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 avx_128_fma_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
48 #include "kernelutil_x86_avx_128_fma_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: CubicSplineTable
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_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 AVX_128, 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 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;
90 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
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 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
97 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
100 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
104 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
106 __m128i ifour = _mm_set1_epi32(4);
107 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
109 __m128 dummy_mask,cutoff_mask;
110 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
111 __m128 one = _mm_set1_ps(1.0);
112 __m128 two = _mm_set1_ps(2.0);
118 jindex = nlist->jindex;
120 shiftidx = nlist->shift;
122 shiftvec = fr->shift_vec[0];
123 fshift = fr->fshift[0];
124 facel = _mm_set1_ps(fr->epsfac);
125 charge = mdatoms->chargeA;
126 nvdwtype = fr->ntype;
128 vdwtype = mdatoms->typeA;
130 vftab = kernel_data->table_elec_vdw->data;
131 vftabscale = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
133 /* Setup water-specific parameters */
134 inr = nlist->iinr[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 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
138 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
140 /* Avoid stupid compiler warnings */
141 jnrA = jnrB = jnrC = jnrD = 0;
150 for(iidx=0;iidx<4*DIM;iidx++)
155 /* Start outer loop over neighborlists */
156 for(iidx=0; iidx<nri; iidx++)
158 /* Load shift vector for this list */
159 i_shift_offset = DIM*shiftidx[iidx];
161 /* Load limits for loop over neighbors */
162 j_index_start = jindex[iidx];
163 j_index_end = jindex[iidx+1];
165 /* Get outer coordinate index */
167 i_coord_offset = DIM*inr;
169 /* Load i particle coords and add shift vector */
170 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
171 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
173 fix0 = _mm_setzero_ps();
174 fiy0 = _mm_setzero_ps();
175 fiz0 = _mm_setzero_ps();
176 fix1 = _mm_setzero_ps();
177 fiy1 = _mm_setzero_ps();
178 fiz1 = _mm_setzero_ps();
179 fix2 = _mm_setzero_ps();
180 fiy2 = _mm_setzero_ps();
181 fiz2 = _mm_setzero_ps();
182 fix3 = _mm_setzero_ps();
183 fiy3 = _mm_setzero_ps();
184 fiz3 = _mm_setzero_ps();
186 /* Reset potential sums */
187 velecsum = _mm_setzero_ps();
188 vvdwsum = _mm_setzero_ps();
190 /* Start inner kernel loop */
191 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
194 /* Get j neighbor index, and coordinate index */
199 j_coord_offsetA = DIM*jnrA;
200 j_coord_offsetB = DIM*jnrB;
201 j_coord_offsetC = DIM*jnrC;
202 j_coord_offsetD = DIM*jnrD;
204 /* load j atom coordinates */
205 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
206 x+j_coord_offsetC,x+j_coord_offsetD,
209 /* Calculate displacement vector */
210 dx00 = _mm_sub_ps(ix0,jx0);
211 dy00 = _mm_sub_ps(iy0,jy0);
212 dz00 = _mm_sub_ps(iz0,jz0);
213 dx10 = _mm_sub_ps(ix1,jx0);
214 dy10 = _mm_sub_ps(iy1,jy0);
215 dz10 = _mm_sub_ps(iz1,jz0);
216 dx20 = _mm_sub_ps(ix2,jx0);
217 dy20 = _mm_sub_ps(iy2,jy0);
218 dz20 = _mm_sub_ps(iz2,jz0);
219 dx30 = _mm_sub_ps(ix3,jx0);
220 dy30 = _mm_sub_ps(iy3,jy0);
221 dz30 = _mm_sub_ps(iz3,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);
227 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
229 rinv00 = gmx_mm_invsqrt_ps(rsq00);
230 rinv10 = gmx_mm_invsqrt_ps(rsq10);
231 rinv20 = gmx_mm_invsqrt_ps(rsq20);
232 rinv30 = gmx_mm_invsqrt_ps(rsq30);
234 /* Load parameters for j particles */
235 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
236 charge+jnrC+0,charge+jnrD+0);
237 vdwjidx0A = 2*vdwtype[jnrA+0];
238 vdwjidx0B = 2*vdwtype[jnrB+0];
239 vdwjidx0C = 2*vdwtype[jnrC+0];
240 vdwjidx0D = 2*vdwtype[jnrD+0];
242 fjx0 = _mm_setzero_ps();
243 fjy0 = _mm_setzero_ps();
244 fjz0 = _mm_setzero_ps();
246 /**************************
247 * CALCULATE INTERACTIONS *
248 **************************/
250 r00 = _mm_mul_ps(rsq00,rinv00);
252 /* Compute parameters for interactions between i and j atoms */
253 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
254 vdwparam+vdwioffset0+vdwjidx0B,
255 vdwparam+vdwioffset0+vdwjidx0C,
256 vdwparam+vdwioffset0+vdwjidx0D,
259 /* Calculate table index by multiplying r with table scale and truncate to integer */
260 rt = _mm_mul_ps(r00,vftabscale);
261 vfitab = _mm_cvttps_epi32(rt);
263 vfeps = _mm_frcz_ps(rt);
265 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
267 twovfeps = _mm_add_ps(vfeps,vfeps);
268 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
270 /* CUBIC SPLINE TABLE DISPERSION */
271 vfitab = _mm_add_epi32(vfitab,ifour);
272 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
273 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
274 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
275 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
276 _MM_TRANSPOSE4_PS(Y,F,G,H);
277 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
278 VV = _mm_macc_ps(vfeps,Fp,Y);
279 vvdw6 = _mm_mul_ps(c6_00,VV);
280 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
281 fvdw6 = _mm_mul_ps(c6_00,FF);
283 /* CUBIC SPLINE TABLE REPULSION */
284 vfitab = _mm_add_epi32(vfitab,ifour);
285 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
286 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
287 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
288 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
289 _MM_TRANSPOSE4_PS(Y,F,G,H);
290 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
291 VV = _mm_macc_ps(vfeps,Fp,Y);
292 vvdw12 = _mm_mul_ps(c12_00,VV);
293 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
294 fvdw12 = _mm_mul_ps(c12_00,FF);
295 vvdw = _mm_add_ps(vvdw12,vvdw6);
296 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
298 /* Update potential sum for this i atom from the interaction with this j atom. */
299 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
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 r10 = _mm_mul_ps(rsq10,rinv10);
318 /* Compute parameters for interactions between i and j atoms */
319 qq10 = _mm_mul_ps(iq1,jq0);
321 /* Calculate table index by multiplying r with table scale and truncate to integer */
322 rt = _mm_mul_ps(r10,vftabscale);
323 vfitab = _mm_cvttps_epi32(rt);
325 vfeps = _mm_frcz_ps(rt);
327 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
329 twovfeps = _mm_add_ps(vfeps,vfeps);
330 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
332 /* CUBIC SPLINE TABLE ELECTROSTATICS */
333 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
334 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
335 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
336 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
337 _MM_TRANSPOSE4_PS(Y,F,G,H);
338 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
339 VV = _mm_macc_ps(vfeps,Fp,Y);
340 velec = _mm_mul_ps(qq10,VV);
341 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
342 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
344 /* Update potential sum for this i atom from the interaction with this j atom. */
345 velecsum = _mm_add_ps(velecsum,velec);
349 /* Update vectorial force */
350 fix1 = _mm_macc_ps(dx10,fscal,fix1);
351 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
352 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
354 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
355 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
356 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
358 /**************************
359 * CALCULATE INTERACTIONS *
360 **************************/
362 r20 = _mm_mul_ps(rsq20,rinv20);
364 /* Compute parameters for interactions between i and j atoms */
365 qq20 = _mm_mul_ps(iq2,jq0);
367 /* Calculate table index by multiplying r with table scale and truncate to integer */
368 rt = _mm_mul_ps(r20,vftabscale);
369 vfitab = _mm_cvttps_epi32(rt);
371 vfeps = _mm_frcz_ps(rt);
373 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
375 twovfeps = _mm_add_ps(vfeps,vfeps);
376 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
378 /* CUBIC SPLINE TABLE ELECTROSTATICS */
379 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
380 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
381 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
382 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
383 _MM_TRANSPOSE4_PS(Y,F,G,H);
384 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
385 VV = _mm_macc_ps(vfeps,Fp,Y);
386 velec = _mm_mul_ps(qq20,VV);
387 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
388 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
390 /* Update potential sum for this i atom from the interaction with this j atom. */
391 velecsum = _mm_add_ps(velecsum,velec);
395 /* Update vectorial force */
396 fix2 = _mm_macc_ps(dx20,fscal,fix2);
397 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
398 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
400 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
401 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
402 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
404 /**************************
405 * CALCULATE INTERACTIONS *
406 **************************/
408 r30 = _mm_mul_ps(rsq30,rinv30);
410 /* Compute parameters for interactions between i and j atoms */
411 qq30 = _mm_mul_ps(iq3,jq0);
413 /* Calculate table index by multiplying r with table scale and truncate to integer */
414 rt = _mm_mul_ps(r30,vftabscale);
415 vfitab = _mm_cvttps_epi32(rt);
417 vfeps = _mm_frcz_ps(rt);
419 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
421 twovfeps = _mm_add_ps(vfeps,vfeps);
422 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
424 /* CUBIC SPLINE TABLE ELECTROSTATICS */
425 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
426 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
427 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
428 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
429 _MM_TRANSPOSE4_PS(Y,F,G,H);
430 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
431 VV = _mm_macc_ps(vfeps,Fp,Y);
432 velec = _mm_mul_ps(qq30,VV);
433 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
434 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
436 /* Update potential sum for this i atom from the interaction with this j atom. */
437 velecsum = _mm_add_ps(velecsum,velec);
441 /* Update vectorial force */
442 fix3 = _mm_macc_ps(dx30,fscal,fix3);
443 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
444 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
446 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
447 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
448 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
450 fjptrA = f+j_coord_offsetA;
451 fjptrB = f+j_coord_offsetB;
452 fjptrC = f+j_coord_offsetC;
453 fjptrD = f+j_coord_offsetD;
455 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
457 /* Inner loop uses 197 flops */
463 /* Get j neighbor index, and coordinate index */
464 jnrlistA = jjnr[jidx];
465 jnrlistB = jjnr[jidx+1];
466 jnrlistC = jjnr[jidx+2];
467 jnrlistD = jjnr[jidx+3];
468 /* Sign of each element will be negative for non-real atoms.
469 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
470 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
472 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
473 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
474 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
475 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
476 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
477 j_coord_offsetA = DIM*jnrA;
478 j_coord_offsetB = DIM*jnrB;
479 j_coord_offsetC = DIM*jnrC;
480 j_coord_offsetD = DIM*jnrD;
482 /* load j atom coordinates */
483 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
484 x+j_coord_offsetC,x+j_coord_offsetD,
487 /* Calculate displacement vector */
488 dx00 = _mm_sub_ps(ix0,jx0);
489 dy00 = _mm_sub_ps(iy0,jy0);
490 dz00 = _mm_sub_ps(iz0,jz0);
491 dx10 = _mm_sub_ps(ix1,jx0);
492 dy10 = _mm_sub_ps(iy1,jy0);
493 dz10 = _mm_sub_ps(iz1,jz0);
494 dx20 = _mm_sub_ps(ix2,jx0);
495 dy20 = _mm_sub_ps(iy2,jy0);
496 dz20 = _mm_sub_ps(iz2,jz0);
497 dx30 = _mm_sub_ps(ix3,jx0);
498 dy30 = _mm_sub_ps(iy3,jy0);
499 dz30 = _mm_sub_ps(iz3,jz0);
501 /* Calculate squared distance and things based on it */
502 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
503 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
504 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
505 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
507 rinv00 = gmx_mm_invsqrt_ps(rsq00);
508 rinv10 = gmx_mm_invsqrt_ps(rsq10);
509 rinv20 = gmx_mm_invsqrt_ps(rsq20);
510 rinv30 = gmx_mm_invsqrt_ps(rsq30);
512 /* Load parameters for j particles */
513 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
514 charge+jnrC+0,charge+jnrD+0);
515 vdwjidx0A = 2*vdwtype[jnrA+0];
516 vdwjidx0B = 2*vdwtype[jnrB+0];
517 vdwjidx0C = 2*vdwtype[jnrC+0];
518 vdwjidx0D = 2*vdwtype[jnrD+0];
520 fjx0 = _mm_setzero_ps();
521 fjy0 = _mm_setzero_ps();
522 fjz0 = _mm_setzero_ps();
524 /**************************
525 * CALCULATE INTERACTIONS *
526 **************************/
528 r00 = _mm_mul_ps(rsq00,rinv00);
529 r00 = _mm_andnot_ps(dummy_mask,r00);
531 /* Compute parameters for interactions between i and j atoms */
532 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
533 vdwparam+vdwioffset0+vdwjidx0B,
534 vdwparam+vdwioffset0+vdwjidx0C,
535 vdwparam+vdwioffset0+vdwjidx0D,
538 /* Calculate table index by multiplying r with table scale and truncate to integer */
539 rt = _mm_mul_ps(r00,vftabscale);
540 vfitab = _mm_cvttps_epi32(rt);
542 vfeps = _mm_frcz_ps(rt);
544 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
546 twovfeps = _mm_add_ps(vfeps,vfeps);
547 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
549 /* CUBIC SPLINE TABLE DISPERSION */
550 vfitab = _mm_add_epi32(vfitab,ifour);
551 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
552 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
553 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
554 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
555 _MM_TRANSPOSE4_PS(Y,F,G,H);
556 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
557 VV = _mm_macc_ps(vfeps,Fp,Y);
558 vvdw6 = _mm_mul_ps(c6_00,VV);
559 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
560 fvdw6 = _mm_mul_ps(c6_00,FF);
562 /* CUBIC SPLINE TABLE REPULSION */
563 vfitab = _mm_add_epi32(vfitab,ifour);
564 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
565 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
566 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
567 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
568 _MM_TRANSPOSE4_PS(Y,F,G,H);
569 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
570 VV = _mm_macc_ps(vfeps,Fp,Y);
571 vvdw12 = _mm_mul_ps(c12_00,VV);
572 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
573 fvdw12 = _mm_mul_ps(c12_00,FF);
574 vvdw = _mm_add_ps(vvdw12,vvdw6);
575 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
577 /* Update potential sum for this i atom from the interaction with this j atom. */
578 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
579 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
583 fscal = _mm_andnot_ps(dummy_mask,fscal);
585 /* Update vectorial force */
586 fix0 = _mm_macc_ps(dx00,fscal,fix0);
587 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
588 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
590 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
591 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
592 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
594 /**************************
595 * CALCULATE INTERACTIONS *
596 **************************/
598 r10 = _mm_mul_ps(rsq10,rinv10);
599 r10 = _mm_andnot_ps(dummy_mask,r10);
601 /* Compute parameters for interactions between i and j atoms */
602 qq10 = _mm_mul_ps(iq1,jq0);
604 /* Calculate table index by multiplying r with table scale and truncate to integer */
605 rt = _mm_mul_ps(r10,vftabscale);
606 vfitab = _mm_cvttps_epi32(rt);
608 vfeps = _mm_frcz_ps(rt);
610 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
612 twovfeps = _mm_add_ps(vfeps,vfeps);
613 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
615 /* CUBIC SPLINE TABLE ELECTROSTATICS */
616 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
617 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
618 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
619 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
620 _MM_TRANSPOSE4_PS(Y,F,G,H);
621 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
622 VV = _mm_macc_ps(vfeps,Fp,Y);
623 velec = _mm_mul_ps(qq10,VV);
624 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
625 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
627 /* Update potential sum for this i atom from the interaction with this j atom. */
628 velec = _mm_andnot_ps(dummy_mask,velec);
629 velecsum = _mm_add_ps(velecsum,velec);
633 fscal = _mm_andnot_ps(dummy_mask,fscal);
635 /* Update vectorial force */
636 fix1 = _mm_macc_ps(dx10,fscal,fix1);
637 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
638 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
640 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
641 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
642 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
644 /**************************
645 * CALCULATE INTERACTIONS *
646 **************************/
648 r20 = _mm_mul_ps(rsq20,rinv20);
649 r20 = _mm_andnot_ps(dummy_mask,r20);
651 /* Compute parameters for interactions between i and j atoms */
652 qq20 = _mm_mul_ps(iq2,jq0);
654 /* Calculate table index by multiplying r with table scale and truncate to integer */
655 rt = _mm_mul_ps(r20,vftabscale);
656 vfitab = _mm_cvttps_epi32(rt);
658 vfeps = _mm_frcz_ps(rt);
660 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
662 twovfeps = _mm_add_ps(vfeps,vfeps);
663 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
665 /* CUBIC SPLINE TABLE ELECTROSTATICS */
666 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
667 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
668 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
669 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
670 _MM_TRANSPOSE4_PS(Y,F,G,H);
671 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
672 VV = _mm_macc_ps(vfeps,Fp,Y);
673 velec = _mm_mul_ps(qq20,VV);
674 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
675 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
677 /* Update potential sum for this i atom from the interaction with this j atom. */
678 velec = _mm_andnot_ps(dummy_mask,velec);
679 velecsum = _mm_add_ps(velecsum,velec);
683 fscal = _mm_andnot_ps(dummy_mask,fscal);
685 /* Update vectorial force */
686 fix2 = _mm_macc_ps(dx20,fscal,fix2);
687 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
688 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
690 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
691 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
692 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
694 /**************************
695 * CALCULATE INTERACTIONS *
696 **************************/
698 r30 = _mm_mul_ps(rsq30,rinv30);
699 r30 = _mm_andnot_ps(dummy_mask,r30);
701 /* Compute parameters for interactions between i and j atoms */
702 qq30 = _mm_mul_ps(iq3,jq0);
704 /* Calculate table index by multiplying r with table scale and truncate to integer */
705 rt = _mm_mul_ps(r30,vftabscale);
706 vfitab = _mm_cvttps_epi32(rt);
708 vfeps = _mm_frcz_ps(rt);
710 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
712 twovfeps = _mm_add_ps(vfeps,vfeps);
713 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
715 /* CUBIC SPLINE TABLE ELECTROSTATICS */
716 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
717 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
718 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
719 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
720 _MM_TRANSPOSE4_PS(Y,F,G,H);
721 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
722 VV = _mm_macc_ps(vfeps,Fp,Y);
723 velec = _mm_mul_ps(qq30,VV);
724 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
725 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
727 /* Update potential sum for this i atom from the interaction with this j atom. */
728 velec = _mm_andnot_ps(dummy_mask,velec);
729 velecsum = _mm_add_ps(velecsum,velec);
733 fscal = _mm_andnot_ps(dummy_mask,fscal);
735 /* Update vectorial force */
736 fix3 = _mm_macc_ps(dx30,fscal,fix3);
737 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
738 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
740 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
741 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
742 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
744 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
745 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
746 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
747 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
749 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
751 /* Inner loop uses 201 flops */
754 /* End of innermost loop */
756 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
757 f+i_coord_offset,fshift+i_shift_offset);
760 /* Update potential energies */
761 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
762 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
764 /* Increment number of inner iterations */
765 inneriter += j_index_end - j_index_start;
767 /* Outer loop uses 26 flops */
770 /* Increment number of outer iterations */
773 /* Update outer/inner flops */
775 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*201);
778 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
779 * Electrostatics interaction: CubicSplineTable
780 * VdW interaction: CubicSplineTable
781 * Geometry: Water4-Particle
782 * Calculate force/pot: Force
785 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
786 (t_nblist * gmx_restrict nlist,
787 rvec * gmx_restrict xx,
788 rvec * gmx_restrict ff,
789 t_forcerec * gmx_restrict fr,
790 t_mdatoms * gmx_restrict mdatoms,
791 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
792 t_nrnb * gmx_restrict nrnb)
794 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
795 * just 0 for non-waters.
796 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
797 * jnr indices corresponding to data put in the four positions in the SIMD register.
799 int i_shift_offset,i_coord_offset,outeriter,inneriter;
800 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
801 int jnrA,jnrB,jnrC,jnrD;
802 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
803 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
804 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
806 real *shiftvec,*fshift,*x,*f;
807 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
809 __m128 fscal,rcutoff,rcutoff2,jidxall;
811 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
813 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
815 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
817 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
818 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
819 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
820 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
821 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
822 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
823 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
824 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
827 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
830 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
831 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
833 __m128i ifour = _mm_set1_epi32(4);
834 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
836 __m128 dummy_mask,cutoff_mask;
837 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
838 __m128 one = _mm_set1_ps(1.0);
839 __m128 two = _mm_set1_ps(2.0);
845 jindex = nlist->jindex;
847 shiftidx = nlist->shift;
849 shiftvec = fr->shift_vec[0];
850 fshift = fr->fshift[0];
851 facel = _mm_set1_ps(fr->epsfac);
852 charge = mdatoms->chargeA;
853 nvdwtype = fr->ntype;
855 vdwtype = mdatoms->typeA;
857 vftab = kernel_data->table_elec_vdw->data;
858 vftabscale = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
860 /* Setup water-specific parameters */
861 inr = nlist->iinr[0];
862 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
863 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
864 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
865 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
867 /* Avoid stupid compiler warnings */
868 jnrA = jnrB = jnrC = jnrD = 0;
877 for(iidx=0;iidx<4*DIM;iidx++)
882 /* Start outer loop over neighborlists */
883 for(iidx=0; iidx<nri; iidx++)
885 /* Load shift vector for this list */
886 i_shift_offset = DIM*shiftidx[iidx];
888 /* Load limits for loop over neighbors */
889 j_index_start = jindex[iidx];
890 j_index_end = jindex[iidx+1];
892 /* Get outer coordinate index */
894 i_coord_offset = DIM*inr;
896 /* Load i particle coords and add shift vector */
897 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
898 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
900 fix0 = _mm_setzero_ps();
901 fiy0 = _mm_setzero_ps();
902 fiz0 = _mm_setzero_ps();
903 fix1 = _mm_setzero_ps();
904 fiy1 = _mm_setzero_ps();
905 fiz1 = _mm_setzero_ps();
906 fix2 = _mm_setzero_ps();
907 fiy2 = _mm_setzero_ps();
908 fiz2 = _mm_setzero_ps();
909 fix3 = _mm_setzero_ps();
910 fiy3 = _mm_setzero_ps();
911 fiz3 = _mm_setzero_ps();
913 /* Start inner kernel loop */
914 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
917 /* Get j neighbor index, and coordinate index */
922 j_coord_offsetA = DIM*jnrA;
923 j_coord_offsetB = DIM*jnrB;
924 j_coord_offsetC = DIM*jnrC;
925 j_coord_offsetD = DIM*jnrD;
927 /* load j atom coordinates */
928 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
929 x+j_coord_offsetC,x+j_coord_offsetD,
932 /* Calculate displacement vector */
933 dx00 = _mm_sub_ps(ix0,jx0);
934 dy00 = _mm_sub_ps(iy0,jy0);
935 dz00 = _mm_sub_ps(iz0,jz0);
936 dx10 = _mm_sub_ps(ix1,jx0);
937 dy10 = _mm_sub_ps(iy1,jy0);
938 dz10 = _mm_sub_ps(iz1,jz0);
939 dx20 = _mm_sub_ps(ix2,jx0);
940 dy20 = _mm_sub_ps(iy2,jy0);
941 dz20 = _mm_sub_ps(iz2,jz0);
942 dx30 = _mm_sub_ps(ix3,jx0);
943 dy30 = _mm_sub_ps(iy3,jy0);
944 dz30 = _mm_sub_ps(iz3,jz0);
946 /* Calculate squared distance and things based on it */
947 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
948 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
949 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
950 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
952 rinv00 = gmx_mm_invsqrt_ps(rsq00);
953 rinv10 = gmx_mm_invsqrt_ps(rsq10);
954 rinv20 = gmx_mm_invsqrt_ps(rsq20);
955 rinv30 = gmx_mm_invsqrt_ps(rsq30);
957 /* Load parameters for j particles */
958 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
959 charge+jnrC+0,charge+jnrD+0);
960 vdwjidx0A = 2*vdwtype[jnrA+0];
961 vdwjidx0B = 2*vdwtype[jnrB+0];
962 vdwjidx0C = 2*vdwtype[jnrC+0];
963 vdwjidx0D = 2*vdwtype[jnrD+0];
965 fjx0 = _mm_setzero_ps();
966 fjy0 = _mm_setzero_ps();
967 fjz0 = _mm_setzero_ps();
969 /**************************
970 * CALCULATE INTERACTIONS *
971 **************************/
973 r00 = _mm_mul_ps(rsq00,rinv00);
975 /* Compute parameters for interactions between i and j atoms */
976 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
977 vdwparam+vdwioffset0+vdwjidx0B,
978 vdwparam+vdwioffset0+vdwjidx0C,
979 vdwparam+vdwioffset0+vdwjidx0D,
982 /* Calculate table index by multiplying r with table scale and truncate to integer */
983 rt = _mm_mul_ps(r00,vftabscale);
984 vfitab = _mm_cvttps_epi32(rt);
986 vfeps = _mm_frcz_ps(rt);
988 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
990 twovfeps = _mm_add_ps(vfeps,vfeps);
991 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
993 /* CUBIC SPLINE TABLE DISPERSION */
994 vfitab = _mm_add_epi32(vfitab,ifour);
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)));
1018 /* Update vectorial force */
1019 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1020 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1021 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1023 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1024 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1025 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1027 /**************************
1028 * CALCULATE INTERACTIONS *
1029 **************************/
1031 r10 = _mm_mul_ps(rsq10,rinv10);
1033 /* Compute parameters for interactions between i and j atoms */
1034 qq10 = _mm_mul_ps(iq1,jq0);
1036 /* Calculate table index by multiplying r with table scale and truncate to integer */
1037 rt = _mm_mul_ps(r10,vftabscale);
1038 vfitab = _mm_cvttps_epi32(rt);
1040 vfeps = _mm_frcz_ps(rt);
1042 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1044 twovfeps = _mm_add_ps(vfeps,vfeps);
1045 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1047 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1048 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1049 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1050 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1051 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1052 _MM_TRANSPOSE4_PS(Y,F,G,H);
1053 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1054 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1055 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1059 /* Update vectorial force */
1060 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1061 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1062 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1064 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1065 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1066 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1068 /**************************
1069 * CALCULATE INTERACTIONS *
1070 **************************/
1072 r20 = _mm_mul_ps(rsq20,rinv20);
1074 /* Compute parameters for interactions between i and j atoms */
1075 qq20 = _mm_mul_ps(iq2,jq0);
1077 /* Calculate table index by multiplying r with table scale and truncate to integer */
1078 rt = _mm_mul_ps(r20,vftabscale);
1079 vfitab = _mm_cvttps_epi32(rt);
1081 vfeps = _mm_frcz_ps(rt);
1083 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1085 twovfeps = _mm_add_ps(vfeps,vfeps);
1086 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1088 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1089 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1090 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1091 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1092 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1093 _MM_TRANSPOSE4_PS(Y,F,G,H);
1094 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1095 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1096 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1100 /* Update vectorial force */
1101 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1102 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1103 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1105 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1106 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1107 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1109 /**************************
1110 * CALCULATE INTERACTIONS *
1111 **************************/
1113 r30 = _mm_mul_ps(rsq30,rinv30);
1115 /* Compute parameters for interactions between i and j atoms */
1116 qq30 = _mm_mul_ps(iq3,jq0);
1118 /* Calculate table index by multiplying r with table scale and truncate to integer */
1119 rt = _mm_mul_ps(r30,vftabscale);
1120 vfitab = _mm_cvttps_epi32(rt);
1122 vfeps = _mm_frcz_ps(rt);
1124 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1126 twovfeps = _mm_add_ps(vfeps,vfeps);
1127 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1129 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1130 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1131 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1132 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1133 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1134 _MM_TRANSPOSE4_PS(Y,F,G,H);
1135 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1136 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1137 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1141 /* Update vectorial force */
1142 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1143 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1144 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1146 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1147 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1148 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1150 fjptrA = f+j_coord_offsetA;
1151 fjptrB = f+j_coord_offsetB;
1152 fjptrC = f+j_coord_offsetC;
1153 fjptrD = f+j_coord_offsetD;
1155 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1157 /* Inner loop uses 177 flops */
1160 if(jidx<j_index_end)
1163 /* Get j neighbor index, and coordinate index */
1164 jnrlistA = jjnr[jidx];
1165 jnrlistB = jjnr[jidx+1];
1166 jnrlistC = jjnr[jidx+2];
1167 jnrlistD = jjnr[jidx+3];
1168 /* Sign of each element will be negative for non-real atoms.
1169 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1170 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1172 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1173 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1174 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1175 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1176 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1177 j_coord_offsetA = DIM*jnrA;
1178 j_coord_offsetB = DIM*jnrB;
1179 j_coord_offsetC = DIM*jnrC;
1180 j_coord_offsetD = DIM*jnrD;
1182 /* load j atom coordinates */
1183 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1184 x+j_coord_offsetC,x+j_coord_offsetD,
1187 /* Calculate displacement vector */
1188 dx00 = _mm_sub_ps(ix0,jx0);
1189 dy00 = _mm_sub_ps(iy0,jy0);
1190 dz00 = _mm_sub_ps(iz0,jz0);
1191 dx10 = _mm_sub_ps(ix1,jx0);
1192 dy10 = _mm_sub_ps(iy1,jy0);
1193 dz10 = _mm_sub_ps(iz1,jz0);
1194 dx20 = _mm_sub_ps(ix2,jx0);
1195 dy20 = _mm_sub_ps(iy2,jy0);
1196 dz20 = _mm_sub_ps(iz2,jz0);
1197 dx30 = _mm_sub_ps(ix3,jx0);
1198 dy30 = _mm_sub_ps(iy3,jy0);
1199 dz30 = _mm_sub_ps(iz3,jz0);
1201 /* Calculate squared distance and things based on it */
1202 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1203 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1204 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1205 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1207 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1208 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1209 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1210 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1212 /* Load parameters for j particles */
1213 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1214 charge+jnrC+0,charge+jnrD+0);
1215 vdwjidx0A = 2*vdwtype[jnrA+0];
1216 vdwjidx0B = 2*vdwtype[jnrB+0];
1217 vdwjidx0C = 2*vdwtype[jnrC+0];
1218 vdwjidx0D = 2*vdwtype[jnrD+0];
1220 fjx0 = _mm_setzero_ps();
1221 fjy0 = _mm_setzero_ps();
1222 fjz0 = _mm_setzero_ps();
1224 /**************************
1225 * CALCULATE INTERACTIONS *
1226 **************************/
1228 r00 = _mm_mul_ps(rsq00,rinv00);
1229 r00 = _mm_andnot_ps(dummy_mask,r00);
1231 /* Compute parameters for interactions between i and j atoms */
1232 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1233 vdwparam+vdwioffset0+vdwjidx0B,
1234 vdwparam+vdwioffset0+vdwjidx0C,
1235 vdwparam+vdwioffset0+vdwjidx0D,
1238 /* Calculate table index by multiplying r with table scale and truncate to integer */
1239 rt = _mm_mul_ps(r00,vftabscale);
1240 vfitab = _mm_cvttps_epi32(rt);
1242 vfeps = _mm_frcz_ps(rt);
1244 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1246 twovfeps = _mm_add_ps(vfeps,vfeps);
1247 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1249 /* CUBIC SPLINE TABLE DISPERSION */
1250 vfitab = _mm_add_epi32(vfitab,ifour);
1251 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1252 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1253 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1254 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1255 _MM_TRANSPOSE4_PS(Y,F,G,H);
1256 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1257 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1258 fvdw6 = _mm_mul_ps(c6_00,FF);
1260 /* CUBIC SPLINE TABLE REPULSION */
1261 vfitab = _mm_add_epi32(vfitab,ifour);
1262 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1263 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1264 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1265 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1266 _MM_TRANSPOSE4_PS(Y,F,G,H);
1267 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1268 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1269 fvdw12 = _mm_mul_ps(c12_00,FF);
1270 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1274 fscal = _mm_andnot_ps(dummy_mask,fscal);
1276 /* Update vectorial force */
1277 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1278 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1279 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1281 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1282 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1283 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1285 /**************************
1286 * CALCULATE INTERACTIONS *
1287 **************************/
1289 r10 = _mm_mul_ps(rsq10,rinv10);
1290 r10 = _mm_andnot_ps(dummy_mask,r10);
1292 /* Compute parameters for interactions between i and j atoms */
1293 qq10 = _mm_mul_ps(iq1,jq0);
1295 /* Calculate table index by multiplying r with table scale and truncate to integer */
1296 rt = _mm_mul_ps(r10,vftabscale);
1297 vfitab = _mm_cvttps_epi32(rt);
1299 vfeps = _mm_frcz_ps(rt);
1301 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1303 twovfeps = _mm_add_ps(vfeps,vfeps);
1304 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1306 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1307 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1308 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1309 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1310 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1311 _MM_TRANSPOSE4_PS(Y,F,G,H);
1312 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1313 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1314 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1318 fscal = _mm_andnot_ps(dummy_mask,fscal);
1320 /* Update vectorial force */
1321 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1322 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1323 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1325 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1326 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1327 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1329 /**************************
1330 * CALCULATE INTERACTIONS *
1331 **************************/
1333 r20 = _mm_mul_ps(rsq20,rinv20);
1334 r20 = _mm_andnot_ps(dummy_mask,r20);
1336 /* Compute parameters for interactions between i and j atoms */
1337 qq20 = _mm_mul_ps(iq2,jq0);
1339 /* Calculate table index by multiplying r with table scale and truncate to integer */
1340 rt = _mm_mul_ps(r20,vftabscale);
1341 vfitab = _mm_cvttps_epi32(rt);
1343 vfeps = _mm_frcz_ps(rt);
1345 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1347 twovfeps = _mm_add_ps(vfeps,vfeps);
1348 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1350 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1351 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1352 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1353 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1354 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1355 _MM_TRANSPOSE4_PS(Y,F,G,H);
1356 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1357 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1358 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1362 fscal = _mm_andnot_ps(dummy_mask,fscal);
1364 /* Update vectorial force */
1365 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1366 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1367 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1369 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1370 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1371 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1373 /**************************
1374 * CALCULATE INTERACTIONS *
1375 **************************/
1377 r30 = _mm_mul_ps(rsq30,rinv30);
1378 r30 = _mm_andnot_ps(dummy_mask,r30);
1380 /* Compute parameters for interactions between i and j atoms */
1381 qq30 = _mm_mul_ps(iq3,jq0);
1383 /* Calculate table index by multiplying r with table scale and truncate to integer */
1384 rt = _mm_mul_ps(r30,vftabscale);
1385 vfitab = _mm_cvttps_epi32(rt);
1387 vfeps = _mm_frcz_ps(rt);
1389 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1391 twovfeps = _mm_add_ps(vfeps,vfeps);
1392 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1394 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1395 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1396 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1397 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1398 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1399 _MM_TRANSPOSE4_PS(Y,F,G,H);
1400 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1401 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1402 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1406 fscal = _mm_andnot_ps(dummy_mask,fscal);
1408 /* Update vectorial force */
1409 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1410 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1411 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1413 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1414 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1415 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1417 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1418 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1419 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1420 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1422 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1424 /* Inner loop uses 181 flops */
1427 /* End of innermost loop */
1429 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1430 f+i_coord_offset,fshift+i_shift_offset);
1432 /* Increment number of inner iterations */
1433 inneriter += j_index_end - j_index_start;
1435 /* Outer loop uses 24 flops */
1438 /* Increment number of outer iterations */
1441 /* Update outer/inner flops */
1443 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*181);