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_GeomW4P1_VF_sse2_single
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_VdwNone_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 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
72 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
74 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
75 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
76 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
77 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
78 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
79 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
80 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
83 __m128i ifour = _mm_set1_epi32(4);
84 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
86 __m128 dummy_mask,cutoff_mask;
87 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
88 __m128 one = _mm_set1_ps(1.0);
89 __m128 two = _mm_set1_ps(2.0);
95 jindex = nlist->jindex;
97 shiftidx = nlist->shift;
99 shiftvec = fr->shift_vec[0];
100 fshift = fr->fshift[0];
101 facel = _mm_set1_ps(fr->epsfac);
102 charge = mdatoms->chargeA;
104 vftab = kernel_data->table_elec->data;
105 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
107 /* Setup water-specific parameters */
108 inr = nlist->iinr[0];
109 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
110 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
111 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
113 /* Avoid stupid compiler warnings */
114 jnrA = jnrB = jnrC = jnrD = 0;
123 for(iidx=0;iidx<4*DIM;iidx++)
128 /* Start outer loop over neighborlists */
129 for(iidx=0; iidx<nri; iidx++)
131 /* Load shift vector for this list */
132 i_shift_offset = DIM*shiftidx[iidx];
134 /* Load limits for loop over neighbors */
135 j_index_start = jindex[iidx];
136 j_index_end = jindex[iidx+1];
138 /* Get outer coordinate index */
140 i_coord_offset = DIM*inr;
142 /* Load i particle coords and add shift vector */
143 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
144 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
146 fix1 = _mm_setzero_ps();
147 fiy1 = _mm_setzero_ps();
148 fiz1 = _mm_setzero_ps();
149 fix2 = _mm_setzero_ps();
150 fiy2 = _mm_setzero_ps();
151 fiz2 = _mm_setzero_ps();
152 fix3 = _mm_setzero_ps();
153 fiy3 = _mm_setzero_ps();
154 fiz3 = _mm_setzero_ps();
156 /* Reset potential sums */
157 velecsum = _mm_setzero_ps();
159 /* Start inner kernel loop */
160 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
163 /* Get j neighbor index, and coordinate index */
168 j_coord_offsetA = DIM*jnrA;
169 j_coord_offsetB = DIM*jnrB;
170 j_coord_offsetC = DIM*jnrC;
171 j_coord_offsetD = DIM*jnrD;
173 /* load j atom coordinates */
174 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
175 x+j_coord_offsetC,x+j_coord_offsetD,
178 /* Calculate displacement vector */
179 dx10 = _mm_sub_ps(ix1,jx0);
180 dy10 = _mm_sub_ps(iy1,jy0);
181 dz10 = _mm_sub_ps(iz1,jz0);
182 dx20 = _mm_sub_ps(ix2,jx0);
183 dy20 = _mm_sub_ps(iy2,jy0);
184 dz20 = _mm_sub_ps(iz2,jz0);
185 dx30 = _mm_sub_ps(ix3,jx0);
186 dy30 = _mm_sub_ps(iy3,jy0);
187 dz30 = _mm_sub_ps(iz3,jz0);
189 /* Calculate squared distance and things based on it */
190 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
191 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
192 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
194 rinv10 = gmx_mm_invsqrt_ps(rsq10);
195 rinv20 = gmx_mm_invsqrt_ps(rsq20);
196 rinv30 = gmx_mm_invsqrt_ps(rsq30);
198 /* Load parameters for j particles */
199 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
200 charge+jnrC+0,charge+jnrD+0);
202 /**************************
203 * CALCULATE INTERACTIONS *
204 **************************/
206 r10 = _mm_mul_ps(rsq10,rinv10);
208 /* Compute parameters for interactions between i and j atoms */
209 qq10 = _mm_mul_ps(iq1,jq0);
211 /* Calculate table index by multiplying r with table scale and truncate to integer */
212 rt = _mm_mul_ps(r10,vftabscale);
213 vfitab = _mm_cvttps_epi32(rt);
214 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
215 vfitab = _mm_slli_epi32(vfitab,2);
217 /* CUBIC SPLINE TABLE ELECTROSTATICS */
218 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
219 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
220 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
221 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
222 _MM_TRANSPOSE4_PS(Y,F,G,H);
223 Heps = _mm_mul_ps(vfeps,H);
224 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
225 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
226 velec = _mm_mul_ps(qq10,VV);
227 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
228 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
230 /* Update potential sum for this i atom from the interaction with this j atom. */
231 velecsum = _mm_add_ps(velecsum,velec);
235 /* Calculate temporary vectorial force */
236 tx = _mm_mul_ps(fscal,dx10);
237 ty = _mm_mul_ps(fscal,dy10);
238 tz = _mm_mul_ps(fscal,dz10);
240 /* Update vectorial force */
241 fix1 = _mm_add_ps(fix1,tx);
242 fiy1 = _mm_add_ps(fiy1,ty);
243 fiz1 = _mm_add_ps(fiz1,tz);
245 fjptrA = f+j_coord_offsetA;
246 fjptrB = f+j_coord_offsetB;
247 fjptrC = f+j_coord_offsetC;
248 fjptrD = f+j_coord_offsetD;
249 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
251 /**************************
252 * CALCULATE INTERACTIONS *
253 **************************/
255 r20 = _mm_mul_ps(rsq20,rinv20);
257 /* Compute parameters for interactions between i and j atoms */
258 qq20 = _mm_mul_ps(iq2,jq0);
260 /* Calculate table index by multiplying r with table scale and truncate to integer */
261 rt = _mm_mul_ps(r20,vftabscale);
262 vfitab = _mm_cvttps_epi32(rt);
263 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
264 vfitab = _mm_slli_epi32(vfitab,2);
266 /* CUBIC SPLINE TABLE ELECTROSTATICS */
267 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
268 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
269 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
270 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
271 _MM_TRANSPOSE4_PS(Y,F,G,H);
272 Heps = _mm_mul_ps(vfeps,H);
273 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
274 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
275 velec = _mm_mul_ps(qq20,VV);
276 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
277 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
279 /* Update potential sum for this i atom from the interaction with this j atom. */
280 velecsum = _mm_add_ps(velecsum,velec);
284 /* Calculate temporary vectorial force */
285 tx = _mm_mul_ps(fscal,dx20);
286 ty = _mm_mul_ps(fscal,dy20);
287 tz = _mm_mul_ps(fscal,dz20);
289 /* Update vectorial force */
290 fix2 = _mm_add_ps(fix2,tx);
291 fiy2 = _mm_add_ps(fiy2,ty);
292 fiz2 = _mm_add_ps(fiz2,tz);
294 fjptrA = f+j_coord_offsetA;
295 fjptrB = f+j_coord_offsetB;
296 fjptrC = f+j_coord_offsetC;
297 fjptrD = f+j_coord_offsetD;
298 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
300 /**************************
301 * CALCULATE INTERACTIONS *
302 **************************/
304 r30 = _mm_mul_ps(rsq30,rinv30);
306 /* Compute parameters for interactions between i and j atoms */
307 qq30 = _mm_mul_ps(iq3,jq0);
309 /* Calculate table index by multiplying r with table scale and truncate to integer */
310 rt = _mm_mul_ps(r30,vftabscale);
311 vfitab = _mm_cvttps_epi32(rt);
312 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
313 vfitab = _mm_slli_epi32(vfitab,2);
315 /* CUBIC SPLINE TABLE ELECTROSTATICS */
316 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
317 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
318 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
319 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
320 _MM_TRANSPOSE4_PS(Y,F,G,H);
321 Heps = _mm_mul_ps(vfeps,H);
322 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
323 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
324 velec = _mm_mul_ps(qq30,VV);
325 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
326 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
328 /* Update potential sum for this i atom from the interaction with this j atom. */
329 velecsum = _mm_add_ps(velecsum,velec);
333 /* Calculate temporary vectorial force */
334 tx = _mm_mul_ps(fscal,dx30);
335 ty = _mm_mul_ps(fscal,dy30);
336 tz = _mm_mul_ps(fscal,dz30);
338 /* Update vectorial force */
339 fix3 = _mm_add_ps(fix3,tx);
340 fiy3 = _mm_add_ps(fiy3,ty);
341 fiz3 = _mm_add_ps(fiz3,tz);
343 fjptrA = f+j_coord_offsetA;
344 fjptrB = f+j_coord_offsetB;
345 fjptrC = f+j_coord_offsetC;
346 fjptrD = f+j_coord_offsetD;
347 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
349 /* Inner loop uses 129 flops */
355 /* Get j neighbor index, and coordinate index */
356 jnrlistA = jjnr[jidx];
357 jnrlistB = jjnr[jidx+1];
358 jnrlistC = jjnr[jidx+2];
359 jnrlistD = jjnr[jidx+3];
360 /* Sign of each element will be negative for non-real atoms.
361 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
362 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
364 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
365 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
366 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
367 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
368 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
369 j_coord_offsetA = DIM*jnrA;
370 j_coord_offsetB = DIM*jnrB;
371 j_coord_offsetC = DIM*jnrC;
372 j_coord_offsetD = DIM*jnrD;
374 /* load j atom coordinates */
375 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
376 x+j_coord_offsetC,x+j_coord_offsetD,
379 /* Calculate displacement vector */
380 dx10 = _mm_sub_ps(ix1,jx0);
381 dy10 = _mm_sub_ps(iy1,jy0);
382 dz10 = _mm_sub_ps(iz1,jz0);
383 dx20 = _mm_sub_ps(ix2,jx0);
384 dy20 = _mm_sub_ps(iy2,jy0);
385 dz20 = _mm_sub_ps(iz2,jz0);
386 dx30 = _mm_sub_ps(ix3,jx0);
387 dy30 = _mm_sub_ps(iy3,jy0);
388 dz30 = _mm_sub_ps(iz3,jz0);
390 /* Calculate squared distance and things based on it */
391 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
392 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
393 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
395 rinv10 = gmx_mm_invsqrt_ps(rsq10);
396 rinv20 = gmx_mm_invsqrt_ps(rsq20);
397 rinv30 = gmx_mm_invsqrt_ps(rsq30);
399 /* Load parameters for j particles */
400 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
401 charge+jnrC+0,charge+jnrD+0);
403 /**************************
404 * CALCULATE INTERACTIONS *
405 **************************/
407 r10 = _mm_mul_ps(rsq10,rinv10);
408 r10 = _mm_andnot_ps(dummy_mask,r10);
410 /* Compute parameters for interactions between i and j atoms */
411 qq10 = _mm_mul_ps(iq1,jq0);
413 /* Calculate table index by multiplying r with table scale and truncate to integer */
414 rt = _mm_mul_ps(r10,vftabscale);
415 vfitab = _mm_cvttps_epi32(rt);
416 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
417 vfitab = _mm_slli_epi32(vfitab,2);
419 /* CUBIC SPLINE TABLE ELECTROSTATICS */
420 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
421 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
422 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
423 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
424 _MM_TRANSPOSE4_PS(Y,F,G,H);
425 Heps = _mm_mul_ps(vfeps,H);
426 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
427 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
428 velec = _mm_mul_ps(qq10,VV);
429 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
430 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
432 /* Update potential sum for this i atom from the interaction with this j atom. */
433 velec = _mm_andnot_ps(dummy_mask,velec);
434 velecsum = _mm_add_ps(velecsum,velec);
438 fscal = _mm_andnot_ps(dummy_mask,fscal);
440 /* Calculate temporary vectorial force */
441 tx = _mm_mul_ps(fscal,dx10);
442 ty = _mm_mul_ps(fscal,dy10);
443 tz = _mm_mul_ps(fscal,dz10);
445 /* Update vectorial force */
446 fix1 = _mm_add_ps(fix1,tx);
447 fiy1 = _mm_add_ps(fiy1,ty);
448 fiz1 = _mm_add_ps(fiz1,tz);
450 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
451 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
452 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
453 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
454 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
456 /**************************
457 * CALCULATE INTERACTIONS *
458 **************************/
460 r20 = _mm_mul_ps(rsq20,rinv20);
461 r20 = _mm_andnot_ps(dummy_mask,r20);
463 /* Compute parameters for interactions between i and j atoms */
464 qq20 = _mm_mul_ps(iq2,jq0);
466 /* Calculate table index by multiplying r with table scale and truncate to integer */
467 rt = _mm_mul_ps(r20,vftabscale);
468 vfitab = _mm_cvttps_epi32(rt);
469 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
470 vfitab = _mm_slli_epi32(vfitab,2);
472 /* CUBIC SPLINE TABLE ELECTROSTATICS */
473 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
474 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
475 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
476 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
477 _MM_TRANSPOSE4_PS(Y,F,G,H);
478 Heps = _mm_mul_ps(vfeps,H);
479 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
480 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
481 velec = _mm_mul_ps(qq20,VV);
482 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
483 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
485 /* Update potential sum for this i atom from the interaction with this j atom. */
486 velec = _mm_andnot_ps(dummy_mask,velec);
487 velecsum = _mm_add_ps(velecsum,velec);
491 fscal = _mm_andnot_ps(dummy_mask,fscal);
493 /* Calculate temporary vectorial force */
494 tx = _mm_mul_ps(fscal,dx20);
495 ty = _mm_mul_ps(fscal,dy20);
496 tz = _mm_mul_ps(fscal,dz20);
498 /* Update vectorial force */
499 fix2 = _mm_add_ps(fix2,tx);
500 fiy2 = _mm_add_ps(fiy2,ty);
501 fiz2 = _mm_add_ps(fiz2,tz);
503 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
504 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
505 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
506 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
507 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
509 /**************************
510 * CALCULATE INTERACTIONS *
511 **************************/
513 r30 = _mm_mul_ps(rsq30,rinv30);
514 r30 = _mm_andnot_ps(dummy_mask,r30);
516 /* Compute parameters for interactions between i and j atoms */
517 qq30 = _mm_mul_ps(iq3,jq0);
519 /* Calculate table index by multiplying r with table scale and truncate to integer */
520 rt = _mm_mul_ps(r30,vftabscale);
521 vfitab = _mm_cvttps_epi32(rt);
522 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
523 vfitab = _mm_slli_epi32(vfitab,2);
525 /* CUBIC SPLINE TABLE ELECTROSTATICS */
526 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
527 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
528 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
529 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
530 _MM_TRANSPOSE4_PS(Y,F,G,H);
531 Heps = _mm_mul_ps(vfeps,H);
532 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
533 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
534 velec = _mm_mul_ps(qq30,VV);
535 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
536 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
538 /* Update potential sum for this i atom from the interaction with this j atom. */
539 velec = _mm_andnot_ps(dummy_mask,velec);
540 velecsum = _mm_add_ps(velecsum,velec);
544 fscal = _mm_andnot_ps(dummy_mask,fscal);
546 /* Calculate temporary vectorial force */
547 tx = _mm_mul_ps(fscal,dx30);
548 ty = _mm_mul_ps(fscal,dy30);
549 tz = _mm_mul_ps(fscal,dz30);
551 /* Update vectorial force */
552 fix3 = _mm_add_ps(fix3,tx);
553 fiy3 = _mm_add_ps(fiy3,ty);
554 fiz3 = _mm_add_ps(fiz3,tz);
556 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
557 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
558 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
559 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
560 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
562 /* Inner loop uses 132 flops */
565 /* End of innermost loop */
567 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
568 f+i_coord_offset+DIM,fshift+i_shift_offset);
571 /* Update potential energies */
572 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
574 /* Increment number of inner iterations */
575 inneriter += j_index_end - j_index_start;
577 /* Outer loop uses 19 flops */
580 /* Increment number of outer iterations */
583 /* Update outer/inner flops */
585 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*132);
588 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse2_single
589 * Electrostatics interaction: CubicSplineTable
590 * VdW interaction: None
591 * Geometry: Water4-Particle
592 * Calculate force/pot: Force
595 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse2_single
596 (t_nblist * gmx_restrict nlist,
597 rvec * gmx_restrict xx,
598 rvec * gmx_restrict ff,
599 t_forcerec * gmx_restrict fr,
600 t_mdatoms * gmx_restrict mdatoms,
601 nb_kernel_data_t * gmx_restrict kernel_data,
602 t_nrnb * gmx_restrict nrnb)
604 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
605 * just 0 for non-waters.
606 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
607 * jnr indices corresponding to data put in the four positions in the SIMD register.
609 int i_shift_offset,i_coord_offset,outeriter,inneriter;
610 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
611 int jnrA,jnrB,jnrC,jnrD;
612 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
613 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
614 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
616 real *shiftvec,*fshift,*x,*f;
617 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
619 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
621 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
623 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
625 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
626 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
627 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
628 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
629 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
630 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
631 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
634 __m128i ifour = _mm_set1_epi32(4);
635 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
637 __m128 dummy_mask,cutoff_mask;
638 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
639 __m128 one = _mm_set1_ps(1.0);
640 __m128 two = _mm_set1_ps(2.0);
646 jindex = nlist->jindex;
648 shiftidx = nlist->shift;
650 shiftvec = fr->shift_vec[0];
651 fshift = fr->fshift[0];
652 facel = _mm_set1_ps(fr->epsfac);
653 charge = mdatoms->chargeA;
655 vftab = kernel_data->table_elec->data;
656 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
658 /* Setup water-specific parameters */
659 inr = nlist->iinr[0];
660 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
661 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
662 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
664 /* Avoid stupid compiler warnings */
665 jnrA = jnrB = jnrC = jnrD = 0;
674 for(iidx=0;iidx<4*DIM;iidx++)
679 /* Start outer loop over neighborlists */
680 for(iidx=0; iidx<nri; iidx++)
682 /* Load shift vector for this list */
683 i_shift_offset = DIM*shiftidx[iidx];
685 /* Load limits for loop over neighbors */
686 j_index_start = jindex[iidx];
687 j_index_end = jindex[iidx+1];
689 /* Get outer coordinate index */
691 i_coord_offset = DIM*inr;
693 /* Load i particle coords and add shift vector */
694 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
695 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
697 fix1 = _mm_setzero_ps();
698 fiy1 = _mm_setzero_ps();
699 fiz1 = _mm_setzero_ps();
700 fix2 = _mm_setzero_ps();
701 fiy2 = _mm_setzero_ps();
702 fiz2 = _mm_setzero_ps();
703 fix3 = _mm_setzero_ps();
704 fiy3 = _mm_setzero_ps();
705 fiz3 = _mm_setzero_ps();
707 /* Start inner kernel loop */
708 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
711 /* Get j neighbor index, and coordinate index */
716 j_coord_offsetA = DIM*jnrA;
717 j_coord_offsetB = DIM*jnrB;
718 j_coord_offsetC = DIM*jnrC;
719 j_coord_offsetD = DIM*jnrD;
721 /* load j atom coordinates */
722 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
723 x+j_coord_offsetC,x+j_coord_offsetD,
726 /* Calculate displacement vector */
727 dx10 = _mm_sub_ps(ix1,jx0);
728 dy10 = _mm_sub_ps(iy1,jy0);
729 dz10 = _mm_sub_ps(iz1,jz0);
730 dx20 = _mm_sub_ps(ix2,jx0);
731 dy20 = _mm_sub_ps(iy2,jy0);
732 dz20 = _mm_sub_ps(iz2,jz0);
733 dx30 = _mm_sub_ps(ix3,jx0);
734 dy30 = _mm_sub_ps(iy3,jy0);
735 dz30 = _mm_sub_ps(iz3,jz0);
737 /* Calculate squared distance and things based on it */
738 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
739 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
740 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
742 rinv10 = gmx_mm_invsqrt_ps(rsq10);
743 rinv20 = gmx_mm_invsqrt_ps(rsq20);
744 rinv30 = gmx_mm_invsqrt_ps(rsq30);
746 /* Load parameters for j particles */
747 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
748 charge+jnrC+0,charge+jnrD+0);
750 /**************************
751 * CALCULATE INTERACTIONS *
752 **************************/
754 r10 = _mm_mul_ps(rsq10,rinv10);
756 /* Compute parameters for interactions between i and j atoms */
757 qq10 = _mm_mul_ps(iq1,jq0);
759 /* Calculate table index by multiplying r with table scale and truncate to integer */
760 rt = _mm_mul_ps(r10,vftabscale);
761 vfitab = _mm_cvttps_epi32(rt);
762 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
763 vfitab = _mm_slli_epi32(vfitab,2);
765 /* CUBIC SPLINE TABLE ELECTROSTATICS */
766 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
767 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
768 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
769 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
770 _MM_TRANSPOSE4_PS(Y,F,G,H);
771 Heps = _mm_mul_ps(vfeps,H);
772 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
773 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
774 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
778 /* Calculate temporary vectorial force */
779 tx = _mm_mul_ps(fscal,dx10);
780 ty = _mm_mul_ps(fscal,dy10);
781 tz = _mm_mul_ps(fscal,dz10);
783 /* Update vectorial force */
784 fix1 = _mm_add_ps(fix1,tx);
785 fiy1 = _mm_add_ps(fiy1,ty);
786 fiz1 = _mm_add_ps(fiz1,tz);
788 fjptrA = f+j_coord_offsetA;
789 fjptrB = f+j_coord_offsetB;
790 fjptrC = f+j_coord_offsetC;
791 fjptrD = f+j_coord_offsetD;
792 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
794 /**************************
795 * CALCULATE INTERACTIONS *
796 **************************/
798 r20 = _mm_mul_ps(rsq20,rinv20);
800 /* Compute parameters for interactions between i and j atoms */
801 qq20 = _mm_mul_ps(iq2,jq0);
803 /* Calculate table index by multiplying r with table scale and truncate to integer */
804 rt = _mm_mul_ps(r20,vftabscale);
805 vfitab = _mm_cvttps_epi32(rt);
806 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
807 vfitab = _mm_slli_epi32(vfitab,2);
809 /* CUBIC SPLINE TABLE ELECTROSTATICS */
810 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
811 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
812 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
813 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
814 _MM_TRANSPOSE4_PS(Y,F,G,H);
815 Heps = _mm_mul_ps(vfeps,H);
816 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
817 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
818 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
822 /* Calculate temporary vectorial force */
823 tx = _mm_mul_ps(fscal,dx20);
824 ty = _mm_mul_ps(fscal,dy20);
825 tz = _mm_mul_ps(fscal,dz20);
827 /* Update vectorial force */
828 fix2 = _mm_add_ps(fix2,tx);
829 fiy2 = _mm_add_ps(fiy2,ty);
830 fiz2 = _mm_add_ps(fiz2,tz);
832 fjptrA = f+j_coord_offsetA;
833 fjptrB = f+j_coord_offsetB;
834 fjptrC = f+j_coord_offsetC;
835 fjptrD = f+j_coord_offsetD;
836 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
838 /**************************
839 * CALCULATE INTERACTIONS *
840 **************************/
842 r30 = _mm_mul_ps(rsq30,rinv30);
844 /* Compute parameters for interactions between i and j atoms */
845 qq30 = _mm_mul_ps(iq3,jq0);
847 /* Calculate table index by multiplying r with table scale and truncate to integer */
848 rt = _mm_mul_ps(r30,vftabscale);
849 vfitab = _mm_cvttps_epi32(rt);
850 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
851 vfitab = _mm_slli_epi32(vfitab,2);
853 /* CUBIC SPLINE TABLE ELECTROSTATICS */
854 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
855 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
856 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
857 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
858 _MM_TRANSPOSE4_PS(Y,F,G,H);
859 Heps = _mm_mul_ps(vfeps,H);
860 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
861 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
862 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
866 /* Calculate temporary vectorial force */
867 tx = _mm_mul_ps(fscal,dx30);
868 ty = _mm_mul_ps(fscal,dy30);
869 tz = _mm_mul_ps(fscal,dz30);
871 /* Update vectorial force */
872 fix3 = _mm_add_ps(fix3,tx);
873 fiy3 = _mm_add_ps(fiy3,ty);
874 fiz3 = _mm_add_ps(fiz3,tz);
876 fjptrA = f+j_coord_offsetA;
877 fjptrB = f+j_coord_offsetB;
878 fjptrC = f+j_coord_offsetC;
879 fjptrD = f+j_coord_offsetD;
880 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
882 /* Inner loop uses 117 flops */
888 /* Get j neighbor index, and coordinate index */
889 jnrlistA = jjnr[jidx];
890 jnrlistB = jjnr[jidx+1];
891 jnrlistC = jjnr[jidx+2];
892 jnrlistD = jjnr[jidx+3];
893 /* Sign of each element will be negative for non-real atoms.
894 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
895 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
897 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
898 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
899 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
900 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
901 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
902 j_coord_offsetA = DIM*jnrA;
903 j_coord_offsetB = DIM*jnrB;
904 j_coord_offsetC = DIM*jnrC;
905 j_coord_offsetD = DIM*jnrD;
907 /* load j atom coordinates */
908 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
909 x+j_coord_offsetC,x+j_coord_offsetD,
912 /* Calculate displacement vector */
913 dx10 = _mm_sub_ps(ix1,jx0);
914 dy10 = _mm_sub_ps(iy1,jy0);
915 dz10 = _mm_sub_ps(iz1,jz0);
916 dx20 = _mm_sub_ps(ix2,jx0);
917 dy20 = _mm_sub_ps(iy2,jy0);
918 dz20 = _mm_sub_ps(iz2,jz0);
919 dx30 = _mm_sub_ps(ix3,jx0);
920 dy30 = _mm_sub_ps(iy3,jy0);
921 dz30 = _mm_sub_ps(iz3,jz0);
923 /* Calculate squared distance and things based on it */
924 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
925 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
926 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
928 rinv10 = gmx_mm_invsqrt_ps(rsq10);
929 rinv20 = gmx_mm_invsqrt_ps(rsq20);
930 rinv30 = gmx_mm_invsqrt_ps(rsq30);
932 /* Load parameters for j particles */
933 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
934 charge+jnrC+0,charge+jnrD+0);
936 /**************************
937 * CALCULATE INTERACTIONS *
938 **************************/
940 r10 = _mm_mul_ps(rsq10,rinv10);
941 r10 = _mm_andnot_ps(dummy_mask,r10);
943 /* Compute parameters for interactions between i and j atoms */
944 qq10 = _mm_mul_ps(iq1,jq0);
946 /* Calculate table index by multiplying r with table scale and truncate to integer */
947 rt = _mm_mul_ps(r10,vftabscale);
948 vfitab = _mm_cvttps_epi32(rt);
949 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
950 vfitab = _mm_slli_epi32(vfitab,2);
952 /* CUBIC SPLINE TABLE ELECTROSTATICS */
953 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
954 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
955 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
956 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
957 _MM_TRANSPOSE4_PS(Y,F,G,H);
958 Heps = _mm_mul_ps(vfeps,H);
959 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
960 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
961 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
965 fscal = _mm_andnot_ps(dummy_mask,fscal);
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 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
978 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
979 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
980 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
981 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
983 /**************************
984 * CALCULATE INTERACTIONS *
985 **************************/
987 r20 = _mm_mul_ps(rsq20,rinv20);
988 r20 = _mm_andnot_ps(dummy_mask,r20);
990 /* Compute parameters for interactions between i and j atoms */
991 qq20 = _mm_mul_ps(iq2,jq0);
993 /* Calculate table index by multiplying r with table scale and truncate to integer */
994 rt = _mm_mul_ps(r20,vftabscale);
995 vfitab = _mm_cvttps_epi32(rt);
996 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
997 vfitab = _mm_slli_epi32(vfitab,2);
999 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1000 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1001 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1002 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1003 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1004 _MM_TRANSPOSE4_PS(Y,F,G,H);
1005 Heps = _mm_mul_ps(vfeps,H);
1006 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1007 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1008 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1012 fscal = _mm_andnot_ps(dummy_mask,fscal);
1014 /* Calculate temporary vectorial force */
1015 tx = _mm_mul_ps(fscal,dx20);
1016 ty = _mm_mul_ps(fscal,dy20);
1017 tz = _mm_mul_ps(fscal,dz20);
1019 /* Update vectorial force */
1020 fix2 = _mm_add_ps(fix2,tx);
1021 fiy2 = _mm_add_ps(fiy2,ty);
1022 fiz2 = _mm_add_ps(fiz2,tz);
1024 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1025 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1026 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1027 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1028 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1030 /**************************
1031 * CALCULATE INTERACTIONS *
1032 **************************/
1034 r30 = _mm_mul_ps(rsq30,rinv30);
1035 r30 = _mm_andnot_ps(dummy_mask,r30);
1037 /* Compute parameters for interactions between i and j atoms */
1038 qq30 = _mm_mul_ps(iq3,jq0);
1040 /* Calculate table index by multiplying r with table scale and truncate to integer */
1041 rt = _mm_mul_ps(r30,vftabscale);
1042 vfitab = _mm_cvttps_epi32(rt);
1043 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1044 vfitab = _mm_slli_epi32(vfitab,2);
1046 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1047 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1048 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1049 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1050 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1051 _MM_TRANSPOSE4_PS(Y,F,G,H);
1052 Heps = _mm_mul_ps(vfeps,H);
1053 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1054 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1055 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1059 fscal = _mm_andnot_ps(dummy_mask,fscal);
1061 /* Calculate temporary vectorial force */
1062 tx = _mm_mul_ps(fscal,dx30);
1063 ty = _mm_mul_ps(fscal,dy30);
1064 tz = _mm_mul_ps(fscal,dz30);
1066 /* Update vectorial force */
1067 fix3 = _mm_add_ps(fix3,tx);
1068 fiy3 = _mm_add_ps(fiy3,ty);
1069 fiz3 = _mm_add_ps(fiz3,tz);
1071 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1072 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1073 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1074 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1075 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1077 /* Inner loop uses 120 flops */
1080 /* End of innermost loop */
1082 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1083 f+i_coord_offset+DIM,fshift+i_shift_offset);
1085 /* Increment number of inner iterations */
1086 inneriter += j_index_end - j_index_start;
1088 /* Outer loop uses 18 flops */
1091 /* Increment number of outer iterations */
1094 /* Update outer/inner flops */
1096 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);