2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW3W3_VF_avx_128_fma_double
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: None
54 * Geometry: Water3-Water3
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_VF_avx_128_fma_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 int vdwjidx0A,vdwjidx0B;
87 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 int vdwjidx1A,vdwjidx1B;
89 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
90 int vdwjidx2A,vdwjidx2B;
91 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
92 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
94 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
95 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
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 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
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 velec,felec,velecsum,facel,crf,krf,krf2;
104 __m128i ifour = _mm_set1_epi32(4);
105 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
107 __m128d dummy_mask,cutoff_mask;
108 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
109 __m128d one = _mm_set1_pd(1.0);
110 __m128d two = _mm_set1_pd(2.0);
116 jindex = nlist->jindex;
118 shiftidx = nlist->shift;
120 shiftvec = fr->shift_vec[0];
121 fshift = fr->fshift[0];
122 facel = _mm_set1_pd(fr->epsfac);
123 charge = mdatoms->chargeA;
125 vftab = kernel_data->table_elec->data;
126 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
128 /* Setup water-specific parameters */
129 inr = nlist->iinr[0];
130 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
131 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
132 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
134 jq0 = _mm_set1_pd(charge[inr+0]);
135 jq1 = _mm_set1_pd(charge[inr+1]);
136 jq2 = _mm_set1_pd(charge[inr+2]);
137 qq00 = _mm_mul_pd(iq0,jq0);
138 qq01 = _mm_mul_pd(iq0,jq1);
139 qq02 = _mm_mul_pd(iq0,jq2);
140 qq10 = _mm_mul_pd(iq1,jq0);
141 qq11 = _mm_mul_pd(iq1,jq1);
142 qq12 = _mm_mul_pd(iq1,jq2);
143 qq20 = _mm_mul_pd(iq2,jq0);
144 qq21 = _mm_mul_pd(iq2,jq1);
145 qq22 = _mm_mul_pd(iq2,jq2);
147 /* Avoid stupid compiler warnings */
155 /* Start outer loop over neighborlists */
156 for(iidx=0; iidx<nri; iidx++)
158 /* Load shift vector for this list */
159 i_shift_offset = DIM*shiftidx[iidx];
161 /* Load limits for loop over neighbors */
162 j_index_start = jindex[iidx];
163 j_index_end = jindex[iidx+1];
165 /* Get outer coordinate index */
167 i_coord_offset = DIM*inr;
169 /* Load i particle coords and add shift vector */
170 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
171 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
173 fix0 = _mm_setzero_pd();
174 fiy0 = _mm_setzero_pd();
175 fiz0 = _mm_setzero_pd();
176 fix1 = _mm_setzero_pd();
177 fiy1 = _mm_setzero_pd();
178 fiz1 = _mm_setzero_pd();
179 fix2 = _mm_setzero_pd();
180 fiy2 = _mm_setzero_pd();
181 fiz2 = _mm_setzero_pd();
183 /* Reset potential sums */
184 velecsum = _mm_setzero_pd();
186 /* Start inner kernel loop */
187 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
190 /* Get j neighbor index, and coordinate index */
193 j_coord_offsetA = DIM*jnrA;
194 j_coord_offsetB = DIM*jnrB;
196 /* load j atom coordinates */
197 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
198 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
200 /* Calculate displacement vector */
201 dx00 = _mm_sub_pd(ix0,jx0);
202 dy00 = _mm_sub_pd(iy0,jy0);
203 dz00 = _mm_sub_pd(iz0,jz0);
204 dx01 = _mm_sub_pd(ix0,jx1);
205 dy01 = _mm_sub_pd(iy0,jy1);
206 dz01 = _mm_sub_pd(iz0,jz1);
207 dx02 = _mm_sub_pd(ix0,jx2);
208 dy02 = _mm_sub_pd(iy0,jy2);
209 dz02 = _mm_sub_pd(iz0,jz2);
210 dx10 = _mm_sub_pd(ix1,jx0);
211 dy10 = _mm_sub_pd(iy1,jy0);
212 dz10 = _mm_sub_pd(iz1,jz0);
213 dx11 = _mm_sub_pd(ix1,jx1);
214 dy11 = _mm_sub_pd(iy1,jy1);
215 dz11 = _mm_sub_pd(iz1,jz1);
216 dx12 = _mm_sub_pd(ix1,jx2);
217 dy12 = _mm_sub_pd(iy1,jy2);
218 dz12 = _mm_sub_pd(iz1,jz2);
219 dx20 = _mm_sub_pd(ix2,jx0);
220 dy20 = _mm_sub_pd(iy2,jy0);
221 dz20 = _mm_sub_pd(iz2,jz0);
222 dx21 = _mm_sub_pd(ix2,jx1);
223 dy21 = _mm_sub_pd(iy2,jy1);
224 dz21 = _mm_sub_pd(iz2,jz1);
225 dx22 = _mm_sub_pd(ix2,jx2);
226 dy22 = _mm_sub_pd(iy2,jy2);
227 dz22 = _mm_sub_pd(iz2,jz2);
229 /* Calculate squared distance and things based on it */
230 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
231 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
232 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
233 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
234 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
235 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
236 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
237 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
238 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
240 rinv00 = gmx_mm_invsqrt_pd(rsq00);
241 rinv01 = gmx_mm_invsqrt_pd(rsq01);
242 rinv02 = gmx_mm_invsqrt_pd(rsq02);
243 rinv10 = gmx_mm_invsqrt_pd(rsq10);
244 rinv11 = gmx_mm_invsqrt_pd(rsq11);
245 rinv12 = gmx_mm_invsqrt_pd(rsq12);
246 rinv20 = gmx_mm_invsqrt_pd(rsq20);
247 rinv21 = gmx_mm_invsqrt_pd(rsq21);
248 rinv22 = gmx_mm_invsqrt_pd(rsq22);
250 fjx0 = _mm_setzero_pd();
251 fjy0 = _mm_setzero_pd();
252 fjz0 = _mm_setzero_pd();
253 fjx1 = _mm_setzero_pd();
254 fjy1 = _mm_setzero_pd();
255 fjz1 = _mm_setzero_pd();
256 fjx2 = _mm_setzero_pd();
257 fjy2 = _mm_setzero_pd();
258 fjz2 = _mm_setzero_pd();
260 /**************************
261 * CALCULATE INTERACTIONS *
262 **************************/
264 r00 = _mm_mul_pd(rsq00,rinv00);
266 /* Calculate table index by multiplying r with table scale and truncate to integer */
267 rt = _mm_mul_pd(r00,vftabscale);
268 vfitab = _mm_cvttpd_epi32(rt);
270 vfeps = _mm_frcz_pd(rt);
272 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
274 twovfeps = _mm_add_pd(vfeps,vfeps);
275 vfitab = _mm_slli_epi32(vfitab,2);
277 /* CUBIC SPLINE TABLE ELECTROSTATICS */
278 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
279 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
280 GMX_MM_TRANSPOSE2_PD(Y,F);
281 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
282 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
283 GMX_MM_TRANSPOSE2_PD(G,H);
284 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
285 VV = _mm_macc_pd(vfeps,Fp,Y);
286 velec = _mm_mul_pd(qq00,VV);
287 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
288 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
290 /* Update potential sum for this i atom from the interaction with this j atom. */
291 velecsum = _mm_add_pd(velecsum,velec);
295 /* Update vectorial force */
296 fix0 = _mm_macc_pd(dx00,fscal,fix0);
297 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
298 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
300 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
301 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
302 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
304 /**************************
305 * CALCULATE INTERACTIONS *
306 **************************/
308 r01 = _mm_mul_pd(rsq01,rinv01);
310 /* Calculate table index by multiplying r with table scale and truncate to integer */
311 rt = _mm_mul_pd(r01,vftabscale);
312 vfitab = _mm_cvttpd_epi32(rt);
314 vfeps = _mm_frcz_pd(rt);
316 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
318 twovfeps = _mm_add_pd(vfeps,vfeps);
319 vfitab = _mm_slli_epi32(vfitab,2);
321 /* CUBIC SPLINE TABLE ELECTROSTATICS */
322 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
323 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
324 GMX_MM_TRANSPOSE2_PD(Y,F);
325 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
326 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
327 GMX_MM_TRANSPOSE2_PD(G,H);
328 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
329 VV = _mm_macc_pd(vfeps,Fp,Y);
330 velec = _mm_mul_pd(qq01,VV);
331 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
332 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
334 /* Update potential sum for this i atom from the interaction with this j atom. */
335 velecsum = _mm_add_pd(velecsum,velec);
339 /* Update vectorial force */
340 fix0 = _mm_macc_pd(dx01,fscal,fix0);
341 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
342 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
344 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
345 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
346 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
348 /**************************
349 * CALCULATE INTERACTIONS *
350 **************************/
352 r02 = _mm_mul_pd(rsq02,rinv02);
354 /* Calculate table index by multiplying r with table scale and truncate to integer */
355 rt = _mm_mul_pd(r02,vftabscale);
356 vfitab = _mm_cvttpd_epi32(rt);
358 vfeps = _mm_frcz_pd(rt);
360 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
362 twovfeps = _mm_add_pd(vfeps,vfeps);
363 vfitab = _mm_slli_epi32(vfitab,2);
365 /* CUBIC SPLINE TABLE ELECTROSTATICS */
366 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
367 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
368 GMX_MM_TRANSPOSE2_PD(Y,F);
369 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
370 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
371 GMX_MM_TRANSPOSE2_PD(G,H);
372 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
373 VV = _mm_macc_pd(vfeps,Fp,Y);
374 velec = _mm_mul_pd(qq02,VV);
375 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
376 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
378 /* Update potential sum for this i atom from the interaction with this j atom. */
379 velecsum = _mm_add_pd(velecsum,velec);
383 /* Update vectorial force */
384 fix0 = _mm_macc_pd(dx02,fscal,fix0);
385 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
386 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
388 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
389 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
390 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
392 /**************************
393 * CALCULATE INTERACTIONS *
394 **************************/
396 r10 = _mm_mul_pd(rsq10,rinv10);
398 /* Calculate table index by multiplying r with table scale and truncate to integer */
399 rt = _mm_mul_pd(r10,vftabscale);
400 vfitab = _mm_cvttpd_epi32(rt);
402 vfeps = _mm_frcz_pd(rt);
404 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
406 twovfeps = _mm_add_pd(vfeps,vfeps);
407 vfitab = _mm_slli_epi32(vfitab,2);
409 /* CUBIC SPLINE TABLE ELECTROSTATICS */
410 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
411 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
412 GMX_MM_TRANSPOSE2_PD(Y,F);
413 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
414 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
415 GMX_MM_TRANSPOSE2_PD(G,H);
416 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
417 VV = _mm_macc_pd(vfeps,Fp,Y);
418 velec = _mm_mul_pd(qq10,VV);
419 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
420 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
422 /* Update potential sum for this i atom from the interaction with this j atom. */
423 velecsum = _mm_add_pd(velecsum,velec);
427 /* Update vectorial force */
428 fix1 = _mm_macc_pd(dx10,fscal,fix1);
429 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
430 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
432 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
433 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
434 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
436 /**************************
437 * CALCULATE INTERACTIONS *
438 **************************/
440 r11 = _mm_mul_pd(rsq11,rinv11);
442 /* Calculate table index by multiplying r with table scale and truncate to integer */
443 rt = _mm_mul_pd(r11,vftabscale);
444 vfitab = _mm_cvttpd_epi32(rt);
446 vfeps = _mm_frcz_pd(rt);
448 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
450 twovfeps = _mm_add_pd(vfeps,vfeps);
451 vfitab = _mm_slli_epi32(vfitab,2);
453 /* CUBIC SPLINE TABLE ELECTROSTATICS */
454 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
455 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
456 GMX_MM_TRANSPOSE2_PD(Y,F);
457 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
458 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
459 GMX_MM_TRANSPOSE2_PD(G,H);
460 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
461 VV = _mm_macc_pd(vfeps,Fp,Y);
462 velec = _mm_mul_pd(qq11,VV);
463 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
464 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
466 /* Update potential sum for this i atom from the interaction with this j atom. */
467 velecsum = _mm_add_pd(velecsum,velec);
471 /* Update vectorial force */
472 fix1 = _mm_macc_pd(dx11,fscal,fix1);
473 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
474 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
476 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
477 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
478 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
480 /**************************
481 * CALCULATE INTERACTIONS *
482 **************************/
484 r12 = _mm_mul_pd(rsq12,rinv12);
486 /* Calculate table index by multiplying r with table scale and truncate to integer */
487 rt = _mm_mul_pd(r12,vftabscale);
488 vfitab = _mm_cvttpd_epi32(rt);
490 vfeps = _mm_frcz_pd(rt);
492 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
494 twovfeps = _mm_add_pd(vfeps,vfeps);
495 vfitab = _mm_slli_epi32(vfitab,2);
497 /* CUBIC SPLINE TABLE ELECTROSTATICS */
498 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
499 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
500 GMX_MM_TRANSPOSE2_PD(Y,F);
501 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
502 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
503 GMX_MM_TRANSPOSE2_PD(G,H);
504 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
505 VV = _mm_macc_pd(vfeps,Fp,Y);
506 velec = _mm_mul_pd(qq12,VV);
507 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
508 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
510 /* Update potential sum for this i atom from the interaction with this j atom. */
511 velecsum = _mm_add_pd(velecsum,velec);
515 /* Update vectorial force */
516 fix1 = _mm_macc_pd(dx12,fscal,fix1);
517 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
518 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
520 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
521 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
522 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
524 /**************************
525 * CALCULATE INTERACTIONS *
526 **************************/
528 r20 = _mm_mul_pd(rsq20,rinv20);
530 /* Calculate table index by multiplying r with table scale and truncate to integer */
531 rt = _mm_mul_pd(r20,vftabscale);
532 vfitab = _mm_cvttpd_epi32(rt);
534 vfeps = _mm_frcz_pd(rt);
536 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
538 twovfeps = _mm_add_pd(vfeps,vfeps);
539 vfitab = _mm_slli_epi32(vfitab,2);
541 /* CUBIC SPLINE TABLE ELECTROSTATICS */
542 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
543 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
544 GMX_MM_TRANSPOSE2_PD(Y,F);
545 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
546 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
547 GMX_MM_TRANSPOSE2_PD(G,H);
548 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
549 VV = _mm_macc_pd(vfeps,Fp,Y);
550 velec = _mm_mul_pd(qq20,VV);
551 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
552 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
554 /* Update potential sum for this i atom from the interaction with this j atom. */
555 velecsum = _mm_add_pd(velecsum,velec);
559 /* Update vectorial force */
560 fix2 = _mm_macc_pd(dx20,fscal,fix2);
561 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
562 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
564 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
565 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
566 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
568 /**************************
569 * CALCULATE INTERACTIONS *
570 **************************/
572 r21 = _mm_mul_pd(rsq21,rinv21);
574 /* Calculate table index by multiplying r with table scale and truncate to integer */
575 rt = _mm_mul_pd(r21,vftabscale);
576 vfitab = _mm_cvttpd_epi32(rt);
578 vfeps = _mm_frcz_pd(rt);
580 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
582 twovfeps = _mm_add_pd(vfeps,vfeps);
583 vfitab = _mm_slli_epi32(vfitab,2);
585 /* CUBIC SPLINE TABLE ELECTROSTATICS */
586 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
587 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
588 GMX_MM_TRANSPOSE2_PD(Y,F);
589 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
590 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
591 GMX_MM_TRANSPOSE2_PD(G,H);
592 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
593 VV = _mm_macc_pd(vfeps,Fp,Y);
594 velec = _mm_mul_pd(qq21,VV);
595 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
596 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
598 /* Update potential sum for this i atom from the interaction with this j atom. */
599 velecsum = _mm_add_pd(velecsum,velec);
603 /* Update vectorial force */
604 fix2 = _mm_macc_pd(dx21,fscal,fix2);
605 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
606 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
608 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
609 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
610 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
612 /**************************
613 * CALCULATE INTERACTIONS *
614 **************************/
616 r22 = _mm_mul_pd(rsq22,rinv22);
618 /* Calculate table index by multiplying r with table scale and truncate to integer */
619 rt = _mm_mul_pd(r22,vftabscale);
620 vfitab = _mm_cvttpd_epi32(rt);
622 vfeps = _mm_frcz_pd(rt);
624 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
626 twovfeps = _mm_add_pd(vfeps,vfeps);
627 vfitab = _mm_slli_epi32(vfitab,2);
629 /* CUBIC SPLINE TABLE ELECTROSTATICS */
630 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
631 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
632 GMX_MM_TRANSPOSE2_PD(Y,F);
633 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
634 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
635 GMX_MM_TRANSPOSE2_PD(G,H);
636 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
637 VV = _mm_macc_pd(vfeps,Fp,Y);
638 velec = _mm_mul_pd(qq22,VV);
639 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
640 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
642 /* Update potential sum for this i atom from the interaction with this j atom. */
643 velecsum = _mm_add_pd(velecsum,velec);
647 /* Update vectorial force */
648 fix2 = _mm_macc_pd(dx22,fscal,fix2);
649 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
650 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
652 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
653 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
654 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
656 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
658 /* Inner loop uses 414 flops */
665 j_coord_offsetA = DIM*jnrA;
667 /* load j atom coordinates */
668 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
669 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
671 /* Calculate displacement vector */
672 dx00 = _mm_sub_pd(ix0,jx0);
673 dy00 = _mm_sub_pd(iy0,jy0);
674 dz00 = _mm_sub_pd(iz0,jz0);
675 dx01 = _mm_sub_pd(ix0,jx1);
676 dy01 = _mm_sub_pd(iy0,jy1);
677 dz01 = _mm_sub_pd(iz0,jz1);
678 dx02 = _mm_sub_pd(ix0,jx2);
679 dy02 = _mm_sub_pd(iy0,jy2);
680 dz02 = _mm_sub_pd(iz0,jz2);
681 dx10 = _mm_sub_pd(ix1,jx0);
682 dy10 = _mm_sub_pd(iy1,jy0);
683 dz10 = _mm_sub_pd(iz1,jz0);
684 dx11 = _mm_sub_pd(ix1,jx1);
685 dy11 = _mm_sub_pd(iy1,jy1);
686 dz11 = _mm_sub_pd(iz1,jz1);
687 dx12 = _mm_sub_pd(ix1,jx2);
688 dy12 = _mm_sub_pd(iy1,jy2);
689 dz12 = _mm_sub_pd(iz1,jz2);
690 dx20 = _mm_sub_pd(ix2,jx0);
691 dy20 = _mm_sub_pd(iy2,jy0);
692 dz20 = _mm_sub_pd(iz2,jz0);
693 dx21 = _mm_sub_pd(ix2,jx1);
694 dy21 = _mm_sub_pd(iy2,jy1);
695 dz21 = _mm_sub_pd(iz2,jz1);
696 dx22 = _mm_sub_pd(ix2,jx2);
697 dy22 = _mm_sub_pd(iy2,jy2);
698 dz22 = _mm_sub_pd(iz2,jz2);
700 /* Calculate squared distance and things based on it */
701 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
702 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
703 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
704 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
705 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
706 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
707 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
708 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
709 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
711 rinv00 = gmx_mm_invsqrt_pd(rsq00);
712 rinv01 = gmx_mm_invsqrt_pd(rsq01);
713 rinv02 = gmx_mm_invsqrt_pd(rsq02);
714 rinv10 = gmx_mm_invsqrt_pd(rsq10);
715 rinv11 = gmx_mm_invsqrt_pd(rsq11);
716 rinv12 = gmx_mm_invsqrt_pd(rsq12);
717 rinv20 = gmx_mm_invsqrt_pd(rsq20);
718 rinv21 = gmx_mm_invsqrt_pd(rsq21);
719 rinv22 = gmx_mm_invsqrt_pd(rsq22);
721 fjx0 = _mm_setzero_pd();
722 fjy0 = _mm_setzero_pd();
723 fjz0 = _mm_setzero_pd();
724 fjx1 = _mm_setzero_pd();
725 fjy1 = _mm_setzero_pd();
726 fjz1 = _mm_setzero_pd();
727 fjx2 = _mm_setzero_pd();
728 fjy2 = _mm_setzero_pd();
729 fjz2 = _mm_setzero_pd();
731 /**************************
732 * CALCULATE INTERACTIONS *
733 **************************/
735 r00 = _mm_mul_pd(rsq00,rinv00);
737 /* Calculate table index by multiplying r with table scale and truncate to integer */
738 rt = _mm_mul_pd(r00,vftabscale);
739 vfitab = _mm_cvttpd_epi32(rt);
741 vfeps = _mm_frcz_pd(rt);
743 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
745 twovfeps = _mm_add_pd(vfeps,vfeps);
746 vfitab = _mm_slli_epi32(vfitab,2);
748 /* CUBIC SPLINE TABLE ELECTROSTATICS */
749 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
750 F = _mm_setzero_pd();
751 GMX_MM_TRANSPOSE2_PD(Y,F);
752 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
753 H = _mm_setzero_pd();
754 GMX_MM_TRANSPOSE2_PD(G,H);
755 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
756 VV = _mm_macc_pd(vfeps,Fp,Y);
757 velec = _mm_mul_pd(qq00,VV);
758 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
759 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
761 /* Update potential sum for this i atom from the interaction with this j atom. */
762 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
763 velecsum = _mm_add_pd(velecsum,velec);
767 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
769 /* Update vectorial force */
770 fix0 = _mm_macc_pd(dx00,fscal,fix0);
771 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
772 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
774 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
775 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
776 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
778 /**************************
779 * CALCULATE INTERACTIONS *
780 **************************/
782 r01 = _mm_mul_pd(rsq01,rinv01);
784 /* Calculate table index by multiplying r with table scale and truncate to integer */
785 rt = _mm_mul_pd(r01,vftabscale);
786 vfitab = _mm_cvttpd_epi32(rt);
788 vfeps = _mm_frcz_pd(rt);
790 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
792 twovfeps = _mm_add_pd(vfeps,vfeps);
793 vfitab = _mm_slli_epi32(vfitab,2);
795 /* CUBIC SPLINE TABLE ELECTROSTATICS */
796 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
797 F = _mm_setzero_pd();
798 GMX_MM_TRANSPOSE2_PD(Y,F);
799 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
800 H = _mm_setzero_pd();
801 GMX_MM_TRANSPOSE2_PD(G,H);
802 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
803 VV = _mm_macc_pd(vfeps,Fp,Y);
804 velec = _mm_mul_pd(qq01,VV);
805 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
806 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
808 /* Update potential sum for this i atom from the interaction with this j atom. */
809 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
810 velecsum = _mm_add_pd(velecsum,velec);
814 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
816 /* Update vectorial force */
817 fix0 = _mm_macc_pd(dx01,fscal,fix0);
818 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
819 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
821 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
822 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
823 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
825 /**************************
826 * CALCULATE INTERACTIONS *
827 **************************/
829 r02 = _mm_mul_pd(rsq02,rinv02);
831 /* Calculate table index by multiplying r with table scale and truncate to integer */
832 rt = _mm_mul_pd(r02,vftabscale);
833 vfitab = _mm_cvttpd_epi32(rt);
835 vfeps = _mm_frcz_pd(rt);
837 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
839 twovfeps = _mm_add_pd(vfeps,vfeps);
840 vfitab = _mm_slli_epi32(vfitab,2);
842 /* CUBIC SPLINE TABLE ELECTROSTATICS */
843 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
844 F = _mm_setzero_pd();
845 GMX_MM_TRANSPOSE2_PD(Y,F);
846 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
847 H = _mm_setzero_pd();
848 GMX_MM_TRANSPOSE2_PD(G,H);
849 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
850 VV = _mm_macc_pd(vfeps,Fp,Y);
851 velec = _mm_mul_pd(qq02,VV);
852 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
853 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
855 /* Update potential sum for this i atom from the interaction with this j atom. */
856 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
857 velecsum = _mm_add_pd(velecsum,velec);
861 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
863 /* Update vectorial force */
864 fix0 = _mm_macc_pd(dx02,fscal,fix0);
865 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
866 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
868 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
869 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
870 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
872 /**************************
873 * CALCULATE INTERACTIONS *
874 **************************/
876 r10 = _mm_mul_pd(rsq10,rinv10);
878 /* Calculate table index by multiplying r with table scale and truncate to integer */
879 rt = _mm_mul_pd(r10,vftabscale);
880 vfitab = _mm_cvttpd_epi32(rt);
882 vfeps = _mm_frcz_pd(rt);
884 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
886 twovfeps = _mm_add_pd(vfeps,vfeps);
887 vfitab = _mm_slli_epi32(vfitab,2);
889 /* CUBIC SPLINE TABLE ELECTROSTATICS */
890 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
891 F = _mm_setzero_pd();
892 GMX_MM_TRANSPOSE2_PD(Y,F);
893 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
894 H = _mm_setzero_pd();
895 GMX_MM_TRANSPOSE2_PD(G,H);
896 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
897 VV = _mm_macc_pd(vfeps,Fp,Y);
898 velec = _mm_mul_pd(qq10,VV);
899 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
900 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
902 /* Update potential sum for this i atom from the interaction with this j atom. */
903 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
904 velecsum = _mm_add_pd(velecsum,velec);
908 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
910 /* Update vectorial force */
911 fix1 = _mm_macc_pd(dx10,fscal,fix1);
912 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
913 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
915 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
916 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
917 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
919 /**************************
920 * CALCULATE INTERACTIONS *
921 **************************/
923 r11 = _mm_mul_pd(rsq11,rinv11);
925 /* Calculate table index by multiplying r with table scale and truncate to integer */
926 rt = _mm_mul_pd(r11,vftabscale);
927 vfitab = _mm_cvttpd_epi32(rt);
929 vfeps = _mm_frcz_pd(rt);
931 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
933 twovfeps = _mm_add_pd(vfeps,vfeps);
934 vfitab = _mm_slli_epi32(vfitab,2);
936 /* CUBIC SPLINE TABLE ELECTROSTATICS */
937 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
938 F = _mm_setzero_pd();
939 GMX_MM_TRANSPOSE2_PD(Y,F);
940 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
941 H = _mm_setzero_pd();
942 GMX_MM_TRANSPOSE2_PD(G,H);
943 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
944 VV = _mm_macc_pd(vfeps,Fp,Y);
945 velec = _mm_mul_pd(qq11,VV);
946 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
947 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
949 /* Update potential sum for this i atom from the interaction with this j atom. */
950 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
951 velecsum = _mm_add_pd(velecsum,velec);
955 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
957 /* Update vectorial force */
958 fix1 = _mm_macc_pd(dx11,fscal,fix1);
959 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
960 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
962 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
963 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
964 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
966 /**************************
967 * CALCULATE INTERACTIONS *
968 **************************/
970 r12 = _mm_mul_pd(rsq12,rinv12);
972 /* Calculate table index by multiplying r with table scale and truncate to integer */
973 rt = _mm_mul_pd(r12,vftabscale);
974 vfitab = _mm_cvttpd_epi32(rt);
976 vfeps = _mm_frcz_pd(rt);
978 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
980 twovfeps = _mm_add_pd(vfeps,vfeps);
981 vfitab = _mm_slli_epi32(vfitab,2);
983 /* CUBIC SPLINE TABLE ELECTROSTATICS */
984 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
985 F = _mm_setzero_pd();
986 GMX_MM_TRANSPOSE2_PD(Y,F);
987 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
988 H = _mm_setzero_pd();
989 GMX_MM_TRANSPOSE2_PD(G,H);
990 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
991 VV = _mm_macc_pd(vfeps,Fp,Y);
992 velec = _mm_mul_pd(qq12,VV);
993 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
994 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
996 /* Update potential sum for this i atom from the interaction with this j atom. */
997 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
998 velecsum = _mm_add_pd(velecsum,velec);
1002 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1004 /* Update vectorial force */
1005 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1006 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1007 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1009 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1010 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1011 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1013 /**************************
1014 * CALCULATE INTERACTIONS *
1015 **************************/
1017 r20 = _mm_mul_pd(rsq20,rinv20);
1019 /* Calculate table index by multiplying r with table scale and truncate to integer */
1020 rt = _mm_mul_pd(r20,vftabscale);
1021 vfitab = _mm_cvttpd_epi32(rt);
1023 vfeps = _mm_frcz_pd(rt);
1025 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1027 twovfeps = _mm_add_pd(vfeps,vfeps);
1028 vfitab = _mm_slli_epi32(vfitab,2);
1030 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1031 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1032 F = _mm_setzero_pd();
1033 GMX_MM_TRANSPOSE2_PD(Y,F);
1034 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1035 H = _mm_setzero_pd();
1036 GMX_MM_TRANSPOSE2_PD(G,H);
1037 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1038 VV = _mm_macc_pd(vfeps,Fp,Y);
1039 velec = _mm_mul_pd(qq20,VV);
1040 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1041 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1043 /* Update potential sum for this i atom from the interaction with this j atom. */
1044 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1045 velecsum = _mm_add_pd(velecsum,velec);
1049 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1051 /* Update vectorial force */
1052 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1053 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1054 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1056 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1057 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1058 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1060 /**************************
1061 * CALCULATE INTERACTIONS *
1062 **************************/
1064 r21 = _mm_mul_pd(rsq21,rinv21);
1066 /* Calculate table index by multiplying r with table scale and truncate to integer */
1067 rt = _mm_mul_pd(r21,vftabscale);
1068 vfitab = _mm_cvttpd_epi32(rt);
1070 vfeps = _mm_frcz_pd(rt);
1072 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1074 twovfeps = _mm_add_pd(vfeps,vfeps);
1075 vfitab = _mm_slli_epi32(vfitab,2);
1077 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1078 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1079 F = _mm_setzero_pd();
1080 GMX_MM_TRANSPOSE2_PD(Y,F);
1081 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1082 H = _mm_setzero_pd();
1083 GMX_MM_TRANSPOSE2_PD(G,H);
1084 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1085 VV = _mm_macc_pd(vfeps,Fp,Y);
1086 velec = _mm_mul_pd(qq21,VV);
1087 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1088 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1090 /* Update potential sum for this i atom from the interaction with this j atom. */
1091 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1092 velecsum = _mm_add_pd(velecsum,velec);
1096 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1098 /* Update vectorial force */
1099 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1100 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1101 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1103 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1104 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1105 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1107 /**************************
1108 * CALCULATE INTERACTIONS *
1109 **************************/
1111 r22 = _mm_mul_pd(rsq22,rinv22);
1113 /* Calculate table index by multiplying r with table scale and truncate to integer */
1114 rt = _mm_mul_pd(r22,vftabscale);
1115 vfitab = _mm_cvttpd_epi32(rt);
1117 vfeps = _mm_frcz_pd(rt);
1119 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1121 twovfeps = _mm_add_pd(vfeps,vfeps);
1122 vfitab = _mm_slli_epi32(vfitab,2);
1124 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1125 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1126 F = _mm_setzero_pd();
1127 GMX_MM_TRANSPOSE2_PD(Y,F);
1128 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1129 H = _mm_setzero_pd();
1130 GMX_MM_TRANSPOSE2_PD(G,H);
1131 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1132 VV = _mm_macc_pd(vfeps,Fp,Y);
1133 velec = _mm_mul_pd(qq22,VV);
1134 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1135 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1137 /* Update potential sum for this i atom from the interaction with this j atom. */
1138 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1139 velecsum = _mm_add_pd(velecsum,velec);
1143 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1145 /* Update vectorial force */
1146 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1147 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1148 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1150 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1151 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1152 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1154 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1156 /* Inner loop uses 414 flops */
1159 /* End of innermost loop */
1161 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1162 f+i_coord_offset,fshift+i_shift_offset);
1165 /* Update potential energies */
1166 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1168 /* Increment number of inner iterations */
1169 inneriter += j_index_end - j_index_start;
1171 /* Outer loop uses 19 flops */
1174 /* Increment number of outer iterations */
1177 /* Update outer/inner flops */
1179 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*414);
1182 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_avx_128_fma_double
1183 * Electrostatics interaction: CubicSplineTable
1184 * VdW interaction: None
1185 * Geometry: Water3-Water3
1186 * Calculate force/pot: Force
1189 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_avx_128_fma_double
1190 (t_nblist * gmx_restrict nlist,
1191 rvec * gmx_restrict xx,
1192 rvec * gmx_restrict ff,
1193 t_forcerec * gmx_restrict fr,
1194 t_mdatoms * gmx_restrict mdatoms,
1195 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1196 t_nrnb * gmx_restrict nrnb)
1198 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1199 * just 0 for non-waters.
1200 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1201 * jnr indices corresponding to data put in the four positions in the SIMD register.
1203 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1204 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1206 int j_coord_offsetA,j_coord_offsetB;
1207 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1208 real rcutoff_scalar;
1209 real *shiftvec,*fshift,*x,*f;
1210 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1212 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1214 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1216 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1217 int vdwjidx0A,vdwjidx0B;
1218 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1219 int vdwjidx1A,vdwjidx1B;
1220 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1221 int vdwjidx2A,vdwjidx2B;
1222 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1223 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1224 __m128d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1225 __m128d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1226 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1227 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1228 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1229 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1230 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1231 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1232 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1235 __m128i ifour = _mm_set1_epi32(4);
1236 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1238 __m128d dummy_mask,cutoff_mask;
1239 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1240 __m128d one = _mm_set1_pd(1.0);
1241 __m128d two = _mm_set1_pd(2.0);
1247 jindex = nlist->jindex;
1249 shiftidx = nlist->shift;
1251 shiftvec = fr->shift_vec[0];
1252 fshift = fr->fshift[0];
1253 facel = _mm_set1_pd(fr->epsfac);
1254 charge = mdatoms->chargeA;
1256 vftab = kernel_data->table_elec->data;
1257 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
1259 /* Setup water-specific parameters */
1260 inr = nlist->iinr[0];
1261 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1262 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1263 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1265 jq0 = _mm_set1_pd(charge[inr+0]);
1266 jq1 = _mm_set1_pd(charge[inr+1]);
1267 jq2 = _mm_set1_pd(charge[inr+2]);
1268 qq00 = _mm_mul_pd(iq0,jq0);
1269 qq01 = _mm_mul_pd(iq0,jq1);
1270 qq02 = _mm_mul_pd(iq0,jq2);
1271 qq10 = _mm_mul_pd(iq1,jq0);
1272 qq11 = _mm_mul_pd(iq1,jq1);
1273 qq12 = _mm_mul_pd(iq1,jq2);
1274 qq20 = _mm_mul_pd(iq2,jq0);
1275 qq21 = _mm_mul_pd(iq2,jq1);
1276 qq22 = _mm_mul_pd(iq2,jq2);
1278 /* Avoid stupid compiler warnings */
1280 j_coord_offsetA = 0;
1281 j_coord_offsetB = 0;
1286 /* Start outer loop over neighborlists */
1287 for(iidx=0; iidx<nri; iidx++)
1289 /* Load shift vector for this list */
1290 i_shift_offset = DIM*shiftidx[iidx];
1292 /* Load limits for loop over neighbors */
1293 j_index_start = jindex[iidx];
1294 j_index_end = jindex[iidx+1];
1296 /* Get outer coordinate index */
1298 i_coord_offset = DIM*inr;
1300 /* Load i particle coords and add shift vector */
1301 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1302 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1304 fix0 = _mm_setzero_pd();
1305 fiy0 = _mm_setzero_pd();
1306 fiz0 = _mm_setzero_pd();
1307 fix1 = _mm_setzero_pd();
1308 fiy1 = _mm_setzero_pd();
1309 fiz1 = _mm_setzero_pd();
1310 fix2 = _mm_setzero_pd();
1311 fiy2 = _mm_setzero_pd();
1312 fiz2 = _mm_setzero_pd();
1314 /* Start inner kernel loop */
1315 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1318 /* Get j neighbor index, and coordinate index */
1320 jnrB = jjnr[jidx+1];
1321 j_coord_offsetA = DIM*jnrA;
1322 j_coord_offsetB = DIM*jnrB;
1324 /* load j atom coordinates */
1325 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1326 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1328 /* Calculate displacement vector */
1329 dx00 = _mm_sub_pd(ix0,jx0);
1330 dy00 = _mm_sub_pd(iy0,jy0);
1331 dz00 = _mm_sub_pd(iz0,jz0);
1332 dx01 = _mm_sub_pd(ix0,jx1);
1333 dy01 = _mm_sub_pd(iy0,jy1);
1334 dz01 = _mm_sub_pd(iz0,jz1);
1335 dx02 = _mm_sub_pd(ix0,jx2);
1336 dy02 = _mm_sub_pd(iy0,jy2);
1337 dz02 = _mm_sub_pd(iz0,jz2);
1338 dx10 = _mm_sub_pd(ix1,jx0);
1339 dy10 = _mm_sub_pd(iy1,jy0);
1340 dz10 = _mm_sub_pd(iz1,jz0);
1341 dx11 = _mm_sub_pd(ix1,jx1);
1342 dy11 = _mm_sub_pd(iy1,jy1);
1343 dz11 = _mm_sub_pd(iz1,jz1);
1344 dx12 = _mm_sub_pd(ix1,jx2);
1345 dy12 = _mm_sub_pd(iy1,jy2);
1346 dz12 = _mm_sub_pd(iz1,jz2);
1347 dx20 = _mm_sub_pd(ix2,jx0);
1348 dy20 = _mm_sub_pd(iy2,jy0);
1349 dz20 = _mm_sub_pd(iz2,jz0);
1350 dx21 = _mm_sub_pd(ix2,jx1);
1351 dy21 = _mm_sub_pd(iy2,jy1);
1352 dz21 = _mm_sub_pd(iz2,jz1);
1353 dx22 = _mm_sub_pd(ix2,jx2);
1354 dy22 = _mm_sub_pd(iy2,jy2);
1355 dz22 = _mm_sub_pd(iz2,jz2);
1357 /* Calculate squared distance and things based on it */
1358 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1359 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1360 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1361 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1362 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1363 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1364 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1365 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1366 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1368 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1369 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1370 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1371 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1372 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1373 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1374 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1375 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1376 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1378 fjx0 = _mm_setzero_pd();
1379 fjy0 = _mm_setzero_pd();
1380 fjz0 = _mm_setzero_pd();
1381 fjx1 = _mm_setzero_pd();
1382 fjy1 = _mm_setzero_pd();
1383 fjz1 = _mm_setzero_pd();
1384 fjx2 = _mm_setzero_pd();
1385 fjy2 = _mm_setzero_pd();
1386 fjz2 = _mm_setzero_pd();
1388 /**************************
1389 * CALCULATE INTERACTIONS *
1390 **************************/
1392 r00 = _mm_mul_pd(rsq00,rinv00);
1394 /* Calculate table index by multiplying r with table scale and truncate to integer */
1395 rt = _mm_mul_pd(r00,vftabscale);
1396 vfitab = _mm_cvttpd_epi32(rt);
1398 vfeps = _mm_frcz_pd(rt);
1400 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1402 twovfeps = _mm_add_pd(vfeps,vfeps);
1403 vfitab = _mm_slli_epi32(vfitab,2);
1405 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1406 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1407 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1408 GMX_MM_TRANSPOSE2_PD(Y,F);
1409 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1410 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1411 GMX_MM_TRANSPOSE2_PD(G,H);
1412 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1413 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1414 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1418 /* Update vectorial force */
1419 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1420 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1421 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1423 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1424 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1425 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1427 /**************************
1428 * CALCULATE INTERACTIONS *
1429 **************************/
1431 r01 = _mm_mul_pd(rsq01,rinv01);
1433 /* Calculate table index by multiplying r with table scale and truncate to integer */
1434 rt = _mm_mul_pd(r01,vftabscale);
1435 vfitab = _mm_cvttpd_epi32(rt);
1437 vfeps = _mm_frcz_pd(rt);
1439 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1441 twovfeps = _mm_add_pd(vfeps,vfeps);
1442 vfitab = _mm_slli_epi32(vfitab,2);
1444 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1445 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1446 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1447 GMX_MM_TRANSPOSE2_PD(Y,F);
1448 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1449 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1450 GMX_MM_TRANSPOSE2_PD(G,H);
1451 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1452 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1453 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
1457 /* Update vectorial force */
1458 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1459 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1460 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1462 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1463 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1464 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1466 /**************************
1467 * CALCULATE INTERACTIONS *
1468 **************************/
1470 r02 = _mm_mul_pd(rsq02,rinv02);
1472 /* Calculate table index by multiplying r with table scale and truncate to integer */
1473 rt = _mm_mul_pd(r02,vftabscale);
1474 vfitab = _mm_cvttpd_epi32(rt);
1476 vfeps = _mm_frcz_pd(rt);
1478 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1480 twovfeps = _mm_add_pd(vfeps,vfeps);
1481 vfitab = _mm_slli_epi32(vfitab,2);
1483 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1484 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1485 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1486 GMX_MM_TRANSPOSE2_PD(Y,F);
1487 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1488 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1489 GMX_MM_TRANSPOSE2_PD(G,H);
1490 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1491 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1492 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
1496 /* Update vectorial force */
1497 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1498 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1499 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1501 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1502 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1503 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1505 /**************************
1506 * CALCULATE INTERACTIONS *
1507 **************************/
1509 r10 = _mm_mul_pd(rsq10,rinv10);
1511 /* Calculate table index by multiplying r with table scale and truncate to integer */
1512 rt = _mm_mul_pd(r10,vftabscale);
1513 vfitab = _mm_cvttpd_epi32(rt);
1515 vfeps = _mm_frcz_pd(rt);
1517 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1519 twovfeps = _mm_add_pd(vfeps,vfeps);
1520 vfitab = _mm_slli_epi32(vfitab,2);
1522 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1523 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1524 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1525 GMX_MM_TRANSPOSE2_PD(Y,F);
1526 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1527 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1528 GMX_MM_TRANSPOSE2_PD(G,H);
1529 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1530 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1531 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1535 /* Update vectorial force */
1536 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1537 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1538 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1540 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1541 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1542 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1544 /**************************
1545 * CALCULATE INTERACTIONS *
1546 **************************/
1548 r11 = _mm_mul_pd(rsq11,rinv11);
1550 /* Calculate table index by multiplying r with table scale and truncate to integer */
1551 rt = _mm_mul_pd(r11,vftabscale);
1552 vfitab = _mm_cvttpd_epi32(rt);
1554 vfeps = _mm_frcz_pd(rt);
1556 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1558 twovfeps = _mm_add_pd(vfeps,vfeps);
1559 vfitab = _mm_slli_epi32(vfitab,2);
1561 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1562 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1563 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1564 GMX_MM_TRANSPOSE2_PD(Y,F);
1565 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1566 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1567 GMX_MM_TRANSPOSE2_PD(G,H);
1568 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1569 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1570 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
1574 /* Update vectorial force */
1575 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1576 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1577 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1579 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1580 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1581 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1583 /**************************
1584 * CALCULATE INTERACTIONS *
1585 **************************/
1587 r12 = _mm_mul_pd(rsq12,rinv12);
1589 /* Calculate table index by multiplying r with table scale and truncate to integer */
1590 rt = _mm_mul_pd(r12,vftabscale);
1591 vfitab = _mm_cvttpd_epi32(rt);
1593 vfeps = _mm_frcz_pd(rt);
1595 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1597 twovfeps = _mm_add_pd(vfeps,vfeps);
1598 vfitab = _mm_slli_epi32(vfitab,2);
1600 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1601 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1602 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1603 GMX_MM_TRANSPOSE2_PD(Y,F);
1604 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1605 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1606 GMX_MM_TRANSPOSE2_PD(G,H);
1607 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1608 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1609 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1613 /* Update vectorial force */
1614 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1615 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1616 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1618 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1619 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1620 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1622 /**************************
1623 * CALCULATE INTERACTIONS *
1624 **************************/
1626 r20 = _mm_mul_pd(rsq20,rinv20);
1628 /* Calculate table index by multiplying r with table scale and truncate to integer */
1629 rt = _mm_mul_pd(r20,vftabscale);
1630 vfitab = _mm_cvttpd_epi32(rt);
1632 vfeps = _mm_frcz_pd(rt);
1634 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1636 twovfeps = _mm_add_pd(vfeps,vfeps);
1637 vfitab = _mm_slli_epi32(vfitab,2);
1639 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1640 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1641 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1642 GMX_MM_TRANSPOSE2_PD(Y,F);
1643 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1644 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1645 GMX_MM_TRANSPOSE2_PD(G,H);
1646 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1647 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1648 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1652 /* Update vectorial force */
1653 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1654 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1655 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1657 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1658 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1659 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1661 /**************************
1662 * CALCULATE INTERACTIONS *
1663 **************************/
1665 r21 = _mm_mul_pd(rsq21,rinv21);
1667 /* Calculate table index by multiplying r with table scale and truncate to integer */
1668 rt = _mm_mul_pd(r21,vftabscale);
1669 vfitab = _mm_cvttpd_epi32(rt);
1671 vfeps = _mm_frcz_pd(rt);
1673 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1675 twovfeps = _mm_add_pd(vfeps,vfeps);
1676 vfitab = _mm_slli_epi32(vfitab,2);
1678 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1679 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1680 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1681 GMX_MM_TRANSPOSE2_PD(Y,F);
1682 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1683 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1684 GMX_MM_TRANSPOSE2_PD(G,H);
1685 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1686 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1687 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1691 /* Update vectorial force */
1692 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1693 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1694 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1696 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1697 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1698 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1700 /**************************
1701 * CALCULATE INTERACTIONS *
1702 **************************/
1704 r22 = _mm_mul_pd(rsq22,rinv22);
1706 /* Calculate table index by multiplying r with table scale and truncate to integer */
1707 rt = _mm_mul_pd(r22,vftabscale);
1708 vfitab = _mm_cvttpd_epi32(rt);
1710 vfeps = _mm_frcz_pd(rt);
1712 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1714 twovfeps = _mm_add_pd(vfeps,vfeps);
1715 vfitab = _mm_slli_epi32(vfitab,2);
1717 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1718 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1719 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1720 GMX_MM_TRANSPOSE2_PD(Y,F);
1721 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1722 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1723 GMX_MM_TRANSPOSE2_PD(G,H);
1724 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1725 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1726 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1730 /* Update vectorial force */
1731 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1732 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1733 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1735 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1736 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1737 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1739 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1741 /* Inner loop uses 378 flops */
1744 if(jidx<j_index_end)
1748 j_coord_offsetA = DIM*jnrA;
1750 /* load j atom coordinates */
1751 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1752 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1754 /* Calculate displacement vector */
1755 dx00 = _mm_sub_pd(ix0,jx0);
1756 dy00 = _mm_sub_pd(iy0,jy0);
1757 dz00 = _mm_sub_pd(iz0,jz0);
1758 dx01 = _mm_sub_pd(ix0,jx1);
1759 dy01 = _mm_sub_pd(iy0,jy1);
1760 dz01 = _mm_sub_pd(iz0,jz1);
1761 dx02 = _mm_sub_pd(ix0,jx2);
1762 dy02 = _mm_sub_pd(iy0,jy2);
1763 dz02 = _mm_sub_pd(iz0,jz2);
1764 dx10 = _mm_sub_pd(ix1,jx0);
1765 dy10 = _mm_sub_pd(iy1,jy0);
1766 dz10 = _mm_sub_pd(iz1,jz0);
1767 dx11 = _mm_sub_pd(ix1,jx1);
1768 dy11 = _mm_sub_pd(iy1,jy1);
1769 dz11 = _mm_sub_pd(iz1,jz1);
1770 dx12 = _mm_sub_pd(ix1,jx2);
1771 dy12 = _mm_sub_pd(iy1,jy2);
1772 dz12 = _mm_sub_pd(iz1,jz2);
1773 dx20 = _mm_sub_pd(ix2,jx0);
1774 dy20 = _mm_sub_pd(iy2,jy0);
1775 dz20 = _mm_sub_pd(iz2,jz0);
1776 dx21 = _mm_sub_pd(ix2,jx1);
1777 dy21 = _mm_sub_pd(iy2,jy1);
1778 dz21 = _mm_sub_pd(iz2,jz1);
1779 dx22 = _mm_sub_pd(ix2,jx2);
1780 dy22 = _mm_sub_pd(iy2,jy2);
1781 dz22 = _mm_sub_pd(iz2,jz2);
1783 /* Calculate squared distance and things based on it */
1784 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1785 rsq01 = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1786 rsq02 = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1787 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1788 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1789 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1790 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1791 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1792 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1794 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1795 rinv01 = gmx_mm_invsqrt_pd(rsq01);
1796 rinv02 = gmx_mm_invsqrt_pd(rsq02);
1797 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1798 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1799 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1800 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1801 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1802 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1804 fjx0 = _mm_setzero_pd();
1805 fjy0 = _mm_setzero_pd();
1806 fjz0 = _mm_setzero_pd();
1807 fjx1 = _mm_setzero_pd();
1808 fjy1 = _mm_setzero_pd();
1809 fjz1 = _mm_setzero_pd();
1810 fjx2 = _mm_setzero_pd();
1811 fjy2 = _mm_setzero_pd();
1812 fjz2 = _mm_setzero_pd();
1814 /**************************
1815 * CALCULATE INTERACTIONS *
1816 **************************/
1818 r00 = _mm_mul_pd(rsq00,rinv00);
1820 /* Calculate table index by multiplying r with table scale and truncate to integer */
1821 rt = _mm_mul_pd(r00,vftabscale);
1822 vfitab = _mm_cvttpd_epi32(rt);
1824 vfeps = _mm_frcz_pd(rt);
1826 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1828 twovfeps = _mm_add_pd(vfeps,vfeps);
1829 vfitab = _mm_slli_epi32(vfitab,2);
1831 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1832 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1833 F = _mm_setzero_pd();
1834 GMX_MM_TRANSPOSE2_PD(Y,F);
1835 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1836 H = _mm_setzero_pd();
1837 GMX_MM_TRANSPOSE2_PD(G,H);
1838 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1839 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1840 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1844 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1846 /* Update vectorial force */
1847 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1848 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1849 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1851 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1852 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1853 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1855 /**************************
1856 * CALCULATE INTERACTIONS *
1857 **************************/
1859 r01 = _mm_mul_pd(rsq01,rinv01);
1861 /* Calculate table index by multiplying r with table scale and truncate to integer */
1862 rt = _mm_mul_pd(r01,vftabscale);
1863 vfitab = _mm_cvttpd_epi32(rt);
1865 vfeps = _mm_frcz_pd(rt);
1867 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1869 twovfeps = _mm_add_pd(vfeps,vfeps);
1870 vfitab = _mm_slli_epi32(vfitab,2);
1872 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1873 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1874 F = _mm_setzero_pd();
1875 GMX_MM_TRANSPOSE2_PD(Y,F);
1876 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1877 H = _mm_setzero_pd();
1878 GMX_MM_TRANSPOSE2_PD(G,H);
1879 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1880 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1881 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
1885 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1887 /* Update vectorial force */
1888 fix0 = _mm_macc_pd(dx01,fscal,fix0);
1889 fiy0 = _mm_macc_pd(dy01,fscal,fiy0);
1890 fiz0 = _mm_macc_pd(dz01,fscal,fiz0);
1892 fjx1 = _mm_macc_pd(dx01,fscal,fjx1);
1893 fjy1 = _mm_macc_pd(dy01,fscal,fjy1);
1894 fjz1 = _mm_macc_pd(dz01,fscal,fjz1);
1896 /**************************
1897 * CALCULATE INTERACTIONS *
1898 **************************/
1900 r02 = _mm_mul_pd(rsq02,rinv02);
1902 /* Calculate table index by multiplying r with table scale and truncate to integer */
1903 rt = _mm_mul_pd(r02,vftabscale);
1904 vfitab = _mm_cvttpd_epi32(rt);
1906 vfeps = _mm_frcz_pd(rt);
1908 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1910 twovfeps = _mm_add_pd(vfeps,vfeps);
1911 vfitab = _mm_slli_epi32(vfitab,2);
1913 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1914 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1915 F = _mm_setzero_pd();
1916 GMX_MM_TRANSPOSE2_PD(Y,F);
1917 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1918 H = _mm_setzero_pd();
1919 GMX_MM_TRANSPOSE2_PD(G,H);
1920 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1921 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1922 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
1926 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1928 /* Update vectorial force */
1929 fix0 = _mm_macc_pd(dx02,fscal,fix0);
1930 fiy0 = _mm_macc_pd(dy02,fscal,fiy0);
1931 fiz0 = _mm_macc_pd(dz02,fscal,fiz0);
1933 fjx2 = _mm_macc_pd(dx02,fscal,fjx2);
1934 fjy2 = _mm_macc_pd(dy02,fscal,fjy2);
1935 fjz2 = _mm_macc_pd(dz02,fscal,fjz2);
1937 /**************************
1938 * CALCULATE INTERACTIONS *
1939 **************************/
1941 r10 = _mm_mul_pd(rsq10,rinv10);
1943 /* Calculate table index by multiplying r with table scale and truncate to integer */
1944 rt = _mm_mul_pd(r10,vftabscale);
1945 vfitab = _mm_cvttpd_epi32(rt);
1947 vfeps = _mm_frcz_pd(rt);
1949 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1951 twovfeps = _mm_add_pd(vfeps,vfeps);
1952 vfitab = _mm_slli_epi32(vfitab,2);
1954 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1955 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1956 F = _mm_setzero_pd();
1957 GMX_MM_TRANSPOSE2_PD(Y,F);
1958 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1959 H = _mm_setzero_pd();
1960 GMX_MM_TRANSPOSE2_PD(G,H);
1961 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1962 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1963 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1967 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1969 /* Update vectorial force */
1970 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1971 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1972 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1974 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1975 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1976 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1978 /**************************
1979 * CALCULATE INTERACTIONS *
1980 **************************/
1982 r11 = _mm_mul_pd(rsq11,rinv11);
1984 /* Calculate table index by multiplying r with table scale and truncate to integer */
1985 rt = _mm_mul_pd(r11,vftabscale);
1986 vfitab = _mm_cvttpd_epi32(rt);
1988 vfeps = _mm_frcz_pd(rt);
1990 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1992 twovfeps = _mm_add_pd(vfeps,vfeps);
1993 vfitab = _mm_slli_epi32(vfitab,2);
1995 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1996 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1997 F = _mm_setzero_pd();
1998 GMX_MM_TRANSPOSE2_PD(Y,F);
1999 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2000 H = _mm_setzero_pd();
2001 GMX_MM_TRANSPOSE2_PD(G,H);
2002 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2003 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2004 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
2008 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2010 /* Update vectorial force */
2011 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2012 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2013 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2015 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2016 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2017 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2019 /**************************
2020 * CALCULATE INTERACTIONS *
2021 **************************/
2023 r12 = _mm_mul_pd(rsq12,rinv12);
2025 /* Calculate table index by multiplying r with table scale and truncate to integer */
2026 rt = _mm_mul_pd(r12,vftabscale);
2027 vfitab = _mm_cvttpd_epi32(rt);
2029 vfeps = _mm_frcz_pd(rt);
2031 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2033 twovfeps = _mm_add_pd(vfeps,vfeps);
2034 vfitab = _mm_slli_epi32(vfitab,2);
2036 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2037 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2038 F = _mm_setzero_pd();
2039 GMX_MM_TRANSPOSE2_PD(Y,F);
2040 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2041 H = _mm_setzero_pd();
2042 GMX_MM_TRANSPOSE2_PD(G,H);
2043 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2044 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2045 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
2049 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2051 /* Update vectorial force */
2052 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2053 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2054 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2056 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2057 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2058 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2060 /**************************
2061 * CALCULATE INTERACTIONS *
2062 **************************/
2064 r20 = _mm_mul_pd(rsq20,rinv20);
2066 /* Calculate table index by multiplying r with table scale and truncate to integer */
2067 rt = _mm_mul_pd(r20,vftabscale);
2068 vfitab = _mm_cvttpd_epi32(rt);
2070 vfeps = _mm_frcz_pd(rt);
2072 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2074 twovfeps = _mm_add_pd(vfeps,vfeps);
2075 vfitab = _mm_slli_epi32(vfitab,2);
2077 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2078 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2079 F = _mm_setzero_pd();
2080 GMX_MM_TRANSPOSE2_PD(Y,F);
2081 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2082 H = _mm_setzero_pd();
2083 GMX_MM_TRANSPOSE2_PD(G,H);
2084 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2085 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2086 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
2090 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2092 /* Update vectorial force */
2093 fix2 = _mm_macc_pd(dx20,fscal,fix2);
2094 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
2095 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
2097 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
2098 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
2099 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
2101 /**************************
2102 * CALCULATE INTERACTIONS *
2103 **************************/
2105 r21 = _mm_mul_pd(rsq21,rinv21);
2107 /* Calculate table index by multiplying r with table scale and truncate to integer */
2108 rt = _mm_mul_pd(r21,vftabscale);
2109 vfitab = _mm_cvttpd_epi32(rt);
2111 vfeps = _mm_frcz_pd(rt);
2113 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2115 twovfeps = _mm_add_pd(vfeps,vfeps);
2116 vfitab = _mm_slli_epi32(vfitab,2);
2118 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2119 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2120 F = _mm_setzero_pd();
2121 GMX_MM_TRANSPOSE2_PD(Y,F);
2122 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2123 H = _mm_setzero_pd();
2124 GMX_MM_TRANSPOSE2_PD(G,H);
2125 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2126 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2127 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
2131 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2133 /* Update vectorial force */
2134 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2135 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2136 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2138 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2139 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2140 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2142 /**************************
2143 * CALCULATE INTERACTIONS *
2144 **************************/
2146 r22 = _mm_mul_pd(rsq22,rinv22);
2148 /* Calculate table index by multiplying r with table scale and truncate to integer */
2149 rt = _mm_mul_pd(r22,vftabscale);
2150 vfitab = _mm_cvttpd_epi32(rt);
2152 vfeps = _mm_frcz_pd(rt);
2154 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2156 twovfeps = _mm_add_pd(vfeps,vfeps);
2157 vfitab = _mm_slli_epi32(vfitab,2);
2159 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2160 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2161 F = _mm_setzero_pd();
2162 GMX_MM_TRANSPOSE2_PD(Y,F);
2163 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2164 H = _mm_setzero_pd();
2165 GMX_MM_TRANSPOSE2_PD(G,H);
2166 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2167 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2168 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
2172 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2174 /* Update vectorial force */
2175 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2176 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2177 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2179 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2180 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2181 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2183 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2185 /* Inner loop uses 378 flops */
2188 /* End of innermost loop */
2190 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2191 f+i_coord_offset,fshift+i_shift_offset);
2193 /* Increment number of inner iterations */
2194 inneriter += j_index_end - j_index_start;
2196 /* Outer loop uses 18 flops */
2199 /* Increment number of outer iterations */
2202 /* Update outer/inner flops */
2204 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*378);