2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
50 #include "kernelutil_x86_avx_128_fma_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
54 * Electrostatics interaction: Coulomb
55 * VdW interaction: CubicSplineTable
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
102 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
103 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
105 __m128i ifour = _mm_set1_epi32(4);
106 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
108 __m128 dummy_mask,cutoff_mask;
109 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
110 __m128 one = _mm_set1_ps(1.0);
111 __m128 two = _mm_set1_ps(2.0);
117 jindex = nlist->jindex;
119 shiftidx = nlist->shift;
121 shiftvec = fr->shift_vec[0];
122 fshift = fr->fshift[0];
123 facel = _mm_set1_ps(fr->epsfac);
124 charge = mdatoms->chargeA;
125 nvdwtype = fr->ntype;
127 vdwtype = mdatoms->typeA;
129 vftab = kernel_data->table_vdw->data;
130 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
132 /* Setup water-specific parameters */
133 inr = nlist->iinr[0];
134 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
135 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
136 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
137 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
139 /* Avoid stupid compiler warnings */
140 jnrA = jnrB = jnrC = jnrD = 0;
149 for(iidx=0;iidx<4*DIM;iidx++)
154 /* Start outer loop over neighborlists */
155 for(iidx=0; iidx<nri; iidx++)
157 /* Load shift vector for this list */
158 i_shift_offset = DIM*shiftidx[iidx];
160 /* Load limits for loop over neighbors */
161 j_index_start = jindex[iidx];
162 j_index_end = jindex[iidx+1];
164 /* Get outer coordinate index */
166 i_coord_offset = DIM*inr;
168 /* Load i particle coords and add shift vector */
169 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
170 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
172 fix0 = _mm_setzero_ps();
173 fiy0 = _mm_setzero_ps();
174 fiz0 = _mm_setzero_ps();
175 fix1 = _mm_setzero_ps();
176 fiy1 = _mm_setzero_ps();
177 fiz1 = _mm_setzero_ps();
178 fix2 = _mm_setzero_ps();
179 fiy2 = _mm_setzero_ps();
180 fiz2 = _mm_setzero_ps();
182 /* Reset potential sums */
183 velecsum = _mm_setzero_ps();
184 vvdwsum = _mm_setzero_ps();
186 /* Start inner kernel loop */
187 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
190 /* Get j neighbor index, and coordinate index */
195 j_coord_offsetA = DIM*jnrA;
196 j_coord_offsetB = DIM*jnrB;
197 j_coord_offsetC = DIM*jnrC;
198 j_coord_offsetD = DIM*jnrD;
200 /* load j atom coordinates */
201 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
202 x+j_coord_offsetC,x+j_coord_offsetD,
205 /* Calculate displacement vector */
206 dx00 = _mm_sub_ps(ix0,jx0);
207 dy00 = _mm_sub_ps(iy0,jy0);
208 dz00 = _mm_sub_ps(iz0,jz0);
209 dx10 = _mm_sub_ps(ix1,jx0);
210 dy10 = _mm_sub_ps(iy1,jy0);
211 dz10 = _mm_sub_ps(iz1,jz0);
212 dx20 = _mm_sub_ps(ix2,jx0);
213 dy20 = _mm_sub_ps(iy2,jy0);
214 dz20 = _mm_sub_ps(iz2,jz0);
216 /* Calculate squared distance and things based on it */
217 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
218 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
219 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
221 rinv00 = gmx_mm_invsqrt_ps(rsq00);
222 rinv10 = gmx_mm_invsqrt_ps(rsq10);
223 rinv20 = gmx_mm_invsqrt_ps(rsq20);
225 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
226 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
227 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
229 /* Load parameters for j particles */
230 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
231 charge+jnrC+0,charge+jnrD+0);
232 vdwjidx0A = 2*vdwtype[jnrA+0];
233 vdwjidx0B = 2*vdwtype[jnrB+0];
234 vdwjidx0C = 2*vdwtype[jnrC+0];
235 vdwjidx0D = 2*vdwtype[jnrD+0];
237 fjx0 = _mm_setzero_ps();
238 fjy0 = _mm_setzero_ps();
239 fjz0 = _mm_setzero_ps();
241 /**************************
242 * CALCULATE INTERACTIONS *
243 **************************/
245 r00 = _mm_mul_ps(rsq00,rinv00);
247 /* Compute parameters for interactions between i and j atoms */
248 qq00 = _mm_mul_ps(iq0,jq0);
249 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
250 vdwparam+vdwioffset0+vdwjidx0B,
251 vdwparam+vdwioffset0+vdwjidx0C,
252 vdwparam+vdwioffset0+vdwjidx0D,
255 /* Calculate table index by multiplying r with table scale and truncate to integer */
256 rt = _mm_mul_ps(r00,vftabscale);
257 vfitab = _mm_cvttps_epi32(rt);
259 vfeps = _mm_frcz_ps(rt);
261 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
263 twovfeps = _mm_add_ps(vfeps,vfeps);
264 vfitab = _mm_slli_epi32(vfitab,3);
266 /* COULOMB ELECTROSTATICS */
267 velec = _mm_mul_ps(qq00,rinv00);
268 felec = _mm_mul_ps(velec,rinvsq00);
270 /* CUBIC SPLINE TABLE DISPERSION */
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 vvdw6 = _mm_mul_ps(c6_00,VV);
279 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
280 fvdw6 = _mm_mul_ps(c6_00,FF);
282 /* CUBIC SPLINE TABLE REPULSION */
283 vfitab = _mm_add_epi32(vfitab,ifour);
284 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
285 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
286 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
287 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
288 _MM_TRANSPOSE4_PS(Y,F,G,H);
289 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
290 VV = _mm_macc_ps(vfeps,Fp,Y);
291 vvdw12 = _mm_mul_ps(c12_00,VV);
292 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
293 fvdw12 = _mm_mul_ps(c12_00,FF);
294 vvdw = _mm_add_ps(vvdw12,vvdw6);
295 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
297 /* Update potential sum for this i atom from the interaction with this j atom. */
298 velecsum = _mm_add_ps(velecsum,velec);
299 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
301 fscal = _mm_add_ps(felec,fvdw);
303 /* Update vectorial force */
304 fix0 = _mm_macc_ps(dx00,fscal,fix0);
305 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
306 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
308 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
309 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
310 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
312 /**************************
313 * CALCULATE INTERACTIONS *
314 **************************/
316 /* Compute parameters for interactions between i and j atoms */
317 qq10 = _mm_mul_ps(iq1,jq0);
319 /* COULOMB ELECTROSTATICS */
320 velec = _mm_mul_ps(qq10,rinv10);
321 felec = _mm_mul_ps(velec,rinvsq10);
323 /* Update potential sum for this i atom from the interaction with this j atom. */
324 velecsum = _mm_add_ps(velecsum,velec);
328 /* Update vectorial force */
329 fix1 = _mm_macc_ps(dx10,fscal,fix1);
330 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
331 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
333 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
334 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
335 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
337 /**************************
338 * CALCULATE INTERACTIONS *
339 **************************/
341 /* Compute parameters for interactions between i and j atoms */
342 qq20 = _mm_mul_ps(iq2,jq0);
344 /* COULOMB ELECTROSTATICS */
345 velec = _mm_mul_ps(qq20,rinv20);
346 felec = _mm_mul_ps(velec,rinvsq20);
348 /* Update potential sum for this i atom from the interaction with this j atom. */
349 velecsum = _mm_add_ps(velecsum,velec);
353 /* Update vectorial force */
354 fix2 = _mm_macc_ps(dx20,fscal,fix2);
355 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
356 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
358 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
359 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
360 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
362 fjptrA = f+j_coord_offsetA;
363 fjptrB = f+j_coord_offsetB;
364 fjptrC = f+j_coord_offsetC;
365 fjptrD = f+j_coord_offsetD;
367 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
369 /* Inner loop uses 128 flops */
375 /* Get j neighbor index, and coordinate index */
376 jnrlistA = jjnr[jidx];
377 jnrlistB = jjnr[jidx+1];
378 jnrlistC = jjnr[jidx+2];
379 jnrlistD = jjnr[jidx+3];
380 /* Sign of each element will be negative for non-real atoms.
381 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
382 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
384 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
385 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
386 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
387 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
388 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
389 j_coord_offsetA = DIM*jnrA;
390 j_coord_offsetB = DIM*jnrB;
391 j_coord_offsetC = DIM*jnrC;
392 j_coord_offsetD = DIM*jnrD;
394 /* load j atom coordinates */
395 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
396 x+j_coord_offsetC,x+j_coord_offsetD,
399 /* Calculate displacement vector */
400 dx00 = _mm_sub_ps(ix0,jx0);
401 dy00 = _mm_sub_ps(iy0,jy0);
402 dz00 = _mm_sub_ps(iz0,jz0);
403 dx10 = _mm_sub_ps(ix1,jx0);
404 dy10 = _mm_sub_ps(iy1,jy0);
405 dz10 = _mm_sub_ps(iz1,jz0);
406 dx20 = _mm_sub_ps(ix2,jx0);
407 dy20 = _mm_sub_ps(iy2,jy0);
408 dz20 = _mm_sub_ps(iz2,jz0);
410 /* Calculate squared distance and things based on it */
411 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
412 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
413 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
415 rinv00 = gmx_mm_invsqrt_ps(rsq00);
416 rinv10 = gmx_mm_invsqrt_ps(rsq10);
417 rinv20 = gmx_mm_invsqrt_ps(rsq20);
419 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
420 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
421 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
423 /* Load parameters for j particles */
424 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
425 charge+jnrC+0,charge+jnrD+0);
426 vdwjidx0A = 2*vdwtype[jnrA+0];
427 vdwjidx0B = 2*vdwtype[jnrB+0];
428 vdwjidx0C = 2*vdwtype[jnrC+0];
429 vdwjidx0D = 2*vdwtype[jnrD+0];
431 fjx0 = _mm_setzero_ps();
432 fjy0 = _mm_setzero_ps();
433 fjz0 = _mm_setzero_ps();
435 /**************************
436 * CALCULATE INTERACTIONS *
437 **************************/
439 r00 = _mm_mul_ps(rsq00,rinv00);
440 r00 = _mm_andnot_ps(dummy_mask,r00);
442 /* Compute parameters for interactions between i and j atoms */
443 qq00 = _mm_mul_ps(iq0,jq0);
444 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
445 vdwparam+vdwioffset0+vdwjidx0B,
446 vdwparam+vdwioffset0+vdwjidx0C,
447 vdwparam+vdwioffset0+vdwjidx0D,
450 /* Calculate table index by multiplying r with table scale and truncate to integer */
451 rt = _mm_mul_ps(r00,vftabscale);
452 vfitab = _mm_cvttps_epi32(rt);
454 vfeps = _mm_frcz_ps(rt);
456 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
458 twovfeps = _mm_add_ps(vfeps,vfeps);
459 vfitab = _mm_slli_epi32(vfitab,3);
461 /* COULOMB ELECTROSTATICS */
462 velec = _mm_mul_ps(qq00,rinv00);
463 felec = _mm_mul_ps(velec,rinvsq00);
465 /* CUBIC SPLINE TABLE DISPERSION */
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 vvdw6 = _mm_mul_ps(c6_00,VV);
474 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
475 fvdw6 = _mm_mul_ps(c6_00,FF);
477 /* CUBIC SPLINE TABLE REPULSION */
478 vfitab = _mm_add_epi32(vfitab,ifour);
479 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
480 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
481 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
482 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
483 _MM_TRANSPOSE4_PS(Y,F,G,H);
484 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
485 VV = _mm_macc_ps(vfeps,Fp,Y);
486 vvdw12 = _mm_mul_ps(c12_00,VV);
487 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
488 fvdw12 = _mm_mul_ps(c12_00,FF);
489 vvdw = _mm_add_ps(vvdw12,vvdw6);
490 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
492 /* Update potential sum for this i atom from the interaction with this j atom. */
493 velec = _mm_andnot_ps(dummy_mask,velec);
494 velecsum = _mm_add_ps(velecsum,velec);
495 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
496 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
498 fscal = _mm_add_ps(felec,fvdw);
500 fscal = _mm_andnot_ps(dummy_mask,fscal);
502 /* Update vectorial force */
503 fix0 = _mm_macc_ps(dx00,fscal,fix0);
504 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
505 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
507 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
508 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
509 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
511 /**************************
512 * CALCULATE INTERACTIONS *
513 **************************/
515 /* Compute parameters for interactions between i and j atoms */
516 qq10 = _mm_mul_ps(iq1,jq0);
518 /* COULOMB ELECTROSTATICS */
519 velec = _mm_mul_ps(qq10,rinv10);
520 felec = _mm_mul_ps(velec,rinvsq10);
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velec = _mm_andnot_ps(dummy_mask,velec);
524 velecsum = _mm_add_ps(velecsum,velec);
528 fscal = _mm_andnot_ps(dummy_mask,fscal);
530 /* Update vectorial force */
531 fix1 = _mm_macc_ps(dx10,fscal,fix1);
532 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
533 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
535 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
536 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
537 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
539 /**************************
540 * CALCULATE INTERACTIONS *
541 **************************/
543 /* Compute parameters for interactions between i and j atoms */
544 qq20 = _mm_mul_ps(iq2,jq0);
546 /* COULOMB ELECTROSTATICS */
547 velec = _mm_mul_ps(qq20,rinv20);
548 felec = _mm_mul_ps(velec,rinvsq20);
550 /* Update potential sum for this i atom from the interaction with this j atom. */
551 velec = _mm_andnot_ps(dummy_mask,velec);
552 velecsum = _mm_add_ps(velecsum,velec);
556 fscal = _mm_andnot_ps(dummy_mask,fscal);
558 /* Update vectorial force */
559 fix2 = _mm_macc_ps(dx20,fscal,fix2);
560 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
561 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
563 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
564 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
565 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
567 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
568 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
569 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
570 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
572 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
574 /* Inner loop uses 129 flops */
577 /* End of innermost loop */
579 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
580 f+i_coord_offset,fshift+i_shift_offset);
583 /* Update potential energies */
584 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
585 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
587 /* Increment number of inner iterations */
588 inneriter += j_index_end - j_index_start;
590 /* Outer loop uses 20 flops */
593 /* Increment number of outer iterations */
596 /* Update outer/inner flops */
598 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
601 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_single
602 * Electrostatics interaction: Coulomb
603 * VdW interaction: CubicSplineTable
604 * Geometry: Water3-Particle
605 * Calculate force/pot: Force
608 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_single
609 (t_nblist * gmx_restrict nlist,
610 rvec * gmx_restrict xx,
611 rvec * gmx_restrict ff,
612 t_forcerec * gmx_restrict fr,
613 t_mdatoms * gmx_restrict mdatoms,
614 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
615 t_nrnb * gmx_restrict nrnb)
617 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
618 * just 0 for non-waters.
619 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
620 * jnr indices corresponding to data put in the four positions in the SIMD register.
622 int i_shift_offset,i_coord_offset,outeriter,inneriter;
623 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
624 int jnrA,jnrB,jnrC,jnrD;
625 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
626 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
627 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
629 real *shiftvec,*fshift,*x,*f;
630 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
632 __m128 fscal,rcutoff,rcutoff2,jidxall;
634 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
636 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
638 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
639 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
640 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
641 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
642 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
643 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
644 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
647 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
650 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
651 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
653 __m128i ifour = _mm_set1_epi32(4);
654 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
656 __m128 dummy_mask,cutoff_mask;
657 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
658 __m128 one = _mm_set1_ps(1.0);
659 __m128 two = _mm_set1_ps(2.0);
665 jindex = nlist->jindex;
667 shiftidx = nlist->shift;
669 shiftvec = fr->shift_vec[0];
670 fshift = fr->fshift[0];
671 facel = _mm_set1_ps(fr->epsfac);
672 charge = mdatoms->chargeA;
673 nvdwtype = fr->ntype;
675 vdwtype = mdatoms->typeA;
677 vftab = kernel_data->table_vdw->data;
678 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
680 /* Setup water-specific parameters */
681 inr = nlist->iinr[0];
682 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
683 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
684 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
685 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
687 /* Avoid stupid compiler warnings */
688 jnrA = jnrB = jnrC = jnrD = 0;
697 for(iidx=0;iidx<4*DIM;iidx++)
702 /* Start outer loop over neighborlists */
703 for(iidx=0; iidx<nri; iidx++)
705 /* Load shift vector for this list */
706 i_shift_offset = DIM*shiftidx[iidx];
708 /* Load limits for loop over neighbors */
709 j_index_start = jindex[iidx];
710 j_index_end = jindex[iidx+1];
712 /* Get outer coordinate index */
714 i_coord_offset = DIM*inr;
716 /* Load i particle coords and add shift vector */
717 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
718 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
720 fix0 = _mm_setzero_ps();
721 fiy0 = _mm_setzero_ps();
722 fiz0 = _mm_setzero_ps();
723 fix1 = _mm_setzero_ps();
724 fiy1 = _mm_setzero_ps();
725 fiz1 = _mm_setzero_ps();
726 fix2 = _mm_setzero_ps();
727 fiy2 = _mm_setzero_ps();
728 fiz2 = _mm_setzero_ps();
730 /* Start inner kernel loop */
731 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
734 /* Get j neighbor index, and coordinate index */
739 j_coord_offsetA = DIM*jnrA;
740 j_coord_offsetB = DIM*jnrB;
741 j_coord_offsetC = DIM*jnrC;
742 j_coord_offsetD = DIM*jnrD;
744 /* load j atom coordinates */
745 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
746 x+j_coord_offsetC,x+j_coord_offsetD,
749 /* Calculate displacement vector */
750 dx00 = _mm_sub_ps(ix0,jx0);
751 dy00 = _mm_sub_ps(iy0,jy0);
752 dz00 = _mm_sub_ps(iz0,jz0);
753 dx10 = _mm_sub_ps(ix1,jx0);
754 dy10 = _mm_sub_ps(iy1,jy0);
755 dz10 = _mm_sub_ps(iz1,jz0);
756 dx20 = _mm_sub_ps(ix2,jx0);
757 dy20 = _mm_sub_ps(iy2,jy0);
758 dz20 = _mm_sub_ps(iz2,jz0);
760 /* Calculate squared distance and things based on it */
761 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
762 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
763 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
765 rinv00 = gmx_mm_invsqrt_ps(rsq00);
766 rinv10 = gmx_mm_invsqrt_ps(rsq10);
767 rinv20 = gmx_mm_invsqrt_ps(rsq20);
769 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
770 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
771 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
773 /* Load parameters for j particles */
774 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
775 charge+jnrC+0,charge+jnrD+0);
776 vdwjidx0A = 2*vdwtype[jnrA+0];
777 vdwjidx0B = 2*vdwtype[jnrB+0];
778 vdwjidx0C = 2*vdwtype[jnrC+0];
779 vdwjidx0D = 2*vdwtype[jnrD+0];
781 fjx0 = _mm_setzero_ps();
782 fjy0 = _mm_setzero_ps();
783 fjz0 = _mm_setzero_ps();
785 /**************************
786 * CALCULATE INTERACTIONS *
787 **************************/
789 r00 = _mm_mul_ps(rsq00,rinv00);
791 /* Compute parameters for interactions between i and j atoms */
792 qq00 = _mm_mul_ps(iq0,jq0);
793 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
794 vdwparam+vdwioffset0+vdwjidx0B,
795 vdwparam+vdwioffset0+vdwjidx0C,
796 vdwparam+vdwioffset0+vdwjidx0D,
799 /* Calculate table index by multiplying r with table scale and truncate to integer */
800 rt = _mm_mul_ps(r00,vftabscale);
801 vfitab = _mm_cvttps_epi32(rt);
803 vfeps = _mm_frcz_ps(rt);
805 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
807 twovfeps = _mm_add_ps(vfeps,vfeps);
808 vfitab = _mm_slli_epi32(vfitab,3);
810 /* COULOMB ELECTROSTATICS */
811 velec = _mm_mul_ps(qq00,rinv00);
812 felec = _mm_mul_ps(velec,rinvsq00);
814 /* CUBIC SPLINE TABLE DISPERSION */
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 fvdw6 = _mm_mul_ps(c6_00,FF);
824 /* CUBIC SPLINE TABLE REPULSION */
825 vfitab = _mm_add_epi32(vfitab,ifour);
826 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
827 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
828 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
829 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
830 _MM_TRANSPOSE4_PS(Y,F,G,H);
831 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
832 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
833 fvdw12 = _mm_mul_ps(c12_00,FF);
834 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
836 fscal = _mm_add_ps(felec,fvdw);
838 /* Update vectorial force */
839 fix0 = _mm_macc_ps(dx00,fscal,fix0);
840 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
841 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
843 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
844 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
845 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
847 /**************************
848 * CALCULATE INTERACTIONS *
849 **************************/
851 /* Compute parameters for interactions between i and j atoms */
852 qq10 = _mm_mul_ps(iq1,jq0);
854 /* COULOMB ELECTROSTATICS */
855 velec = _mm_mul_ps(qq10,rinv10);
856 felec = _mm_mul_ps(velec,rinvsq10);
860 /* Update vectorial force */
861 fix1 = _mm_macc_ps(dx10,fscal,fix1);
862 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
863 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
865 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
866 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
867 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
869 /**************************
870 * CALCULATE INTERACTIONS *
871 **************************/
873 /* Compute parameters for interactions between i and j atoms */
874 qq20 = _mm_mul_ps(iq2,jq0);
876 /* COULOMB ELECTROSTATICS */
877 velec = _mm_mul_ps(qq20,rinv20);
878 felec = _mm_mul_ps(velec,rinvsq20);
882 /* Update vectorial force */
883 fix2 = _mm_macc_ps(dx20,fscal,fix2);
884 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
885 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
887 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
888 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
889 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
891 fjptrA = f+j_coord_offsetA;
892 fjptrB = f+j_coord_offsetB;
893 fjptrC = f+j_coord_offsetC;
894 fjptrD = f+j_coord_offsetD;
896 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
898 /* Inner loop uses 117 flops */
904 /* Get j neighbor index, and coordinate index */
905 jnrlistA = jjnr[jidx];
906 jnrlistB = jjnr[jidx+1];
907 jnrlistC = jjnr[jidx+2];
908 jnrlistD = jjnr[jidx+3];
909 /* Sign of each element will be negative for non-real atoms.
910 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
911 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
913 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
914 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
915 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
916 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
917 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
918 j_coord_offsetA = DIM*jnrA;
919 j_coord_offsetB = DIM*jnrB;
920 j_coord_offsetC = DIM*jnrC;
921 j_coord_offsetD = DIM*jnrD;
923 /* load j atom coordinates */
924 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
925 x+j_coord_offsetC,x+j_coord_offsetD,
928 /* Calculate displacement vector */
929 dx00 = _mm_sub_ps(ix0,jx0);
930 dy00 = _mm_sub_ps(iy0,jy0);
931 dz00 = _mm_sub_ps(iz0,jz0);
932 dx10 = _mm_sub_ps(ix1,jx0);
933 dy10 = _mm_sub_ps(iy1,jy0);
934 dz10 = _mm_sub_ps(iz1,jz0);
935 dx20 = _mm_sub_ps(ix2,jx0);
936 dy20 = _mm_sub_ps(iy2,jy0);
937 dz20 = _mm_sub_ps(iz2,jz0);
939 /* Calculate squared distance and things based on it */
940 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
941 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
942 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
944 rinv00 = gmx_mm_invsqrt_ps(rsq00);
945 rinv10 = gmx_mm_invsqrt_ps(rsq10);
946 rinv20 = gmx_mm_invsqrt_ps(rsq20);
948 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
949 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
950 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
952 /* Load parameters for j particles */
953 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
954 charge+jnrC+0,charge+jnrD+0);
955 vdwjidx0A = 2*vdwtype[jnrA+0];
956 vdwjidx0B = 2*vdwtype[jnrB+0];
957 vdwjidx0C = 2*vdwtype[jnrC+0];
958 vdwjidx0D = 2*vdwtype[jnrD+0];
960 fjx0 = _mm_setzero_ps();
961 fjy0 = _mm_setzero_ps();
962 fjz0 = _mm_setzero_ps();
964 /**************************
965 * CALCULATE INTERACTIONS *
966 **************************/
968 r00 = _mm_mul_ps(rsq00,rinv00);
969 r00 = _mm_andnot_ps(dummy_mask,r00);
971 /* Compute parameters for interactions between i and j atoms */
972 qq00 = _mm_mul_ps(iq0,jq0);
973 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
974 vdwparam+vdwioffset0+vdwjidx0B,
975 vdwparam+vdwioffset0+vdwjidx0C,
976 vdwparam+vdwioffset0+vdwjidx0D,
979 /* Calculate table index by multiplying r with table scale and truncate to integer */
980 rt = _mm_mul_ps(r00,vftabscale);
981 vfitab = _mm_cvttps_epi32(rt);
983 vfeps = _mm_frcz_ps(rt);
985 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
987 twovfeps = _mm_add_ps(vfeps,vfeps);
988 vfitab = _mm_slli_epi32(vfitab,3);
990 /* COULOMB ELECTROSTATICS */
991 velec = _mm_mul_ps(qq00,rinv00);
992 felec = _mm_mul_ps(velec,rinvsq00);
994 /* CUBIC SPLINE TABLE DISPERSION */
995 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
996 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
997 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
998 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
999 _MM_TRANSPOSE4_PS(Y,F,G,H);
1000 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1001 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1002 fvdw6 = _mm_mul_ps(c6_00,FF);
1004 /* CUBIC SPLINE TABLE REPULSION */
1005 vfitab = _mm_add_epi32(vfitab,ifour);
1006 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1007 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1008 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1009 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1010 _MM_TRANSPOSE4_PS(Y,F,G,H);
1011 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1012 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1013 fvdw12 = _mm_mul_ps(c12_00,FF);
1014 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1016 fscal = _mm_add_ps(felec,fvdw);
1018 fscal = _mm_andnot_ps(dummy_mask,fscal);
1020 /* Update vectorial force */
1021 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1022 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1023 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1025 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1026 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1027 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1029 /**************************
1030 * CALCULATE INTERACTIONS *
1031 **************************/
1033 /* Compute parameters for interactions between i and j atoms */
1034 qq10 = _mm_mul_ps(iq1,jq0);
1036 /* COULOMB ELECTROSTATICS */
1037 velec = _mm_mul_ps(qq10,rinv10);
1038 felec = _mm_mul_ps(velec,rinvsq10);
1042 fscal = _mm_andnot_ps(dummy_mask,fscal);
1044 /* Update vectorial force */
1045 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1046 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1047 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1049 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1050 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1051 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1053 /**************************
1054 * CALCULATE INTERACTIONS *
1055 **************************/
1057 /* Compute parameters for interactions between i and j atoms */
1058 qq20 = _mm_mul_ps(iq2,jq0);
1060 /* COULOMB ELECTROSTATICS */
1061 velec = _mm_mul_ps(qq20,rinv20);
1062 felec = _mm_mul_ps(velec,rinvsq20);
1066 fscal = _mm_andnot_ps(dummy_mask,fscal);
1068 /* Update vectorial force */
1069 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1070 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1071 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1073 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1074 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1075 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1077 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1078 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1079 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1080 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1082 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1084 /* Inner loop uses 118 flops */
1087 /* End of innermost loop */
1089 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1090 f+i_coord_offset,fshift+i_shift_offset);
1092 /* Increment number of inner iterations */
1093 inneriter += j_index_end - j_index_start;
1095 /* Outer loop uses 18 flops */
1098 /* Increment number of outer iterations */
1101 /* Update outer/inner flops */
1103 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);