2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2017, 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_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_avx_128_fma_double.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW4W4_VF_avx_128_fma_double
51 * Electrostatics interaction: Coulomb
52 * VdW interaction: CubicSplineTable
53 * Geometry: Water4-Water4
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecCoul_VdwCSTab_GeomW4W4_VF_avx_128_fma_double
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int j_coord_offsetA,j_coord_offsetB;
75 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
77 real *shiftvec,*fshift,*x,*f;
78 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
80 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
82 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
84 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
87 int vdwjidx0A,vdwjidx0B;
88 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 int vdwjidx1A,vdwjidx1B;
90 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
91 int vdwjidx2A,vdwjidx2B;
92 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
93 int vdwjidx3A,vdwjidx3B;
94 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
95 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
96 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
97 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
98 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
99 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
100 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
101 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
102 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
103 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
104 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
105 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
108 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
111 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
112 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
114 __m128i ifour = _mm_set1_epi32(4);
115 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
117 __m128d dummy_mask,cutoff_mask;
118 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
119 __m128d one = _mm_set1_pd(1.0);
120 __m128d two = _mm_set1_pd(2.0);
126 jindex = nlist->jindex;
128 shiftidx = nlist->shift;
130 shiftvec = fr->shift_vec[0];
131 fshift = fr->fshift[0];
132 facel = _mm_set1_pd(fr->ic->epsfac);
133 charge = mdatoms->chargeA;
134 nvdwtype = fr->ntype;
136 vdwtype = mdatoms->typeA;
138 vftab = kernel_data->table_vdw->data;
139 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
141 /* Setup water-specific parameters */
142 inr = nlist->iinr[0];
143 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
144 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
145 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
146 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
148 jq1 = _mm_set1_pd(charge[inr+1]);
149 jq2 = _mm_set1_pd(charge[inr+2]);
150 jq3 = _mm_set1_pd(charge[inr+3]);
151 vdwjidx0A = 2*vdwtype[inr+0];
152 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
153 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
154 qq11 = _mm_mul_pd(iq1,jq1);
155 qq12 = _mm_mul_pd(iq1,jq2);
156 qq13 = _mm_mul_pd(iq1,jq3);
157 qq21 = _mm_mul_pd(iq2,jq1);
158 qq22 = _mm_mul_pd(iq2,jq2);
159 qq23 = _mm_mul_pd(iq2,jq3);
160 qq31 = _mm_mul_pd(iq3,jq1);
161 qq32 = _mm_mul_pd(iq3,jq2);
162 qq33 = _mm_mul_pd(iq3,jq3);
164 /* Avoid stupid compiler warnings */
172 /* Start outer loop over neighborlists */
173 for(iidx=0; iidx<nri; iidx++)
175 /* Load shift vector for this list */
176 i_shift_offset = DIM*shiftidx[iidx];
178 /* Load limits for loop over neighbors */
179 j_index_start = jindex[iidx];
180 j_index_end = jindex[iidx+1];
182 /* Get outer coordinate index */
184 i_coord_offset = DIM*inr;
186 /* Load i particle coords and add shift vector */
187 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
188 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
190 fix0 = _mm_setzero_pd();
191 fiy0 = _mm_setzero_pd();
192 fiz0 = _mm_setzero_pd();
193 fix1 = _mm_setzero_pd();
194 fiy1 = _mm_setzero_pd();
195 fiz1 = _mm_setzero_pd();
196 fix2 = _mm_setzero_pd();
197 fiy2 = _mm_setzero_pd();
198 fiz2 = _mm_setzero_pd();
199 fix3 = _mm_setzero_pd();
200 fiy3 = _mm_setzero_pd();
201 fiz3 = _mm_setzero_pd();
203 /* Reset potential sums */
204 velecsum = _mm_setzero_pd();
205 vvdwsum = _mm_setzero_pd();
207 /* Start inner kernel loop */
208 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
211 /* Get j neighbor index, and coordinate index */
214 j_coord_offsetA = DIM*jnrA;
215 j_coord_offsetB = DIM*jnrB;
217 /* load j atom coordinates */
218 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
219 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
220 &jy2,&jz2,&jx3,&jy3,&jz3);
222 /* Calculate displacement vector */
223 dx00 = _mm_sub_pd(ix0,jx0);
224 dy00 = _mm_sub_pd(iy0,jy0);
225 dz00 = _mm_sub_pd(iz0,jz0);
226 dx11 = _mm_sub_pd(ix1,jx1);
227 dy11 = _mm_sub_pd(iy1,jy1);
228 dz11 = _mm_sub_pd(iz1,jz1);
229 dx12 = _mm_sub_pd(ix1,jx2);
230 dy12 = _mm_sub_pd(iy1,jy2);
231 dz12 = _mm_sub_pd(iz1,jz2);
232 dx13 = _mm_sub_pd(ix1,jx3);
233 dy13 = _mm_sub_pd(iy1,jy3);
234 dz13 = _mm_sub_pd(iz1,jz3);
235 dx21 = _mm_sub_pd(ix2,jx1);
236 dy21 = _mm_sub_pd(iy2,jy1);
237 dz21 = _mm_sub_pd(iz2,jz1);
238 dx22 = _mm_sub_pd(ix2,jx2);
239 dy22 = _mm_sub_pd(iy2,jy2);
240 dz22 = _mm_sub_pd(iz2,jz2);
241 dx23 = _mm_sub_pd(ix2,jx3);
242 dy23 = _mm_sub_pd(iy2,jy3);
243 dz23 = _mm_sub_pd(iz2,jz3);
244 dx31 = _mm_sub_pd(ix3,jx1);
245 dy31 = _mm_sub_pd(iy3,jy1);
246 dz31 = _mm_sub_pd(iz3,jz1);
247 dx32 = _mm_sub_pd(ix3,jx2);
248 dy32 = _mm_sub_pd(iy3,jy2);
249 dz32 = _mm_sub_pd(iz3,jz2);
250 dx33 = _mm_sub_pd(ix3,jx3);
251 dy33 = _mm_sub_pd(iy3,jy3);
252 dz33 = _mm_sub_pd(iz3,jz3);
254 /* Calculate squared distance and things based on it */
255 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
256 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
257 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
258 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
259 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
260 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
261 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
262 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
263 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
264 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
266 rinv00 = avx128fma_invsqrt_d(rsq00);
267 rinv11 = avx128fma_invsqrt_d(rsq11);
268 rinv12 = avx128fma_invsqrt_d(rsq12);
269 rinv13 = avx128fma_invsqrt_d(rsq13);
270 rinv21 = avx128fma_invsqrt_d(rsq21);
271 rinv22 = avx128fma_invsqrt_d(rsq22);
272 rinv23 = avx128fma_invsqrt_d(rsq23);
273 rinv31 = avx128fma_invsqrt_d(rsq31);
274 rinv32 = avx128fma_invsqrt_d(rsq32);
275 rinv33 = avx128fma_invsqrt_d(rsq33);
277 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
278 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
279 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
280 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
281 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
282 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
283 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
284 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
285 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
287 fjx0 = _mm_setzero_pd();
288 fjy0 = _mm_setzero_pd();
289 fjz0 = _mm_setzero_pd();
290 fjx1 = _mm_setzero_pd();
291 fjy1 = _mm_setzero_pd();
292 fjz1 = _mm_setzero_pd();
293 fjx2 = _mm_setzero_pd();
294 fjy2 = _mm_setzero_pd();
295 fjz2 = _mm_setzero_pd();
296 fjx3 = _mm_setzero_pd();
297 fjy3 = _mm_setzero_pd();
298 fjz3 = _mm_setzero_pd();
300 /**************************
301 * CALCULATE INTERACTIONS *
302 **************************/
304 r00 = _mm_mul_pd(rsq00,rinv00);
306 /* Calculate table index by multiplying r with table scale and truncate to integer */
307 rt = _mm_mul_pd(r00,vftabscale);
308 vfitab = _mm_cvttpd_epi32(rt);
310 vfeps = _mm_frcz_pd(rt);
312 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
314 twovfeps = _mm_add_pd(vfeps,vfeps);
315 vfitab = _mm_slli_epi32(vfitab,3);
317 /* CUBIC SPLINE TABLE DISPERSION */
318 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
319 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
320 GMX_MM_TRANSPOSE2_PD(Y,F);
321 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
322 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
323 GMX_MM_TRANSPOSE2_PD(G,H);
324 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
325 VV = _mm_macc_pd(vfeps,Fp,Y);
326 vvdw6 = _mm_mul_pd(c6_00,VV);
327 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
328 fvdw6 = _mm_mul_pd(c6_00,FF);
330 /* CUBIC SPLINE TABLE REPULSION */
331 vfitab = _mm_add_epi32(vfitab,ifour);
332 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
333 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
334 GMX_MM_TRANSPOSE2_PD(Y,F);
335 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
336 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
337 GMX_MM_TRANSPOSE2_PD(G,H);
338 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
339 VV = _mm_macc_pd(vfeps,Fp,Y);
340 vvdw12 = _mm_mul_pd(c12_00,VV);
341 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
342 fvdw12 = _mm_mul_pd(c12_00,FF);
343 vvdw = _mm_add_pd(vvdw12,vvdw6);
344 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
346 /* Update potential sum for this i atom from the interaction with this j atom. */
347 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
351 /* Update vectorial force */
352 fix0 = _mm_macc_pd(dx00,fscal,fix0);
353 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
354 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
356 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
357 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
358 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
360 /**************************
361 * CALCULATE INTERACTIONS *
362 **************************/
364 /* COULOMB ELECTROSTATICS */
365 velec = _mm_mul_pd(qq11,rinv11);
366 felec = _mm_mul_pd(velec,rinvsq11);
368 /* Update potential sum for this i atom from the interaction with this j atom. */
369 velecsum = _mm_add_pd(velecsum,velec);
373 /* Update vectorial force */
374 fix1 = _mm_macc_pd(dx11,fscal,fix1);
375 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
376 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
378 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
379 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
380 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
382 /**************************
383 * CALCULATE INTERACTIONS *
384 **************************/
386 /* COULOMB ELECTROSTATICS */
387 velec = _mm_mul_pd(qq12,rinv12);
388 felec = _mm_mul_pd(velec,rinvsq12);
390 /* Update potential sum for this i atom from the interaction with this j atom. */
391 velecsum = _mm_add_pd(velecsum,velec);
395 /* Update vectorial force */
396 fix1 = _mm_macc_pd(dx12,fscal,fix1);
397 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
398 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
400 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
401 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
402 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
404 /**************************
405 * CALCULATE INTERACTIONS *
406 **************************/
408 /* COULOMB ELECTROSTATICS */
409 velec = _mm_mul_pd(qq13,rinv13);
410 felec = _mm_mul_pd(velec,rinvsq13);
412 /* Update potential sum for this i atom from the interaction with this j atom. */
413 velecsum = _mm_add_pd(velecsum,velec);
417 /* Update vectorial force */
418 fix1 = _mm_macc_pd(dx13,fscal,fix1);
419 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
420 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
422 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
423 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
424 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
426 /**************************
427 * CALCULATE INTERACTIONS *
428 **************************/
430 /* COULOMB ELECTROSTATICS */
431 velec = _mm_mul_pd(qq21,rinv21);
432 felec = _mm_mul_pd(velec,rinvsq21);
434 /* Update potential sum for this i atom from the interaction with this j atom. */
435 velecsum = _mm_add_pd(velecsum,velec);
439 /* Update vectorial force */
440 fix2 = _mm_macc_pd(dx21,fscal,fix2);
441 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
442 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
444 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
445 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
446 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
448 /**************************
449 * CALCULATE INTERACTIONS *
450 **************************/
452 /* COULOMB ELECTROSTATICS */
453 velec = _mm_mul_pd(qq22,rinv22);
454 felec = _mm_mul_pd(velec,rinvsq22);
456 /* Update potential sum for this i atom from the interaction with this j atom. */
457 velecsum = _mm_add_pd(velecsum,velec);
461 /* Update vectorial force */
462 fix2 = _mm_macc_pd(dx22,fscal,fix2);
463 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
464 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
466 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
467 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
468 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
470 /**************************
471 * CALCULATE INTERACTIONS *
472 **************************/
474 /* COULOMB ELECTROSTATICS */
475 velec = _mm_mul_pd(qq23,rinv23);
476 felec = _mm_mul_pd(velec,rinvsq23);
478 /* Update potential sum for this i atom from the interaction with this j atom. */
479 velecsum = _mm_add_pd(velecsum,velec);
483 /* Update vectorial force */
484 fix2 = _mm_macc_pd(dx23,fscal,fix2);
485 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
486 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
488 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
489 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
490 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
492 /**************************
493 * CALCULATE INTERACTIONS *
494 **************************/
496 /* COULOMB ELECTROSTATICS */
497 velec = _mm_mul_pd(qq31,rinv31);
498 felec = _mm_mul_pd(velec,rinvsq31);
500 /* Update potential sum for this i atom from the interaction with this j atom. */
501 velecsum = _mm_add_pd(velecsum,velec);
505 /* Update vectorial force */
506 fix3 = _mm_macc_pd(dx31,fscal,fix3);
507 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
508 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
510 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
511 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
512 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
514 /**************************
515 * CALCULATE INTERACTIONS *
516 **************************/
518 /* COULOMB ELECTROSTATICS */
519 velec = _mm_mul_pd(qq32,rinv32);
520 felec = _mm_mul_pd(velec,rinvsq32);
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velecsum = _mm_add_pd(velecsum,velec);
527 /* Update vectorial force */
528 fix3 = _mm_macc_pd(dx32,fscal,fix3);
529 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
530 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
532 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
533 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
534 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
536 /**************************
537 * CALCULATE INTERACTIONS *
538 **************************/
540 /* COULOMB ELECTROSTATICS */
541 velec = _mm_mul_pd(qq33,rinv33);
542 felec = _mm_mul_pd(velec,rinvsq33);
544 /* Update potential sum for this i atom from the interaction with this j atom. */
545 velecsum = _mm_add_pd(velecsum,velec);
549 /* Update vectorial force */
550 fix3 = _mm_macc_pd(dx33,fscal,fix3);
551 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
552 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
554 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
555 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
556 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
558 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
560 /* Inner loop uses 341 flops */
567 j_coord_offsetA = DIM*jnrA;
569 /* load j atom coordinates */
570 gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
571 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
572 &jy2,&jz2,&jx3,&jy3,&jz3);
574 /* Calculate displacement vector */
575 dx00 = _mm_sub_pd(ix0,jx0);
576 dy00 = _mm_sub_pd(iy0,jy0);
577 dz00 = _mm_sub_pd(iz0,jz0);
578 dx11 = _mm_sub_pd(ix1,jx1);
579 dy11 = _mm_sub_pd(iy1,jy1);
580 dz11 = _mm_sub_pd(iz1,jz1);
581 dx12 = _mm_sub_pd(ix1,jx2);
582 dy12 = _mm_sub_pd(iy1,jy2);
583 dz12 = _mm_sub_pd(iz1,jz2);
584 dx13 = _mm_sub_pd(ix1,jx3);
585 dy13 = _mm_sub_pd(iy1,jy3);
586 dz13 = _mm_sub_pd(iz1,jz3);
587 dx21 = _mm_sub_pd(ix2,jx1);
588 dy21 = _mm_sub_pd(iy2,jy1);
589 dz21 = _mm_sub_pd(iz2,jz1);
590 dx22 = _mm_sub_pd(ix2,jx2);
591 dy22 = _mm_sub_pd(iy2,jy2);
592 dz22 = _mm_sub_pd(iz2,jz2);
593 dx23 = _mm_sub_pd(ix2,jx3);
594 dy23 = _mm_sub_pd(iy2,jy3);
595 dz23 = _mm_sub_pd(iz2,jz3);
596 dx31 = _mm_sub_pd(ix3,jx1);
597 dy31 = _mm_sub_pd(iy3,jy1);
598 dz31 = _mm_sub_pd(iz3,jz1);
599 dx32 = _mm_sub_pd(ix3,jx2);
600 dy32 = _mm_sub_pd(iy3,jy2);
601 dz32 = _mm_sub_pd(iz3,jz2);
602 dx33 = _mm_sub_pd(ix3,jx3);
603 dy33 = _mm_sub_pd(iy3,jy3);
604 dz33 = _mm_sub_pd(iz3,jz3);
606 /* Calculate squared distance and things based on it */
607 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
608 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
609 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
610 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
611 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
612 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
613 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
614 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
615 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
616 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
618 rinv00 = avx128fma_invsqrt_d(rsq00);
619 rinv11 = avx128fma_invsqrt_d(rsq11);
620 rinv12 = avx128fma_invsqrt_d(rsq12);
621 rinv13 = avx128fma_invsqrt_d(rsq13);
622 rinv21 = avx128fma_invsqrt_d(rsq21);
623 rinv22 = avx128fma_invsqrt_d(rsq22);
624 rinv23 = avx128fma_invsqrt_d(rsq23);
625 rinv31 = avx128fma_invsqrt_d(rsq31);
626 rinv32 = avx128fma_invsqrt_d(rsq32);
627 rinv33 = avx128fma_invsqrt_d(rsq33);
629 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
630 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
631 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
632 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
633 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
634 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
635 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
636 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
637 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
639 fjx0 = _mm_setzero_pd();
640 fjy0 = _mm_setzero_pd();
641 fjz0 = _mm_setzero_pd();
642 fjx1 = _mm_setzero_pd();
643 fjy1 = _mm_setzero_pd();
644 fjz1 = _mm_setzero_pd();
645 fjx2 = _mm_setzero_pd();
646 fjy2 = _mm_setzero_pd();
647 fjz2 = _mm_setzero_pd();
648 fjx3 = _mm_setzero_pd();
649 fjy3 = _mm_setzero_pd();
650 fjz3 = _mm_setzero_pd();
652 /**************************
653 * CALCULATE INTERACTIONS *
654 **************************/
656 r00 = _mm_mul_pd(rsq00,rinv00);
658 /* Calculate table index by multiplying r with table scale and truncate to integer */
659 rt = _mm_mul_pd(r00,vftabscale);
660 vfitab = _mm_cvttpd_epi32(rt);
662 vfeps = _mm_frcz_pd(rt);
664 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
666 twovfeps = _mm_add_pd(vfeps,vfeps);
667 vfitab = _mm_slli_epi32(vfitab,3);
669 /* CUBIC SPLINE TABLE DISPERSION */
670 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
671 F = _mm_setzero_pd();
672 GMX_MM_TRANSPOSE2_PD(Y,F);
673 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
674 H = _mm_setzero_pd();
675 GMX_MM_TRANSPOSE2_PD(G,H);
676 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
677 VV = _mm_macc_pd(vfeps,Fp,Y);
678 vvdw6 = _mm_mul_pd(c6_00,VV);
679 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
680 fvdw6 = _mm_mul_pd(c6_00,FF);
682 /* CUBIC SPLINE TABLE REPULSION */
683 vfitab = _mm_add_epi32(vfitab,ifour);
684 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
685 F = _mm_setzero_pd();
686 GMX_MM_TRANSPOSE2_PD(Y,F);
687 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
688 H = _mm_setzero_pd();
689 GMX_MM_TRANSPOSE2_PD(G,H);
690 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
691 VV = _mm_macc_pd(vfeps,Fp,Y);
692 vvdw12 = _mm_mul_pd(c12_00,VV);
693 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
694 fvdw12 = _mm_mul_pd(c12_00,FF);
695 vvdw = _mm_add_pd(vvdw12,vvdw6);
696 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
698 /* Update potential sum for this i atom from the interaction with this j atom. */
699 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
700 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
704 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
706 /* Update vectorial force */
707 fix0 = _mm_macc_pd(dx00,fscal,fix0);
708 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
709 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
711 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
712 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
713 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
715 /**************************
716 * CALCULATE INTERACTIONS *
717 **************************/
719 /* COULOMB ELECTROSTATICS */
720 velec = _mm_mul_pd(qq11,rinv11);
721 felec = _mm_mul_pd(velec,rinvsq11);
723 /* Update potential sum for this i atom from the interaction with this j atom. */
724 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
725 velecsum = _mm_add_pd(velecsum,velec);
729 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
731 /* Update vectorial force */
732 fix1 = _mm_macc_pd(dx11,fscal,fix1);
733 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
734 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
736 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
737 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
738 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
740 /**************************
741 * CALCULATE INTERACTIONS *
742 **************************/
744 /* COULOMB ELECTROSTATICS */
745 velec = _mm_mul_pd(qq12,rinv12);
746 felec = _mm_mul_pd(velec,rinvsq12);
748 /* Update potential sum for this i atom from the interaction with this j atom. */
749 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
750 velecsum = _mm_add_pd(velecsum,velec);
754 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
756 /* Update vectorial force */
757 fix1 = _mm_macc_pd(dx12,fscal,fix1);
758 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
759 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
761 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
762 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
763 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
765 /**************************
766 * CALCULATE INTERACTIONS *
767 **************************/
769 /* COULOMB ELECTROSTATICS */
770 velec = _mm_mul_pd(qq13,rinv13);
771 felec = _mm_mul_pd(velec,rinvsq13);
773 /* Update potential sum for this i atom from the interaction with this j atom. */
774 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
775 velecsum = _mm_add_pd(velecsum,velec);
779 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
781 /* Update vectorial force */
782 fix1 = _mm_macc_pd(dx13,fscal,fix1);
783 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
784 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
786 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
787 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
788 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
790 /**************************
791 * CALCULATE INTERACTIONS *
792 **************************/
794 /* COULOMB ELECTROSTATICS */
795 velec = _mm_mul_pd(qq21,rinv21);
796 felec = _mm_mul_pd(velec,rinvsq21);
798 /* Update potential sum for this i atom from the interaction with this j atom. */
799 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
800 velecsum = _mm_add_pd(velecsum,velec);
804 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
806 /* Update vectorial force */
807 fix2 = _mm_macc_pd(dx21,fscal,fix2);
808 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
809 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
811 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
812 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
813 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
815 /**************************
816 * CALCULATE INTERACTIONS *
817 **************************/
819 /* COULOMB ELECTROSTATICS */
820 velec = _mm_mul_pd(qq22,rinv22);
821 felec = _mm_mul_pd(velec,rinvsq22);
823 /* Update potential sum for this i atom from the interaction with this j atom. */
824 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
825 velecsum = _mm_add_pd(velecsum,velec);
829 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
831 /* Update vectorial force */
832 fix2 = _mm_macc_pd(dx22,fscal,fix2);
833 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
834 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
836 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
837 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
838 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
840 /**************************
841 * CALCULATE INTERACTIONS *
842 **************************/
844 /* COULOMB ELECTROSTATICS */
845 velec = _mm_mul_pd(qq23,rinv23);
846 felec = _mm_mul_pd(velec,rinvsq23);
848 /* Update potential sum for this i atom from the interaction with this j atom. */
849 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
850 velecsum = _mm_add_pd(velecsum,velec);
854 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
856 /* Update vectorial force */
857 fix2 = _mm_macc_pd(dx23,fscal,fix2);
858 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
859 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
861 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
862 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
863 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
865 /**************************
866 * CALCULATE INTERACTIONS *
867 **************************/
869 /* COULOMB ELECTROSTATICS */
870 velec = _mm_mul_pd(qq31,rinv31);
871 felec = _mm_mul_pd(velec,rinvsq31);
873 /* Update potential sum for this i atom from the interaction with this j atom. */
874 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
875 velecsum = _mm_add_pd(velecsum,velec);
879 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
881 /* Update vectorial force */
882 fix3 = _mm_macc_pd(dx31,fscal,fix3);
883 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
884 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
886 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
887 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
888 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
890 /**************************
891 * CALCULATE INTERACTIONS *
892 **************************/
894 /* COULOMB ELECTROSTATICS */
895 velec = _mm_mul_pd(qq32,rinv32);
896 felec = _mm_mul_pd(velec,rinvsq32);
898 /* Update potential sum for this i atom from the interaction with this j atom. */
899 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
900 velecsum = _mm_add_pd(velecsum,velec);
904 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
906 /* Update vectorial force */
907 fix3 = _mm_macc_pd(dx32,fscal,fix3);
908 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
909 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
911 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
912 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
913 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
915 /**************************
916 * CALCULATE INTERACTIONS *
917 **************************/
919 /* COULOMB ELECTROSTATICS */
920 velec = _mm_mul_pd(qq33,rinv33);
921 felec = _mm_mul_pd(velec,rinvsq33);
923 /* Update potential sum for this i atom from the interaction with this j atom. */
924 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
925 velecsum = _mm_add_pd(velecsum,velec);
929 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
931 /* Update vectorial force */
932 fix3 = _mm_macc_pd(dx33,fscal,fix3);
933 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
934 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
936 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
937 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
938 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
940 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
942 /* Inner loop uses 341 flops */
945 /* End of innermost loop */
947 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
948 f+i_coord_offset,fshift+i_shift_offset);
951 /* Update potential energies */
952 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
953 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
955 /* Increment number of inner iterations */
956 inneriter += j_index_end - j_index_start;
958 /* Outer loop uses 26 flops */
961 /* Increment number of outer iterations */
964 /* Update outer/inner flops */
966 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*341);
969 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW4W4_F_avx_128_fma_double
970 * Electrostatics interaction: Coulomb
971 * VdW interaction: CubicSplineTable
972 * Geometry: Water4-Water4
973 * Calculate force/pot: Force
976 nb_kernel_ElecCoul_VdwCSTab_GeomW4W4_F_avx_128_fma_double
977 (t_nblist * gmx_restrict nlist,
978 rvec * gmx_restrict xx,
979 rvec * gmx_restrict ff,
980 struct t_forcerec * gmx_restrict fr,
981 t_mdatoms * gmx_restrict mdatoms,
982 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
983 t_nrnb * gmx_restrict nrnb)
985 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
986 * just 0 for non-waters.
987 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
988 * jnr indices corresponding to data put in the four positions in the SIMD register.
990 int i_shift_offset,i_coord_offset,outeriter,inneriter;
991 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
993 int j_coord_offsetA,j_coord_offsetB;
994 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
996 real *shiftvec,*fshift,*x,*f;
997 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
999 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1001 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1003 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1005 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1006 int vdwjidx0A,vdwjidx0B;
1007 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1008 int vdwjidx1A,vdwjidx1B;
1009 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1010 int vdwjidx2A,vdwjidx2B;
1011 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1012 int vdwjidx3A,vdwjidx3B;
1013 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1014 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1015 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1016 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1017 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1018 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1019 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1020 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1021 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1022 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1023 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1024 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1027 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1030 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
1031 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
1033 __m128i ifour = _mm_set1_epi32(4);
1034 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1036 __m128d dummy_mask,cutoff_mask;
1037 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1038 __m128d one = _mm_set1_pd(1.0);
1039 __m128d two = _mm_set1_pd(2.0);
1045 jindex = nlist->jindex;
1047 shiftidx = nlist->shift;
1049 shiftvec = fr->shift_vec[0];
1050 fshift = fr->fshift[0];
1051 facel = _mm_set1_pd(fr->ic->epsfac);
1052 charge = mdatoms->chargeA;
1053 nvdwtype = fr->ntype;
1054 vdwparam = fr->nbfp;
1055 vdwtype = mdatoms->typeA;
1057 vftab = kernel_data->table_vdw->data;
1058 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
1060 /* Setup water-specific parameters */
1061 inr = nlist->iinr[0];
1062 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1063 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1064 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1065 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1067 jq1 = _mm_set1_pd(charge[inr+1]);
1068 jq2 = _mm_set1_pd(charge[inr+2]);
1069 jq3 = _mm_set1_pd(charge[inr+3]);
1070 vdwjidx0A = 2*vdwtype[inr+0];
1071 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1072 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1073 qq11 = _mm_mul_pd(iq1,jq1);
1074 qq12 = _mm_mul_pd(iq1,jq2);
1075 qq13 = _mm_mul_pd(iq1,jq3);
1076 qq21 = _mm_mul_pd(iq2,jq1);
1077 qq22 = _mm_mul_pd(iq2,jq2);
1078 qq23 = _mm_mul_pd(iq2,jq3);
1079 qq31 = _mm_mul_pd(iq3,jq1);
1080 qq32 = _mm_mul_pd(iq3,jq2);
1081 qq33 = _mm_mul_pd(iq3,jq3);
1083 /* Avoid stupid compiler warnings */
1085 j_coord_offsetA = 0;
1086 j_coord_offsetB = 0;
1091 /* Start outer loop over neighborlists */
1092 for(iidx=0; iidx<nri; iidx++)
1094 /* Load shift vector for this list */
1095 i_shift_offset = DIM*shiftidx[iidx];
1097 /* Load limits for loop over neighbors */
1098 j_index_start = jindex[iidx];
1099 j_index_end = jindex[iidx+1];
1101 /* Get outer coordinate index */
1103 i_coord_offset = DIM*inr;
1105 /* Load i particle coords and add shift vector */
1106 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1107 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1109 fix0 = _mm_setzero_pd();
1110 fiy0 = _mm_setzero_pd();
1111 fiz0 = _mm_setzero_pd();
1112 fix1 = _mm_setzero_pd();
1113 fiy1 = _mm_setzero_pd();
1114 fiz1 = _mm_setzero_pd();
1115 fix2 = _mm_setzero_pd();
1116 fiy2 = _mm_setzero_pd();
1117 fiz2 = _mm_setzero_pd();
1118 fix3 = _mm_setzero_pd();
1119 fiy3 = _mm_setzero_pd();
1120 fiz3 = _mm_setzero_pd();
1122 /* Start inner kernel loop */
1123 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1126 /* Get j neighbor index, and coordinate index */
1128 jnrB = jjnr[jidx+1];
1129 j_coord_offsetA = DIM*jnrA;
1130 j_coord_offsetB = DIM*jnrB;
1132 /* load j atom coordinates */
1133 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1134 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1135 &jy2,&jz2,&jx3,&jy3,&jz3);
1137 /* Calculate displacement vector */
1138 dx00 = _mm_sub_pd(ix0,jx0);
1139 dy00 = _mm_sub_pd(iy0,jy0);
1140 dz00 = _mm_sub_pd(iz0,jz0);
1141 dx11 = _mm_sub_pd(ix1,jx1);
1142 dy11 = _mm_sub_pd(iy1,jy1);
1143 dz11 = _mm_sub_pd(iz1,jz1);
1144 dx12 = _mm_sub_pd(ix1,jx2);
1145 dy12 = _mm_sub_pd(iy1,jy2);
1146 dz12 = _mm_sub_pd(iz1,jz2);
1147 dx13 = _mm_sub_pd(ix1,jx3);
1148 dy13 = _mm_sub_pd(iy1,jy3);
1149 dz13 = _mm_sub_pd(iz1,jz3);
1150 dx21 = _mm_sub_pd(ix2,jx1);
1151 dy21 = _mm_sub_pd(iy2,jy1);
1152 dz21 = _mm_sub_pd(iz2,jz1);
1153 dx22 = _mm_sub_pd(ix2,jx2);
1154 dy22 = _mm_sub_pd(iy2,jy2);
1155 dz22 = _mm_sub_pd(iz2,jz2);
1156 dx23 = _mm_sub_pd(ix2,jx3);
1157 dy23 = _mm_sub_pd(iy2,jy3);
1158 dz23 = _mm_sub_pd(iz2,jz3);
1159 dx31 = _mm_sub_pd(ix3,jx1);
1160 dy31 = _mm_sub_pd(iy3,jy1);
1161 dz31 = _mm_sub_pd(iz3,jz1);
1162 dx32 = _mm_sub_pd(ix3,jx2);
1163 dy32 = _mm_sub_pd(iy3,jy2);
1164 dz32 = _mm_sub_pd(iz3,jz2);
1165 dx33 = _mm_sub_pd(ix3,jx3);
1166 dy33 = _mm_sub_pd(iy3,jy3);
1167 dz33 = _mm_sub_pd(iz3,jz3);
1169 /* Calculate squared distance and things based on it */
1170 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1171 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1172 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1173 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1174 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1175 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1176 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1177 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1178 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1179 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1181 rinv00 = avx128fma_invsqrt_d(rsq00);
1182 rinv11 = avx128fma_invsqrt_d(rsq11);
1183 rinv12 = avx128fma_invsqrt_d(rsq12);
1184 rinv13 = avx128fma_invsqrt_d(rsq13);
1185 rinv21 = avx128fma_invsqrt_d(rsq21);
1186 rinv22 = avx128fma_invsqrt_d(rsq22);
1187 rinv23 = avx128fma_invsqrt_d(rsq23);
1188 rinv31 = avx128fma_invsqrt_d(rsq31);
1189 rinv32 = avx128fma_invsqrt_d(rsq32);
1190 rinv33 = avx128fma_invsqrt_d(rsq33);
1192 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1193 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1194 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
1195 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1196 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1197 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
1198 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
1199 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
1200 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
1202 fjx0 = _mm_setzero_pd();
1203 fjy0 = _mm_setzero_pd();
1204 fjz0 = _mm_setzero_pd();
1205 fjx1 = _mm_setzero_pd();
1206 fjy1 = _mm_setzero_pd();
1207 fjz1 = _mm_setzero_pd();
1208 fjx2 = _mm_setzero_pd();
1209 fjy2 = _mm_setzero_pd();
1210 fjz2 = _mm_setzero_pd();
1211 fjx3 = _mm_setzero_pd();
1212 fjy3 = _mm_setzero_pd();
1213 fjz3 = _mm_setzero_pd();
1215 /**************************
1216 * CALCULATE INTERACTIONS *
1217 **************************/
1219 r00 = _mm_mul_pd(rsq00,rinv00);
1221 /* Calculate table index by multiplying r with table scale and truncate to integer */
1222 rt = _mm_mul_pd(r00,vftabscale);
1223 vfitab = _mm_cvttpd_epi32(rt);
1225 vfeps = _mm_frcz_pd(rt);
1227 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1229 twovfeps = _mm_add_pd(vfeps,vfeps);
1230 vfitab = _mm_slli_epi32(vfitab,3);
1232 /* CUBIC SPLINE TABLE DISPERSION */
1233 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1234 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1235 GMX_MM_TRANSPOSE2_PD(Y,F);
1236 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1237 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1238 GMX_MM_TRANSPOSE2_PD(G,H);
1239 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1240 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1241 fvdw6 = _mm_mul_pd(c6_00,FF);
1243 /* CUBIC SPLINE TABLE REPULSION */
1244 vfitab = _mm_add_epi32(vfitab,ifour);
1245 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1246 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1247 GMX_MM_TRANSPOSE2_PD(Y,F);
1248 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1249 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1250 GMX_MM_TRANSPOSE2_PD(G,H);
1251 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1252 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1253 fvdw12 = _mm_mul_pd(c12_00,FF);
1254 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1258 /* Update vectorial force */
1259 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1260 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1261 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1263 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1264 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1265 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1267 /**************************
1268 * CALCULATE INTERACTIONS *
1269 **************************/
1271 /* COULOMB ELECTROSTATICS */
1272 velec = _mm_mul_pd(qq11,rinv11);
1273 felec = _mm_mul_pd(velec,rinvsq11);
1277 /* Update vectorial force */
1278 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1279 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1280 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1282 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1283 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1284 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1286 /**************************
1287 * CALCULATE INTERACTIONS *
1288 **************************/
1290 /* COULOMB ELECTROSTATICS */
1291 velec = _mm_mul_pd(qq12,rinv12);
1292 felec = _mm_mul_pd(velec,rinvsq12);
1296 /* Update vectorial force */
1297 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1298 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1299 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1301 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1302 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1303 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1305 /**************************
1306 * CALCULATE INTERACTIONS *
1307 **************************/
1309 /* COULOMB ELECTROSTATICS */
1310 velec = _mm_mul_pd(qq13,rinv13);
1311 felec = _mm_mul_pd(velec,rinvsq13);
1315 /* Update vectorial force */
1316 fix1 = _mm_macc_pd(dx13,fscal,fix1);
1317 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
1318 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1320 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1321 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1322 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1324 /**************************
1325 * CALCULATE INTERACTIONS *
1326 **************************/
1328 /* COULOMB ELECTROSTATICS */
1329 velec = _mm_mul_pd(qq21,rinv21);
1330 felec = _mm_mul_pd(velec,rinvsq21);
1334 /* Update vectorial force */
1335 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1336 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1337 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1339 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1340 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1341 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1343 /**************************
1344 * CALCULATE INTERACTIONS *
1345 **************************/
1347 /* COULOMB ELECTROSTATICS */
1348 velec = _mm_mul_pd(qq22,rinv22);
1349 felec = _mm_mul_pd(velec,rinvsq22);
1353 /* Update vectorial force */
1354 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1355 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1356 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1358 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1359 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1360 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1362 /**************************
1363 * CALCULATE INTERACTIONS *
1364 **************************/
1366 /* COULOMB ELECTROSTATICS */
1367 velec = _mm_mul_pd(qq23,rinv23);
1368 felec = _mm_mul_pd(velec,rinvsq23);
1372 /* Update vectorial force */
1373 fix2 = _mm_macc_pd(dx23,fscal,fix2);
1374 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
1375 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
1377 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
1378 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
1379 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
1381 /**************************
1382 * CALCULATE INTERACTIONS *
1383 **************************/
1385 /* COULOMB ELECTROSTATICS */
1386 velec = _mm_mul_pd(qq31,rinv31);
1387 felec = _mm_mul_pd(velec,rinvsq31);
1391 /* Update vectorial force */
1392 fix3 = _mm_macc_pd(dx31,fscal,fix3);
1393 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
1394 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
1396 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
1397 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
1398 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
1400 /**************************
1401 * CALCULATE INTERACTIONS *
1402 **************************/
1404 /* COULOMB ELECTROSTATICS */
1405 velec = _mm_mul_pd(qq32,rinv32);
1406 felec = _mm_mul_pd(velec,rinvsq32);
1410 /* Update vectorial force */
1411 fix3 = _mm_macc_pd(dx32,fscal,fix3);
1412 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
1413 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
1415 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
1416 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
1417 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
1419 /**************************
1420 * CALCULATE INTERACTIONS *
1421 **************************/
1423 /* COULOMB ELECTROSTATICS */
1424 velec = _mm_mul_pd(qq33,rinv33);
1425 felec = _mm_mul_pd(velec,rinvsq33);
1429 /* Update vectorial force */
1430 fix3 = _mm_macc_pd(dx33,fscal,fix3);
1431 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
1432 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
1434 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
1435 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
1436 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
1438 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1440 /* Inner loop uses 324 flops */
1443 if(jidx<j_index_end)
1447 j_coord_offsetA = DIM*jnrA;
1449 /* load j atom coordinates */
1450 gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1451 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1452 &jy2,&jz2,&jx3,&jy3,&jz3);
1454 /* Calculate displacement vector */
1455 dx00 = _mm_sub_pd(ix0,jx0);
1456 dy00 = _mm_sub_pd(iy0,jy0);
1457 dz00 = _mm_sub_pd(iz0,jz0);
1458 dx11 = _mm_sub_pd(ix1,jx1);
1459 dy11 = _mm_sub_pd(iy1,jy1);
1460 dz11 = _mm_sub_pd(iz1,jz1);
1461 dx12 = _mm_sub_pd(ix1,jx2);
1462 dy12 = _mm_sub_pd(iy1,jy2);
1463 dz12 = _mm_sub_pd(iz1,jz2);
1464 dx13 = _mm_sub_pd(ix1,jx3);
1465 dy13 = _mm_sub_pd(iy1,jy3);
1466 dz13 = _mm_sub_pd(iz1,jz3);
1467 dx21 = _mm_sub_pd(ix2,jx1);
1468 dy21 = _mm_sub_pd(iy2,jy1);
1469 dz21 = _mm_sub_pd(iz2,jz1);
1470 dx22 = _mm_sub_pd(ix2,jx2);
1471 dy22 = _mm_sub_pd(iy2,jy2);
1472 dz22 = _mm_sub_pd(iz2,jz2);
1473 dx23 = _mm_sub_pd(ix2,jx3);
1474 dy23 = _mm_sub_pd(iy2,jy3);
1475 dz23 = _mm_sub_pd(iz2,jz3);
1476 dx31 = _mm_sub_pd(ix3,jx1);
1477 dy31 = _mm_sub_pd(iy3,jy1);
1478 dz31 = _mm_sub_pd(iz3,jz1);
1479 dx32 = _mm_sub_pd(ix3,jx2);
1480 dy32 = _mm_sub_pd(iy3,jy2);
1481 dz32 = _mm_sub_pd(iz3,jz2);
1482 dx33 = _mm_sub_pd(ix3,jx3);
1483 dy33 = _mm_sub_pd(iy3,jy3);
1484 dz33 = _mm_sub_pd(iz3,jz3);
1486 /* Calculate squared distance and things based on it */
1487 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1488 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1489 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1490 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1491 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1492 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1493 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1494 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1495 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1496 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1498 rinv00 = avx128fma_invsqrt_d(rsq00);
1499 rinv11 = avx128fma_invsqrt_d(rsq11);
1500 rinv12 = avx128fma_invsqrt_d(rsq12);
1501 rinv13 = avx128fma_invsqrt_d(rsq13);
1502 rinv21 = avx128fma_invsqrt_d(rsq21);
1503 rinv22 = avx128fma_invsqrt_d(rsq22);
1504 rinv23 = avx128fma_invsqrt_d(rsq23);
1505 rinv31 = avx128fma_invsqrt_d(rsq31);
1506 rinv32 = avx128fma_invsqrt_d(rsq32);
1507 rinv33 = avx128fma_invsqrt_d(rsq33);
1509 rinvsq11 = _mm_mul_pd(rinv11,rinv11);
1510 rinvsq12 = _mm_mul_pd(rinv12,rinv12);
1511 rinvsq13 = _mm_mul_pd(rinv13,rinv13);
1512 rinvsq21 = _mm_mul_pd(rinv21,rinv21);
1513 rinvsq22 = _mm_mul_pd(rinv22,rinv22);
1514 rinvsq23 = _mm_mul_pd(rinv23,rinv23);
1515 rinvsq31 = _mm_mul_pd(rinv31,rinv31);
1516 rinvsq32 = _mm_mul_pd(rinv32,rinv32);
1517 rinvsq33 = _mm_mul_pd(rinv33,rinv33);
1519 fjx0 = _mm_setzero_pd();
1520 fjy0 = _mm_setzero_pd();
1521 fjz0 = _mm_setzero_pd();
1522 fjx1 = _mm_setzero_pd();
1523 fjy1 = _mm_setzero_pd();
1524 fjz1 = _mm_setzero_pd();
1525 fjx2 = _mm_setzero_pd();
1526 fjy2 = _mm_setzero_pd();
1527 fjz2 = _mm_setzero_pd();
1528 fjx3 = _mm_setzero_pd();
1529 fjy3 = _mm_setzero_pd();
1530 fjz3 = _mm_setzero_pd();
1532 /**************************
1533 * CALCULATE INTERACTIONS *
1534 **************************/
1536 r00 = _mm_mul_pd(rsq00,rinv00);
1538 /* Calculate table index by multiplying r with table scale and truncate to integer */
1539 rt = _mm_mul_pd(r00,vftabscale);
1540 vfitab = _mm_cvttpd_epi32(rt);
1542 vfeps = _mm_frcz_pd(rt);
1544 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1546 twovfeps = _mm_add_pd(vfeps,vfeps);
1547 vfitab = _mm_slli_epi32(vfitab,3);
1549 /* CUBIC SPLINE TABLE DISPERSION */
1550 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1551 F = _mm_setzero_pd();
1552 GMX_MM_TRANSPOSE2_PD(Y,F);
1553 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1554 H = _mm_setzero_pd();
1555 GMX_MM_TRANSPOSE2_PD(G,H);
1556 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1557 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1558 fvdw6 = _mm_mul_pd(c6_00,FF);
1560 /* CUBIC SPLINE TABLE REPULSION */
1561 vfitab = _mm_add_epi32(vfitab,ifour);
1562 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1563 F = _mm_setzero_pd();
1564 GMX_MM_TRANSPOSE2_PD(Y,F);
1565 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1566 H = _mm_setzero_pd();
1567 GMX_MM_TRANSPOSE2_PD(G,H);
1568 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1569 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1570 fvdw12 = _mm_mul_pd(c12_00,FF);
1571 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1575 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1577 /* Update vectorial force */
1578 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1579 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1580 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1582 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1583 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1584 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1586 /**************************
1587 * CALCULATE INTERACTIONS *
1588 **************************/
1590 /* COULOMB ELECTROSTATICS */
1591 velec = _mm_mul_pd(qq11,rinv11);
1592 felec = _mm_mul_pd(velec,rinvsq11);
1596 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1598 /* Update vectorial force */
1599 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1600 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1601 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1603 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1604 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1605 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1607 /**************************
1608 * CALCULATE INTERACTIONS *
1609 **************************/
1611 /* COULOMB ELECTROSTATICS */
1612 velec = _mm_mul_pd(qq12,rinv12);
1613 felec = _mm_mul_pd(velec,rinvsq12);
1617 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1619 /* Update vectorial force */
1620 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1621 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1622 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1624 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1625 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1626 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1628 /**************************
1629 * CALCULATE INTERACTIONS *
1630 **************************/
1632 /* COULOMB ELECTROSTATICS */
1633 velec = _mm_mul_pd(qq13,rinv13);
1634 felec = _mm_mul_pd(velec,rinvsq13);
1638 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1640 /* Update vectorial force */
1641 fix1 = _mm_macc_pd(dx13,fscal,fix1);
1642 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
1643 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1645 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1646 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1647 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1649 /**************************
1650 * CALCULATE INTERACTIONS *
1651 **************************/
1653 /* COULOMB ELECTROSTATICS */
1654 velec = _mm_mul_pd(qq21,rinv21);
1655 felec = _mm_mul_pd(velec,rinvsq21);
1659 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1661 /* Update vectorial force */
1662 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1663 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1664 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1666 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1667 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1668 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1670 /**************************
1671 * CALCULATE INTERACTIONS *
1672 **************************/
1674 /* COULOMB ELECTROSTATICS */
1675 velec = _mm_mul_pd(qq22,rinv22);
1676 felec = _mm_mul_pd(velec,rinvsq22);
1680 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1682 /* Update vectorial force */
1683 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1684 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1685 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1687 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1688 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1689 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1691 /**************************
1692 * CALCULATE INTERACTIONS *
1693 **************************/
1695 /* COULOMB ELECTROSTATICS */
1696 velec = _mm_mul_pd(qq23,rinv23);
1697 felec = _mm_mul_pd(velec,rinvsq23);
1701 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1703 /* Update vectorial force */
1704 fix2 = _mm_macc_pd(dx23,fscal,fix2);
1705 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
1706 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
1708 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
1709 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
1710 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
1712 /**************************
1713 * CALCULATE INTERACTIONS *
1714 **************************/
1716 /* COULOMB ELECTROSTATICS */
1717 velec = _mm_mul_pd(qq31,rinv31);
1718 felec = _mm_mul_pd(velec,rinvsq31);
1722 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1724 /* Update vectorial force */
1725 fix3 = _mm_macc_pd(dx31,fscal,fix3);
1726 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
1727 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
1729 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
1730 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
1731 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
1733 /**************************
1734 * CALCULATE INTERACTIONS *
1735 **************************/
1737 /* COULOMB ELECTROSTATICS */
1738 velec = _mm_mul_pd(qq32,rinv32);
1739 felec = _mm_mul_pd(velec,rinvsq32);
1743 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1745 /* Update vectorial force */
1746 fix3 = _mm_macc_pd(dx32,fscal,fix3);
1747 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
1748 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
1750 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
1751 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
1752 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
1754 /**************************
1755 * CALCULATE INTERACTIONS *
1756 **************************/
1758 /* COULOMB ELECTROSTATICS */
1759 velec = _mm_mul_pd(qq33,rinv33);
1760 felec = _mm_mul_pd(velec,rinvsq33);
1764 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1766 /* Update vectorial force */
1767 fix3 = _mm_macc_pd(dx33,fscal,fix3);
1768 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
1769 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
1771 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
1772 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
1773 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
1775 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1777 /* Inner loop uses 324 flops */
1780 /* End of innermost loop */
1782 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1783 f+i_coord_offset,fshift+i_shift_offset);
1785 /* Increment number of inner iterations */
1786 inneriter += j_index_end - j_index_start;
1788 /* Outer loop uses 24 flops */
1791 /* Increment number of outer iterations */
1794 /* Update outer/inner flops */
1796 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*324);