2 * Note: this file was generated by the Gromacs sse2_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse2_double.h"
34 #include "kernelutil_x86_sse2_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: CubicSplineTable
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
68 int vdwjidx0A,vdwjidx0B;
69 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
70 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
71 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
74 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
77 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
78 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
80 __m128i ifour = _mm_set1_epi32(4);
81 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
83 __m128d dummy_mask,cutoff_mask;
84 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
85 __m128d one = _mm_set1_pd(1.0);
86 __m128d two = _mm_set1_pd(2.0);
92 jindex = nlist->jindex;
94 shiftidx = nlist->shift;
96 shiftvec = fr->shift_vec[0];
97 fshift = fr->fshift[0];
98 facel = _mm_set1_pd(fr->epsfac);
99 charge = mdatoms->chargeA;
100 krf = _mm_set1_pd(fr->ic->k_rf);
101 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
102 crf = _mm_set1_pd(fr->ic->c_rf);
103 nvdwtype = fr->ntype;
105 vdwtype = mdatoms->typeA;
107 vftab = kernel_data->table_vdw->data;
108 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
110 /* Avoid stupid compiler warnings */
118 /* Start outer loop over neighborlists */
119 for(iidx=0; iidx<nri; iidx++)
121 /* Load shift vector for this list */
122 i_shift_offset = DIM*shiftidx[iidx];
124 /* Load limits for loop over neighbors */
125 j_index_start = jindex[iidx];
126 j_index_end = jindex[iidx+1];
128 /* Get outer coordinate index */
130 i_coord_offset = DIM*inr;
132 /* Load i particle coords and add shift vector */
133 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
135 fix0 = _mm_setzero_pd();
136 fiy0 = _mm_setzero_pd();
137 fiz0 = _mm_setzero_pd();
139 /* Load parameters for i particles */
140 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
141 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
143 /* Reset potential sums */
144 velecsum = _mm_setzero_pd();
145 vvdwsum = _mm_setzero_pd();
147 /* Start inner kernel loop */
148 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
151 /* Get j neighbor index, and coordinate index */
154 j_coord_offsetA = DIM*jnrA;
155 j_coord_offsetB = DIM*jnrB;
157 /* load j atom coordinates */
158 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
161 /* Calculate displacement vector */
162 dx00 = _mm_sub_pd(ix0,jx0);
163 dy00 = _mm_sub_pd(iy0,jy0);
164 dz00 = _mm_sub_pd(iz0,jz0);
166 /* Calculate squared distance and things based on it */
167 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
169 rinv00 = gmx_mm_invsqrt_pd(rsq00);
171 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
173 /* Load parameters for j particles */
174 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
175 vdwjidx0A = 2*vdwtype[jnrA+0];
176 vdwjidx0B = 2*vdwtype[jnrB+0];
178 /**************************
179 * CALCULATE INTERACTIONS *
180 **************************/
182 r00 = _mm_mul_pd(rsq00,rinv00);
184 /* Compute parameters for interactions between i and j atoms */
185 qq00 = _mm_mul_pd(iq0,jq0);
186 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
187 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
189 /* Calculate table index by multiplying r with table scale and truncate to integer */
190 rt = _mm_mul_pd(r00,vftabscale);
191 vfitab = _mm_cvttpd_epi32(rt);
192 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
193 vfitab = _mm_slli_epi32(vfitab,3);
195 /* REACTION-FIELD ELECTROSTATICS */
196 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
197 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
199 /* CUBIC SPLINE TABLE DISPERSION */
200 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
201 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
202 GMX_MM_TRANSPOSE2_PD(Y,F);
203 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
204 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
205 GMX_MM_TRANSPOSE2_PD(G,H);
206 Heps = _mm_mul_pd(vfeps,H);
207 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
208 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
209 vvdw6 = _mm_mul_pd(c6_00,VV);
210 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
211 fvdw6 = _mm_mul_pd(c6_00,FF);
213 /* CUBIC SPLINE TABLE REPULSION */
214 vfitab = _mm_add_epi32(vfitab,ifour);
215 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
216 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
217 GMX_MM_TRANSPOSE2_PD(Y,F);
218 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
219 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
220 GMX_MM_TRANSPOSE2_PD(G,H);
221 Heps = _mm_mul_pd(vfeps,H);
222 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
223 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
224 vvdw12 = _mm_mul_pd(c12_00,VV);
225 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
226 fvdw12 = _mm_mul_pd(c12_00,FF);
227 vvdw = _mm_add_pd(vvdw12,vvdw6);
228 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
230 /* Update potential sum for this i atom from the interaction with this j atom. */
231 velecsum = _mm_add_pd(velecsum,velec);
232 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
234 fscal = _mm_add_pd(felec,fvdw);
236 /* Calculate temporary vectorial force */
237 tx = _mm_mul_pd(fscal,dx00);
238 ty = _mm_mul_pd(fscal,dy00);
239 tz = _mm_mul_pd(fscal,dz00);
241 /* Update vectorial force */
242 fix0 = _mm_add_pd(fix0,tx);
243 fiy0 = _mm_add_pd(fiy0,ty);
244 fiz0 = _mm_add_pd(fiz0,tz);
246 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
248 /* Inner loop uses 67 flops */
255 j_coord_offsetA = DIM*jnrA;
257 /* load j atom coordinates */
258 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
261 /* Calculate displacement vector */
262 dx00 = _mm_sub_pd(ix0,jx0);
263 dy00 = _mm_sub_pd(iy0,jy0);
264 dz00 = _mm_sub_pd(iz0,jz0);
266 /* Calculate squared distance and things based on it */
267 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
269 rinv00 = gmx_mm_invsqrt_pd(rsq00);
271 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
273 /* Load parameters for j particles */
274 jq0 = _mm_load_sd(charge+jnrA+0);
275 vdwjidx0A = 2*vdwtype[jnrA+0];
277 /**************************
278 * CALCULATE INTERACTIONS *
279 **************************/
281 r00 = _mm_mul_pd(rsq00,rinv00);
283 /* Compute parameters for interactions between i and j atoms */
284 qq00 = _mm_mul_pd(iq0,jq0);
285 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
287 /* Calculate table index by multiplying r with table scale and truncate to integer */
288 rt = _mm_mul_pd(r00,vftabscale);
289 vfitab = _mm_cvttpd_epi32(rt);
290 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
291 vfitab = _mm_slli_epi32(vfitab,3);
293 /* REACTION-FIELD ELECTROSTATICS */
294 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
295 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
297 /* CUBIC SPLINE TABLE DISPERSION */
298 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
299 F = _mm_setzero_pd();
300 GMX_MM_TRANSPOSE2_PD(Y,F);
301 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
302 H = _mm_setzero_pd();
303 GMX_MM_TRANSPOSE2_PD(G,H);
304 Heps = _mm_mul_pd(vfeps,H);
305 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
306 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
307 vvdw6 = _mm_mul_pd(c6_00,VV);
308 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
309 fvdw6 = _mm_mul_pd(c6_00,FF);
311 /* CUBIC SPLINE TABLE REPULSION */
312 vfitab = _mm_add_epi32(vfitab,ifour);
313 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
314 F = _mm_setzero_pd();
315 GMX_MM_TRANSPOSE2_PD(Y,F);
316 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
317 H = _mm_setzero_pd();
318 GMX_MM_TRANSPOSE2_PD(G,H);
319 Heps = _mm_mul_pd(vfeps,H);
320 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
321 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
322 vvdw12 = _mm_mul_pd(c12_00,VV);
323 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
324 fvdw12 = _mm_mul_pd(c12_00,FF);
325 vvdw = _mm_add_pd(vvdw12,vvdw6);
326 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
328 /* Update potential sum for this i atom from the interaction with this j atom. */
329 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
330 velecsum = _mm_add_pd(velecsum,velec);
331 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
332 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
334 fscal = _mm_add_pd(felec,fvdw);
336 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
338 /* Calculate temporary vectorial force */
339 tx = _mm_mul_pd(fscal,dx00);
340 ty = _mm_mul_pd(fscal,dy00);
341 tz = _mm_mul_pd(fscal,dz00);
343 /* Update vectorial force */
344 fix0 = _mm_add_pd(fix0,tx);
345 fiy0 = _mm_add_pd(fiy0,ty);
346 fiz0 = _mm_add_pd(fiz0,tz);
348 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
350 /* Inner loop uses 67 flops */
353 /* End of innermost loop */
355 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
356 f+i_coord_offset,fshift+i_shift_offset);
359 /* Update potential energies */
360 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
361 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
363 /* Increment number of inner iterations */
364 inneriter += j_index_end - j_index_start;
366 /* Outer loop uses 9 flops */
369 /* Increment number of outer iterations */
372 /* Update outer/inner flops */
374 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
377 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_double
378 * Electrostatics interaction: ReactionField
379 * VdW interaction: CubicSplineTable
380 * Geometry: Particle-Particle
381 * Calculate force/pot: Force
384 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_double
385 (t_nblist * gmx_restrict nlist,
386 rvec * gmx_restrict xx,
387 rvec * gmx_restrict ff,
388 t_forcerec * gmx_restrict fr,
389 t_mdatoms * gmx_restrict mdatoms,
390 nb_kernel_data_t * gmx_restrict kernel_data,
391 t_nrnb * gmx_restrict nrnb)
393 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
394 * just 0 for non-waters.
395 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
396 * jnr indices corresponding to data put in the four positions in the SIMD register.
398 int i_shift_offset,i_coord_offset,outeriter,inneriter;
399 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
401 int j_coord_offsetA,j_coord_offsetB;
402 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
404 real *shiftvec,*fshift,*x,*f;
405 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
407 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
408 int vdwjidx0A,vdwjidx0B;
409 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
410 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
411 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
414 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
417 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
418 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
420 __m128i ifour = _mm_set1_epi32(4);
421 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
423 __m128d dummy_mask,cutoff_mask;
424 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
425 __m128d one = _mm_set1_pd(1.0);
426 __m128d two = _mm_set1_pd(2.0);
432 jindex = nlist->jindex;
434 shiftidx = nlist->shift;
436 shiftvec = fr->shift_vec[0];
437 fshift = fr->fshift[0];
438 facel = _mm_set1_pd(fr->epsfac);
439 charge = mdatoms->chargeA;
440 krf = _mm_set1_pd(fr->ic->k_rf);
441 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
442 crf = _mm_set1_pd(fr->ic->c_rf);
443 nvdwtype = fr->ntype;
445 vdwtype = mdatoms->typeA;
447 vftab = kernel_data->table_vdw->data;
448 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
450 /* Avoid stupid compiler warnings */
458 /* Start outer loop over neighborlists */
459 for(iidx=0; iidx<nri; iidx++)
461 /* Load shift vector for this list */
462 i_shift_offset = DIM*shiftidx[iidx];
464 /* Load limits for loop over neighbors */
465 j_index_start = jindex[iidx];
466 j_index_end = jindex[iidx+1];
468 /* Get outer coordinate index */
470 i_coord_offset = DIM*inr;
472 /* Load i particle coords and add shift vector */
473 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
475 fix0 = _mm_setzero_pd();
476 fiy0 = _mm_setzero_pd();
477 fiz0 = _mm_setzero_pd();
479 /* Load parameters for i particles */
480 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
481 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
483 /* Start inner kernel loop */
484 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
487 /* Get j neighbor index, and coordinate index */
490 j_coord_offsetA = DIM*jnrA;
491 j_coord_offsetB = DIM*jnrB;
493 /* load j atom coordinates */
494 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
497 /* Calculate displacement vector */
498 dx00 = _mm_sub_pd(ix0,jx0);
499 dy00 = _mm_sub_pd(iy0,jy0);
500 dz00 = _mm_sub_pd(iz0,jz0);
502 /* Calculate squared distance and things based on it */
503 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
505 rinv00 = gmx_mm_invsqrt_pd(rsq00);
507 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
509 /* Load parameters for j particles */
510 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
511 vdwjidx0A = 2*vdwtype[jnrA+0];
512 vdwjidx0B = 2*vdwtype[jnrB+0];
514 /**************************
515 * CALCULATE INTERACTIONS *
516 **************************/
518 r00 = _mm_mul_pd(rsq00,rinv00);
520 /* Compute parameters for interactions between i and j atoms */
521 qq00 = _mm_mul_pd(iq0,jq0);
522 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
523 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
525 /* Calculate table index by multiplying r with table scale and truncate to integer */
526 rt = _mm_mul_pd(r00,vftabscale);
527 vfitab = _mm_cvttpd_epi32(rt);
528 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
529 vfitab = _mm_slli_epi32(vfitab,3);
531 /* REACTION-FIELD ELECTROSTATICS */
532 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
534 /* CUBIC SPLINE TABLE DISPERSION */
535 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
536 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
537 GMX_MM_TRANSPOSE2_PD(Y,F);
538 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
539 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
540 GMX_MM_TRANSPOSE2_PD(G,H);
541 Heps = _mm_mul_pd(vfeps,H);
542 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
543 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
544 fvdw6 = _mm_mul_pd(c6_00,FF);
546 /* CUBIC SPLINE TABLE REPULSION */
547 vfitab = _mm_add_epi32(vfitab,ifour);
548 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
549 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
550 GMX_MM_TRANSPOSE2_PD(Y,F);
551 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
552 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
553 GMX_MM_TRANSPOSE2_PD(G,H);
554 Heps = _mm_mul_pd(vfeps,H);
555 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
556 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
557 fvdw12 = _mm_mul_pd(c12_00,FF);
558 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
560 fscal = _mm_add_pd(felec,fvdw);
562 /* Calculate temporary vectorial force */
563 tx = _mm_mul_pd(fscal,dx00);
564 ty = _mm_mul_pd(fscal,dy00);
565 tz = _mm_mul_pd(fscal,dz00);
567 /* Update vectorial force */
568 fix0 = _mm_add_pd(fix0,tx);
569 fiy0 = _mm_add_pd(fiy0,ty);
570 fiz0 = _mm_add_pd(fiz0,tz);
572 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
574 /* Inner loop uses 54 flops */
581 j_coord_offsetA = DIM*jnrA;
583 /* load j atom coordinates */
584 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
587 /* Calculate displacement vector */
588 dx00 = _mm_sub_pd(ix0,jx0);
589 dy00 = _mm_sub_pd(iy0,jy0);
590 dz00 = _mm_sub_pd(iz0,jz0);
592 /* Calculate squared distance and things based on it */
593 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
595 rinv00 = gmx_mm_invsqrt_pd(rsq00);
597 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
599 /* Load parameters for j particles */
600 jq0 = _mm_load_sd(charge+jnrA+0);
601 vdwjidx0A = 2*vdwtype[jnrA+0];
603 /**************************
604 * CALCULATE INTERACTIONS *
605 **************************/
607 r00 = _mm_mul_pd(rsq00,rinv00);
609 /* Compute parameters for interactions between i and j atoms */
610 qq00 = _mm_mul_pd(iq0,jq0);
611 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
613 /* Calculate table index by multiplying r with table scale and truncate to integer */
614 rt = _mm_mul_pd(r00,vftabscale);
615 vfitab = _mm_cvttpd_epi32(rt);
616 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
617 vfitab = _mm_slli_epi32(vfitab,3);
619 /* REACTION-FIELD ELECTROSTATICS */
620 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
622 /* CUBIC SPLINE TABLE DISPERSION */
623 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
624 F = _mm_setzero_pd();
625 GMX_MM_TRANSPOSE2_PD(Y,F);
626 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
627 H = _mm_setzero_pd();
628 GMX_MM_TRANSPOSE2_PD(G,H);
629 Heps = _mm_mul_pd(vfeps,H);
630 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
631 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
632 fvdw6 = _mm_mul_pd(c6_00,FF);
634 /* CUBIC SPLINE TABLE REPULSION */
635 vfitab = _mm_add_epi32(vfitab,ifour);
636 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
637 F = _mm_setzero_pd();
638 GMX_MM_TRANSPOSE2_PD(Y,F);
639 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
640 H = _mm_setzero_pd();
641 GMX_MM_TRANSPOSE2_PD(G,H);
642 Heps = _mm_mul_pd(vfeps,H);
643 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
644 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
645 fvdw12 = _mm_mul_pd(c12_00,FF);
646 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
648 fscal = _mm_add_pd(felec,fvdw);
650 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
652 /* Calculate temporary vectorial force */
653 tx = _mm_mul_pd(fscal,dx00);
654 ty = _mm_mul_pd(fscal,dy00);
655 tz = _mm_mul_pd(fscal,dz00);
657 /* Update vectorial force */
658 fix0 = _mm_add_pd(fix0,tx);
659 fiy0 = _mm_add_pd(fiy0,ty);
660 fiz0 = _mm_add_pd(fiz0,tz);
662 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
664 /* Inner loop uses 54 flops */
667 /* End of innermost loop */
669 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
670 f+i_coord_offset,fshift+i_shift_offset);
672 /* Increment number of inner iterations */
673 inneriter += j_index_end - j_index_start;
675 /* Outer loop uses 7 flops */
678 /* Increment number of outer iterations */
681 /* Update outer/inner flops */
683 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);