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_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
47 #include "kernelutil_x86_avx_128_fma_single.h"
50 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3W3_VF_avx_128_fma_single
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: CubicSplineTable
53 * Geometry: Water3-Water3
54 * Calculate force/pot: PotentialAndForce
57 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_VF_avx_128_fma_single
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,C,D refer to j loop unrolling done with AVX_128, e.g. for the four 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;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
75 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
81 __m128 fscal,rcutoff,rcutoff2,jidxall;
83 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
89 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
91 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
92 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
93 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
94 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
95 __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
96 __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
97 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
98 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
99 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
100 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
101 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
102 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
103 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
106 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
109 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
110 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
112 __m128i ifour = _mm_set1_epi32(4);
113 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
115 __m128 dummy_mask,cutoff_mask;
116 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
117 __m128 one = _mm_set1_ps(1.0);
118 __m128 two = _mm_set1_ps(2.0);
124 jindex = nlist->jindex;
126 shiftidx = nlist->shift;
128 shiftvec = fr->shift_vec[0];
129 fshift = fr->fshift[0];
130 facel = _mm_set1_ps(fr->ic->epsfac);
131 charge = mdatoms->chargeA;
132 krf = _mm_set1_ps(fr->ic->k_rf);
133 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
134 crf = _mm_set1_ps(fr->ic->c_rf);
135 nvdwtype = fr->ntype;
137 vdwtype = mdatoms->typeA;
139 vftab = kernel_data->table_vdw->data;
140 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
142 /* Setup water-specific parameters */
143 inr = nlist->iinr[0];
144 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
145 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
146 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
147 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
149 jq0 = _mm_set1_ps(charge[inr+0]);
150 jq1 = _mm_set1_ps(charge[inr+1]);
151 jq2 = _mm_set1_ps(charge[inr+2]);
152 vdwjidx0A = 2*vdwtype[inr+0];
153 qq00 = _mm_mul_ps(iq0,jq0);
154 c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
155 c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
156 qq01 = _mm_mul_ps(iq0,jq1);
157 qq02 = _mm_mul_ps(iq0,jq2);
158 qq10 = _mm_mul_ps(iq1,jq0);
159 qq11 = _mm_mul_ps(iq1,jq1);
160 qq12 = _mm_mul_ps(iq1,jq2);
161 qq20 = _mm_mul_ps(iq2,jq0);
162 qq21 = _mm_mul_ps(iq2,jq1);
163 qq22 = _mm_mul_ps(iq2,jq2);
165 /* Avoid stupid compiler warnings */
166 jnrA = jnrB = jnrC = jnrD = 0;
175 for(iidx=0;iidx<4*DIM;iidx++)
180 /* Start outer loop over neighborlists */
181 for(iidx=0; iidx<nri; iidx++)
183 /* Load shift vector for this list */
184 i_shift_offset = DIM*shiftidx[iidx];
186 /* Load limits for loop over neighbors */
187 j_index_start = jindex[iidx];
188 j_index_end = jindex[iidx+1];
190 /* Get outer coordinate index */
192 i_coord_offset = DIM*inr;
194 /* Load i particle coords and add shift vector */
195 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
196 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
198 fix0 = _mm_setzero_ps();
199 fiy0 = _mm_setzero_ps();
200 fiz0 = _mm_setzero_ps();
201 fix1 = _mm_setzero_ps();
202 fiy1 = _mm_setzero_ps();
203 fiz1 = _mm_setzero_ps();
204 fix2 = _mm_setzero_ps();
205 fiy2 = _mm_setzero_ps();
206 fiz2 = _mm_setzero_ps();
208 /* Reset potential sums */
209 velecsum = _mm_setzero_ps();
210 vvdwsum = _mm_setzero_ps();
212 /* Start inner kernel loop */
213 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
216 /* Get j neighbor index, and coordinate index */
221 j_coord_offsetA = DIM*jnrA;
222 j_coord_offsetB = DIM*jnrB;
223 j_coord_offsetC = DIM*jnrC;
224 j_coord_offsetD = DIM*jnrD;
226 /* load j atom coordinates */
227 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
228 x+j_coord_offsetC,x+j_coord_offsetD,
229 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
231 /* Calculate displacement vector */
232 dx00 = _mm_sub_ps(ix0,jx0);
233 dy00 = _mm_sub_ps(iy0,jy0);
234 dz00 = _mm_sub_ps(iz0,jz0);
235 dx01 = _mm_sub_ps(ix0,jx1);
236 dy01 = _mm_sub_ps(iy0,jy1);
237 dz01 = _mm_sub_ps(iz0,jz1);
238 dx02 = _mm_sub_ps(ix0,jx2);
239 dy02 = _mm_sub_ps(iy0,jy2);
240 dz02 = _mm_sub_ps(iz0,jz2);
241 dx10 = _mm_sub_ps(ix1,jx0);
242 dy10 = _mm_sub_ps(iy1,jy0);
243 dz10 = _mm_sub_ps(iz1,jz0);
244 dx11 = _mm_sub_ps(ix1,jx1);
245 dy11 = _mm_sub_ps(iy1,jy1);
246 dz11 = _mm_sub_ps(iz1,jz1);
247 dx12 = _mm_sub_ps(ix1,jx2);
248 dy12 = _mm_sub_ps(iy1,jy2);
249 dz12 = _mm_sub_ps(iz1,jz2);
250 dx20 = _mm_sub_ps(ix2,jx0);
251 dy20 = _mm_sub_ps(iy2,jy0);
252 dz20 = _mm_sub_ps(iz2,jz0);
253 dx21 = _mm_sub_ps(ix2,jx1);
254 dy21 = _mm_sub_ps(iy2,jy1);
255 dz21 = _mm_sub_ps(iz2,jz1);
256 dx22 = _mm_sub_ps(ix2,jx2);
257 dy22 = _mm_sub_ps(iy2,jy2);
258 dz22 = _mm_sub_ps(iz2,jz2);
260 /* Calculate squared distance and things based on it */
261 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
262 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
263 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
264 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
265 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
266 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
267 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
268 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
269 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
271 rinv00 = avx128fma_invsqrt_f(rsq00);
272 rinv01 = avx128fma_invsqrt_f(rsq01);
273 rinv02 = avx128fma_invsqrt_f(rsq02);
274 rinv10 = avx128fma_invsqrt_f(rsq10);
275 rinv11 = avx128fma_invsqrt_f(rsq11);
276 rinv12 = avx128fma_invsqrt_f(rsq12);
277 rinv20 = avx128fma_invsqrt_f(rsq20);
278 rinv21 = avx128fma_invsqrt_f(rsq21);
279 rinv22 = avx128fma_invsqrt_f(rsq22);
281 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
282 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
283 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
284 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
285 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
286 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
287 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
288 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
289 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
291 fjx0 = _mm_setzero_ps();
292 fjy0 = _mm_setzero_ps();
293 fjz0 = _mm_setzero_ps();
294 fjx1 = _mm_setzero_ps();
295 fjy1 = _mm_setzero_ps();
296 fjz1 = _mm_setzero_ps();
297 fjx2 = _mm_setzero_ps();
298 fjy2 = _mm_setzero_ps();
299 fjz2 = _mm_setzero_ps();
301 /**************************
302 * CALCULATE INTERACTIONS *
303 **************************/
305 r00 = _mm_mul_ps(rsq00,rinv00);
307 /* Calculate table index by multiplying r with table scale and truncate to integer */
308 rt = _mm_mul_ps(r00,vftabscale);
309 vfitab = _mm_cvttps_epi32(rt);
311 vfeps = _mm_frcz_ps(rt);
313 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
315 twovfeps = _mm_add_ps(vfeps,vfeps);
316 vfitab = _mm_slli_epi32(vfitab,3);
318 /* REACTION-FIELD ELECTROSTATICS */
319 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
320 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
322 /* CUBIC SPLINE TABLE DISPERSION */
323 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
324 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
325 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
326 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
327 _MM_TRANSPOSE4_PS(Y,F,G,H);
328 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
329 VV = _mm_macc_ps(vfeps,Fp,Y);
330 vvdw6 = _mm_mul_ps(c6_00,VV);
331 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
332 fvdw6 = _mm_mul_ps(c6_00,FF);
334 /* CUBIC SPLINE TABLE REPULSION */
335 vfitab = _mm_add_epi32(vfitab,ifour);
336 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
337 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
338 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
339 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
340 _MM_TRANSPOSE4_PS(Y,F,G,H);
341 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
342 VV = _mm_macc_ps(vfeps,Fp,Y);
343 vvdw12 = _mm_mul_ps(c12_00,VV);
344 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
345 fvdw12 = _mm_mul_ps(c12_00,FF);
346 vvdw = _mm_add_ps(vvdw12,vvdw6);
347 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
349 /* Update potential sum for this i atom from the interaction with this j atom. */
350 velecsum = _mm_add_ps(velecsum,velec);
351 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
353 fscal = _mm_add_ps(felec,fvdw);
355 /* Update vectorial force */
356 fix0 = _mm_macc_ps(dx00,fscal,fix0);
357 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
358 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
360 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
361 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
362 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
364 /**************************
365 * CALCULATE INTERACTIONS *
366 **************************/
368 /* REACTION-FIELD ELECTROSTATICS */
369 velec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
370 felec = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
372 /* Update potential sum for this i atom from the interaction with this j atom. */
373 velecsum = _mm_add_ps(velecsum,velec);
377 /* Update vectorial force */
378 fix0 = _mm_macc_ps(dx01,fscal,fix0);
379 fiy0 = _mm_macc_ps(dy01,fscal,fiy0);
380 fiz0 = _mm_macc_ps(dz01,fscal,fiz0);
382 fjx1 = _mm_macc_ps(dx01,fscal,fjx1);
383 fjy1 = _mm_macc_ps(dy01,fscal,fjy1);
384 fjz1 = _mm_macc_ps(dz01,fscal,fjz1);
386 /**************************
387 * CALCULATE INTERACTIONS *
388 **************************/
390 /* REACTION-FIELD ELECTROSTATICS */
391 velec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
392 felec = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
394 /* Update potential sum for this i atom from the interaction with this j atom. */
395 velecsum = _mm_add_ps(velecsum,velec);
399 /* Update vectorial force */
400 fix0 = _mm_macc_ps(dx02,fscal,fix0);
401 fiy0 = _mm_macc_ps(dy02,fscal,fiy0);
402 fiz0 = _mm_macc_ps(dz02,fscal,fiz0);
404 fjx2 = _mm_macc_ps(dx02,fscal,fjx2);
405 fjy2 = _mm_macc_ps(dy02,fscal,fjy2);
406 fjz2 = _mm_macc_ps(dz02,fscal,fjz2);
408 /**************************
409 * CALCULATE INTERACTIONS *
410 **************************/
412 /* REACTION-FIELD ELECTROSTATICS */
413 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
414 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
416 /* Update potential sum for this i atom from the interaction with this j atom. */
417 velecsum = _mm_add_ps(velecsum,velec);
421 /* Update vectorial force */
422 fix1 = _mm_macc_ps(dx10,fscal,fix1);
423 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
424 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
426 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
427 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
428 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
430 /**************************
431 * CALCULATE INTERACTIONS *
432 **************************/
434 /* REACTION-FIELD ELECTROSTATICS */
435 velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
436 felec = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
438 /* Update potential sum for this i atom from the interaction with this j atom. */
439 velecsum = _mm_add_ps(velecsum,velec);
443 /* Update vectorial force */
444 fix1 = _mm_macc_ps(dx11,fscal,fix1);
445 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
446 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
448 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
449 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
450 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
452 /**************************
453 * CALCULATE INTERACTIONS *
454 **************************/
456 /* REACTION-FIELD ELECTROSTATICS */
457 velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
458 felec = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
460 /* Update potential sum for this i atom from the interaction with this j atom. */
461 velecsum = _mm_add_ps(velecsum,velec);
465 /* Update vectorial force */
466 fix1 = _mm_macc_ps(dx12,fscal,fix1);
467 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
468 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
470 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
471 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
472 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
474 /**************************
475 * CALCULATE INTERACTIONS *
476 **************************/
478 /* REACTION-FIELD ELECTROSTATICS */
479 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
480 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
482 /* Update potential sum for this i atom from the interaction with this j atom. */
483 velecsum = _mm_add_ps(velecsum,velec);
487 /* Update vectorial force */
488 fix2 = _mm_macc_ps(dx20,fscal,fix2);
489 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
490 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
492 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
493 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
494 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
496 /**************************
497 * CALCULATE INTERACTIONS *
498 **************************/
500 /* REACTION-FIELD ELECTROSTATICS */
501 velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
502 felec = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
504 /* Update potential sum for this i atom from the interaction with this j atom. */
505 velecsum = _mm_add_ps(velecsum,velec);
509 /* Update vectorial force */
510 fix2 = _mm_macc_ps(dx21,fscal,fix2);
511 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
512 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
514 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
515 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
516 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
518 /**************************
519 * CALCULATE INTERACTIONS *
520 **************************/
522 /* REACTION-FIELD ELECTROSTATICS */
523 velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
524 felec = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
526 /* Update potential sum for this i atom from the interaction with this j atom. */
527 velecsum = _mm_add_ps(velecsum,velec);
531 /* Update vectorial force */
532 fix2 = _mm_macc_ps(dx22,fscal,fix2);
533 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
534 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
536 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
537 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
538 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
540 fjptrA = f+j_coord_offsetA;
541 fjptrB = f+j_coord_offsetB;
542 fjptrC = f+j_coord_offsetC;
543 fjptrD = f+j_coord_offsetD;
545 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
546 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
548 /* Inner loop uses 350 flops */
554 /* Get j neighbor index, and coordinate index */
555 jnrlistA = jjnr[jidx];
556 jnrlistB = jjnr[jidx+1];
557 jnrlistC = jjnr[jidx+2];
558 jnrlistD = jjnr[jidx+3];
559 /* Sign of each element will be negative for non-real atoms.
560 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
561 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
563 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
564 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
565 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
566 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
567 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
568 j_coord_offsetA = DIM*jnrA;
569 j_coord_offsetB = DIM*jnrB;
570 j_coord_offsetC = DIM*jnrC;
571 j_coord_offsetD = DIM*jnrD;
573 /* load j atom coordinates */
574 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
575 x+j_coord_offsetC,x+j_coord_offsetD,
576 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
578 /* Calculate displacement vector */
579 dx00 = _mm_sub_ps(ix0,jx0);
580 dy00 = _mm_sub_ps(iy0,jy0);
581 dz00 = _mm_sub_ps(iz0,jz0);
582 dx01 = _mm_sub_ps(ix0,jx1);
583 dy01 = _mm_sub_ps(iy0,jy1);
584 dz01 = _mm_sub_ps(iz0,jz1);
585 dx02 = _mm_sub_ps(ix0,jx2);
586 dy02 = _mm_sub_ps(iy0,jy2);
587 dz02 = _mm_sub_ps(iz0,jz2);
588 dx10 = _mm_sub_ps(ix1,jx0);
589 dy10 = _mm_sub_ps(iy1,jy0);
590 dz10 = _mm_sub_ps(iz1,jz0);
591 dx11 = _mm_sub_ps(ix1,jx1);
592 dy11 = _mm_sub_ps(iy1,jy1);
593 dz11 = _mm_sub_ps(iz1,jz1);
594 dx12 = _mm_sub_ps(ix1,jx2);
595 dy12 = _mm_sub_ps(iy1,jy2);
596 dz12 = _mm_sub_ps(iz1,jz2);
597 dx20 = _mm_sub_ps(ix2,jx0);
598 dy20 = _mm_sub_ps(iy2,jy0);
599 dz20 = _mm_sub_ps(iz2,jz0);
600 dx21 = _mm_sub_ps(ix2,jx1);
601 dy21 = _mm_sub_ps(iy2,jy1);
602 dz21 = _mm_sub_ps(iz2,jz1);
603 dx22 = _mm_sub_ps(ix2,jx2);
604 dy22 = _mm_sub_ps(iy2,jy2);
605 dz22 = _mm_sub_ps(iz2,jz2);
607 /* Calculate squared distance and things based on it */
608 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
609 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
610 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
611 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
612 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
613 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
614 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
615 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
616 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
618 rinv00 = avx128fma_invsqrt_f(rsq00);
619 rinv01 = avx128fma_invsqrt_f(rsq01);
620 rinv02 = avx128fma_invsqrt_f(rsq02);
621 rinv10 = avx128fma_invsqrt_f(rsq10);
622 rinv11 = avx128fma_invsqrt_f(rsq11);
623 rinv12 = avx128fma_invsqrt_f(rsq12);
624 rinv20 = avx128fma_invsqrt_f(rsq20);
625 rinv21 = avx128fma_invsqrt_f(rsq21);
626 rinv22 = avx128fma_invsqrt_f(rsq22);
628 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
629 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
630 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
631 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
632 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
633 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
634 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
635 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
636 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
638 fjx0 = _mm_setzero_ps();
639 fjy0 = _mm_setzero_ps();
640 fjz0 = _mm_setzero_ps();
641 fjx1 = _mm_setzero_ps();
642 fjy1 = _mm_setzero_ps();
643 fjz1 = _mm_setzero_ps();
644 fjx2 = _mm_setzero_ps();
645 fjy2 = _mm_setzero_ps();
646 fjz2 = _mm_setzero_ps();
648 /**************************
649 * CALCULATE INTERACTIONS *
650 **************************/
652 r00 = _mm_mul_ps(rsq00,rinv00);
653 r00 = _mm_andnot_ps(dummy_mask,r00);
655 /* Calculate table index by multiplying r with table scale and truncate to integer */
656 rt = _mm_mul_ps(r00,vftabscale);
657 vfitab = _mm_cvttps_epi32(rt);
659 vfeps = _mm_frcz_ps(rt);
661 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
663 twovfeps = _mm_add_ps(vfeps,vfeps);
664 vfitab = _mm_slli_epi32(vfitab,3);
666 /* REACTION-FIELD ELECTROSTATICS */
667 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
668 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
670 /* CUBIC SPLINE TABLE DISPERSION */
671 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
672 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
673 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
674 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
675 _MM_TRANSPOSE4_PS(Y,F,G,H);
676 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
677 VV = _mm_macc_ps(vfeps,Fp,Y);
678 vvdw6 = _mm_mul_ps(c6_00,VV);
679 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
680 fvdw6 = _mm_mul_ps(c6_00,FF);
682 /* CUBIC SPLINE TABLE REPULSION */
683 vfitab = _mm_add_epi32(vfitab,ifour);
684 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
685 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
686 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
687 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
688 _MM_TRANSPOSE4_PS(Y,F,G,H);
689 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
690 VV = _mm_macc_ps(vfeps,Fp,Y);
691 vvdw12 = _mm_mul_ps(c12_00,VV);
692 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
693 fvdw12 = _mm_mul_ps(c12_00,FF);
694 vvdw = _mm_add_ps(vvdw12,vvdw6);
695 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
697 /* Update potential sum for this i atom from the interaction with this j atom. */
698 velec = _mm_andnot_ps(dummy_mask,velec);
699 velecsum = _mm_add_ps(velecsum,velec);
700 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
701 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
703 fscal = _mm_add_ps(felec,fvdw);
705 fscal = _mm_andnot_ps(dummy_mask,fscal);
707 /* Update vectorial force */
708 fix0 = _mm_macc_ps(dx00,fscal,fix0);
709 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
710 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
712 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
713 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
714 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
716 /**************************
717 * CALCULATE INTERACTIONS *
718 **************************/
720 /* REACTION-FIELD ELECTROSTATICS */
721 velec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
722 felec = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
724 /* Update potential sum for this i atom from the interaction with this j atom. */
725 velec = _mm_andnot_ps(dummy_mask,velec);
726 velecsum = _mm_add_ps(velecsum,velec);
730 fscal = _mm_andnot_ps(dummy_mask,fscal);
732 /* Update vectorial force */
733 fix0 = _mm_macc_ps(dx01,fscal,fix0);
734 fiy0 = _mm_macc_ps(dy01,fscal,fiy0);
735 fiz0 = _mm_macc_ps(dz01,fscal,fiz0);
737 fjx1 = _mm_macc_ps(dx01,fscal,fjx1);
738 fjy1 = _mm_macc_ps(dy01,fscal,fjy1);
739 fjz1 = _mm_macc_ps(dz01,fscal,fjz1);
741 /**************************
742 * CALCULATE INTERACTIONS *
743 **************************/
745 /* REACTION-FIELD ELECTROSTATICS */
746 velec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
747 felec = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
749 /* Update potential sum for this i atom from the interaction with this j atom. */
750 velec = _mm_andnot_ps(dummy_mask,velec);
751 velecsum = _mm_add_ps(velecsum,velec);
755 fscal = _mm_andnot_ps(dummy_mask,fscal);
757 /* Update vectorial force */
758 fix0 = _mm_macc_ps(dx02,fscal,fix0);
759 fiy0 = _mm_macc_ps(dy02,fscal,fiy0);
760 fiz0 = _mm_macc_ps(dz02,fscal,fiz0);
762 fjx2 = _mm_macc_ps(dx02,fscal,fjx2);
763 fjy2 = _mm_macc_ps(dy02,fscal,fjy2);
764 fjz2 = _mm_macc_ps(dz02,fscal,fjz2);
766 /**************************
767 * CALCULATE INTERACTIONS *
768 **************************/
770 /* REACTION-FIELD ELECTROSTATICS */
771 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
772 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
774 /* Update potential sum for this i atom from the interaction with this j atom. */
775 velec = _mm_andnot_ps(dummy_mask,velec);
776 velecsum = _mm_add_ps(velecsum,velec);
780 fscal = _mm_andnot_ps(dummy_mask,fscal);
782 /* Update vectorial force */
783 fix1 = _mm_macc_ps(dx10,fscal,fix1);
784 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
785 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
787 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
788 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
789 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
791 /**************************
792 * CALCULATE INTERACTIONS *
793 **************************/
795 /* REACTION-FIELD ELECTROSTATICS */
796 velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
797 felec = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
799 /* Update potential sum for this i atom from the interaction with this j atom. */
800 velec = _mm_andnot_ps(dummy_mask,velec);
801 velecsum = _mm_add_ps(velecsum,velec);
805 fscal = _mm_andnot_ps(dummy_mask,fscal);
807 /* Update vectorial force */
808 fix1 = _mm_macc_ps(dx11,fscal,fix1);
809 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
810 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
812 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
813 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
814 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
816 /**************************
817 * CALCULATE INTERACTIONS *
818 **************************/
820 /* REACTION-FIELD ELECTROSTATICS */
821 velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
822 felec = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
824 /* Update potential sum for this i atom from the interaction with this j atom. */
825 velec = _mm_andnot_ps(dummy_mask,velec);
826 velecsum = _mm_add_ps(velecsum,velec);
830 fscal = _mm_andnot_ps(dummy_mask,fscal);
832 /* Update vectorial force */
833 fix1 = _mm_macc_ps(dx12,fscal,fix1);
834 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
835 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
837 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
838 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
839 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
841 /**************************
842 * CALCULATE INTERACTIONS *
843 **************************/
845 /* REACTION-FIELD ELECTROSTATICS */
846 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
847 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
849 /* Update potential sum for this i atom from the interaction with this j atom. */
850 velec = _mm_andnot_ps(dummy_mask,velec);
851 velecsum = _mm_add_ps(velecsum,velec);
855 fscal = _mm_andnot_ps(dummy_mask,fscal);
857 /* Update vectorial force */
858 fix2 = _mm_macc_ps(dx20,fscal,fix2);
859 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
860 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
862 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
863 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
864 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
866 /**************************
867 * CALCULATE INTERACTIONS *
868 **************************/
870 /* REACTION-FIELD ELECTROSTATICS */
871 velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
872 felec = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
874 /* Update potential sum for this i atom from the interaction with this j atom. */
875 velec = _mm_andnot_ps(dummy_mask,velec);
876 velecsum = _mm_add_ps(velecsum,velec);
880 fscal = _mm_andnot_ps(dummy_mask,fscal);
882 /* Update vectorial force */
883 fix2 = _mm_macc_ps(dx21,fscal,fix2);
884 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
885 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
887 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
888 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
889 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
891 /**************************
892 * CALCULATE INTERACTIONS *
893 **************************/
895 /* REACTION-FIELD ELECTROSTATICS */
896 velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
897 felec = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
899 /* Update potential sum for this i atom from the interaction with this j atom. */
900 velec = _mm_andnot_ps(dummy_mask,velec);
901 velecsum = _mm_add_ps(velecsum,velec);
905 fscal = _mm_andnot_ps(dummy_mask,fscal);
907 /* Update vectorial force */
908 fix2 = _mm_macc_ps(dx22,fscal,fix2);
909 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
910 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
912 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
913 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
914 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
916 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
917 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
918 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
919 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
921 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
922 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
924 /* Inner loop uses 351 flops */
927 /* End of innermost loop */
929 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
930 f+i_coord_offset,fshift+i_shift_offset);
933 /* Update potential energies */
934 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
935 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
937 /* Increment number of inner iterations */
938 inneriter += j_index_end - j_index_start;
940 /* Outer loop uses 20 flops */
943 /* Increment number of outer iterations */
946 /* Update outer/inner flops */
948 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*351);
951 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_128_fma_single
952 * Electrostatics interaction: ReactionField
953 * VdW interaction: CubicSplineTable
954 * Geometry: Water3-Water3
955 * Calculate force/pot: Force
958 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_128_fma_single
959 (t_nblist * gmx_restrict nlist,
960 rvec * gmx_restrict xx,
961 rvec * gmx_restrict ff,
962 struct t_forcerec * gmx_restrict fr,
963 t_mdatoms * gmx_restrict mdatoms,
964 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
965 t_nrnb * gmx_restrict nrnb)
967 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
968 * just 0 for non-waters.
969 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
970 * jnr indices corresponding to data put in the four positions in the SIMD register.
972 int i_shift_offset,i_coord_offset,outeriter,inneriter;
973 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
974 int jnrA,jnrB,jnrC,jnrD;
975 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
976 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
977 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
979 real *shiftvec,*fshift,*x,*f;
980 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
982 __m128 fscal,rcutoff,rcutoff2,jidxall;
984 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
986 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
988 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
989 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
990 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
991 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
992 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
993 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
994 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
995 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
996 __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
997 __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
998 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
999 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1000 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1001 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1002 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1003 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1004 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1007 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1010 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
1011 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
1013 __m128i ifour = _mm_set1_epi32(4);
1014 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1016 __m128 dummy_mask,cutoff_mask;
1017 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1018 __m128 one = _mm_set1_ps(1.0);
1019 __m128 two = _mm_set1_ps(2.0);
1025 jindex = nlist->jindex;
1027 shiftidx = nlist->shift;
1029 shiftvec = fr->shift_vec[0];
1030 fshift = fr->fshift[0];
1031 facel = _mm_set1_ps(fr->ic->epsfac);
1032 charge = mdatoms->chargeA;
1033 krf = _mm_set1_ps(fr->ic->k_rf);
1034 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
1035 crf = _mm_set1_ps(fr->ic->c_rf);
1036 nvdwtype = fr->ntype;
1037 vdwparam = fr->nbfp;
1038 vdwtype = mdatoms->typeA;
1040 vftab = kernel_data->table_vdw->data;
1041 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
1043 /* Setup water-specific parameters */
1044 inr = nlist->iinr[0];
1045 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1046 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1047 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1048 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1050 jq0 = _mm_set1_ps(charge[inr+0]);
1051 jq1 = _mm_set1_ps(charge[inr+1]);
1052 jq2 = _mm_set1_ps(charge[inr+2]);
1053 vdwjidx0A = 2*vdwtype[inr+0];
1054 qq00 = _mm_mul_ps(iq0,jq0);
1055 c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1056 c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1057 qq01 = _mm_mul_ps(iq0,jq1);
1058 qq02 = _mm_mul_ps(iq0,jq2);
1059 qq10 = _mm_mul_ps(iq1,jq0);
1060 qq11 = _mm_mul_ps(iq1,jq1);
1061 qq12 = _mm_mul_ps(iq1,jq2);
1062 qq20 = _mm_mul_ps(iq2,jq0);
1063 qq21 = _mm_mul_ps(iq2,jq1);
1064 qq22 = _mm_mul_ps(iq2,jq2);
1066 /* Avoid stupid compiler warnings */
1067 jnrA = jnrB = jnrC = jnrD = 0;
1068 j_coord_offsetA = 0;
1069 j_coord_offsetB = 0;
1070 j_coord_offsetC = 0;
1071 j_coord_offsetD = 0;
1076 for(iidx=0;iidx<4*DIM;iidx++)
1078 scratch[iidx] = 0.0;
1081 /* Start outer loop over neighborlists */
1082 for(iidx=0; iidx<nri; iidx++)
1084 /* Load shift vector for this list */
1085 i_shift_offset = DIM*shiftidx[iidx];
1087 /* Load limits for loop over neighbors */
1088 j_index_start = jindex[iidx];
1089 j_index_end = jindex[iidx+1];
1091 /* Get outer coordinate index */
1093 i_coord_offset = DIM*inr;
1095 /* Load i particle coords and add shift vector */
1096 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1097 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1099 fix0 = _mm_setzero_ps();
1100 fiy0 = _mm_setzero_ps();
1101 fiz0 = _mm_setzero_ps();
1102 fix1 = _mm_setzero_ps();
1103 fiy1 = _mm_setzero_ps();
1104 fiz1 = _mm_setzero_ps();
1105 fix2 = _mm_setzero_ps();
1106 fiy2 = _mm_setzero_ps();
1107 fiz2 = _mm_setzero_ps();
1109 /* Start inner kernel loop */
1110 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1113 /* Get j neighbor index, and coordinate index */
1115 jnrB = jjnr[jidx+1];
1116 jnrC = jjnr[jidx+2];
1117 jnrD = jjnr[jidx+3];
1118 j_coord_offsetA = DIM*jnrA;
1119 j_coord_offsetB = DIM*jnrB;
1120 j_coord_offsetC = DIM*jnrC;
1121 j_coord_offsetD = DIM*jnrD;
1123 /* load j atom coordinates */
1124 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1125 x+j_coord_offsetC,x+j_coord_offsetD,
1126 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1128 /* Calculate displacement vector */
1129 dx00 = _mm_sub_ps(ix0,jx0);
1130 dy00 = _mm_sub_ps(iy0,jy0);
1131 dz00 = _mm_sub_ps(iz0,jz0);
1132 dx01 = _mm_sub_ps(ix0,jx1);
1133 dy01 = _mm_sub_ps(iy0,jy1);
1134 dz01 = _mm_sub_ps(iz0,jz1);
1135 dx02 = _mm_sub_ps(ix0,jx2);
1136 dy02 = _mm_sub_ps(iy0,jy2);
1137 dz02 = _mm_sub_ps(iz0,jz2);
1138 dx10 = _mm_sub_ps(ix1,jx0);
1139 dy10 = _mm_sub_ps(iy1,jy0);
1140 dz10 = _mm_sub_ps(iz1,jz0);
1141 dx11 = _mm_sub_ps(ix1,jx1);
1142 dy11 = _mm_sub_ps(iy1,jy1);
1143 dz11 = _mm_sub_ps(iz1,jz1);
1144 dx12 = _mm_sub_ps(ix1,jx2);
1145 dy12 = _mm_sub_ps(iy1,jy2);
1146 dz12 = _mm_sub_ps(iz1,jz2);
1147 dx20 = _mm_sub_ps(ix2,jx0);
1148 dy20 = _mm_sub_ps(iy2,jy0);
1149 dz20 = _mm_sub_ps(iz2,jz0);
1150 dx21 = _mm_sub_ps(ix2,jx1);
1151 dy21 = _mm_sub_ps(iy2,jy1);
1152 dz21 = _mm_sub_ps(iz2,jz1);
1153 dx22 = _mm_sub_ps(ix2,jx2);
1154 dy22 = _mm_sub_ps(iy2,jy2);
1155 dz22 = _mm_sub_ps(iz2,jz2);
1157 /* Calculate squared distance and things based on it */
1158 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1159 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1160 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1161 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1162 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1163 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1164 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1165 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1166 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1168 rinv00 = avx128fma_invsqrt_f(rsq00);
1169 rinv01 = avx128fma_invsqrt_f(rsq01);
1170 rinv02 = avx128fma_invsqrt_f(rsq02);
1171 rinv10 = avx128fma_invsqrt_f(rsq10);
1172 rinv11 = avx128fma_invsqrt_f(rsq11);
1173 rinv12 = avx128fma_invsqrt_f(rsq12);
1174 rinv20 = avx128fma_invsqrt_f(rsq20);
1175 rinv21 = avx128fma_invsqrt_f(rsq21);
1176 rinv22 = avx128fma_invsqrt_f(rsq22);
1178 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1179 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
1180 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
1181 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1182 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1183 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1184 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1185 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1186 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1188 fjx0 = _mm_setzero_ps();
1189 fjy0 = _mm_setzero_ps();
1190 fjz0 = _mm_setzero_ps();
1191 fjx1 = _mm_setzero_ps();
1192 fjy1 = _mm_setzero_ps();
1193 fjz1 = _mm_setzero_ps();
1194 fjx2 = _mm_setzero_ps();
1195 fjy2 = _mm_setzero_ps();
1196 fjz2 = _mm_setzero_ps();
1198 /**************************
1199 * CALCULATE INTERACTIONS *
1200 **************************/
1202 r00 = _mm_mul_ps(rsq00,rinv00);
1204 /* Calculate table index by multiplying r with table scale and truncate to integer */
1205 rt = _mm_mul_ps(r00,vftabscale);
1206 vfitab = _mm_cvttps_epi32(rt);
1208 vfeps = _mm_frcz_ps(rt);
1210 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1212 twovfeps = _mm_add_ps(vfeps,vfeps);
1213 vfitab = _mm_slli_epi32(vfitab,3);
1215 /* REACTION-FIELD ELECTROSTATICS */
1216 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1218 /* CUBIC SPLINE TABLE DISPERSION */
1219 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1220 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1221 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1222 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1223 _MM_TRANSPOSE4_PS(Y,F,G,H);
1224 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1225 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1226 fvdw6 = _mm_mul_ps(c6_00,FF);
1228 /* CUBIC SPLINE TABLE REPULSION */
1229 vfitab = _mm_add_epi32(vfitab,ifour);
1230 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1231 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1232 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1233 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1234 _MM_TRANSPOSE4_PS(Y,F,G,H);
1235 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1236 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1237 fvdw12 = _mm_mul_ps(c12_00,FF);
1238 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1240 fscal = _mm_add_ps(felec,fvdw);
1242 /* Update vectorial force */
1243 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1244 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1245 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1247 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1248 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1249 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1251 /**************************
1252 * CALCULATE INTERACTIONS *
1253 **************************/
1255 /* REACTION-FIELD ELECTROSTATICS */
1256 felec = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1260 /* Update vectorial force */
1261 fix0 = _mm_macc_ps(dx01,fscal,fix0);
1262 fiy0 = _mm_macc_ps(dy01,fscal,fiy0);
1263 fiz0 = _mm_macc_ps(dz01,fscal,fiz0);
1265 fjx1 = _mm_macc_ps(dx01,fscal,fjx1);
1266 fjy1 = _mm_macc_ps(dy01,fscal,fjy1);
1267 fjz1 = _mm_macc_ps(dz01,fscal,fjz1);
1269 /**************************
1270 * CALCULATE INTERACTIONS *
1271 **************************/
1273 /* REACTION-FIELD ELECTROSTATICS */
1274 felec = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1278 /* Update vectorial force */
1279 fix0 = _mm_macc_ps(dx02,fscal,fix0);
1280 fiy0 = _mm_macc_ps(dy02,fscal,fiy0);
1281 fiz0 = _mm_macc_ps(dz02,fscal,fiz0);
1283 fjx2 = _mm_macc_ps(dx02,fscal,fjx2);
1284 fjy2 = _mm_macc_ps(dy02,fscal,fjy2);
1285 fjz2 = _mm_macc_ps(dz02,fscal,fjz2);
1287 /**************************
1288 * CALCULATE INTERACTIONS *
1289 **************************/
1291 /* REACTION-FIELD ELECTROSTATICS */
1292 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1296 /* Update vectorial force */
1297 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1298 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1299 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1301 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1302 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1303 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1305 /**************************
1306 * CALCULATE INTERACTIONS *
1307 **************************/
1309 /* REACTION-FIELD ELECTROSTATICS */
1310 felec = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1314 /* Update vectorial force */
1315 fix1 = _mm_macc_ps(dx11,fscal,fix1);
1316 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
1317 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
1319 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
1320 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
1321 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
1323 /**************************
1324 * CALCULATE INTERACTIONS *
1325 **************************/
1327 /* REACTION-FIELD ELECTROSTATICS */
1328 felec = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1332 /* Update vectorial force */
1333 fix1 = _mm_macc_ps(dx12,fscal,fix1);
1334 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
1335 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
1337 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
1338 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
1339 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
1341 /**************************
1342 * CALCULATE INTERACTIONS *
1343 **************************/
1345 /* REACTION-FIELD ELECTROSTATICS */
1346 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1350 /* Update vectorial force */
1351 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1352 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1353 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1355 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1356 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1357 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1359 /**************************
1360 * CALCULATE INTERACTIONS *
1361 **************************/
1363 /* REACTION-FIELD ELECTROSTATICS */
1364 felec = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1368 /* Update vectorial force */
1369 fix2 = _mm_macc_ps(dx21,fscal,fix2);
1370 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
1371 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
1373 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
1374 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
1375 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
1377 /**************************
1378 * CALCULATE INTERACTIONS *
1379 **************************/
1381 /* REACTION-FIELD ELECTROSTATICS */
1382 felec = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1386 /* Update vectorial force */
1387 fix2 = _mm_macc_ps(dx22,fscal,fix2);
1388 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
1389 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
1391 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
1392 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
1393 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
1395 fjptrA = f+j_coord_offsetA;
1396 fjptrB = f+j_coord_offsetB;
1397 fjptrC = f+j_coord_offsetC;
1398 fjptrD = f+j_coord_offsetD;
1400 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1401 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1403 /* Inner loop uses 297 flops */
1406 if(jidx<j_index_end)
1409 /* Get j neighbor index, and coordinate index */
1410 jnrlistA = jjnr[jidx];
1411 jnrlistB = jjnr[jidx+1];
1412 jnrlistC = jjnr[jidx+2];
1413 jnrlistD = jjnr[jidx+3];
1414 /* Sign of each element will be negative for non-real atoms.
1415 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1416 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1418 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1419 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1420 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1421 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1422 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1423 j_coord_offsetA = DIM*jnrA;
1424 j_coord_offsetB = DIM*jnrB;
1425 j_coord_offsetC = DIM*jnrC;
1426 j_coord_offsetD = DIM*jnrD;
1428 /* load j atom coordinates */
1429 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1430 x+j_coord_offsetC,x+j_coord_offsetD,
1431 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1433 /* Calculate displacement vector */
1434 dx00 = _mm_sub_ps(ix0,jx0);
1435 dy00 = _mm_sub_ps(iy0,jy0);
1436 dz00 = _mm_sub_ps(iz0,jz0);
1437 dx01 = _mm_sub_ps(ix0,jx1);
1438 dy01 = _mm_sub_ps(iy0,jy1);
1439 dz01 = _mm_sub_ps(iz0,jz1);
1440 dx02 = _mm_sub_ps(ix0,jx2);
1441 dy02 = _mm_sub_ps(iy0,jy2);
1442 dz02 = _mm_sub_ps(iz0,jz2);
1443 dx10 = _mm_sub_ps(ix1,jx0);
1444 dy10 = _mm_sub_ps(iy1,jy0);
1445 dz10 = _mm_sub_ps(iz1,jz0);
1446 dx11 = _mm_sub_ps(ix1,jx1);
1447 dy11 = _mm_sub_ps(iy1,jy1);
1448 dz11 = _mm_sub_ps(iz1,jz1);
1449 dx12 = _mm_sub_ps(ix1,jx2);
1450 dy12 = _mm_sub_ps(iy1,jy2);
1451 dz12 = _mm_sub_ps(iz1,jz2);
1452 dx20 = _mm_sub_ps(ix2,jx0);
1453 dy20 = _mm_sub_ps(iy2,jy0);
1454 dz20 = _mm_sub_ps(iz2,jz0);
1455 dx21 = _mm_sub_ps(ix2,jx1);
1456 dy21 = _mm_sub_ps(iy2,jy1);
1457 dz21 = _mm_sub_ps(iz2,jz1);
1458 dx22 = _mm_sub_ps(ix2,jx2);
1459 dy22 = _mm_sub_ps(iy2,jy2);
1460 dz22 = _mm_sub_ps(iz2,jz2);
1462 /* Calculate squared distance and things based on it */
1463 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1464 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1465 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1466 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1467 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1468 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1469 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1470 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1471 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1473 rinv00 = avx128fma_invsqrt_f(rsq00);
1474 rinv01 = avx128fma_invsqrt_f(rsq01);
1475 rinv02 = avx128fma_invsqrt_f(rsq02);
1476 rinv10 = avx128fma_invsqrt_f(rsq10);
1477 rinv11 = avx128fma_invsqrt_f(rsq11);
1478 rinv12 = avx128fma_invsqrt_f(rsq12);
1479 rinv20 = avx128fma_invsqrt_f(rsq20);
1480 rinv21 = avx128fma_invsqrt_f(rsq21);
1481 rinv22 = avx128fma_invsqrt_f(rsq22);
1483 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1484 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
1485 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
1486 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1487 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1488 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1489 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1490 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1491 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1493 fjx0 = _mm_setzero_ps();
1494 fjy0 = _mm_setzero_ps();
1495 fjz0 = _mm_setzero_ps();
1496 fjx1 = _mm_setzero_ps();
1497 fjy1 = _mm_setzero_ps();
1498 fjz1 = _mm_setzero_ps();
1499 fjx2 = _mm_setzero_ps();
1500 fjy2 = _mm_setzero_ps();
1501 fjz2 = _mm_setzero_ps();
1503 /**************************
1504 * CALCULATE INTERACTIONS *
1505 **************************/
1507 r00 = _mm_mul_ps(rsq00,rinv00);
1508 r00 = _mm_andnot_ps(dummy_mask,r00);
1510 /* Calculate table index by multiplying r with table scale and truncate to integer */
1511 rt = _mm_mul_ps(r00,vftabscale);
1512 vfitab = _mm_cvttps_epi32(rt);
1514 vfeps = _mm_frcz_ps(rt);
1516 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1518 twovfeps = _mm_add_ps(vfeps,vfeps);
1519 vfitab = _mm_slli_epi32(vfitab,3);
1521 /* REACTION-FIELD ELECTROSTATICS */
1522 felec = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1524 /* CUBIC SPLINE TABLE DISPERSION */
1525 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1526 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1527 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1528 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1529 _MM_TRANSPOSE4_PS(Y,F,G,H);
1530 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1531 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1532 fvdw6 = _mm_mul_ps(c6_00,FF);
1534 /* CUBIC SPLINE TABLE REPULSION */
1535 vfitab = _mm_add_epi32(vfitab,ifour);
1536 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1537 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1538 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1539 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1540 _MM_TRANSPOSE4_PS(Y,F,G,H);
1541 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1542 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1543 fvdw12 = _mm_mul_ps(c12_00,FF);
1544 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1546 fscal = _mm_add_ps(felec,fvdw);
1548 fscal = _mm_andnot_ps(dummy_mask,fscal);
1550 /* Update vectorial force */
1551 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1552 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1553 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1555 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1556 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1557 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1559 /**************************
1560 * CALCULATE INTERACTIONS *
1561 **************************/
1563 /* REACTION-FIELD ELECTROSTATICS */
1564 felec = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1568 fscal = _mm_andnot_ps(dummy_mask,fscal);
1570 /* Update vectorial force */
1571 fix0 = _mm_macc_ps(dx01,fscal,fix0);
1572 fiy0 = _mm_macc_ps(dy01,fscal,fiy0);
1573 fiz0 = _mm_macc_ps(dz01,fscal,fiz0);
1575 fjx1 = _mm_macc_ps(dx01,fscal,fjx1);
1576 fjy1 = _mm_macc_ps(dy01,fscal,fjy1);
1577 fjz1 = _mm_macc_ps(dz01,fscal,fjz1);
1579 /**************************
1580 * CALCULATE INTERACTIONS *
1581 **************************/
1583 /* REACTION-FIELD ELECTROSTATICS */
1584 felec = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1588 fscal = _mm_andnot_ps(dummy_mask,fscal);
1590 /* Update vectorial force */
1591 fix0 = _mm_macc_ps(dx02,fscal,fix0);
1592 fiy0 = _mm_macc_ps(dy02,fscal,fiy0);
1593 fiz0 = _mm_macc_ps(dz02,fscal,fiz0);
1595 fjx2 = _mm_macc_ps(dx02,fscal,fjx2);
1596 fjy2 = _mm_macc_ps(dy02,fscal,fjy2);
1597 fjz2 = _mm_macc_ps(dz02,fscal,fjz2);
1599 /**************************
1600 * CALCULATE INTERACTIONS *
1601 **************************/
1603 /* REACTION-FIELD ELECTROSTATICS */
1604 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1608 fscal = _mm_andnot_ps(dummy_mask,fscal);
1610 /* Update vectorial force */
1611 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1612 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1613 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1615 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1616 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1617 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1619 /**************************
1620 * CALCULATE INTERACTIONS *
1621 **************************/
1623 /* REACTION-FIELD ELECTROSTATICS */
1624 felec = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1628 fscal = _mm_andnot_ps(dummy_mask,fscal);
1630 /* Update vectorial force */
1631 fix1 = _mm_macc_ps(dx11,fscal,fix1);
1632 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
1633 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
1635 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
1636 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
1637 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
1639 /**************************
1640 * CALCULATE INTERACTIONS *
1641 **************************/
1643 /* REACTION-FIELD ELECTROSTATICS */
1644 felec = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1648 fscal = _mm_andnot_ps(dummy_mask,fscal);
1650 /* Update vectorial force */
1651 fix1 = _mm_macc_ps(dx12,fscal,fix1);
1652 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
1653 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
1655 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
1656 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
1657 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
1659 /**************************
1660 * CALCULATE INTERACTIONS *
1661 **************************/
1663 /* REACTION-FIELD ELECTROSTATICS */
1664 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1668 fscal = _mm_andnot_ps(dummy_mask,fscal);
1670 /* Update vectorial force */
1671 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1672 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1673 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1675 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1676 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1677 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1679 /**************************
1680 * CALCULATE INTERACTIONS *
1681 **************************/
1683 /* REACTION-FIELD ELECTROSTATICS */
1684 felec = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1688 fscal = _mm_andnot_ps(dummy_mask,fscal);
1690 /* Update vectorial force */
1691 fix2 = _mm_macc_ps(dx21,fscal,fix2);
1692 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
1693 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
1695 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
1696 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
1697 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
1699 /**************************
1700 * CALCULATE INTERACTIONS *
1701 **************************/
1703 /* REACTION-FIELD ELECTROSTATICS */
1704 felec = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1708 fscal = _mm_andnot_ps(dummy_mask,fscal);
1710 /* Update vectorial force */
1711 fix2 = _mm_macc_ps(dx22,fscal,fix2);
1712 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
1713 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
1715 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
1716 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
1717 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
1719 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1720 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1721 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1722 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1724 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1725 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1727 /* Inner loop uses 298 flops */
1730 /* End of innermost loop */
1732 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1733 f+i_coord_offset,fshift+i_shift_offset);
1735 /* Increment number of inner iterations */
1736 inneriter += j_index_end - j_index_start;
1738 /* Outer loop uses 18 flops */
1741 /* Increment number of outer iterations */
1744 /* Update outer/inner flops */
1746 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*298);