2 * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
34 #include "kernelutil_x86_sse4_1_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse4_1_single
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse4_1_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;
75 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
76 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
77 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
78 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
83 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
85 __m128 dummy_mask,cutoff_mask;
86 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
87 __m128 one = _mm_set1_ps(1.0);
88 __m128 two = _mm_set1_ps(2.0);
94 jindex = nlist->jindex;
96 shiftidx = nlist->shift;
98 shiftvec = fr->shift_vec[0];
99 fshift = fr->fshift[0];
100 facel = _mm_set1_ps(fr->epsfac);
101 charge = mdatoms->chargeA;
103 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
104 ewtab = fr->ic->tabq_coul_FDV0;
105 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
106 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
108 /* Setup water-specific parameters */
109 inr = nlist->iinr[0];
110 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
111 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
112 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
114 /* Avoid stupid compiler warnings */
115 jnrA = jnrB = jnrC = jnrD = 0;
124 for(iidx=0;iidx<4*DIM;iidx++)
129 /* Start outer loop over neighborlists */
130 for(iidx=0; iidx<nri; iidx++)
132 /* Load shift vector for this list */
133 i_shift_offset = DIM*shiftidx[iidx];
135 /* Load limits for loop over neighbors */
136 j_index_start = jindex[iidx];
137 j_index_end = jindex[iidx+1];
139 /* Get outer coordinate index */
141 i_coord_offset = DIM*inr;
143 /* Load i particle coords and add shift vector */
144 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
145 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
147 fix0 = _mm_setzero_ps();
148 fiy0 = _mm_setzero_ps();
149 fiz0 = _mm_setzero_ps();
150 fix1 = _mm_setzero_ps();
151 fiy1 = _mm_setzero_ps();
152 fiz1 = _mm_setzero_ps();
153 fix2 = _mm_setzero_ps();
154 fiy2 = _mm_setzero_ps();
155 fiz2 = _mm_setzero_ps();
157 /* Reset potential sums */
158 velecsum = _mm_setzero_ps();
160 /* Start inner kernel loop */
161 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
164 /* Get j neighbor index, and coordinate index */
169 j_coord_offsetA = DIM*jnrA;
170 j_coord_offsetB = DIM*jnrB;
171 j_coord_offsetC = DIM*jnrC;
172 j_coord_offsetD = DIM*jnrD;
174 /* load j atom coordinates */
175 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
176 x+j_coord_offsetC,x+j_coord_offsetD,
179 /* Calculate displacement vector */
180 dx00 = _mm_sub_ps(ix0,jx0);
181 dy00 = _mm_sub_ps(iy0,jy0);
182 dz00 = _mm_sub_ps(iz0,jz0);
183 dx10 = _mm_sub_ps(ix1,jx0);
184 dy10 = _mm_sub_ps(iy1,jy0);
185 dz10 = _mm_sub_ps(iz1,jz0);
186 dx20 = _mm_sub_ps(ix2,jx0);
187 dy20 = _mm_sub_ps(iy2,jy0);
188 dz20 = _mm_sub_ps(iz2,jz0);
190 /* Calculate squared distance and things based on it */
191 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
192 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
193 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
195 rinv00 = gmx_mm_invsqrt_ps(rsq00);
196 rinv10 = gmx_mm_invsqrt_ps(rsq10);
197 rinv20 = gmx_mm_invsqrt_ps(rsq20);
199 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
200 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
201 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
203 /* Load parameters for j particles */
204 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
205 charge+jnrC+0,charge+jnrD+0);
207 /**************************
208 * CALCULATE INTERACTIONS *
209 **************************/
211 r00 = _mm_mul_ps(rsq00,rinv00);
213 /* Compute parameters for interactions between i and j atoms */
214 qq00 = _mm_mul_ps(iq0,jq0);
216 /* EWALD ELECTROSTATICS */
218 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
219 ewrt = _mm_mul_ps(r00,ewtabscale);
220 ewitab = _mm_cvttps_epi32(ewrt);
221 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
222 ewitab = _mm_slli_epi32(ewitab,2);
223 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
224 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
225 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
226 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
227 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
228 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
229 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
230 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
231 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
233 /* Update potential sum for this i atom from the interaction with this j atom. */
234 velecsum = _mm_add_ps(velecsum,velec);
238 /* Calculate temporary vectorial force */
239 tx = _mm_mul_ps(fscal,dx00);
240 ty = _mm_mul_ps(fscal,dy00);
241 tz = _mm_mul_ps(fscal,dz00);
243 /* Update vectorial force */
244 fix0 = _mm_add_ps(fix0,tx);
245 fiy0 = _mm_add_ps(fiy0,ty);
246 fiz0 = _mm_add_ps(fiz0,tz);
248 fjptrA = f+j_coord_offsetA;
249 fjptrB = f+j_coord_offsetB;
250 fjptrC = f+j_coord_offsetC;
251 fjptrD = f+j_coord_offsetD;
252 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
254 /**************************
255 * CALCULATE INTERACTIONS *
256 **************************/
258 r10 = _mm_mul_ps(rsq10,rinv10);
260 /* Compute parameters for interactions between i and j atoms */
261 qq10 = _mm_mul_ps(iq1,jq0);
263 /* EWALD ELECTROSTATICS */
265 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
266 ewrt = _mm_mul_ps(r10,ewtabscale);
267 ewitab = _mm_cvttps_epi32(ewrt);
268 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
269 ewitab = _mm_slli_epi32(ewitab,2);
270 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
271 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
272 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
273 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
274 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
275 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
276 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
277 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
278 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velecsum = _mm_add_ps(velecsum,velec);
285 /* Calculate temporary vectorial force */
286 tx = _mm_mul_ps(fscal,dx10);
287 ty = _mm_mul_ps(fscal,dy10);
288 tz = _mm_mul_ps(fscal,dz10);
290 /* Update vectorial force */
291 fix1 = _mm_add_ps(fix1,tx);
292 fiy1 = _mm_add_ps(fiy1,ty);
293 fiz1 = _mm_add_ps(fiz1,tz);
295 fjptrA = f+j_coord_offsetA;
296 fjptrB = f+j_coord_offsetB;
297 fjptrC = f+j_coord_offsetC;
298 fjptrD = f+j_coord_offsetD;
299 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
301 /**************************
302 * CALCULATE INTERACTIONS *
303 **************************/
305 r20 = _mm_mul_ps(rsq20,rinv20);
307 /* Compute parameters for interactions between i and j atoms */
308 qq20 = _mm_mul_ps(iq2,jq0);
310 /* EWALD ELECTROSTATICS */
312 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
313 ewrt = _mm_mul_ps(r20,ewtabscale);
314 ewitab = _mm_cvttps_epi32(ewrt);
315 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
316 ewitab = _mm_slli_epi32(ewitab,2);
317 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
318 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
319 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
320 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
321 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
322 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
323 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
324 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
325 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
327 /* Update potential sum for this i atom from the interaction with this j atom. */
328 velecsum = _mm_add_ps(velecsum,velec);
332 /* Calculate temporary vectorial force */
333 tx = _mm_mul_ps(fscal,dx20);
334 ty = _mm_mul_ps(fscal,dy20);
335 tz = _mm_mul_ps(fscal,dz20);
337 /* Update vectorial force */
338 fix2 = _mm_add_ps(fix2,tx);
339 fiy2 = _mm_add_ps(fiy2,ty);
340 fiz2 = _mm_add_ps(fiz2,tz);
342 fjptrA = f+j_coord_offsetA;
343 fjptrB = f+j_coord_offsetB;
344 fjptrC = f+j_coord_offsetC;
345 fjptrD = f+j_coord_offsetD;
346 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
348 /* Inner loop uses 123 flops */
354 /* Get j neighbor index, and coordinate index */
355 jnrlistA = jjnr[jidx];
356 jnrlistB = jjnr[jidx+1];
357 jnrlistC = jjnr[jidx+2];
358 jnrlistD = jjnr[jidx+3];
359 /* Sign of each element will be negative for non-real atoms.
360 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
361 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
363 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
364 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
365 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
366 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
367 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
368 j_coord_offsetA = DIM*jnrA;
369 j_coord_offsetB = DIM*jnrB;
370 j_coord_offsetC = DIM*jnrC;
371 j_coord_offsetD = DIM*jnrD;
373 /* load j atom coordinates */
374 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
375 x+j_coord_offsetC,x+j_coord_offsetD,
378 /* Calculate displacement vector */
379 dx00 = _mm_sub_ps(ix0,jx0);
380 dy00 = _mm_sub_ps(iy0,jy0);
381 dz00 = _mm_sub_ps(iz0,jz0);
382 dx10 = _mm_sub_ps(ix1,jx0);
383 dy10 = _mm_sub_ps(iy1,jy0);
384 dz10 = _mm_sub_ps(iz1,jz0);
385 dx20 = _mm_sub_ps(ix2,jx0);
386 dy20 = _mm_sub_ps(iy2,jy0);
387 dz20 = _mm_sub_ps(iz2,jz0);
389 /* Calculate squared distance and things based on it */
390 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
391 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
392 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
394 rinv00 = gmx_mm_invsqrt_ps(rsq00);
395 rinv10 = gmx_mm_invsqrt_ps(rsq10);
396 rinv20 = gmx_mm_invsqrt_ps(rsq20);
398 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
399 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
400 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
402 /* Load parameters for j particles */
403 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
404 charge+jnrC+0,charge+jnrD+0);
406 /**************************
407 * CALCULATE INTERACTIONS *
408 **************************/
410 r00 = _mm_mul_ps(rsq00,rinv00);
411 r00 = _mm_andnot_ps(dummy_mask,r00);
413 /* Compute parameters for interactions between i and j atoms */
414 qq00 = _mm_mul_ps(iq0,jq0);
416 /* EWALD ELECTROSTATICS */
418 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
419 ewrt = _mm_mul_ps(r00,ewtabscale);
420 ewitab = _mm_cvttps_epi32(ewrt);
421 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
422 ewitab = _mm_slli_epi32(ewitab,2);
423 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
424 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
425 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
426 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
427 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
428 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
429 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
430 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
431 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
433 /* Update potential sum for this i atom from the interaction with this j atom. */
434 velec = _mm_andnot_ps(dummy_mask,velec);
435 velecsum = _mm_add_ps(velecsum,velec);
439 fscal = _mm_andnot_ps(dummy_mask,fscal);
441 /* Calculate temporary vectorial force */
442 tx = _mm_mul_ps(fscal,dx00);
443 ty = _mm_mul_ps(fscal,dy00);
444 tz = _mm_mul_ps(fscal,dz00);
446 /* Update vectorial force */
447 fix0 = _mm_add_ps(fix0,tx);
448 fiy0 = _mm_add_ps(fiy0,ty);
449 fiz0 = _mm_add_ps(fiz0,tz);
451 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
452 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
453 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
454 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
455 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
457 /**************************
458 * CALCULATE INTERACTIONS *
459 **************************/
461 r10 = _mm_mul_ps(rsq10,rinv10);
462 r10 = _mm_andnot_ps(dummy_mask,r10);
464 /* Compute parameters for interactions between i and j atoms */
465 qq10 = _mm_mul_ps(iq1,jq0);
467 /* EWALD ELECTROSTATICS */
469 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
470 ewrt = _mm_mul_ps(r10,ewtabscale);
471 ewitab = _mm_cvttps_epi32(ewrt);
472 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
473 ewitab = _mm_slli_epi32(ewitab,2);
474 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
475 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
476 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
477 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
478 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
479 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
480 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
481 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
482 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
484 /* Update potential sum for this i atom from the interaction with this j atom. */
485 velec = _mm_andnot_ps(dummy_mask,velec);
486 velecsum = _mm_add_ps(velecsum,velec);
490 fscal = _mm_andnot_ps(dummy_mask,fscal);
492 /* Calculate temporary vectorial force */
493 tx = _mm_mul_ps(fscal,dx10);
494 ty = _mm_mul_ps(fscal,dy10);
495 tz = _mm_mul_ps(fscal,dz10);
497 /* Update vectorial force */
498 fix1 = _mm_add_ps(fix1,tx);
499 fiy1 = _mm_add_ps(fiy1,ty);
500 fiz1 = _mm_add_ps(fiz1,tz);
502 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
503 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
504 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
505 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
506 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
508 /**************************
509 * CALCULATE INTERACTIONS *
510 **************************/
512 r20 = _mm_mul_ps(rsq20,rinv20);
513 r20 = _mm_andnot_ps(dummy_mask,r20);
515 /* Compute parameters for interactions between i and j atoms */
516 qq20 = _mm_mul_ps(iq2,jq0);
518 /* EWALD ELECTROSTATICS */
520 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
521 ewrt = _mm_mul_ps(r20,ewtabscale);
522 ewitab = _mm_cvttps_epi32(ewrt);
523 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
524 ewitab = _mm_slli_epi32(ewitab,2);
525 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
526 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
527 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
528 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
529 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
530 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
531 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
532 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
533 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
535 /* Update potential sum for this i atom from the interaction with this j atom. */
536 velec = _mm_andnot_ps(dummy_mask,velec);
537 velecsum = _mm_add_ps(velecsum,velec);
541 fscal = _mm_andnot_ps(dummy_mask,fscal);
543 /* Calculate temporary vectorial force */
544 tx = _mm_mul_ps(fscal,dx20);
545 ty = _mm_mul_ps(fscal,dy20);
546 tz = _mm_mul_ps(fscal,dz20);
548 /* Update vectorial force */
549 fix2 = _mm_add_ps(fix2,tx);
550 fiy2 = _mm_add_ps(fiy2,ty);
551 fiz2 = _mm_add_ps(fiz2,tz);
553 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
554 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
555 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
556 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
557 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
559 /* Inner loop uses 126 flops */
562 /* End of innermost loop */
564 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
565 f+i_coord_offset,fshift+i_shift_offset);
568 /* Update potential energies */
569 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
571 /* Increment number of inner iterations */
572 inneriter += j_index_end - j_index_start;
574 /* Outer loop uses 19 flops */
577 /* Increment number of outer iterations */
580 /* Update outer/inner flops */
582 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*126);
585 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse4_1_single
586 * Electrostatics interaction: Ewald
587 * VdW interaction: None
588 * Geometry: Water3-Particle
589 * Calculate force/pot: Force
592 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse4_1_single
593 (t_nblist * gmx_restrict nlist,
594 rvec * gmx_restrict xx,
595 rvec * gmx_restrict ff,
596 t_forcerec * gmx_restrict fr,
597 t_mdatoms * gmx_restrict mdatoms,
598 nb_kernel_data_t * gmx_restrict kernel_data,
599 t_nrnb * gmx_restrict nrnb)
601 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
602 * just 0 for non-waters.
603 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
604 * jnr indices corresponding to data put in the four positions in the SIMD register.
606 int i_shift_offset,i_coord_offset,outeriter,inneriter;
607 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
608 int jnrA,jnrB,jnrC,jnrD;
609 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
610 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
611 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
613 real *shiftvec,*fshift,*x,*f;
614 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
616 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
618 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
620 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
622 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
623 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
624 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
625 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
626 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
627 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
628 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
631 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
633 __m128 dummy_mask,cutoff_mask;
634 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
635 __m128 one = _mm_set1_ps(1.0);
636 __m128 two = _mm_set1_ps(2.0);
642 jindex = nlist->jindex;
644 shiftidx = nlist->shift;
646 shiftvec = fr->shift_vec[0];
647 fshift = fr->fshift[0];
648 facel = _mm_set1_ps(fr->epsfac);
649 charge = mdatoms->chargeA;
651 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
652 ewtab = fr->ic->tabq_coul_F;
653 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
654 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
656 /* Setup water-specific parameters */
657 inr = nlist->iinr[0];
658 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
659 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
660 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
662 /* Avoid stupid compiler warnings */
663 jnrA = jnrB = jnrC = jnrD = 0;
672 for(iidx=0;iidx<4*DIM;iidx++)
677 /* Start outer loop over neighborlists */
678 for(iidx=0; iidx<nri; iidx++)
680 /* Load shift vector for this list */
681 i_shift_offset = DIM*shiftidx[iidx];
683 /* Load limits for loop over neighbors */
684 j_index_start = jindex[iidx];
685 j_index_end = jindex[iidx+1];
687 /* Get outer coordinate index */
689 i_coord_offset = DIM*inr;
691 /* Load i particle coords and add shift vector */
692 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
693 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
695 fix0 = _mm_setzero_ps();
696 fiy0 = _mm_setzero_ps();
697 fiz0 = _mm_setzero_ps();
698 fix1 = _mm_setzero_ps();
699 fiy1 = _mm_setzero_ps();
700 fiz1 = _mm_setzero_ps();
701 fix2 = _mm_setzero_ps();
702 fiy2 = _mm_setzero_ps();
703 fiz2 = _mm_setzero_ps();
705 /* Start inner kernel loop */
706 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
709 /* Get j neighbor index, and coordinate index */
714 j_coord_offsetA = DIM*jnrA;
715 j_coord_offsetB = DIM*jnrB;
716 j_coord_offsetC = DIM*jnrC;
717 j_coord_offsetD = DIM*jnrD;
719 /* load j atom coordinates */
720 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
721 x+j_coord_offsetC,x+j_coord_offsetD,
724 /* Calculate displacement vector */
725 dx00 = _mm_sub_ps(ix0,jx0);
726 dy00 = _mm_sub_ps(iy0,jy0);
727 dz00 = _mm_sub_ps(iz0,jz0);
728 dx10 = _mm_sub_ps(ix1,jx0);
729 dy10 = _mm_sub_ps(iy1,jy0);
730 dz10 = _mm_sub_ps(iz1,jz0);
731 dx20 = _mm_sub_ps(ix2,jx0);
732 dy20 = _mm_sub_ps(iy2,jy0);
733 dz20 = _mm_sub_ps(iz2,jz0);
735 /* Calculate squared distance and things based on it */
736 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
737 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
738 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
740 rinv00 = gmx_mm_invsqrt_ps(rsq00);
741 rinv10 = gmx_mm_invsqrt_ps(rsq10);
742 rinv20 = gmx_mm_invsqrt_ps(rsq20);
744 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
745 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
746 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
748 /* Load parameters for j particles */
749 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
750 charge+jnrC+0,charge+jnrD+0);
752 /**************************
753 * CALCULATE INTERACTIONS *
754 **************************/
756 r00 = _mm_mul_ps(rsq00,rinv00);
758 /* Compute parameters for interactions between i and j atoms */
759 qq00 = _mm_mul_ps(iq0,jq0);
761 /* EWALD ELECTROSTATICS */
763 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
764 ewrt = _mm_mul_ps(r00,ewtabscale);
765 ewitab = _mm_cvttps_epi32(ewrt);
766 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
767 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
768 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
770 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
771 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
775 /* Calculate temporary vectorial force */
776 tx = _mm_mul_ps(fscal,dx00);
777 ty = _mm_mul_ps(fscal,dy00);
778 tz = _mm_mul_ps(fscal,dz00);
780 /* Update vectorial force */
781 fix0 = _mm_add_ps(fix0,tx);
782 fiy0 = _mm_add_ps(fiy0,ty);
783 fiz0 = _mm_add_ps(fiz0,tz);
785 fjptrA = f+j_coord_offsetA;
786 fjptrB = f+j_coord_offsetB;
787 fjptrC = f+j_coord_offsetC;
788 fjptrD = f+j_coord_offsetD;
789 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
791 /**************************
792 * CALCULATE INTERACTIONS *
793 **************************/
795 r10 = _mm_mul_ps(rsq10,rinv10);
797 /* Compute parameters for interactions between i and j atoms */
798 qq10 = _mm_mul_ps(iq1,jq0);
800 /* EWALD ELECTROSTATICS */
802 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
803 ewrt = _mm_mul_ps(r10,ewtabscale);
804 ewitab = _mm_cvttps_epi32(ewrt);
805 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
806 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
807 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
809 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
810 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
814 /* Calculate temporary vectorial force */
815 tx = _mm_mul_ps(fscal,dx10);
816 ty = _mm_mul_ps(fscal,dy10);
817 tz = _mm_mul_ps(fscal,dz10);
819 /* Update vectorial force */
820 fix1 = _mm_add_ps(fix1,tx);
821 fiy1 = _mm_add_ps(fiy1,ty);
822 fiz1 = _mm_add_ps(fiz1,tz);
824 fjptrA = f+j_coord_offsetA;
825 fjptrB = f+j_coord_offsetB;
826 fjptrC = f+j_coord_offsetC;
827 fjptrD = f+j_coord_offsetD;
828 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
830 /**************************
831 * CALCULATE INTERACTIONS *
832 **************************/
834 r20 = _mm_mul_ps(rsq20,rinv20);
836 /* Compute parameters for interactions between i and j atoms */
837 qq20 = _mm_mul_ps(iq2,jq0);
839 /* EWALD ELECTROSTATICS */
841 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
842 ewrt = _mm_mul_ps(r20,ewtabscale);
843 ewitab = _mm_cvttps_epi32(ewrt);
844 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
845 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
846 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
848 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
849 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
853 /* Calculate temporary vectorial force */
854 tx = _mm_mul_ps(fscal,dx20);
855 ty = _mm_mul_ps(fscal,dy20);
856 tz = _mm_mul_ps(fscal,dz20);
858 /* Update vectorial force */
859 fix2 = _mm_add_ps(fix2,tx);
860 fiy2 = _mm_add_ps(fiy2,ty);
861 fiz2 = _mm_add_ps(fiz2,tz);
863 fjptrA = f+j_coord_offsetA;
864 fjptrB = f+j_coord_offsetB;
865 fjptrC = f+j_coord_offsetC;
866 fjptrD = f+j_coord_offsetD;
867 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
869 /* Inner loop uses 108 flops */
875 /* Get j neighbor index, and coordinate index */
876 jnrlistA = jjnr[jidx];
877 jnrlistB = jjnr[jidx+1];
878 jnrlistC = jjnr[jidx+2];
879 jnrlistD = jjnr[jidx+3];
880 /* Sign of each element will be negative for non-real atoms.
881 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
882 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
884 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
885 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
886 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
887 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
888 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
889 j_coord_offsetA = DIM*jnrA;
890 j_coord_offsetB = DIM*jnrB;
891 j_coord_offsetC = DIM*jnrC;
892 j_coord_offsetD = DIM*jnrD;
894 /* load j atom coordinates */
895 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
896 x+j_coord_offsetC,x+j_coord_offsetD,
899 /* Calculate displacement vector */
900 dx00 = _mm_sub_ps(ix0,jx0);
901 dy00 = _mm_sub_ps(iy0,jy0);
902 dz00 = _mm_sub_ps(iz0,jz0);
903 dx10 = _mm_sub_ps(ix1,jx0);
904 dy10 = _mm_sub_ps(iy1,jy0);
905 dz10 = _mm_sub_ps(iz1,jz0);
906 dx20 = _mm_sub_ps(ix2,jx0);
907 dy20 = _mm_sub_ps(iy2,jy0);
908 dz20 = _mm_sub_ps(iz2,jz0);
910 /* Calculate squared distance and things based on it */
911 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
912 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
913 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
915 rinv00 = gmx_mm_invsqrt_ps(rsq00);
916 rinv10 = gmx_mm_invsqrt_ps(rsq10);
917 rinv20 = gmx_mm_invsqrt_ps(rsq20);
919 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
920 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
921 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
923 /* Load parameters for j particles */
924 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
925 charge+jnrC+0,charge+jnrD+0);
927 /**************************
928 * CALCULATE INTERACTIONS *
929 **************************/
931 r00 = _mm_mul_ps(rsq00,rinv00);
932 r00 = _mm_andnot_ps(dummy_mask,r00);
934 /* Compute parameters for interactions between i and j atoms */
935 qq00 = _mm_mul_ps(iq0,jq0);
937 /* EWALD ELECTROSTATICS */
939 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
940 ewrt = _mm_mul_ps(r00,ewtabscale);
941 ewitab = _mm_cvttps_epi32(ewrt);
942 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
943 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
944 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
946 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
947 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
951 fscal = _mm_andnot_ps(dummy_mask,fscal);
953 /* Calculate temporary vectorial force */
954 tx = _mm_mul_ps(fscal,dx00);
955 ty = _mm_mul_ps(fscal,dy00);
956 tz = _mm_mul_ps(fscal,dz00);
958 /* Update vectorial force */
959 fix0 = _mm_add_ps(fix0,tx);
960 fiy0 = _mm_add_ps(fiy0,ty);
961 fiz0 = _mm_add_ps(fiz0,tz);
963 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
964 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
965 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
966 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
967 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
969 /**************************
970 * CALCULATE INTERACTIONS *
971 **************************/
973 r10 = _mm_mul_ps(rsq10,rinv10);
974 r10 = _mm_andnot_ps(dummy_mask,r10);
976 /* Compute parameters for interactions between i and j atoms */
977 qq10 = _mm_mul_ps(iq1,jq0);
979 /* EWALD ELECTROSTATICS */
981 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
982 ewrt = _mm_mul_ps(r10,ewtabscale);
983 ewitab = _mm_cvttps_epi32(ewrt);
984 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
985 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
986 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
988 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
989 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
993 fscal = _mm_andnot_ps(dummy_mask,fscal);
995 /* Calculate temporary vectorial force */
996 tx = _mm_mul_ps(fscal,dx10);
997 ty = _mm_mul_ps(fscal,dy10);
998 tz = _mm_mul_ps(fscal,dz10);
1000 /* Update vectorial force */
1001 fix1 = _mm_add_ps(fix1,tx);
1002 fiy1 = _mm_add_ps(fiy1,ty);
1003 fiz1 = _mm_add_ps(fiz1,tz);
1005 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1006 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1007 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1008 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1009 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1011 /**************************
1012 * CALCULATE INTERACTIONS *
1013 **************************/
1015 r20 = _mm_mul_ps(rsq20,rinv20);
1016 r20 = _mm_andnot_ps(dummy_mask,r20);
1018 /* Compute parameters for interactions between i and j atoms */
1019 qq20 = _mm_mul_ps(iq2,jq0);
1021 /* EWALD ELECTROSTATICS */
1023 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1024 ewrt = _mm_mul_ps(r20,ewtabscale);
1025 ewitab = _mm_cvttps_epi32(ewrt);
1026 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1027 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1028 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1030 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1031 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1035 fscal = _mm_andnot_ps(dummy_mask,fscal);
1037 /* Calculate temporary vectorial force */
1038 tx = _mm_mul_ps(fscal,dx20);
1039 ty = _mm_mul_ps(fscal,dy20);
1040 tz = _mm_mul_ps(fscal,dz20);
1042 /* Update vectorial force */
1043 fix2 = _mm_add_ps(fix2,tx);
1044 fiy2 = _mm_add_ps(fiy2,ty);
1045 fiz2 = _mm_add_ps(fiz2,tz);
1047 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1048 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1049 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1050 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1051 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1053 /* Inner loop uses 111 flops */
1056 /* End of innermost loop */
1058 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1059 f+i_coord_offset,fshift+i_shift_offset);
1061 /* Increment number of inner iterations */
1062 inneriter += j_index_end - j_index_start;
1064 /* Outer loop uses 18 flops */
1067 /* Increment number of outer iterations */
1070 /* Update outer/inner flops */
1072 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*111);