2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_VF_avx_128_fma_double
54 * Electrostatics interaction: CubicSplineTable
55 * VdW interaction: CubicSplineTable
56 * Geometry: Water4-Water4
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_VF_avx_128_fma_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
89 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
90 int vdwjidx0A,vdwjidx0B;
91 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 int vdwjidx1A,vdwjidx1B;
93 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
94 int vdwjidx2A,vdwjidx2B;
95 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
96 int vdwjidx3A,vdwjidx3B;
97 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
98 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
99 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
100 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
101 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
102 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
103 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
104 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
105 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
106 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
107 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
108 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
111 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
114 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
115 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
117 __m128i ifour = _mm_set1_epi32(4);
118 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
120 __m128d dummy_mask,cutoff_mask;
121 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
122 __m128d one = _mm_set1_pd(1.0);
123 __m128d two = _mm_set1_pd(2.0);
129 jindex = nlist->jindex;
131 shiftidx = nlist->shift;
133 shiftvec = fr->shift_vec[0];
134 fshift = fr->fshift[0];
135 facel = _mm_set1_pd(fr->epsfac);
136 charge = mdatoms->chargeA;
137 nvdwtype = fr->ntype;
139 vdwtype = mdatoms->typeA;
141 vftab = kernel_data->table_elec_vdw->data;
142 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
144 /* Setup water-specific parameters */
145 inr = nlist->iinr[0];
146 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
147 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
148 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
149 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
151 jq1 = _mm_set1_pd(charge[inr+1]);
152 jq2 = _mm_set1_pd(charge[inr+2]);
153 jq3 = _mm_set1_pd(charge[inr+3]);
154 vdwjidx0A = 2*vdwtype[inr+0];
155 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
156 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
157 qq11 = _mm_mul_pd(iq1,jq1);
158 qq12 = _mm_mul_pd(iq1,jq2);
159 qq13 = _mm_mul_pd(iq1,jq3);
160 qq21 = _mm_mul_pd(iq2,jq1);
161 qq22 = _mm_mul_pd(iq2,jq2);
162 qq23 = _mm_mul_pd(iq2,jq3);
163 qq31 = _mm_mul_pd(iq3,jq1);
164 qq32 = _mm_mul_pd(iq3,jq2);
165 qq33 = _mm_mul_pd(iq3,jq3);
167 /* Avoid stupid compiler warnings */
175 /* Start outer loop over neighborlists */
176 for(iidx=0; iidx<nri; iidx++)
178 /* Load shift vector for this list */
179 i_shift_offset = DIM*shiftidx[iidx];
181 /* Load limits for loop over neighbors */
182 j_index_start = jindex[iidx];
183 j_index_end = jindex[iidx+1];
185 /* Get outer coordinate index */
187 i_coord_offset = DIM*inr;
189 /* Load i particle coords and add shift vector */
190 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
191 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
193 fix0 = _mm_setzero_pd();
194 fiy0 = _mm_setzero_pd();
195 fiz0 = _mm_setzero_pd();
196 fix1 = _mm_setzero_pd();
197 fiy1 = _mm_setzero_pd();
198 fiz1 = _mm_setzero_pd();
199 fix2 = _mm_setzero_pd();
200 fiy2 = _mm_setzero_pd();
201 fiz2 = _mm_setzero_pd();
202 fix3 = _mm_setzero_pd();
203 fiy3 = _mm_setzero_pd();
204 fiz3 = _mm_setzero_pd();
206 /* Reset potential sums */
207 velecsum = _mm_setzero_pd();
208 vvdwsum = _mm_setzero_pd();
210 /* Start inner kernel loop */
211 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
214 /* Get j neighbor index, and coordinate index */
217 j_coord_offsetA = DIM*jnrA;
218 j_coord_offsetB = DIM*jnrB;
220 /* load j atom coordinates */
221 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
222 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
223 &jy2,&jz2,&jx3,&jy3,&jz3);
225 /* Calculate displacement vector */
226 dx00 = _mm_sub_pd(ix0,jx0);
227 dy00 = _mm_sub_pd(iy0,jy0);
228 dz00 = _mm_sub_pd(iz0,jz0);
229 dx11 = _mm_sub_pd(ix1,jx1);
230 dy11 = _mm_sub_pd(iy1,jy1);
231 dz11 = _mm_sub_pd(iz1,jz1);
232 dx12 = _mm_sub_pd(ix1,jx2);
233 dy12 = _mm_sub_pd(iy1,jy2);
234 dz12 = _mm_sub_pd(iz1,jz2);
235 dx13 = _mm_sub_pd(ix1,jx3);
236 dy13 = _mm_sub_pd(iy1,jy3);
237 dz13 = _mm_sub_pd(iz1,jz3);
238 dx21 = _mm_sub_pd(ix2,jx1);
239 dy21 = _mm_sub_pd(iy2,jy1);
240 dz21 = _mm_sub_pd(iz2,jz1);
241 dx22 = _mm_sub_pd(ix2,jx2);
242 dy22 = _mm_sub_pd(iy2,jy2);
243 dz22 = _mm_sub_pd(iz2,jz2);
244 dx23 = _mm_sub_pd(ix2,jx3);
245 dy23 = _mm_sub_pd(iy2,jy3);
246 dz23 = _mm_sub_pd(iz2,jz3);
247 dx31 = _mm_sub_pd(ix3,jx1);
248 dy31 = _mm_sub_pd(iy3,jy1);
249 dz31 = _mm_sub_pd(iz3,jz1);
250 dx32 = _mm_sub_pd(ix3,jx2);
251 dy32 = _mm_sub_pd(iy3,jy2);
252 dz32 = _mm_sub_pd(iz3,jz2);
253 dx33 = _mm_sub_pd(ix3,jx3);
254 dy33 = _mm_sub_pd(iy3,jy3);
255 dz33 = _mm_sub_pd(iz3,jz3);
257 /* Calculate squared distance and things based on it */
258 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
259 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
260 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
261 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
262 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
263 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
264 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
265 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
266 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
267 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
269 rinv00 = gmx_mm_invsqrt_pd(rsq00);
270 rinv11 = gmx_mm_invsqrt_pd(rsq11);
271 rinv12 = gmx_mm_invsqrt_pd(rsq12);
272 rinv13 = gmx_mm_invsqrt_pd(rsq13);
273 rinv21 = gmx_mm_invsqrt_pd(rsq21);
274 rinv22 = gmx_mm_invsqrt_pd(rsq22);
275 rinv23 = gmx_mm_invsqrt_pd(rsq23);
276 rinv31 = gmx_mm_invsqrt_pd(rsq31);
277 rinv32 = gmx_mm_invsqrt_pd(rsq32);
278 rinv33 = gmx_mm_invsqrt_pd(rsq33);
280 fjx0 = _mm_setzero_pd();
281 fjy0 = _mm_setzero_pd();
282 fjz0 = _mm_setzero_pd();
283 fjx1 = _mm_setzero_pd();
284 fjy1 = _mm_setzero_pd();
285 fjz1 = _mm_setzero_pd();
286 fjx2 = _mm_setzero_pd();
287 fjy2 = _mm_setzero_pd();
288 fjz2 = _mm_setzero_pd();
289 fjx3 = _mm_setzero_pd();
290 fjy3 = _mm_setzero_pd();
291 fjz3 = _mm_setzero_pd();
293 /**************************
294 * CALCULATE INTERACTIONS *
295 **************************/
297 r00 = _mm_mul_pd(rsq00,rinv00);
299 /* Calculate table index by multiplying r with table scale and truncate to integer */
300 rt = _mm_mul_pd(r00,vftabscale);
301 vfitab = _mm_cvttpd_epi32(rt);
303 vfeps = _mm_frcz_pd(rt);
305 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
307 twovfeps = _mm_add_pd(vfeps,vfeps);
308 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
310 /* CUBIC SPLINE TABLE DISPERSION */
311 vfitab = _mm_add_epi32(vfitab,ifour);
312 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
313 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
314 GMX_MM_TRANSPOSE2_PD(Y,F);
315 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
316 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
317 GMX_MM_TRANSPOSE2_PD(G,H);
318 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
319 VV = _mm_macc_pd(vfeps,Fp,Y);
320 vvdw6 = _mm_mul_pd(c6_00,VV);
321 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
322 fvdw6 = _mm_mul_pd(c6_00,FF);
324 /* CUBIC SPLINE TABLE REPULSION */
325 vfitab = _mm_add_epi32(vfitab,ifour);
326 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
327 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
328 GMX_MM_TRANSPOSE2_PD(Y,F);
329 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
330 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
331 GMX_MM_TRANSPOSE2_PD(G,H);
332 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
333 VV = _mm_macc_pd(vfeps,Fp,Y);
334 vvdw12 = _mm_mul_pd(c12_00,VV);
335 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
336 fvdw12 = _mm_mul_pd(c12_00,FF);
337 vvdw = _mm_add_pd(vvdw12,vvdw6);
338 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
345 /* Update vectorial force */
346 fix0 = _mm_macc_pd(dx00,fscal,fix0);
347 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
348 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
350 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
351 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
352 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
354 /**************************
355 * CALCULATE INTERACTIONS *
356 **************************/
358 r11 = _mm_mul_pd(rsq11,rinv11);
360 /* Calculate table index by multiplying r with table scale and truncate to integer */
361 rt = _mm_mul_pd(r11,vftabscale);
362 vfitab = _mm_cvttpd_epi32(rt);
364 vfeps = _mm_frcz_pd(rt);
366 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
368 twovfeps = _mm_add_pd(vfeps,vfeps);
369 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
371 /* CUBIC SPLINE TABLE ELECTROSTATICS */
372 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
373 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
374 GMX_MM_TRANSPOSE2_PD(Y,F);
375 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
376 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
377 GMX_MM_TRANSPOSE2_PD(G,H);
378 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
379 VV = _mm_macc_pd(vfeps,Fp,Y);
380 velec = _mm_mul_pd(qq11,VV);
381 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
382 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
384 /* Update potential sum for this i atom from the interaction with this j atom. */
385 velecsum = _mm_add_pd(velecsum,velec);
389 /* Update vectorial force */
390 fix1 = _mm_macc_pd(dx11,fscal,fix1);
391 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
392 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
394 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
395 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
396 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
398 /**************************
399 * CALCULATE INTERACTIONS *
400 **************************/
402 r12 = _mm_mul_pd(rsq12,rinv12);
404 /* Calculate table index by multiplying r with table scale and truncate to integer */
405 rt = _mm_mul_pd(r12,vftabscale);
406 vfitab = _mm_cvttpd_epi32(rt);
408 vfeps = _mm_frcz_pd(rt);
410 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
412 twovfeps = _mm_add_pd(vfeps,vfeps);
413 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
415 /* CUBIC SPLINE TABLE ELECTROSTATICS */
416 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
417 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
418 GMX_MM_TRANSPOSE2_PD(Y,F);
419 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
420 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
421 GMX_MM_TRANSPOSE2_PD(G,H);
422 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
423 VV = _mm_macc_pd(vfeps,Fp,Y);
424 velec = _mm_mul_pd(qq12,VV);
425 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
426 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
428 /* Update potential sum for this i atom from the interaction with this j atom. */
429 velecsum = _mm_add_pd(velecsum,velec);
433 /* Update vectorial force */
434 fix1 = _mm_macc_pd(dx12,fscal,fix1);
435 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
436 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
438 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
439 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
440 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
442 /**************************
443 * CALCULATE INTERACTIONS *
444 **************************/
446 r13 = _mm_mul_pd(rsq13,rinv13);
448 /* Calculate table index by multiplying r with table scale and truncate to integer */
449 rt = _mm_mul_pd(r13,vftabscale);
450 vfitab = _mm_cvttpd_epi32(rt);
452 vfeps = _mm_frcz_pd(rt);
454 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
456 twovfeps = _mm_add_pd(vfeps,vfeps);
457 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
459 /* CUBIC SPLINE TABLE ELECTROSTATICS */
460 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
461 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
462 GMX_MM_TRANSPOSE2_PD(Y,F);
463 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
464 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
465 GMX_MM_TRANSPOSE2_PD(G,H);
466 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
467 VV = _mm_macc_pd(vfeps,Fp,Y);
468 velec = _mm_mul_pd(qq13,VV);
469 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
470 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
472 /* Update potential sum for this i atom from the interaction with this j atom. */
473 velecsum = _mm_add_pd(velecsum,velec);
477 /* Update vectorial force */
478 fix1 = _mm_macc_pd(dx13,fscal,fix1);
479 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
480 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
482 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
483 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
484 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
486 /**************************
487 * CALCULATE INTERACTIONS *
488 **************************/
490 r21 = _mm_mul_pd(rsq21,rinv21);
492 /* Calculate table index by multiplying r with table scale and truncate to integer */
493 rt = _mm_mul_pd(r21,vftabscale);
494 vfitab = _mm_cvttpd_epi32(rt);
496 vfeps = _mm_frcz_pd(rt);
498 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
500 twovfeps = _mm_add_pd(vfeps,vfeps);
501 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
503 /* CUBIC SPLINE TABLE ELECTROSTATICS */
504 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
505 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
506 GMX_MM_TRANSPOSE2_PD(Y,F);
507 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
508 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
509 GMX_MM_TRANSPOSE2_PD(G,H);
510 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
511 VV = _mm_macc_pd(vfeps,Fp,Y);
512 velec = _mm_mul_pd(qq21,VV);
513 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
514 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
516 /* Update potential sum for this i atom from the interaction with this j atom. */
517 velecsum = _mm_add_pd(velecsum,velec);
521 /* Update vectorial force */
522 fix2 = _mm_macc_pd(dx21,fscal,fix2);
523 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
524 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
526 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
527 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
528 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
530 /**************************
531 * CALCULATE INTERACTIONS *
532 **************************/
534 r22 = _mm_mul_pd(rsq22,rinv22);
536 /* Calculate table index by multiplying r with table scale and truncate to integer */
537 rt = _mm_mul_pd(r22,vftabscale);
538 vfitab = _mm_cvttpd_epi32(rt);
540 vfeps = _mm_frcz_pd(rt);
542 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
544 twovfeps = _mm_add_pd(vfeps,vfeps);
545 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
547 /* CUBIC SPLINE TABLE ELECTROSTATICS */
548 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
549 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
550 GMX_MM_TRANSPOSE2_PD(Y,F);
551 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
552 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
553 GMX_MM_TRANSPOSE2_PD(G,H);
554 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
555 VV = _mm_macc_pd(vfeps,Fp,Y);
556 velec = _mm_mul_pd(qq22,VV);
557 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
558 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
560 /* Update potential sum for this i atom from the interaction with this j atom. */
561 velecsum = _mm_add_pd(velecsum,velec);
565 /* Update vectorial force */
566 fix2 = _mm_macc_pd(dx22,fscal,fix2);
567 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
568 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
570 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
571 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
572 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
574 /**************************
575 * CALCULATE INTERACTIONS *
576 **************************/
578 r23 = _mm_mul_pd(rsq23,rinv23);
580 /* Calculate table index by multiplying r with table scale and truncate to integer */
581 rt = _mm_mul_pd(r23,vftabscale);
582 vfitab = _mm_cvttpd_epi32(rt);
584 vfeps = _mm_frcz_pd(rt);
586 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
588 twovfeps = _mm_add_pd(vfeps,vfeps);
589 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
591 /* CUBIC SPLINE TABLE ELECTROSTATICS */
592 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
593 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
594 GMX_MM_TRANSPOSE2_PD(Y,F);
595 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
596 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
597 GMX_MM_TRANSPOSE2_PD(G,H);
598 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
599 VV = _mm_macc_pd(vfeps,Fp,Y);
600 velec = _mm_mul_pd(qq23,VV);
601 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
602 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
604 /* Update potential sum for this i atom from the interaction with this j atom. */
605 velecsum = _mm_add_pd(velecsum,velec);
609 /* Update vectorial force */
610 fix2 = _mm_macc_pd(dx23,fscal,fix2);
611 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
612 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
614 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
615 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
616 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
618 /**************************
619 * CALCULATE INTERACTIONS *
620 **************************/
622 r31 = _mm_mul_pd(rsq31,rinv31);
624 /* Calculate table index by multiplying r with table scale and truncate to integer */
625 rt = _mm_mul_pd(r31,vftabscale);
626 vfitab = _mm_cvttpd_epi32(rt);
628 vfeps = _mm_frcz_pd(rt);
630 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
632 twovfeps = _mm_add_pd(vfeps,vfeps);
633 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
635 /* CUBIC SPLINE TABLE ELECTROSTATICS */
636 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
637 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
638 GMX_MM_TRANSPOSE2_PD(Y,F);
639 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
640 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
641 GMX_MM_TRANSPOSE2_PD(G,H);
642 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
643 VV = _mm_macc_pd(vfeps,Fp,Y);
644 velec = _mm_mul_pd(qq31,VV);
645 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
646 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
648 /* Update potential sum for this i atom from the interaction with this j atom. */
649 velecsum = _mm_add_pd(velecsum,velec);
653 /* Update vectorial force */
654 fix3 = _mm_macc_pd(dx31,fscal,fix3);
655 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
656 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
658 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
659 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
660 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
662 /**************************
663 * CALCULATE INTERACTIONS *
664 **************************/
666 r32 = _mm_mul_pd(rsq32,rinv32);
668 /* Calculate table index by multiplying r with table scale and truncate to integer */
669 rt = _mm_mul_pd(r32,vftabscale);
670 vfitab = _mm_cvttpd_epi32(rt);
672 vfeps = _mm_frcz_pd(rt);
674 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
676 twovfeps = _mm_add_pd(vfeps,vfeps);
677 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
679 /* CUBIC SPLINE TABLE ELECTROSTATICS */
680 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
681 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
682 GMX_MM_TRANSPOSE2_PD(Y,F);
683 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
684 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
685 GMX_MM_TRANSPOSE2_PD(G,H);
686 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
687 VV = _mm_macc_pd(vfeps,Fp,Y);
688 velec = _mm_mul_pd(qq32,VV);
689 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
690 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
692 /* Update potential sum for this i atom from the interaction with this j atom. */
693 velecsum = _mm_add_pd(velecsum,velec);
697 /* Update vectorial force */
698 fix3 = _mm_macc_pd(dx32,fscal,fix3);
699 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
700 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
702 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
703 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
704 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
706 /**************************
707 * CALCULATE INTERACTIONS *
708 **************************/
710 r33 = _mm_mul_pd(rsq33,rinv33);
712 /* Calculate table index by multiplying r with table scale and truncate to integer */
713 rt = _mm_mul_pd(r33,vftabscale);
714 vfitab = _mm_cvttpd_epi32(rt);
716 vfeps = _mm_frcz_pd(rt);
718 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
720 twovfeps = _mm_add_pd(vfeps,vfeps);
721 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
723 /* CUBIC SPLINE TABLE ELECTROSTATICS */
724 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
725 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
726 GMX_MM_TRANSPOSE2_PD(Y,F);
727 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
728 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
729 GMX_MM_TRANSPOSE2_PD(G,H);
730 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
731 VV = _mm_macc_pd(vfeps,Fp,Y);
732 velec = _mm_mul_pd(qq33,VV);
733 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
734 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
736 /* Update potential sum for this i atom from the interaction with this j atom. */
737 velecsum = _mm_add_pd(velecsum,velec);
741 /* Update vectorial force */
742 fix3 = _mm_macc_pd(dx33,fscal,fix3);
743 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
744 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
746 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
747 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
748 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
750 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
752 /* Inner loop uses 476 flops */
759 j_coord_offsetA = DIM*jnrA;
761 /* load j atom coordinates */
762 gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
763 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
764 &jy2,&jz2,&jx3,&jy3,&jz3);
766 /* Calculate displacement vector */
767 dx00 = _mm_sub_pd(ix0,jx0);
768 dy00 = _mm_sub_pd(iy0,jy0);
769 dz00 = _mm_sub_pd(iz0,jz0);
770 dx11 = _mm_sub_pd(ix1,jx1);
771 dy11 = _mm_sub_pd(iy1,jy1);
772 dz11 = _mm_sub_pd(iz1,jz1);
773 dx12 = _mm_sub_pd(ix1,jx2);
774 dy12 = _mm_sub_pd(iy1,jy2);
775 dz12 = _mm_sub_pd(iz1,jz2);
776 dx13 = _mm_sub_pd(ix1,jx3);
777 dy13 = _mm_sub_pd(iy1,jy3);
778 dz13 = _mm_sub_pd(iz1,jz3);
779 dx21 = _mm_sub_pd(ix2,jx1);
780 dy21 = _mm_sub_pd(iy2,jy1);
781 dz21 = _mm_sub_pd(iz2,jz1);
782 dx22 = _mm_sub_pd(ix2,jx2);
783 dy22 = _mm_sub_pd(iy2,jy2);
784 dz22 = _mm_sub_pd(iz2,jz2);
785 dx23 = _mm_sub_pd(ix2,jx3);
786 dy23 = _mm_sub_pd(iy2,jy3);
787 dz23 = _mm_sub_pd(iz2,jz3);
788 dx31 = _mm_sub_pd(ix3,jx1);
789 dy31 = _mm_sub_pd(iy3,jy1);
790 dz31 = _mm_sub_pd(iz3,jz1);
791 dx32 = _mm_sub_pd(ix3,jx2);
792 dy32 = _mm_sub_pd(iy3,jy2);
793 dz32 = _mm_sub_pd(iz3,jz2);
794 dx33 = _mm_sub_pd(ix3,jx3);
795 dy33 = _mm_sub_pd(iy3,jy3);
796 dz33 = _mm_sub_pd(iz3,jz3);
798 /* Calculate squared distance and things based on it */
799 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
800 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
801 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
802 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
803 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
804 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
805 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
806 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
807 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
808 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
810 rinv00 = gmx_mm_invsqrt_pd(rsq00);
811 rinv11 = gmx_mm_invsqrt_pd(rsq11);
812 rinv12 = gmx_mm_invsqrt_pd(rsq12);
813 rinv13 = gmx_mm_invsqrt_pd(rsq13);
814 rinv21 = gmx_mm_invsqrt_pd(rsq21);
815 rinv22 = gmx_mm_invsqrt_pd(rsq22);
816 rinv23 = gmx_mm_invsqrt_pd(rsq23);
817 rinv31 = gmx_mm_invsqrt_pd(rsq31);
818 rinv32 = gmx_mm_invsqrt_pd(rsq32);
819 rinv33 = gmx_mm_invsqrt_pd(rsq33);
821 fjx0 = _mm_setzero_pd();
822 fjy0 = _mm_setzero_pd();
823 fjz0 = _mm_setzero_pd();
824 fjx1 = _mm_setzero_pd();
825 fjy1 = _mm_setzero_pd();
826 fjz1 = _mm_setzero_pd();
827 fjx2 = _mm_setzero_pd();
828 fjy2 = _mm_setzero_pd();
829 fjz2 = _mm_setzero_pd();
830 fjx3 = _mm_setzero_pd();
831 fjy3 = _mm_setzero_pd();
832 fjz3 = _mm_setzero_pd();
834 /**************************
835 * CALCULATE INTERACTIONS *
836 **************************/
838 r00 = _mm_mul_pd(rsq00,rinv00);
840 /* Calculate table index by multiplying r with table scale and truncate to integer */
841 rt = _mm_mul_pd(r00,vftabscale);
842 vfitab = _mm_cvttpd_epi32(rt);
844 vfeps = _mm_frcz_pd(rt);
846 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
848 twovfeps = _mm_add_pd(vfeps,vfeps);
849 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
851 /* CUBIC SPLINE TABLE DISPERSION */
852 vfitab = _mm_add_epi32(vfitab,ifour);
853 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
854 F = _mm_setzero_pd();
855 GMX_MM_TRANSPOSE2_PD(Y,F);
856 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
857 H = _mm_setzero_pd();
858 GMX_MM_TRANSPOSE2_PD(G,H);
859 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
860 VV = _mm_macc_pd(vfeps,Fp,Y);
861 vvdw6 = _mm_mul_pd(c6_00,VV);
862 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
863 fvdw6 = _mm_mul_pd(c6_00,FF);
865 /* CUBIC SPLINE TABLE REPULSION */
866 vfitab = _mm_add_epi32(vfitab,ifour);
867 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
868 F = _mm_setzero_pd();
869 GMX_MM_TRANSPOSE2_PD(Y,F);
870 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
871 H = _mm_setzero_pd();
872 GMX_MM_TRANSPOSE2_PD(G,H);
873 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
874 VV = _mm_macc_pd(vfeps,Fp,Y);
875 vvdw12 = _mm_mul_pd(c12_00,VV);
876 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
877 fvdw12 = _mm_mul_pd(c12_00,FF);
878 vvdw = _mm_add_pd(vvdw12,vvdw6);
879 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
881 /* Update potential sum for this i atom from the interaction with this j atom. */
882 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
883 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
887 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
889 /* Update vectorial force */
890 fix0 = _mm_macc_pd(dx00,fscal,fix0);
891 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
892 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
894 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
895 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
896 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
898 /**************************
899 * CALCULATE INTERACTIONS *
900 **************************/
902 r11 = _mm_mul_pd(rsq11,rinv11);
904 /* Calculate table index by multiplying r with table scale and truncate to integer */
905 rt = _mm_mul_pd(r11,vftabscale);
906 vfitab = _mm_cvttpd_epi32(rt);
908 vfeps = _mm_frcz_pd(rt);
910 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
912 twovfeps = _mm_add_pd(vfeps,vfeps);
913 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
915 /* CUBIC SPLINE TABLE ELECTROSTATICS */
916 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
917 F = _mm_setzero_pd();
918 GMX_MM_TRANSPOSE2_PD(Y,F);
919 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
920 H = _mm_setzero_pd();
921 GMX_MM_TRANSPOSE2_PD(G,H);
922 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
923 VV = _mm_macc_pd(vfeps,Fp,Y);
924 velec = _mm_mul_pd(qq11,VV);
925 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
926 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
928 /* Update potential sum for this i atom from the interaction with this j atom. */
929 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
930 velecsum = _mm_add_pd(velecsum,velec);
934 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
936 /* Update vectorial force */
937 fix1 = _mm_macc_pd(dx11,fscal,fix1);
938 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
939 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
941 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
942 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
943 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
945 /**************************
946 * CALCULATE INTERACTIONS *
947 **************************/
949 r12 = _mm_mul_pd(rsq12,rinv12);
951 /* Calculate table index by multiplying r with table scale and truncate to integer */
952 rt = _mm_mul_pd(r12,vftabscale);
953 vfitab = _mm_cvttpd_epi32(rt);
955 vfeps = _mm_frcz_pd(rt);
957 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
959 twovfeps = _mm_add_pd(vfeps,vfeps);
960 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
962 /* CUBIC SPLINE TABLE ELECTROSTATICS */
963 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
964 F = _mm_setzero_pd();
965 GMX_MM_TRANSPOSE2_PD(Y,F);
966 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
967 H = _mm_setzero_pd();
968 GMX_MM_TRANSPOSE2_PD(G,H);
969 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
970 VV = _mm_macc_pd(vfeps,Fp,Y);
971 velec = _mm_mul_pd(qq12,VV);
972 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
973 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
975 /* Update potential sum for this i atom from the interaction with this j atom. */
976 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
977 velecsum = _mm_add_pd(velecsum,velec);
981 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
983 /* Update vectorial force */
984 fix1 = _mm_macc_pd(dx12,fscal,fix1);
985 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
986 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
988 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
989 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
990 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
992 /**************************
993 * CALCULATE INTERACTIONS *
994 **************************/
996 r13 = _mm_mul_pd(rsq13,rinv13);
998 /* Calculate table index by multiplying r with table scale and truncate to integer */
999 rt = _mm_mul_pd(r13,vftabscale);
1000 vfitab = _mm_cvttpd_epi32(rt);
1002 vfeps = _mm_frcz_pd(rt);
1004 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1006 twovfeps = _mm_add_pd(vfeps,vfeps);
1007 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1009 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1010 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1011 F = _mm_setzero_pd();
1012 GMX_MM_TRANSPOSE2_PD(Y,F);
1013 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1014 H = _mm_setzero_pd();
1015 GMX_MM_TRANSPOSE2_PD(G,H);
1016 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1017 VV = _mm_macc_pd(vfeps,Fp,Y);
1018 velec = _mm_mul_pd(qq13,VV);
1019 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1020 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
1022 /* Update potential sum for this i atom from the interaction with this j atom. */
1023 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1024 velecsum = _mm_add_pd(velecsum,velec);
1028 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1030 /* Update vectorial force */
1031 fix1 = _mm_macc_pd(dx13,fscal,fix1);
1032 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
1033 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1035 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1036 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1037 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1039 /**************************
1040 * CALCULATE INTERACTIONS *
1041 **************************/
1043 r21 = _mm_mul_pd(rsq21,rinv21);
1045 /* Calculate table index by multiplying r with table scale and truncate to integer */
1046 rt = _mm_mul_pd(r21,vftabscale);
1047 vfitab = _mm_cvttpd_epi32(rt);
1049 vfeps = _mm_frcz_pd(rt);
1051 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1053 twovfeps = _mm_add_pd(vfeps,vfeps);
1054 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1056 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1057 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1058 F = _mm_setzero_pd();
1059 GMX_MM_TRANSPOSE2_PD(Y,F);
1060 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1061 H = _mm_setzero_pd();
1062 GMX_MM_TRANSPOSE2_PD(G,H);
1063 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1064 VV = _mm_macc_pd(vfeps,Fp,Y);
1065 velec = _mm_mul_pd(qq21,VV);
1066 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1067 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1069 /* Update potential sum for this i atom from the interaction with this j atom. */
1070 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1071 velecsum = _mm_add_pd(velecsum,velec);
1075 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1077 /* Update vectorial force */
1078 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1079 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1080 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1082 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1083 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1084 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1086 /**************************
1087 * CALCULATE INTERACTIONS *
1088 **************************/
1090 r22 = _mm_mul_pd(rsq22,rinv22);
1092 /* Calculate table index by multiplying r with table scale and truncate to integer */
1093 rt = _mm_mul_pd(r22,vftabscale);
1094 vfitab = _mm_cvttpd_epi32(rt);
1096 vfeps = _mm_frcz_pd(rt);
1098 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1100 twovfeps = _mm_add_pd(vfeps,vfeps);
1101 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1103 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1104 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1105 F = _mm_setzero_pd();
1106 GMX_MM_TRANSPOSE2_PD(Y,F);
1107 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1108 H = _mm_setzero_pd();
1109 GMX_MM_TRANSPOSE2_PD(G,H);
1110 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1111 VV = _mm_macc_pd(vfeps,Fp,Y);
1112 velec = _mm_mul_pd(qq22,VV);
1113 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1114 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1116 /* Update potential sum for this i atom from the interaction with this j atom. */
1117 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1118 velecsum = _mm_add_pd(velecsum,velec);
1122 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1124 /* Update vectorial force */
1125 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1126 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1127 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1129 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1130 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1131 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1133 /**************************
1134 * CALCULATE INTERACTIONS *
1135 **************************/
1137 r23 = _mm_mul_pd(rsq23,rinv23);
1139 /* Calculate table index by multiplying r with table scale and truncate to integer */
1140 rt = _mm_mul_pd(r23,vftabscale);
1141 vfitab = _mm_cvttpd_epi32(rt);
1143 vfeps = _mm_frcz_pd(rt);
1145 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1147 twovfeps = _mm_add_pd(vfeps,vfeps);
1148 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1150 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1151 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1152 F = _mm_setzero_pd();
1153 GMX_MM_TRANSPOSE2_PD(Y,F);
1154 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1155 H = _mm_setzero_pd();
1156 GMX_MM_TRANSPOSE2_PD(G,H);
1157 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1158 VV = _mm_macc_pd(vfeps,Fp,Y);
1159 velec = _mm_mul_pd(qq23,VV);
1160 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1161 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
1163 /* Update potential sum for this i atom from the interaction with this j atom. */
1164 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1165 velecsum = _mm_add_pd(velecsum,velec);
1169 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1171 /* Update vectorial force */
1172 fix2 = _mm_macc_pd(dx23,fscal,fix2);
1173 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
1174 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
1176 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
1177 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
1178 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
1180 /**************************
1181 * CALCULATE INTERACTIONS *
1182 **************************/
1184 r31 = _mm_mul_pd(rsq31,rinv31);
1186 /* Calculate table index by multiplying r with table scale and truncate to integer */
1187 rt = _mm_mul_pd(r31,vftabscale);
1188 vfitab = _mm_cvttpd_epi32(rt);
1190 vfeps = _mm_frcz_pd(rt);
1192 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1194 twovfeps = _mm_add_pd(vfeps,vfeps);
1195 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1197 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1198 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1199 F = _mm_setzero_pd();
1200 GMX_MM_TRANSPOSE2_PD(Y,F);
1201 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1202 H = _mm_setzero_pd();
1203 GMX_MM_TRANSPOSE2_PD(G,H);
1204 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1205 VV = _mm_macc_pd(vfeps,Fp,Y);
1206 velec = _mm_mul_pd(qq31,VV);
1207 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1208 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
1210 /* Update potential sum for this i atom from the interaction with this j atom. */
1211 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1212 velecsum = _mm_add_pd(velecsum,velec);
1216 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1218 /* Update vectorial force */
1219 fix3 = _mm_macc_pd(dx31,fscal,fix3);
1220 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
1221 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
1223 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
1224 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
1225 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
1227 /**************************
1228 * CALCULATE INTERACTIONS *
1229 **************************/
1231 r32 = _mm_mul_pd(rsq32,rinv32);
1233 /* Calculate table index by multiplying r with table scale and truncate to integer */
1234 rt = _mm_mul_pd(r32,vftabscale);
1235 vfitab = _mm_cvttpd_epi32(rt);
1237 vfeps = _mm_frcz_pd(rt);
1239 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1241 twovfeps = _mm_add_pd(vfeps,vfeps);
1242 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1244 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1245 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1246 F = _mm_setzero_pd();
1247 GMX_MM_TRANSPOSE2_PD(Y,F);
1248 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1249 H = _mm_setzero_pd();
1250 GMX_MM_TRANSPOSE2_PD(G,H);
1251 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1252 VV = _mm_macc_pd(vfeps,Fp,Y);
1253 velec = _mm_mul_pd(qq32,VV);
1254 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1255 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
1257 /* Update potential sum for this i atom from the interaction with this j atom. */
1258 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1259 velecsum = _mm_add_pd(velecsum,velec);
1263 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1265 /* Update vectorial force */
1266 fix3 = _mm_macc_pd(dx32,fscal,fix3);
1267 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
1268 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
1270 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
1271 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
1272 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
1274 /**************************
1275 * CALCULATE INTERACTIONS *
1276 **************************/
1278 r33 = _mm_mul_pd(rsq33,rinv33);
1280 /* Calculate table index by multiplying r with table scale and truncate to integer */
1281 rt = _mm_mul_pd(r33,vftabscale);
1282 vfitab = _mm_cvttpd_epi32(rt);
1284 vfeps = _mm_frcz_pd(rt);
1286 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1288 twovfeps = _mm_add_pd(vfeps,vfeps);
1289 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1291 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1292 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1293 F = _mm_setzero_pd();
1294 GMX_MM_TRANSPOSE2_PD(Y,F);
1295 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1296 H = _mm_setzero_pd();
1297 GMX_MM_TRANSPOSE2_PD(G,H);
1298 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1299 VV = _mm_macc_pd(vfeps,Fp,Y);
1300 velec = _mm_mul_pd(qq33,VV);
1301 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1302 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
1304 /* Update potential sum for this i atom from the interaction with this j atom. */
1305 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1306 velecsum = _mm_add_pd(velecsum,velec);
1310 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1312 /* Update vectorial force */
1313 fix3 = _mm_macc_pd(dx33,fscal,fix3);
1314 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
1315 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
1317 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
1318 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
1319 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
1321 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1323 /* Inner loop uses 476 flops */
1326 /* End of innermost loop */
1328 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1329 f+i_coord_offset,fshift+i_shift_offset);
1332 /* Update potential energies */
1333 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1334 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1336 /* Increment number of inner iterations */
1337 inneriter += j_index_end - j_index_start;
1339 /* Outer loop uses 26 flops */
1342 /* Increment number of outer iterations */
1345 /* Update outer/inner flops */
1347 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*476);
1350 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_F_avx_128_fma_double
1351 * Electrostatics interaction: CubicSplineTable
1352 * VdW interaction: CubicSplineTable
1353 * Geometry: Water4-Water4
1354 * Calculate force/pot: Force
1357 nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_F_avx_128_fma_double
1358 (t_nblist * gmx_restrict nlist,
1359 rvec * gmx_restrict xx,
1360 rvec * gmx_restrict ff,
1361 t_forcerec * gmx_restrict fr,
1362 t_mdatoms * gmx_restrict mdatoms,
1363 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1364 t_nrnb * gmx_restrict nrnb)
1366 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1367 * just 0 for non-waters.
1368 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1369 * jnr indices corresponding to data put in the four positions in the SIMD register.
1371 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1372 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1374 int j_coord_offsetA,j_coord_offsetB;
1375 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1376 real rcutoff_scalar;
1377 real *shiftvec,*fshift,*x,*f;
1378 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1380 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1382 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1384 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1386 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1387 int vdwjidx0A,vdwjidx0B;
1388 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1389 int vdwjidx1A,vdwjidx1B;
1390 __m128d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1391 int vdwjidx2A,vdwjidx2B;
1392 __m128d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1393 int vdwjidx3A,vdwjidx3B;
1394 __m128d jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1395 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1396 __m128d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1397 __m128d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1398 __m128d dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1399 __m128d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1400 __m128d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1401 __m128d dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1402 __m128d dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1403 __m128d dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1404 __m128d dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1405 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
1408 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1411 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
1412 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
1414 __m128i ifour = _mm_set1_epi32(4);
1415 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1417 __m128d dummy_mask,cutoff_mask;
1418 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1419 __m128d one = _mm_set1_pd(1.0);
1420 __m128d two = _mm_set1_pd(2.0);
1426 jindex = nlist->jindex;
1428 shiftidx = nlist->shift;
1430 shiftvec = fr->shift_vec[0];
1431 fshift = fr->fshift[0];
1432 facel = _mm_set1_pd(fr->epsfac);
1433 charge = mdatoms->chargeA;
1434 nvdwtype = fr->ntype;
1435 vdwparam = fr->nbfp;
1436 vdwtype = mdatoms->typeA;
1438 vftab = kernel_data->table_elec_vdw->data;
1439 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
1441 /* Setup water-specific parameters */
1442 inr = nlist->iinr[0];
1443 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1444 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1445 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1446 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1448 jq1 = _mm_set1_pd(charge[inr+1]);
1449 jq2 = _mm_set1_pd(charge[inr+2]);
1450 jq3 = _mm_set1_pd(charge[inr+3]);
1451 vdwjidx0A = 2*vdwtype[inr+0];
1452 c6_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1453 c12_00 = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1454 qq11 = _mm_mul_pd(iq1,jq1);
1455 qq12 = _mm_mul_pd(iq1,jq2);
1456 qq13 = _mm_mul_pd(iq1,jq3);
1457 qq21 = _mm_mul_pd(iq2,jq1);
1458 qq22 = _mm_mul_pd(iq2,jq2);
1459 qq23 = _mm_mul_pd(iq2,jq3);
1460 qq31 = _mm_mul_pd(iq3,jq1);
1461 qq32 = _mm_mul_pd(iq3,jq2);
1462 qq33 = _mm_mul_pd(iq3,jq3);
1464 /* Avoid stupid compiler warnings */
1466 j_coord_offsetA = 0;
1467 j_coord_offsetB = 0;
1472 /* Start outer loop over neighborlists */
1473 for(iidx=0; iidx<nri; iidx++)
1475 /* Load shift vector for this list */
1476 i_shift_offset = DIM*shiftidx[iidx];
1478 /* Load limits for loop over neighbors */
1479 j_index_start = jindex[iidx];
1480 j_index_end = jindex[iidx+1];
1482 /* Get outer coordinate index */
1484 i_coord_offset = DIM*inr;
1486 /* Load i particle coords and add shift vector */
1487 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1488 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1490 fix0 = _mm_setzero_pd();
1491 fiy0 = _mm_setzero_pd();
1492 fiz0 = _mm_setzero_pd();
1493 fix1 = _mm_setzero_pd();
1494 fiy1 = _mm_setzero_pd();
1495 fiz1 = _mm_setzero_pd();
1496 fix2 = _mm_setzero_pd();
1497 fiy2 = _mm_setzero_pd();
1498 fiz2 = _mm_setzero_pd();
1499 fix3 = _mm_setzero_pd();
1500 fiy3 = _mm_setzero_pd();
1501 fiz3 = _mm_setzero_pd();
1503 /* Start inner kernel loop */
1504 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1507 /* Get j neighbor index, and coordinate index */
1509 jnrB = jjnr[jidx+1];
1510 j_coord_offsetA = DIM*jnrA;
1511 j_coord_offsetB = DIM*jnrB;
1513 /* load j atom coordinates */
1514 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1515 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1516 &jy2,&jz2,&jx3,&jy3,&jz3);
1518 /* Calculate displacement vector */
1519 dx00 = _mm_sub_pd(ix0,jx0);
1520 dy00 = _mm_sub_pd(iy0,jy0);
1521 dz00 = _mm_sub_pd(iz0,jz0);
1522 dx11 = _mm_sub_pd(ix1,jx1);
1523 dy11 = _mm_sub_pd(iy1,jy1);
1524 dz11 = _mm_sub_pd(iz1,jz1);
1525 dx12 = _mm_sub_pd(ix1,jx2);
1526 dy12 = _mm_sub_pd(iy1,jy2);
1527 dz12 = _mm_sub_pd(iz1,jz2);
1528 dx13 = _mm_sub_pd(ix1,jx3);
1529 dy13 = _mm_sub_pd(iy1,jy3);
1530 dz13 = _mm_sub_pd(iz1,jz3);
1531 dx21 = _mm_sub_pd(ix2,jx1);
1532 dy21 = _mm_sub_pd(iy2,jy1);
1533 dz21 = _mm_sub_pd(iz2,jz1);
1534 dx22 = _mm_sub_pd(ix2,jx2);
1535 dy22 = _mm_sub_pd(iy2,jy2);
1536 dz22 = _mm_sub_pd(iz2,jz2);
1537 dx23 = _mm_sub_pd(ix2,jx3);
1538 dy23 = _mm_sub_pd(iy2,jy3);
1539 dz23 = _mm_sub_pd(iz2,jz3);
1540 dx31 = _mm_sub_pd(ix3,jx1);
1541 dy31 = _mm_sub_pd(iy3,jy1);
1542 dz31 = _mm_sub_pd(iz3,jz1);
1543 dx32 = _mm_sub_pd(ix3,jx2);
1544 dy32 = _mm_sub_pd(iy3,jy2);
1545 dz32 = _mm_sub_pd(iz3,jz2);
1546 dx33 = _mm_sub_pd(ix3,jx3);
1547 dy33 = _mm_sub_pd(iy3,jy3);
1548 dz33 = _mm_sub_pd(iz3,jz3);
1550 /* Calculate squared distance and things based on it */
1551 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1552 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1553 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1554 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1555 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1556 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1557 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1558 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1559 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1560 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1562 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1563 rinv11 = gmx_mm_invsqrt_pd(rsq11);
1564 rinv12 = gmx_mm_invsqrt_pd(rsq12);
1565 rinv13 = gmx_mm_invsqrt_pd(rsq13);
1566 rinv21 = gmx_mm_invsqrt_pd(rsq21);
1567 rinv22 = gmx_mm_invsqrt_pd(rsq22);
1568 rinv23 = gmx_mm_invsqrt_pd(rsq23);
1569 rinv31 = gmx_mm_invsqrt_pd(rsq31);
1570 rinv32 = gmx_mm_invsqrt_pd(rsq32);
1571 rinv33 = gmx_mm_invsqrt_pd(rsq33);
1573 fjx0 = _mm_setzero_pd();
1574 fjy0 = _mm_setzero_pd();
1575 fjz0 = _mm_setzero_pd();
1576 fjx1 = _mm_setzero_pd();
1577 fjy1 = _mm_setzero_pd();
1578 fjz1 = _mm_setzero_pd();
1579 fjx2 = _mm_setzero_pd();
1580 fjy2 = _mm_setzero_pd();
1581 fjz2 = _mm_setzero_pd();
1582 fjx3 = _mm_setzero_pd();
1583 fjy3 = _mm_setzero_pd();
1584 fjz3 = _mm_setzero_pd();
1586 /**************************
1587 * CALCULATE INTERACTIONS *
1588 **************************/
1590 r00 = _mm_mul_pd(rsq00,rinv00);
1592 /* Calculate table index by multiplying r with table scale and truncate to integer */
1593 rt = _mm_mul_pd(r00,vftabscale);
1594 vfitab = _mm_cvttpd_epi32(rt);
1596 vfeps = _mm_frcz_pd(rt);
1598 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1600 twovfeps = _mm_add_pd(vfeps,vfeps);
1601 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1603 /* CUBIC SPLINE TABLE DISPERSION */
1604 vfitab = _mm_add_epi32(vfitab,ifour);
1605 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1606 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1607 GMX_MM_TRANSPOSE2_PD(Y,F);
1608 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1609 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1610 GMX_MM_TRANSPOSE2_PD(G,H);
1611 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1612 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1613 fvdw6 = _mm_mul_pd(c6_00,FF);
1615 /* CUBIC SPLINE TABLE REPULSION */
1616 vfitab = _mm_add_epi32(vfitab,ifour);
1617 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1618 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1619 GMX_MM_TRANSPOSE2_PD(Y,F);
1620 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1621 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1622 GMX_MM_TRANSPOSE2_PD(G,H);
1623 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1624 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1625 fvdw12 = _mm_mul_pd(c12_00,FF);
1626 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1630 /* Update vectorial force */
1631 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1632 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1633 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1635 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1636 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1637 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1639 /**************************
1640 * CALCULATE INTERACTIONS *
1641 **************************/
1643 r11 = _mm_mul_pd(rsq11,rinv11);
1645 /* Calculate table index by multiplying r with table scale and truncate to integer */
1646 rt = _mm_mul_pd(r11,vftabscale);
1647 vfitab = _mm_cvttpd_epi32(rt);
1649 vfeps = _mm_frcz_pd(rt);
1651 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1653 twovfeps = _mm_add_pd(vfeps,vfeps);
1654 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1656 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1657 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1658 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1659 GMX_MM_TRANSPOSE2_PD(Y,F);
1660 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1661 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1662 GMX_MM_TRANSPOSE2_PD(G,H);
1663 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1664 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1665 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
1669 /* Update vectorial force */
1670 fix1 = _mm_macc_pd(dx11,fscal,fix1);
1671 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
1672 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
1674 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
1675 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
1676 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
1678 /**************************
1679 * CALCULATE INTERACTIONS *
1680 **************************/
1682 r12 = _mm_mul_pd(rsq12,rinv12);
1684 /* Calculate table index by multiplying r with table scale and truncate to integer */
1685 rt = _mm_mul_pd(r12,vftabscale);
1686 vfitab = _mm_cvttpd_epi32(rt);
1688 vfeps = _mm_frcz_pd(rt);
1690 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1692 twovfeps = _mm_add_pd(vfeps,vfeps);
1693 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1695 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1696 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1697 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1698 GMX_MM_TRANSPOSE2_PD(Y,F);
1699 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1700 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1701 GMX_MM_TRANSPOSE2_PD(G,H);
1702 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1703 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1704 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1708 /* Update vectorial force */
1709 fix1 = _mm_macc_pd(dx12,fscal,fix1);
1710 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
1711 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
1713 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
1714 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
1715 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
1717 /**************************
1718 * CALCULATE INTERACTIONS *
1719 **************************/
1721 r13 = _mm_mul_pd(rsq13,rinv13);
1723 /* Calculate table index by multiplying r with table scale and truncate to integer */
1724 rt = _mm_mul_pd(r13,vftabscale);
1725 vfitab = _mm_cvttpd_epi32(rt);
1727 vfeps = _mm_frcz_pd(rt);
1729 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1731 twovfeps = _mm_add_pd(vfeps,vfeps);
1732 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1734 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1735 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1736 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1737 GMX_MM_TRANSPOSE2_PD(Y,F);
1738 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1739 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1740 GMX_MM_TRANSPOSE2_PD(G,H);
1741 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1742 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1743 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
1747 /* Update vectorial force */
1748 fix1 = _mm_macc_pd(dx13,fscal,fix1);
1749 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
1750 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
1752 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
1753 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
1754 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
1756 /**************************
1757 * CALCULATE INTERACTIONS *
1758 **************************/
1760 r21 = _mm_mul_pd(rsq21,rinv21);
1762 /* Calculate table index by multiplying r with table scale and truncate to integer */
1763 rt = _mm_mul_pd(r21,vftabscale);
1764 vfitab = _mm_cvttpd_epi32(rt);
1766 vfeps = _mm_frcz_pd(rt);
1768 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1770 twovfeps = _mm_add_pd(vfeps,vfeps);
1771 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1773 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1774 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1775 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1776 GMX_MM_TRANSPOSE2_PD(Y,F);
1777 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1778 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1779 GMX_MM_TRANSPOSE2_PD(G,H);
1780 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1781 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1782 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1786 /* Update vectorial force */
1787 fix2 = _mm_macc_pd(dx21,fscal,fix2);
1788 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
1789 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
1791 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
1792 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
1793 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
1795 /**************************
1796 * CALCULATE INTERACTIONS *
1797 **************************/
1799 r22 = _mm_mul_pd(rsq22,rinv22);
1801 /* Calculate table index by multiplying r with table scale and truncate to integer */
1802 rt = _mm_mul_pd(r22,vftabscale);
1803 vfitab = _mm_cvttpd_epi32(rt);
1805 vfeps = _mm_frcz_pd(rt);
1807 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1809 twovfeps = _mm_add_pd(vfeps,vfeps);
1810 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1812 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1813 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1814 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1815 GMX_MM_TRANSPOSE2_PD(Y,F);
1816 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1817 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1818 GMX_MM_TRANSPOSE2_PD(G,H);
1819 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1820 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1821 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1825 /* Update vectorial force */
1826 fix2 = _mm_macc_pd(dx22,fscal,fix2);
1827 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
1828 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
1830 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
1831 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
1832 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
1834 /**************************
1835 * CALCULATE INTERACTIONS *
1836 **************************/
1838 r23 = _mm_mul_pd(rsq23,rinv23);
1840 /* Calculate table index by multiplying r with table scale and truncate to integer */
1841 rt = _mm_mul_pd(r23,vftabscale);
1842 vfitab = _mm_cvttpd_epi32(rt);
1844 vfeps = _mm_frcz_pd(rt);
1846 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1848 twovfeps = _mm_add_pd(vfeps,vfeps);
1849 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1851 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1852 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1853 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1854 GMX_MM_TRANSPOSE2_PD(Y,F);
1855 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1856 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1857 GMX_MM_TRANSPOSE2_PD(G,H);
1858 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1859 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1860 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
1864 /* Update vectorial force */
1865 fix2 = _mm_macc_pd(dx23,fscal,fix2);
1866 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
1867 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
1869 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
1870 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
1871 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
1873 /**************************
1874 * CALCULATE INTERACTIONS *
1875 **************************/
1877 r31 = _mm_mul_pd(rsq31,rinv31);
1879 /* Calculate table index by multiplying r with table scale and truncate to integer */
1880 rt = _mm_mul_pd(r31,vftabscale);
1881 vfitab = _mm_cvttpd_epi32(rt);
1883 vfeps = _mm_frcz_pd(rt);
1885 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1887 twovfeps = _mm_add_pd(vfeps,vfeps);
1888 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1890 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1891 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1892 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1893 GMX_MM_TRANSPOSE2_PD(Y,F);
1894 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1895 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1896 GMX_MM_TRANSPOSE2_PD(G,H);
1897 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1898 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1899 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
1903 /* Update vectorial force */
1904 fix3 = _mm_macc_pd(dx31,fscal,fix3);
1905 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
1906 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
1908 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
1909 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
1910 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
1912 /**************************
1913 * CALCULATE INTERACTIONS *
1914 **************************/
1916 r32 = _mm_mul_pd(rsq32,rinv32);
1918 /* Calculate table index by multiplying r with table scale and truncate to integer */
1919 rt = _mm_mul_pd(r32,vftabscale);
1920 vfitab = _mm_cvttpd_epi32(rt);
1922 vfeps = _mm_frcz_pd(rt);
1924 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1926 twovfeps = _mm_add_pd(vfeps,vfeps);
1927 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1929 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1930 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1931 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1932 GMX_MM_TRANSPOSE2_PD(Y,F);
1933 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1934 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1935 GMX_MM_TRANSPOSE2_PD(G,H);
1936 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1937 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1938 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
1942 /* Update vectorial force */
1943 fix3 = _mm_macc_pd(dx32,fscal,fix3);
1944 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
1945 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
1947 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
1948 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
1949 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
1951 /**************************
1952 * CALCULATE INTERACTIONS *
1953 **************************/
1955 r33 = _mm_mul_pd(rsq33,rinv33);
1957 /* Calculate table index by multiplying r with table scale and truncate to integer */
1958 rt = _mm_mul_pd(r33,vftabscale);
1959 vfitab = _mm_cvttpd_epi32(rt);
1961 vfeps = _mm_frcz_pd(rt);
1963 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1965 twovfeps = _mm_add_pd(vfeps,vfeps);
1966 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1968 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1969 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1970 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1971 GMX_MM_TRANSPOSE2_PD(Y,F);
1972 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1973 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1974 GMX_MM_TRANSPOSE2_PD(G,H);
1975 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1976 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1977 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
1981 /* Update vectorial force */
1982 fix3 = _mm_macc_pd(dx33,fscal,fix3);
1983 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
1984 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
1986 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
1987 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
1988 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
1990 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1992 /* Inner loop uses 432 flops */
1995 if(jidx<j_index_end)
1999 j_coord_offsetA = DIM*jnrA;
2001 /* load j atom coordinates */
2002 gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2003 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2004 &jy2,&jz2,&jx3,&jy3,&jz3);
2006 /* Calculate displacement vector */
2007 dx00 = _mm_sub_pd(ix0,jx0);
2008 dy00 = _mm_sub_pd(iy0,jy0);
2009 dz00 = _mm_sub_pd(iz0,jz0);
2010 dx11 = _mm_sub_pd(ix1,jx1);
2011 dy11 = _mm_sub_pd(iy1,jy1);
2012 dz11 = _mm_sub_pd(iz1,jz1);
2013 dx12 = _mm_sub_pd(ix1,jx2);
2014 dy12 = _mm_sub_pd(iy1,jy2);
2015 dz12 = _mm_sub_pd(iz1,jz2);
2016 dx13 = _mm_sub_pd(ix1,jx3);
2017 dy13 = _mm_sub_pd(iy1,jy3);
2018 dz13 = _mm_sub_pd(iz1,jz3);
2019 dx21 = _mm_sub_pd(ix2,jx1);
2020 dy21 = _mm_sub_pd(iy2,jy1);
2021 dz21 = _mm_sub_pd(iz2,jz1);
2022 dx22 = _mm_sub_pd(ix2,jx2);
2023 dy22 = _mm_sub_pd(iy2,jy2);
2024 dz22 = _mm_sub_pd(iz2,jz2);
2025 dx23 = _mm_sub_pd(ix2,jx3);
2026 dy23 = _mm_sub_pd(iy2,jy3);
2027 dz23 = _mm_sub_pd(iz2,jz3);
2028 dx31 = _mm_sub_pd(ix3,jx1);
2029 dy31 = _mm_sub_pd(iy3,jy1);
2030 dz31 = _mm_sub_pd(iz3,jz1);
2031 dx32 = _mm_sub_pd(ix3,jx2);
2032 dy32 = _mm_sub_pd(iy3,jy2);
2033 dz32 = _mm_sub_pd(iz3,jz2);
2034 dx33 = _mm_sub_pd(ix3,jx3);
2035 dy33 = _mm_sub_pd(iy3,jy3);
2036 dz33 = _mm_sub_pd(iz3,jz3);
2038 /* Calculate squared distance and things based on it */
2039 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2040 rsq11 = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2041 rsq12 = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2042 rsq13 = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
2043 rsq21 = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2044 rsq22 = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2045 rsq23 = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
2046 rsq31 = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
2047 rsq32 = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
2048 rsq33 = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
2050 rinv00 = gmx_mm_invsqrt_pd(rsq00);
2051 rinv11 = gmx_mm_invsqrt_pd(rsq11);
2052 rinv12 = gmx_mm_invsqrt_pd(rsq12);
2053 rinv13 = gmx_mm_invsqrt_pd(rsq13);
2054 rinv21 = gmx_mm_invsqrt_pd(rsq21);
2055 rinv22 = gmx_mm_invsqrt_pd(rsq22);
2056 rinv23 = gmx_mm_invsqrt_pd(rsq23);
2057 rinv31 = gmx_mm_invsqrt_pd(rsq31);
2058 rinv32 = gmx_mm_invsqrt_pd(rsq32);
2059 rinv33 = gmx_mm_invsqrt_pd(rsq33);
2061 fjx0 = _mm_setzero_pd();
2062 fjy0 = _mm_setzero_pd();
2063 fjz0 = _mm_setzero_pd();
2064 fjx1 = _mm_setzero_pd();
2065 fjy1 = _mm_setzero_pd();
2066 fjz1 = _mm_setzero_pd();
2067 fjx2 = _mm_setzero_pd();
2068 fjy2 = _mm_setzero_pd();
2069 fjz2 = _mm_setzero_pd();
2070 fjx3 = _mm_setzero_pd();
2071 fjy3 = _mm_setzero_pd();
2072 fjz3 = _mm_setzero_pd();
2074 /**************************
2075 * CALCULATE INTERACTIONS *
2076 **************************/
2078 r00 = _mm_mul_pd(rsq00,rinv00);
2080 /* Calculate table index by multiplying r with table scale and truncate to integer */
2081 rt = _mm_mul_pd(r00,vftabscale);
2082 vfitab = _mm_cvttpd_epi32(rt);
2084 vfeps = _mm_frcz_pd(rt);
2086 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2088 twovfeps = _mm_add_pd(vfeps,vfeps);
2089 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2091 /* CUBIC SPLINE TABLE DISPERSION */
2092 vfitab = _mm_add_epi32(vfitab,ifour);
2093 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2094 F = _mm_setzero_pd();
2095 GMX_MM_TRANSPOSE2_PD(Y,F);
2096 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2097 H = _mm_setzero_pd();
2098 GMX_MM_TRANSPOSE2_PD(G,H);
2099 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
2100 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
2101 fvdw6 = _mm_mul_pd(c6_00,FF);
2103 /* CUBIC SPLINE TABLE REPULSION */
2104 vfitab = _mm_add_epi32(vfitab,ifour);
2105 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2106 F = _mm_setzero_pd();
2107 GMX_MM_TRANSPOSE2_PD(Y,F);
2108 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2109 H = _mm_setzero_pd();
2110 GMX_MM_TRANSPOSE2_PD(G,H);
2111 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
2112 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
2113 fvdw12 = _mm_mul_pd(c12_00,FF);
2114 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
2118 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2120 /* Update vectorial force */
2121 fix0 = _mm_macc_pd(dx00,fscal,fix0);
2122 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
2123 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
2125 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
2126 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
2127 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
2129 /**************************
2130 * CALCULATE INTERACTIONS *
2131 **************************/
2133 r11 = _mm_mul_pd(rsq11,rinv11);
2135 /* Calculate table index by multiplying r with table scale and truncate to integer */
2136 rt = _mm_mul_pd(r11,vftabscale);
2137 vfitab = _mm_cvttpd_epi32(rt);
2139 vfeps = _mm_frcz_pd(rt);
2141 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2143 twovfeps = _mm_add_pd(vfeps,vfeps);
2144 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2146 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2147 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2148 F = _mm_setzero_pd();
2149 GMX_MM_TRANSPOSE2_PD(Y,F);
2150 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2151 H = _mm_setzero_pd();
2152 GMX_MM_TRANSPOSE2_PD(G,H);
2153 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2154 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2155 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
2159 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2161 /* Update vectorial force */
2162 fix1 = _mm_macc_pd(dx11,fscal,fix1);
2163 fiy1 = _mm_macc_pd(dy11,fscal,fiy1);
2164 fiz1 = _mm_macc_pd(dz11,fscal,fiz1);
2166 fjx1 = _mm_macc_pd(dx11,fscal,fjx1);
2167 fjy1 = _mm_macc_pd(dy11,fscal,fjy1);
2168 fjz1 = _mm_macc_pd(dz11,fscal,fjz1);
2170 /**************************
2171 * CALCULATE INTERACTIONS *
2172 **************************/
2174 r12 = _mm_mul_pd(rsq12,rinv12);
2176 /* Calculate table index by multiplying r with table scale and truncate to integer */
2177 rt = _mm_mul_pd(r12,vftabscale);
2178 vfitab = _mm_cvttpd_epi32(rt);
2180 vfeps = _mm_frcz_pd(rt);
2182 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2184 twovfeps = _mm_add_pd(vfeps,vfeps);
2185 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2187 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2188 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2189 F = _mm_setzero_pd();
2190 GMX_MM_TRANSPOSE2_PD(Y,F);
2191 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2192 H = _mm_setzero_pd();
2193 GMX_MM_TRANSPOSE2_PD(G,H);
2194 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2195 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2196 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
2200 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2202 /* Update vectorial force */
2203 fix1 = _mm_macc_pd(dx12,fscal,fix1);
2204 fiy1 = _mm_macc_pd(dy12,fscal,fiy1);
2205 fiz1 = _mm_macc_pd(dz12,fscal,fiz1);
2207 fjx2 = _mm_macc_pd(dx12,fscal,fjx2);
2208 fjy2 = _mm_macc_pd(dy12,fscal,fjy2);
2209 fjz2 = _mm_macc_pd(dz12,fscal,fjz2);
2211 /**************************
2212 * CALCULATE INTERACTIONS *
2213 **************************/
2215 r13 = _mm_mul_pd(rsq13,rinv13);
2217 /* Calculate table index by multiplying r with table scale and truncate to integer */
2218 rt = _mm_mul_pd(r13,vftabscale);
2219 vfitab = _mm_cvttpd_epi32(rt);
2221 vfeps = _mm_frcz_pd(rt);
2223 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2225 twovfeps = _mm_add_pd(vfeps,vfeps);
2226 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2228 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2229 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2230 F = _mm_setzero_pd();
2231 GMX_MM_TRANSPOSE2_PD(Y,F);
2232 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2233 H = _mm_setzero_pd();
2234 GMX_MM_TRANSPOSE2_PD(G,H);
2235 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2236 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2237 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
2241 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2243 /* Update vectorial force */
2244 fix1 = _mm_macc_pd(dx13,fscal,fix1);
2245 fiy1 = _mm_macc_pd(dy13,fscal,fiy1);
2246 fiz1 = _mm_macc_pd(dz13,fscal,fiz1);
2248 fjx3 = _mm_macc_pd(dx13,fscal,fjx3);
2249 fjy3 = _mm_macc_pd(dy13,fscal,fjy3);
2250 fjz3 = _mm_macc_pd(dz13,fscal,fjz3);
2252 /**************************
2253 * CALCULATE INTERACTIONS *
2254 **************************/
2256 r21 = _mm_mul_pd(rsq21,rinv21);
2258 /* Calculate table index by multiplying r with table scale and truncate to integer */
2259 rt = _mm_mul_pd(r21,vftabscale);
2260 vfitab = _mm_cvttpd_epi32(rt);
2262 vfeps = _mm_frcz_pd(rt);
2264 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2266 twovfeps = _mm_add_pd(vfeps,vfeps);
2267 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2269 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2270 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2271 F = _mm_setzero_pd();
2272 GMX_MM_TRANSPOSE2_PD(Y,F);
2273 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2274 H = _mm_setzero_pd();
2275 GMX_MM_TRANSPOSE2_PD(G,H);
2276 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2277 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2278 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
2282 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2284 /* Update vectorial force */
2285 fix2 = _mm_macc_pd(dx21,fscal,fix2);
2286 fiy2 = _mm_macc_pd(dy21,fscal,fiy2);
2287 fiz2 = _mm_macc_pd(dz21,fscal,fiz2);
2289 fjx1 = _mm_macc_pd(dx21,fscal,fjx1);
2290 fjy1 = _mm_macc_pd(dy21,fscal,fjy1);
2291 fjz1 = _mm_macc_pd(dz21,fscal,fjz1);
2293 /**************************
2294 * CALCULATE INTERACTIONS *
2295 **************************/
2297 r22 = _mm_mul_pd(rsq22,rinv22);
2299 /* Calculate table index by multiplying r with table scale and truncate to integer */
2300 rt = _mm_mul_pd(r22,vftabscale);
2301 vfitab = _mm_cvttpd_epi32(rt);
2303 vfeps = _mm_frcz_pd(rt);
2305 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2307 twovfeps = _mm_add_pd(vfeps,vfeps);
2308 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2310 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2311 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2312 F = _mm_setzero_pd();
2313 GMX_MM_TRANSPOSE2_PD(Y,F);
2314 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2315 H = _mm_setzero_pd();
2316 GMX_MM_TRANSPOSE2_PD(G,H);
2317 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2318 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2319 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
2323 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2325 /* Update vectorial force */
2326 fix2 = _mm_macc_pd(dx22,fscal,fix2);
2327 fiy2 = _mm_macc_pd(dy22,fscal,fiy2);
2328 fiz2 = _mm_macc_pd(dz22,fscal,fiz2);
2330 fjx2 = _mm_macc_pd(dx22,fscal,fjx2);
2331 fjy2 = _mm_macc_pd(dy22,fscal,fjy2);
2332 fjz2 = _mm_macc_pd(dz22,fscal,fjz2);
2334 /**************************
2335 * CALCULATE INTERACTIONS *
2336 **************************/
2338 r23 = _mm_mul_pd(rsq23,rinv23);
2340 /* Calculate table index by multiplying r with table scale and truncate to integer */
2341 rt = _mm_mul_pd(r23,vftabscale);
2342 vfitab = _mm_cvttpd_epi32(rt);
2344 vfeps = _mm_frcz_pd(rt);
2346 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2348 twovfeps = _mm_add_pd(vfeps,vfeps);
2349 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2351 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2352 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2353 F = _mm_setzero_pd();
2354 GMX_MM_TRANSPOSE2_PD(Y,F);
2355 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2356 H = _mm_setzero_pd();
2357 GMX_MM_TRANSPOSE2_PD(G,H);
2358 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2359 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2360 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
2364 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2366 /* Update vectorial force */
2367 fix2 = _mm_macc_pd(dx23,fscal,fix2);
2368 fiy2 = _mm_macc_pd(dy23,fscal,fiy2);
2369 fiz2 = _mm_macc_pd(dz23,fscal,fiz2);
2371 fjx3 = _mm_macc_pd(dx23,fscal,fjx3);
2372 fjy3 = _mm_macc_pd(dy23,fscal,fjy3);
2373 fjz3 = _mm_macc_pd(dz23,fscal,fjz3);
2375 /**************************
2376 * CALCULATE INTERACTIONS *
2377 **************************/
2379 r31 = _mm_mul_pd(rsq31,rinv31);
2381 /* Calculate table index by multiplying r with table scale and truncate to integer */
2382 rt = _mm_mul_pd(r31,vftabscale);
2383 vfitab = _mm_cvttpd_epi32(rt);
2385 vfeps = _mm_frcz_pd(rt);
2387 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2389 twovfeps = _mm_add_pd(vfeps,vfeps);
2390 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2392 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2393 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2394 F = _mm_setzero_pd();
2395 GMX_MM_TRANSPOSE2_PD(Y,F);
2396 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2397 H = _mm_setzero_pd();
2398 GMX_MM_TRANSPOSE2_PD(G,H);
2399 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2400 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2401 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
2405 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2407 /* Update vectorial force */
2408 fix3 = _mm_macc_pd(dx31,fscal,fix3);
2409 fiy3 = _mm_macc_pd(dy31,fscal,fiy3);
2410 fiz3 = _mm_macc_pd(dz31,fscal,fiz3);
2412 fjx1 = _mm_macc_pd(dx31,fscal,fjx1);
2413 fjy1 = _mm_macc_pd(dy31,fscal,fjy1);
2414 fjz1 = _mm_macc_pd(dz31,fscal,fjz1);
2416 /**************************
2417 * CALCULATE INTERACTIONS *
2418 **************************/
2420 r32 = _mm_mul_pd(rsq32,rinv32);
2422 /* Calculate table index by multiplying r with table scale and truncate to integer */
2423 rt = _mm_mul_pd(r32,vftabscale);
2424 vfitab = _mm_cvttpd_epi32(rt);
2426 vfeps = _mm_frcz_pd(rt);
2428 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2430 twovfeps = _mm_add_pd(vfeps,vfeps);
2431 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2433 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2434 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2435 F = _mm_setzero_pd();
2436 GMX_MM_TRANSPOSE2_PD(Y,F);
2437 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2438 H = _mm_setzero_pd();
2439 GMX_MM_TRANSPOSE2_PD(G,H);
2440 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2441 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2442 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
2446 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2448 /* Update vectorial force */
2449 fix3 = _mm_macc_pd(dx32,fscal,fix3);
2450 fiy3 = _mm_macc_pd(dy32,fscal,fiy3);
2451 fiz3 = _mm_macc_pd(dz32,fscal,fiz3);
2453 fjx2 = _mm_macc_pd(dx32,fscal,fjx2);
2454 fjy2 = _mm_macc_pd(dy32,fscal,fjy2);
2455 fjz2 = _mm_macc_pd(dz32,fscal,fjz2);
2457 /**************************
2458 * CALCULATE INTERACTIONS *
2459 **************************/
2461 r33 = _mm_mul_pd(rsq33,rinv33);
2463 /* Calculate table index by multiplying r with table scale and truncate to integer */
2464 rt = _mm_mul_pd(r33,vftabscale);
2465 vfitab = _mm_cvttpd_epi32(rt);
2467 vfeps = _mm_frcz_pd(rt);
2469 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2471 twovfeps = _mm_add_pd(vfeps,vfeps);
2472 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2474 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2475 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2476 F = _mm_setzero_pd();
2477 GMX_MM_TRANSPOSE2_PD(Y,F);
2478 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2479 H = _mm_setzero_pd();
2480 GMX_MM_TRANSPOSE2_PD(G,H);
2481 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2482 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2483 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
2487 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2489 /* Update vectorial force */
2490 fix3 = _mm_macc_pd(dx33,fscal,fix3);
2491 fiy3 = _mm_macc_pd(dy33,fscal,fiy3);
2492 fiz3 = _mm_macc_pd(dz33,fscal,fiz3);
2494 fjx3 = _mm_macc_pd(dx33,fscal,fjx3);
2495 fjy3 = _mm_macc_pd(dy33,fscal,fjy3);
2496 fjz3 = _mm_macc_pd(dz33,fscal,fjz3);
2498 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2500 /* Inner loop uses 432 flops */
2503 /* End of innermost loop */
2505 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2506 f+i_coord_offset,fshift+i_shift_offset);
2508 /* Increment number of inner iterations */
2509 inneriter += j_index_end - j_index_start;
2511 /* Outer loop uses 24 flops */
2514 /* Increment number of outer iterations */
2517 /* Update outer/inner flops */
2519 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*432);