ty);
633 fjz0 = _mm_add_ps(fjz0,tz);
637 /**************************
638 * CALCULATE INTERACTIONS *
639 **************************/
641 if (gmx_mm_any_lt(rsq20,rcutoff2))
644 r20 = _mm_mul_ps(rsq20,rinv20);
645 r20 = _mm_andnot_ps(dummy_mask,r20);
647 /* Compute parameters for interactions between i and j atoms */
648 qq20 = _mm_mul_ps(iq2,jq0);
650 /* EWALD ELECTROSTATICS */
652 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
653 ewrt = _mm_mul_ps(r20,ewtabscale);
654 ewitab = _mm_cvttps_epi32(ewrt);
655 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
656 ewitab = _mm_slli_epi32(ewitab,2);
657 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
658 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
659 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
660 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
661 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
662 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
663 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
664 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
665 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
667 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
669 /* Update potential sum for this i atom from the interaction with this j atom. */
670 velec = _mm_and_ps(velec,cutoff_mask);
671 velec = _mm_andnot_ps(dummy_mask,velec);
672 velecsum = _mm_add_ps(velecsum,velec);
676 fscal = _mm_and_ps(fscal,cutoff_mask);
678 fscal = _mm_andnot_ps(dummy_mask,fscal);
680 /* Calculate temporary vectorial force */
681 tx = _mm_mul_ps(fscal,dx20);
682 ty = _mm_mul_ps(fscal,dy20);
683 tz = _mm_mul_ps(fscal,dz20);
685 /* Update vectorial force */
686 fix2 = _mm_add_ps(fix2,tx);
687 fiy2 = _mm_add_ps(fiy2,ty);
688 fiz2 = _mm_add_ps(fiz2,tz);
690 fjx0 = _mm_add_ps(fjx0,tx);
691 fjy0 = _mm_add_ps(fjy0,ty);
692 fjz0 = _mm_add_ps(fjz0,tz);
696 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
697 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
698 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
699 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
701 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
703 /* Inner loop uses 159 flops */
706 /* End of innermost loop */
708 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
709 f+i_coord_offset,fshift+i_shift_offset);
712 /* Update potential energies */
713 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
714 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
716 /* Increment number of inner iterations */
717 inneriter += j_index_end - j_index_start;
719 /* Outer loop uses 20 flops */
722 /* Increment number of outer iterations */
725 /* Update outer/inner flops */
727 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
730 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
731 * Electrostatics interaction: Ewald
732 * VdW interaction: LennardJones
733 * Geometry: Water3-Particle
734 * Calculate force/pot: Force
737 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
738 (t_nblist * gmx_restrict nlist,
739 rvec * gmx_restrict xx,
740 rvec * gmx_restrict ff,
741 t_forcerec * gmx_restrict fr,
742 t_mdatoms * gmx_restrict mdatoms,
743 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
744 t_nrnb * gmx_restrict nrnb)
746 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
747 * just 0 for non-waters.
748 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
749 * jnr indices corresponding to data put in the four positions in the SIMD register.
751 int i_shift_offset,i_coord_offset,outeriter,inneriter;
752 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
753 int jnrA,jnrB,jnrC,jnrD;
754 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
755 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
756 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
758 real *shiftvec,*fshift,*x,*f;
759 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
761 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
763 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
765 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
767 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
768 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
769 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
770 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
771 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
772 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
773 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
776 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
779 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
780 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
782 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
784 __m128 dummy_mask,cutoff_mask;
785 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
786 __m128 one = _mm_set1_ps(1.0);
787 __m128 two = _mm_set1_ps(2.0);
793 jindex = nlist->jindex;
795 shiftidx = nlist->shift;
797 shiftvec = fr->shift_vec[0];
798 fshift = fr->fshift[0];
799 facel = _mm_set1_ps(fr->epsfac);
800 charge = mdatoms->chargeA;
801 nvdwtype = fr->ntype;
803 vdwtype = mdatoms->typeA;
805 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
806 ewtab = fr->ic->tabq_coul_F;
807 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
808 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
810 /* Setup water-specific parameters */
811 inr = nlist->iinr[0];
812 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
813 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
814 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
815 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
817 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
818 rcutoff_scalar = fr->rcoulomb;
819 rcutoff = _mm_set1_ps(rcutoff_scalar);
820 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
822 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
823 rvdw = _mm_set1_ps(fr->rvdw);
825 /* Avoid stupid compiler warnings */
826 jnrA = jnrB = jnrC = jnrD = 0;
835 for(iidx=0;iidx<4*DIM;iidx++)
840 /* Start outer loop over neighborlists */
841 for(iidx=0; iidx<nri; iidx++)
843 /* Load shift vector for this list */
844 i_shift_offset = DIM*shiftidx[iidx];
846 /* Load limits for loop over neighbors */
847 j_index_start = jindex[iidx];
848 j_index_end = jindex[iidx+1];
850 /* Get outer coordinate index */
852 i_coord_offset = DIM*inr;
854 /* Load i particle coords and add shift vector */
855 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
856 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
858 fix0 = _mm_setzero_ps();
859 fiy0 = _mm_setzero_ps();
860 fiz0 = _mm_setzero_ps();
861 fix1 = _mm_setzero_ps();
862 fiy1 = _mm_setzero_ps();
863 fiz1 = _mm_setzero_ps();
864 fix2 = _mm_setzero_ps();
865 fiy2 = _mm_setzero_ps();
866 fiz2 = _mm_setzero_ps();
868 /* Start inner kernel loop */
869 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
872 /* Get j neighbor index, and coordinate index */
877 j_coord_offsetA = DIM*jnrA;
878 j_coord_offsetB = DIM*jnrB;
879 j_coord_offsetC = DIM*jnrC;
880 j_coord_offsetD = DIM*jnrD;
882 /* load j atom coordinates */
883 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
884 x+j_coord_offsetC,x+j_coord_offsetD,
887 /* Calculate displacement vector */
888 dx00 = _mm_sub_ps(ix0,jx0);
889 dy00 = _mm_sub_ps(iy0,jy0);
890 dz00 = _mm_sub_ps(iz0,jz0);
891 dx10 = _mm_sub_ps(ix1,jx0);
892 dy10 = _mm_sub_ps(iy1,jy0);
893 dz10 = _mm_sub_ps(iz1,jz0);
894 dx20 = _mm_sub_ps(ix2,jx0);
895 dy20 = _mm_sub_ps(iy2,jy0);
896 dz20 = _mm_sub_ps(iz2,jz0);
898 /* Calculate squared distance and things based on it */
899 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
900 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
901 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
903 rinv00 = gmx_mm_invsqrt_ps(rsq00);
904 rinv10 = gmx_mm_invsqrt_ps(rsq10);
905 rinv20 = gmx_mm_invsqrt_ps(rsq20);
907 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
908 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
909 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
911 /* Load parameters for j particles */
912 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
913 charge+jnrC+0,charge+jnrD+0);
914 vdwjidx0A = 2*vdwtype[jnrA+0];
915 vdwjidx0B = 2*vdwtype[jnrB+0];
916 vdwjidx0C = 2*vdwtype[jnrC+0];
917 vdwjidx0D = 2*vdwtype[jnrD+0];
919 fjx0 = _mm_setzero_ps();
920 fjy0 = _mm_setzero_ps();
921 fjz0 = _mm_setzero_ps();
923 /**************************
924 * CALCULATE INTERACTIONS *
925 **************************/
927 if (gmx_mm_any_lt(rsq00,rcutoff2))
930 r00 = _mm_mul_ps(rsq00,rinv00);
932 /* Compute parameters for interactions between i and j atoms */
933 qq00 = _mm_mul_ps(iq0,jq0);
934 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
935 vdwparam+vdwioffset0+vdwjidx0B,
936 vdwparam+vdwioffset0+vdwjidx0C,
937 vdwparam+vdwioffset0+vdwjidx0D,
940 /* EWALD ELECTROSTATICS */
942 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
943 ewrt = _mm_mul_ps(r00,ewtabscale);
944 ewitab = _mm_cvttps_epi32(ewrt);
945 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
946 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
947 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
949 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
950 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
952 /* LENNARD-JONES DISPERSION/REPULSION */
954 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
955 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
957 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
959 fscal = _mm_add_ps(felec,fvdw);
961 fscal = _mm_and_ps(fscal,cutoff_mask);
963 /* Calculate temporary vectorial force */
964 tx = _mm_mul_ps(fscal,dx00);
965 ty = _mm_mul_ps(fscal,dy00);
966 tz = _mm_mul_ps(fscal,dz00);
968 /* Update vectorial force */
969 fix0 = _mm_add_ps(fix0,tx);
970 fiy0 = _mm_add_ps(fiy0,ty);
971 fiz0 = _mm_add_ps(fiz0,tz);
973 fjx0 = _mm_add_ps(fjx0,tx);
974 fjy0 = _mm_add_ps(fjy0,ty);
975 fjz0 = _mm_add_ps(fjz0,tz);
979 /**************************
980 * CALCULATE INTERACTIONS *
981 **************************/
983 if (gmx_mm_any_lt(rsq10,rcutoff2))
986 r10 = _mm_mul_ps(rsq10,rinv10);
988 /* Compute parameters for interactions between i and j atoms */
989 qq10 = _mm_mul_ps(iq1,jq0);
991 /* EWALD ELECTROSTATICS */
993 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
994 ewrt = _mm_mul_ps(r10,ewtabscale);
995 ewitab = _mm_cvttps_epi32(ewrt);
996 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
997 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
998 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1000 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1001 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1003 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1007 fscal = _mm_and_ps(fscal,cutoff_mask);
1009 /* Calculate temporary vectorial force */
1010 tx = _mm_mul_ps(fscal,dx10);
1011 ty = _mm_mul_ps(fscal,dy10);
1012 tz = _mm_mul_ps(fscal,dz10);
1014 /* Update vectorial force */
1015 fix1 = _mm_add_ps(fix1,tx);
1016 fiy1 = _mm_add_ps(fiy1,ty);
1017 fiz1 = _mm_add_ps(fiz1,tz);
1019 fjx0 = _mm_add_ps(fjx0,tx);
1020 fjy0 = _mm_add_ps(fjy0,ty);
1021 fjz0 = _mm_add_ps(fjz0,tz);
1025 /**************************
1026 * CALCULATE INTERACTIONS *
1027 **************************/
1029 if (gmx_mm_any_lt(rsq20,rcutoff2))
1032 r20 = _mm_mul_ps(rsq20,rinv20);
1034 /* Compute parameters for interactions between i and j atoms */
1035 qq20 = _mm_mul_ps(iq2,jq0);
1037 /* EWALD ELECTROSTATICS */
1039 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1040 ewrt = _mm_mul_ps(r20,ewtabscale);
1041 ewitab = _mm_cvttps_epi32(ewrt);
1042 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1043 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1044 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1046 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1047 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1049 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1053 fscal = _mm_and_ps(fscal,cutoff_mask);
1055 /* Calculate temporary vectorial force */
1056 tx = _mm_mul_ps(fscal,dx20);
1057 ty = _mm_mul_ps(fscal,dy20);
1058 tz = _mm_mul_ps(fscal,dz20);
1060 /* Update vectorial force */
1061 fix2 = _mm_add_ps(fix2,tx);
1062 fiy2 = _mm_add_ps(fiy2,ty);
1063 fiz2 = _mm_add_ps(fiz2,tz);
1065 fjx0 = _mm_add_ps(fjx0,tx);
1066 fjy0 = _mm_add_ps(fjy0,ty);
1067 fjz0 = _mm_add_ps(fjz0,tz);
1071 fjptrA = f+j_coord_offsetA;
1072 fjptrB = f+j_coord_offsetB;
1073 fjptrC = f+j_coord_offsetC;
1074 fjptrD = f+j_coord_offsetD;
1076 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1078 /* Inner loop uses 124 flops */
1081 if(jidx<j_index_end)
1084 /* Get j neighbor index, and coordinate index */
1085 jnrlistA = jjnr[jidx];
1086 jnrlistB = jjnr[jidx+1];
1087 jnrlistC = jjnr[jidx+2];
1088 jnrlistD = jjnr[jidx+3];
1089 /* Sign of each element will be negative for non-real atoms.
1090 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1091 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1093 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1094 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1095 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1096 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1097 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1098 j_coord_offsetA = DIM*jnrA;
1099 j_coord_offsetB = DIM*jnrB;
1100 j_coord_offsetC = DIM*jnrC;
1101 j_coord_offsetD = DIM*jnrD;
1103 /* load j atom coordinates */
1104 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1105 x+j_coord_offsetC,x+j_coord_offsetD,
1108 /* Calculate displacement vector */
1109 dx00 = _mm_sub_ps(ix0,jx0);
1110 dy00 = _mm_sub_ps(iy0,jy0);
1111 dz00 = _mm_sub_ps(iz0,jz0);
1112 dx10 = _mm_sub_ps(ix1,jx0);
1113 dy10 = _mm_sub_ps(iy1,jy0);
1114 dz10 = _mm_sub_ps(iz1,jz0);
1115 dx20 = _mm_sub_ps(ix2,jx0);
1116 dy20 = _mm_sub_ps(iy2,jy0);
1117 dz20 = _mm_sub_ps(iz2,jz0);
1119 /* Calculate squared distance and things based on it */
1120 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1121 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1122 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1124 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1125 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1126 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1128 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1129 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1130 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1132 /* Load parameters for j particles */
1133 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1134 charge+jnrC+0,charge+jnrD+0);
1135 vdwjidx0A = 2*vdwtype[jnrA+0];
1136 vdwjidx0B = 2*vdwtype[jnrB+0];
1137 vdwjidx0C = 2*vdwtype[jnrC+0];
1138 vdwjidx0D = 2*vdwtype[jnrD+0];
1140 fjx0 = _mm_setzero_ps();
1141 fjy0 = _mm_setzero_ps();
1142 fjz0 = _mm_setzero_ps();
1144 /**************************
1145 * CALCULATE INTERACTIONS *
1146 **************************/
1148 if (gmx_mm_any_lt(rsq00,rcutoff2))
1151 r00 = _mm_mul_ps(rsq00,rinv00);
1152 r00 = _mm_andnot_ps(dummy_mask,r00);
1154 /* Compute parameters for interactions between i and j atoms */
1155 qq00 = _mm_mul_ps(iq0,jq0);
1156 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1157 vdwparam+vdwioffset0+vdwjidx0B,
1158 vdwparam+vdwioffset0+vdwjidx0C,
1159 vdwparam+vdwioffset0+vdwjidx0D,
1162 /* EWALD ELECTROSTATICS */
1164 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1165 ewrt = _mm_mul_ps(r00,ewtabscale);
1166 ewitab = _mm_cvttps_epi32(ewrt);
1167 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1168 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1169 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1171 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1172 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1174 /* LENNARD-JONES DISPERSION/REPULSION */
1176 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1177 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1179 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1181 fscal = _mm_add_ps(felec,fvdw);
1183 fscal = _mm_and_ps(fscal,cutoff_mask);
1185 fscal = _mm_andnot_ps(dummy_mask,fscal);
1187 /* Calculate temporary vectorial force */
1188 tx = _mm_mul_ps(fscal,dx00);
1189 ty = _mm_mul_ps(fscal,dy00);
1190 tz = _mm_mul_ps(fscal,dz00);
1192 /* Update vectorial force */
1193 fix0 = _mm_add_ps(fix0,tx);
1194 fiy0 = _mm_add_ps(fiy0,ty);
1195 fiz0 = _mm_add_ps(fiz0,tz);
1197 fjx0 = _mm_add_ps(fjx0,tx);
1198 fjy0 = _mm_add_ps(fjy0,ty);
1199 fjz0 = _mm_add_ps(fjz0,tz);
1203 /**************************
1204 * CALCULATE INTERACTIONS *
1205 **************************/
1207 if (gmx_mm_any_lt(rsq10,rcutoff2))
1210 r10 = _mm_mul_ps(rsq10,rinv10);
1211 r10 = _mm_andnot_ps(dummy_mask,r10);
1213 /* Compute parameters for interactions between i and j atoms */
1214 qq10 = _mm_mul_ps(iq1,jq0);
1216 /* EWALD ELECTROSTATICS */
1218 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1219 ewrt = _mm_mul_ps(r10,ewtabscale);
1220 ewitab = _mm_cvttps_epi32(ewrt);
1221 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1222 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1223 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1225 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1226 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1228 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1232 fscal = _mm_and_ps(fscal,cutoff_mask);
1234 fscal = _mm_andnot_ps(dummy_mask,fscal);
1236 /* Calculate temporary vectorial force */
1237 tx = _mm_mul_ps(fscal,dx10);
1238 ty = _mm_mul_ps(fscal,dy10);
1239 tz = _mm_mul_ps(fscal,dz10);
1241 /* Update vectorial force */
1242 fix1 = _mm_add_ps(fix1,tx);
1243 fiy1 = _mm_add_ps(fiy1,ty);
1244 fiz1 = _mm_add_ps(fiz1,tz);
1246 fjx0 = _mm_add_ps(fjx0,tx);
1247 fjy0 = _mm_add_ps(fjy0,ty);
1248 fjz0 = _mm_add_ps(fjz0,tz);
1252 /**************************
1253 * CALCULATE INTERACTIONS *
1254 **************************/
1256 if (gmx_mm_any_lt(rsq20,rcutoff2))
1259 r20 = _mm_mul_ps(rsq20,rinv20);
1260 r20 = _mm_andnot_ps(dummy_mask,r20);
1262 /* Compute parameters for interactions between i and j atoms */
1263 qq20 = _mm_mul_ps(iq2,jq0);
1265 /* EWALD ELECTROSTATICS */
1267 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1268 ewrt = _mm_mul_ps(r20,ewtabscale);
1269 ewitab = _mm_cvttps_epi32(ewrt);
1270 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1271 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1272 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1274 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1275 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1277 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1281 fscal = _mm_and_ps(fscal,cutoff_mask);
1283 fscal = _mm_andnot_ps(dummy_mask,fscal);
1285 /* Calculate temporary vectorial force */
1286 tx = _mm_mul_ps(fscal,dx20);
1287 ty = _mm_mul_ps(fscal,dy20);
1288 tz = _mm_mul_ps(fscal,dz20);
1290 /* Update vectorial force */
1291 fix2 = _mm_add_ps(fix2,tx);
1292 fiy2 = _mm_add_ps(fiy2,ty);
1293 fiz2 = _mm_add_ps(fiz2,tz);
1295 fjx0 = _mm_add_ps(fjx0,tx);
1296 fjy0 = _mm_add_ps(fjy0,ty);
1297 fjz0 = _mm_add_ps(fjz0,tz);
1301 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1302 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1303 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1304 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1306 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1308 /* Inner loop uses 127 flops */
1311 /* End of innermost loop */
1313 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1314 f+i_coord_offset,fshift+i_shift_offset);
1316 /* Increment number of inner iterations */
1317 inneriter += j_index_end - j_index_start;
1319 /* Outer loop uses 18 flops */
1322 /* Increment number of outer iterations */
1325 /* Update outer/inner flops */
1327 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);