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_ElecEw_VdwCSTab_GeomW3P1_VF_sse2_single
38 * Electrostatics interaction: Ewald
39 * VdW interaction: CubicSplineTable
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEw_VdwCSTab_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 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
83 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
84 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
86 __m128i ifour = _mm_set1_epi32(4);
87 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
90 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
92 __m128 dummy_mask,cutoff_mask;
93 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
94 __m128 one = _mm_set1_ps(1.0);
95 __m128 two = _mm_set1_ps(2.0);
101 jindex = nlist->jindex;
103 shiftidx = nlist->shift;
105 shiftvec = fr->shift_vec[0];
106 fshift = fr->fshift[0];
107 facel = _mm_set1_ps(fr->epsfac);
108 charge = mdatoms->chargeA;
109 nvdwtype = fr->ntype;
111 vdwtype = mdatoms->typeA;
113 vftab = kernel_data->table_vdw->data;
114 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
116 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
117 ewtab = fr->ic->tabq_coul_FDV0;
118 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
119 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
121 /* Setup water-specific parameters */
122 inr = nlist->iinr[0];
123 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
124 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
125 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
126 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
128 /* Avoid stupid compiler warnings */
129 jnrA = jnrB = jnrC = jnrD = 0;
138 /* Start outer loop over neighborlists */
139 for(iidx=0; iidx<nri; iidx++)
141 /* Load shift vector for this list */
142 i_shift_offset = DIM*shiftidx[iidx];
143 shX = shiftvec[i_shift_offset+XX];
144 shY = shiftvec[i_shift_offset+YY];
145 shZ = shiftvec[i_shift_offset+ZZ];
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 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
157 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
158 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
159 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
160 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
161 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
162 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
163 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
164 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
166 fix0 = _mm_setzero_ps();
167 fiy0 = _mm_setzero_ps();
168 fiz0 = _mm_setzero_ps();
169 fix1 = _mm_setzero_ps();
170 fiy1 = _mm_setzero_ps();
171 fiz1 = _mm_setzero_ps();
172 fix2 = _mm_setzero_ps();
173 fiy2 = _mm_setzero_ps();
174 fiz2 = _mm_setzero_ps();
176 /* Reset potential sums */
177 velecsum = _mm_setzero_ps();
178 vvdwsum = _mm_setzero_ps();
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
184 /* Get j neighbor index, and coordinate index */
190 j_coord_offsetA = DIM*jnrA;
191 j_coord_offsetB = DIM*jnrB;
192 j_coord_offsetC = DIM*jnrC;
193 j_coord_offsetD = DIM*jnrD;
195 /* load j atom coordinates */
196 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
197 x+j_coord_offsetC,x+j_coord_offsetD,
200 /* Calculate displacement vector */
201 dx00 = _mm_sub_ps(ix0,jx0);
202 dy00 = _mm_sub_ps(iy0,jy0);
203 dz00 = _mm_sub_ps(iz0,jz0);
204 dx10 = _mm_sub_ps(ix1,jx0);
205 dy10 = _mm_sub_ps(iy1,jy0);
206 dz10 = _mm_sub_ps(iz1,jz0);
207 dx20 = _mm_sub_ps(ix2,jx0);
208 dy20 = _mm_sub_ps(iy2,jy0);
209 dz20 = _mm_sub_ps(iz2,jz0);
211 /* Calculate squared distance and things based on it */
212 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
213 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
214 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
216 rinv00 = gmx_mm_invsqrt_ps(rsq00);
217 rinv10 = gmx_mm_invsqrt_ps(rsq10);
218 rinv20 = gmx_mm_invsqrt_ps(rsq20);
220 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
221 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
222 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
224 /* Load parameters for j particles */
225 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
226 charge+jnrC+0,charge+jnrD+0);
227 vdwjidx0A = 2*vdwtype[jnrA+0];
228 vdwjidx0B = 2*vdwtype[jnrB+0];
229 vdwjidx0C = 2*vdwtype[jnrC+0];
230 vdwjidx0D = 2*vdwtype[jnrD+0];
232 /**************************
233 * CALCULATE INTERACTIONS *
234 **************************/
236 r00 = _mm_mul_ps(rsq00,rinv00);
238 /* Compute parameters for interactions between i and j atoms */
239 qq00 = _mm_mul_ps(iq0,jq0);
240 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
241 vdwparam+vdwioffset0+vdwjidx0B,
242 vdwparam+vdwioffset0+vdwjidx0C,
243 vdwparam+vdwioffset0+vdwjidx0D,
246 /* Calculate table index by multiplying r with table scale and truncate to integer */
247 rt = _mm_mul_ps(r00,vftabscale);
248 vfitab = _mm_cvttps_epi32(rt);
249 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
250 vfitab = _mm_slli_epi32(vfitab,3);
252 /* EWALD ELECTROSTATICS */
254 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
255 ewrt = _mm_mul_ps(r00,ewtabscale);
256 ewitab = _mm_cvttps_epi32(ewrt);
257 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
258 ewitab = _mm_slli_epi32(ewitab,2);
259 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
260 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
261 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
262 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
263 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
264 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
265 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
266 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
267 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
269 /* CUBIC SPLINE TABLE DISPERSION */
270 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
271 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
272 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
273 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
274 _MM_TRANSPOSE4_PS(Y,F,G,H);
275 Heps = _mm_mul_ps(vfeps,H);
276 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
277 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
278 vvdw6 = _mm_mul_ps(c6_00,VV);
279 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
280 fvdw6 = _mm_mul_ps(c6_00,FF);
282 /* CUBIC SPLINE TABLE REPULSION */
283 vfitab = _mm_add_epi32(vfitab,ifour);
284 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
285 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
286 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
287 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
288 _MM_TRANSPOSE4_PS(Y,F,G,H);
289 Heps = _mm_mul_ps(vfeps,H);
290 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
291 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
292 vvdw12 = _mm_mul_ps(c12_00,VV);
293 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
294 fvdw12 = _mm_mul_ps(c12_00,FF);
295 vvdw = _mm_add_ps(vvdw12,vvdw6);
296 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
298 /* Update potential sum for this i atom from the interaction with this j atom. */
299 velecsum = _mm_add_ps(velecsum,velec);
300 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
302 fscal = _mm_add_ps(felec,fvdw);
304 /* Calculate temporary vectorial force */
305 tx = _mm_mul_ps(fscal,dx00);
306 ty = _mm_mul_ps(fscal,dy00);
307 tz = _mm_mul_ps(fscal,dz00);
309 /* Update vectorial force */
310 fix0 = _mm_add_ps(fix0,tx);
311 fiy0 = _mm_add_ps(fiy0,ty);
312 fiz0 = _mm_add_ps(fiz0,tz);
314 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
315 f+j_coord_offsetC,f+j_coord_offsetD,
318 /**************************
319 * CALCULATE INTERACTIONS *
320 **************************/
322 r10 = _mm_mul_ps(rsq10,rinv10);
324 /* Compute parameters for interactions between i and j atoms */
325 qq10 = _mm_mul_ps(iq1,jq0);
327 /* EWALD ELECTROSTATICS */
329 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
330 ewrt = _mm_mul_ps(r10,ewtabscale);
331 ewitab = _mm_cvttps_epi32(ewrt);
332 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
333 ewitab = _mm_slli_epi32(ewitab,2);
334 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
335 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
336 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
337 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
338 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
339 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
340 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
341 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
342 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
344 /* Update potential sum for this i atom from the interaction with this j atom. */
345 velecsum = _mm_add_ps(velecsum,velec);
349 /* Calculate temporary vectorial force */
350 tx = _mm_mul_ps(fscal,dx10);
351 ty = _mm_mul_ps(fscal,dy10);
352 tz = _mm_mul_ps(fscal,dz10);
354 /* Update vectorial force */
355 fix1 = _mm_add_ps(fix1,tx);
356 fiy1 = _mm_add_ps(fiy1,ty);
357 fiz1 = _mm_add_ps(fiz1,tz);
359 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
360 f+j_coord_offsetC,f+j_coord_offsetD,
363 /**************************
364 * CALCULATE INTERACTIONS *
365 **************************/
367 r20 = _mm_mul_ps(rsq20,rinv20);
369 /* Compute parameters for interactions between i and j atoms */
370 qq20 = _mm_mul_ps(iq2,jq0);
372 /* EWALD ELECTROSTATICS */
374 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
375 ewrt = _mm_mul_ps(r20,ewtabscale);
376 ewitab = _mm_cvttps_epi32(ewrt);
377 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
378 ewitab = _mm_slli_epi32(ewitab,2);
379 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
380 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
381 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
382 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
383 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
384 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
385 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
386 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
387 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
389 /* Update potential sum for this i atom from the interaction with this j atom. */
390 velecsum = _mm_add_ps(velecsum,velec);
394 /* Calculate temporary vectorial force */
395 tx = _mm_mul_ps(fscal,dx20);
396 ty = _mm_mul_ps(fscal,dy20);
397 tz = _mm_mul_ps(fscal,dz20);
399 /* Update vectorial force */
400 fix2 = _mm_add_ps(fix2,tx);
401 fiy2 = _mm_add_ps(fiy2,ty);
402 fiz2 = _mm_add_ps(fiz2,tz);
404 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
405 f+j_coord_offsetC,f+j_coord_offsetD,
408 /* Inner loop uses 157 flops */
414 /* Get j neighbor index, and coordinate index */
420 /* Sign of each element will be negative for non-real atoms.
421 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
422 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
424 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
425 jnrA = (jnrA>=0) ? jnrA : 0;
426 jnrB = (jnrB>=0) ? jnrB : 0;
427 jnrC = (jnrC>=0) ? jnrC : 0;
428 jnrD = (jnrD>=0) ? jnrD : 0;
430 j_coord_offsetA = DIM*jnrA;
431 j_coord_offsetB = DIM*jnrB;
432 j_coord_offsetC = DIM*jnrC;
433 j_coord_offsetD = DIM*jnrD;
435 /* load j atom coordinates */
436 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
437 x+j_coord_offsetC,x+j_coord_offsetD,
440 /* Calculate displacement vector */
441 dx00 = _mm_sub_ps(ix0,jx0);
442 dy00 = _mm_sub_ps(iy0,jy0);
443 dz00 = _mm_sub_ps(iz0,jz0);
444 dx10 = _mm_sub_ps(ix1,jx0);
445 dy10 = _mm_sub_ps(iy1,jy0);
446 dz10 = _mm_sub_ps(iz1,jz0);
447 dx20 = _mm_sub_ps(ix2,jx0);
448 dy20 = _mm_sub_ps(iy2,jy0);
449 dz20 = _mm_sub_ps(iz2,jz0);
451 /* Calculate squared distance and things based on it */
452 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
453 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
454 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
456 rinv00 = gmx_mm_invsqrt_ps(rsq00);
457 rinv10 = gmx_mm_invsqrt_ps(rsq10);
458 rinv20 = gmx_mm_invsqrt_ps(rsq20);
460 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
461 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
462 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
464 /* Load parameters for j particles */
465 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
466 charge+jnrC+0,charge+jnrD+0);
467 vdwjidx0A = 2*vdwtype[jnrA+0];
468 vdwjidx0B = 2*vdwtype[jnrB+0];
469 vdwjidx0C = 2*vdwtype[jnrC+0];
470 vdwjidx0D = 2*vdwtype[jnrD+0];
472 /**************************
473 * CALCULATE INTERACTIONS *
474 **************************/
476 r00 = _mm_mul_ps(rsq00,rinv00);
477 r00 = _mm_andnot_ps(dummy_mask,r00);
479 /* Compute parameters for interactions between i and j atoms */
480 qq00 = _mm_mul_ps(iq0,jq0);
481 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
482 vdwparam+vdwioffset0+vdwjidx0B,
483 vdwparam+vdwioffset0+vdwjidx0C,
484 vdwparam+vdwioffset0+vdwjidx0D,
487 /* Calculate table index by multiplying r with table scale and truncate to integer */
488 rt = _mm_mul_ps(r00,vftabscale);
489 vfitab = _mm_cvttps_epi32(rt);
490 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
491 vfitab = _mm_slli_epi32(vfitab,3);
493 /* EWALD ELECTROSTATICS */
495 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
496 ewrt = _mm_mul_ps(r00,ewtabscale);
497 ewitab = _mm_cvttps_epi32(ewrt);
498 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
499 ewitab = _mm_slli_epi32(ewitab,2);
500 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
501 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
502 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
503 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
504 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
505 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
506 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
507 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
508 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
510 /* CUBIC SPLINE TABLE DISPERSION */
511 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
512 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
513 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
514 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
515 _MM_TRANSPOSE4_PS(Y,F,G,H);
516 Heps = _mm_mul_ps(vfeps,H);
517 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
518 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
519 vvdw6 = _mm_mul_ps(c6_00,VV);
520 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
521 fvdw6 = _mm_mul_ps(c6_00,FF);
523 /* CUBIC SPLINE TABLE REPULSION */
524 vfitab = _mm_add_epi32(vfitab,ifour);
525 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
526 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
527 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
528 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
529 _MM_TRANSPOSE4_PS(Y,F,G,H);
530 Heps = _mm_mul_ps(vfeps,H);
531 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
532 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
533 vvdw12 = _mm_mul_ps(c12_00,VV);
534 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
535 fvdw12 = _mm_mul_ps(c12_00,FF);
536 vvdw = _mm_add_ps(vvdw12,vvdw6);
537 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
539 /* Update potential sum for this i atom from the interaction with this j atom. */
540 velec = _mm_andnot_ps(dummy_mask,velec);
541 velecsum = _mm_add_ps(velecsum,velec);
542 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
543 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
545 fscal = _mm_add_ps(felec,fvdw);
547 fscal = _mm_andnot_ps(dummy_mask,fscal);
549 /* Calculate temporary vectorial force */
550 tx = _mm_mul_ps(fscal,dx00);
551 ty = _mm_mul_ps(fscal,dy00);
552 tz = _mm_mul_ps(fscal,dz00);
554 /* Update vectorial force */
555 fix0 = _mm_add_ps(fix0,tx);
556 fiy0 = _mm_add_ps(fiy0,ty);
557 fiz0 = _mm_add_ps(fiz0,tz);
559 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
560 f+j_coord_offsetC,f+j_coord_offsetD,
563 /**************************
564 * CALCULATE INTERACTIONS *
565 **************************/
567 r10 = _mm_mul_ps(rsq10,rinv10);
568 r10 = _mm_andnot_ps(dummy_mask,r10);
570 /* Compute parameters for interactions between i and j atoms */
571 qq10 = _mm_mul_ps(iq1,jq0);
573 /* EWALD ELECTROSTATICS */
575 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
576 ewrt = _mm_mul_ps(r10,ewtabscale);
577 ewitab = _mm_cvttps_epi32(ewrt);
578 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
579 ewitab = _mm_slli_epi32(ewitab,2);
580 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
581 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
582 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
583 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
584 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
585 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
586 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
587 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
588 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
590 /* Update potential sum for this i atom from the interaction with this j atom. */
591 velec = _mm_andnot_ps(dummy_mask,velec);
592 velecsum = _mm_add_ps(velecsum,velec);
596 fscal = _mm_andnot_ps(dummy_mask,fscal);
598 /* Calculate temporary vectorial force */
599 tx = _mm_mul_ps(fscal,dx10);
600 ty = _mm_mul_ps(fscal,dy10);
601 tz = _mm_mul_ps(fscal,dz10);
603 /* Update vectorial force */
604 fix1 = _mm_add_ps(fix1,tx);
605 fiy1 = _mm_add_ps(fiy1,ty);
606 fiz1 = _mm_add_ps(fiz1,tz);
608 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
609 f+j_coord_offsetC,f+j_coord_offsetD,
612 /**************************
613 * CALCULATE INTERACTIONS *
614 **************************/
616 r20 = _mm_mul_ps(rsq20,rinv20);
617 r20 = _mm_andnot_ps(dummy_mask,r20);
619 /* Compute parameters for interactions between i and j atoms */
620 qq20 = _mm_mul_ps(iq2,jq0);
622 /* EWALD ELECTROSTATICS */
624 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
625 ewrt = _mm_mul_ps(r20,ewtabscale);
626 ewitab = _mm_cvttps_epi32(ewrt);
627 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
628 ewitab = _mm_slli_epi32(ewitab,2);
629 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
630 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
631 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
632 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
633 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
634 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
635 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
636 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
637 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
639 /* Update potential sum for this i atom from the interaction with this j atom. */
640 velec = _mm_andnot_ps(dummy_mask,velec);
641 velecsum = _mm_add_ps(velecsum,velec);
645 fscal = _mm_andnot_ps(dummy_mask,fscal);
647 /* Calculate temporary vectorial force */
648 tx = _mm_mul_ps(fscal,dx20);
649 ty = _mm_mul_ps(fscal,dy20);
650 tz = _mm_mul_ps(fscal,dz20);
652 /* Update vectorial force */
653 fix2 = _mm_add_ps(fix2,tx);
654 fiy2 = _mm_add_ps(fiy2,ty);
655 fiz2 = _mm_add_ps(fiz2,tz);
657 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
658 f+j_coord_offsetC,f+j_coord_offsetD,
661 /* Inner loop uses 160 flops */
664 /* End of innermost loop */
666 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
667 f+i_coord_offset,fshift+i_shift_offset);
670 /* Update potential energies */
671 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
672 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
674 /* Increment number of inner iterations */
675 inneriter += j_index_end - j_index_start;
677 /* Outer loop uses 29 flops */
680 /* Increment number of outer iterations */
683 /* Update outer/inner flops */
685 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*160);
688 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_single
689 * Electrostatics interaction: Ewald
690 * VdW interaction: CubicSplineTable
691 * Geometry: Water3-Particle
692 * Calculate force/pot: Force
695 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_single
696 (t_nblist * gmx_restrict nlist,
697 rvec * gmx_restrict xx,
698 rvec * gmx_restrict ff,
699 t_forcerec * gmx_restrict fr,
700 t_mdatoms * gmx_restrict mdatoms,
701 nb_kernel_data_t * gmx_restrict kernel_data,
702 t_nrnb * gmx_restrict nrnb)
704 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
705 * just 0 for non-waters.
706 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
707 * jnr indices corresponding to data put in the four positions in the SIMD register.
709 int i_shift_offset,i_coord_offset,outeriter,inneriter;
710 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
711 int jnrA,jnrB,jnrC,jnrD;
712 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
713 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
714 real shX,shY,shZ,rcutoff_scalar;
715 real *shiftvec,*fshift,*x,*f;
716 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
718 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
720 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
722 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
723 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
724 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
725 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
726 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
727 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
728 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
731 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
734 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
735 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
737 __m128i ifour = _mm_set1_epi32(4);
738 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
741 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
743 __m128 dummy_mask,cutoff_mask;
744 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
745 __m128 one = _mm_set1_ps(1.0);
746 __m128 two = _mm_set1_ps(2.0);
752 jindex = nlist->jindex;
754 shiftidx = nlist->shift;
756 shiftvec = fr->shift_vec[0];
757 fshift = fr->fshift[0];
758 facel = _mm_set1_ps(fr->epsfac);
759 charge = mdatoms->chargeA;
760 nvdwtype = fr->ntype;
762 vdwtype = mdatoms->typeA;
764 vftab = kernel_data->table_vdw->data;
765 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
767 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
768 ewtab = fr->ic->tabq_coul_F;
769 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
770 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
772 /* Setup water-specific parameters */
773 inr = nlist->iinr[0];
774 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
775 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
776 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
777 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
779 /* Avoid stupid compiler warnings */
780 jnrA = jnrB = jnrC = jnrD = 0;
789 /* Start outer loop over neighborlists */
790 for(iidx=0; iidx<nri; iidx++)
792 /* Load shift vector for this list */
793 i_shift_offset = DIM*shiftidx[iidx];
794 shX = shiftvec[i_shift_offset+XX];
795 shY = shiftvec[i_shift_offset+YY];
796 shZ = shiftvec[i_shift_offset+ZZ];
798 /* Load limits for loop over neighbors */
799 j_index_start = jindex[iidx];
800 j_index_end = jindex[iidx+1];
802 /* Get outer coordinate index */
804 i_coord_offset = DIM*inr;
806 /* Load i particle coords and add shift vector */
807 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
808 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
809 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
810 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
811 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
812 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
813 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
814 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
815 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
817 fix0 = _mm_setzero_ps();
818 fiy0 = _mm_setzero_ps();
819 fiz0 = _mm_setzero_ps();
820 fix1 = _mm_setzero_ps();
821 fiy1 = _mm_setzero_ps();
822 fiz1 = _mm_setzero_ps();
823 fix2 = _mm_setzero_ps();
824 fiy2 = _mm_setzero_ps();
825 fiz2 = _mm_setzero_ps();
827 /* Start inner kernel loop */
828 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
831 /* Get j neighbor index, and coordinate index */
837 j_coord_offsetA = DIM*jnrA;
838 j_coord_offsetB = DIM*jnrB;
839 j_coord_offsetC = DIM*jnrC;
840 j_coord_offsetD = DIM*jnrD;
842 /* load j atom coordinates */
843 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
844 x+j_coord_offsetC,x+j_coord_offsetD,
847 /* Calculate displacement vector */
848 dx00 = _mm_sub_ps(ix0,jx0);
849 dy00 = _mm_sub_ps(iy0,jy0);
850 dz00 = _mm_sub_ps(iz0,jz0);
851 dx10 = _mm_sub_ps(ix1,jx0);
852 dy10 = _mm_sub_ps(iy1,jy0);
853 dz10 = _mm_sub_ps(iz1,jz0);
854 dx20 = _mm_sub_ps(ix2,jx0);
855 dy20 = _mm_sub_ps(iy2,jy0);
856 dz20 = _mm_sub_ps(iz2,jz0);
858 /* Calculate squared distance and things based on it */
859 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
860 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
861 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
863 rinv00 = gmx_mm_invsqrt_ps(rsq00);
864 rinv10 = gmx_mm_invsqrt_ps(rsq10);
865 rinv20 = gmx_mm_invsqrt_ps(rsq20);
867 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
868 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
869 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
871 /* Load parameters for j particles */
872 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
873 charge+jnrC+0,charge+jnrD+0);
874 vdwjidx0A = 2*vdwtype[jnrA+0];
875 vdwjidx0B = 2*vdwtype[jnrB+0];
876 vdwjidx0C = 2*vdwtype[jnrC+0];
877 vdwjidx0D = 2*vdwtype[jnrD+0];
879 /**************************
880 * CALCULATE INTERACTIONS *
881 **************************/
883 r00 = _mm_mul_ps(rsq00,rinv00);
885 /* Compute parameters for interactions between i and j atoms */
886 qq00 = _mm_mul_ps(iq0,jq0);
887 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
888 vdwparam+vdwioffset0+vdwjidx0B,
889 vdwparam+vdwioffset0+vdwjidx0C,
890 vdwparam+vdwioffset0+vdwjidx0D,
893 /* Calculate table index by multiplying r with table scale and truncate to integer */
894 rt = _mm_mul_ps(r00,vftabscale);
895 vfitab = _mm_cvttps_epi32(rt);
896 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
897 vfitab = _mm_slli_epi32(vfitab,3);
899 /* EWALD ELECTROSTATICS */
901 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
902 ewrt = _mm_mul_ps(r00,ewtabscale);
903 ewitab = _mm_cvttps_epi32(ewrt);
904 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
905 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
906 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
908 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
909 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
911 /* CUBIC SPLINE TABLE DISPERSION */
912 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
913 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
914 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
915 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
916 _MM_TRANSPOSE4_PS(Y,F,G,H);
917 Heps = _mm_mul_ps(vfeps,H);
918 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
919 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
920 fvdw6 = _mm_mul_ps(c6_00,FF);
922 /* CUBIC SPLINE TABLE REPULSION */
923 vfitab = _mm_add_epi32(vfitab,ifour);
924 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
925 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
926 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
927 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
928 _MM_TRANSPOSE4_PS(Y,F,G,H);
929 Heps = _mm_mul_ps(vfeps,H);
930 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
931 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
932 fvdw12 = _mm_mul_ps(c12_00,FF);
933 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
935 fscal = _mm_add_ps(felec,fvdw);
937 /* Calculate temporary vectorial force */
938 tx = _mm_mul_ps(fscal,dx00);
939 ty = _mm_mul_ps(fscal,dy00);
940 tz = _mm_mul_ps(fscal,dz00);
942 /* Update vectorial force */
943 fix0 = _mm_add_ps(fix0,tx);
944 fiy0 = _mm_add_ps(fiy0,ty);
945 fiz0 = _mm_add_ps(fiz0,tz);
947 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
948 f+j_coord_offsetC,f+j_coord_offsetD,
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 r10 = _mm_mul_ps(rsq10,rinv10);
957 /* Compute parameters for interactions between i and j atoms */
958 qq10 = _mm_mul_ps(iq1,jq0);
960 /* EWALD ELECTROSTATICS */
962 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
963 ewrt = _mm_mul_ps(r10,ewtabscale);
964 ewitab = _mm_cvttps_epi32(ewrt);
965 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
966 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
967 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
969 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
970 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
974 /* Calculate temporary vectorial force */
975 tx = _mm_mul_ps(fscal,dx10);
976 ty = _mm_mul_ps(fscal,dy10);
977 tz = _mm_mul_ps(fscal,dz10);
979 /* Update vectorial force */
980 fix1 = _mm_add_ps(fix1,tx);
981 fiy1 = _mm_add_ps(fiy1,ty);
982 fiz1 = _mm_add_ps(fiz1,tz);
984 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
985 f+j_coord_offsetC,f+j_coord_offsetD,
988 /**************************
989 * CALCULATE INTERACTIONS *
990 **************************/
992 r20 = _mm_mul_ps(rsq20,rinv20);
994 /* Compute parameters for interactions between i and j atoms */
995 qq20 = _mm_mul_ps(iq2,jq0);
997 /* EWALD ELECTROSTATICS */
999 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1000 ewrt = _mm_mul_ps(r20,ewtabscale);
1001 ewitab = _mm_cvttps_epi32(ewrt);
1002 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1003 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1004 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1006 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1007 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1011 /* Calculate temporary vectorial force */
1012 tx = _mm_mul_ps(fscal,dx20);
1013 ty = _mm_mul_ps(fscal,dy20);
1014 tz = _mm_mul_ps(fscal,dz20);
1016 /* Update vectorial force */
1017 fix2 = _mm_add_ps(fix2,tx);
1018 fiy2 = _mm_add_ps(fiy2,ty);
1019 fiz2 = _mm_add_ps(fiz2,tz);
1021 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1022 f+j_coord_offsetC,f+j_coord_offsetD,
1025 /* Inner loop uses 134 flops */
1028 if(jidx<j_index_end)
1031 /* Get j neighbor index, and coordinate index */
1033 jnrB = jjnr[jidx+1];
1034 jnrC = jjnr[jidx+2];
1035 jnrD = jjnr[jidx+3];
1037 /* Sign of each element will be negative for non-real atoms.
1038 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1039 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1041 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1042 jnrA = (jnrA>=0) ? jnrA : 0;
1043 jnrB = (jnrB>=0) ? jnrB : 0;
1044 jnrC = (jnrC>=0) ? jnrC : 0;
1045 jnrD = (jnrD>=0) ? jnrD : 0;
1047 j_coord_offsetA = DIM*jnrA;
1048 j_coord_offsetB = DIM*jnrB;
1049 j_coord_offsetC = DIM*jnrC;
1050 j_coord_offsetD = DIM*jnrD;
1052 /* load j atom coordinates */
1053 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1054 x+j_coord_offsetC,x+j_coord_offsetD,
1057 /* Calculate displacement vector */
1058 dx00 = _mm_sub_ps(ix0,jx0);
1059 dy00 = _mm_sub_ps(iy0,jy0);
1060 dz00 = _mm_sub_ps(iz0,jz0);
1061 dx10 = _mm_sub_ps(ix1,jx0);
1062 dy10 = _mm_sub_ps(iy1,jy0);
1063 dz10 = _mm_sub_ps(iz1,jz0);
1064 dx20 = _mm_sub_ps(ix2,jx0);
1065 dy20 = _mm_sub_ps(iy2,jy0);
1066 dz20 = _mm_sub_ps(iz2,jz0);
1068 /* Calculate squared distance and things based on it */
1069 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1070 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1071 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1073 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1074 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1075 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1077 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1078 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1079 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1081 /* Load parameters for j particles */
1082 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1083 charge+jnrC+0,charge+jnrD+0);
1084 vdwjidx0A = 2*vdwtype[jnrA+0];
1085 vdwjidx0B = 2*vdwtype[jnrB+0];
1086 vdwjidx0C = 2*vdwtype[jnrC+0];
1087 vdwjidx0D = 2*vdwtype[jnrD+0];
1089 /**************************
1090 * CALCULATE INTERACTIONS *
1091 **************************/
1093 r00 = _mm_mul_ps(rsq00,rinv00);
1094 r00 = _mm_andnot_ps(dummy_mask,r00);
1096 /* Compute parameters for interactions between i and j atoms */
1097 qq00 = _mm_mul_ps(iq0,jq0);
1098 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1099 vdwparam+vdwioffset0+vdwjidx0B,
1100 vdwparam+vdwioffset0+vdwjidx0C,
1101 vdwparam+vdwioffset0+vdwjidx0D,
1104 /* Calculate table index by multiplying r with table scale and truncate to integer */
1105 rt = _mm_mul_ps(r00,vftabscale);
1106 vfitab = _mm_cvttps_epi32(rt);
1107 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1108 vfitab = _mm_slli_epi32(vfitab,3);
1110 /* EWALD ELECTROSTATICS */
1112 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1113 ewrt = _mm_mul_ps(r00,ewtabscale);
1114 ewitab = _mm_cvttps_epi32(ewrt);
1115 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1116 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1117 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1119 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1120 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1122 /* CUBIC SPLINE TABLE DISPERSION */
1123 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1124 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1125 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1126 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1127 _MM_TRANSPOSE4_PS(Y,F,G,H);
1128 Heps = _mm_mul_ps(vfeps,H);
1129 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1130 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1131 fvdw6 = _mm_mul_ps(c6_00,FF);
1133 /* CUBIC SPLINE TABLE REPULSION */
1134 vfitab = _mm_add_epi32(vfitab,ifour);
1135 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1136 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1137 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1138 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1139 _MM_TRANSPOSE4_PS(Y,F,G,H);
1140 Heps = _mm_mul_ps(vfeps,H);
1141 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1142 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1143 fvdw12 = _mm_mul_ps(c12_00,FF);
1144 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1146 fscal = _mm_add_ps(felec,fvdw);
1148 fscal = _mm_andnot_ps(dummy_mask,fscal);
1150 /* Calculate temporary vectorial force */
1151 tx = _mm_mul_ps(fscal,dx00);
1152 ty = _mm_mul_ps(fscal,dy00);
1153 tz = _mm_mul_ps(fscal,dz00);
1155 /* Update vectorial force */
1156 fix0 = _mm_add_ps(fix0,tx);
1157 fiy0 = _mm_add_ps(fiy0,ty);
1158 fiz0 = _mm_add_ps(fiz0,tz);
1160 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1161 f+j_coord_offsetC,f+j_coord_offsetD,
1164 /**************************
1165 * CALCULATE INTERACTIONS *
1166 **************************/
1168 r10 = _mm_mul_ps(rsq10,rinv10);
1169 r10 = _mm_andnot_ps(dummy_mask,r10);
1171 /* Compute parameters for interactions between i and j atoms */
1172 qq10 = _mm_mul_ps(iq1,jq0);
1174 /* EWALD ELECTROSTATICS */
1176 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1177 ewrt = _mm_mul_ps(r10,ewtabscale);
1178 ewitab = _mm_cvttps_epi32(ewrt);
1179 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1180 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1181 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1183 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1184 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1188 fscal = _mm_andnot_ps(dummy_mask,fscal);
1190 /* Calculate temporary vectorial force */
1191 tx = _mm_mul_ps(fscal,dx10);
1192 ty = _mm_mul_ps(fscal,dy10);
1193 tz = _mm_mul_ps(fscal,dz10);
1195 /* Update vectorial force */
1196 fix1 = _mm_add_ps(fix1,tx);
1197 fiy1 = _mm_add_ps(fiy1,ty);
1198 fiz1 = _mm_add_ps(fiz1,tz);
1200 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1201 f+j_coord_offsetC,f+j_coord_offsetD,
1204 /**************************
1205 * CALCULATE INTERACTIONS *
1206 **************************/
1208 r20 = _mm_mul_ps(rsq20,rinv20);
1209 r20 = _mm_andnot_ps(dummy_mask,r20);
1211 /* Compute parameters for interactions between i and j atoms */
1212 qq20 = _mm_mul_ps(iq2,jq0);
1214 /* EWALD ELECTROSTATICS */
1216 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1217 ewrt = _mm_mul_ps(r20,ewtabscale);
1218 ewitab = _mm_cvttps_epi32(ewrt);
1219 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1220 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1221 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1223 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1224 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1228 fscal = _mm_andnot_ps(dummy_mask,fscal);
1230 /* Calculate temporary vectorial force */
1231 tx = _mm_mul_ps(fscal,dx20);
1232 ty = _mm_mul_ps(fscal,dy20);
1233 tz = _mm_mul_ps(fscal,dz20);
1235 /* Update vectorial force */
1236 fix2 = _mm_add_ps(fix2,tx);
1237 fiy2 = _mm_add_ps(fiy2,ty);
1238 fiz2 = _mm_add_ps(fiz2,tz);
1240 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1241 f+j_coord_offsetC,f+j_coord_offsetD,
1244 /* Inner loop uses 137 flops */
1247 /* End of innermost loop */
1249 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1250 f+i_coord_offset,fshift+i_shift_offset);
1252 /* Increment number of inner iterations */
1253 inneriter += j_index_end - j_index_start;
1255 /* Outer loop uses 27 flops */
1258 /* Increment number of outer iterations */
1261 /* Update outer/inner flops */
1263 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*137);