2 * Note: this file was generated by the Gromacs sse2_single 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_single.h"
34 #include "kernelutil_x86_sse2_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW3P1_VF_sse2_single
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: None
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_VF_sse2_single
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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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;
60 int jnrA,jnrB,jnrC,jnrD;
61 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
63 real shX,shY,shZ,rcutoff_scalar;
64 real *shiftvec,*fshift,*x,*f;
65 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
73 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
80 __m128i ifour = _mm_set1_epi32(4);
81 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
83 __m128 dummy_mask,cutoff_mask;
84 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
85 __m128 one = _mm_set1_ps(1.0);
86 __m128 two = _mm_set1_ps(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_ps(fr->epsfac);
99 charge = mdatoms->chargeA;
101 vftab = kernel_data->table_elec->data;
102 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
104 /* Setup water-specific parameters */
105 inr = nlist->iinr[0];
106 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
107 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
108 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
110 /* Avoid stupid compiler warnings */
111 jnrA = jnrB = jnrC = jnrD = 0;
120 /* Start outer loop over neighborlists */
121 for(iidx=0; iidx<nri; iidx++)
123 /* Load shift vector for this list */
124 i_shift_offset = DIM*shiftidx[iidx];
125 shX = shiftvec[i_shift_offset+XX];
126 shY = shiftvec[i_shift_offset+YY];
127 shZ = shiftvec[i_shift_offset+ZZ];
129 /* Load limits for loop over neighbors */
130 j_index_start = jindex[iidx];
131 j_index_end = jindex[iidx+1];
133 /* Get outer coordinate index */
135 i_coord_offset = DIM*inr;
137 /* Load i particle coords and add shift vector */
138 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
139 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
140 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
141 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
142 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
143 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
144 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
145 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
146 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
148 fix0 = _mm_setzero_ps();
149 fiy0 = _mm_setzero_ps();
150 fiz0 = _mm_setzero_ps();
151 fix1 = _mm_setzero_ps();
152 fiy1 = _mm_setzero_ps();
153 fiz1 = _mm_setzero_ps();
154 fix2 = _mm_setzero_ps();
155 fiy2 = _mm_setzero_ps();
156 fiz2 = _mm_setzero_ps();
158 /* Reset potential sums */
159 velecsum = _mm_setzero_ps();
161 /* Start inner kernel loop */
162 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
165 /* Get j neighbor index, and coordinate index */
171 j_coord_offsetA = DIM*jnrA;
172 j_coord_offsetB = DIM*jnrB;
173 j_coord_offsetC = DIM*jnrC;
174 j_coord_offsetD = DIM*jnrD;
176 /* load j atom coordinates */
177 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
178 x+j_coord_offsetC,x+j_coord_offsetD,
181 /* Calculate displacement vector */
182 dx00 = _mm_sub_ps(ix0,jx0);
183 dy00 = _mm_sub_ps(iy0,jy0);
184 dz00 = _mm_sub_ps(iz0,jz0);
185 dx10 = _mm_sub_ps(ix1,jx0);
186 dy10 = _mm_sub_ps(iy1,jy0);
187 dz10 = _mm_sub_ps(iz1,jz0);
188 dx20 = _mm_sub_ps(ix2,jx0);
189 dy20 = _mm_sub_ps(iy2,jy0);
190 dz20 = _mm_sub_ps(iz2,jz0);
192 /* Calculate squared distance and things based on it */
193 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
194 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
195 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
197 rinv00 = gmx_mm_invsqrt_ps(rsq00);
198 rinv10 = gmx_mm_invsqrt_ps(rsq10);
199 rinv20 = gmx_mm_invsqrt_ps(rsq20);
201 /* Load parameters for j particles */
202 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
203 charge+jnrC+0,charge+jnrD+0);
205 /**************************
206 * CALCULATE INTERACTIONS *
207 **************************/
209 r00 = _mm_mul_ps(rsq00,rinv00);
211 /* Compute parameters for interactions between i and j atoms */
212 qq00 = _mm_mul_ps(iq0,jq0);
214 /* Calculate table index by multiplying r with table scale and truncate to integer */
215 rt = _mm_mul_ps(r00,vftabscale);
216 vfitab = _mm_cvttps_epi32(rt);
217 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
218 vfitab = _mm_slli_epi32(vfitab,2);
220 /* CUBIC SPLINE TABLE ELECTROSTATICS */
221 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
222 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
223 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
224 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
225 _MM_TRANSPOSE4_PS(Y,F,G,H);
226 Heps = _mm_mul_ps(vfeps,H);
227 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
228 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
229 velec = _mm_mul_ps(qq00,VV);
230 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
231 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
233 /* Update potential sum for this i atom from the interaction with this j atom. */
234 velecsum = _mm_add_ps(velecsum,velec);
238 /* Calculate temporary vectorial force */
239 tx = _mm_mul_ps(fscal,dx00);
240 ty = _mm_mul_ps(fscal,dy00);
241 tz = _mm_mul_ps(fscal,dz00);
243 /* Update vectorial force */
244 fix0 = _mm_add_ps(fix0,tx);
245 fiy0 = _mm_add_ps(fiy0,ty);
246 fiz0 = _mm_add_ps(fiz0,tz);
248 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
249 f+j_coord_offsetC,f+j_coord_offsetD,
252 /**************************
253 * CALCULATE INTERACTIONS *
254 **************************/
256 r10 = _mm_mul_ps(rsq10,rinv10);
258 /* Compute parameters for interactions between i and j atoms */
259 qq10 = _mm_mul_ps(iq1,jq0);
261 /* Calculate table index by multiplying r with table scale and truncate to integer */
262 rt = _mm_mul_ps(r10,vftabscale);
263 vfitab = _mm_cvttps_epi32(rt);
264 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
265 vfitab = _mm_slli_epi32(vfitab,2);
267 /* CUBIC SPLINE TABLE ELECTROSTATICS */
268 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
269 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
270 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
271 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
272 _MM_TRANSPOSE4_PS(Y,F,G,H);
273 Heps = _mm_mul_ps(vfeps,H);
274 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
275 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
276 velec = _mm_mul_ps(qq10,VV);
277 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
278 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velecsum = _mm_add_ps(velecsum,velec);
285 /* Calculate temporary vectorial force */
286 tx = _mm_mul_ps(fscal,dx10);
287 ty = _mm_mul_ps(fscal,dy10);
288 tz = _mm_mul_ps(fscal,dz10);
290 /* Update vectorial force */
291 fix1 = _mm_add_ps(fix1,tx);
292 fiy1 = _mm_add_ps(fiy1,ty);
293 fiz1 = _mm_add_ps(fiz1,tz);
295 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
296 f+j_coord_offsetC,f+j_coord_offsetD,
299 /**************************
300 * CALCULATE INTERACTIONS *
301 **************************/
303 r20 = _mm_mul_ps(rsq20,rinv20);
305 /* Compute parameters for interactions between i and j atoms */
306 qq20 = _mm_mul_ps(iq2,jq0);
308 /* Calculate table index by multiplying r with table scale and truncate to integer */
309 rt = _mm_mul_ps(r20,vftabscale);
310 vfitab = _mm_cvttps_epi32(rt);
311 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
312 vfitab = _mm_slli_epi32(vfitab,2);
314 /* CUBIC SPLINE TABLE ELECTROSTATICS */
315 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
316 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
317 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
318 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
319 _MM_TRANSPOSE4_PS(Y,F,G,H);
320 Heps = _mm_mul_ps(vfeps,H);
321 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
322 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
323 velec = _mm_mul_ps(qq20,VV);
324 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
325 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
327 /* Update potential sum for this i atom from the interaction with this j atom. */
328 velecsum = _mm_add_ps(velecsum,velec);
332 /* Calculate temporary vectorial force */
333 tx = _mm_mul_ps(fscal,dx20);
334 ty = _mm_mul_ps(fscal,dy20);
335 tz = _mm_mul_ps(fscal,dz20);
337 /* Update vectorial force */
338 fix2 = _mm_add_ps(fix2,tx);
339 fiy2 = _mm_add_ps(fiy2,ty);
340 fiz2 = _mm_add_ps(fiz2,tz);
342 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
343 f+j_coord_offsetC,f+j_coord_offsetD,
346 /* Inner loop uses 129 flops */
352 /* Get j neighbor index, and coordinate index */
358 /* Sign of each element will be negative for non-real atoms.
359 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
360 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
362 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
363 jnrA = (jnrA>=0) ? jnrA : 0;
364 jnrB = (jnrB>=0) ? jnrB : 0;
365 jnrC = (jnrC>=0) ? jnrC : 0;
366 jnrD = (jnrD>=0) ? jnrD : 0;
368 j_coord_offsetA = DIM*jnrA;
369 j_coord_offsetB = DIM*jnrB;
370 j_coord_offsetC = DIM*jnrC;
371 j_coord_offsetD = DIM*jnrD;
373 /* load j atom coordinates */
374 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
375 x+j_coord_offsetC,x+j_coord_offsetD,
378 /* Calculate displacement vector */
379 dx00 = _mm_sub_ps(ix0,jx0);
380 dy00 = _mm_sub_ps(iy0,jy0);
381 dz00 = _mm_sub_ps(iz0,jz0);
382 dx10 = _mm_sub_ps(ix1,jx0);
383 dy10 = _mm_sub_ps(iy1,jy0);
384 dz10 = _mm_sub_ps(iz1,jz0);
385 dx20 = _mm_sub_ps(ix2,jx0);
386 dy20 = _mm_sub_ps(iy2,jy0);
387 dz20 = _mm_sub_ps(iz2,jz0);
389 /* Calculate squared distance and things based on it */
390 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
391 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
392 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
394 rinv00 = gmx_mm_invsqrt_ps(rsq00);
395 rinv10 = gmx_mm_invsqrt_ps(rsq10);
396 rinv20 = gmx_mm_invsqrt_ps(rsq20);
398 /* Load parameters for j particles */
399 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
400 charge+jnrC+0,charge+jnrD+0);
402 /**************************
403 * CALCULATE INTERACTIONS *
404 **************************/
406 r00 = _mm_mul_ps(rsq00,rinv00);
407 r00 = _mm_andnot_ps(dummy_mask,r00);
409 /* Compute parameters for interactions between i and j atoms */
410 qq00 = _mm_mul_ps(iq0,jq0);
412 /* Calculate table index by multiplying r with table scale and truncate to integer */
413 rt = _mm_mul_ps(r00,vftabscale);
414 vfitab = _mm_cvttps_epi32(rt);
415 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
416 vfitab = _mm_slli_epi32(vfitab,2);
418 /* CUBIC SPLINE TABLE ELECTROSTATICS */
419 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
420 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
421 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
422 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
423 _MM_TRANSPOSE4_PS(Y,F,G,H);
424 Heps = _mm_mul_ps(vfeps,H);
425 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
426 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
427 velec = _mm_mul_ps(qq00,VV);
428 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
429 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
431 /* Update potential sum for this i atom from the interaction with this j atom. */
432 velec = _mm_andnot_ps(dummy_mask,velec);
433 velecsum = _mm_add_ps(velecsum,velec);
437 fscal = _mm_andnot_ps(dummy_mask,fscal);
439 /* Calculate temporary vectorial force */
440 tx = _mm_mul_ps(fscal,dx00);
441 ty = _mm_mul_ps(fscal,dy00);
442 tz = _mm_mul_ps(fscal,dz00);
444 /* Update vectorial force */
445 fix0 = _mm_add_ps(fix0,tx);
446 fiy0 = _mm_add_ps(fiy0,ty);
447 fiz0 = _mm_add_ps(fiz0,tz);
449 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
450 f+j_coord_offsetC,f+j_coord_offsetD,
453 /**************************
454 * CALCULATE INTERACTIONS *
455 **************************/
457 r10 = _mm_mul_ps(rsq10,rinv10);
458 r10 = _mm_andnot_ps(dummy_mask,r10);
460 /* Compute parameters for interactions between i and j atoms */
461 qq10 = _mm_mul_ps(iq1,jq0);
463 /* Calculate table index by multiplying r with table scale and truncate to integer */
464 rt = _mm_mul_ps(r10,vftabscale);
465 vfitab = _mm_cvttps_epi32(rt);
466 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
467 vfitab = _mm_slli_epi32(vfitab,2);
469 /* CUBIC SPLINE TABLE ELECTROSTATICS */
470 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
471 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
472 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
473 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
474 _MM_TRANSPOSE4_PS(Y,F,G,H);
475 Heps = _mm_mul_ps(vfeps,H);
476 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
477 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
478 velec = _mm_mul_ps(qq10,VV);
479 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
480 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
482 /* Update potential sum for this i atom from the interaction with this j atom. */
483 velec = _mm_andnot_ps(dummy_mask,velec);
484 velecsum = _mm_add_ps(velecsum,velec);
488 fscal = _mm_andnot_ps(dummy_mask,fscal);
490 /* Calculate temporary vectorial force */
491 tx = _mm_mul_ps(fscal,dx10);
492 ty = _mm_mul_ps(fscal,dy10);
493 tz = _mm_mul_ps(fscal,dz10);
495 /* Update vectorial force */
496 fix1 = _mm_add_ps(fix1,tx);
497 fiy1 = _mm_add_ps(fiy1,ty);
498 fiz1 = _mm_add_ps(fiz1,tz);
500 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
501 f+j_coord_offsetC,f+j_coord_offsetD,
504 /**************************
505 * CALCULATE INTERACTIONS *
506 **************************/
508 r20 = _mm_mul_ps(rsq20,rinv20);
509 r20 = _mm_andnot_ps(dummy_mask,r20);
511 /* Compute parameters for interactions between i and j atoms */
512 qq20 = _mm_mul_ps(iq2,jq0);
514 /* Calculate table index by multiplying r with table scale and truncate to integer */
515 rt = _mm_mul_ps(r20,vftabscale);
516 vfitab = _mm_cvttps_epi32(rt);
517 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
518 vfitab = _mm_slli_epi32(vfitab,2);
520 /* CUBIC SPLINE TABLE ELECTROSTATICS */
521 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
522 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
523 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
524 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
525 _MM_TRANSPOSE4_PS(Y,F,G,H);
526 Heps = _mm_mul_ps(vfeps,H);
527 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
528 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
529 velec = _mm_mul_ps(qq20,VV);
530 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
531 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
533 /* Update potential sum for this i atom from the interaction with this j atom. */
534 velec = _mm_andnot_ps(dummy_mask,velec);
535 velecsum = _mm_add_ps(velecsum,velec);
539 fscal = _mm_andnot_ps(dummy_mask,fscal);
541 /* Calculate temporary vectorial force */
542 tx = _mm_mul_ps(fscal,dx20);
543 ty = _mm_mul_ps(fscal,dy20);
544 tz = _mm_mul_ps(fscal,dz20);
546 /* Update vectorial force */
547 fix2 = _mm_add_ps(fix2,tx);
548 fiy2 = _mm_add_ps(fiy2,ty);
549 fiz2 = _mm_add_ps(fiz2,tz);
551 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
552 f+j_coord_offsetC,f+j_coord_offsetD,
555 /* Inner loop uses 132 flops */
558 /* End of innermost loop */
560 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
561 f+i_coord_offset,fshift+i_shift_offset);
564 /* Update potential energies */
565 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
567 /* Increment number of inner iterations */
568 inneriter += j_index_end - j_index_start;
570 /* Outer loop uses 28 flops */
573 /* Increment number of outer iterations */
576 /* Update outer/inner flops */
578 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*28 + inneriter*132);
581 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_sse2_single
582 * Electrostatics interaction: CubicSplineTable
583 * VdW interaction: None
584 * Geometry: Water3-Particle
585 * Calculate force/pot: Force
588 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_sse2_single
589 (t_nblist * gmx_restrict nlist,
590 rvec * gmx_restrict xx,
591 rvec * gmx_restrict ff,
592 t_forcerec * gmx_restrict fr,
593 t_mdatoms * gmx_restrict mdatoms,
594 nb_kernel_data_t * gmx_restrict kernel_data,
595 t_nrnb * gmx_restrict nrnb)
597 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
598 * just 0 for non-waters.
599 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
600 * jnr indices corresponding to data put in the four positions in the SIMD register.
602 int i_shift_offset,i_coord_offset,outeriter,inneriter;
603 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
604 int jnrA,jnrB,jnrC,jnrD;
605 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
606 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
607 real shX,shY,shZ,rcutoff_scalar;
608 real *shiftvec,*fshift,*x,*f;
609 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
611 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
613 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
615 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
616 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
617 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
618 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
619 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
620 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
621 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
624 __m128i ifour = _mm_set1_epi32(4);
625 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
627 __m128 dummy_mask,cutoff_mask;
628 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
629 __m128 one = _mm_set1_ps(1.0);
630 __m128 two = _mm_set1_ps(2.0);
636 jindex = nlist->jindex;
638 shiftidx = nlist->shift;
640 shiftvec = fr->shift_vec[0];
641 fshift = fr->fshift[0];
642 facel = _mm_set1_ps(fr->epsfac);
643 charge = mdatoms->chargeA;
645 vftab = kernel_data->table_elec->data;
646 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
648 /* Setup water-specific parameters */
649 inr = nlist->iinr[0];
650 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
651 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
652 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
654 /* Avoid stupid compiler warnings */
655 jnrA = jnrB = jnrC = jnrD = 0;
664 /* Start outer loop over neighborlists */
665 for(iidx=0; iidx<nri; iidx++)
667 /* Load shift vector for this list */
668 i_shift_offset = DIM*shiftidx[iidx];
669 shX = shiftvec[i_shift_offset+XX];
670 shY = shiftvec[i_shift_offset+YY];
671 shZ = shiftvec[i_shift_offset+ZZ];
673 /* Load limits for loop over neighbors */
674 j_index_start = jindex[iidx];
675 j_index_end = jindex[iidx+1];
677 /* Get outer coordinate index */
679 i_coord_offset = DIM*inr;
681 /* Load i particle coords and add shift vector */
682 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
683 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
684 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
685 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
686 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
687 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
688 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
689 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
690 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
692 fix0 = _mm_setzero_ps();
693 fiy0 = _mm_setzero_ps();
694 fiz0 = _mm_setzero_ps();
695 fix1 = _mm_setzero_ps();
696 fiy1 = _mm_setzero_ps();
697 fiz1 = _mm_setzero_ps();
698 fix2 = _mm_setzero_ps();
699 fiy2 = _mm_setzero_ps();
700 fiz2 = _mm_setzero_ps();
702 /* Start inner kernel loop */
703 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
706 /* Get j neighbor index, and coordinate index */
712 j_coord_offsetA = DIM*jnrA;
713 j_coord_offsetB = DIM*jnrB;
714 j_coord_offsetC = DIM*jnrC;
715 j_coord_offsetD = DIM*jnrD;
717 /* load j atom coordinates */
718 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
719 x+j_coord_offsetC,x+j_coord_offsetD,
722 /* Calculate displacement vector */
723 dx00 = _mm_sub_ps(ix0,jx0);
724 dy00 = _mm_sub_ps(iy0,jy0);
725 dz00 = _mm_sub_ps(iz0,jz0);
726 dx10 = _mm_sub_ps(ix1,jx0);
727 dy10 = _mm_sub_ps(iy1,jy0);
728 dz10 = _mm_sub_ps(iz1,jz0);
729 dx20 = _mm_sub_ps(ix2,jx0);
730 dy20 = _mm_sub_ps(iy2,jy0);
731 dz20 = _mm_sub_ps(iz2,jz0);
733 /* Calculate squared distance and things based on it */
734 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
735 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
736 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
738 rinv00 = gmx_mm_invsqrt_ps(rsq00);
739 rinv10 = gmx_mm_invsqrt_ps(rsq10);
740 rinv20 = gmx_mm_invsqrt_ps(rsq20);
742 /* Load parameters for j particles */
743 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
744 charge+jnrC+0,charge+jnrD+0);
746 /**************************
747 * CALCULATE INTERACTIONS *
748 **************************/
750 r00 = _mm_mul_ps(rsq00,rinv00);
752 /* Compute parameters for interactions between i and j atoms */
753 qq00 = _mm_mul_ps(iq0,jq0);
755 /* Calculate table index by multiplying r with table scale and truncate to integer */
756 rt = _mm_mul_ps(r00,vftabscale);
757 vfitab = _mm_cvttps_epi32(rt);
758 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
759 vfitab = _mm_slli_epi32(vfitab,2);
761 /* CUBIC SPLINE TABLE ELECTROSTATICS */
762 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
763 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
764 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
765 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
766 _MM_TRANSPOSE4_PS(Y,F,G,H);
767 Heps = _mm_mul_ps(vfeps,H);
768 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
769 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
770 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
774 /* Calculate temporary vectorial force */
775 tx = _mm_mul_ps(fscal,dx00);
776 ty = _mm_mul_ps(fscal,dy00);
777 tz = _mm_mul_ps(fscal,dz00);
779 /* Update vectorial force */
780 fix0 = _mm_add_ps(fix0,tx);
781 fiy0 = _mm_add_ps(fiy0,ty);
782 fiz0 = _mm_add_ps(fiz0,tz);
784 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
785 f+j_coord_offsetC,f+j_coord_offsetD,
788 /**************************
789 * CALCULATE INTERACTIONS *
790 **************************/
792 r10 = _mm_mul_ps(rsq10,rinv10);
794 /* Compute parameters for interactions between i and j atoms */
795 qq10 = _mm_mul_ps(iq1,jq0);
797 /* Calculate table index by multiplying r with table scale and truncate to integer */
798 rt = _mm_mul_ps(r10,vftabscale);
799 vfitab = _mm_cvttps_epi32(rt);
800 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
801 vfitab = _mm_slli_epi32(vfitab,2);
803 /* CUBIC SPLINE TABLE ELECTROSTATICS */
804 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
805 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
806 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
807 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
808 _MM_TRANSPOSE4_PS(Y,F,G,H);
809 Heps = _mm_mul_ps(vfeps,H);
810 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
811 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
812 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
816 /* Calculate temporary vectorial force */
817 tx = _mm_mul_ps(fscal,dx10);
818 ty = _mm_mul_ps(fscal,dy10);
819 tz = _mm_mul_ps(fscal,dz10);
821 /* Update vectorial force */
822 fix1 = _mm_add_ps(fix1,tx);
823 fiy1 = _mm_add_ps(fiy1,ty);
824 fiz1 = _mm_add_ps(fiz1,tz);
826 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
827 f+j_coord_offsetC,f+j_coord_offsetD,
830 /**************************
831 * CALCULATE INTERACTIONS *
832 **************************/
834 r20 = _mm_mul_ps(rsq20,rinv20);
836 /* Compute parameters for interactions between i and j atoms */
837 qq20 = _mm_mul_ps(iq2,jq0);
839 /* Calculate table index by multiplying r with table scale and truncate to integer */
840 rt = _mm_mul_ps(r20,vftabscale);
841 vfitab = _mm_cvttps_epi32(rt);
842 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
843 vfitab = _mm_slli_epi32(vfitab,2);
845 /* CUBIC SPLINE TABLE ELECTROSTATICS */
846 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
847 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
848 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
849 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
850 _MM_TRANSPOSE4_PS(Y,F,G,H);
851 Heps = _mm_mul_ps(vfeps,H);
852 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
853 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
854 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
858 /* Calculate temporary vectorial force */
859 tx = _mm_mul_ps(fscal,dx20);
860 ty = _mm_mul_ps(fscal,dy20);
861 tz = _mm_mul_ps(fscal,dz20);
863 /* Update vectorial force */
864 fix2 = _mm_add_ps(fix2,tx);
865 fiy2 = _mm_add_ps(fiy2,ty);
866 fiz2 = _mm_add_ps(fiz2,tz);
868 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
869 f+j_coord_offsetC,f+j_coord_offsetD,
872 /* Inner loop uses 117 flops */
878 /* Get j neighbor index, and coordinate index */
884 /* Sign of each element will be negative for non-real atoms.
885 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
886 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
888 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
889 jnrA = (jnrA>=0) ? jnrA : 0;
890 jnrB = (jnrB>=0) ? jnrB : 0;
891 jnrC = (jnrC>=0) ? jnrC : 0;
892 jnrD = (jnrD>=0) ? jnrD : 0;
894 j_coord_offsetA = DIM*jnrA;
895 j_coord_offsetB = DIM*jnrB;
896 j_coord_offsetC = DIM*jnrC;
897 j_coord_offsetD = DIM*jnrD;
899 /* load j atom coordinates */
900 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
901 x+j_coord_offsetC,x+j_coord_offsetD,
904 /* Calculate displacement vector */
905 dx00 = _mm_sub_ps(ix0,jx0);
906 dy00 = _mm_sub_ps(iy0,jy0);
907 dz00 = _mm_sub_ps(iz0,jz0);
908 dx10 = _mm_sub_ps(ix1,jx0);
909 dy10 = _mm_sub_ps(iy1,jy0);
910 dz10 = _mm_sub_ps(iz1,jz0);
911 dx20 = _mm_sub_ps(ix2,jx0);
912 dy20 = _mm_sub_ps(iy2,jy0);
913 dz20 = _mm_sub_ps(iz2,jz0);
915 /* Calculate squared distance and things based on it */
916 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
917 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
918 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
920 rinv00 = gmx_mm_invsqrt_ps(rsq00);
921 rinv10 = gmx_mm_invsqrt_ps(rsq10);
922 rinv20 = gmx_mm_invsqrt_ps(rsq20);
924 /* Load parameters for j particles */
925 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
926 charge+jnrC+0,charge+jnrD+0);
928 /**************************
929 * CALCULATE INTERACTIONS *
930 **************************/
932 r00 = _mm_mul_ps(rsq00,rinv00);
933 r00 = _mm_andnot_ps(dummy_mask,r00);
935 /* Compute parameters for interactions between i and j atoms */
936 qq00 = _mm_mul_ps(iq0,jq0);
938 /* Calculate table index by multiplying r with table scale and truncate to integer */
939 rt = _mm_mul_ps(r00,vftabscale);
940 vfitab = _mm_cvttps_epi32(rt);
941 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
942 vfitab = _mm_slli_epi32(vfitab,2);
944 /* CUBIC SPLINE TABLE ELECTROSTATICS */
945 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
946 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
947 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
948 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
949 _MM_TRANSPOSE4_PS(Y,F,G,H);
950 Heps = _mm_mul_ps(vfeps,H);
951 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
952 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
953 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
957 fscal = _mm_andnot_ps(dummy_mask,fscal);
959 /* Calculate temporary vectorial force */
960 tx = _mm_mul_ps(fscal,dx00);
961 ty = _mm_mul_ps(fscal,dy00);
962 tz = _mm_mul_ps(fscal,dz00);
964 /* Update vectorial force */
965 fix0 = _mm_add_ps(fix0,tx);
966 fiy0 = _mm_add_ps(fiy0,ty);
967 fiz0 = _mm_add_ps(fiz0,tz);
969 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
970 f+j_coord_offsetC,f+j_coord_offsetD,
973 /**************************
974 * CALCULATE INTERACTIONS *
975 **************************/
977 r10 = _mm_mul_ps(rsq10,rinv10);
978 r10 = _mm_andnot_ps(dummy_mask,r10);
980 /* Compute parameters for interactions between i and j atoms */
981 qq10 = _mm_mul_ps(iq1,jq0);
983 /* Calculate table index by multiplying r with table scale and truncate to integer */
984 rt = _mm_mul_ps(r10,vftabscale);
985 vfitab = _mm_cvttps_epi32(rt);
986 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
987 vfitab = _mm_slli_epi32(vfitab,2);
989 /* CUBIC SPLINE TABLE ELECTROSTATICS */
990 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
991 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
992 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
993 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
994 _MM_TRANSPOSE4_PS(Y,F,G,H);
995 Heps = _mm_mul_ps(vfeps,H);
996 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
997 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
998 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1002 fscal = _mm_andnot_ps(dummy_mask,fscal);
1004 /* Calculate temporary vectorial force */
1005 tx = _mm_mul_ps(fscal,dx10);
1006 ty = _mm_mul_ps(fscal,dy10);
1007 tz = _mm_mul_ps(fscal,dz10);
1009 /* Update vectorial force */
1010 fix1 = _mm_add_ps(fix1,tx);
1011 fiy1 = _mm_add_ps(fiy1,ty);
1012 fiz1 = _mm_add_ps(fiz1,tz);
1014 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1015 f+j_coord_offsetC,f+j_coord_offsetD,
1018 /**************************
1019 * CALCULATE INTERACTIONS *
1020 **************************/
1022 r20 = _mm_mul_ps(rsq20,rinv20);
1023 r20 = _mm_andnot_ps(dummy_mask,r20);
1025 /* Compute parameters for interactions between i and j atoms */
1026 qq20 = _mm_mul_ps(iq2,jq0);
1028 /* Calculate table index by multiplying r with table scale and truncate to integer */
1029 rt = _mm_mul_ps(r20,vftabscale);
1030 vfitab = _mm_cvttps_epi32(rt);
1031 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1032 vfitab = _mm_slli_epi32(vfitab,2);
1034 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1035 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1036 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1037 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1038 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1039 _MM_TRANSPOSE4_PS(Y,F,G,H);
1040 Heps = _mm_mul_ps(vfeps,H);
1041 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1042 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1043 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1047 fscal = _mm_andnot_ps(dummy_mask,fscal);
1049 /* Calculate temporary vectorial force */
1050 tx = _mm_mul_ps(fscal,dx20);
1051 ty = _mm_mul_ps(fscal,dy20);
1052 tz = _mm_mul_ps(fscal,dz20);
1054 /* Update vectorial force */
1055 fix2 = _mm_add_ps(fix2,tx);
1056 fiy2 = _mm_add_ps(fiy2,ty);
1057 fiz2 = _mm_add_ps(fiz2,tz);
1059 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1060 f+j_coord_offsetC,f+j_coord_offsetD,
1063 /* Inner loop uses 120 flops */
1066 /* End of innermost loop */
1068 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1069 f+i_coord_offset,fshift+i_shift_offset);
1071 /* Increment number of inner iterations */
1072 inneriter += j_index_end - j_index_start;
1074 /* Outer loop uses 27 flops */
1077 /* Increment number of outer iterations */
1080 /* Update outer/inner flops */
1082 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*27 + inneriter*120);