2 * Note: this file was generated by the Gromacs sse2_single kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse2_single.h"
34 #include "kernelutil_x86_sse2_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_sse2_single
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_sse2_single
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
71 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
72 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
73 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
74 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
77 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
79 __m128 dummy_mask,cutoff_mask;
80 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
81 __m128 one = _mm_set1_ps(1.0);
82 __m128 two = _mm_set1_ps(2.0);
88 jindex = nlist->jindex;
90 shiftidx = nlist->shift;
92 shiftvec = fr->shift_vec[0];
93 fshift = fr->fshift[0];
94 facel = _mm_set1_ps(fr->epsfac);
95 charge = mdatoms->chargeA;
97 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
98 ewtab = fr->ic->tabq_coul_FDV0;
99 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
100 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
102 /* Avoid stupid compiler warnings */
103 jnrA = jnrB = jnrC = jnrD = 0;
112 for(iidx=0;iidx<4*DIM;iidx++)
117 /* Start outer loop over neighborlists */
118 for(iidx=0; iidx<nri; iidx++)
120 /* Load shift vector for this list */
121 i_shift_offset = DIM*shiftidx[iidx];
123 /* Load limits for loop over neighbors */
124 j_index_start = jindex[iidx];
125 j_index_end = jindex[iidx+1];
127 /* Get outer coordinate index */
129 i_coord_offset = DIM*inr;
131 /* Load i particle coords and add shift vector */
132 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
134 fix0 = _mm_setzero_ps();
135 fiy0 = _mm_setzero_ps();
136 fiz0 = _mm_setzero_ps();
138 /* Load parameters for i particles */
139 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
141 /* Reset potential sums */
142 velecsum = _mm_setzero_ps();
144 /* Start inner kernel loop */
145 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
148 /* Get j neighbor index, and coordinate index */
153 j_coord_offsetA = DIM*jnrA;
154 j_coord_offsetB = DIM*jnrB;
155 j_coord_offsetC = DIM*jnrC;
156 j_coord_offsetD = DIM*jnrD;
158 /* load j atom coordinates */
159 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
160 x+j_coord_offsetC,x+j_coord_offsetD,
163 /* Calculate displacement vector */
164 dx00 = _mm_sub_ps(ix0,jx0);
165 dy00 = _mm_sub_ps(iy0,jy0);
166 dz00 = _mm_sub_ps(iz0,jz0);
168 /* Calculate squared distance and things based on it */
169 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
171 rinv00 = gmx_mm_invsqrt_ps(rsq00);
173 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
175 /* Load parameters for j particles */
176 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
177 charge+jnrC+0,charge+jnrD+0);
179 /**************************
180 * CALCULATE INTERACTIONS *
181 **************************/
183 r00 = _mm_mul_ps(rsq00,rinv00);
185 /* Compute parameters for interactions between i and j atoms */
186 qq00 = _mm_mul_ps(iq0,jq0);
188 /* EWALD ELECTROSTATICS */
190 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
191 ewrt = _mm_mul_ps(r00,ewtabscale);
192 ewitab = _mm_cvttps_epi32(ewrt);
193 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
194 ewitab = _mm_slli_epi32(ewitab,2);
195 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
196 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
197 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
198 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
199 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
200 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
201 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
202 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
203 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
205 /* Update potential sum for this i atom from the interaction with this j atom. */
206 velecsum = _mm_add_ps(velecsum,velec);
210 /* Calculate temporary vectorial force */
211 tx = _mm_mul_ps(fscal,dx00);
212 ty = _mm_mul_ps(fscal,dy00);
213 tz = _mm_mul_ps(fscal,dz00);
215 /* Update vectorial force */
216 fix0 = _mm_add_ps(fix0,tx);
217 fiy0 = _mm_add_ps(fiy0,ty);
218 fiz0 = _mm_add_ps(fiz0,tz);
220 fjptrA = f+j_coord_offsetA;
221 fjptrB = f+j_coord_offsetB;
222 fjptrC = f+j_coord_offsetC;
223 fjptrD = f+j_coord_offsetD;
224 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
226 /* Inner loop uses 41 flops */
232 /* Get j neighbor index, and coordinate index */
233 jnrlistA = jjnr[jidx];
234 jnrlistB = jjnr[jidx+1];
235 jnrlistC = jjnr[jidx+2];
236 jnrlistD = jjnr[jidx+3];
237 /* Sign of each element will be negative for non-real atoms.
238 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
239 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
241 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
242 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
243 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
244 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
245 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
246 j_coord_offsetA = DIM*jnrA;
247 j_coord_offsetB = DIM*jnrB;
248 j_coord_offsetC = DIM*jnrC;
249 j_coord_offsetD = DIM*jnrD;
251 /* load j atom coordinates */
252 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
253 x+j_coord_offsetC,x+j_coord_offsetD,
256 /* Calculate displacement vector */
257 dx00 = _mm_sub_ps(ix0,jx0);
258 dy00 = _mm_sub_ps(iy0,jy0);
259 dz00 = _mm_sub_ps(iz0,jz0);
261 /* Calculate squared distance and things based on it */
262 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
264 rinv00 = gmx_mm_invsqrt_ps(rsq00);
266 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
268 /* Load parameters for j particles */
269 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
270 charge+jnrC+0,charge+jnrD+0);
272 /**************************
273 * CALCULATE INTERACTIONS *
274 **************************/
276 r00 = _mm_mul_ps(rsq00,rinv00);
277 r00 = _mm_andnot_ps(dummy_mask,r00);
279 /* Compute parameters for interactions between i and j atoms */
280 qq00 = _mm_mul_ps(iq0,jq0);
282 /* EWALD ELECTROSTATICS */
284 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
285 ewrt = _mm_mul_ps(r00,ewtabscale);
286 ewitab = _mm_cvttps_epi32(ewrt);
287 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
288 ewitab = _mm_slli_epi32(ewitab,2);
289 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
290 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
291 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
292 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
293 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
294 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
295 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
296 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
297 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
299 /* Update potential sum for this i atom from the interaction with this j atom. */
300 velec = _mm_andnot_ps(dummy_mask,velec);
301 velecsum = _mm_add_ps(velecsum,velec);
305 fscal = _mm_andnot_ps(dummy_mask,fscal);
307 /* Calculate temporary vectorial force */
308 tx = _mm_mul_ps(fscal,dx00);
309 ty = _mm_mul_ps(fscal,dy00);
310 tz = _mm_mul_ps(fscal,dz00);
312 /* Update vectorial force */
313 fix0 = _mm_add_ps(fix0,tx);
314 fiy0 = _mm_add_ps(fiy0,ty);
315 fiz0 = _mm_add_ps(fiz0,tz);
317 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
318 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
319 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
320 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
321 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
323 /* Inner loop uses 42 flops */
326 /* End of innermost loop */
328 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
329 f+i_coord_offset,fshift+i_shift_offset);
332 /* Update potential energies */
333 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
335 /* Increment number of inner iterations */
336 inneriter += j_index_end - j_index_start;
338 /* Outer loop uses 8 flops */
341 /* Increment number of outer iterations */
344 /* Update outer/inner flops */
346 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*42);
349 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse2_single
350 * Electrostatics interaction: Ewald
351 * VdW interaction: None
352 * Geometry: Particle-Particle
353 * Calculate force/pot: Force
356 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse2_single
357 (t_nblist * gmx_restrict nlist,
358 rvec * gmx_restrict xx,
359 rvec * gmx_restrict ff,
360 t_forcerec * gmx_restrict fr,
361 t_mdatoms * gmx_restrict mdatoms,
362 nb_kernel_data_t * gmx_restrict kernel_data,
363 t_nrnb * gmx_restrict nrnb)
365 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
366 * just 0 for non-waters.
367 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
368 * jnr indices corresponding to data put in the four positions in the SIMD register.
370 int i_shift_offset,i_coord_offset,outeriter,inneriter;
371 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
372 int jnrA,jnrB,jnrC,jnrD;
373 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
374 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
375 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
377 real *shiftvec,*fshift,*x,*f;
378 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
380 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
382 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
383 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
384 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
385 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
386 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
389 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
391 __m128 dummy_mask,cutoff_mask;
392 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
393 __m128 one = _mm_set1_ps(1.0);
394 __m128 two = _mm_set1_ps(2.0);
400 jindex = nlist->jindex;
402 shiftidx = nlist->shift;
404 shiftvec = fr->shift_vec[0];
405 fshift = fr->fshift[0];
406 facel = _mm_set1_ps(fr->epsfac);
407 charge = mdatoms->chargeA;
409 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
410 ewtab = fr->ic->tabq_coul_F;
411 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
412 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
414 /* Avoid stupid compiler warnings */
415 jnrA = jnrB = jnrC = jnrD = 0;
424 for(iidx=0;iidx<4*DIM;iidx++)
429 /* Start outer loop over neighborlists */
430 for(iidx=0; iidx<nri; iidx++)
432 /* Load shift vector for this list */
433 i_shift_offset = DIM*shiftidx[iidx];
435 /* Load limits for loop over neighbors */
436 j_index_start = jindex[iidx];
437 j_index_end = jindex[iidx+1];
439 /* Get outer coordinate index */
441 i_coord_offset = DIM*inr;
443 /* Load i particle coords and add shift vector */
444 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
446 fix0 = _mm_setzero_ps();
447 fiy0 = _mm_setzero_ps();
448 fiz0 = _mm_setzero_ps();
450 /* Load parameters for i particles */
451 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
453 /* Start inner kernel loop */
454 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
457 /* Get j neighbor index, and coordinate index */
462 j_coord_offsetA = DIM*jnrA;
463 j_coord_offsetB = DIM*jnrB;
464 j_coord_offsetC = DIM*jnrC;
465 j_coord_offsetD = DIM*jnrD;
467 /* load j atom coordinates */
468 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
469 x+j_coord_offsetC,x+j_coord_offsetD,
472 /* Calculate displacement vector */
473 dx00 = _mm_sub_ps(ix0,jx0);
474 dy00 = _mm_sub_ps(iy0,jy0);
475 dz00 = _mm_sub_ps(iz0,jz0);
477 /* Calculate squared distance and things based on it */
478 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
480 rinv00 = gmx_mm_invsqrt_ps(rsq00);
482 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
484 /* Load parameters for j particles */
485 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
486 charge+jnrC+0,charge+jnrD+0);
488 /**************************
489 * CALCULATE INTERACTIONS *
490 **************************/
492 r00 = _mm_mul_ps(rsq00,rinv00);
494 /* Compute parameters for interactions between i and j atoms */
495 qq00 = _mm_mul_ps(iq0,jq0);
497 /* EWALD ELECTROSTATICS */
499 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
500 ewrt = _mm_mul_ps(r00,ewtabscale);
501 ewitab = _mm_cvttps_epi32(ewrt);
502 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
503 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
504 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
506 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
507 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
511 /* Calculate temporary vectorial force */
512 tx = _mm_mul_ps(fscal,dx00);
513 ty = _mm_mul_ps(fscal,dy00);
514 tz = _mm_mul_ps(fscal,dz00);
516 /* Update vectorial force */
517 fix0 = _mm_add_ps(fix0,tx);
518 fiy0 = _mm_add_ps(fiy0,ty);
519 fiz0 = _mm_add_ps(fiz0,tz);
521 fjptrA = f+j_coord_offsetA;
522 fjptrB = f+j_coord_offsetB;
523 fjptrC = f+j_coord_offsetC;
524 fjptrD = f+j_coord_offsetD;
525 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
527 /* Inner loop uses 36 flops */
533 /* Get j neighbor index, and coordinate index */
534 jnrlistA = jjnr[jidx];
535 jnrlistB = jjnr[jidx+1];
536 jnrlistC = jjnr[jidx+2];
537 jnrlistD = jjnr[jidx+3];
538 /* Sign of each element will be negative for non-real atoms.
539 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
540 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
542 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
543 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
544 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
545 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
546 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
547 j_coord_offsetA = DIM*jnrA;
548 j_coord_offsetB = DIM*jnrB;
549 j_coord_offsetC = DIM*jnrC;
550 j_coord_offsetD = DIM*jnrD;
552 /* load j atom coordinates */
553 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
554 x+j_coord_offsetC,x+j_coord_offsetD,
557 /* Calculate displacement vector */
558 dx00 = _mm_sub_ps(ix0,jx0);
559 dy00 = _mm_sub_ps(iy0,jy0);
560 dz00 = _mm_sub_ps(iz0,jz0);
562 /* Calculate squared distance and things based on it */
563 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
565 rinv00 = gmx_mm_invsqrt_ps(rsq00);
567 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
569 /* Load parameters for j particles */
570 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
571 charge+jnrC+0,charge+jnrD+0);
573 /**************************
574 * CALCULATE INTERACTIONS *
575 **************************/
577 r00 = _mm_mul_ps(rsq00,rinv00);
578 r00 = _mm_andnot_ps(dummy_mask,r00);
580 /* Compute parameters for interactions between i and j atoms */
581 qq00 = _mm_mul_ps(iq0,jq0);
583 /* EWALD ELECTROSTATICS */
585 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
586 ewrt = _mm_mul_ps(r00,ewtabscale);
587 ewitab = _mm_cvttps_epi32(ewrt);
588 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
589 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
590 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
592 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
593 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
597 fscal = _mm_andnot_ps(dummy_mask,fscal);
599 /* Calculate temporary vectorial force */
600 tx = _mm_mul_ps(fscal,dx00);
601 ty = _mm_mul_ps(fscal,dy00);
602 tz = _mm_mul_ps(fscal,dz00);
604 /* Update vectorial force */
605 fix0 = _mm_add_ps(fix0,tx);
606 fiy0 = _mm_add_ps(fiy0,ty);
607 fiz0 = _mm_add_ps(fiz0,tz);
609 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
610 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
611 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
612 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
613 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
615 /* Inner loop uses 37 flops */
618 /* End of innermost loop */
620 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
621 f+i_coord_offset,fshift+i_shift_offset);
623 /* Increment number of inner iterations */
624 inneriter += j_index_end - j_index_start;
626 /* Outer loop uses 7 flops */
629 /* Increment number of outer iterations */
632 /* Update outer/inner flops */
634 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*37);