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_VdwLJ_GeomW4P1_VF_sse2_single
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: LennardJones
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_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 jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
72 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
76 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
77 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
78 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
79 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
80 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
81 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
82 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
83 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
86 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
89 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
90 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
92 __m128i ifour = _mm_set1_epi32(4);
93 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
95 __m128 dummy_mask,cutoff_mask;
96 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
97 __m128 one = _mm_set1_ps(1.0);
98 __m128 two = _mm_set1_ps(2.0);
104 jindex = nlist->jindex;
106 shiftidx = nlist->shift;
108 shiftvec = fr->shift_vec[0];
109 fshift = fr->fshift[0];
110 facel = _mm_set1_ps(fr->epsfac);
111 charge = mdatoms->chargeA;
112 nvdwtype = fr->ntype;
114 vdwtype = mdatoms->typeA;
116 vftab = kernel_data->table_elec->data;
117 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
119 /* Setup water-specific parameters */
120 inr = nlist->iinr[0];
121 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
122 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
123 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
124 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
126 /* Avoid stupid compiler warnings */
127 jnrA = jnrB = jnrC = jnrD = 0;
136 for(iidx=0;iidx<4*DIM;iidx++)
141 /* Start outer loop over neighborlists */
142 for(iidx=0; iidx<nri; iidx++)
144 /* Load shift vector for this list */
145 i_shift_offset = DIM*shiftidx[iidx];
147 /* Load limits for loop over neighbors */
148 j_index_start = jindex[iidx];
149 j_index_end = jindex[iidx+1];
151 /* Get outer coordinate index */
153 i_coord_offset = DIM*inr;
155 /* Load i particle coords and add shift vector */
156 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
157 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
159 fix0 = _mm_setzero_ps();
160 fiy0 = _mm_setzero_ps();
161 fiz0 = _mm_setzero_ps();
162 fix1 = _mm_setzero_ps();
163 fiy1 = _mm_setzero_ps();
164 fiz1 = _mm_setzero_ps();
165 fix2 = _mm_setzero_ps();
166 fiy2 = _mm_setzero_ps();
167 fiz2 = _mm_setzero_ps();
168 fix3 = _mm_setzero_ps();
169 fiy3 = _mm_setzero_ps();
170 fiz3 = _mm_setzero_ps();
172 /* Reset potential sums */
173 velecsum = _mm_setzero_ps();
174 vvdwsum = _mm_setzero_ps();
176 /* Start inner kernel loop */
177 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
180 /* Get j neighbor index, and coordinate index */
185 j_coord_offsetA = DIM*jnrA;
186 j_coord_offsetB = DIM*jnrB;
187 j_coord_offsetC = DIM*jnrC;
188 j_coord_offsetD = DIM*jnrD;
190 /* load j atom coordinates */
191 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
192 x+j_coord_offsetC,x+j_coord_offsetD,
195 /* Calculate displacement vector */
196 dx00 = _mm_sub_ps(ix0,jx0);
197 dy00 = _mm_sub_ps(iy0,jy0);
198 dz00 = _mm_sub_ps(iz0,jz0);
199 dx10 = _mm_sub_ps(ix1,jx0);
200 dy10 = _mm_sub_ps(iy1,jy0);
201 dz10 = _mm_sub_ps(iz1,jz0);
202 dx20 = _mm_sub_ps(ix2,jx0);
203 dy20 = _mm_sub_ps(iy2,jy0);
204 dz20 = _mm_sub_ps(iz2,jz0);
205 dx30 = _mm_sub_ps(ix3,jx0);
206 dy30 = _mm_sub_ps(iy3,jy0);
207 dz30 = _mm_sub_ps(iz3,jz0);
209 /* Calculate squared distance and things based on it */
210 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
211 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
212 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
213 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
215 rinv10 = gmx_mm_invsqrt_ps(rsq10);
216 rinv20 = gmx_mm_invsqrt_ps(rsq20);
217 rinv30 = gmx_mm_invsqrt_ps(rsq30);
219 rinvsq00 = gmx_mm_inv_ps(rsq00);
221 /* Load parameters for j particles */
222 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
223 charge+jnrC+0,charge+jnrD+0);
224 vdwjidx0A = 2*vdwtype[jnrA+0];
225 vdwjidx0B = 2*vdwtype[jnrB+0];
226 vdwjidx0C = 2*vdwtype[jnrC+0];
227 vdwjidx0D = 2*vdwtype[jnrD+0];
229 fjx0 = _mm_setzero_ps();
230 fjy0 = _mm_setzero_ps();
231 fjz0 = _mm_setzero_ps();
233 /**************************
234 * CALCULATE INTERACTIONS *
235 **************************/
237 /* Compute parameters for interactions between i and j atoms */
238 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
239 vdwparam+vdwioffset0+vdwjidx0B,
240 vdwparam+vdwioffset0+vdwjidx0C,
241 vdwparam+vdwioffset0+vdwjidx0D,
244 /* LENNARD-JONES DISPERSION/REPULSION */
246 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
247 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
248 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
249 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
250 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
252 /* Update potential sum for this i atom from the interaction with this j atom. */
253 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
257 /* Calculate temporary vectorial force */
258 tx = _mm_mul_ps(fscal,dx00);
259 ty = _mm_mul_ps(fscal,dy00);
260 tz = _mm_mul_ps(fscal,dz00);
262 /* Update vectorial force */
263 fix0 = _mm_add_ps(fix0,tx);
264 fiy0 = _mm_add_ps(fiy0,ty);
265 fiz0 = _mm_add_ps(fiz0,tz);
267 fjx0 = _mm_add_ps(fjx0,tx);
268 fjy0 = _mm_add_ps(fjy0,ty);
269 fjz0 = _mm_add_ps(fjz0,tz);
271 /**************************
272 * CALCULATE INTERACTIONS *
273 **************************/
275 r10 = _mm_mul_ps(rsq10,rinv10);
277 /* Compute parameters for interactions between i and j atoms */
278 qq10 = _mm_mul_ps(iq1,jq0);
280 /* Calculate table index by multiplying r with table scale and truncate to integer */
281 rt = _mm_mul_ps(r10,vftabscale);
282 vfitab = _mm_cvttps_epi32(rt);
283 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
284 vfitab = _mm_slli_epi32(vfitab,2);
286 /* CUBIC SPLINE TABLE ELECTROSTATICS */
287 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
288 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
289 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
290 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
291 _MM_TRANSPOSE4_PS(Y,F,G,H);
292 Heps = _mm_mul_ps(vfeps,H);
293 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
294 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
295 velec = _mm_mul_ps(qq10,VV);
296 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
297 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
299 /* Update potential sum for this i atom from the interaction with this j atom. */
300 velecsum = _mm_add_ps(velecsum,velec);
304 /* Calculate temporary vectorial force */
305 tx = _mm_mul_ps(fscal,dx10);
306 ty = _mm_mul_ps(fscal,dy10);
307 tz = _mm_mul_ps(fscal,dz10);
309 /* Update vectorial force */
310 fix1 = _mm_add_ps(fix1,tx);
311 fiy1 = _mm_add_ps(fiy1,ty);
312 fiz1 = _mm_add_ps(fiz1,tz);
314 fjx0 = _mm_add_ps(fjx0,tx);
315 fjy0 = _mm_add_ps(fjy0,ty);
316 fjz0 = _mm_add_ps(fjz0,tz);
318 /**************************
319 * CALCULATE INTERACTIONS *
320 **************************/
322 r20 = _mm_mul_ps(rsq20,rinv20);
324 /* Compute parameters for interactions between i and j atoms */
325 qq20 = _mm_mul_ps(iq2,jq0);
327 /* Calculate table index by multiplying r with table scale and truncate to integer */
328 rt = _mm_mul_ps(r20,vftabscale);
329 vfitab = _mm_cvttps_epi32(rt);
330 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
331 vfitab = _mm_slli_epi32(vfitab,2);
333 /* CUBIC SPLINE TABLE ELECTROSTATICS */
334 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
335 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
336 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
337 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
338 _MM_TRANSPOSE4_PS(Y,F,G,H);
339 Heps = _mm_mul_ps(vfeps,H);
340 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
341 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
342 velec = _mm_mul_ps(qq20,VV);
343 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
344 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
346 /* Update potential sum for this i atom from the interaction with this j atom. */
347 velecsum = _mm_add_ps(velecsum,velec);
351 /* Calculate temporary vectorial force */
352 tx = _mm_mul_ps(fscal,dx20);
353 ty = _mm_mul_ps(fscal,dy20);
354 tz = _mm_mul_ps(fscal,dz20);
356 /* Update vectorial force */
357 fix2 = _mm_add_ps(fix2,tx);
358 fiy2 = _mm_add_ps(fiy2,ty);
359 fiz2 = _mm_add_ps(fiz2,tz);
361 fjx0 = _mm_add_ps(fjx0,tx);
362 fjy0 = _mm_add_ps(fjy0,ty);
363 fjz0 = _mm_add_ps(fjz0,tz);
365 /**************************
366 * CALCULATE INTERACTIONS *
367 **************************/
369 r30 = _mm_mul_ps(rsq30,rinv30);
371 /* Compute parameters for interactions between i and j atoms */
372 qq30 = _mm_mul_ps(iq3,jq0);
374 /* Calculate table index by multiplying r with table scale and truncate to integer */
375 rt = _mm_mul_ps(r30,vftabscale);
376 vfitab = _mm_cvttps_epi32(rt);
377 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
378 vfitab = _mm_slli_epi32(vfitab,2);
380 /* CUBIC SPLINE TABLE ELECTROSTATICS */
381 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
382 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
383 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
384 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
385 _MM_TRANSPOSE4_PS(Y,F,G,H);
386 Heps = _mm_mul_ps(vfeps,H);
387 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
388 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
389 velec = _mm_mul_ps(qq30,VV);
390 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
391 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
393 /* Update potential sum for this i atom from the interaction with this j atom. */
394 velecsum = _mm_add_ps(velecsum,velec);
398 /* Calculate temporary vectorial force */
399 tx = _mm_mul_ps(fscal,dx30);
400 ty = _mm_mul_ps(fscal,dy30);
401 tz = _mm_mul_ps(fscal,dz30);
403 /* Update vectorial force */
404 fix3 = _mm_add_ps(fix3,tx);
405 fiy3 = _mm_add_ps(fiy3,ty);
406 fiz3 = _mm_add_ps(fiz3,tz);
408 fjx0 = _mm_add_ps(fjx0,tx);
409 fjy0 = _mm_add_ps(fjy0,ty);
410 fjz0 = _mm_add_ps(fjz0,tz);
412 fjptrA = f+j_coord_offsetA;
413 fjptrB = f+j_coord_offsetB;
414 fjptrC = f+j_coord_offsetC;
415 fjptrD = f+j_coord_offsetD;
417 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
419 /* Inner loop uses 161 flops */
425 /* Get j neighbor index, and coordinate index */
426 jnrlistA = jjnr[jidx];
427 jnrlistB = jjnr[jidx+1];
428 jnrlistC = jjnr[jidx+2];
429 jnrlistD = jjnr[jidx+3];
430 /* Sign of each element will be negative for non-real atoms.
431 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
432 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
434 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
435 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
436 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
437 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
438 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
439 j_coord_offsetA = DIM*jnrA;
440 j_coord_offsetB = DIM*jnrB;
441 j_coord_offsetC = DIM*jnrC;
442 j_coord_offsetD = DIM*jnrD;
444 /* load j atom coordinates */
445 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
446 x+j_coord_offsetC,x+j_coord_offsetD,
449 /* Calculate displacement vector */
450 dx00 = _mm_sub_ps(ix0,jx0);
451 dy00 = _mm_sub_ps(iy0,jy0);
452 dz00 = _mm_sub_ps(iz0,jz0);
453 dx10 = _mm_sub_ps(ix1,jx0);
454 dy10 = _mm_sub_ps(iy1,jy0);
455 dz10 = _mm_sub_ps(iz1,jz0);
456 dx20 = _mm_sub_ps(ix2,jx0);
457 dy20 = _mm_sub_ps(iy2,jy0);
458 dz20 = _mm_sub_ps(iz2,jz0);
459 dx30 = _mm_sub_ps(ix3,jx0);
460 dy30 = _mm_sub_ps(iy3,jy0);
461 dz30 = _mm_sub_ps(iz3,jz0);
463 /* Calculate squared distance and things based on it */
464 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
465 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
466 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
467 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
469 rinv10 = gmx_mm_invsqrt_ps(rsq10);
470 rinv20 = gmx_mm_invsqrt_ps(rsq20);
471 rinv30 = gmx_mm_invsqrt_ps(rsq30);
473 rinvsq00 = gmx_mm_inv_ps(rsq00);
475 /* Load parameters for j particles */
476 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
477 charge+jnrC+0,charge+jnrD+0);
478 vdwjidx0A = 2*vdwtype[jnrA+0];
479 vdwjidx0B = 2*vdwtype[jnrB+0];
480 vdwjidx0C = 2*vdwtype[jnrC+0];
481 vdwjidx0D = 2*vdwtype[jnrD+0];
483 fjx0 = _mm_setzero_ps();
484 fjy0 = _mm_setzero_ps();
485 fjz0 = _mm_setzero_ps();
487 /**************************
488 * CALCULATE INTERACTIONS *
489 **************************/
491 /* Compute parameters for interactions between i and j atoms */
492 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
493 vdwparam+vdwioffset0+vdwjidx0B,
494 vdwparam+vdwioffset0+vdwjidx0C,
495 vdwparam+vdwioffset0+vdwjidx0D,
498 /* LENNARD-JONES DISPERSION/REPULSION */
500 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
501 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
502 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
503 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
504 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
506 /* Update potential sum for this i atom from the interaction with this j atom. */
507 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
508 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
512 fscal = _mm_andnot_ps(dummy_mask,fscal);
514 /* Calculate temporary vectorial force */
515 tx = _mm_mul_ps(fscal,dx00);
516 ty = _mm_mul_ps(fscal,dy00);
517 tz = _mm_mul_ps(fscal,dz00);
519 /* Update vectorial force */
520 fix0 = _mm_add_ps(fix0,tx);
521 fiy0 = _mm_add_ps(fiy0,ty);
522 fiz0 = _mm_add_ps(fiz0,tz);
524 fjx0 = _mm_add_ps(fjx0,tx);
525 fjy0 = _mm_add_ps(fjy0,ty);
526 fjz0 = _mm_add_ps(fjz0,tz);
528 /**************************
529 * CALCULATE INTERACTIONS *
530 **************************/
532 r10 = _mm_mul_ps(rsq10,rinv10);
533 r10 = _mm_andnot_ps(dummy_mask,r10);
535 /* Compute parameters for interactions between i and j atoms */
536 qq10 = _mm_mul_ps(iq1,jq0);
538 /* Calculate table index by multiplying r with table scale and truncate to integer */
539 rt = _mm_mul_ps(r10,vftabscale);
540 vfitab = _mm_cvttps_epi32(rt);
541 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
542 vfitab = _mm_slli_epi32(vfitab,2);
544 /* CUBIC SPLINE TABLE ELECTROSTATICS */
545 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
546 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
547 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
548 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
549 _MM_TRANSPOSE4_PS(Y,F,G,H);
550 Heps = _mm_mul_ps(vfeps,H);
551 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
552 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
553 velec = _mm_mul_ps(qq10,VV);
554 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
555 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
557 /* Update potential sum for this i atom from the interaction with this j atom. */
558 velec = _mm_andnot_ps(dummy_mask,velec);
559 velecsum = _mm_add_ps(velecsum,velec);
563 fscal = _mm_andnot_ps(dummy_mask,fscal);
565 /* Calculate temporary vectorial force */
566 tx = _mm_mul_ps(fscal,dx10);
567 ty = _mm_mul_ps(fscal,dy10);
568 tz = _mm_mul_ps(fscal,dz10);
570 /* Update vectorial force */
571 fix1 = _mm_add_ps(fix1,tx);
572 fiy1 = _mm_add_ps(fiy1,ty);
573 fiz1 = _mm_add_ps(fiz1,tz);
575 fjx0 = _mm_add_ps(fjx0,tx);
576 fjy0 = _mm_add_ps(fjy0,ty);
577 fjz0 = _mm_add_ps(fjz0,tz);
579 /**************************
580 * CALCULATE INTERACTIONS *
581 **************************/
583 r20 = _mm_mul_ps(rsq20,rinv20);
584 r20 = _mm_andnot_ps(dummy_mask,r20);
586 /* Compute parameters for interactions between i and j atoms */
587 qq20 = _mm_mul_ps(iq2,jq0);
589 /* Calculate table index by multiplying r with table scale and truncate to integer */
590 rt = _mm_mul_ps(r20,vftabscale);
591 vfitab = _mm_cvttps_epi32(rt);
592 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
593 vfitab = _mm_slli_epi32(vfitab,2);
595 /* CUBIC SPLINE TABLE ELECTROSTATICS */
596 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
597 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
598 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
599 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
600 _MM_TRANSPOSE4_PS(Y,F,G,H);
601 Heps = _mm_mul_ps(vfeps,H);
602 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
603 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
604 velec = _mm_mul_ps(qq20,VV);
605 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
606 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
608 /* Update potential sum for this i atom from the interaction with this j atom. */
609 velec = _mm_andnot_ps(dummy_mask,velec);
610 velecsum = _mm_add_ps(velecsum,velec);
614 fscal = _mm_andnot_ps(dummy_mask,fscal);
616 /* Calculate temporary vectorial force */
617 tx = _mm_mul_ps(fscal,dx20);
618 ty = _mm_mul_ps(fscal,dy20);
619 tz = _mm_mul_ps(fscal,dz20);
621 /* Update vectorial force */
622 fix2 = _mm_add_ps(fix2,tx);
623 fiy2 = _mm_add_ps(fiy2,ty);
624 fiz2 = _mm_add_ps(fiz2,tz);
626 fjx0 = _mm_add_ps(fjx0,tx);
627 fjy0 = _mm_add_ps(fjy0,ty);
628 fjz0 = _mm_add_ps(fjz0,tz);
630 /**************************
631 * CALCULATE INTERACTIONS *
632 **************************/
634 r30 = _mm_mul_ps(rsq30,rinv30);
635 r30 = _mm_andnot_ps(dummy_mask,r30);
637 /* Compute parameters for interactions between i and j atoms */
638 qq30 = _mm_mul_ps(iq3,jq0);
640 /* Calculate table index by multiplying r with table scale and truncate to integer */
641 rt = _mm_mul_ps(r30,vftabscale);
642 vfitab = _mm_cvttps_epi32(rt);
643 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
644 vfitab = _mm_slli_epi32(vfitab,2);
646 /* CUBIC SPLINE TABLE ELECTROSTATICS */
647 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
648 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
649 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
650 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
651 _MM_TRANSPOSE4_PS(Y,F,G,H);
652 Heps = _mm_mul_ps(vfeps,H);
653 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
654 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
655 velec = _mm_mul_ps(qq30,VV);
656 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
657 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
659 /* Update potential sum for this i atom from the interaction with this j atom. */
660 velec = _mm_andnot_ps(dummy_mask,velec);
661 velecsum = _mm_add_ps(velecsum,velec);
665 fscal = _mm_andnot_ps(dummy_mask,fscal);
667 /* Calculate temporary vectorial force */
668 tx = _mm_mul_ps(fscal,dx30);
669 ty = _mm_mul_ps(fscal,dy30);
670 tz = _mm_mul_ps(fscal,dz30);
672 /* Update vectorial force */
673 fix3 = _mm_add_ps(fix3,tx);
674 fiy3 = _mm_add_ps(fiy3,ty);
675 fiz3 = _mm_add_ps(fiz3,tz);
677 fjx0 = _mm_add_ps(fjx0,tx);
678 fjy0 = _mm_add_ps(fjy0,ty);
679 fjz0 = _mm_add_ps(fjz0,tz);
681 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
682 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
683 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
684 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
686 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
688 /* Inner loop uses 164 flops */
691 /* End of innermost loop */
693 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
694 f+i_coord_offset,fshift+i_shift_offset);
697 /* Update potential energies */
698 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
699 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
701 /* Increment number of inner iterations */
702 inneriter += j_index_end - j_index_start;
704 /* Outer loop uses 26 flops */
707 /* Increment number of outer iterations */
710 /* Update outer/inner flops */
712 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
715 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_single
716 * Electrostatics interaction: CubicSplineTable
717 * VdW interaction: LennardJones
718 * Geometry: Water4-Particle
719 * Calculate force/pot: Force
722 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_single
723 (t_nblist * gmx_restrict nlist,
724 rvec * gmx_restrict xx,
725 rvec * gmx_restrict ff,
726 t_forcerec * gmx_restrict fr,
727 t_mdatoms * gmx_restrict mdatoms,
728 nb_kernel_data_t * gmx_restrict kernel_data,
729 t_nrnb * gmx_restrict nrnb)
731 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
732 * just 0 for non-waters.
733 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
734 * jnr indices corresponding to data put in the four positions in the SIMD register.
736 int i_shift_offset,i_coord_offset,outeriter,inneriter;
737 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
738 int jnrA,jnrB,jnrC,jnrD;
739 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
740 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
741 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
743 real *shiftvec,*fshift,*x,*f;
744 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
746 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
748 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
750 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
752 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
754 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
755 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
756 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
757 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
758 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
759 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
760 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
761 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
764 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
767 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
768 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
770 __m128i ifour = _mm_set1_epi32(4);
771 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
773 __m128 dummy_mask,cutoff_mask;
774 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
775 __m128 one = _mm_set1_ps(1.0);
776 __m128 two = _mm_set1_ps(2.0);
782 jindex = nlist->jindex;
784 shiftidx = nlist->shift;
786 shiftvec = fr->shift_vec[0];
787 fshift = fr->fshift[0];
788 facel = _mm_set1_ps(fr->epsfac);
789 charge = mdatoms->chargeA;
790 nvdwtype = fr->ntype;
792 vdwtype = mdatoms->typeA;
794 vftab = kernel_data->table_elec->data;
795 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
797 /* Setup water-specific parameters */
798 inr = nlist->iinr[0];
799 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
800 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
801 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
802 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
804 /* Avoid stupid compiler warnings */
805 jnrA = jnrB = jnrC = jnrD = 0;
814 for(iidx=0;iidx<4*DIM;iidx++)
819 /* Start outer loop over neighborlists */
820 for(iidx=0; iidx<nri; iidx++)
822 /* Load shift vector for this list */
823 i_shift_offset = DIM*shiftidx[iidx];
825 /* Load limits for loop over neighbors */
826 j_index_start = jindex[iidx];
827 j_index_end = jindex[iidx+1];
829 /* Get outer coordinate index */
831 i_coord_offset = DIM*inr;
833 /* Load i particle coords and add shift vector */
834 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
835 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
837 fix0 = _mm_setzero_ps();
838 fiy0 = _mm_setzero_ps();
839 fiz0 = _mm_setzero_ps();
840 fix1 = _mm_setzero_ps();
841 fiy1 = _mm_setzero_ps();
842 fiz1 = _mm_setzero_ps();
843 fix2 = _mm_setzero_ps();
844 fiy2 = _mm_setzero_ps();
845 fiz2 = _mm_setzero_ps();
846 fix3 = _mm_setzero_ps();
847 fiy3 = _mm_setzero_ps();
848 fiz3 = _mm_setzero_ps();
850 /* Start inner kernel loop */
851 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
854 /* Get j neighbor index, and coordinate index */
859 j_coord_offsetA = DIM*jnrA;
860 j_coord_offsetB = DIM*jnrB;
861 j_coord_offsetC = DIM*jnrC;
862 j_coord_offsetD = DIM*jnrD;
864 /* load j atom coordinates */
865 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
866 x+j_coord_offsetC,x+j_coord_offsetD,
869 /* Calculate displacement vector */
870 dx00 = _mm_sub_ps(ix0,jx0);
871 dy00 = _mm_sub_ps(iy0,jy0);
872 dz00 = _mm_sub_ps(iz0,jz0);
873 dx10 = _mm_sub_ps(ix1,jx0);
874 dy10 = _mm_sub_ps(iy1,jy0);
875 dz10 = _mm_sub_ps(iz1,jz0);
876 dx20 = _mm_sub_ps(ix2,jx0);
877 dy20 = _mm_sub_ps(iy2,jy0);
878 dz20 = _mm_sub_ps(iz2,jz0);
879 dx30 = _mm_sub_ps(ix3,jx0);
880 dy30 = _mm_sub_ps(iy3,jy0);
881 dz30 = _mm_sub_ps(iz3,jz0);
883 /* Calculate squared distance and things based on it */
884 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
885 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
886 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
887 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
889 rinv10 = gmx_mm_invsqrt_ps(rsq10);
890 rinv20 = gmx_mm_invsqrt_ps(rsq20);
891 rinv30 = gmx_mm_invsqrt_ps(rsq30);
893 rinvsq00 = gmx_mm_inv_ps(rsq00);
895 /* Load parameters for j particles */
896 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
897 charge+jnrC+0,charge+jnrD+0);
898 vdwjidx0A = 2*vdwtype[jnrA+0];
899 vdwjidx0B = 2*vdwtype[jnrB+0];
900 vdwjidx0C = 2*vdwtype[jnrC+0];
901 vdwjidx0D = 2*vdwtype[jnrD+0];
903 fjx0 = _mm_setzero_ps();
904 fjy0 = _mm_setzero_ps();
905 fjz0 = _mm_setzero_ps();
907 /**************************
908 * CALCULATE INTERACTIONS *
909 **************************/
911 /* Compute parameters for interactions between i and j atoms */
912 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
913 vdwparam+vdwioffset0+vdwjidx0B,
914 vdwparam+vdwioffset0+vdwjidx0C,
915 vdwparam+vdwioffset0+vdwjidx0D,
918 /* LENNARD-JONES DISPERSION/REPULSION */
920 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
921 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
925 /* Calculate temporary vectorial force */
926 tx = _mm_mul_ps(fscal,dx00);
927 ty = _mm_mul_ps(fscal,dy00);
928 tz = _mm_mul_ps(fscal,dz00);
930 /* Update vectorial force */
931 fix0 = _mm_add_ps(fix0,tx);
932 fiy0 = _mm_add_ps(fiy0,ty);
933 fiz0 = _mm_add_ps(fiz0,tz);
935 fjx0 = _mm_add_ps(fjx0,tx);
936 fjy0 = _mm_add_ps(fjy0,ty);
937 fjz0 = _mm_add_ps(fjz0,tz);
939 /**************************
940 * CALCULATE INTERACTIONS *
941 **************************/
943 r10 = _mm_mul_ps(rsq10,rinv10);
945 /* Compute parameters for interactions between i and j atoms */
946 qq10 = _mm_mul_ps(iq1,jq0);
948 /* Calculate table index by multiplying r with table scale and truncate to integer */
949 rt = _mm_mul_ps(r10,vftabscale);
950 vfitab = _mm_cvttps_epi32(rt);
951 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
952 vfitab = _mm_slli_epi32(vfitab,2);
954 /* CUBIC SPLINE TABLE ELECTROSTATICS */
955 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
956 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
957 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
958 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
959 _MM_TRANSPOSE4_PS(Y,F,G,H);
960 Heps = _mm_mul_ps(vfeps,H);
961 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
962 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
963 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
967 /* Calculate temporary vectorial force */
968 tx = _mm_mul_ps(fscal,dx10);
969 ty = _mm_mul_ps(fscal,dy10);
970 tz = _mm_mul_ps(fscal,dz10);
972 /* Update vectorial force */
973 fix1 = _mm_add_ps(fix1,tx);
974 fiy1 = _mm_add_ps(fiy1,ty);
975 fiz1 = _mm_add_ps(fiz1,tz);
977 fjx0 = _mm_add_ps(fjx0,tx);
978 fjy0 = _mm_add_ps(fjy0,ty);
979 fjz0 = _mm_add_ps(fjz0,tz);
981 /**************************
982 * CALCULATE INTERACTIONS *
983 **************************/
985 r20 = _mm_mul_ps(rsq20,rinv20);
987 /* Compute parameters for interactions between i and j atoms */
988 qq20 = _mm_mul_ps(iq2,jq0);
990 /* Calculate table index by multiplying r with table scale and truncate to integer */
991 rt = _mm_mul_ps(r20,vftabscale);
992 vfitab = _mm_cvttps_epi32(rt);
993 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
994 vfitab = _mm_slli_epi32(vfitab,2);
996 /* CUBIC SPLINE TABLE ELECTROSTATICS */
997 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
998 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
999 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1000 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1001 _MM_TRANSPOSE4_PS(Y,F,G,H);
1002 Heps = _mm_mul_ps(vfeps,H);
1003 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1004 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1005 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1009 /* Calculate temporary vectorial force */
1010 tx = _mm_mul_ps(fscal,dx20);
1011 ty = _mm_mul_ps(fscal,dy20);
1012 tz = _mm_mul_ps(fscal,dz20);
1014 /* Update vectorial force */
1015 fix2 = _mm_add_ps(fix2,tx);
1016 fiy2 = _mm_add_ps(fiy2,ty);
1017 fiz2 = _mm_add_ps(fiz2,tz);
1019 fjx0 = _mm_add_ps(fjx0,tx);
1020 fjy0 = _mm_add_ps(fjy0,ty);
1021 fjz0 = _mm_add_ps(fjz0,tz);
1023 /**************************
1024 * CALCULATE INTERACTIONS *
1025 **************************/
1027 r30 = _mm_mul_ps(rsq30,rinv30);
1029 /* Compute parameters for interactions between i and j atoms */
1030 qq30 = _mm_mul_ps(iq3,jq0);
1032 /* Calculate table index by multiplying r with table scale and truncate to integer */
1033 rt = _mm_mul_ps(r30,vftabscale);
1034 vfitab = _mm_cvttps_epi32(rt);
1035 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1036 vfitab = _mm_slli_epi32(vfitab,2);
1038 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1039 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1040 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1041 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1042 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1043 _MM_TRANSPOSE4_PS(Y,F,G,H);
1044 Heps = _mm_mul_ps(vfeps,H);
1045 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1046 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1047 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1051 /* Calculate temporary vectorial force */
1052 tx = _mm_mul_ps(fscal,dx30);
1053 ty = _mm_mul_ps(fscal,dy30);
1054 tz = _mm_mul_ps(fscal,dz30);
1056 /* Update vectorial force */
1057 fix3 = _mm_add_ps(fix3,tx);
1058 fiy3 = _mm_add_ps(fiy3,ty);
1059 fiz3 = _mm_add_ps(fiz3,tz);
1061 fjx0 = _mm_add_ps(fjx0,tx);
1062 fjy0 = _mm_add_ps(fjy0,ty);
1063 fjz0 = _mm_add_ps(fjz0,tz);
1065 fjptrA = f+j_coord_offsetA;
1066 fjptrB = f+j_coord_offsetB;
1067 fjptrC = f+j_coord_offsetC;
1068 fjptrD = f+j_coord_offsetD;
1070 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1072 /* Inner loop uses 144 flops */
1075 if(jidx<j_index_end)
1078 /* Get j neighbor index, and coordinate index */
1079 jnrlistA = jjnr[jidx];
1080 jnrlistB = jjnr[jidx+1];
1081 jnrlistC = jjnr[jidx+2];
1082 jnrlistD = jjnr[jidx+3];
1083 /* Sign of each element will be negative for non-real atoms.
1084 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1085 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1087 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1088 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1089 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1090 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1091 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1092 j_coord_offsetA = DIM*jnrA;
1093 j_coord_offsetB = DIM*jnrB;
1094 j_coord_offsetC = DIM*jnrC;
1095 j_coord_offsetD = DIM*jnrD;
1097 /* load j atom coordinates */
1098 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1099 x+j_coord_offsetC,x+j_coord_offsetD,
1102 /* Calculate displacement vector */
1103 dx00 = _mm_sub_ps(ix0,jx0);
1104 dy00 = _mm_sub_ps(iy0,jy0);
1105 dz00 = _mm_sub_ps(iz0,jz0);
1106 dx10 = _mm_sub_ps(ix1,jx0);
1107 dy10 = _mm_sub_ps(iy1,jy0);
1108 dz10 = _mm_sub_ps(iz1,jz0);
1109 dx20 = _mm_sub_ps(ix2,jx0);
1110 dy20 = _mm_sub_ps(iy2,jy0);
1111 dz20 = _mm_sub_ps(iz2,jz0);
1112 dx30 = _mm_sub_ps(ix3,jx0);
1113 dy30 = _mm_sub_ps(iy3,jy0);
1114 dz30 = _mm_sub_ps(iz3,jz0);
1116 /* Calculate squared distance and things based on it */
1117 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1118 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1119 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1120 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1122 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1123 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1124 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1126 rinvsq00 = gmx_mm_inv_ps(rsq00);
1128 /* Load parameters for j particles */
1129 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1130 charge+jnrC+0,charge+jnrD+0);
1131 vdwjidx0A = 2*vdwtype[jnrA+0];
1132 vdwjidx0B = 2*vdwtype[jnrB+0];
1133 vdwjidx0C = 2*vdwtype[jnrC+0];
1134 vdwjidx0D = 2*vdwtype[jnrD+0];
1136 fjx0 = _mm_setzero_ps();
1137 fjy0 = _mm_setzero_ps();
1138 fjz0 = _mm_setzero_ps();
1140 /**************************
1141 * CALCULATE INTERACTIONS *
1142 **************************/
1144 /* Compute parameters for interactions between i and j atoms */
1145 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1146 vdwparam+vdwioffset0+vdwjidx0B,
1147 vdwparam+vdwioffset0+vdwjidx0C,
1148 vdwparam+vdwioffset0+vdwjidx0D,
1151 /* LENNARD-JONES DISPERSION/REPULSION */
1153 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1154 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1158 fscal = _mm_andnot_ps(dummy_mask,fscal);
1160 /* Calculate temporary vectorial force */
1161 tx = _mm_mul_ps(fscal,dx00);
1162 ty = _mm_mul_ps(fscal,dy00);
1163 tz = _mm_mul_ps(fscal,dz00);
1165 /* Update vectorial force */
1166 fix0 = _mm_add_ps(fix0,tx);
1167 fiy0 = _mm_add_ps(fiy0,ty);
1168 fiz0 = _mm_add_ps(fiz0,tz);
1170 fjx0 = _mm_add_ps(fjx0,tx);
1171 fjy0 = _mm_add_ps(fjy0,ty);
1172 fjz0 = _mm_add_ps(fjz0,tz);
1174 /**************************
1175 * CALCULATE INTERACTIONS *
1176 **************************/
1178 r10 = _mm_mul_ps(rsq10,rinv10);
1179 r10 = _mm_andnot_ps(dummy_mask,r10);
1181 /* Compute parameters for interactions between i and j atoms */
1182 qq10 = _mm_mul_ps(iq1,jq0);
1184 /* Calculate table index by multiplying r with table scale and truncate to integer */
1185 rt = _mm_mul_ps(r10,vftabscale);
1186 vfitab = _mm_cvttps_epi32(rt);
1187 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1188 vfitab = _mm_slli_epi32(vfitab,2);
1190 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1191 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1192 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1193 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1194 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1195 _MM_TRANSPOSE4_PS(Y,F,G,H);
1196 Heps = _mm_mul_ps(vfeps,H);
1197 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1198 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1199 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1203 fscal = _mm_andnot_ps(dummy_mask,fscal);
1205 /* Calculate temporary vectorial force */
1206 tx = _mm_mul_ps(fscal,dx10);
1207 ty = _mm_mul_ps(fscal,dy10);
1208 tz = _mm_mul_ps(fscal,dz10);
1210 /* Update vectorial force */
1211 fix1 = _mm_add_ps(fix1,tx);
1212 fiy1 = _mm_add_ps(fiy1,ty);
1213 fiz1 = _mm_add_ps(fiz1,tz);
1215 fjx0 = _mm_add_ps(fjx0,tx);
1216 fjy0 = _mm_add_ps(fjy0,ty);
1217 fjz0 = _mm_add_ps(fjz0,tz);
1219 /**************************
1220 * CALCULATE INTERACTIONS *
1221 **************************/
1223 r20 = _mm_mul_ps(rsq20,rinv20);
1224 r20 = _mm_andnot_ps(dummy_mask,r20);
1226 /* Compute parameters for interactions between i and j atoms */
1227 qq20 = _mm_mul_ps(iq2,jq0);
1229 /* Calculate table index by multiplying r with table scale and truncate to integer */
1230 rt = _mm_mul_ps(r20,vftabscale);
1231 vfitab = _mm_cvttps_epi32(rt);
1232 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1233 vfitab = _mm_slli_epi32(vfitab,2);
1235 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1236 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1237 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1238 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1239 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1240 _MM_TRANSPOSE4_PS(Y,F,G,H);
1241 Heps = _mm_mul_ps(vfeps,H);
1242 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1243 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1244 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1248 fscal = _mm_andnot_ps(dummy_mask,fscal);
1250 /* Calculate temporary vectorial force */
1251 tx = _mm_mul_ps(fscal,dx20);
1252 ty = _mm_mul_ps(fscal,dy20);
1253 tz = _mm_mul_ps(fscal,dz20);
1255 /* Update vectorial force */
1256 fix2 = _mm_add_ps(fix2,tx);
1257 fiy2 = _mm_add_ps(fiy2,ty);
1258 fiz2 = _mm_add_ps(fiz2,tz);
1260 fjx0 = _mm_add_ps(fjx0,tx);
1261 fjy0 = _mm_add_ps(fjy0,ty);
1262 fjz0 = _mm_add_ps(fjz0,tz);
1264 /**************************
1265 * CALCULATE INTERACTIONS *
1266 **************************/
1268 r30 = _mm_mul_ps(rsq30,rinv30);
1269 r30 = _mm_andnot_ps(dummy_mask,r30);
1271 /* Compute parameters for interactions between i and j atoms */
1272 qq30 = _mm_mul_ps(iq3,jq0);
1274 /* Calculate table index by multiplying r with table scale and truncate to integer */
1275 rt = _mm_mul_ps(r30,vftabscale);
1276 vfitab = _mm_cvttps_epi32(rt);
1277 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1278 vfitab = _mm_slli_epi32(vfitab,2);
1280 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1281 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1282 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1283 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1284 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1285 _MM_TRANSPOSE4_PS(Y,F,G,H);
1286 Heps = _mm_mul_ps(vfeps,H);
1287 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1288 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1289 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1293 fscal = _mm_andnot_ps(dummy_mask,fscal);
1295 /* Calculate temporary vectorial force */
1296 tx = _mm_mul_ps(fscal,dx30);
1297 ty = _mm_mul_ps(fscal,dy30);
1298 tz = _mm_mul_ps(fscal,dz30);
1300 /* Update vectorial force */
1301 fix3 = _mm_add_ps(fix3,tx);
1302 fiy3 = _mm_add_ps(fiy3,ty);
1303 fiz3 = _mm_add_ps(fiz3,tz);
1305 fjx0 = _mm_add_ps(fjx0,tx);
1306 fjy0 = _mm_add_ps(fjy0,ty);
1307 fjz0 = _mm_add_ps(fjz0,tz);
1309 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1310 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1311 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1312 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1314 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1316 /* Inner loop uses 147 flops */
1319 /* End of innermost loop */
1321 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1322 f+i_coord_offset,fshift+i_shift_offset);
1324 /* Increment number of inner iterations */
1325 inneriter += j_index_end - j_index_start;
1327 /* Outer loop uses 24 flops */
1330 /* Increment number of outer iterations */
1333 /* Update outer/inner flops */
1335 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);