2 * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_single.h"
34 #include "kernelutil_x86_avx_128_fma_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: CubicSplineTable
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_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 AVX_128, 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 fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
72 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
75 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
76 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
77 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
78 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
83 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
86 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
87 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
89 __m128i ifour = _mm_set1_epi32(4);
90 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
92 __m128 dummy_mask,cutoff_mask;
93 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
94 __m128 one = _mm_set1_ps(1.0);
95 __m128 two = _mm_set1_ps(2.0);
101 jindex = nlist->jindex;
103 shiftidx = nlist->shift;
105 shiftvec = fr->shift_vec[0];
106 fshift = fr->fshift[0];
107 facel = _mm_set1_ps(fr->epsfac);
108 charge = mdatoms->chargeA;
109 krf = _mm_set1_ps(fr->ic->k_rf);
110 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
111 crf = _mm_set1_ps(fr->ic->c_rf);
112 nvdwtype = fr->ntype;
114 vdwtype = mdatoms->typeA;
116 vftab = kernel_data->table_vdw->data;
117 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
119 /* Setup water-specific parameters */
120 inr = nlist->iinr[0];
121 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
122 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
123 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
124 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
126 /* Avoid stupid compiler warnings */
127 jnrA = jnrB = jnrC = jnrD = 0;
136 for(iidx=0;iidx<4*DIM;iidx++)
141 /* Start outer loop over neighborlists */
142 for(iidx=0; iidx<nri; iidx++)
144 /* Load shift vector for this list */
145 i_shift_offset = DIM*shiftidx[iidx];
147 /* Load limits for loop over neighbors */
148 j_index_start = jindex[iidx];
149 j_index_end = jindex[iidx+1];
151 /* Get outer coordinate index */
153 i_coord_offset = DIM*inr;
155 /* Load i particle coords and add shift vector */
156 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
157 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
159 fix0 = _mm_setzero_ps();
160 fiy0 = _mm_setzero_ps();
161 fiz0 = _mm_setzero_ps();
162 fix1 = _mm_setzero_ps();
163 fiy1 = _mm_setzero_ps();
164 fiz1 = _mm_setzero_ps();
165 fix2 = _mm_setzero_ps();
166 fiy2 = _mm_setzero_ps();
167 fiz2 = _mm_setzero_ps();
169 /* Reset potential sums */
170 velecsum = _mm_setzero_ps();
171 vvdwsum = _mm_setzero_ps();
173 /* Start inner kernel loop */
174 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
177 /* Get j neighbor index, and coordinate index */
182 j_coord_offsetA = DIM*jnrA;
183 j_coord_offsetB = DIM*jnrB;
184 j_coord_offsetC = DIM*jnrC;
185 j_coord_offsetD = DIM*jnrD;
187 /* load j atom coordinates */
188 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
189 x+j_coord_offsetC,x+j_coord_offsetD,
192 /* Calculate displacement vector */
193 dx00 = _mm_sub_ps(ix0,jx0);
194 dy00 = _mm_sub_ps(iy0,jy0);
195 dz00 = _mm_sub_ps(iz0,jz0);
196 dx10 = _mm_sub_ps(ix1,jx0);
197 dy10 = _mm_sub_ps(iy1,jy0);
198 dz10 = _mm_sub_ps(iz1,jz0);
199 dx20 = _mm_sub_ps(ix2,jx0);
200 dy20 = _mm_sub_ps(iy2,jy0);
201 dz20 = _mm_sub_ps(iz2,jz0);
203 /* Calculate squared distance and things based on it */
204 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
205 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
206 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
208 rinv00 = gmx_mm_invsqrt_ps(rsq00);
209 rinv10 = gmx_mm_invsqrt_ps(rsq10);
210 rinv20 = gmx_mm_invsqrt_ps(rsq20);
212 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
213 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
214 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
216 /* Load parameters for j particles */
217 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
218 charge+jnrC+0,charge+jnrD+0);
219 vdwjidx0A = 2*vdwtype[jnrA+0];
220 vdwjidx0B = 2*vdwtype[jnrB+0];
221 vdwjidx0C = 2*vdwtype[jnrC+0];
222 vdwjidx0D = 2*vdwtype[jnrD+0];
224 fjx0 = _mm_setzero_ps();
225 fjy0 = _mm_setzero_ps();
226 fjz0 = _mm_setzero_ps();
228 /**************************
229 * CALCULATE INTERACTIONS *
230 **************************/
232 r00 = _mm_mul_ps(rsq00,rinv00);
234 /* Compute parameters for interactions between i and j atoms */
235 qq00 = _mm_mul_ps(iq0,jq0);
236 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
237 vdwparam+vdwioffset0+vdwjidx0B,
238 vdwparam+vdwioffset0+vdwjidx0C,
239 vdwparam+vdwioffset0+vdwjidx0D,
242 /* Calculate table index by multiplying r with table scale and truncate to integer */
243 rt = _mm_mul_ps(r00,vftabscale);
244 vfitab = _mm_cvttps_epi32(rt);
246 vfeps = _mm_frcz_ps(rt);
248 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
250 twovfeps = _mm_add_ps(vfeps,vfeps);
251 vfitab = _mm_slli_epi32(vfitab,3);
253 /* REACTION-FIELD ELECTROSTATICS */
254 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
255 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
257 /* CUBIC SPLINE TABLE DISPERSION */
258 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
259 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
260 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
261 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
262 _MM_TRANSPOSE4_PS(Y,F,G,H);
263 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
264 VV = _mm_macc_ps(vfeps,Fp,Y);
265 vvdw6 = _mm_mul_ps(c6_00,VV);
266 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
267 fvdw6 = _mm_mul_ps(c6_00,FF);
269 /* CUBIC SPLINE TABLE REPULSION */
270 vfitab = _mm_add_epi32(vfitab,ifour);
271 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
272 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
273 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
274 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
275 _MM_TRANSPOSE4_PS(Y,F,G,H);
276 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
277 VV = _mm_macc_ps(vfeps,Fp,Y);
278 vvdw12 = _mm_mul_ps(c12_00,VV);
279 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
280 fvdw12 = _mm_mul_ps(c12_00,FF);
281 vvdw = _mm_add_ps(vvdw12,vvdw6);
282 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
284 /* Update potential sum for this i atom from the interaction with this j atom. */
285 velecsum = _mm_add_ps(velecsum,velec);
286 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
288 fscal = _mm_add_ps(felec,fvdw);
290 /* Update vectorial force */
291 fix0 = _mm_macc_ps(dx00,fscal,fix0);
292 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
293 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
295 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
296 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
297 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
299 /**************************
300 * CALCULATE INTERACTIONS *
301 **************************/
303 /* Compute parameters for interactions between i and j atoms */
304 qq10 = _mm_mul_ps(iq1,jq0);
306 /* REACTION-FIELD ELECTROSTATICS */
307 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
308 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
310 /* Update potential sum for this i atom from the interaction with this j atom. */
311 velecsum = _mm_add_ps(velecsum,velec);
315 /* Update vectorial force */
316 fix1 = _mm_macc_ps(dx10,fscal,fix1);
317 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
318 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
320 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
321 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
322 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
324 /**************************
325 * CALCULATE INTERACTIONS *
326 **************************/
328 /* Compute parameters for interactions between i and j atoms */
329 qq20 = _mm_mul_ps(iq2,jq0);
331 /* REACTION-FIELD ELECTROSTATICS */
332 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
333 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
335 /* Update potential sum for this i atom from the interaction with this j atom. */
336 velecsum = _mm_add_ps(velecsum,velec);
340 /* Update vectorial force */
341 fix2 = _mm_macc_ps(dx20,fscal,fix2);
342 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
343 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
345 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
346 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
347 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
349 fjptrA = f+j_coord_offsetA;
350 fjptrB = f+j_coord_offsetB;
351 fjptrC = f+j_coord_offsetC;
352 fjptrD = f+j_coord_offsetD;
354 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
356 /* Inner loop uses 140 flops */
362 /* Get j neighbor index, and coordinate index */
363 jnrlistA = jjnr[jidx];
364 jnrlistB = jjnr[jidx+1];
365 jnrlistC = jjnr[jidx+2];
366 jnrlistD = jjnr[jidx+3];
367 /* Sign of each element will be negative for non-real atoms.
368 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
369 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
371 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
372 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
373 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
374 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
375 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
376 j_coord_offsetA = DIM*jnrA;
377 j_coord_offsetB = DIM*jnrB;
378 j_coord_offsetC = DIM*jnrC;
379 j_coord_offsetD = DIM*jnrD;
381 /* load j atom coordinates */
382 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
383 x+j_coord_offsetC,x+j_coord_offsetD,
386 /* Calculate displacement vector */
387 dx00 = _mm_sub_ps(ix0,jx0);
388 dy00 = _mm_sub_ps(iy0,jy0);
389 dz00 = _mm_sub_ps(iz0,jz0);
390 dx10 = _mm_sub_ps(ix1,jx0);
391 dy10 = _mm_sub_ps(iy1,jy0);
392 dz10 = _mm_sub_ps(iz1,jz0);
393 dx20 = _mm_sub_ps(ix2,jx0);
394 dy20 = _mm_sub_ps(iy2,jy0);
395 dz20 = _mm_sub_ps(iz2,jz0);
397 /* Calculate squared distance and things based on it */
398 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
399 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
400 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
402 rinv00 = gmx_mm_invsqrt_ps(rsq00);
403 rinv10 = gmx_mm_invsqrt_ps(rsq10);
404 rinv20 = gmx_mm_invsqrt_ps(rsq20);
406 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
407 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
408 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
410 /* Load parameters for j particles */
411 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
412 charge+jnrC+0,charge+jnrD+0);
413 vdwjidx0A = 2*vdwtype[jnrA+0];
414 vdwjidx0B = 2*vdwtype[jnrB+0];
415 vdwjidx0C = 2*vdwtype[jnrC+0];
416 vdwjidx0D = 2*vdwtype[jnrD+0];
418 fjx0 = _mm_setzero_ps();
419 fjy0 = _mm_setzero_ps();
420 fjz0 = _mm_setzero_ps();
422 /**************************
423 * CALCULATE INTERACTIONS *
424 **************************/
426 r00 = _mm_mul_ps(rsq00,rinv00);
427 r00 = _mm_andnot_ps(dummy_mask,r00);
429 /* Compute parameters for interactions between i and j atoms */
430 qq00 = _mm_mul_ps(iq0,jq0);
431 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
432 vdwparam+vdwioffset0+vdwjidx0B,
433 vdwparam+vdwioffset0+vdwjidx0C,
434 vdwparam+vdwioffset0+vdwjidx0D,
437 /* Calculate table index by multiplying r with table scale and truncate to integer */
438 rt = _mm_mul_ps(r00,vftabscale);
439 vfitab = _mm_cvttps_epi32(rt);
441 vfeps = _mm_frcz_ps(rt);
443 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
445 twovfeps = _mm_add_ps(vfeps,vfeps);
446 vfitab = _mm_slli_epi32(vfitab,3);
448 /* REACTION-FIELD ELECTROSTATICS */
449 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
450 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
452 /* CUBIC SPLINE TABLE DISPERSION */
453 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
454 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
455 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
456 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
457 _MM_TRANSPOSE4_PS(Y,F,G,H);
458 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
459 VV = _mm_macc_ps(vfeps,Fp,Y);
460 vvdw6 = _mm_mul_ps(c6_00,VV);
461 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
462 fvdw6 = _mm_mul_ps(c6_00,FF);
464 /* CUBIC SPLINE TABLE REPULSION */
465 vfitab = _mm_add_epi32(vfitab,ifour);
466 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
467 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
468 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
469 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
470 _MM_TRANSPOSE4_PS(Y,F,G,H);
471 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
472 VV = _mm_macc_ps(vfeps,Fp,Y);
473 vvdw12 = _mm_mul_ps(c12_00,VV);
474 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
475 fvdw12 = _mm_mul_ps(c12_00,FF);
476 vvdw = _mm_add_ps(vvdw12,vvdw6);
477 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
479 /* Update potential sum for this i atom from the interaction with this j atom. */
480 velec = _mm_andnot_ps(dummy_mask,velec);
481 velecsum = _mm_add_ps(velecsum,velec);
482 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
483 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
485 fscal = _mm_add_ps(felec,fvdw);
487 fscal = _mm_andnot_ps(dummy_mask,fscal);
489 /* Update vectorial force */
490 fix0 = _mm_macc_ps(dx00,fscal,fix0);
491 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
492 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
494 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
495 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
496 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
498 /**************************
499 * CALCULATE INTERACTIONS *
500 **************************/
502 /* Compute parameters for interactions between i and j atoms */
503 qq10 = _mm_mul_ps(iq1,jq0);
505 /* REACTION-FIELD ELECTROSTATICS */
506 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
507 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
509 /* Update potential sum for this i atom from the interaction with this j atom. */
510 velec = _mm_andnot_ps(dummy_mask,velec);
511 velecsum = _mm_add_ps(velecsum,velec);
515 fscal = _mm_andnot_ps(dummy_mask,fscal);
517 /* Update vectorial force */
518 fix1 = _mm_macc_ps(dx10,fscal,fix1);
519 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
520 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
522 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
523 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
524 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
526 /**************************
527 * CALCULATE INTERACTIONS *
528 **************************/
530 /* Compute parameters for interactions between i and j atoms */
531 qq20 = _mm_mul_ps(iq2,jq0);
533 /* REACTION-FIELD ELECTROSTATICS */
534 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
535 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
537 /* Update potential sum for this i atom from the interaction with this j atom. */
538 velec = _mm_andnot_ps(dummy_mask,velec);
539 velecsum = _mm_add_ps(velecsum,velec);
543 fscal = _mm_andnot_ps(dummy_mask,fscal);
545 /* Update vectorial force */
546 fix2 = _mm_macc_ps(dx20,fscal,fix2);
547 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
548 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
550 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
551 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
552 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
554 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
555 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
556 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
557 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
559 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
561 /* Inner loop uses 141 flops */
564 /* End of innermost loop */
566 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
567 f+i_coord_offset,fshift+i_shift_offset);
570 /* Update potential energies */
571 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
572 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
574 /* Increment number of inner iterations */
575 inneriter += j_index_end - j_index_start;
577 /* Outer loop uses 20 flops */
580 /* Increment number of outer iterations */
583 /* Update outer/inner flops */
585 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*141);
588 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
589 * Electrostatics interaction: ReactionField
590 * VdW interaction: CubicSplineTable
591 * Geometry: Water3-Particle
592 * Calculate force/pot: Force
595 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
596 (t_nblist * gmx_restrict nlist,
597 rvec * gmx_restrict xx,
598 rvec * gmx_restrict ff,
599 t_forcerec * gmx_restrict fr,
600 t_mdatoms * gmx_restrict mdatoms,
601 nb_kernel_data_t * gmx_restrict kernel_data,
602 t_nrnb * gmx_restrict nrnb)
604 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
605 * just 0 for non-waters.
606 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
607 * jnr indices corresponding to data put in the four positions in the SIMD register.
609 int i_shift_offset,i_coord_offset,outeriter,inneriter;
610 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
611 int jnrA,jnrB,jnrC,jnrD;
612 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
613 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
614 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
616 real *shiftvec,*fshift,*x,*f;
617 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
619 __m128 fscal,rcutoff,rcutoff2,jidxall;
621 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
623 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
625 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
626 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
627 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
628 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
629 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
630 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
631 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
634 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
637 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
638 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
640 __m128i ifour = _mm_set1_epi32(4);
641 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
643 __m128 dummy_mask,cutoff_mask;
644 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
645 __m128 one = _mm_set1_ps(1.0);
646 __m128 two = _mm_set1_ps(2.0);
652 jindex = nlist->jindex;
654 shiftidx = nlist->shift;
656 shiftvec = fr->shift_vec[0];
657 fshift = fr->fshift[0];
658 facel = _mm_set1_ps(fr->epsfac);
659 charge = mdatoms->chargeA;
660 krf = _mm_set1_ps(fr->ic->k_rf);
661 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
662 crf = _mm_set1_ps(fr->ic->c_rf);
663 nvdwtype = fr->ntype;
665 vdwtype = mdatoms->typeA;
667 vftab = kernel_data->table_vdw->data;
668 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
670 /* Setup water-specific parameters */
671 inr = nlist->iinr[0];
672 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
673 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
674 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
675 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
677 /* Avoid stupid compiler warnings */
678 jnrA = jnrB = jnrC = jnrD = 0;
687 for(iidx=0;iidx<4*DIM;iidx++)
692 /* Start outer loop over neighborlists */
693 for(iidx=0; iidx<nri; iidx++)
695 /* Load shift vector for this list */
696 i_shift_offset = DIM*shiftidx[iidx];
698 /* Load limits for loop over neighbors */
699 j_index_start = jindex[iidx];
700 j_index_end = jindex[iidx+1];
702 /* Get outer coordinate index */
704 i_coord_offset = DIM*inr;
706 /* Load i particle coords and add shift vector */
707 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
708 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
710 fix0 = _mm_setzero_ps();
711 fiy0 = _mm_setzero_ps();
712 fiz0 = _mm_setzero_ps();
713 fix1 = _mm_setzero_ps();
714 fiy1 = _mm_setzero_ps();
715 fiz1 = _mm_setzero_ps();
716 fix2 = _mm_setzero_ps();
717 fiy2 = _mm_setzero_ps();
718 fiz2 = _mm_setzero_ps();
720 /* Start inner kernel loop */
721 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
724 /* Get j neighbor index, and coordinate index */
729 j_coord_offsetA = DIM*jnrA;
730 j_coord_offsetB = DIM*jnrB;
731 j_coord_offsetC = DIM*jnrC;
732 j_coord_offsetD = DIM*jnrD;
734 /* load j atom coordinates */
735 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
736 x+j_coord_offsetC,x+j_coord_offsetD,
739 /* Calculate displacement vector */
740 dx00 = _mm_sub_ps(ix0,jx0);
741 dy00 = _mm_sub_ps(iy0,jy0);
742 dz00 = _mm_sub_ps(iz0,jz0);
743 dx10 = _mm_sub_ps(ix1,jx0);
744 dy10 = _mm_sub_ps(iy1,jy0);
745 dz10 = _mm_sub_ps(iz1,jz0);
746 dx20 = _mm_sub_ps(ix2,jx0);
747 dy20 = _mm_sub_ps(iy2,jy0);
748 dz20 = _mm_sub_ps(iz2,jz0);
750 /* Calculate squared distance and things based on it */
751 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
752 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
753 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
755 rinv00 = gmx_mm_invsqrt_ps(rsq00);
756 rinv10 = gmx_mm_invsqrt_ps(rsq10);
757 rinv20 = gmx_mm_invsqrt_ps(rsq20);
759 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
760 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
761 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
763 /* Load parameters for j particles */
764 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
765 charge+jnrC+0,charge+jnrD+0);
766 vdwjidx0A = 2*vdwtype[jnrA+0];
767 vdwjidx0B = 2*vdwtype[jnrB+0];
768 vdwjidx0C = 2*vdwtype[jnrC+0];
769 vdwjidx0D = 2*vdwtype[jnrD+0];
771 fjx0 = _mm_setzero_ps();
772 fjy0 = _mm_setzero_ps();
773 fjz0 = _mm_setzero_ps();
775 /**************************
776 * CALCULATE INTERACTIONS *
777 **************************/
779 r00 = _mm_mul_ps(rsq00,rinv00);
781 /* Compute parameters for interactions between i and j atoms */
782 qq00 = _mm_mul_ps(iq0,jq0);
783 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
784 vdwparam+vdwioffset0+vdwjidx0B,
785 vdwparam+vdwioffset0+vdwjidx0C,
786 vdwparam+vdwioffset0+vdwjidx0D,
789 /* Calculate table index by multiplying r with table scale and truncate to integer */
790 rt = _mm_mul_ps(r00,vftabscale);
791 vfitab = _mm_cvttps_epi32(rt);
793 vfeps = _mm_frcz_ps(rt);
795 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
797 twovfeps = _mm_add_ps(vfeps,vfeps);
798 vfitab = _mm_slli_epi32(vfitab,3);
800 /* REACTION-FIELD ELECTROSTATICS */
801 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
803 /* CUBIC SPLINE TABLE DISPERSION */
804 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
805 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
806 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
807 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
808 _MM_TRANSPOSE4_PS(Y,F,G,H);
809 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
810 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
811 fvdw6 = _mm_mul_ps(c6_00,FF);
813 /* CUBIC SPLINE TABLE REPULSION */
814 vfitab = _mm_add_epi32(vfitab,ifour);
815 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
816 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
817 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
818 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
819 _MM_TRANSPOSE4_PS(Y,F,G,H);
820 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
821 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
822 fvdw12 = _mm_mul_ps(c12_00,FF);
823 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
825 fscal = _mm_add_ps(felec,fvdw);
827 /* Update vectorial force */
828 fix0 = _mm_macc_ps(dx00,fscal,fix0);
829 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
830 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
832 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
833 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
834 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
836 /**************************
837 * CALCULATE INTERACTIONS *
838 **************************/
840 /* Compute parameters for interactions between i and j atoms */
841 qq10 = _mm_mul_ps(iq1,jq0);
843 /* REACTION-FIELD ELECTROSTATICS */
844 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
848 /* Update vectorial force */
849 fix1 = _mm_macc_ps(dx10,fscal,fix1);
850 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
851 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
853 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
854 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
855 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
857 /**************************
858 * CALCULATE INTERACTIONS *
859 **************************/
861 /* Compute parameters for interactions between i and j atoms */
862 qq20 = _mm_mul_ps(iq2,jq0);
864 /* REACTION-FIELD ELECTROSTATICS */
865 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
869 /* Update vectorial force */
870 fix2 = _mm_macc_ps(dx20,fscal,fix2);
871 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
872 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
874 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
875 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
876 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
878 fjptrA = f+j_coord_offsetA;
879 fjptrB = f+j_coord_offsetB;
880 fjptrC = f+j_coord_offsetC;
881 fjptrD = f+j_coord_offsetD;
883 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
885 /* Inner loop uses 117 flops */
891 /* Get j neighbor index, and coordinate index */
892 jnrlistA = jjnr[jidx];
893 jnrlistB = jjnr[jidx+1];
894 jnrlistC = jjnr[jidx+2];
895 jnrlistD = jjnr[jidx+3];
896 /* Sign of each element will be negative for non-real atoms.
897 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
898 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
900 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
901 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
902 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
903 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
904 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
905 j_coord_offsetA = DIM*jnrA;
906 j_coord_offsetB = DIM*jnrB;
907 j_coord_offsetC = DIM*jnrC;
908 j_coord_offsetD = DIM*jnrD;
910 /* load j atom coordinates */
911 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
912 x+j_coord_offsetC,x+j_coord_offsetD,
915 /* Calculate displacement vector */
916 dx00 = _mm_sub_ps(ix0,jx0);
917 dy00 = _mm_sub_ps(iy0,jy0);
918 dz00 = _mm_sub_ps(iz0,jz0);
919 dx10 = _mm_sub_ps(ix1,jx0);
920 dy10 = _mm_sub_ps(iy1,jy0);
921 dz10 = _mm_sub_ps(iz1,jz0);
922 dx20 = _mm_sub_ps(ix2,jx0);
923 dy20 = _mm_sub_ps(iy2,jy0);
924 dz20 = _mm_sub_ps(iz2,jz0);
926 /* Calculate squared distance and things based on it */
927 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
928 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
929 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
931 rinv00 = gmx_mm_invsqrt_ps(rsq00);
932 rinv10 = gmx_mm_invsqrt_ps(rsq10);
933 rinv20 = gmx_mm_invsqrt_ps(rsq20);
935 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
936 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
937 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
939 /* Load parameters for j particles */
940 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
941 charge+jnrC+0,charge+jnrD+0);
942 vdwjidx0A = 2*vdwtype[jnrA+0];
943 vdwjidx0B = 2*vdwtype[jnrB+0];
944 vdwjidx0C = 2*vdwtype[jnrC+0];
945 vdwjidx0D = 2*vdwtype[jnrD+0];
947 fjx0 = _mm_setzero_ps();
948 fjy0 = _mm_setzero_ps();
949 fjz0 = _mm_setzero_ps();
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 r00 = _mm_mul_ps(rsq00,rinv00);
956 r00 = _mm_andnot_ps(dummy_mask,r00);
958 /* Compute parameters for interactions between i and j atoms */
959 qq00 = _mm_mul_ps(iq0,jq0);
960 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
961 vdwparam+vdwioffset0+vdwjidx0B,
962 vdwparam+vdwioffset0+vdwjidx0C,
963 vdwparam+vdwioffset0+vdwjidx0D,
966 /* Calculate table index by multiplying r with table scale and truncate to integer */
967 rt = _mm_mul_ps(r00,vftabscale);
968 vfitab = _mm_cvttps_epi32(rt);
970 vfeps = _mm_frcz_ps(rt);
972 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
974 twovfeps = _mm_add_ps(vfeps,vfeps);
975 vfitab = _mm_slli_epi32(vfitab,3);
977 /* REACTION-FIELD ELECTROSTATICS */
978 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
980 /* CUBIC SPLINE TABLE DISPERSION */
981 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
982 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
983 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
984 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
985 _MM_TRANSPOSE4_PS(Y,F,G,H);
986 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
987 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
988 fvdw6 = _mm_mul_ps(c6_00,FF);
990 /* CUBIC SPLINE TABLE REPULSION */
991 vfitab = _mm_add_epi32(vfitab,ifour);
992 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
993 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
994 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
995 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
996 _MM_TRANSPOSE4_PS(Y,F,G,H);
997 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
998 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
999 fvdw12 = _mm_mul_ps(c12_00,FF);
1000 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1002 fscal = _mm_add_ps(felec,fvdw);
1004 fscal = _mm_andnot_ps(dummy_mask,fscal);
1006 /* Update vectorial force */
1007 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1008 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1009 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1011 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1012 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1013 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1015 /**************************
1016 * CALCULATE INTERACTIONS *
1017 **************************/
1019 /* Compute parameters for interactions between i and j atoms */
1020 qq10 = _mm_mul_ps(iq1,jq0);
1022 /* REACTION-FIELD ELECTROSTATICS */
1023 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1027 fscal = _mm_andnot_ps(dummy_mask,fscal);
1029 /* Update vectorial force */
1030 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1031 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1032 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1034 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1035 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1036 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1038 /**************************
1039 * CALCULATE INTERACTIONS *
1040 **************************/
1042 /* Compute parameters for interactions between i and j atoms */
1043 qq20 = _mm_mul_ps(iq2,jq0);
1045 /* REACTION-FIELD ELECTROSTATICS */
1046 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1050 fscal = _mm_andnot_ps(dummy_mask,fscal);
1052 /* Update vectorial force */
1053 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1054 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1055 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1057 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1058 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1059 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1061 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1062 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1063 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1064 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1066 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1068 /* Inner loop uses 118 flops */
1071 /* End of innermost loop */
1073 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1074 f+i_coord_offset,fshift+i_shift_offset);
1076 /* Increment number of inner iterations */
1077 inneriter += j_index_end - j_index_start;
1079 /* Outer loop uses 18 flops */
1082 /* Increment number of outer iterations */
1085 /* Update outer/inner flops */
1087 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);